fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #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 ocfs2_dinode *fe,
  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         vfs_dq_init(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         /* get our super block */
 248         osb = OCFS2_SB(dir->i_sb);
 249
 250         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 251         if (status < 0) {
 252                 if (status != -ENOENT)
 253                         mlog_errno(status);
 254                 return status;
 255         }
 256
 257         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 258                 status = -EMLINK;
 259                 goto leave;
 260         }
 261
 262         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 263         if (!ocfs2_read_links_count(dirfe)) {
 264                 /* can't make a file in a deleted directory. */
 265                 status = -ENOENT;
 266                 goto leave;
 267         }
 268
 269         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 270                                            dentry->d_name.len);
 271         if (status)
 272                 goto leave;
 273
 274         /* get a spot inside the dir. */
 275         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 276                                               dentry->d_name.name,
 277                                               dentry->d_name.len, &lookup);
 278         if (status < 0) {
 279                 mlog_errno(status);
 280                 goto leave;
 281         }
 282
 283         /* reserve an inode spot */
 284         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 285         if (status < 0) {
 286                 if (status != -ENOSPC)
 287                         mlog_errno(status);
 288                 goto leave;
 289         }
 290
 291         inode = ocfs2_get_init_inode(dir, mode);
 292         if (!inode) {
 293                 status = -ENOMEM;
 294                 mlog_errno(status);
 295                 goto leave;
 296         }
 297
 298         /* get security xattr */
 299         status = ocfs2_init_security_get(inode, dir, &si);
 300         if (status) {
 301                 if (status == -EOPNOTSUPP)
 302                         si.enable = 0;
 303                 else {
 304                         mlog_errno(status);
 305                         goto leave;
 306                 }
 307         }
 308
 309         /* calculate meta data/clusters for setting security and acl xattr */
 310         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 311                                        &si, &want_clusters,
 312                                        &xattr_credits, &want_meta);
 313         if (status < 0) {
 314                 mlog_errno(status);
 315                 goto leave;
 316         }
 317
 318         /* Reserve a cluster if creating an extent based directory. */
 319         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 320                 want_clusters += 1;
 321
 322                 /* Dir indexing requires extra space as well */
 323                 if (ocfs2_supports_indexed_dirs(osb))
 324                         want_meta++;
 325         }
 326
 327         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 328         if (status < 0) {
 329                 if (status != -ENOSPC)
 330                         mlog_errno(status);
 331                 goto leave;
 332         }
 333
 334         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 335         if (status < 0) {
 336                 if (status != -ENOSPC)
 337                         mlog_errno(status);
 338                 goto leave;
 339         }
 340
 341         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 342                                                             S_ISDIR(mode),
 343                                                             xattr_credits));
 344         if (IS_ERR(handle)) {
 345                 status = PTR_ERR(handle);
 346                 handle = NULL;
 347                 mlog_errno(status);
 348                 goto leave;
 349         }
 350
 351         /* We don't use standard VFS wrapper because we don't want vfs_dq_init
 352          * to be called. */
 353         if (sb_any_quota_active(osb->sb) &&
 354             osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
 355                 status = -EDQUOT;
 356                 goto leave;
 357         }
 358         did_quota_inode = 1;
 359
 360         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
 361                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
 362                    dentry->d_name.name);
 363
 364         /* do the real work now. */
 365         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 366                                     &new_fe_bh, parent_fe_bh, handle,
 367                                     inode_ac);
 368         if (status < 0) {
 369                 mlog_errno(status);
 370                 goto leave;
 371         }
 372
 373         if (S_ISDIR(mode)) {
 374                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 375                                             new_fe_bh, data_ac, meta_ac);
 376                 if (status < 0) {
 377                         mlog_errno(status);
 378                         goto leave;
 379                 }
 380
 381                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 382                                                  parent_fe_bh,
 383                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 384                 if (status < 0) {
 385                         mlog_errno(status);
 386                         goto leave;
 387                 }
 388                 ocfs2_add_links_count(dirfe, 1);
 389                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 390                 if (status < 0) {
 391                         mlog_errno(status);
 392                         goto leave;
 393                 }
 394                 inc_nlink(dir);
 395         }
 396
 397         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 398                                 meta_ac, data_ac);
 399         if (status < 0) {
 400                 mlog_errno(status);
 401                 goto leave;
 402         }
 403
 404         if (si.enable) {
 405                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 406                                                  meta_ac, data_ac);
 407                 if (status < 0) {
 408                         mlog_errno(status);
 409                         goto leave;
 410                 }
 411         }
 412
 413         status = ocfs2_add_entry(handle, dentry, inode,
 414                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 415                                  &lookup);
 416         if (status < 0) {
 417                 mlog_errno(status);
 418                 goto leave;
 419         }
 420
 421         status = ocfs2_dentry_attach_lock(dentry, inode,
 422                                           OCFS2_I(dir)->ip_blkno);
 423         if (status) {
 424                 mlog_errno(status);
 425                 goto leave;
 426         }
 427
 428         insert_inode_hash(inode);
 429         dentry->d_op = &ocfs2_dentry_ops;
 430         d_instantiate(dentry, inode);
 431         status = 0;
 432 leave:
 433         if (status < 0 && did_quota_inode)
 434                 vfs_dq_free_inode(inode);
 435         if (handle)
 436                 ocfs2_commit_trans(osb, handle);
 437
 438         ocfs2_inode_unlock(dir, 1);
 439
 440         if (status == -ENOSPC)
 441                 mlog(0, "Disk is full\n");
 442
 443         brelse(new_fe_bh);
 444         brelse(parent_fe_bh);
 445         kfree(si.name);
 446         kfree(si.value);
 447
 448         ocfs2_free_dir_lookup_result(&lookup);
 449
 450         if ((status < 0) && inode) {
 451                 clear_nlink(inode);
 452                 iput(inode);
 453         }
 454
 455         if (inode_ac)
 456                 ocfs2_free_alloc_context(inode_ac);
 457
 458         if (data_ac)
 459                 ocfs2_free_alloc_context(data_ac);
 460
 461         if (meta_ac)
 462                 ocfs2_free_alloc_context(meta_ac);
 463
 464         mlog_exit(status);
 465
 466         return status;
 467 }
 468
 469 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 470                               struct inode *dir,
 471                               struct inode *inode,
 472                               dev_t dev,
 473                               struct buffer_head **new_fe_bh,
 474                               struct buffer_head *parent_fe_bh,
 475                               handle_t *handle,
 476                               struct ocfs2_alloc_context *inode_ac)
 477 {
 478         int status = 0;
 479         struct ocfs2_dinode *fe = NULL;
 480         struct ocfs2_extent_list *fel;
 481         u64 fe_blkno = 0;
 482         u16 suballoc_bit;
 483         u16 feat;
 484
 485         *new_fe_bh = NULL;
 486
 487         status = ocfs2_claim_new_inode(osb, handle, dir, parent_fe_bh,
 488                                        inode_ac, &suballoc_bit, &fe_blkno);
 489         if (status < 0) {
 490                 mlog_errno(status);
 491                 goto leave;
 492         }
 493
 494         /* populate as many fields early on as possible - many of
 495          * these are used by the support functions here and in
 496          * callers. */
 497         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 498         OCFS2_I(inode)->ip_blkno = fe_blkno;
 499         spin_lock(&osb->osb_lock);
 500         inode->i_generation = osb->s_next_generation++;
 501         spin_unlock(&osb->osb_lock);
 502
 503         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 504         if (!*new_fe_bh) {
 505                 status = -EIO;
 506                 mlog_errno(status);
 507                 goto leave;
 508         }
 509         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 510
 511         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 512                                          *new_fe_bh,
 513                                          OCFS2_JOURNAL_ACCESS_CREATE);
 514         if (status < 0) {
 515                 mlog_errno(status);
 516                 goto leave;
 517         }
 518
 519         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 520         memset(fe, 0, osb->sb->s_blocksize);
 521
 522         fe->i_generation = cpu_to_le32(inode->i_generation);
 523         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 524         fe->i_blkno = cpu_to_le64(fe_blkno);
 525         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 526         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 527         fe->i_uid = cpu_to_le32(inode->i_uid);
 528         fe->i_gid = cpu_to_le32(inode->i_gid);
 529         fe->i_mode = cpu_to_le16(inode->i_mode);
 530         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 531                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 532
 533         ocfs2_set_links_count(fe, inode->i_nlink);
 534
 535         fe->i_last_eb_blk = 0;
 536         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 537         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 538         fe->i_atime = fe->i_ctime = fe->i_mtime =
 539                 cpu_to_le64(CURRENT_TIME.tv_sec);
 540         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 541                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 542         fe->i_dtime = 0;
 543
 544         /*
 545          * If supported, directories start with inline data. If inline
 546          * isn't supported, but indexing is, we start them as indexed.
 547          */
 548         feat = le16_to_cpu(fe->i_dyn_features);
 549         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 550                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 551
 552                 fe->id2.i_data.id_count = cpu_to_le16(
 553                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 554         } else {
 555                 fel = &fe->id2.i_list;
 556                 fel->l_tree_depth = 0;
 557                 fel->l_next_free_rec = 0;
 558                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 559         }
 560
 561         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 562         if (status < 0) {
 563                 mlog_errno(status);
 564                 goto leave;
 565         }
 566
 567         ocfs2_populate_inode(inode, fe, 1);
 568         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 569         if (!ocfs2_mount_local(osb)) {
 570                 status = ocfs2_create_new_inode_locks(inode);
 571                 if (status < 0)
 572                         mlog_errno(status);
 573         }
 574
 575         status = 0; /* error in ocfs2_create_new_inode_locks is not
 576                      * critical */
 577
 578 leave:
 579         if (status < 0) {
 580                 if (*new_fe_bh) {
 581                         brelse(*new_fe_bh);
 582                         *new_fe_bh = NULL;
 583                 }
 584         }
 585
 586         mlog_exit(status);
 587         return status;
 588 }
 589
 590 static int ocfs2_mkdir(struct inode *dir,
 591                        struct dentry *dentry,
 592                        int mode)
 593 {
 594         int ret;
 595
 596         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 597                    dentry->d_name.len, dentry->d_name.name);
 598         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 599         mlog_exit(ret);
 600
 601         return ret;
 602 }
 603
 604 static int ocfs2_create(struct inode *dir,
 605                         struct dentry *dentry,
 606                         int mode,
 607                         struct nameidata *nd)
 608 {
 609         int ret;
 610
 611         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 612                    dentry->d_name.len, dentry->d_name.name);
 613         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 614         mlog_exit(ret);
 615
 616         return ret;
 617 }
 618
 619 static int ocfs2_link(struct dentry *old_dentry,
 620                       struct inode *dir,
 621                       struct dentry *dentry)
 622 {
 623         handle_t *handle;
 624         struct inode *inode = old_dentry->d_inode;
 625         int err;
 626         struct buffer_head *fe_bh = NULL;
 627         struct buffer_head *parent_fe_bh = NULL;
 628         struct ocfs2_dinode *fe = NULL;
 629         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 630         struct ocfs2_dir_lookup_result lookup = { NULL, };
 631
 632         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 633                    old_dentry->d_name.len, old_dentry->d_name.name,
 634                    dentry->d_name.len, dentry->d_name.name);
 635
 636         if (S_ISDIR(inode->i_mode))
 637                 return -EPERM;
 638
 639         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 640         if (err < 0) {
 641                 if (err != -ENOENT)
 642                         mlog_errno(err);
 643                 return err;
 644         }
 645
 646         if (!dir->i_nlink) {
 647                 err = -ENOENT;
 648                 goto out;
 649         }
 650
 651         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 652                                         dentry->d_name.len);
 653         if (err)
 654                 goto out;
 655
 656         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 657                                            dentry->d_name.name,
 658                                            dentry->d_name.len, &lookup);
 659         if (err < 0) {
 660                 mlog_errno(err);
 661                 goto out;
 662         }
 663
 664         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 665         if (err < 0) {
 666                 if (err != -ENOENT)
 667                         mlog_errno(err);
 668                 goto out;
 669         }
 670
 671         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 672         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 673                 err = -EMLINK;
 674                 goto out_unlock_inode;
 675         }
 676
 677         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 678         if (IS_ERR(handle)) {
 679                 err = PTR_ERR(handle);
 680                 handle = NULL;
 681                 mlog_errno(err);
 682                 goto out_unlock_inode;
 683         }
 684
 685         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 686                                       OCFS2_JOURNAL_ACCESS_WRITE);
 687         if (err < 0) {
 688                 mlog_errno(err);
 689                 goto out_commit;
 690         }
 691
 692         inc_nlink(inode);
 693         inode->i_ctime = CURRENT_TIME;
 694         ocfs2_set_links_count(fe, inode->i_nlink);
 695         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 696         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 697
 698         err = ocfs2_journal_dirty(handle, fe_bh);
 699         if (err < 0) {
 700                 ocfs2_add_links_count(fe, -1);
 701                 drop_nlink(inode);
 702                 mlog_errno(err);
 703                 goto out_commit;
 704         }
 705
 706         err = ocfs2_add_entry(handle, dentry, inode,
 707                               OCFS2_I(inode)->ip_blkno,
 708                               parent_fe_bh, &lookup);
 709         if (err) {
 710                 ocfs2_add_links_count(fe, -1);
 711                 drop_nlink(inode);
 712                 mlog_errno(err);
 713                 goto out_commit;
 714         }
 715
 716         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 717         if (err) {
 718                 mlog_errno(err);
 719                 goto out_commit;
 720         }
 721
 722         atomic_inc(&inode->i_count);
 723         dentry->d_op = &ocfs2_dentry_ops;
 724         d_instantiate(dentry, inode);
 725
 726 out_commit:
 727         ocfs2_commit_trans(osb, handle);
 728 out_unlock_inode:
 729         ocfs2_inode_unlock(inode, 1);
 730
 731 out:
 732         ocfs2_inode_unlock(dir, 1);
 733
 734         brelse(fe_bh);
 735         brelse(parent_fe_bh);
 736
 737         ocfs2_free_dir_lookup_result(&lookup);
 738
 739         mlog_exit(err);
 740
 741         return err;
 742 }
 743
 744 /*
 745  * Takes and drops an exclusive lock on the given dentry. This will
 746  * force other nodes to drop it.
 747  */
 748 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 749 {
 750         int ret;
 751
 752         ret = ocfs2_dentry_lock(dentry, 1);
 753         if (ret)
 754                 mlog_errno(ret);
 755         else
 756                 ocfs2_dentry_unlock(dentry, 1);
 757
 758         return ret;
 759 }
 760
 761 static inline int inode_is_unlinkable(struct inode *inode)
 762 {
 763         if (S_ISDIR(inode->i_mode)) {
 764                 if (inode->i_nlink == 2)
 765                         return 1;
 766                 return 0;
 767         }
 768
 769         if (inode->i_nlink == 1)
 770                 return 1;
 771         return 0;
 772 }
 773
 774 static int ocfs2_unlink(struct inode *dir,
 775                         struct dentry *dentry)
 776 {
 777         int status;
 778         int child_locked = 0;
 779         struct inode *inode = dentry->d_inode;
 780         struct inode *orphan_dir = NULL;
 781         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 782         u64 blkno;
 783         struct ocfs2_dinode *fe = NULL;
 784         struct buffer_head *fe_bh = NULL;
 785         struct buffer_head *parent_node_bh = NULL;
 786         handle_t *handle = NULL;
 787         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 788         struct ocfs2_dir_lookup_result lookup = { NULL, };
 789         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 790
 791         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 792                    dentry->d_name.len, dentry->d_name.name);
 793
 794         BUG_ON(dentry->d_parent->d_inode != dir);
 795
 796         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 797
 798         if (inode == osb->root_inode) {
 799                 mlog(0, "Cannot delete the root directory\n");
 800                 return -EPERM;
 801         }
 802
 803         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 804                                          OI_LS_PARENT);
 805         if (status < 0) {
 806                 if (status != -ENOENT)
 807                         mlog_errno(status);
 808                 return status;
 809         }
 810
 811         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 812                                           dentry->d_name.len, &blkno, dir,
 813                                           &lookup);
 814         if (status < 0) {
 815                 if (status != -ENOENT)
 816                         mlog_errno(status);
 817                 goto leave;
 818         }
 819
 820         if (OCFS2_I(inode)->ip_blkno != blkno) {
 821                 status = -ENOENT;
 822
 823                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 824                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 825                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 826                 goto leave;
 827         }
 828
 829         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 830         if (status < 0) {
 831                 if (status != -ENOENT)
 832                         mlog_errno(status);
 833                 goto leave;
 834         }
 835         child_locked = 1;
 836
 837         if (S_ISDIR(inode->i_mode)) {
 838                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 839                         status = -ENOTEMPTY;
 840                         goto leave;
 841                 }
 842         }
 843
 844         status = ocfs2_remote_dentry_delete(dentry);
 845         if (status < 0) {
 846                 /* This remote delete should succeed under all normal
 847                  * circumstances. */
 848                 mlog_errno(status);
 849                 goto leave;
 850         }
 851
 852         if (inode_is_unlinkable(inode)) {
 853                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 854                                                   OCFS2_I(inode)->ip_blkno,
 855                                                   orphan_name, &orphan_insert);
 856                 if (status < 0) {
 857                         mlog_errno(status);
 858                         goto leave;
 859                 }
 860         }
 861
 862         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 863         if (IS_ERR(handle)) {
 864                 status = PTR_ERR(handle);
 865                 handle = NULL;
 866                 mlog_errno(status);
 867                 goto leave;
 868         }
 869
 870         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 871                                          OCFS2_JOURNAL_ACCESS_WRITE);
 872         if (status < 0) {
 873                 mlog_errno(status);
 874                 goto leave;
 875         }
 876
 877         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 878
 879         if (inode_is_unlinkable(inode)) {
 880                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 881                                           &orphan_insert, orphan_dir);
 882                 if (status < 0) {
 883                         mlog_errno(status);
 884                         goto leave;
 885                 }
 886         }
 887
 888         /* delete the name from the parent dir */
 889         status = ocfs2_delete_entry(handle, dir, &lookup);
 890         if (status < 0) {
 891                 mlog_errno(status);
 892                 goto leave;
 893         }
 894
 895         if (S_ISDIR(inode->i_mode))
 896                 drop_nlink(inode);
 897         drop_nlink(inode);
 898         ocfs2_set_links_count(fe, inode->i_nlink);
 899
 900         status = ocfs2_journal_dirty(handle, fe_bh);
 901         if (status < 0) {
 902                 mlog_errno(status);
 903                 goto leave;
 904         }
 905
 906         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 907         if (S_ISDIR(inode->i_mode))
 908                 drop_nlink(dir);
 909
 910         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 911         if (status < 0) {
 912                 mlog_errno(status);
 913                 if (S_ISDIR(inode->i_mode))
 914                         inc_nlink(dir);
 915         }
 916
 917 leave:
 918         if (handle)
 919                 ocfs2_commit_trans(osb, handle);
 920
 921         if (child_locked)
 922                 ocfs2_inode_unlock(inode, 1);
 923
 924         ocfs2_inode_unlock(dir, 1);
 925
 926         if (orphan_dir) {
 927                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 928                 ocfs2_inode_unlock(orphan_dir, 1);
 929                 mutex_unlock(&orphan_dir->i_mutex);
 930                 iput(orphan_dir);
 931         }
 932
 933         brelse(fe_bh);
 934         brelse(parent_node_bh);
 935
 936         ocfs2_free_dir_lookup_result(&orphan_insert);
 937         ocfs2_free_dir_lookup_result(&lookup);
 938
 939         mlog_exit(status);
 940
 941         return status;
 942 }
 943
 944 /*
 945  * The only place this should be used is rename!
 946  * if they have the same id, then the 1st one is the only one locked.
 947  */
 948 static int ocfs2_double_lock(struct ocfs2_super *osb,
 949                              struct buffer_head **bh1,
 950                              struct inode *inode1,
 951                              struct buffer_head **bh2,
 952                              struct inode *inode2)
 953 {
 954         int status;
 955         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 956         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 957         struct buffer_head **tmpbh;
 958         struct inode *tmpinode;
 959
 960         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 961                    (unsigned long long)oi1->ip_blkno,
 962                    (unsigned long long)oi2->ip_blkno);
 963
 964         if (*bh1)
 965                 *bh1 = NULL;
 966         if (*bh2)
 967                 *bh2 = NULL;
 968
 969         /* we always want to lock the one with the lower lockid first. */
 970         if (oi1->ip_blkno != oi2->ip_blkno) {
 971                 if (oi1->ip_blkno < oi2->ip_blkno) {
 972                         /* switch id1 and id2 around */
 973                         mlog(0, "switching them around...\n");
 974                         tmpbh = bh2;
 975                         bh2 = bh1;
 976                         bh1 = tmpbh;
 977
 978                         tmpinode = inode2;
 979                         inode2 = inode1;
 980                         inode1 = tmpinode;
 981                 }
 982                 /* lock id2 */
 983                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 984                                                  OI_LS_RENAME1);
 985                 if (status < 0) {
 986                         if (status != -ENOENT)
 987                                 mlog_errno(status);
 988                         goto bail;
 989                 }
 990         }
 991
 992         /* lock id1 */
 993         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
 994         if (status < 0) {
 995                 /*
 996                  * An error return must mean that no cluster locks
 997                  * were held on function exit.
 998                  */
 999                 if (oi1->ip_blkno != oi2->ip_blkno)
1000                         ocfs2_inode_unlock(inode2, 1);
1001
1002                 if (status != -ENOENT)
1003                         mlog_errno(status);
1004         }
1005
1006 bail:
1007         mlog_exit(status);
1008         return status;
1009 }
1010
1011 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1012 {
1013         ocfs2_inode_unlock(inode1, 1);
1014
1015         if (inode1 != inode2)
1016                 ocfs2_inode_unlock(inode2, 1);
1017 }
1018
1019 static int ocfs2_rename(struct inode *old_dir,
1020                         struct dentry *old_dentry,
1021                         struct inode *new_dir,
1022                         struct dentry *new_dentry)
1023 {
1024         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1025         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1026         struct inode *old_inode = old_dentry->d_inode;
1027         struct inode *new_inode = new_dentry->d_inode;
1028         struct inode *orphan_dir = NULL;
1029         struct ocfs2_dinode *newfe = NULL;
1030         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1031         struct buffer_head *newfe_bh = NULL;
1032         struct buffer_head *old_inode_bh = NULL;
1033         struct ocfs2_super *osb = NULL;
1034         u64 newfe_blkno, old_de_ino;
1035         handle_t *handle = NULL;
1036         struct buffer_head *old_dir_bh = NULL;
1037         struct buffer_head *new_dir_bh = NULL;
1038         nlink_t old_dir_nlink = old_dir->i_nlink;
1039         struct ocfs2_dinode *old_di;
1040         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1041         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1042         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1043         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1044         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1045
1046         /* At some point it might be nice to break this function up a
1047          * bit. */
1048
1049         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1050                    old_dir, old_dentry, new_dir, new_dentry,
1051                    old_dentry->d_name.len, old_dentry->d_name.name,
1052                    new_dentry->d_name.len, new_dentry->d_name.name);
1053
1054         osb = OCFS2_SB(old_dir->i_sb);
1055
1056         if (new_inode) {
1057                 if (!igrab(new_inode))
1058                         BUG();
1059         }
1060
1061         /* Assume a directory hierarchy thusly:
1062          * a/b/c
1063          * a/d
1064          * a,b,c, and d are all directories.
1065          *
1066          * from cwd of 'a' on both nodes:
1067          * node1: mv b/c d
1068          * node2: mv d   b/c
1069          *
1070          * And that's why, just like the VFS, we need a file system
1071          * rename lock. */
1072         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1073                 status = ocfs2_rename_lock(osb);
1074                 if (status < 0) {
1075                         mlog_errno(status);
1076                         goto bail;
1077                 }
1078                 rename_lock = 1;
1079         }
1080
1081         /* if old and new are the same, this'll just do one lock. */
1082         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1083                                    &new_dir_bh, new_dir);
1084         if (status < 0) {
1085                 mlog_errno(status);
1086                 goto bail;
1087         }
1088         parents_locked = 1;
1089
1090         /* make sure both dirs have bhs
1091          * get an extra ref on old_dir_bh if old==new */
1092         if (!new_dir_bh) {
1093                 if (old_dir_bh) {
1094                         new_dir_bh = old_dir_bh;
1095                         get_bh(new_dir_bh);
1096                 } else {
1097                         mlog(ML_ERROR, "no old_dir_bh!\n");
1098                         status = -EIO;
1099                         goto bail;
1100                 }
1101         }
1102
1103         /*
1104          * Aside from allowing a meta data update, the locking here
1105          * also ensures that the downconvert thread on other nodes
1106          * won't have to concurrently downconvert the inode and the
1107          * dentry locks.
1108          */
1109         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1110                                          OI_LS_PARENT);
1111         if (status < 0) {
1112                 if (status != -ENOENT)
1113                         mlog_errno(status);
1114                 goto bail;
1115         }
1116         old_child_locked = 1;
1117
1118         status = ocfs2_remote_dentry_delete(old_dentry);
1119         if (status < 0) {
1120                 mlog_errno(status);
1121                 goto bail;
1122         }
1123
1124         if (S_ISDIR(old_inode->i_mode)) {
1125                 u64 old_inode_parent;
1126
1127                 update_dot_dot = 1;
1128                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1129                                                   old_inode,
1130                                                   &old_inode_dot_dot_res);
1131                 if (status) {
1132                         status = -EIO;
1133                         goto bail;
1134                 }
1135
1136                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1137                         status = -EIO;
1138                         goto bail;
1139                 }
1140
1141                 if (!new_inode && new_dir != old_dir &&
1142                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1143                         status = -EMLINK;
1144                         goto bail;
1145                 }
1146         }
1147
1148         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1149                                             old_dentry->d_name.len,
1150                                             &old_de_ino);
1151         if (status) {
1152                 status = -ENOENT;
1153                 goto bail;
1154         }
1155
1156         /*
1157          *  Check for inode number is _not_ due to possible IO errors.
1158          *  We might rmdir the source, keep it as pwd of some process
1159          *  and merrily kill the link to whatever was created under the
1160          *  same name. Goodbye sticky bit ;-<
1161          */
1162         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1163                 status = -ENOENT;
1164                 goto bail;
1165         }
1166
1167         /* check if the target already exists (in which case we need
1168          * to delete it */
1169         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1170                                           new_dentry->d_name.len,
1171                                           &newfe_blkno, new_dir,
1172                                           &target_lookup_res);
1173         /* The only error we allow here is -ENOENT because the new
1174          * file not existing is perfectly valid. */
1175         if ((status < 0) && (status != -ENOENT)) {
1176                 /* If we cannot find the file specified we should just */
1177                 /* return the error... */
1178                 mlog_errno(status);
1179                 goto bail;
1180         }
1181         if (status == 0)
1182                 target_exists = 1;
1183
1184         if (!target_exists && new_inode) {
1185                 /*
1186                  * Target was unlinked by another node while we were
1187                  * waiting to get to ocfs2_rename(). There isn't
1188                  * anything we can do here to help the situation, so
1189                  * bubble up the appropriate error.
1190                  */
1191                 status = -ENOENT;
1192                 goto bail;
1193         }
1194
1195         /* In case we need to overwrite an existing file, we blow it
1196          * away first */
1197         if (target_exists) {
1198                 /* VFS didn't think there existed an inode here, but
1199                  * someone else in the cluster must have raced our
1200                  * rename to create one. Today we error cleanly, in
1201                  * the future we should consider calling iget to build
1202                  * a new struct inode for this entry. */
1203                 if (!new_inode) {
1204                         status = -EACCES;
1205
1206                         mlog(0, "We found an inode for name %.*s but VFS "
1207                              "didn't give us one.\n", new_dentry->d_name.len,
1208                              new_dentry->d_name.name);
1209                         goto bail;
1210                 }
1211
1212                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1213                         status = -EACCES;
1214
1215                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1216                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1217                              (unsigned long long)newfe_blkno,
1218                              OCFS2_I(new_inode)->ip_flags);
1219                         goto bail;
1220                 }
1221
1222                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1223                 if (status < 0) {
1224                         if (status != -ENOENT)
1225                                 mlog_errno(status);
1226                         goto bail;
1227                 }
1228                 new_child_locked = 1;
1229
1230                 status = ocfs2_remote_dentry_delete(new_dentry);
1231                 if (status < 0) {
1232                         mlog_errno(status);
1233                         goto bail;
1234                 }
1235
1236                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1237
1238                 mlog(0, "aha rename over existing... new_blkno=%llu "
1239                      "newfebh=%p bhblocknr=%llu\n",
1240                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1241                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1242
1243                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1244                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1245                                                 OCFS2_I(new_inode)->ip_blkno,
1246                                                 orphan_name, &orphan_insert);
1247                         if (status < 0) {
1248                                 mlog_errno(status);
1249                                 goto bail;
1250                         }
1251                 }
1252         } else {
1253                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1254
1255                 status = ocfs2_check_dir_for_entry(new_dir,
1256                                                    new_dentry->d_name.name,
1257                                                    new_dentry->d_name.len);
1258                 if (status)
1259                         goto bail;
1260
1261                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1262                                                       new_dentry->d_name.name,
1263                                                       new_dentry->d_name.len,
1264                                                       &target_insert);
1265                 if (status < 0) {
1266                         mlog_errno(status);
1267                         goto bail;
1268                 }
1269         }
1270
1271         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1272         if (IS_ERR(handle)) {
1273                 status = PTR_ERR(handle);
1274                 handle = NULL;
1275                 mlog_errno(status);
1276                 goto bail;
1277         }
1278
1279         if (target_exists) {
1280                 if (S_ISDIR(new_inode->i_mode)) {
1281                         if (new_inode->i_nlink != 2 ||
1282                             !ocfs2_empty_dir(new_inode)) {
1283                                 status = -ENOTEMPTY;
1284                                 goto bail;
1285                         }
1286                 }
1287                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1288                                                  newfe_bh,
1289                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1290                 if (status < 0) {
1291                         mlog_errno(status);
1292                         goto bail;
1293                 }
1294
1295                 if (S_ISDIR(new_inode->i_mode) ||
1296                     (ocfs2_read_links_count(newfe) == 1)) {
1297                         status = ocfs2_orphan_add(osb, handle, new_inode,
1298                                                   newfe, orphan_name,
1299                                                   &orphan_insert, orphan_dir);
1300                         if (status < 0) {
1301                                 mlog_errno(status);
1302                                 goto bail;
1303                         }
1304                 }
1305
1306                 /* change the dirent to point to the correct inode */
1307                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1308                                             old_inode);
1309                 if (status < 0) {
1310                         mlog_errno(status);
1311                         goto bail;
1312                 }
1313                 new_dir->i_version++;
1314
1315                 if (S_ISDIR(new_inode->i_mode))
1316                         ocfs2_set_links_count(newfe, 0);
1317                 else
1318                         ocfs2_add_links_count(newfe, -1);
1319
1320                 status = ocfs2_journal_dirty(handle, newfe_bh);
1321                 if (status < 0) {
1322                         mlog_errno(status);
1323                         goto bail;
1324                 }
1325         } else {
1326                 /* if the name was not found in new_dir, add it now */
1327                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1328                                          OCFS2_I(old_inode)->ip_blkno,
1329                                          new_dir_bh, &target_insert);
1330         }
1331
1332         old_inode->i_ctime = CURRENT_TIME;
1333         mark_inode_dirty(old_inode);
1334
1335         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1336                                          old_inode_bh,
1337                                          OCFS2_JOURNAL_ACCESS_WRITE);
1338         if (status >= 0) {
1339                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1340
1341                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1342                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1343
1344                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1345                 if (status < 0)
1346                         mlog_errno(status);
1347         } else
1348                 mlog_errno(status);
1349
1350         /*
1351          * Now that the name has been added to new_dir, remove the old name.
1352          *
1353          * We don't keep any directory entry context around until now
1354          * because the insert might have changed the type of directory
1355          * we're dealing with.
1356          */
1357         status = ocfs2_find_entry(old_dentry->d_name.name,
1358                                   old_dentry->d_name.len, old_dir,
1359                                   &old_entry_lookup);
1360         if (status)
1361                 goto bail;
1362
1363         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1364         if (status < 0) {
1365                 mlog_errno(status);
1366                 goto bail;
1367         }
1368
1369         if (new_inode) {
1370                 new_inode->i_nlink--;
1371                 new_inode->i_ctime = CURRENT_TIME;
1372         }
1373         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1374
1375         if (update_dot_dot) {
1376                 status = ocfs2_update_entry(old_inode, handle,
1377                                             &old_inode_dot_dot_res, new_dir);
1378                 old_dir->i_nlink--;
1379                 if (new_inode) {
1380                         new_inode->i_nlink--;
1381                 } else {
1382                         inc_nlink(new_dir);
1383                         mark_inode_dirty(new_dir);
1384                 }
1385         }
1386         mark_inode_dirty(old_dir);
1387         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1388         if (new_inode) {
1389                 mark_inode_dirty(new_inode);
1390                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1391         }
1392
1393         if (old_dir != new_dir) {
1394                 /* Keep the same times on both directories.*/
1395                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1396
1397                 /*
1398                  * This will also pick up the i_nlink change from the
1399                  * block above.
1400                  */
1401                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1402         }
1403
1404         if (old_dir_nlink != old_dir->i_nlink) {
1405                 if (!old_dir_bh) {
1406                         mlog(ML_ERROR, "need to change nlink for old dir "
1407                              "%llu from %d to %d but bh is NULL!\n",
1408                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1409                              (int)old_dir_nlink, old_dir->i_nlink);
1410                 } else {
1411                         struct ocfs2_dinode *fe;
1412                         status = ocfs2_journal_access_di(handle,
1413                                                          INODE_CACHE(old_dir),
1414                                                          old_dir_bh,
1415                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1416                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1417                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1418                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1419                 }
1420         }
1421         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1422         status = 0;
1423 bail:
1424         if (rename_lock)
1425                 ocfs2_rename_unlock(osb);
1426
1427         if (handle)
1428                 ocfs2_commit_trans(osb, handle);
1429
1430         if (parents_locked)
1431                 ocfs2_double_unlock(old_dir, new_dir);
1432
1433         if (old_child_locked)
1434                 ocfs2_inode_unlock(old_inode, 1);
1435
1436         if (new_child_locked)
1437                 ocfs2_inode_unlock(new_inode, 1);
1438
1439         if (orphan_dir) {
1440                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1441                 ocfs2_inode_unlock(orphan_dir, 1);
1442                 mutex_unlock(&orphan_dir->i_mutex);
1443                 iput(orphan_dir);
1444         }
1445
1446         if (new_inode)
1447                 sync_mapping_buffers(old_inode->i_mapping);
1448
1449         if (new_inode)
1450                 iput(new_inode);
1451
1452         ocfs2_free_dir_lookup_result(&target_lookup_res);
1453         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1454         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1455         ocfs2_free_dir_lookup_result(&orphan_insert);
1456         ocfs2_free_dir_lookup_result(&target_insert);
1457
1458         brelse(newfe_bh);
1459         brelse(old_inode_bh);
1460         brelse(old_dir_bh);
1461         brelse(new_dir_bh);
1462
1463         mlog_exit(status);
1464
1465         return status;
1466 }
1467
1468 /*
1469  * we expect i_size = strlen(symname). Copy symname into the file
1470  * data, including the null terminator.
1471  */
1472 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1473                                      handle_t *handle,
1474                                      struct inode *inode,
1475                                      const char *symname)
1476 {
1477         struct buffer_head **bhs = NULL;
1478         const char *c;
1479         struct super_block *sb = osb->sb;
1480         u64 p_blkno, p_blocks;
1481         int virtual, blocks, status, i, bytes_left;
1482
1483         bytes_left = i_size_read(inode) + 1;
1484         /* we can't trust i_blocks because we're actually going to
1485          * write i_size + 1 bytes. */
1486         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1487
1488         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1489                         (unsigned long long)inode->i_blocks,
1490                         i_size_read(inode), blocks);
1491
1492         /* Sanity check -- make sure we're going to fit. */
1493         if (bytes_left >
1494             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1495                 status = -EIO;
1496                 mlog_errno(status);
1497                 goto bail;
1498         }
1499
1500         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1501         if (!bhs) {
1502                 status = -ENOMEM;
1503                 mlog_errno(status);
1504                 goto bail;
1505         }
1506
1507         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1508                                              NULL);
1509         if (status < 0) {
1510                 mlog_errno(status);
1511                 goto bail;
1512         }
1513
1514         /* links can never be larger than one cluster so we know this
1515          * is all going to be contiguous, but do a sanity check
1516          * anyway. */
1517         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1518                 status = -EIO;
1519                 mlog_errno(status);
1520                 goto bail;
1521         }
1522
1523         virtual = 0;
1524         while(bytes_left > 0) {
1525                 c = &symname[virtual * sb->s_blocksize];
1526
1527                 bhs[virtual] = sb_getblk(sb, p_blkno);
1528                 if (!bhs[virtual]) {
1529                         status = -ENOMEM;
1530                         mlog_errno(status);
1531                         goto bail;
1532                 }
1533                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1534                                               bhs[virtual]);
1535
1536                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1537                                               bhs[virtual],
1538                                               OCFS2_JOURNAL_ACCESS_CREATE);
1539                 if (status < 0) {
1540                         mlog_errno(status);
1541                         goto bail;
1542                 }
1543
1544                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1545
1546                 memcpy(bhs[virtual]->b_data, c,
1547                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1548                        bytes_left);
1549
1550                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1551                 if (status < 0) {
1552                         mlog_errno(status);
1553                         goto bail;
1554                 }
1555
1556                 virtual++;
1557                 p_blkno++;
1558                 bytes_left -= sb->s_blocksize;
1559         }
1560
1561         status = 0;
1562 bail:
1563
1564         if (bhs) {
1565                 for(i = 0; i < blocks; i++)
1566                         brelse(bhs[i]);
1567                 kfree(bhs);
1568         }
1569
1570         mlog_exit(status);
1571         return status;
1572 }
1573
1574 static int ocfs2_symlink(struct inode *dir,
1575                          struct dentry *dentry,
1576                          const char *symname)
1577 {
1578         int status, l, credits;
1579         u64 newsize;
1580         struct ocfs2_super *osb = NULL;
1581         struct inode *inode = NULL;
1582         struct super_block *sb;
1583         struct buffer_head *new_fe_bh = NULL;
1584         struct buffer_head *parent_fe_bh = NULL;
1585         struct ocfs2_dinode *fe = NULL;
1586         struct ocfs2_dinode *dirfe;
1587         handle_t *handle = NULL;
1588         struct ocfs2_alloc_context *inode_ac = NULL;
1589         struct ocfs2_alloc_context *data_ac = NULL;
1590         struct ocfs2_alloc_context *xattr_ac = NULL;
1591         int want_clusters = 0;
1592         int xattr_credits = 0;
1593         struct ocfs2_security_xattr_info si = {
1594                 .enable = 1,
1595         };
1596         int did_quota = 0, did_quota_inode = 0;
1597         struct ocfs2_dir_lookup_result lookup = { NULL, };
1598
1599         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1600                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1601
1602         sb = dir->i_sb;
1603         osb = OCFS2_SB(sb);
1604
1605         l = strlen(symname) + 1;
1606
1607         credits = ocfs2_calc_symlink_credits(sb);
1608
1609         /* lock the parent directory */
1610         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1611         if (status < 0) {
1612                 if (status != -ENOENT)
1613                         mlog_errno(status);
1614                 return status;
1615         }
1616
1617         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1618         if (!ocfs2_read_links_count(dirfe)) {
1619                 /* can't make a file in a deleted directory. */
1620                 status = -ENOENT;
1621                 goto bail;
1622         }
1623
1624         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1625                                            dentry->d_name.len);
1626         if (status)
1627                 goto bail;
1628
1629         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1630                                               dentry->d_name.name,
1631                                               dentry->d_name.len, &lookup);
1632         if (status < 0) {
1633                 mlog_errno(status);
1634                 goto bail;
1635         }
1636
1637         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1638         if (status < 0) {
1639                 if (status != -ENOSPC)
1640                         mlog_errno(status);
1641                 goto bail;
1642         }
1643
1644         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1645         if (!inode) {
1646                 status = -ENOMEM;
1647                 mlog_errno(status);
1648                 goto bail;
1649         }
1650
1651         /* get security xattr */
1652         status = ocfs2_init_security_get(inode, dir, &si);
1653         if (status) {
1654                 if (status == -EOPNOTSUPP)
1655                         si.enable = 0;
1656                 else {
1657                         mlog_errno(status);
1658                         goto bail;
1659                 }
1660         }
1661
1662         /* calculate meta data/clusters for setting security xattr */
1663         if (si.enable) {
1664                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1665                                                   &xattr_credits, &xattr_ac);
1666                 if (status < 0) {
1667                         mlog_errno(status);
1668                         goto bail;
1669                 }
1670         }
1671
1672         /* don't reserve bitmap space for fast symlinks. */
1673         if (l > ocfs2_fast_symlink_chars(sb))
1674                 want_clusters += 1;
1675
1676         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1677         if (status < 0) {
1678                 if (status != -ENOSPC)
1679                         mlog_errno(status);
1680                 goto bail;
1681         }
1682
1683         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1684         if (IS_ERR(handle)) {
1685                 status = PTR_ERR(handle);
1686                 handle = NULL;
1687                 mlog_errno(status);
1688                 goto bail;
1689         }
1690
1691         /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1692          * to be called. */
1693         if (sb_any_quota_active(osb->sb) &&
1694             osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1695                 status = -EDQUOT;
1696                 goto bail;
1697         }
1698         did_quota_inode = 1;
1699
1700         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1701                    inode->i_mode, dentry->d_name.len,
1702                    dentry->d_name.name);
1703
1704         status = ocfs2_mknod_locked(osb, dir, inode,
1705                                     0, &new_fe_bh, parent_fe_bh, handle,
1706                                     inode_ac);
1707         if (status < 0) {
1708                 mlog_errno(status);
1709                 goto bail;
1710         }
1711
1712         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1713         inode->i_rdev = 0;
1714         newsize = l - 1;
1715         if (l > ocfs2_fast_symlink_chars(sb)) {
1716                 u32 offset = 0;
1717
1718                 inode->i_op = &ocfs2_symlink_inode_operations;
1719                 if (vfs_dq_alloc_space_nodirty(inode,
1720                     ocfs2_clusters_to_bytes(osb->sb, 1))) {
1721                         status = -EDQUOT;
1722                         goto bail;
1723                 }
1724                 did_quota = 1;
1725                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1726                                               new_fe_bh,
1727                                               handle, data_ac, NULL,
1728                                               NULL);
1729                 if (status < 0) {
1730                         if (status != -ENOSPC && status != -EINTR) {
1731                                 mlog(ML_ERROR,
1732                                      "Failed to extend file to %llu\n",
1733                                      (unsigned long long)newsize);
1734                                 mlog_errno(status);
1735                                 status = -ENOSPC;
1736                         }
1737                         goto bail;
1738                 }
1739                 i_size_write(inode, newsize);
1740                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1741         } else {
1742                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1743                 memcpy((char *) fe->id2.i_symlink, symname, l);
1744                 i_size_write(inode, newsize);
1745                 inode->i_blocks = 0;
1746         }
1747
1748         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1749         if (status < 0) {
1750                 mlog_errno(status);
1751                 goto bail;
1752         }
1753
1754         if (!ocfs2_inode_is_fast_symlink(inode)) {
1755                 status = ocfs2_create_symlink_data(osb, handle, inode,
1756                                                    symname);
1757                 if (status < 0) {
1758                         mlog_errno(status);
1759                         goto bail;
1760                 }
1761         }
1762
1763         if (si.enable) {
1764                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1765                                                  xattr_ac, data_ac);
1766                 if (status < 0) {
1767                         mlog_errno(status);
1768                         goto bail;
1769                 }
1770         }
1771
1772         status = ocfs2_add_entry(handle, dentry, inode,
1773                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1774                                  &lookup);
1775         if (status < 0) {
1776                 mlog_errno(status);
1777                 goto bail;
1778         }
1779
1780         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1781         if (status) {
1782                 mlog_errno(status);
1783                 goto bail;
1784         }
1785
1786         insert_inode_hash(inode);
1787         dentry->d_op = &ocfs2_dentry_ops;
1788         d_instantiate(dentry, inode);
1789 bail:
1790         if (status < 0 && did_quota)
1791                 vfs_dq_free_space_nodirty(inode,
1792                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1793         if (status < 0 && did_quota_inode)
1794                 vfs_dq_free_inode(inode);
1795         if (handle)
1796                 ocfs2_commit_trans(osb, handle);
1797
1798         ocfs2_inode_unlock(dir, 1);
1799
1800         brelse(new_fe_bh);
1801         brelse(parent_fe_bh);
1802         kfree(si.name);
1803         kfree(si.value);
1804         ocfs2_free_dir_lookup_result(&lookup);
1805         if (inode_ac)
1806                 ocfs2_free_alloc_context(inode_ac);
1807         if (data_ac)
1808                 ocfs2_free_alloc_context(data_ac);
1809         if (xattr_ac)
1810                 ocfs2_free_alloc_context(xattr_ac);
1811         if ((status < 0) && inode) {
1812                 clear_nlink(inode);
1813                 iput(inode);
1814         }
1815
1816         mlog_exit(status);
1817
1818         return status;
1819 }
1820
1821 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1822 {
1823         int status, namelen;
1824
1825         mlog_entry_void();
1826
1827         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1828                            (long long)blkno);
1829         if (namelen <= 0) {
1830                 if (namelen)
1831                         status = namelen;
1832                 else
1833                         status = -EINVAL;
1834                 mlog_errno(status);
1835                 goto bail;
1836         }
1837         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1838                 status = -EINVAL;
1839                 mlog_errno(status);
1840                 goto bail;
1841         }
1842
1843         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1844              namelen);
1845
1846         status = 0;
1847 bail:
1848         mlog_exit(status);
1849         return status;
1850 }
1851
1852 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1853                                     struct inode **ret_orphan_dir,
1854                                     u64 blkno,
1855                                     char *name,
1856                                     struct ocfs2_dir_lookup_result *lookup)
1857 {
1858         struct inode *orphan_dir_inode;
1859         struct buffer_head *orphan_dir_bh = NULL;
1860         int status = 0;
1861
1862         status = ocfs2_blkno_stringify(blkno, name);
1863         if (status < 0) {
1864                 mlog_errno(status);
1865                 return status;
1866         }
1867
1868         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1869                                                        ORPHAN_DIR_SYSTEM_INODE,
1870                                                        osb->slot_num);
1871         if (!orphan_dir_inode) {
1872                 status = -ENOENT;
1873                 mlog_errno(status);
1874                 return status;
1875         }
1876
1877         mutex_lock(&orphan_dir_inode->i_mutex);
1878
1879         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1880         if (status < 0) {
1881                 mlog_errno(status);
1882                 goto leave;
1883         }
1884
1885         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1886                                               orphan_dir_bh, name,
1887                                               OCFS2_ORPHAN_NAMELEN, lookup);
1888         if (status < 0) {
1889                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1890
1891                 mlog_errno(status);
1892                 goto leave;
1893         }
1894
1895         *ret_orphan_dir = orphan_dir_inode;
1896
1897 leave:
1898         if (status) {
1899                 mutex_unlock(&orphan_dir_inode->i_mutex);
1900                 iput(orphan_dir_inode);
1901         }
1902
1903         brelse(orphan_dir_bh);
1904
1905         mlog_exit(status);
1906         return status;
1907 }
1908
1909 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1910                             handle_t *handle,
1911                             struct inode *inode,
1912                             struct ocfs2_dinode *fe,
1913                             char *name,
1914                             struct ocfs2_dir_lookup_result *lookup,
1915                             struct inode *orphan_dir_inode)
1916 {
1917         struct buffer_head *orphan_dir_bh = NULL;
1918         int status = 0;
1919         struct ocfs2_dinode *orphan_fe;
1920
1921         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1922
1923         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1924         if (status < 0) {
1925                 mlog_errno(status);
1926                 goto leave;
1927         }
1928
1929         status = ocfs2_journal_access_di(handle,
1930                                          INODE_CACHE(orphan_dir_inode),
1931                                          orphan_dir_bh,
1932                                          OCFS2_JOURNAL_ACCESS_WRITE);
1933         if (status < 0) {
1934                 mlog_errno(status);
1935                 goto leave;
1936         }
1937
1938         /* we're a cluster, and nlink can change on disk from
1939          * underneath us... */
1940         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1941         if (S_ISDIR(inode->i_mode))
1942                 ocfs2_add_links_count(orphan_fe, 1);
1943         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1944
1945         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1946         if (status < 0) {
1947                 mlog_errno(status);
1948                 goto leave;
1949         }
1950
1951         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1952                                    OCFS2_ORPHAN_NAMELEN, inode,
1953                                    OCFS2_I(inode)->ip_blkno,
1954                                    orphan_dir_bh, lookup);
1955         if (status < 0) {
1956                 mlog_errno(status);
1957                 goto leave;
1958         }
1959
1960         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1961
1962         /* Record which orphan dir our inode now resides
1963          * in. delete_inode will use this to determine which orphan
1964          * dir to lock. */
1965         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1966
1967         mlog(0, "Inode %llu orphaned in slot %d\n",
1968              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1969
1970 leave:
1971         brelse(orphan_dir_bh);
1972
1973         mlog_exit(status);
1974         return status;
1975 }
1976
1977 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1978 int ocfs2_orphan_del(struct ocfs2_super *osb,
1979                      handle_t *handle,
1980                      struct inode *orphan_dir_inode,
1981                      struct inode *inode,
1982                      struct buffer_head *orphan_dir_bh)
1983 {
1984         char name[OCFS2_ORPHAN_NAMELEN + 1];
1985         struct ocfs2_dinode *orphan_fe;
1986         int status = 0;
1987         struct ocfs2_dir_lookup_result lookup = { NULL, };
1988
1989         mlog_entry_void();
1990
1991         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1992         if (status < 0) {
1993                 mlog_errno(status);
1994                 goto leave;
1995         }
1996
1997         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1998              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1999              OCFS2_ORPHAN_NAMELEN);
2000
2001         /* find it's spot in the orphan directory */
2002         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2003                                   &lookup);
2004         if (status) {
2005                 mlog_errno(status);
2006                 goto leave;
2007         }
2008
2009         /* remove it from the orphan directory */
2010         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2011         if (status < 0) {
2012                 mlog_errno(status);
2013                 goto leave;
2014         }
2015
2016         status = ocfs2_journal_access_di(handle,
2017                                          INODE_CACHE(orphan_dir_inode),
2018                                          orphan_dir_bh,
2019                                          OCFS2_JOURNAL_ACCESS_WRITE);
2020         if (status < 0) {
2021                 mlog_errno(status);
2022                 goto leave;
2023         }
2024
2025         /* do the i_nlink dance! :) */
2026         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2027         if (S_ISDIR(inode->i_mode))
2028                 ocfs2_add_links_count(orphan_fe, -1);
2029         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2030
2031         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2032         if (status < 0) {
2033                 mlog_errno(status);
2034                 goto leave;
2035         }
2036
2037 leave:
2038         ocfs2_free_dir_lookup_result(&lookup);
2039
2040         mlog_exit(status);
2041         return status;
2042 }
2043
2044 int ocfs2_create_inode_in_orphan(struct inode *dir,
2045                                  int mode,
2046                                  struct inode **new_inode)
2047 {
2048         int status, did_quota_inode = 0;
2049         struct inode *inode = NULL;
2050         struct inode *orphan_dir = NULL;
2051         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2052         struct ocfs2_dinode *di = NULL;
2053         handle_t *handle = NULL;
2054         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2055         struct buffer_head *parent_di_bh = NULL;
2056         struct buffer_head *new_di_bh = NULL;
2057         struct ocfs2_alloc_context *inode_ac = NULL;
2058         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2059
2060         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2061         if (status < 0) {
2062                 if (status != -ENOENT)
2063                         mlog_errno(status);
2064                 return status;
2065         }
2066
2067         /*
2068          * We give the orphan dir the root blkno to fake an orphan name,
2069          * and allocate enough space for our insertion.
2070          */
2071         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
2072                                           osb->root_blkno,
2073                                           orphan_name, &orphan_insert);
2074         if (status < 0) {
2075                 mlog_errno(status);
2076                 goto leave;
2077         }
2078
2079         /* reserve an inode spot */
2080         status = ocfs2_reserve_new_inode(osb, &inode_ac);
2081         if (status < 0) {
2082                 if (status != -ENOSPC)
2083                         mlog_errno(status);
2084                 goto leave;
2085         }
2086
2087         inode = ocfs2_get_init_inode(dir, mode);
2088         if (!inode) {
2089                 status = -ENOMEM;
2090                 mlog_errno(status);
2091                 goto leave;
2092         }
2093
2094         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2095         if (IS_ERR(handle)) {
2096                 status = PTR_ERR(handle);
2097                 handle = NULL;
2098                 mlog_errno(status);
2099                 goto leave;
2100         }
2101
2102         /* We don't use standard VFS wrapper because we don't want vfs_dq_init
2103          * to be called. */
2104         if (sb_any_quota_active(osb->sb) &&
2105             osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
2106                 status = -EDQUOT;
2107                 goto leave;
2108         }
2109         did_quota_inode = 1;
2110
2111         /* do the real work now. */
2112         status = ocfs2_mknod_locked(osb, dir, inode,
2113                                     0, &new_di_bh, parent_di_bh, handle,
2114                                     inode_ac);
2115         if (status < 0) {
2116                 mlog_errno(status);
2117                 goto leave;
2118         }
2119
2120         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
2121         if (status < 0) {
2122                 mlog_errno(status);
2123                 goto leave;
2124         }
2125
2126         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2127         status = ocfs2_orphan_add(osb, handle, inode, di, orphan_name,
2128                                   &orphan_insert, orphan_dir);
2129         if (status < 0) {
2130                 mlog_errno(status);
2131                 goto leave;
2132         }
2133
2134         /* get open lock so that only nodes can't remove it from orphan dir. */
2135         status = ocfs2_open_lock(inode);
2136         if (status < 0)
2137                 mlog_errno(status);
2138
2139 leave:
2140         if (status < 0 && did_quota_inode)
2141                 vfs_dq_free_inode(inode);
2142         if (handle)
2143                 ocfs2_commit_trans(osb, handle);
2144
2145         if (orphan_dir) {
2146                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2147                 ocfs2_inode_unlock(orphan_dir, 1);
2148                 mutex_unlock(&orphan_dir->i_mutex);
2149                 iput(orphan_dir);
2150         }
2151
2152         if (status == -ENOSPC)
2153                 mlog(0, "Disk is full\n");
2154
2155         if ((status < 0) && inode) {
2156                 clear_nlink(inode);
2157                 iput(inode);
2158         }
2159
2160         if (inode_ac)
2161                 ocfs2_free_alloc_context(inode_ac);
2162
2163         brelse(new_di_bh);
2164
2165         if (!status)
2166                 *new_inode = inode;
2167
2168         ocfs2_free_dir_lookup_result(&orphan_insert);
2169
2170         ocfs2_inode_unlock(dir, 1);
2171         brelse(parent_di_bh);
2172         return status;
2173 }
2174
2175 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2176                                    struct inode *inode,
2177                                    struct dentry *dentry)
2178 {
2179         int status = 0;
2180         struct buffer_head *parent_di_bh = NULL;
2181         handle_t *handle = NULL;
2182         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2183         struct ocfs2_dinode *dir_di, *di;
2184         struct inode *orphan_dir_inode = NULL;
2185         struct buffer_head *orphan_dir_bh = NULL;
2186         struct buffer_head *di_bh = NULL;
2187         struct ocfs2_dir_lookup_result lookup = { NULL, };
2188
2189         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2190                    dentry->d_name.len, dentry->d_name.name);
2191
2192         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2193         if (status < 0) {
2194                 if (status != -ENOENT)
2195                         mlog_errno(status);
2196                 return status;
2197         }
2198
2199         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2200         if (!dir_di->i_links_count) {
2201                 /* can't make a file in a deleted directory. */
2202                 status = -ENOENT;
2203                 goto leave;
2204         }
2205
2206         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2207                                            dentry->d_name.len);
2208         if (status)
2209                 goto leave;
2210
2211         /* get a spot inside the dir. */
2212         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2213                                               dentry->d_name.name,
2214                                               dentry->d_name.len, &lookup);
2215         if (status < 0) {
2216                 mlog_errno(status);
2217                 goto leave;
2218         }
2219
2220         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2221                                                        ORPHAN_DIR_SYSTEM_INODE,
2222                                                        osb->slot_num);
2223         if (!orphan_dir_inode) {
2224                 status = -EEXIST;
2225                 mlog_errno(status);
2226                 goto leave;
2227         }
2228
2229         mutex_lock(&orphan_dir_inode->i_mutex);
2230
2231         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2232         if (status < 0) {
2233                 mlog_errno(status);
2234                 mutex_unlock(&orphan_dir_inode->i_mutex);
2235                 iput(orphan_dir_inode);
2236                 goto leave;
2237         }
2238
2239         status = ocfs2_read_inode_block(inode, &di_bh);
2240         if (status < 0) {
2241                 mlog_errno(status);
2242                 goto orphan_unlock;
2243         }
2244
2245         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2246         if (IS_ERR(handle)) {
2247                 status = PTR_ERR(handle);
2248                 handle = NULL;
2249                 mlog_errno(status);
2250                 goto orphan_unlock;
2251         }
2252
2253         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2254                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2255         if (status < 0) {
2256                 mlog_errno(status);
2257                 goto out_commit;
2258         }
2259
2260         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2261                                   orphan_dir_bh);
2262         if (status < 0) {
2263                 mlog_errno(status);
2264                 goto out_commit;
2265         }
2266
2267         di = (struct ocfs2_dinode *)di_bh->b_data;
2268         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2269         di->i_orphaned_slot = 0;
2270         ocfs2_journal_dirty(handle, di_bh);
2271
2272         status = ocfs2_add_entry(handle, dentry, inode,
2273                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2274                                  &lookup);
2275         if (status < 0) {
2276                 mlog_errno(status);
2277                 goto out_commit;
2278         }
2279
2280         status = ocfs2_dentry_attach_lock(dentry, inode,
2281                                           OCFS2_I(dir)->ip_blkno);
2282         if (status) {
2283                 mlog_errno(status);
2284                 goto out_commit;
2285         }
2286
2287         insert_inode_hash(inode);
2288         dentry->d_op = &ocfs2_dentry_ops;
2289         d_instantiate(dentry, inode);
2290         status = 0;
2291 out_commit:
2292         ocfs2_commit_trans(osb, handle);
2293 orphan_unlock:
2294         ocfs2_inode_unlock(orphan_dir_inode, 1);
2295         mutex_unlock(&orphan_dir_inode->i_mutex);
2296         iput(orphan_dir_inode);
2297 leave:
2298
2299         ocfs2_inode_unlock(dir, 1);
2300
2301         brelse(di_bh);
2302         brelse(parent_di_bh);
2303         brelse(orphan_dir_bh);
2304
2305         ocfs2_free_dir_lookup_result(&lookup);
2306
2307         mlog_exit(status);
2308
2309         return status;
2310 }
2311
2312 const struct inode_operations ocfs2_dir_iops = {
2313         .create         = ocfs2_create,
2314         .lookup         = ocfs2_lookup,
2315         .link           = ocfs2_link,
2316         .unlink         = ocfs2_unlink,
2317         .rmdir          = ocfs2_unlink,
2318         .symlink        = ocfs2_symlink,
2319         .mkdir          = ocfs2_mkdir,
2320         .mknod          = ocfs2_mknod,
2321         .rename         = ocfs2_rename,
2322         .setattr        = ocfs2_setattr,
2323         .getattr        = ocfs2_getattr,
2324         .permission     = ocfs2_permission,
2325         .setxattr       = generic_setxattr,
2326         .getxattr       = generic_getxattr,
2327         .listxattr      = ocfs2_listxattr,
2328         .removexattr    = generic_removexattr,
2329 };