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

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #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         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                 ocfs2_journal_dirty(handle, parent_fe_bh);
 388                 inc_nlink(dir);
 389         }
 390
 391         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 392                                 meta_ac, data_ac);
 393         if (status < 0) {
 394                 mlog_errno(status);
 395                 goto leave;
 396         }
 397
 398         if (si.enable) {
 399                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 400                                                  meta_ac, data_ac);
 401                 if (status < 0) {
 402                         mlog_errno(status);
 403                         goto leave;
 404                 }
 405         }
 406
 407         status = ocfs2_add_entry(handle, dentry, inode,
 408                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 409                                  &lookup);
 410         if (status < 0) {
 411                 mlog_errno(status);
 412                 goto leave;
 413         }
 414
 415         status = ocfs2_dentry_attach_lock(dentry, inode,
 416                                           OCFS2_I(dir)->ip_blkno);
 417         if (status) {
 418                 mlog_errno(status);
 419                 goto leave;
 420         }
 421
 422         insert_inode_hash(inode);
 423         dentry->d_op = &ocfs2_dentry_ops;
 424         d_instantiate(dentry, inode);
 425         status = 0;
 426 leave:
 427         if (status < 0 && did_quota_inode)
 428                 dquot_free_inode(inode);
 429         if (handle)
 430                 ocfs2_commit_trans(osb, handle);
 431
 432         ocfs2_inode_unlock(dir, 1);
 433
 434         if (status == -ENOSPC)
 435                 mlog(0, "Disk is full\n");
 436
 437         brelse(new_fe_bh);
 438         brelse(parent_fe_bh);
 439         kfree(si.name);
 440         kfree(si.value);
 441
 442         ocfs2_free_dir_lookup_result(&lookup);
 443
 444         if ((status < 0) && inode) {
 445                 clear_nlink(inode);
 446                 iput(inode);
 447         }
 448
 449         if (inode_ac)
 450                 ocfs2_free_alloc_context(inode_ac);
 451
 452         if (data_ac)
 453                 ocfs2_free_alloc_context(data_ac);
 454
 455         if (meta_ac)
 456                 ocfs2_free_alloc_context(meta_ac);
 457
 458         mlog_exit(status);
 459
 460         return status;
 461 }
 462
 463 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 464                               struct inode *dir,
 465                               struct inode *inode,
 466                               dev_t dev,
 467                               struct buffer_head **new_fe_bh,
 468                               struct buffer_head *parent_fe_bh,
 469                               handle_t *handle,
 470                               struct ocfs2_alloc_context *inode_ac)
 471 {
 472         int status = 0;
 473         struct ocfs2_dinode *fe = NULL;
 474         struct ocfs2_extent_list *fel;
 475         u64 fe_blkno = 0;
 476         u16 suballoc_bit;
 477         u16 feat;
 478
 479         *new_fe_bh = NULL;
 480
 481         status = ocfs2_claim_new_inode(osb, handle, dir, parent_fe_bh,
 482                                        inode_ac, &suballoc_bit, &fe_blkno);
 483         if (status < 0) {
 484                 mlog_errno(status);
 485                 goto leave;
 486         }
 487
 488         /* populate as many fields early on as possible - many of
 489          * these are used by the support functions here and in
 490          * callers. */
 491         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 492         OCFS2_I(inode)->ip_blkno = fe_blkno;
 493         spin_lock(&osb->osb_lock);
 494         inode->i_generation = osb->s_next_generation++;
 495         spin_unlock(&osb->osb_lock);
 496
 497         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 498         if (!*new_fe_bh) {
 499                 status = -EIO;
 500                 mlog_errno(status);
 501                 goto leave;
 502         }
 503         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 504
 505         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 506                                          *new_fe_bh,
 507                                          OCFS2_JOURNAL_ACCESS_CREATE);
 508         if (status < 0) {
 509                 mlog_errno(status);
 510                 goto leave;
 511         }
 512
 513         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 514         memset(fe, 0, osb->sb->s_blocksize);
 515
 516         fe->i_generation = cpu_to_le32(inode->i_generation);
 517         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 518         fe->i_blkno = cpu_to_le64(fe_blkno);
 519         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 520         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 521         fe->i_uid = cpu_to_le32(inode->i_uid);
 522         fe->i_gid = cpu_to_le32(inode->i_gid);
 523         fe->i_mode = cpu_to_le16(inode->i_mode);
 524         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 525                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 526
 527         ocfs2_set_links_count(fe, inode->i_nlink);
 528
 529         fe->i_last_eb_blk = 0;
 530         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 531         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 532         fe->i_atime = fe->i_ctime = fe->i_mtime =
 533                 cpu_to_le64(CURRENT_TIME.tv_sec);
 534         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 535                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 536         fe->i_dtime = 0;
 537
 538         /*
 539          * If supported, directories start with inline data. If inline
 540          * isn't supported, but indexing is, we start them as indexed.
 541          */
 542         feat = le16_to_cpu(fe->i_dyn_features);
 543         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 544                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 545
 546                 fe->id2.i_data.id_count = cpu_to_le16(
 547                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 548         } else {
 549                 fel = &fe->id2.i_list;
 550                 fel->l_tree_depth = 0;
 551                 fel->l_next_free_rec = 0;
 552                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 553         }
 554
 555         ocfs2_journal_dirty(handle, *new_fe_bh);
 556
 557         ocfs2_populate_inode(inode, fe, 1);
 558         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 559         if (!ocfs2_mount_local(osb)) {
 560                 status = ocfs2_create_new_inode_locks(inode);
 561                 if (status < 0)
 562                         mlog_errno(status);
 563         }
 564
 565         status = 0; /* error in ocfs2_create_new_inode_locks is not
 566                      * critical */
 567
 568 leave:
 569         if (status < 0) {
 570                 if (*new_fe_bh) {
 571                         brelse(*new_fe_bh);
 572                         *new_fe_bh = NULL;
 573                 }
 574         }
 575
 576         mlog_exit(status);
 577         return status;
 578 }
 579
 580 static int ocfs2_mkdir(struct inode *dir,
 581                        struct dentry *dentry,
 582                        int mode)
 583 {
 584         int ret;
 585
 586         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 587                    dentry->d_name.len, dentry->d_name.name);
 588         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 589         mlog_exit(ret);
 590
 591         return ret;
 592 }
 593
 594 static int ocfs2_create(struct inode *dir,
 595                         struct dentry *dentry,
 596                         int mode,
 597                         struct nameidata *nd)
 598 {
 599         int ret;
 600
 601         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 602                    dentry->d_name.len, dentry->d_name.name);
 603         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 604         mlog_exit(ret);
 605
 606         return ret;
 607 }
 608
 609 static int ocfs2_link(struct dentry *old_dentry,
 610                       struct inode *dir,
 611                       struct dentry *dentry)
 612 {
 613         handle_t *handle;
 614         struct inode *inode = old_dentry->d_inode;
 615         int err;
 616         struct buffer_head *fe_bh = NULL;
 617         struct buffer_head *parent_fe_bh = NULL;
 618         struct ocfs2_dinode *fe = NULL;
 619         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 620         struct ocfs2_dir_lookup_result lookup = { NULL, };
 621
 622         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 623                    old_dentry->d_name.len, old_dentry->d_name.name,
 624                    dentry->d_name.len, dentry->d_name.name);
 625
 626         if (S_ISDIR(inode->i_mode))
 627                 return -EPERM;
 628
 629         dquot_initialize(dir);
 630
 631         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 632         if (err < 0) {
 633                 if (err != -ENOENT)
 634                         mlog_errno(err);
 635                 return err;
 636         }
 637
 638         if (!dir->i_nlink) {
 639                 err = -ENOENT;
 640                 goto out;
 641         }
 642
 643         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 644                                         dentry->d_name.len);
 645         if (err)
 646                 goto out;
 647
 648         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 649                                            dentry->d_name.name,
 650                                            dentry->d_name.len, &lookup);
 651         if (err < 0) {
 652                 mlog_errno(err);
 653                 goto out;
 654         }
 655
 656         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 657         if (err < 0) {
 658                 if (err != -ENOENT)
 659                         mlog_errno(err);
 660                 goto out;
 661         }
 662
 663         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 664         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 665                 err = -EMLINK;
 666                 goto out_unlock_inode;
 667         }
 668
 669         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 670         if (IS_ERR(handle)) {
 671                 err = PTR_ERR(handle);
 672                 handle = NULL;
 673                 mlog_errno(err);
 674                 goto out_unlock_inode;
 675         }
 676
 677         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 678                                       OCFS2_JOURNAL_ACCESS_WRITE);
 679         if (err < 0) {
 680                 mlog_errno(err);
 681                 goto out_commit;
 682         }
 683
 684         inc_nlink(inode);
 685         inode->i_ctime = CURRENT_TIME;
 686         ocfs2_set_links_count(fe, inode->i_nlink);
 687         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 688         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 689         ocfs2_journal_dirty(handle, fe_bh);
 690
 691         err = ocfs2_add_entry(handle, dentry, inode,
 692                               OCFS2_I(inode)->ip_blkno,
 693                               parent_fe_bh, &lookup);
 694         if (err) {
 695                 ocfs2_add_links_count(fe, -1);
 696                 drop_nlink(inode);
 697                 mlog_errno(err);
 698                 goto out_commit;
 699         }
 700
 701         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 702         if (err) {
 703                 mlog_errno(err);
 704                 goto out_commit;
 705         }
 706
 707         atomic_inc(&inode->i_count);
 708         dentry->d_op = &ocfs2_dentry_ops;
 709         d_instantiate(dentry, inode);
 710
 711 out_commit:
 712         ocfs2_commit_trans(osb, handle);
 713 out_unlock_inode:
 714         ocfs2_inode_unlock(inode, 1);
 715
 716 out:
 717         ocfs2_inode_unlock(dir, 1);
 718
 719         brelse(fe_bh);
 720         brelse(parent_fe_bh);
 721
 722         ocfs2_free_dir_lookup_result(&lookup);
 723
 724         mlog_exit(err);
 725
 726         return err;
 727 }
 728
 729 /*
 730  * Takes and drops an exclusive lock on the given dentry. This will
 731  * force other nodes to drop it.
 732  */
 733 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 734 {
 735         int ret;
 736
 737         ret = ocfs2_dentry_lock(dentry, 1);
 738         if (ret)
 739                 mlog_errno(ret);
 740         else
 741                 ocfs2_dentry_unlock(dentry, 1);
 742
 743         return ret;
 744 }
 745
 746 static inline int inode_is_unlinkable(struct inode *inode)
 747 {
 748         if (S_ISDIR(inode->i_mode)) {
 749                 if (inode->i_nlink == 2)
 750                         return 1;
 751                 return 0;
 752         }
 753
 754         if (inode->i_nlink == 1)
 755                 return 1;
 756         return 0;
 757 }
 758
 759 static int ocfs2_unlink(struct inode *dir,
 760                         struct dentry *dentry)
 761 {
 762         int status;
 763         int child_locked = 0;
 764         struct inode *inode = dentry->d_inode;
 765         struct inode *orphan_dir = NULL;
 766         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 767         u64 blkno;
 768         struct ocfs2_dinode *fe = NULL;
 769         struct buffer_head *fe_bh = NULL;
 770         struct buffer_head *parent_node_bh = NULL;
 771         handle_t *handle = NULL;
 772         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 773         struct ocfs2_dir_lookup_result lookup = { NULL, };
 774         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 775
 776         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 777                    dentry->d_name.len, dentry->d_name.name);
 778
 779         dquot_initialize(dir);
 780
 781         BUG_ON(dentry->d_parent->d_inode != dir);
 782
 783         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 784
 785         if (inode == osb->root_inode) {
 786                 mlog(0, "Cannot delete the root directory\n");
 787                 return -EPERM;
 788         }
 789
 790         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 791                                          OI_LS_PARENT);
 792         if (status < 0) {
 793                 if (status != -ENOENT)
 794                         mlog_errno(status);
 795                 return status;
 796         }
 797
 798         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 799                                           dentry->d_name.len, &blkno, dir,
 800                                           &lookup);
 801         if (status < 0) {
 802                 if (status != -ENOENT)
 803                         mlog_errno(status);
 804                 goto leave;
 805         }
 806
 807         if (OCFS2_I(inode)->ip_blkno != blkno) {
 808                 status = -ENOENT;
 809
 810                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 811                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 812                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 813                 goto leave;
 814         }
 815
 816         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 817         if (status < 0) {
 818                 if (status != -ENOENT)
 819                         mlog_errno(status);
 820                 goto leave;
 821         }
 822         child_locked = 1;
 823
 824         if (S_ISDIR(inode->i_mode)) {
 825                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 826                         status = -ENOTEMPTY;
 827                         goto leave;
 828                 }
 829         }
 830
 831         status = ocfs2_remote_dentry_delete(dentry);
 832         if (status < 0) {
 833                 /* This remote delete should succeed under all normal
 834                  * circumstances. */
 835                 mlog_errno(status);
 836                 goto leave;
 837         }
 838
 839         if (inode_is_unlinkable(inode)) {
 840                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 841                                                   OCFS2_I(inode)->ip_blkno,
 842                                                   orphan_name, &orphan_insert);
 843                 if (status < 0) {
 844                         mlog_errno(status);
 845                         goto leave;
 846                 }
 847         }
 848
 849         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 850         if (IS_ERR(handle)) {
 851                 status = PTR_ERR(handle);
 852                 handle = NULL;
 853                 mlog_errno(status);
 854                 goto leave;
 855         }
 856
 857         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 858                                          OCFS2_JOURNAL_ACCESS_WRITE);
 859         if (status < 0) {
 860                 mlog_errno(status);
 861                 goto leave;
 862         }
 863
 864         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 865
 866         if (inode_is_unlinkable(inode)) {
 867                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 868                                           &orphan_insert, orphan_dir);
 869                 if (status < 0) {
 870                         mlog_errno(status);
 871                         goto leave;
 872                 }
 873         }
 874
 875         /* delete the name from the parent dir */
 876         status = ocfs2_delete_entry(handle, dir, &lookup);
 877         if (status < 0) {
 878                 mlog_errno(status);
 879                 goto leave;
 880         }
 881
 882         if (S_ISDIR(inode->i_mode))
 883                 drop_nlink(inode);
 884         drop_nlink(inode);
 885         ocfs2_set_links_count(fe, inode->i_nlink);
 886         ocfs2_journal_dirty(handle, fe_bh);
 887
 888         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 889         if (S_ISDIR(inode->i_mode))
 890                 drop_nlink(dir);
 891
 892         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 893         if (status < 0) {
 894                 mlog_errno(status);
 895                 if (S_ISDIR(inode->i_mode))
 896                         inc_nlink(dir);
 897         }
 898
 899 leave:
 900         if (handle)
 901                 ocfs2_commit_trans(osb, handle);
 902
 903         if (child_locked)
 904                 ocfs2_inode_unlock(inode, 1);
 905
 906         ocfs2_inode_unlock(dir, 1);
 907
 908         if (orphan_dir) {
 909                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 910                 ocfs2_inode_unlock(orphan_dir, 1);
 911                 mutex_unlock(&orphan_dir->i_mutex);
 912                 iput(orphan_dir);
 913         }
 914
 915         brelse(fe_bh);
 916         brelse(parent_node_bh);
 917
 918         ocfs2_free_dir_lookup_result(&orphan_insert);
 919         ocfs2_free_dir_lookup_result(&lookup);
 920
 921         mlog_exit(status);
 922
 923         return status;
 924 }
 925
 926 /*
 927  * The only place this should be used is rename!
 928  * if they have the same id, then the 1st one is the only one locked.
 929  */
 930 static int ocfs2_double_lock(struct ocfs2_super *osb,
 931                              struct buffer_head **bh1,
 932                              struct inode *inode1,
 933                              struct buffer_head **bh2,
 934                              struct inode *inode2)
 935 {
 936         int status;
 937         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 938         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 939         struct buffer_head **tmpbh;
 940         struct inode *tmpinode;
 941
 942         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 943                    (unsigned long long)oi1->ip_blkno,
 944                    (unsigned long long)oi2->ip_blkno);
 945
 946         if (*bh1)
 947                 *bh1 = NULL;
 948         if (*bh2)
 949                 *bh2 = NULL;
 950
 951         /* we always want to lock the one with the lower lockid first. */
 952         if (oi1->ip_blkno != oi2->ip_blkno) {
 953                 if (oi1->ip_blkno < oi2->ip_blkno) {
 954                         /* switch id1 and id2 around */
 955                         mlog(0, "switching them around...\n");
 956                         tmpbh = bh2;
 957                         bh2 = bh1;
 958                         bh1 = tmpbh;
 959
 960                         tmpinode = inode2;
 961                         inode2 = inode1;
 962                         inode1 = tmpinode;
 963                 }
 964                 /* lock id2 */
 965                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 966                                                  OI_LS_RENAME1);
 967                 if (status < 0) {
 968                         if (status != -ENOENT)
 969                                 mlog_errno(status);
 970                         goto bail;
 971                 }
 972         }
 973
 974         /* lock id1 */
 975         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
 976         if (status < 0) {
 977                 /*
 978                  * An error return must mean that no cluster locks
 979                  * were held on function exit.
 980                  */
 981                 if (oi1->ip_blkno != oi2->ip_blkno)
 982                         ocfs2_inode_unlock(inode2, 1);
 983
 984                 if (status != -ENOENT)
 985                         mlog_errno(status);
 986         }
 987
 988 bail:
 989         mlog_exit(status);
 990         return status;
 991 }
 992
 993 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
 994 {
 995         ocfs2_inode_unlock(inode1, 1);
 996
 997         if (inode1 != inode2)
 998                 ocfs2_inode_unlock(inode2, 1);
 999 }
1000
1001 static int ocfs2_rename(struct inode *old_dir,
1002                         struct dentry *old_dentry,
1003                         struct inode *new_dir,
1004                         struct dentry *new_dentry)
1005 {
1006         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1007         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1008         struct inode *old_inode = old_dentry->d_inode;
1009         struct inode *new_inode = new_dentry->d_inode;
1010         struct inode *orphan_dir = NULL;
1011         struct ocfs2_dinode *newfe = NULL;
1012         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1013         struct buffer_head *newfe_bh = NULL;
1014         struct buffer_head *old_inode_bh = NULL;
1015         struct ocfs2_super *osb = NULL;
1016         u64 newfe_blkno, old_de_ino;
1017         handle_t *handle = NULL;
1018         struct buffer_head *old_dir_bh = NULL;
1019         struct buffer_head *new_dir_bh = NULL;
1020         nlink_t old_dir_nlink = old_dir->i_nlink;
1021         struct ocfs2_dinode *old_di;
1022         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1023         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1024         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1025         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1026         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1027
1028         /* At some point it might be nice to break this function up a
1029          * bit. */
1030
1031         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1032                    old_dir, old_dentry, new_dir, new_dentry,
1033                    old_dentry->d_name.len, old_dentry->d_name.name,
1034                    new_dentry->d_name.len, new_dentry->d_name.name);
1035
1036         dquot_initialize(old_dir);
1037         dquot_initialize(new_dir);
1038
1039         osb = OCFS2_SB(old_dir->i_sb);
1040
1041         if (new_inode) {
1042                 if (!igrab(new_inode))
1043                         BUG();
1044         }
1045
1046         /* Assume a directory hierarchy thusly:
1047          * a/b/c
1048          * a/d
1049          * a,b,c, and d are all directories.
1050          *
1051          * from cwd of 'a' on both nodes:
1052          * node1: mv b/c d
1053          * node2: mv d   b/c
1054          *
1055          * And that's why, just like the VFS, we need a file system
1056          * rename lock. */
1057         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1058                 status = ocfs2_rename_lock(osb);
1059                 if (status < 0) {
1060                         mlog_errno(status);
1061                         goto bail;
1062                 }
1063                 rename_lock = 1;
1064         }
1065
1066         /* if old and new are the same, this'll just do one lock. */
1067         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1068                                    &new_dir_bh, new_dir);
1069         if (status < 0) {
1070                 mlog_errno(status);
1071                 goto bail;
1072         }
1073         parents_locked = 1;
1074
1075         /* make sure both dirs have bhs
1076          * get an extra ref on old_dir_bh if old==new */
1077         if (!new_dir_bh) {
1078                 if (old_dir_bh) {
1079                         new_dir_bh = old_dir_bh;
1080                         get_bh(new_dir_bh);
1081                 } else {
1082                         mlog(ML_ERROR, "no old_dir_bh!\n");
1083                         status = -EIO;
1084                         goto bail;
1085                 }
1086         }
1087
1088         /*
1089          * Aside from allowing a meta data update, the locking here
1090          * also ensures that the downconvert thread on other nodes
1091          * won't have to concurrently downconvert the inode and the
1092          * dentry locks.
1093          */
1094         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1095                                          OI_LS_PARENT);
1096         if (status < 0) {
1097                 if (status != -ENOENT)
1098                         mlog_errno(status);
1099                 goto bail;
1100         }
1101         old_child_locked = 1;
1102
1103         status = ocfs2_remote_dentry_delete(old_dentry);
1104         if (status < 0) {
1105                 mlog_errno(status);
1106                 goto bail;
1107         }
1108
1109         if (S_ISDIR(old_inode->i_mode)) {
1110                 u64 old_inode_parent;
1111
1112                 update_dot_dot = 1;
1113                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1114                                                   old_inode,
1115                                                   &old_inode_dot_dot_res);
1116                 if (status) {
1117                         status = -EIO;
1118                         goto bail;
1119                 }
1120
1121                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1122                         status = -EIO;
1123                         goto bail;
1124                 }
1125
1126                 if (!new_inode && new_dir != old_dir &&
1127                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1128                         status = -EMLINK;
1129                         goto bail;
1130                 }
1131         }
1132
1133         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1134                                             old_dentry->d_name.len,
1135                                             &old_de_ino);
1136         if (status) {
1137                 status = -ENOENT;
1138                 goto bail;
1139         }
1140
1141         /*
1142          *  Check for inode number is _not_ due to possible IO errors.
1143          *  We might rmdir the source, keep it as pwd of some process
1144          *  and merrily kill the link to whatever was created under the
1145          *  same name. Goodbye sticky bit ;-<
1146          */
1147         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1148                 status = -ENOENT;
1149                 goto bail;
1150         }
1151
1152         /* check if the target already exists (in which case we need
1153          * to delete it */
1154         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1155                                           new_dentry->d_name.len,
1156                                           &newfe_blkno, new_dir,
1157                                           &target_lookup_res);
1158         /* The only error we allow here is -ENOENT because the new
1159          * file not existing is perfectly valid. */
1160         if ((status < 0) && (status != -ENOENT)) {
1161                 /* If we cannot find the file specified we should just */
1162                 /* return the error... */
1163                 mlog_errno(status);
1164                 goto bail;
1165         }
1166         if (status == 0)
1167                 target_exists = 1;
1168
1169         if (!target_exists && new_inode) {
1170                 /*
1171                  * Target was unlinked by another node while we were
1172                  * waiting to get to ocfs2_rename(). There isn't
1173                  * anything we can do here to help the situation, so
1174                  * bubble up the appropriate error.
1175                  */
1176                 status = -ENOENT;
1177                 goto bail;
1178         }
1179
1180         /* In case we need to overwrite an existing file, we blow it
1181          * away first */
1182         if (target_exists) {
1183                 /* VFS didn't think there existed an inode here, but
1184                  * someone else in the cluster must have raced our
1185                  * rename to create one. Today we error cleanly, in
1186                  * the future we should consider calling iget to build
1187                  * a new struct inode for this entry. */
1188                 if (!new_inode) {
1189                         status = -EACCES;
1190
1191                         mlog(0, "We found an inode for name %.*s but VFS "
1192                              "didn't give us one.\n", new_dentry->d_name.len,
1193                              new_dentry->d_name.name);
1194                         goto bail;
1195                 }
1196
1197                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1198                         status = -EACCES;
1199
1200                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1201                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1202                              (unsigned long long)newfe_blkno,
1203                              OCFS2_I(new_inode)->ip_flags);
1204                         goto bail;
1205                 }
1206
1207                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1208                 if (status < 0) {
1209                         if (status != -ENOENT)
1210                                 mlog_errno(status);
1211                         goto bail;
1212                 }
1213                 new_child_locked = 1;
1214
1215                 status = ocfs2_remote_dentry_delete(new_dentry);
1216                 if (status < 0) {
1217                         mlog_errno(status);
1218                         goto bail;
1219                 }
1220
1221                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1222
1223                 mlog(0, "aha rename over existing... new_blkno=%llu "
1224                      "newfebh=%p bhblocknr=%llu\n",
1225                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1226                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1227
1228                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1229                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1230                                                 OCFS2_I(new_inode)->ip_blkno,
1231                                                 orphan_name, &orphan_insert);
1232                         if (status < 0) {
1233                                 mlog_errno(status);
1234                                 goto bail;
1235                         }
1236                 }
1237         } else {
1238                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1239
1240                 status = ocfs2_check_dir_for_entry(new_dir,
1241                                                    new_dentry->d_name.name,
1242                                                    new_dentry->d_name.len);
1243                 if (status)
1244                         goto bail;
1245
1246                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1247                                                       new_dentry->d_name.name,
1248                                                       new_dentry->d_name.len,
1249                                                       &target_insert);
1250                 if (status < 0) {
1251                         mlog_errno(status);
1252                         goto bail;
1253                 }
1254         }
1255
1256         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1257         if (IS_ERR(handle)) {
1258                 status = PTR_ERR(handle);
1259                 handle = NULL;
1260                 mlog_errno(status);
1261                 goto bail;
1262         }
1263
1264         if (target_exists) {
1265                 if (S_ISDIR(new_inode->i_mode)) {
1266                         if (new_inode->i_nlink != 2 ||
1267                             !ocfs2_empty_dir(new_inode)) {
1268                                 status = -ENOTEMPTY;
1269                                 goto bail;
1270                         }
1271                 }
1272                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1273                                                  newfe_bh,
1274                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1275                 if (status < 0) {
1276                         mlog_errno(status);
1277                         goto bail;
1278                 }
1279
1280                 if (S_ISDIR(new_inode->i_mode) ||
1281                     (ocfs2_read_links_count(newfe) == 1)) {
1282                         status = ocfs2_orphan_add(osb, handle, new_inode,
1283                                                   newfe, orphan_name,
1284                                                   &orphan_insert, orphan_dir);
1285                         if (status < 0) {
1286                                 mlog_errno(status);
1287                                 goto bail;
1288                         }
1289                 }
1290
1291                 /* change the dirent to point to the correct inode */
1292                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1293                                             old_inode);
1294                 if (status < 0) {
1295                         mlog_errno(status);
1296                         goto bail;
1297                 }
1298                 new_dir->i_version++;
1299
1300                 if (S_ISDIR(new_inode->i_mode))
1301                         ocfs2_set_links_count(newfe, 0);
1302                 else
1303                         ocfs2_add_links_count(newfe, -1);
1304                 ocfs2_journal_dirty(handle, newfe_bh);
1305         } else {
1306                 /* if the name was not found in new_dir, add it now */
1307                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1308                                          OCFS2_I(old_inode)->ip_blkno,
1309                                          new_dir_bh, &target_insert);
1310         }
1311
1312         old_inode->i_ctime = CURRENT_TIME;
1313         mark_inode_dirty(old_inode);
1314
1315         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1316                                          old_inode_bh,
1317                                          OCFS2_JOURNAL_ACCESS_WRITE);
1318         if (status >= 0) {
1319                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1320
1321                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1322                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1323                 ocfs2_journal_dirty(handle, old_inode_bh);
1324         } else
1325                 mlog_errno(status);
1326
1327         /*
1328          * Now that the name has been added to new_dir, remove the old name.
1329          *
1330          * We don't keep any directory entry context around until now
1331          * because the insert might have changed the type of directory
1332          * we're dealing with.
1333          */
1334         status = ocfs2_find_entry(old_dentry->d_name.name,
1335                                   old_dentry->d_name.len, old_dir,
1336                                   &old_entry_lookup);
1337         if (status)
1338                 goto bail;
1339
1340         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1341         if (status < 0) {
1342                 mlog_errno(status);
1343                 goto bail;
1344         }
1345
1346         if (new_inode) {
1347                 new_inode->i_nlink--;
1348                 new_inode->i_ctime = CURRENT_TIME;
1349         }
1350         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1351
1352         if (update_dot_dot) {
1353                 status = ocfs2_update_entry(old_inode, handle,
1354                                             &old_inode_dot_dot_res, new_dir);
1355                 old_dir->i_nlink--;
1356                 if (new_inode) {
1357                         new_inode->i_nlink--;
1358                 } else {
1359                         inc_nlink(new_dir);
1360                         mark_inode_dirty(new_dir);
1361                 }
1362         }
1363         mark_inode_dirty(old_dir);
1364         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1365         if (new_inode) {
1366                 mark_inode_dirty(new_inode);
1367                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1368         }
1369
1370         if (old_dir != new_dir) {
1371                 /* Keep the same times on both directories.*/
1372                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1373
1374                 /*
1375                  * This will also pick up the i_nlink change from the
1376                  * block above.
1377                  */
1378                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1379         }
1380
1381         if (old_dir_nlink != old_dir->i_nlink) {
1382                 if (!old_dir_bh) {
1383                         mlog(ML_ERROR, "need to change nlink for old dir "
1384                              "%llu from %d to %d but bh is NULL!\n",
1385                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1386                              (int)old_dir_nlink, old_dir->i_nlink);
1387                 } else {
1388                         struct ocfs2_dinode *fe;
1389                         status = ocfs2_journal_access_di(handle,
1390                                                          INODE_CACHE(old_dir),
1391                                                          old_dir_bh,
1392                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1393                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1394                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1395                         ocfs2_journal_dirty(handle, old_dir_bh);
1396                 }
1397         }
1398         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1399         status = 0;
1400 bail:
1401         if (rename_lock)
1402                 ocfs2_rename_unlock(osb);
1403
1404         if (handle)
1405                 ocfs2_commit_trans(osb, handle);
1406
1407         if (parents_locked)
1408                 ocfs2_double_unlock(old_dir, new_dir);
1409
1410         if (old_child_locked)
1411                 ocfs2_inode_unlock(old_inode, 1);
1412
1413         if (new_child_locked)
1414                 ocfs2_inode_unlock(new_inode, 1);
1415
1416         if (orphan_dir) {
1417                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1418                 ocfs2_inode_unlock(orphan_dir, 1);
1419                 mutex_unlock(&orphan_dir->i_mutex);
1420                 iput(orphan_dir);
1421         }
1422
1423         if (new_inode)
1424                 sync_mapping_buffers(old_inode->i_mapping);
1425
1426         if (new_inode)
1427                 iput(new_inode);
1428
1429         ocfs2_free_dir_lookup_result(&target_lookup_res);
1430         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1431         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1432         ocfs2_free_dir_lookup_result(&orphan_insert);
1433         ocfs2_free_dir_lookup_result(&target_insert);
1434
1435         brelse(newfe_bh);
1436         brelse(old_inode_bh);
1437         brelse(old_dir_bh);
1438         brelse(new_dir_bh);
1439
1440         mlog_exit(status);
1441
1442         return status;
1443 }
1444
1445 /*
1446  * we expect i_size = strlen(symname). Copy symname into the file
1447  * data, including the null terminator.
1448  */
1449 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1450                                      handle_t *handle,
1451                                      struct inode *inode,
1452                                      const char *symname)
1453 {
1454         struct buffer_head **bhs = NULL;
1455         const char *c;
1456         struct super_block *sb = osb->sb;
1457         u64 p_blkno, p_blocks;
1458         int virtual, blocks, status, i, bytes_left;
1459
1460         bytes_left = i_size_read(inode) + 1;
1461         /* we can't trust i_blocks because we're actually going to
1462          * write i_size + 1 bytes. */
1463         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1464
1465         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1466                         (unsigned long long)inode->i_blocks,
1467                         i_size_read(inode), blocks);
1468
1469         /* Sanity check -- make sure we're going to fit. */
1470         if (bytes_left >
1471             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1472                 status = -EIO;
1473                 mlog_errno(status);
1474                 goto bail;
1475         }
1476
1477         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1478         if (!bhs) {
1479                 status = -ENOMEM;
1480                 mlog_errno(status);
1481                 goto bail;
1482         }
1483
1484         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1485                                              NULL);
1486         if (status < 0) {
1487                 mlog_errno(status);
1488                 goto bail;
1489         }
1490
1491         /* links can never be larger than one cluster so we know this
1492          * is all going to be contiguous, but do a sanity check
1493          * anyway. */
1494         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1495                 status = -EIO;
1496                 mlog_errno(status);
1497                 goto bail;
1498         }
1499
1500         virtual = 0;
1501         while(bytes_left > 0) {
1502                 c = &symname[virtual * sb->s_blocksize];
1503
1504                 bhs[virtual] = sb_getblk(sb, p_blkno);
1505                 if (!bhs[virtual]) {
1506                         status = -ENOMEM;
1507                         mlog_errno(status);
1508                         goto bail;
1509                 }
1510                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1511                                               bhs[virtual]);
1512
1513                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1514                                               bhs[virtual],
1515                                               OCFS2_JOURNAL_ACCESS_CREATE);
1516                 if (status < 0) {
1517                         mlog_errno(status);
1518                         goto bail;
1519                 }
1520
1521                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1522
1523                 memcpy(bhs[virtual]->b_data, c,
1524                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1525                        bytes_left);
1526
1527                 ocfs2_journal_dirty(handle, bhs[virtual]);
1528
1529                 virtual++;
1530                 p_blkno++;
1531                 bytes_left -= sb->s_blocksize;
1532         }
1533
1534         status = 0;
1535 bail:
1536
1537         if (bhs) {
1538                 for(i = 0; i < blocks; i++)
1539                         brelse(bhs[i]);
1540                 kfree(bhs);
1541         }
1542
1543         mlog_exit(status);
1544         return status;
1545 }
1546
1547 static int ocfs2_symlink(struct inode *dir,
1548                          struct dentry *dentry,
1549                          const char *symname)
1550 {
1551         int status, l, credits;
1552         u64 newsize;
1553         struct ocfs2_super *osb = NULL;
1554         struct inode *inode = NULL;
1555         struct super_block *sb;
1556         struct buffer_head *new_fe_bh = NULL;
1557         struct buffer_head *parent_fe_bh = NULL;
1558         struct ocfs2_dinode *fe = NULL;
1559         struct ocfs2_dinode *dirfe;
1560         handle_t *handle = NULL;
1561         struct ocfs2_alloc_context *inode_ac = NULL;
1562         struct ocfs2_alloc_context *data_ac = NULL;
1563         struct ocfs2_alloc_context *xattr_ac = NULL;
1564         int want_clusters = 0;
1565         int xattr_credits = 0;
1566         struct ocfs2_security_xattr_info si = {
1567                 .enable = 1,
1568         };
1569         int did_quota = 0, did_quota_inode = 0;
1570         struct ocfs2_dir_lookup_result lookup = { NULL, };
1571
1572         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1573                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1574
1575         dquot_initialize(dir);
1576
1577         sb = dir->i_sb;
1578         osb = OCFS2_SB(sb);
1579
1580         l = strlen(symname) + 1;
1581
1582         credits = ocfs2_calc_symlink_credits(sb);
1583
1584         /* lock the parent directory */
1585         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1586         if (status < 0) {
1587                 if (status != -ENOENT)
1588                         mlog_errno(status);
1589                 return status;
1590         }
1591
1592         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1593         if (!ocfs2_read_links_count(dirfe)) {
1594                 /* can't make a file in a deleted directory. */
1595                 status = -ENOENT;
1596                 goto bail;
1597         }
1598
1599         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1600                                            dentry->d_name.len);
1601         if (status)
1602                 goto bail;
1603
1604         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1605                                               dentry->d_name.name,
1606                                               dentry->d_name.len, &lookup);
1607         if (status < 0) {
1608                 mlog_errno(status);
1609                 goto bail;
1610         }
1611
1612         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1613         if (status < 0) {
1614                 if (status != -ENOSPC)
1615                         mlog_errno(status);
1616                 goto bail;
1617         }
1618
1619         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1620         if (!inode) {
1621                 status = -ENOMEM;
1622                 mlog_errno(status);
1623                 goto bail;
1624         }
1625
1626         /* get security xattr */
1627         status = ocfs2_init_security_get(inode, dir, &si);
1628         if (status) {
1629                 if (status == -EOPNOTSUPP)
1630                         si.enable = 0;
1631                 else {
1632                         mlog_errno(status);
1633                         goto bail;
1634                 }
1635         }
1636
1637         /* calculate meta data/clusters for setting security xattr */
1638         if (si.enable) {
1639                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1640                                                   &xattr_credits, &xattr_ac);
1641                 if (status < 0) {
1642                         mlog_errno(status);
1643                         goto bail;
1644                 }
1645         }
1646
1647         /* don't reserve bitmap space for fast symlinks. */
1648         if (l > ocfs2_fast_symlink_chars(sb))
1649                 want_clusters += 1;
1650
1651         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1652         if (status < 0) {
1653                 if (status != -ENOSPC)
1654                         mlog_errno(status);
1655                 goto bail;
1656         }
1657
1658         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1659         if (IS_ERR(handle)) {
1660                 status = PTR_ERR(handle);
1661                 handle = NULL;
1662                 mlog_errno(status);
1663                 goto bail;
1664         }
1665
1666         status = dquot_alloc_inode(inode);
1667         if (status)
1668                 goto bail;
1669         did_quota_inode = 1;
1670
1671         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1672                    inode->i_mode, dentry->d_name.len,
1673                    dentry->d_name.name);
1674
1675         status = ocfs2_mknod_locked(osb, dir, inode,
1676                                     0, &new_fe_bh, parent_fe_bh, handle,
1677                                     inode_ac);
1678         if (status < 0) {
1679                 mlog_errno(status);
1680                 goto bail;
1681         }
1682
1683         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1684         inode->i_rdev = 0;
1685         newsize = l - 1;
1686         if (l > ocfs2_fast_symlink_chars(sb)) {
1687                 u32 offset = 0;
1688
1689                 inode->i_op = &ocfs2_symlink_inode_operations;
1690                 status = dquot_alloc_space_nodirty(inode,
1691                     ocfs2_clusters_to_bytes(osb->sb, 1));
1692                 if (status)
1693                         goto bail;
1694                 did_quota = 1;
1695                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1696                                               new_fe_bh,
1697                                               handle, data_ac, NULL,
1698                                               NULL);
1699                 if (status < 0) {
1700                         if (status != -ENOSPC && status != -EINTR) {
1701                                 mlog(ML_ERROR,
1702                                      "Failed to extend file to %llu\n",
1703                                      (unsigned long long)newsize);
1704                                 mlog_errno(status);
1705                                 status = -ENOSPC;
1706                         }
1707                         goto bail;
1708                 }
1709                 i_size_write(inode, newsize);
1710                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1711         } else {
1712                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1713                 memcpy((char *) fe->id2.i_symlink, symname, l);
1714                 i_size_write(inode, newsize);
1715                 inode->i_blocks = 0;
1716         }
1717
1718         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1719         if (status < 0) {
1720                 mlog_errno(status);
1721                 goto bail;
1722         }
1723
1724         if (!ocfs2_inode_is_fast_symlink(inode)) {
1725                 status = ocfs2_create_symlink_data(osb, handle, inode,
1726                                                    symname);
1727                 if (status < 0) {
1728                         mlog_errno(status);
1729                         goto bail;
1730                 }
1731         }
1732
1733         if (si.enable) {
1734                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1735                                                  xattr_ac, data_ac);
1736                 if (status < 0) {
1737                         mlog_errno(status);
1738                         goto bail;
1739                 }
1740         }
1741
1742         status = ocfs2_add_entry(handle, dentry, inode,
1743                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1744                                  &lookup);
1745         if (status < 0) {
1746                 mlog_errno(status);
1747                 goto bail;
1748         }
1749
1750         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1751         if (status) {
1752                 mlog_errno(status);
1753                 goto bail;
1754         }
1755
1756         insert_inode_hash(inode);
1757         dentry->d_op = &ocfs2_dentry_ops;
1758         d_instantiate(dentry, inode);
1759 bail:
1760         if (status < 0 && did_quota)
1761                 dquot_free_space_nodirty(inode,
1762                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1763         if (status < 0 && did_quota_inode)
1764                 dquot_free_inode(inode);
1765         if (handle)
1766                 ocfs2_commit_trans(osb, handle);
1767
1768         ocfs2_inode_unlock(dir, 1);
1769
1770         brelse(new_fe_bh);
1771         brelse(parent_fe_bh);
1772         kfree(si.name);
1773         kfree(si.value);
1774         ocfs2_free_dir_lookup_result(&lookup);
1775         if (inode_ac)
1776                 ocfs2_free_alloc_context(inode_ac);
1777         if (data_ac)
1778                 ocfs2_free_alloc_context(data_ac);
1779         if (xattr_ac)
1780                 ocfs2_free_alloc_context(xattr_ac);
1781         if ((status < 0) && inode) {
1782                 clear_nlink(inode);
1783                 iput(inode);
1784         }
1785
1786         mlog_exit(status);
1787
1788         return status;
1789 }
1790
1791 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1792 {
1793         int status, namelen;
1794
1795         mlog_entry_void();
1796
1797         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1798                            (long long)blkno);
1799         if (namelen <= 0) {
1800                 if (namelen)
1801                         status = namelen;
1802                 else
1803                         status = -EINVAL;
1804                 mlog_errno(status);
1805                 goto bail;
1806         }
1807         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1808                 status = -EINVAL;
1809                 mlog_errno(status);
1810                 goto bail;
1811         }
1812
1813         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1814              namelen);
1815
1816         status = 0;
1817 bail:
1818         mlog_exit(status);
1819         return status;
1820 }
1821
1822 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1823                                     struct inode **ret_orphan_dir,
1824                                     u64 blkno,
1825                                     char *name,
1826                                     struct ocfs2_dir_lookup_result *lookup)
1827 {
1828         struct inode *orphan_dir_inode;
1829         struct buffer_head *orphan_dir_bh = NULL;
1830         int status = 0;
1831
1832         status = ocfs2_blkno_stringify(blkno, name);
1833         if (status < 0) {
1834                 mlog_errno(status);
1835                 return status;
1836         }
1837
1838         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1839                                                        ORPHAN_DIR_SYSTEM_INODE,
1840                                                        osb->slot_num);
1841         if (!orphan_dir_inode) {
1842                 status = -ENOENT;
1843                 mlog_errno(status);
1844                 return status;
1845         }
1846
1847         mutex_lock(&orphan_dir_inode->i_mutex);
1848
1849         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1850         if (status < 0) {
1851                 mlog_errno(status);
1852                 goto leave;
1853         }
1854
1855         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1856                                               orphan_dir_bh, name,
1857                                               OCFS2_ORPHAN_NAMELEN, lookup);
1858         if (status < 0) {
1859                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1860
1861                 mlog_errno(status);
1862                 goto leave;
1863         }
1864
1865         *ret_orphan_dir = orphan_dir_inode;
1866
1867 leave:
1868         if (status) {
1869                 mutex_unlock(&orphan_dir_inode->i_mutex);
1870                 iput(orphan_dir_inode);
1871         }
1872
1873         brelse(orphan_dir_bh);
1874
1875         mlog_exit(status);
1876         return status;
1877 }
1878
1879 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1880                             handle_t *handle,
1881                             struct inode *inode,
1882                             struct ocfs2_dinode *fe,
1883                             char *name,
1884                             struct ocfs2_dir_lookup_result *lookup,
1885                             struct inode *orphan_dir_inode)
1886 {
1887         struct buffer_head *orphan_dir_bh = NULL;
1888         int status = 0;
1889         struct ocfs2_dinode *orphan_fe;
1890
1891         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1892
1893         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1894         if (status < 0) {
1895                 mlog_errno(status);
1896                 goto leave;
1897         }
1898
1899         status = ocfs2_journal_access_di(handle,
1900                                          INODE_CACHE(orphan_dir_inode),
1901                                          orphan_dir_bh,
1902                                          OCFS2_JOURNAL_ACCESS_WRITE);
1903         if (status < 0) {
1904                 mlog_errno(status);
1905                 goto leave;
1906         }
1907
1908         /* we're a cluster, and nlink can change on disk from
1909          * underneath us... */
1910         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1911         if (S_ISDIR(inode->i_mode))
1912                 ocfs2_add_links_count(orphan_fe, 1);
1913         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1914         ocfs2_journal_dirty(handle, orphan_dir_bh);
1915
1916         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1917                                    OCFS2_ORPHAN_NAMELEN, inode,
1918                                    OCFS2_I(inode)->ip_blkno,
1919                                    orphan_dir_bh, lookup);
1920         if (status < 0) {
1921                 mlog_errno(status);
1922                 goto leave;
1923         }
1924
1925         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1926
1927         /* Record which orphan dir our inode now resides
1928          * in. delete_inode will use this to determine which orphan
1929          * dir to lock. */
1930         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1931
1932         mlog(0, "Inode %llu orphaned in slot %d\n",
1933              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1934
1935 leave:
1936         brelse(orphan_dir_bh);
1937
1938         mlog_exit(status);
1939         return status;
1940 }
1941
1942 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1943 int ocfs2_orphan_del(struct ocfs2_super *osb,
1944                      handle_t *handle,
1945                      struct inode *orphan_dir_inode,
1946                      struct inode *inode,
1947                      struct buffer_head *orphan_dir_bh)
1948 {
1949         char name[OCFS2_ORPHAN_NAMELEN + 1];
1950         struct ocfs2_dinode *orphan_fe;
1951         int status = 0;
1952         struct ocfs2_dir_lookup_result lookup = { NULL, };
1953
1954         mlog_entry_void();
1955
1956         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1957         if (status < 0) {
1958                 mlog_errno(status);
1959                 goto leave;
1960         }
1961
1962         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1963              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1964              OCFS2_ORPHAN_NAMELEN);
1965
1966         /* find it's spot in the orphan directory */
1967         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
1968                                   &lookup);
1969         if (status) {
1970                 mlog_errno(status);
1971                 goto leave;
1972         }
1973
1974         /* remove it from the orphan directory */
1975         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
1976         if (status < 0) {
1977                 mlog_errno(status);
1978                 goto leave;
1979         }
1980
1981         status = ocfs2_journal_access_di(handle,
1982                                          INODE_CACHE(orphan_dir_inode),
1983                                          orphan_dir_bh,
1984                                          OCFS2_JOURNAL_ACCESS_WRITE);
1985         if (status < 0) {
1986                 mlog_errno(status);
1987                 goto leave;
1988         }
1989
1990         /* do the i_nlink dance! :) */
1991         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1992         if (S_ISDIR(inode->i_mode))
1993                 ocfs2_add_links_count(orphan_fe, -1);
1994         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1995         ocfs2_journal_dirty(handle, orphan_dir_bh);
1996
1997 leave:
1998         ocfs2_free_dir_lookup_result(&lookup);
1999
2000         mlog_exit(status);
2001         return status;
2002 }
2003
2004 int ocfs2_create_inode_in_orphan(struct inode *dir,
2005                                  int mode,
2006                                  struct inode **new_inode)
2007 {
2008         int status, did_quota_inode = 0;
2009         struct inode *inode = NULL;
2010         struct inode *orphan_dir = NULL;
2011         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2012         struct ocfs2_dinode *di = NULL;
2013         handle_t *handle = NULL;
2014         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2015         struct buffer_head *parent_di_bh = NULL;
2016         struct buffer_head *new_di_bh = NULL;
2017         struct ocfs2_alloc_context *inode_ac = NULL;
2018         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2019
2020         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2021         if (status < 0) {
2022                 if (status != -ENOENT)
2023                         mlog_errno(status);
2024                 return status;
2025         }
2026
2027         /*
2028          * We give the orphan dir the root blkno to fake an orphan name,
2029          * and allocate enough space for our insertion.
2030          */
2031         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
2032                                           osb->root_blkno,
2033                                           orphan_name, &orphan_insert);
2034         if (status < 0) {
2035                 mlog_errno(status);
2036                 goto leave;
2037         }
2038
2039         /* reserve an inode spot */
2040         status = ocfs2_reserve_new_inode(osb, &inode_ac);
2041         if (status < 0) {
2042                 if (status != -ENOSPC)
2043                         mlog_errno(status);
2044                 goto leave;
2045         }
2046
2047         inode = ocfs2_get_init_inode(dir, mode);
2048         if (!inode) {
2049                 status = -ENOMEM;
2050                 mlog_errno(status);
2051                 goto leave;
2052         }
2053
2054         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2055         if (IS_ERR(handle)) {
2056                 status = PTR_ERR(handle);
2057                 handle = NULL;
2058                 mlog_errno(status);
2059                 goto leave;
2060         }
2061
2062         status = dquot_alloc_inode(inode);
2063         if (status)
2064                 goto leave;
2065         did_quota_inode = 1;
2066
2067         inode->i_nlink = 0;
2068         /* do the real work now. */
2069         status = ocfs2_mknod_locked(osb, dir, inode,
2070                                     0, &new_di_bh, parent_di_bh, handle,
2071                                     inode_ac);
2072         if (status < 0) {
2073                 mlog_errno(status);
2074                 goto leave;
2075         }
2076
2077         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
2078         if (status < 0) {
2079                 mlog_errno(status);
2080                 goto leave;
2081         }
2082
2083         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2084         status = ocfs2_orphan_add(osb, handle, inode, di, orphan_name,
2085                                   &orphan_insert, orphan_dir);
2086         if (status < 0) {
2087                 mlog_errno(status);
2088                 goto leave;
2089         }
2090
2091         /* get open lock so that only nodes can't remove it from orphan dir. */
2092         status = ocfs2_open_lock(inode);
2093         if (status < 0)
2094                 mlog_errno(status);
2095
2096         insert_inode_hash(inode);
2097 leave:
2098         if (status < 0 && did_quota_inode)
2099                 dquot_free_inode(inode);
2100         if (handle)
2101                 ocfs2_commit_trans(osb, handle);
2102
2103         if (orphan_dir) {
2104                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2105                 ocfs2_inode_unlock(orphan_dir, 1);
2106                 mutex_unlock(&orphan_dir->i_mutex);
2107                 iput(orphan_dir);
2108         }
2109
2110         if (status == -ENOSPC)
2111                 mlog(0, "Disk is full\n");
2112
2113         if ((status < 0) && inode) {
2114                 clear_nlink(inode);
2115                 iput(inode);
2116         }
2117
2118         if (inode_ac)
2119                 ocfs2_free_alloc_context(inode_ac);
2120
2121         brelse(new_di_bh);
2122
2123         if (!status)
2124                 *new_inode = inode;
2125
2126         ocfs2_free_dir_lookup_result(&orphan_insert);
2127
2128         ocfs2_inode_unlock(dir, 1);
2129         brelse(parent_di_bh);
2130         return status;
2131 }
2132
2133 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2134                                    struct inode *inode,
2135                                    struct dentry *dentry)
2136 {
2137         int status = 0;
2138         struct buffer_head *parent_di_bh = NULL;
2139         handle_t *handle = NULL;
2140         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2141         struct ocfs2_dinode *dir_di, *di;
2142         struct inode *orphan_dir_inode = NULL;
2143         struct buffer_head *orphan_dir_bh = NULL;
2144         struct buffer_head *di_bh = NULL;
2145         struct ocfs2_dir_lookup_result lookup = { NULL, };
2146
2147         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2148                    dentry->d_name.len, dentry->d_name.name);
2149
2150         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2151         if (status < 0) {
2152                 if (status != -ENOENT)
2153                         mlog_errno(status);
2154                 return status;
2155         }
2156
2157         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2158         if (!dir_di->i_links_count) {
2159                 /* can't make a file in a deleted directory. */
2160                 status = -ENOENT;
2161                 goto leave;
2162         }
2163
2164         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2165                                            dentry->d_name.len);
2166         if (status)
2167                 goto leave;
2168
2169         /* get a spot inside the dir. */
2170         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2171                                               dentry->d_name.name,
2172                                               dentry->d_name.len, &lookup);
2173         if (status < 0) {
2174                 mlog_errno(status);
2175                 goto leave;
2176         }
2177
2178         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2179                                                        ORPHAN_DIR_SYSTEM_INODE,
2180                                                        osb->slot_num);
2181         if (!orphan_dir_inode) {
2182                 status = -EEXIST;
2183                 mlog_errno(status);
2184                 goto leave;
2185         }
2186
2187         mutex_lock(&orphan_dir_inode->i_mutex);
2188
2189         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2190         if (status < 0) {
2191                 mlog_errno(status);
2192                 mutex_unlock(&orphan_dir_inode->i_mutex);
2193                 iput(orphan_dir_inode);
2194                 goto leave;
2195         }
2196
2197         status = ocfs2_read_inode_block(inode, &di_bh);
2198         if (status < 0) {
2199                 mlog_errno(status);
2200                 goto orphan_unlock;
2201         }
2202
2203         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2204         if (IS_ERR(handle)) {
2205                 status = PTR_ERR(handle);
2206                 handle = NULL;
2207                 mlog_errno(status);
2208                 goto orphan_unlock;
2209         }
2210
2211         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2212                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2213         if (status < 0) {
2214                 mlog_errno(status);
2215                 goto out_commit;
2216         }
2217
2218         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2219                                   orphan_dir_bh);
2220         if (status < 0) {
2221                 mlog_errno(status);
2222                 goto out_commit;
2223         }
2224
2225         di = (struct ocfs2_dinode *)di_bh->b_data;
2226         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2227         di->i_orphaned_slot = 0;
2228         inode->i_nlink = 1;
2229         ocfs2_set_links_count(di, inode->i_nlink);
2230         ocfs2_journal_dirty(handle, di_bh);
2231
2232         status = ocfs2_add_entry(handle, dentry, inode,
2233                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2234                                  &lookup);
2235         if (status < 0) {
2236                 mlog_errno(status);
2237                 goto out_commit;
2238         }
2239
2240         status = ocfs2_dentry_attach_lock(dentry, inode,
2241                                           OCFS2_I(dir)->ip_blkno);
2242         if (status) {
2243                 mlog_errno(status);
2244                 goto out_commit;
2245         }
2246
2247         dentry->d_op = &ocfs2_dentry_ops;
2248         d_instantiate(dentry, inode);
2249         status = 0;
2250 out_commit:
2251         ocfs2_commit_trans(osb, handle);
2252 orphan_unlock:
2253         ocfs2_inode_unlock(orphan_dir_inode, 1);
2254         mutex_unlock(&orphan_dir_inode->i_mutex);
2255         iput(orphan_dir_inode);
2256 leave:
2257
2258         ocfs2_inode_unlock(dir, 1);
2259
2260         brelse(di_bh);
2261         brelse(parent_di_bh);
2262         brelse(orphan_dir_bh);
2263
2264         ocfs2_free_dir_lookup_result(&lookup);
2265
2266         mlog_exit(status);
2267
2268         return status;
2269 }
2270
2271 const struct inode_operations ocfs2_dir_iops = {
2272         .create         = ocfs2_create,
2273         .lookup         = ocfs2_lookup,
2274         .link           = ocfs2_link,
2275         .unlink         = ocfs2_unlink,
2276         .rmdir          = ocfs2_unlink,
2277         .symlink        = ocfs2_symlink,
2278         .mkdir          = ocfs2_mkdir,
2279         .mknod          = ocfs2_mknod,
2280         .rename         = ocfs2_rename,
2281         .setattr        = ocfs2_setattr,
2282         .getattr        = ocfs2_getattr,
2283         .permission     = ocfs2_permission,
2284         .setxattr       = generic_setxattr,
2285         .getxattr       = generic_getxattr,
2286         .listxattr      = ocfs2_listxattr,
2287         .removexattr    = generic_removexattr,
2288         .fiemap         = ocfs2_fiemap,
2289 };