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 ([email protected])
  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 #include <linux/quotaops.h>
  44
  45 #define MLOG_MASK_PREFIX ML_NAMEI
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 #include "alloc.h"
  51 #include "dcache.h"
  52 #include "dir.h"
  53 #include "dlmglue.h"
  54 #include "extent_map.h"
  55 #include "file.h"
  56 #include "inode.h"
  57 #include "journal.h"
  58 #include "namei.h"
  59 #include "suballoc.h"
  60 #include "super.h"
  61 #include "symlink.h"
  62 #include "sysfile.h"
  63 #include "uptodate.h"
  64 #include "xattr.h"
  65 #include "acl.h"
  66
  67 #include "buffer_head_io.h"
  68
  69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  70                               struct inode *dir,
  71                               struct inode *inode,
  72                               dev_t dev,
  73                               struct buffer_head **new_fe_bh,
  74                               struct buffer_head *parent_fe_bh,
  75                               handle_t *handle,
  76                               struct ocfs2_alloc_context *inode_ac);
  77
  78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  79                                     struct inode **ret_orphan_dir,
  80                                     u64 blkno,
  81                                     char *name,
  82                                     struct ocfs2_dir_lookup_result *lookup);
  83
  84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  85                             handle_t *handle,
  86                             struct inode *inode,
  87                             struct buffer_head *fe_bh,
  88                             char *name,
  89                             struct ocfs2_dir_lookup_result *lookup,
  90                             struct inode *orphan_dir_inode);
  91
  92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  93                                      handle_t *handle,
  94                                      struct inode *inode,
  95                                      const char *symname);
  96
  97 /* An orphan dir name is an 8 byte value, printed as a hex string */
  98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  99
 100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 101                                    struct nameidata *nd)
 102 {
 103         int status;
 104         u64 blkno;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_inode_info *oi;
 108
 109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 110                    dentry->d_name.len, dentry->d_name.name);
 111
 112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 113                 ret = ERR_PTR(-ENAMETOOLONG);
 114                 goto bail;
 115         }
 116
 117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 119
 120         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 121         if (status < 0) {
 122                 if (status != -ENOENT)
 123                         mlog_errno(status);
 124                 ret = ERR_PTR(status);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 129                                             dentry->d_name.len, &blkno);
 130         if (status < 0)
 131                 goto bail_add;
 132
 133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 134         if (IS_ERR(inode)) {
 135                 ret = ERR_PTR(-EACCES);
 136                 goto bail_unlock;
 137         }
 138
 139         oi = OCFS2_I(inode);
 140         /* Clear any orphaned state... If we were able to look up the
 141          * inode from a directory, it certainly can't be orphaned. We
 142          * might have the bad state from a node which intended to
 143          * orphan this inode but crashed before it could commit the
 144          * unlink. */
 145         spin_lock(&oi->ip_lock);
 146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 147         spin_unlock(&oi->ip_lock);
 148
 149 bail_add:
 150         dentry->d_op = &ocfs2_dentry_ops;
 151         ret = d_splice_alias(inode, dentry);
 152
 153         if (inode) {
 154                 /*
 155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 156                  * dentry, it will d_move() it on top of ourse. The
 157                  * return value will indicate this however, so in
 158                  * those cases, we switch them around for the locking
 159                  * code.
 160                  *
 161                  * NOTE: This dentry already has ->d_op set from
 162                  * ocfs2_get_parent() and ocfs2_get_dentry()
 163                  */
 164                 if (ret)
 165                         dentry = ret;
 166
 167                 status = ocfs2_dentry_attach_lock(dentry, inode,
 168                                                   OCFS2_I(dir)->ip_blkno);
 169                 if (status) {
 170                         mlog_errno(status);
 171                         ret = ERR_PTR(status);
 172                         goto bail_unlock;
 173                 }
 174         }
 175
 176 bail_unlock:
 177         /* Don't drop the cluster lock until *after* the d_add --
 178          * unlink on another node will message us to remove that
 179          * dentry under this lock so otherwise we can race this with
 180          * the downconvert thread and have a stale dentry. */
 181         ocfs2_inode_unlock(dir, 0);
 182
 183 bail:
 184
 185         mlog_exit_ptr(ret);
 186
 187         return ret;
 188 }
 189
 190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
 191 {
 192         struct inode *inode;
 193
 194         inode = new_inode(dir->i_sb);
 195         if (!inode) {
 196                 mlog(ML_ERROR, "new_inode failed!\n");
 197                 return NULL;
 198         }
 199
 200         /* populate as many fields early on as possible - many of
 201          * these are used by the support functions here and in
 202          * callers. */
 203         if (S_ISDIR(mode))
 204                 inode->i_nlink = 2;
 205         else
 206                 inode->i_nlink = 1;
 207         inode->i_uid = current_fsuid();
 208         if (dir->i_mode & S_ISGID) {
 209                 inode->i_gid = dir->i_gid;
 210                 if (S_ISDIR(mode))
 211                         mode |= S_ISGID;
 212         } else
 213                 inode->i_gid = current_fsgid();
 214         inode->i_mode = mode;
 215         dquot_initialize(inode);
 216         return inode;
 217 }
 218
 219 static int ocfs2_mknod(struct inode *dir,
 220                        struct dentry *dentry,
 221                        int mode,
 222                        dev_t dev)
 223 {
 224         int status = 0;
 225         struct buffer_head *parent_fe_bh = NULL;
 226         handle_t *handle = NULL;
 227         struct ocfs2_super *osb;
 228         struct ocfs2_dinode *dirfe;
 229         struct buffer_head *new_fe_bh = NULL;
 230         struct inode *inode = NULL;
 231         struct ocfs2_alloc_context *inode_ac = NULL;
 232         struct ocfs2_alloc_context *data_ac = NULL;
 233         struct ocfs2_alloc_context *meta_ac = NULL;
 234         int want_clusters = 0;
 235         int want_meta = 0;
 236         int xattr_credits = 0;
 237         struct ocfs2_security_xattr_info si = {
 238                 .enable = 1,
 239         };
 240         int did_quota_inode = 0;
 241         struct ocfs2_dir_lookup_result lookup = { NULL, };
 242
 243         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 244                    (unsigned long)dev, dentry->d_name.len,
 245                    dentry->d_name.name);
 246
 247         dquot_initialize(dir);
 248
 249         /* get our super block */
 250         osb = OCFS2_SB(dir->i_sb);
 251
 252         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 253         if (status < 0) {
 254                 if (status != -ENOENT)
 255                         mlog_errno(status);
 256                 return status;
 257         }
 258
 259         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 260                 status = -EMLINK;
 261                 goto leave;
 262         }
 263
 264         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 265         if (!ocfs2_read_links_count(dirfe)) {
 266                 /* can't make a file in a deleted directory. */
 267                 status = -ENOENT;
 268                 goto leave;
 269         }
 270
 271         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 272                                            dentry->d_name.len);
 273         if (status)
 274                 goto leave;
 275
 276         /* get a spot inside the dir. */
 277         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 278                                               dentry->d_name.name,
 279                                               dentry->d_name.len, &lookup);
 280         if (status < 0) {
 281                 mlog_errno(status);
 282                 goto leave;
 283         }
 284
 285         /* reserve an inode spot */
 286         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 287         if (status < 0) {
 288                 if (status != -ENOSPC)
 289                         mlog_errno(status);
 290                 goto leave;
 291         }
 292
 293         inode = ocfs2_get_init_inode(dir, mode);
 294         if (!inode) {
 295                 status = -ENOMEM;
 296                 mlog_errno(status);
 297                 goto leave;
 298         }
 299
 300         /* get security xattr */
 301         status = ocfs2_init_security_get(inode, dir, &si);
 302         if (status) {
 303                 if (status == -EOPNOTSUPP)
 304                         si.enable = 0;
 305                 else {
 306                         mlog_errno(status);
 307                         goto leave;
 308                 }
 309         }
 310
 311         /* calculate meta data/clusters for setting security and acl xattr */
 312         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 313                                        &si, &want_clusters,
 314                                        &xattr_credits, &want_meta);
 315         if (status < 0) {
 316                 mlog_errno(status);
 317                 goto leave;
 318         }
 319
 320         /* Reserve a cluster if creating an extent based directory. */
 321         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 322                 want_clusters += 1;
 323
 324                 /* Dir indexing requires extra space as well */
 325                 if (ocfs2_supports_indexed_dirs(osb))
 326                         want_meta++;
 327         }
 328
 329         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 330         if (status < 0) {
 331                 if (status != -ENOSPC)
 332                         mlog_errno(status);
 333                 goto leave;
 334         }
 335
 336         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 337         if (status < 0) {
 338                 if (status != -ENOSPC)
 339                         mlog_errno(status);
 340                 goto leave;
 341         }
 342
 343         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 344                                                             S_ISDIR(mode),
 345                                                             xattr_credits));
 346         if (IS_ERR(handle)) {
 347                 status = PTR_ERR(handle);
 348                 handle = NULL;
 349                 mlog_errno(status);
 350                 goto leave;
 351         }
 352
 353         status = dquot_alloc_inode(inode);
 354         if (status)
 355                 goto leave;
 356         did_quota_inode = 1;
 357
 358         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
 359                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
 360                    dentry->d_name.name);
 361
 362         /* do the real work now. */
 363         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 364                                     &new_fe_bh, parent_fe_bh, handle,
 365                                     inode_ac);
 366         if (status < 0) {
 367                 mlog_errno(status);
 368                 goto leave;
 369         }
 370
 371         if (S_ISDIR(mode)) {
 372                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 373                                             new_fe_bh, data_ac, meta_ac);
 374                 if (status < 0) {
 375                         mlog_errno(status);
 376                         goto leave;
 377                 }
 378
 379                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 380                                                  parent_fe_bh,
 381                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 382                 if (status < 0) {
 383                         mlog_errno(status);
 384                         goto leave;
 385                 }
 386                 ocfs2_add_links_count(dirfe, 1);
 387                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 388                 if (status < 0) {
 389                         mlog_errno(status);
 390                         goto leave;
 391                 }
 392                 inc_nlink(dir);
 393         }
 394
 395         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 396                                 meta_ac, data_ac);
 397         if (status < 0) {
 398                 mlog_errno(status);
 399                 goto leave;
 400         }
 401
 402         if (si.enable) {
 403                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 404                                                  meta_ac, data_ac);
 405                 if (status < 0) {
 406                         mlog_errno(status);
 407                         goto leave;
 408                 }
 409         }
 410
 411         /*
 412          * Do this before adding the entry to the directory. We add
 413          * also set d_op after success so that ->d_iput() will cleanup
 414          * the dentry lock even if ocfs2_add_entry() fails below.
 415          */
 416         status = ocfs2_dentry_attach_lock(dentry, inode,
 417                                           OCFS2_I(dir)->ip_blkno);
 418         if (status) {
 419                 mlog_errno(status);
 420                 goto leave;
 421         }
 422         dentry->d_op = &ocfs2_dentry_ops;
 423
 424         status = ocfs2_add_entry(handle, dentry, inode,
 425                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 426                                  &lookup);
 427         if (status < 0) {
 428                 mlog_errno(status);
 429                 goto leave;
 430         }
 431
 432         insert_inode_hash(inode);
 433         d_instantiate(dentry, inode);
 434         status = 0;
 435 leave:
 436         if (status < 0 && did_quota_inode)
 437                 dquot_free_inode(inode);
 438         if (handle)
 439                 ocfs2_commit_trans(osb, handle);
 440
 441         ocfs2_inode_unlock(dir, 1);
 442
 443         if (status == -ENOSPC)
 444                 mlog(0, "Disk is full\n");
 445
 446         brelse(new_fe_bh);
 447         brelse(parent_fe_bh);
 448         kfree(si.name);
 449         kfree(si.value);
 450
 451         ocfs2_free_dir_lookup_result(&lookup);
 452
 453         if (inode_ac)
 454                 ocfs2_free_alloc_context(inode_ac);
 455
 456         if (data_ac)
 457                 ocfs2_free_alloc_context(data_ac);
 458
 459         if (meta_ac)
 460                 ocfs2_free_alloc_context(meta_ac);
 461
 462         /*
 463          * We should call iput after the i_mutex of the bitmap been
 464          * unlocked in ocfs2_free_alloc_context, or the
 465          * ocfs2_delete_inode will mutex_lock again.
 466          */
 467         if ((status < 0) && inode) {
 468                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 469                 clear_nlink(inode);
 470                 iput(inode);
 471         }
 472
 473         mlog_exit(status);
 474
 475         return status;
 476 }
 477
 478 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 479                               struct inode *dir,
 480                               struct inode *inode,
 481                               dev_t dev,
 482                               struct buffer_head **new_fe_bh,
 483                               struct buffer_head *parent_fe_bh,
 484                               handle_t *handle,
 485                               struct ocfs2_alloc_context *inode_ac)
 486 {
 487         int status = 0;
 488         struct ocfs2_dinode *fe = NULL;
 489         struct ocfs2_extent_list *fel;
 490         u64 fe_blkno = 0;
 491         u16 suballoc_bit;
 492         u16 feat;
 493
 494         *new_fe_bh = NULL;
 495
 496         status = ocfs2_claim_new_inode(osb, handle, dir, parent_fe_bh,
 497                                        inode_ac, &suballoc_bit, &fe_blkno);
 498         if (status < 0) {
 499                 mlog_errno(status);
 500                 goto leave;
 501         }
 502
 503         /* populate as many fields early on as possible - many of
 504          * these are used by the support functions here and in
 505          * callers. */
 506         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 507         OCFS2_I(inode)->ip_blkno = fe_blkno;
 508         spin_lock(&osb->osb_lock);
 509         inode->i_generation = osb->s_next_generation++;
 510         spin_unlock(&osb->osb_lock);
 511
 512         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 513         if (!*new_fe_bh) {
 514                 status = -EIO;
 515                 mlog_errno(status);
 516                 goto leave;
 517         }
 518         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 519
 520         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 521                                          *new_fe_bh,
 522                                          OCFS2_JOURNAL_ACCESS_CREATE);
 523         if (status < 0) {
 524                 mlog_errno(status);
 525                 goto leave;
 526         }
 527
 528         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 529         memset(fe, 0, osb->sb->s_blocksize);
 530
 531         fe->i_generation = cpu_to_le32(inode->i_generation);
 532         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 533         fe->i_blkno = cpu_to_le64(fe_blkno);
 534         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 535         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 536         fe->i_uid = cpu_to_le32(inode->i_uid);
 537         fe->i_gid = cpu_to_le32(inode->i_gid);
 538         fe->i_mode = cpu_to_le16(inode->i_mode);
 539         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 540                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 541
 542         ocfs2_set_links_count(fe, inode->i_nlink);
 543
 544         fe->i_last_eb_blk = 0;
 545         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 546         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 547         fe->i_atime = fe->i_ctime = fe->i_mtime =
 548                 cpu_to_le64(CURRENT_TIME.tv_sec);
 549         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 550                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 551         fe->i_dtime = 0;
 552
 553         /*
 554          * If supported, directories start with inline data. If inline
 555          * isn't supported, but indexing is, we start them as indexed.
 556          */
 557         feat = le16_to_cpu(fe->i_dyn_features);
 558         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 559                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 560
 561                 fe->id2.i_data.id_count = cpu_to_le16(
 562                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 563         } else {
 564                 fel = &fe->id2.i_list;
 565                 fel->l_tree_depth = 0;
 566                 fel->l_next_free_rec = 0;
 567                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 568         }
 569
 570         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 571         if (status < 0) {
 572                 mlog_errno(status);
 573                 goto leave;
 574         }
 575
 576         ocfs2_populate_inode(inode, fe, 1);
 577         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 578         if (!ocfs2_mount_local(osb)) {
 579                 status = ocfs2_create_new_inode_locks(inode);
 580                 if (status < 0)
 581                         mlog_errno(status);
 582         }
 583
 584         status = 0; /* error in ocfs2_create_new_inode_locks is not
 585                      * critical */
 586
 587 leave:
 588         if (status < 0) {
 589                 if (*new_fe_bh) {
 590                         brelse(*new_fe_bh);
 591                         *new_fe_bh = NULL;
 592                 }
 593         }
 594
 595         mlog_exit(status);
 596         return status;
 597 }
 598
 599 static int ocfs2_mkdir(struct inode *dir,
 600                        struct dentry *dentry,
 601                        int mode)
 602 {
 603         int ret;
 604
 605         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 606                    dentry->d_name.len, dentry->d_name.name);
 607         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 608         mlog_exit(ret);
 609
 610         return ret;
 611 }
 612
 613 static int ocfs2_create(struct inode *dir,
 614                         struct dentry *dentry,
 615                         int mode,
 616                         struct nameidata *nd)
 617 {
 618         int ret;
 619
 620         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 621                    dentry->d_name.len, dentry->d_name.name);
 622         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 623         mlog_exit(ret);
 624
 625         return ret;
 626 }
 627
 628 static int ocfs2_link(struct dentry *old_dentry,
 629                       struct inode *dir,
 630                       struct dentry *dentry)
 631 {
 632         handle_t *handle;
 633         struct inode *inode = old_dentry->d_inode;
 634         int err;
 635         struct buffer_head *fe_bh = NULL;
 636         struct buffer_head *parent_fe_bh = NULL;
 637         struct ocfs2_dinode *fe = NULL;
 638         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 639         struct ocfs2_dir_lookup_result lookup = { NULL, };
 640
 641         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 642                    old_dentry->d_name.len, old_dentry->d_name.name,
 643                    dentry->d_name.len, dentry->d_name.name);
 644
 645         if (S_ISDIR(inode->i_mode))
 646                 return -EPERM;
 647
 648         dquot_initialize(dir);
 649
 650         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 651         if (err < 0) {
 652                 if (err != -ENOENT)
 653                         mlog_errno(err);
 654                 return err;
 655         }
 656
 657         if (!dir->i_nlink) {
 658                 err = -ENOENT;
 659                 goto out;
 660         }
 661
 662         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 663                                         dentry->d_name.len);
 664         if (err)
 665                 goto out;
 666
 667         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 668                                            dentry->d_name.name,
 669                                            dentry->d_name.len, &lookup);
 670         if (err < 0) {
 671                 mlog_errno(err);
 672                 goto out;
 673         }
 674
 675         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 676         if (err < 0) {
 677                 if (err != -ENOENT)
 678                         mlog_errno(err);
 679                 goto out;
 680         }
 681
 682         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 683         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 684                 err = -EMLINK;
 685                 goto out_unlock_inode;
 686         }
 687
 688         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 689         if (IS_ERR(handle)) {
 690                 err = PTR_ERR(handle);
 691                 handle = NULL;
 692                 mlog_errno(err);
 693                 goto out_unlock_inode;
 694         }
 695
 696         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 697                                       OCFS2_JOURNAL_ACCESS_WRITE);
 698         if (err < 0) {
 699                 mlog_errno(err);
 700                 goto out_commit;
 701         }
 702
 703         inc_nlink(inode);
 704         inode->i_ctime = CURRENT_TIME;
 705         ocfs2_set_links_count(fe, inode->i_nlink);
 706         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 707         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 708
 709         err = ocfs2_journal_dirty(handle, fe_bh);
 710         if (err < 0) {
 711                 ocfs2_add_links_count(fe, -1);
 712                 drop_nlink(inode);
 713                 mlog_errno(err);
 714                 goto out_commit;
 715         }
 716
 717         err = ocfs2_add_entry(handle, dentry, inode,
 718                               OCFS2_I(inode)->ip_blkno,
 719                               parent_fe_bh, &lookup);
 720         if (err) {
 721                 ocfs2_add_links_count(fe, -1);
 722                 drop_nlink(inode);
 723                 mlog_errno(err);
 724                 goto out_commit;
 725         }
 726
 727         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 728         if (err) {
 729                 mlog_errno(err);
 730                 goto out_commit;
 731         }
 732
 733         atomic_inc(&inode->i_count);
 734         dentry->d_op = &ocfs2_dentry_ops;
 735         d_instantiate(dentry, inode);
 736
 737 out_commit:
 738         ocfs2_commit_trans(osb, handle);
 739 out_unlock_inode:
 740         ocfs2_inode_unlock(inode, 1);
 741
 742 out:
 743         ocfs2_inode_unlock(dir, 1);
 744
 745         brelse(fe_bh);
 746         brelse(parent_fe_bh);
 747
 748         ocfs2_free_dir_lookup_result(&lookup);
 749
 750         mlog_exit(err);
 751
 752         return err;
 753 }
 754
 755 /*
 756  * Takes and drops an exclusive lock on the given dentry. This will
 757  * force other nodes to drop it.
 758  */
 759 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 760 {
 761         int ret;
 762
 763         ret = ocfs2_dentry_lock(dentry, 1);
 764         if (ret)
 765                 mlog_errno(ret);
 766         else
 767                 ocfs2_dentry_unlock(dentry, 1);
 768
 769         return ret;
 770 }
 771
 772 static inline int inode_is_unlinkable(struct inode *inode)
 773 {
 774         if (S_ISDIR(inode->i_mode)) {
 775                 if (inode->i_nlink == 2)
 776                         return 1;
 777                 return 0;
 778         }
 779
 780         if (inode->i_nlink == 1)
 781                 return 1;
 782         return 0;
 783 }
 784
 785 static int ocfs2_unlink(struct inode *dir,
 786                         struct dentry *dentry)
 787 {
 788         int status;
 789         int child_locked = 0;
 790         struct inode *inode = dentry->d_inode;
 791         struct inode *orphan_dir = NULL;
 792         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 793         u64 blkno;
 794         struct ocfs2_dinode *fe = NULL;
 795         struct buffer_head *fe_bh = NULL;
 796         struct buffer_head *parent_node_bh = NULL;
 797         handle_t *handle = NULL;
 798         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 799         struct ocfs2_dir_lookup_result lookup = { NULL, };
 800         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 801
 802         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 803                    dentry->d_name.len, dentry->d_name.name);
 804
 805         dquot_initialize(dir);
 806
 807         BUG_ON(dentry->d_parent->d_inode != dir);
 808
 809         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 810
 811         if (inode == osb->root_inode) {
 812                 mlog(0, "Cannot delete the root directory\n");
 813                 return -EPERM;
 814         }
 815
 816         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 817                                          OI_LS_PARENT);
 818         if (status < 0) {
 819                 if (status != -ENOENT)
 820                         mlog_errno(status);
 821                 return status;
 822         }
 823
 824         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 825                                           dentry->d_name.len, &blkno, dir,
 826                                           &lookup);
 827         if (status < 0) {
 828                 if (status != -ENOENT)
 829                         mlog_errno(status);
 830                 goto leave;
 831         }
 832
 833         if (OCFS2_I(inode)->ip_blkno != blkno) {
 834                 status = -ENOENT;
 835
 836                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 837                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 838                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 839                 goto leave;
 840         }
 841
 842         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 843         if (status < 0) {
 844                 if (status != -ENOENT)
 845                         mlog_errno(status);
 846                 goto leave;
 847         }
 848         child_locked = 1;
 849
 850         if (S_ISDIR(inode->i_mode)) {
 851                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 852                         status = -ENOTEMPTY;
 853                         goto leave;
 854                 }
 855         }
 856
 857         status = ocfs2_remote_dentry_delete(dentry);
 858         if (status < 0) {
 859                 /* This remote delete should succeed under all normal
 860                  * circumstances. */
 861                 mlog_errno(status);
 862                 goto leave;
 863         }
 864
 865         if (inode_is_unlinkable(inode)) {
 866                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 867                                                   OCFS2_I(inode)->ip_blkno,
 868                                                   orphan_name, &orphan_insert);
 869                 if (status < 0) {
 870                         mlog_errno(status);
 871                         goto leave;
 872                 }
 873         }
 874
 875         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 876         if (IS_ERR(handle)) {
 877                 status = PTR_ERR(handle);
 878                 handle = NULL;
 879                 mlog_errno(status);
 880                 goto leave;
 881         }
 882
 883         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 884                                          OCFS2_JOURNAL_ACCESS_WRITE);
 885         if (status < 0) {
 886                 mlog_errno(status);
 887                 goto leave;
 888         }
 889
 890         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 891
 892         if (inode_is_unlinkable(inode)) {
 893                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 894                                           &orphan_insert, orphan_dir);
 895                 if (status < 0) {
 896                         mlog_errno(status);
 897                         goto leave;
 898                 }
 899         }
 900
 901         /* delete the name from the parent dir */
 902         status = ocfs2_delete_entry(handle, dir, &lookup);
 903         if (status < 0) {
 904                 mlog_errno(status);
 905                 goto leave;
 906         }
 907
 908         if (S_ISDIR(inode->i_mode))
 909                 drop_nlink(inode);
 910         drop_nlink(inode);
 911         ocfs2_set_links_count(fe, inode->i_nlink);
 912
 913         status = ocfs2_journal_dirty(handle, fe_bh);
 914         if (status < 0) {
 915                 mlog_errno(status);
 916                 goto leave;
 917         }
 918
 919         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 920         if (S_ISDIR(inode->i_mode))
 921                 drop_nlink(dir);
 922
 923         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 924         if (status < 0) {
 925                 mlog_errno(status);
 926                 if (S_ISDIR(inode->i_mode))
 927                         inc_nlink(dir);
 928         }
 929
 930 leave:
 931         if (handle)
 932                 ocfs2_commit_trans(osb, handle);
 933
 934         if (child_locked)
 935                 ocfs2_inode_unlock(inode, 1);
 936
 937         ocfs2_inode_unlock(dir, 1);
 938
 939         if (orphan_dir) {
 940                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 941                 ocfs2_inode_unlock(orphan_dir, 1);
 942                 mutex_unlock(&orphan_dir->i_mutex);
 943                 iput(orphan_dir);
 944         }
 945
 946         brelse(fe_bh);
 947         brelse(parent_node_bh);
 948
 949         ocfs2_free_dir_lookup_result(&orphan_insert);
 950         ocfs2_free_dir_lookup_result(&lookup);
 951
 952         mlog_exit(status);
 953
 954         return status;
 955 }
 956
 957 /*
 958  * The only place this should be used is rename!
 959  * if they have the same id, then the 1st one is the only one locked.
 960  */
 961 static int ocfs2_double_lock(struct ocfs2_super *osb,
 962                              struct buffer_head **bh1,
 963                              struct inode *inode1,
 964                              struct buffer_head **bh2,
 965                              struct inode *inode2)
 966 {
 967         int status;
 968         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 969         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 970         struct buffer_head **tmpbh;
 971         struct inode *tmpinode;
 972
 973         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 974                    (unsigned long long)oi1->ip_blkno,
 975                    (unsigned long long)oi2->ip_blkno);
 976
 977         if (*bh1)
 978                 *bh1 = NULL;
 979         if (*bh2)
 980                 *bh2 = NULL;
 981
 982         /* we always want to lock the one with the lower lockid first. */
 983         if (oi1->ip_blkno != oi2->ip_blkno) {
 984                 if (oi1->ip_blkno < oi2->ip_blkno) {
 985                         /* switch id1 and id2 around */
 986                         mlog(0, "switching them around...\n");
 987                         tmpbh = bh2;
 988                         bh2 = bh1;
 989                         bh1 = tmpbh;
 990
 991                         tmpinode = inode2;
 992                         inode2 = inode1;
 993                         inode1 = tmpinode;
 994                 }
 995                 /* lock id2 */
 996                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 997                                                  OI_LS_RENAME1);
 998                 if (status < 0) {
 999                         if (status != -ENOENT)
1000                                 mlog_errno(status);
1001                         goto bail;
1002                 }
1003         }
1004
1005         /* lock id1 */
1006         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1007         if (status < 0) {
1008                 /*
1009                  * An error return must mean that no cluster locks
1010                  * were held on function exit.
1011                  */
1012                 if (oi1->ip_blkno != oi2->ip_blkno)
1013                         ocfs2_inode_unlock(inode2, 1);
1014
1015                 if (status != -ENOENT)
1016                         mlog_errno(status);
1017         }
1018
1019 bail:
1020         mlog_exit(status);
1021         return status;
1022 }
1023
1024 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1025 {
1026         ocfs2_inode_unlock(inode1, 1);
1027
1028         if (inode1 != inode2)
1029                 ocfs2_inode_unlock(inode2, 1);
1030 }
1031
1032 static int ocfs2_rename(struct inode *old_dir,
1033                         struct dentry *old_dentry,
1034                         struct inode *new_dir,
1035                         struct dentry *new_dentry)
1036 {
1037         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1038         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1039         struct inode *old_inode = old_dentry->d_inode;
1040         struct inode *new_inode = new_dentry->d_inode;
1041         struct inode *orphan_dir = NULL;
1042         struct ocfs2_dinode *newfe = NULL;
1043         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1044         struct buffer_head *newfe_bh = NULL;
1045         struct buffer_head *old_inode_bh = NULL;
1046         struct ocfs2_super *osb = NULL;
1047         u64 newfe_blkno, old_de_ino;
1048         handle_t *handle = NULL;
1049         struct buffer_head *old_dir_bh = NULL;
1050         struct buffer_head *new_dir_bh = NULL;
1051         nlink_t old_dir_nlink = old_dir->i_nlink;
1052         struct ocfs2_dinode *old_di;
1053         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1054         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1055         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1056         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1057         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1058
1059         /* At some point it might be nice to break this function up a
1060          * bit. */
1061
1062         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1063                    old_dir, old_dentry, new_dir, new_dentry,
1064                    old_dentry->d_name.len, old_dentry->d_name.name,
1065                    new_dentry->d_name.len, new_dentry->d_name.name);
1066
1067         dquot_initialize(old_dir);
1068         dquot_initialize(new_dir);
1069
1070         osb = OCFS2_SB(old_dir->i_sb);
1071
1072         if (new_inode) {
1073                 if (!igrab(new_inode))
1074                         BUG();
1075         }
1076
1077         /* Assume a directory hierarchy thusly:
1078          * a/b/c
1079          * a/d
1080          * a,b,c, and d are all directories.
1081          *
1082          * from cwd of 'a' on both nodes:
1083          * node1: mv b/c d
1084          * node2: mv d   b/c
1085          *
1086          * And that's why, just like the VFS, we need a file system
1087          * rename lock. */
1088         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1089                 status = ocfs2_rename_lock(osb);
1090                 if (status < 0) {
1091                         mlog_errno(status);
1092                         goto bail;
1093                 }
1094                 rename_lock = 1;
1095         }
1096
1097         /* if old and new are the same, this'll just do one lock. */
1098         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1099                                    &new_dir_bh, new_dir);
1100         if (status < 0) {
1101                 mlog_errno(status);
1102                 goto bail;
1103         }
1104         parents_locked = 1;
1105
1106         /* make sure both dirs have bhs
1107          * get an extra ref on old_dir_bh if old==new */
1108         if (!new_dir_bh) {
1109                 if (old_dir_bh) {
1110                         new_dir_bh = old_dir_bh;
1111                         get_bh(new_dir_bh);
1112                 } else {
1113                         mlog(ML_ERROR, "no old_dir_bh!\n");
1114                         status = -EIO;
1115                         goto bail;
1116                 }
1117         }
1118
1119         /*
1120          * Aside from allowing a meta data update, the locking here
1121          * also ensures that the downconvert thread on other nodes
1122          * won't have to concurrently downconvert the inode and the
1123          * dentry locks.
1124          */
1125         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1126                                          OI_LS_PARENT);
1127         if (status < 0) {
1128                 if (status != -ENOENT)
1129                         mlog_errno(status);
1130                 goto bail;
1131         }
1132         old_child_locked = 1;
1133
1134         status = ocfs2_remote_dentry_delete(old_dentry);
1135         if (status < 0) {
1136                 mlog_errno(status);
1137                 goto bail;
1138         }
1139
1140         if (S_ISDIR(old_inode->i_mode)) {
1141                 u64 old_inode_parent;
1142
1143                 update_dot_dot = 1;
1144                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1145                                                   old_inode,
1146                                                   &old_inode_dot_dot_res);
1147                 if (status) {
1148                         status = -EIO;
1149                         goto bail;
1150                 }
1151
1152                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1153                         status = -EIO;
1154                         goto bail;
1155                 }
1156
1157                 if (!new_inode && new_dir != old_dir &&
1158                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1159                         status = -EMLINK;
1160                         goto bail;
1161                 }
1162         }
1163
1164         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1165                                             old_dentry->d_name.len,
1166                                             &old_de_ino);
1167         if (status) {
1168                 status = -ENOENT;
1169                 goto bail;
1170         }
1171
1172         /*
1173          *  Check for inode number is _not_ due to possible IO errors.
1174          *  We might rmdir the source, keep it as pwd of some process
1175          *  and merrily kill the link to whatever was created under the
1176          *  same name. Goodbye sticky bit ;-<
1177          */
1178         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1179                 status = -ENOENT;
1180                 goto bail;
1181         }
1182
1183         /* check if the target already exists (in which case we need
1184          * to delete it */
1185         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1186                                           new_dentry->d_name.len,
1187                                           &newfe_blkno, new_dir,
1188                                           &target_lookup_res);
1189         /* The only error we allow here is -ENOENT because the new
1190          * file not existing is perfectly valid. */
1191         if ((status < 0) && (status != -ENOENT)) {
1192                 /* If we cannot find the file specified we should just */
1193                 /* return the error... */
1194                 mlog_errno(status);
1195                 goto bail;
1196         }
1197         if (status == 0)
1198                 target_exists = 1;
1199
1200         if (!target_exists && new_inode) {
1201                 /*
1202                  * Target was unlinked by another node while we were
1203                  * waiting to get to ocfs2_rename(). There isn't
1204                  * anything we can do here to help the situation, so
1205                  * bubble up the appropriate error.
1206                  */
1207                 status = -ENOENT;
1208                 goto bail;
1209         }
1210
1211         /* In case we need to overwrite an existing file, we blow it
1212          * away first */
1213         if (target_exists) {
1214                 /* VFS didn't think there existed an inode here, but
1215                  * someone else in the cluster must have raced our
1216                  * rename to create one. Today we error cleanly, in
1217                  * the future we should consider calling iget to build
1218                  * a new struct inode for this entry. */
1219                 if (!new_inode) {
1220                         status = -EACCES;
1221
1222                         mlog(0, "We found an inode for name %.*s but VFS "
1223                              "didn't give us one.\n", new_dentry->d_name.len,
1224                              new_dentry->d_name.name);
1225                         goto bail;
1226                 }
1227
1228                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1229                         status = -EACCES;
1230
1231                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1232                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1233                              (unsigned long long)newfe_blkno,
1234                              OCFS2_I(new_inode)->ip_flags);
1235                         goto bail;
1236                 }
1237
1238                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1239                 if (status < 0) {
1240                         if (status != -ENOENT)
1241                                 mlog_errno(status);
1242                         goto bail;
1243                 }
1244                 new_child_locked = 1;
1245
1246                 status = ocfs2_remote_dentry_delete(new_dentry);
1247                 if (status < 0) {
1248                         mlog_errno(status);
1249                         goto bail;
1250                 }
1251
1252                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1253
1254                 mlog(0, "aha rename over existing... new_blkno=%llu "
1255                      "newfebh=%p bhblocknr=%llu\n",
1256                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1257                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1258
1259                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1260                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1261                                                 OCFS2_I(new_inode)->ip_blkno,
1262                                                 orphan_name, &orphan_insert);
1263                         if (status < 0) {
1264                                 mlog_errno(status);
1265                                 goto bail;
1266                         }
1267                 }
1268         } else {
1269                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1270
1271                 status = ocfs2_check_dir_for_entry(new_dir,
1272                                                    new_dentry->d_name.name,
1273                                                    new_dentry->d_name.len);
1274                 if (status)
1275                         goto bail;
1276
1277                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1278                                                       new_dentry->d_name.name,
1279                                                       new_dentry->d_name.len,
1280                                                       &target_insert);
1281                 if (status < 0) {
1282                         mlog_errno(status);
1283                         goto bail;
1284                 }
1285         }
1286
1287         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1288         if (IS_ERR(handle)) {
1289                 status = PTR_ERR(handle);
1290                 handle = NULL;
1291                 mlog_errno(status);
1292                 goto bail;
1293         }
1294
1295         if (target_exists) {
1296                 if (S_ISDIR(new_inode->i_mode)) {
1297                         if (new_inode->i_nlink != 2 ||
1298                             !ocfs2_empty_dir(new_inode)) {
1299                                 status = -ENOTEMPTY;
1300                                 goto bail;
1301                         }
1302                 }
1303                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1304                                                  newfe_bh,
1305                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1306                 if (status < 0) {
1307                         mlog_errno(status);
1308                         goto bail;
1309                 }
1310
1311                 if (S_ISDIR(new_inode->i_mode) ||
1312                     (ocfs2_read_links_count(newfe) == 1)) {
1313                         status = ocfs2_orphan_add(osb, handle, new_inode,
1314                                                   newfe_bh, orphan_name,
1315                                                   &orphan_insert, orphan_dir);
1316                         if (status < 0) {
1317                                 mlog_errno(status);
1318                                 goto bail;
1319                         }
1320                 }
1321
1322                 /* change the dirent to point to the correct inode */
1323                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1324                                             old_inode);
1325                 if (status < 0) {
1326                         mlog_errno(status);
1327                         goto bail;
1328                 }
1329                 new_dir->i_version++;
1330
1331                 if (S_ISDIR(new_inode->i_mode))
1332                         ocfs2_set_links_count(newfe, 0);
1333                 else
1334                         ocfs2_add_links_count(newfe, -1);
1335
1336                 status = ocfs2_journal_dirty(handle, newfe_bh);
1337                 if (status < 0) {
1338                         mlog_errno(status);
1339                         goto bail;
1340                 }
1341         } else {
1342                 /* if the name was not found in new_dir, add it now */
1343                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1344                                          OCFS2_I(old_inode)->ip_blkno,
1345                                          new_dir_bh, &target_insert);
1346         }
1347
1348         old_inode->i_ctime = CURRENT_TIME;
1349         mark_inode_dirty(old_inode);
1350
1351         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1352                                          old_inode_bh,
1353                                          OCFS2_JOURNAL_ACCESS_WRITE);
1354         if (status >= 0) {
1355                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1356
1357                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1358                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1359
1360                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1361                 if (status < 0)
1362                         mlog_errno(status);
1363         } else
1364                 mlog_errno(status);
1365
1366         /*
1367          * Now that the name has been added to new_dir, remove the old name.
1368          *
1369          * We don't keep any directory entry context around until now
1370          * because the insert might have changed the type of directory
1371          * we're dealing with.
1372          */
1373         status = ocfs2_find_entry(old_dentry->d_name.name,
1374                                   old_dentry->d_name.len, old_dir,
1375                                   &old_entry_lookup);
1376         if (status)
1377                 goto bail;
1378
1379         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1380         if (status < 0) {
1381                 mlog_errno(status);
1382                 goto bail;
1383         }
1384
1385         if (new_inode) {
1386                 new_inode->i_nlink--;
1387                 new_inode->i_ctime = CURRENT_TIME;
1388         }
1389         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1390
1391         if (update_dot_dot) {
1392                 status = ocfs2_update_entry(old_inode, handle,
1393                                             &old_inode_dot_dot_res, new_dir);
1394                 old_dir->i_nlink--;
1395                 if (new_inode) {
1396                         new_inode->i_nlink--;
1397                 } else {
1398                         inc_nlink(new_dir);
1399                         mark_inode_dirty(new_dir);
1400                 }
1401         }
1402         mark_inode_dirty(old_dir);
1403         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1404         if (new_inode) {
1405                 mark_inode_dirty(new_inode);
1406                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1407         }
1408
1409         if (old_dir != new_dir) {
1410                 /* Keep the same times on both directories.*/
1411                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1412
1413                 /*
1414                  * This will also pick up the i_nlink change from the
1415                  * block above.
1416                  */
1417                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1418         }
1419
1420         if (old_dir_nlink != old_dir->i_nlink) {
1421                 if (!old_dir_bh) {
1422                         mlog(ML_ERROR, "need to change nlink for old dir "
1423                              "%llu from %d to %d but bh is NULL!\n",
1424                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1425                              (int)old_dir_nlink, old_dir->i_nlink);
1426                 } else {
1427                         struct ocfs2_dinode *fe;
1428                         status = ocfs2_journal_access_di(handle,
1429                                                          INODE_CACHE(old_dir),
1430                                                          old_dir_bh,
1431                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1432                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1433                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1434                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1435                 }
1436         }
1437         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1438         status = 0;
1439 bail:
1440         if (rename_lock)
1441                 ocfs2_rename_unlock(osb);
1442
1443         if (handle)
1444                 ocfs2_commit_trans(osb, handle);
1445
1446         if (parents_locked)
1447                 ocfs2_double_unlock(old_dir, new_dir);
1448
1449         if (old_child_locked)
1450                 ocfs2_inode_unlock(old_inode, 1);
1451
1452         if (new_child_locked)
1453                 ocfs2_inode_unlock(new_inode, 1);
1454
1455         if (orphan_dir) {
1456                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1457                 ocfs2_inode_unlock(orphan_dir, 1);
1458                 mutex_unlock(&orphan_dir->i_mutex);
1459                 iput(orphan_dir);
1460         }
1461
1462         if (new_inode)
1463                 sync_mapping_buffers(old_inode->i_mapping);
1464
1465         if (new_inode)
1466                 iput(new_inode);
1467
1468         ocfs2_free_dir_lookup_result(&target_lookup_res);
1469         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1470         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1471         ocfs2_free_dir_lookup_result(&orphan_insert);
1472         ocfs2_free_dir_lookup_result(&target_insert);
1473
1474         brelse(newfe_bh);
1475         brelse(old_inode_bh);
1476         brelse(old_dir_bh);
1477         brelse(new_dir_bh);
1478
1479         mlog_exit(status);
1480
1481         return status;
1482 }
1483
1484 /*
1485  * we expect i_size = strlen(symname). Copy symname into the file
1486  * data, including the null terminator.
1487  */
1488 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1489                                      handle_t *handle,
1490                                      struct inode *inode,
1491                                      const char *symname)
1492 {
1493         struct buffer_head **bhs = NULL;
1494         const char *c;
1495         struct super_block *sb = osb->sb;
1496         u64 p_blkno, p_blocks;
1497         int virtual, blocks, status, i, bytes_left;
1498
1499         bytes_left = i_size_read(inode) + 1;
1500         /* we can't trust i_blocks because we're actually going to
1501          * write i_size + 1 bytes. */
1502         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1503
1504         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1505                         (unsigned long long)inode->i_blocks,
1506                         i_size_read(inode), blocks);
1507
1508         /* Sanity check -- make sure we're going to fit. */
1509         if (bytes_left >
1510             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1511                 status = -EIO;
1512                 mlog_errno(status);
1513                 goto bail;
1514         }
1515
1516         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1517         if (!bhs) {
1518                 status = -ENOMEM;
1519                 mlog_errno(status);
1520                 goto bail;
1521         }
1522
1523         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1524                                              NULL);
1525         if (status < 0) {
1526                 mlog_errno(status);
1527                 goto bail;
1528         }
1529
1530         /* links can never be larger than one cluster so we know this
1531          * is all going to be contiguous, but do a sanity check
1532          * anyway. */
1533         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1534                 status = -EIO;
1535                 mlog_errno(status);
1536                 goto bail;
1537         }
1538
1539         virtual = 0;
1540         while(bytes_left > 0) {
1541                 c = &symname[virtual * sb->s_blocksize];
1542
1543                 bhs[virtual] = sb_getblk(sb, p_blkno);
1544                 if (!bhs[virtual]) {
1545                         status = -ENOMEM;
1546                         mlog_errno(status);
1547                         goto bail;
1548                 }
1549                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1550                                               bhs[virtual]);
1551
1552                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1553                                               bhs[virtual],
1554                                               OCFS2_JOURNAL_ACCESS_CREATE);
1555                 if (status < 0) {
1556                         mlog_errno(status);
1557                         goto bail;
1558                 }
1559
1560                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1561
1562                 memcpy(bhs[virtual]->b_data, c,
1563                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1564                        bytes_left);
1565
1566                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1567                 if (status < 0) {
1568                         mlog_errno(status);
1569                         goto bail;
1570                 }
1571
1572                 virtual++;
1573                 p_blkno++;
1574                 bytes_left -= sb->s_blocksize;
1575         }
1576
1577         status = 0;
1578 bail:
1579
1580         if (bhs) {
1581                 for(i = 0; i < blocks; i++)
1582                         brelse(bhs[i]);
1583                 kfree(bhs);
1584         }
1585
1586         mlog_exit(status);
1587         return status;
1588 }
1589
1590 static int ocfs2_symlink(struct inode *dir,
1591                          struct dentry *dentry,
1592                          const char *symname)
1593 {
1594         int status, l, credits;
1595         u64 newsize;
1596         struct ocfs2_super *osb = NULL;
1597         struct inode *inode = NULL;
1598         struct super_block *sb;
1599         struct buffer_head *new_fe_bh = NULL;
1600         struct buffer_head *parent_fe_bh = NULL;
1601         struct ocfs2_dinode *fe = NULL;
1602         struct ocfs2_dinode *dirfe;
1603         handle_t *handle = NULL;
1604         struct ocfs2_alloc_context *inode_ac = NULL;
1605         struct ocfs2_alloc_context *data_ac = NULL;
1606         struct ocfs2_alloc_context *xattr_ac = NULL;
1607         int want_clusters = 0;
1608         int xattr_credits = 0;
1609         struct ocfs2_security_xattr_info si = {
1610                 .enable = 1,
1611         };
1612         int did_quota = 0, did_quota_inode = 0;
1613         struct ocfs2_dir_lookup_result lookup = { NULL, };
1614
1615         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1616                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1617
1618         dquot_initialize(dir);
1619
1620         sb = dir->i_sb;
1621         osb = OCFS2_SB(sb);
1622
1623         l = strlen(symname) + 1;
1624
1625         credits = ocfs2_calc_symlink_credits(sb);
1626
1627         /* lock the parent directory */
1628         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1629         if (status < 0) {
1630                 if (status != -ENOENT)
1631                         mlog_errno(status);
1632                 return status;
1633         }
1634
1635         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1636         if (!ocfs2_read_links_count(dirfe)) {
1637                 /* can't make a file in a deleted directory. */
1638                 status = -ENOENT;
1639                 goto bail;
1640         }
1641
1642         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1643                                            dentry->d_name.len);
1644         if (status)
1645                 goto bail;
1646
1647         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1648                                               dentry->d_name.name,
1649                                               dentry->d_name.len, &lookup);
1650         if (status < 0) {
1651                 mlog_errno(status);
1652                 goto bail;
1653         }
1654
1655         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1656         if (status < 0) {
1657                 if (status != -ENOSPC)
1658                         mlog_errno(status);
1659                 goto bail;
1660         }
1661
1662         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1663         if (!inode) {
1664                 status = -ENOMEM;
1665                 mlog_errno(status);
1666                 goto bail;
1667         }
1668
1669         /* get security xattr */
1670         status = ocfs2_init_security_get(inode, dir, &si);
1671         if (status) {
1672                 if (status == -EOPNOTSUPP)
1673                         si.enable = 0;
1674                 else {
1675                         mlog_errno(status);
1676                         goto bail;
1677                 }
1678         }
1679
1680         /* calculate meta data/clusters for setting security xattr */
1681         if (si.enable) {
1682                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1683                                                   &xattr_credits, &xattr_ac);
1684                 if (status < 0) {
1685                         mlog_errno(status);
1686                         goto bail;
1687                 }
1688         }
1689
1690         /* don't reserve bitmap space for fast symlinks. */
1691         if (l > ocfs2_fast_symlink_chars(sb))
1692                 want_clusters += 1;
1693
1694         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1695         if (status < 0) {
1696                 if (status != -ENOSPC)
1697                         mlog_errno(status);
1698                 goto bail;
1699         }
1700
1701         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1702         if (IS_ERR(handle)) {
1703                 status = PTR_ERR(handle);
1704                 handle = NULL;
1705                 mlog_errno(status);
1706                 goto bail;
1707         }
1708
1709         status = dquot_alloc_inode(inode);
1710         if (status)
1711                 goto bail;
1712         did_quota_inode = 1;
1713
1714         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1715                    inode->i_mode, dentry->d_name.len,
1716                    dentry->d_name.name);
1717
1718         status = ocfs2_mknod_locked(osb, dir, inode,
1719                                     0, &new_fe_bh, parent_fe_bh, handle,
1720                                     inode_ac);
1721         if (status < 0) {
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1727         inode->i_rdev = 0;
1728         newsize = l - 1;
1729         if (l > ocfs2_fast_symlink_chars(sb)) {
1730                 u32 offset = 0;
1731
1732                 inode->i_op = &ocfs2_symlink_inode_operations;
1733                 status = dquot_alloc_space_nodirty(inode,
1734                     ocfs2_clusters_to_bytes(osb->sb, 1));
1735                 if (status)
1736                         goto bail;
1737                 did_quota = 1;
1738                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1739                                               new_fe_bh,
1740                                               handle, data_ac, NULL,
1741                                               NULL);
1742                 if (status < 0) {
1743                         if (status != -ENOSPC && status != -EINTR) {
1744                                 mlog(ML_ERROR,
1745                                      "Failed to extend file to %llu\n",
1746                                      (unsigned long long)newsize);
1747                                 mlog_errno(status);
1748                                 status = -ENOSPC;
1749                         }
1750                         goto bail;
1751                 }
1752                 i_size_write(inode, newsize);
1753                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1754         } else {
1755                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1756                 memcpy((char *) fe->id2.i_symlink, symname, l);
1757                 i_size_write(inode, newsize);
1758                 inode->i_blocks = 0;
1759         }
1760
1761         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1762         if (status < 0) {
1763                 mlog_errno(status);
1764                 goto bail;
1765         }
1766
1767         if (!ocfs2_inode_is_fast_symlink(inode)) {
1768                 status = ocfs2_create_symlink_data(osb, handle, inode,
1769                                                    symname);
1770                 if (status < 0) {
1771                         mlog_errno(status);
1772                         goto bail;
1773                 }
1774         }
1775
1776         if (si.enable) {
1777                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1778                                                  xattr_ac, data_ac);
1779                 if (status < 0) {
1780                         mlog_errno(status);
1781                         goto bail;
1782                 }
1783         }
1784
1785         /*
1786          * Do this before adding the entry to the directory. We add
1787          * also set d_op after success so that ->d_iput() will cleanup
1788          * the dentry lock even if ocfs2_add_entry() fails below.
1789          */
1790         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1791         if (status) {
1792                 mlog_errno(status);
1793                 goto bail;
1794         }
1795         dentry->d_op = &ocfs2_dentry_ops;
1796
1797         status = ocfs2_add_entry(handle, dentry, inode,
1798                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1799                                  &lookup);
1800         if (status < 0) {
1801                 mlog_errno(status);
1802                 goto bail;
1803         }
1804
1805         insert_inode_hash(inode);
1806         d_instantiate(dentry, inode);
1807 bail:
1808         if (status < 0 && did_quota)
1809                 dquot_free_space_nodirty(inode,
1810                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1811         if (status < 0 && did_quota_inode)
1812                 dquot_free_inode(inode);
1813         if (handle)
1814                 ocfs2_commit_trans(osb, handle);
1815
1816         ocfs2_inode_unlock(dir, 1);
1817
1818         brelse(new_fe_bh);
1819         brelse(parent_fe_bh);
1820         kfree(si.name);
1821         kfree(si.value);
1822         ocfs2_free_dir_lookup_result(&lookup);
1823         if (inode_ac)
1824                 ocfs2_free_alloc_context(inode_ac);
1825         if (data_ac)
1826                 ocfs2_free_alloc_context(data_ac);
1827         if (xattr_ac)
1828                 ocfs2_free_alloc_context(xattr_ac);
1829         if ((status < 0) && inode) {
1830                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1831                 clear_nlink(inode);
1832                 iput(inode);
1833         }
1834
1835         mlog_exit(status);
1836
1837         return status;
1838 }
1839
1840 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1841 {
1842         int status, namelen;
1843
1844         mlog_entry_void();
1845
1846         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1847                            (long long)blkno);
1848         if (namelen <= 0) {
1849                 if (namelen)
1850                         status = namelen;
1851                 else
1852                         status = -EINVAL;
1853                 mlog_errno(status);
1854                 goto bail;
1855         }
1856         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1857                 status = -EINVAL;
1858                 mlog_errno(status);
1859                 goto bail;
1860         }
1861
1862         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1863              namelen);
1864
1865         status = 0;
1866 bail:
1867         mlog_exit(status);
1868         return status;
1869 }
1870
1871 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1872                                     struct inode **ret_orphan_dir,
1873                                     u64 blkno,
1874                                     char *name,
1875                                     struct ocfs2_dir_lookup_result *lookup)
1876 {
1877         struct inode *orphan_dir_inode;
1878         struct buffer_head *orphan_dir_bh = NULL;
1879         int status = 0;
1880
1881         status = ocfs2_blkno_stringify(blkno, name);
1882         if (status < 0) {
1883                 mlog_errno(status);
1884                 return status;
1885         }
1886
1887         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1888                                                        ORPHAN_DIR_SYSTEM_INODE,
1889                                                        osb->slot_num);
1890         if (!orphan_dir_inode) {
1891                 status = -ENOENT;
1892                 mlog_errno(status);
1893                 return status;
1894         }
1895
1896         mutex_lock(&orphan_dir_inode->i_mutex);
1897
1898         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1899         if (status < 0) {
1900                 mlog_errno(status);
1901                 goto leave;
1902         }
1903
1904         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1905                                               orphan_dir_bh, name,
1906                                               OCFS2_ORPHAN_NAMELEN, lookup);
1907         if (status < 0) {
1908                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1909
1910                 mlog_errno(status);
1911                 goto leave;
1912         }
1913
1914         *ret_orphan_dir = orphan_dir_inode;
1915
1916 leave:
1917         if (status) {
1918                 mutex_unlock(&orphan_dir_inode->i_mutex);
1919                 iput(orphan_dir_inode);
1920         }
1921
1922         brelse(orphan_dir_bh);
1923
1924         mlog_exit(status);
1925         return status;
1926 }
1927
1928 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1929                             handle_t *handle,
1930                             struct inode *inode,
1931                             struct buffer_head *fe_bh,
1932                             char *name,
1933                             struct ocfs2_dir_lookup_result *lookup,
1934                             struct inode *orphan_dir_inode)
1935 {
1936         struct buffer_head *orphan_dir_bh = NULL;
1937         int status = 0;
1938         struct ocfs2_dinode *orphan_fe;
1939         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1940
1941         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1942
1943         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1944         if (status < 0) {
1945                 mlog_errno(status);
1946                 goto leave;
1947         }
1948
1949         status = ocfs2_journal_access_di(handle,
1950                                          INODE_CACHE(orphan_dir_inode),
1951                                          orphan_dir_bh,
1952                                          OCFS2_JOURNAL_ACCESS_WRITE);
1953         if (status < 0) {
1954                 mlog_errno(status);
1955                 goto leave;
1956         }
1957
1958         /* we're a cluster, and nlink can change on disk from
1959          * underneath us... */
1960         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1961         if (S_ISDIR(inode->i_mode))
1962                 ocfs2_add_links_count(orphan_fe, 1);
1963         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1964
1965         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1966         if (status < 0) {
1967                 mlog_errno(status);
1968                 goto leave;
1969         }
1970
1971         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1972                                    OCFS2_ORPHAN_NAMELEN, inode,
1973                                    OCFS2_I(inode)->ip_blkno,
1974                                    orphan_dir_bh, lookup);
1975         if (status < 0) {
1976                 mlog_errno(status);
1977                 goto leave;
1978         }
1979
1980         /*
1981          * We're going to journal the change of i_flags and i_orphaned_slot.
1982          * It's safe anyway, though some callers may duplicate the journaling.
1983          * Journaling within the func just make the logic look more
1984          * straightforward.
1985          */
1986         status = ocfs2_journal_access_di(handle,
1987                                          INODE_CACHE(inode),
1988                                          fe_bh,
1989                                          OCFS2_JOURNAL_ACCESS_WRITE);
1990         if (status < 0) {
1991                 mlog_errno(status);
1992                 goto leave;
1993         }
1994
1995         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1996         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
1997
1998         /* Record which orphan dir our inode now resides
1999          * in. delete_inode will use this to determine which orphan
2000          * dir to lock. */
2001         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2002
2003         ocfs2_journal_dirty(handle, fe_bh);
2004
2005         mlog(0, "Inode %llu orphaned in slot %d\n",
2006              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2007
2008 leave:
2009         brelse(orphan_dir_bh);
2010
2011         mlog_exit(status);
2012         return status;
2013 }
2014
2015 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2016 int ocfs2_orphan_del(struct ocfs2_super *osb,
2017                      handle_t *handle,
2018                      struct inode *orphan_dir_inode,
2019                      struct inode *inode,
2020                      struct buffer_head *orphan_dir_bh)
2021 {
2022         char name[OCFS2_ORPHAN_NAMELEN + 1];
2023         struct ocfs2_dinode *orphan_fe;
2024         int status = 0;
2025         struct ocfs2_dir_lookup_result lookup = { NULL, };
2026
2027         mlog_entry_void();
2028
2029         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2030         if (status < 0) {
2031                 mlog_errno(status);
2032                 goto leave;
2033         }
2034
2035         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2036              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2037              OCFS2_ORPHAN_NAMELEN);
2038
2039         /* find it's spot in the orphan directory */
2040         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2041                                   &lookup);
2042         if (status) {
2043                 mlog_errno(status);
2044                 goto leave;
2045         }
2046
2047         /* remove it from the orphan directory */
2048         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2049         if (status < 0) {
2050                 mlog_errno(status);
2051                 goto leave;
2052         }
2053
2054         status = ocfs2_journal_access_di(handle,
2055                                          INODE_CACHE(orphan_dir_inode),
2056                                          orphan_dir_bh,
2057                                          OCFS2_JOURNAL_ACCESS_WRITE);
2058         if (status < 0) {
2059                 mlog_errno(status);
2060                 goto leave;
2061         }
2062
2063         /* do the i_nlink dance! :) */
2064         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2065         if (S_ISDIR(inode->i_mode))
2066                 ocfs2_add_links_count(orphan_fe, -1);
2067         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2068
2069         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2070         if (status < 0) {
2071                 mlog_errno(status);
2072                 goto leave;
2073         }
2074
2075 leave:
2076         ocfs2_free_dir_lookup_result(&lookup);
2077
2078         mlog_exit(status);
2079         return status;
2080 }
2081
2082 int ocfs2_create_inode_in_orphan(struct inode *dir,
2083                                  int mode,
2084                                  struct inode **new_inode)
2085 {
2086         int status, did_quota_inode = 0;
2087         struct inode *inode = NULL;
2088         struct inode *orphan_dir = NULL;
2089         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2090         struct ocfs2_dinode *di = NULL;
2091         handle_t *handle = NULL;
2092         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2093         struct buffer_head *parent_di_bh = NULL;
2094         struct buffer_head *new_di_bh = NULL;
2095         struct ocfs2_alloc_context *inode_ac = NULL;
2096         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2097
2098         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2099         if (status < 0) {
2100                 if (status != -ENOENT)
2101                         mlog_errno(status);
2102                 return status;
2103         }
2104
2105         /*
2106          * We give the orphan dir the root blkno to fake an orphan name,
2107          * and allocate enough space for our insertion.
2108          */
2109         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
2110                                           osb->root_blkno,
2111                                           orphan_name, &orphan_insert);
2112         if (status < 0) {
2113                 mlog_errno(status);
2114                 goto leave;
2115         }
2116
2117         /* reserve an inode spot */
2118         status = ocfs2_reserve_new_inode(osb, &inode_ac);
2119         if (status < 0) {
2120                 if (status != -ENOSPC)
2121                         mlog_errno(status);
2122                 goto leave;
2123         }
2124
2125         inode = ocfs2_get_init_inode(dir, mode);
2126         if (!inode) {
2127                 status = -ENOMEM;
2128                 mlog_errno(status);
2129                 goto leave;
2130         }
2131
2132         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2133         if (IS_ERR(handle)) {
2134                 status = PTR_ERR(handle);
2135                 handle = NULL;
2136                 mlog_errno(status);
2137                 goto leave;
2138         }
2139
2140         status = dquot_alloc_inode(inode);
2141         if (status)
2142                 goto leave;
2143         did_quota_inode = 1;
2144
2145         inode->i_nlink = 0;
2146         /* do the real work now. */
2147         status = ocfs2_mknod_locked(osb, dir, inode,
2148                                     0, &new_di_bh, parent_di_bh, handle,
2149                                     inode_ac);
2150         if (status < 0) {
2151                 mlog_errno(status);
2152                 goto leave;
2153         }
2154
2155         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
2156         if (status < 0) {
2157                 mlog_errno(status);
2158                 goto leave;
2159         }
2160
2161         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2162         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2163                                   &orphan_insert, orphan_dir);
2164         if (status < 0) {
2165                 mlog_errno(status);
2166                 goto leave;
2167         }
2168
2169         /* get open lock so that only nodes can't remove it from orphan dir. */
2170         status = ocfs2_open_lock(inode);
2171         if (status < 0)
2172                 mlog_errno(status);
2173
2174         insert_inode_hash(inode);
2175 leave:
2176         if (status < 0 && did_quota_inode)
2177                 dquot_free_inode(inode);
2178         if (handle)
2179                 ocfs2_commit_trans(osb, handle);
2180
2181         if (orphan_dir) {
2182                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2183                 ocfs2_inode_unlock(orphan_dir, 1);
2184                 mutex_unlock(&orphan_dir->i_mutex);
2185                 iput(orphan_dir);
2186         }
2187
2188         if (status == -ENOSPC)
2189                 mlog(0, "Disk is full\n");
2190
2191         if ((status < 0) && inode) {
2192                 clear_nlink(inode);
2193                 iput(inode);
2194         }
2195
2196         if (inode_ac)
2197                 ocfs2_free_alloc_context(inode_ac);
2198
2199         brelse(new_di_bh);
2200
2201         if (!status)
2202                 *new_inode = inode;
2203
2204         ocfs2_free_dir_lookup_result(&orphan_insert);
2205
2206         ocfs2_inode_unlock(dir, 1);
2207         brelse(parent_di_bh);
2208         return status;
2209 }
2210
2211 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2212                                    struct inode *inode,
2213                                    struct dentry *dentry)
2214 {
2215         int status = 0;
2216         struct buffer_head *parent_di_bh = NULL;
2217         handle_t *handle = NULL;
2218         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2219         struct ocfs2_dinode *dir_di, *di;
2220         struct inode *orphan_dir_inode = NULL;
2221         struct buffer_head *orphan_dir_bh = NULL;
2222         struct buffer_head *di_bh = NULL;
2223         struct ocfs2_dir_lookup_result lookup = { NULL, };
2224
2225         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2226                    dentry->d_name.len, dentry->d_name.name);
2227
2228         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2229         if (status < 0) {
2230                 if (status != -ENOENT)
2231                         mlog_errno(status);
2232                 return status;
2233         }
2234
2235         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2236         if (!dir_di->i_links_count) {
2237                 /* can't make a file in a deleted directory. */
2238                 status = -ENOENT;
2239                 goto leave;
2240         }
2241
2242         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2243                                            dentry->d_name.len);
2244         if (status)
2245                 goto leave;
2246
2247         /* get a spot inside the dir. */
2248         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2249                                               dentry->d_name.name,
2250                                               dentry->d_name.len, &lookup);
2251         if (status < 0) {
2252                 mlog_errno(status);
2253                 goto leave;
2254         }
2255
2256         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2257                                                        ORPHAN_DIR_SYSTEM_INODE,
2258                                                        osb->slot_num);
2259         if (!orphan_dir_inode) {
2260                 status = -EEXIST;
2261                 mlog_errno(status);
2262                 goto leave;
2263         }
2264
2265         mutex_lock(&orphan_dir_inode->i_mutex);
2266
2267         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2268         if (status < 0) {
2269                 mlog_errno(status);
2270                 mutex_unlock(&orphan_dir_inode->i_mutex);
2271                 iput(orphan_dir_inode);
2272                 goto leave;
2273         }
2274
2275         status = ocfs2_read_inode_block(inode, &di_bh);
2276         if (status < 0) {
2277                 mlog_errno(status);
2278                 goto orphan_unlock;
2279         }
2280
2281         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2282         if (IS_ERR(handle)) {
2283                 status = PTR_ERR(handle);
2284                 handle = NULL;
2285                 mlog_errno(status);
2286                 goto orphan_unlock;
2287         }
2288
2289         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2290                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2291         if (status < 0) {
2292                 mlog_errno(status);
2293                 goto out_commit;
2294         }
2295
2296         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2297                                   orphan_dir_bh);
2298         if (status < 0) {
2299                 mlog_errno(status);
2300                 goto out_commit;
2301         }
2302
2303         di = (struct ocfs2_dinode *)di_bh->b_data;
2304         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2305         di->i_orphaned_slot = 0;
2306         inode->i_nlink = 1;
2307         ocfs2_set_links_count(di, inode->i_nlink);
2308         ocfs2_journal_dirty(handle, di_bh);
2309
2310         status = ocfs2_add_entry(handle, dentry, inode,
2311                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2312                                  &lookup);
2313         if (status < 0) {
2314                 mlog_errno(status);
2315                 goto out_commit;
2316         }
2317
2318         status = ocfs2_dentry_attach_lock(dentry, inode,
2319                                           OCFS2_I(dir)->ip_blkno);
2320         if (status) {
2321                 mlog_errno(status);
2322                 goto out_commit;
2323         }
2324
2325         dentry->d_op = &ocfs2_dentry_ops;
2326         d_instantiate(dentry, inode);
2327         status = 0;
2328 out_commit:
2329         ocfs2_commit_trans(osb, handle);
2330 orphan_unlock:
2331         ocfs2_inode_unlock(orphan_dir_inode, 1);
2332         mutex_unlock(&orphan_dir_inode->i_mutex);
2333         iput(orphan_dir_inode);
2334 leave:
2335
2336         ocfs2_inode_unlock(dir, 1);
2337
2338         brelse(di_bh);
2339         brelse(parent_di_bh);
2340         brelse(orphan_dir_bh);
2341
2342         ocfs2_free_dir_lookup_result(&lookup);
2343
2344         mlog_exit(status);
2345
2346         return status;
2347 }
2348
2349 const struct inode_operations ocfs2_dir_iops = {
2350         .create         = ocfs2_create,
2351         .lookup         = ocfs2_lookup,
2352         .link           = ocfs2_link,
2353         .unlink         = ocfs2_unlink,
2354         .rmdir          = ocfs2_unlink,
2355         .symlink        = ocfs2_symlink,
2356         .mkdir          = ocfs2_mkdir,
2357         .mknod          = ocfs2_mknod,
2358         .rename         = ocfs2_rename,
2359         .setattr        = ocfs2_setattr,
2360         .getattr        = ocfs2_getattr,
2361         .permission     = ocfs2_permission,
2362         .setxattr       = generic_setxattr,
2363         .getxattr       = generic_getxattr,
2364         .listxattr      = ocfs2_listxattr,
2365         .removexattr    = generic_removexattr,
2366         .fiemap         = ocfs2_fiemap,
2367 };