fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card ([email protected])
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/quotaops.h>
  44
  45 #define MLOG_MASK_PREFIX ML_NAMEI
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 #include "alloc.h"
  51 #include "dcache.h"
  52 #include "dir.h"
  53 #include "dlmglue.h"
  54 #include "extent_map.h"
  55 #include "file.h"
  56 #include "inode.h"
  57 #include "journal.h"
  58 #include "namei.h"
  59 #include "suballoc.h"
  60 #include "super.h"
  61 #include "symlink.h"
  62 #include "sysfile.h"
  63 #include "uptodate.h"
  64 #include "xattr.h"
  65 #include "acl.h"
  66
  67 #include "buffer_head_io.h"
  68
  69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  70                               struct inode *dir,
  71                               struct inode *inode,
  72                               dev_t dev,
  73                               struct buffer_head **new_fe_bh,
  74                               struct buffer_head *parent_fe_bh,
  75                               handle_t *handle,
  76                               struct ocfs2_alloc_context *inode_ac);
  77
  78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  79                                     struct inode **ret_orphan_dir,
  80                                     u64 blkno,
  81                                     char *name,
  82                                     struct ocfs2_dir_lookup_result *lookup);
  83
  84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  85                             handle_t *handle,
  86                             struct inode *inode,
  87                             struct buffer_head *fe_bh,
  88                             char *name,
  89                             struct ocfs2_dir_lookup_result *lookup,
  90                             struct inode *orphan_dir_inode);
  91
  92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  93                                      handle_t *handle,
  94                                      struct inode *inode,
  95                                      const char *symname);
  96
  97 /* An orphan dir name is an 8 byte value, printed as a hex string */
  98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  99
 100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 101                                    struct nameidata *nd)
 102 {
 103         int status;
 104         u64 blkno;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_inode_info *oi;
 108
 109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 110                    dentry->d_name.len, dentry->d_name.name);
 111
 112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 113                 ret = ERR_PTR(-ENAMETOOLONG);
 114                 goto bail;
 115         }
 116
 117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 119
 120         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 121         if (status < 0) {
 122                 if (status != -ENOENT)
 123                         mlog_errno(status);
 124                 ret = ERR_PTR(status);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 129                                             dentry->d_name.len, &blkno);
 130         if (status < 0)
 131                 goto bail_add;
 132
 133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 134         if (IS_ERR(inode)) {
 135                 ret = ERR_PTR(-EACCES);
 136                 goto bail_unlock;
 137         }
 138
 139         oi = OCFS2_I(inode);
 140         /* Clear any orphaned state... If we were able to look up the
 141          * inode from a directory, it certainly can't be orphaned. We
 142          * might have the bad state from a node which intended to
 143          * orphan this inode but crashed before it could commit the
 144          * unlink. */
 145         spin_lock(&oi->ip_lock);
 146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 147         spin_unlock(&oi->ip_lock);
 148
 149 bail_add:
 150         dentry->d_op = &ocfs2_dentry_ops;
 151         ret = d_splice_alias(inode, dentry);
 152
 153         if (inode) {
 154                 /*
 155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 156                  * dentry, it will d_move() it on top of ourse. The
 157                  * return value will indicate this however, so in
 158                  * those cases, we switch them around for the locking
 159                  * code.
 160                  *
 161                  * NOTE: This dentry already has ->d_op set from
 162                  * ocfs2_get_parent() and ocfs2_get_dentry()
 163                  */
 164                 if (ret)
 165                         dentry = ret;
 166
 167                 status = ocfs2_dentry_attach_lock(dentry, inode,
 168                                                   OCFS2_I(dir)->ip_blkno);
 169                 if (status) {
 170                         mlog_errno(status);
 171                         ret = ERR_PTR(status);
 172                         goto bail_unlock;
 173                 }
 174         }
 175
 176 bail_unlock:
 177         /* Don't drop the cluster lock until *after* the d_add --
 178          * unlink on another node will message us to remove that
 179          * dentry under this lock so otherwise we can race this with
 180          * the downconvert thread and have a stale dentry. */
 181         ocfs2_inode_unlock(dir, 0);
 182
 183 bail:
 184
 185         mlog_exit_ptr(ret);
 186
 187         return ret;
 188 }
 189
 190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
 191 {
 192         struct inode *inode;
 193
 194         inode = new_inode(dir->i_sb);
 195         if (!inode) {
 196                 mlog(ML_ERROR, "new_inode failed!\n");
 197                 return NULL;
 198         }
 199
 200         /* populate as many fields early on as possible - many of
 201          * these are used by the support functions here and in
 202          * callers. */
 203         if (S_ISDIR(mode))
 204                 inode->i_nlink = 2;
 205         else
 206                 inode->i_nlink = 1;
 207         inode->i_uid = current_fsuid();
 208         if (dir->i_mode & S_ISGID) {
 209                 inode->i_gid = dir->i_gid;
 210                 if (S_ISDIR(mode))
 211                         mode |= S_ISGID;
 212         } else
 213                 inode->i_gid = current_fsgid();
 214         inode->i_mode = mode;
 215         dquot_initialize(inode);
 216         return inode;
 217 }
 218
 219 static int ocfs2_mknod(struct inode *dir,
 220                        struct dentry *dentry,
 221                        int mode,
 222                        dev_t dev)
 223 {
 224         int status = 0;
 225         struct buffer_head *parent_fe_bh = NULL;
 226         handle_t *handle = NULL;
 227         struct ocfs2_super *osb;
 228         struct ocfs2_dinode *dirfe;
 229         struct buffer_head *new_fe_bh = NULL;
 230         struct inode *inode = NULL;
 231         struct ocfs2_alloc_context *inode_ac = NULL;
 232         struct ocfs2_alloc_context *data_ac = NULL;
 233         struct ocfs2_alloc_context *meta_ac = NULL;
 234         int want_clusters = 0;
 235         int want_meta = 0;
 236         int xattr_credits = 0;
 237         struct ocfs2_security_xattr_info si = {
 238                 .enable = 1,
 239         };
 240         int did_quota_inode = 0;
 241         struct ocfs2_dir_lookup_result lookup = { NULL, };
 242         sigset_t oldset;
 243         int did_block_signals = 0;
 244
 245         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 246                    (unsigned long)dev, dentry->d_name.len,
 247                    dentry->d_name.name);
 248
 249         dquot_initialize(dir);
 250
 251         /* get our super block */
 252         osb = OCFS2_SB(dir->i_sb);
 253
 254         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 255         if (status < 0) {
 256                 if (status != -ENOENT)
 257                         mlog_errno(status);
 258                 return status;
 259         }
 260
 261         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 262                 status = -EMLINK;
 263                 goto leave;
 264         }
 265
 266         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 267         if (!ocfs2_read_links_count(dirfe)) {
 268                 /* can't make a file in a deleted directory. */
 269                 status = -ENOENT;
 270                 goto leave;
 271         }
 272
 273         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 274                                            dentry->d_name.len);
 275         if (status)
 276                 goto leave;
 277
 278         /* get a spot inside the dir. */
 279         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 280                                               dentry->d_name.name,
 281                                               dentry->d_name.len, &lookup);
 282         if (status < 0) {
 283                 mlog_errno(status);
 284                 goto leave;
 285         }
 286
 287         /* reserve an inode spot */
 288         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 289         if (status < 0) {
 290                 if (status != -ENOSPC)
 291                         mlog_errno(status);
 292                 goto leave;
 293         }
 294
 295         inode = ocfs2_get_init_inode(dir, mode);
 296         if (!inode) {
 297                 status = -ENOMEM;
 298                 mlog_errno(status);
 299                 goto leave;
 300         }
 301
 302         /* get security xattr */
 303         status = ocfs2_init_security_get(inode, dir, &si);
 304         if (status) {
 305                 if (status == -EOPNOTSUPP)
 306                         si.enable = 0;
 307                 else {
 308                         mlog_errno(status);
 309                         goto leave;
 310                 }
 311         }
 312
 313         /* calculate meta data/clusters for setting security and acl xattr */
 314         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 315                                        &si, &want_clusters,
 316                                        &xattr_credits, &want_meta);
 317         if (status < 0) {
 318                 mlog_errno(status);
 319                 goto leave;
 320         }
 321
 322         /* Reserve a cluster if creating an extent based directory. */
 323         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 324                 want_clusters += 1;
 325
 326                 /* Dir indexing requires extra space as well */
 327                 if (ocfs2_supports_indexed_dirs(osb))
 328                         want_meta++;
 329         }
 330
 331         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 332         if (status < 0) {
 333                 if (status != -ENOSPC)
 334                         mlog_errno(status);
 335                 goto leave;
 336         }
 337
 338         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 339         if (status < 0) {
 340                 if (status != -ENOSPC)
 341                         mlog_errno(status);
 342                 goto leave;
 343         }
 344
 345         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 346                                                             S_ISDIR(mode),
 347                                                             xattr_credits));
 348         if (IS_ERR(handle)) {
 349                 status = PTR_ERR(handle);
 350                 handle = NULL;
 351                 mlog_errno(status);
 352                 goto leave;
 353         }
 354
 355         /* Starting to change things, restart is no longer possible. */
 356         ocfs2_block_signals(&oldset);
 357         did_block_signals = 1;
 358
 359         status = dquot_alloc_inode(inode);
 360         if (status)
 361                 goto leave;
 362         did_quota_inode = 1;
 363
 364         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
 365                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
 366                    dentry->d_name.name);
 367
 368         /* do the real work now. */
 369         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 370                                     &new_fe_bh, parent_fe_bh, handle,
 371                                     inode_ac);
 372         if (status < 0) {
 373                 mlog_errno(status);
 374                 goto leave;
 375         }
 376
 377         if (S_ISDIR(mode)) {
 378                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 379                                             new_fe_bh, data_ac, meta_ac);
 380                 if (status < 0) {
 381                         mlog_errno(status);
 382                         goto leave;
 383                 }
 384
 385                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 386                                                  parent_fe_bh,
 387                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 388                 if (status < 0) {
 389                         mlog_errno(status);
 390                         goto leave;
 391                 }
 392                 ocfs2_add_links_count(dirfe, 1);
 393                 ocfs2_journal_dirty(handle, parent_fe_bh);
 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         /*
 414          * Do this before adding the entry to the directory. We add
 415          * also set d_op after success so that ->d_iput() will cleanup
 416          * the dentry lock even if ocfs2_add_entry() fails below.
 417          */
 418         status = ocfs2_dentry_attach_lock(dentry, inode,
 419                                           OCFS2_I(dir)->ip_blkno);
 420         if (status) {
 421                 mlog_errno(status);
 422                 goto leave;
 423         }
 424         dentry->d_op = &ocfs2_dentry_ops;
 425
 426         status = ocfs2_add_entry(handle, dentry, inode,
 427                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 428                                  &lookup);
 429         if (status < 0) {
 430                 mlog_errno(status);
 431                 goto leave;
 432         }
 433
 434         insert_inode_hash(inode);
 435         d_instantiate(dentry, inode);
 436         status = 0;
 437 leave:
 438         if (status < 0 && did_quota_inode)
 439                 dquot_free_inode(inode);
 440         if (handle)
 441                 ocfs2_commit_trans(osb, handle);
 442
 443         ocfs2_inode_unlock(dir, 1);
 444         if (did_block_signals)
 445                 ocfs2_unblock_signals(&oldset);
 446
 447         if (status == -ENOSPC)
 448                 mlog(0, "Disk is full\n");
 449
 450         brelse(new_fe_bh);
 451         brelse(parent_fe_bh);
 452         kfree(si.name);
 453         kfree(si.value);
 454
 455         ocfs2_free_dir_lookup_result(&lookup);
 456
 457         if (inode_ac)
 458                 ocfs2_free_alloc_context(inode_ac);
 459
 460         if (data_ac)
 461                 ocfs2_free_alloc_context(data_ac);
 462
 463         if (meta_ac)
 464                 ocfs2_free_alloc_context(meta_ac);
 465
 466         /*
 467          * We should call iput after the i_mutex of the bitmap been
 468          * unlocked in ocfs2_free_alloc_context, or the
 469          * ocfs2_delete_inode will mutex_lock again.
 470          */
 471         if ((status < 0) && inode) {
 472                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 473                 clear_nlink(inode);
 474                 iput(inode);
 475         }
 476
 477         mlog_exit(status);
 478
 479         return status;
 480 }
 481
 482 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 483                               struct inode *dir,
 484                               struct inode *inode,
 485                               dev_t dev,
 486                               struct buffer_head **new_fe_bh,
 487                               struct buffer_head *parent_fe_bh,
 488                               handle_t *handle,
 489                               struct ocfs2_alloc_context *inode_ac)
 490 {
 491         int status = 0;
 492         struct ocfs2_dinode *fe = NULL;
 493         struct ocfs2_extent_list *fel;
 494         u64 suballoc_loc, fe_blkno = 0;
 495         u16 suballoc_bit;
 496         u16 feat;
 497
 498         *new_fe_bh = NULL;
 499
 500         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 501                                        inode_ac, &suballoc_loc,
 502                                        &suballoc_bit, &fe_blkno);
 503         if (status < 0) {
 504                 mlog_errno(status);
 505                 goto leave;
 506         }
 507
 508         /* populate as many fields early on as possible - many of
 509          * these are used by the support functions here and in
 510          * callers. */
 511         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 512         OCFS2_I(inode)->ip_blkno = fe_blkno;
 513         spin_lock(&osb->osb_lock);
 514         inode->i_generation = osb->s_next_generation++;
 515         spin_unlock(&osb->osb_lock);
 516
 517         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 518         if (!*new_fe_bh) {
 519                 status = -EIO;
 520                 mlog_errno(status);
 521                 goto leave;
 522         }
 523         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 524
 525         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 526                                          *new_fe_bh,
 527                                          OCFS2_JOURNAL_ACCESS_CREATE);
 528         if (status < 0) {
 529                 mlog_errno(status);
 530                 goto leave;
 531         }
 532
 533         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 534         memset(fe, 0, osb->sb->s_blocksize);
 535
 536         fe->i_generation = cpu_to_le32(inode->i_generation);
 537         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 538         fe->i_blkno = cpu_to_le64(fe_blkno);
 539         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 540         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 541         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 542         fe->i_uid = cpu_to_le32(inode->i_uid);
 543         fe->i_gid = cpu_to_le32(inode->i_gid);
 544         fe->i_mode = cpu_to_le16(inode->i_mode);
 545         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 546                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 547
 548         ocfs2_set_links_count(fe, inode->i_nlink);
 549
 550         fe->i_last_eb_blk = 0;
 551         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 552         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 553         fe->i_atime = fe->i_ctime = fe->i_mtime =
 554                 cpu_to_le64(CURRENT_TIME.tv_sec);
 555         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 556                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 557         fe->i_dtime = 0;
 558
 559         /*
 560          * If supported, directories start with inline data. If inline
 561          * isn't supported, but indexing is, we start them as indexed.
 562          */
 563         feat = le16_to_cpu(fe->i_dyn_features);
 564         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 565                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 566
 567                 fe->id2.i_data.id_count = cpu_to_le16(
 568                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 569         } else {
 570                 fel = &fe->id2.i_list;
 571                 fel->l_tree_depth = 0;
 572                 fel->l_next_free_rec = 0;
 573                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 574         }
 575
 576         ocfs2_journal_dirty(handle, *new_fe_bh);
 577
 578         ocfs2_populate_inode(inode, fe, 1);
 579         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 580         if (!ocfs2_mount_local(osb)) {
 581                 status = ocfs2_create_new_inode_locks(inode);
 582                 if (status < 0)
 583                         mlog_errno(status);
 584         }
 585
 586         status = 0; /* error in ocfs2_create_new_inode_locks is not
 587                      * critical */
 588
 589 leave:
 590         if (status < 0) {
 591                 if (*new_fe_bh) {
 592                         brelse(*new_fe_bh);
 593                         *new_fe_bh = NULL;
 594                 }
 595         }
 596
 597         mlog_exit(status);
 598         return status;
 599 }
 600
 601 static int ocfs2_mkdir(struct inode *dir,
 602                        struct dentry *dentry,
 603                        int mode)
 604 {
 605         int ret;
 606
 607         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 608                    dentry->d_name.len, dentry->d_name.name);
 609         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 610         mlog_exit(ret);
 611
 612         return ret;
 613 }
 614
 615 static int ocfs2_create(struct inode *dir,
 616                         struct dentry *dentry,
 617                         int mode,
 618                         struct nameidata *nd)
 619 {
 620         int ret;
 621
 622         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 623                    dentry->d_name.len, dentry->d_name.name);
 624         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 625         mlog_exit(ret);
 626
 627         return ret;
 628 }
 629
 630 static int ocfs2_link(struct dentry *old_dentry,
 631                       struct inode *dir,
 632                       struct dentry *dentry)
 633 {
 634         handle_t *handle;
 635         struct inode *inode = old_dentry->d_inode;
 636         int err;
 637         struct buffer_head *fe_bh = NULL;
 638         struct buffer_head *parent_fe_bh = NULL;
 639         struct ocfs2_dinode *fe = NULL;
 640         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 641         struct ocfs2_dir_lookup_result lookup = { NULL, };
 642         sigset_t oldset;
 643
 644         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 645                    old_dentry->d_name.len, old_dentry->d_name.name,
 646                    dentry->d_name.len, dentry->d_name.name);
 647
 648         if (S_ISDIR(inode->i_mode))
 649                 return -EPERM;
 650
 651         dquot_initialize(dir);
 652
 653         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 654         if (err < 0) {
 655                 if (err != -ENOENT)
 656                         mlog_errno(err);
 657                 return err;
 658         }
 659
 660         if (!dir->i_nlink) {
 661                 err = -ENOENT;
 662                 goto out;
 663         }
 664
 665         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 666                                         dentry->d_name.len);
 667         if (err)
 668                 goto out;
 669
 670         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 671                                            dentry->d_name.name,
 672                                            dentry->d_name.len, &lookup);
 673         if (err < 0) {
 674                 mlog_errno(err);
 675                 goto out;
 676         }
 677
 678         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 679         if (err < 0) {
 680                 if (err != -ENOENT)
 681                         mlog_errno(err);
 682                 goto out;
 683         }
 684
 685         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 686         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 687                 err = -EMLINK;
 688                 goto out_unlock_inode;
 689         }
 690
 691         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 692         if (IS_ERR(handle)) {
 693                 err = PTR_ERR(handle);
 694                 handle = NULL;
 695                 mlog_errno(err);
 696                 goto out_unlock_inode;
 697         }
 698
 699         /* Starting to change things, restart is no longer possible. */
 700         ocfs2_block_signals(&oldset);
 701
 702         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 703                                       OCFS2_JOURNAL_ACCESS_WRITE);
 704         if (err < 0) {
 705                 mlog_errno(err);
 706                 goto out_commit;
 707         }
 708
 709         inc_nlink(inode);
 710         inode->i_ctime = CURRENT_TIME;
 711         ocfs2_set_links_count(fe, inode->i_nlink);
 712         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 713         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 714         ocfs2_journal_dirty(handle, fe_bh);
 715
 716         err = ocfs2_add_entry(handle, dentry, inode,
 717                               OCFS2_I(inode)->ip_blkno,
 718                               parent_fe_bh, &lookup);
 719         if (err) {
 720                 ocfs2_add_links_count(fe, -1);
 721                 drop_nlink(inode);
 722                 mlog_errno(err);
 723                 goto out_commit;
 724         }
 725
 726         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 727         if (err) {
 728                 mlog_errno(err);
 729                 goto out_commit;
 730         }
 731
 732         atomic_inc(&inode->i_count);
 733         dentry->d_op = &ocfs2_dentry_ops;
 734         d_instantiate(dentry, inode);
 735
 736 out_commit:
 737         ocfs2_commit_trans(osb, handle);
 738         ocfs2_unblock_signals(&oldset);
 739 out_unlock_inode:
 740         ocfs2_inode_unlock(inode, 1);
 741
 742 out:
 743         ocfs2_inode_unlock(dir, 1);
 744
 745         brelse(fe_bh);
 746         brelse(parent_fe_bh);
 747
 748         ocfs2_free_dir_lookup_result(&lookup);
 749
 750         mlog_exit(err);
 751
 752         return err;
 753 }
 754
 755 /*
 756  * Takes and drops an exclusive lock on the given dentry. This will
 757  * force other nodes to drop it.
 758  */
 759 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 760 {
 761         int ret;
 762
 763         ret = ocfs2_dentry_lock(dentry, 1);
 764         if (ret)
 765                 mlog_errno(ret);
 766         else
 767                 ocfs2_dentry_unlock(dentry, 1);
 768
 769         return ret;
 770 }
 771
 772 static inline int inode_is_unlinkable(struct inode *inode)
 773 {
 774         if (S_ISDIR(inode->i_mode)) {
 775                 if (inode->i_nlink == 2)
 776                         return 1;
 777                 return 0;
 778         }
 779
 780         if (inode->i_nlink == 1)
 781                 return 1;
 782         return 0;
 783 }
 784
 785 static int ocfs2_unlink(struct inode *dir,
 786                         struct dentry *dentry)
 787 {
 788         int status;
 789         int child_locked = 0;
 790         struct inode *inode = dentry->d_inode;
 791         struct inode *orphan_dir = NULL;
 792         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 793         u64 blkno;
 794         struct ocfs2_dinode *fe = NULL;
 795         struct buffer_head *fe_bh = NULL;
 796         struct buffer_head *parent_node_bh = NULL;
 797         handle_t *handle = NULL;
 798         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 799         struct ocfs2_dir_lookup_result lookup = { NULL, };
 800         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 801
 802         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 803                    dentry->d_name.len, dentry->d_name.name);
 804
 805         dquot_initialize(dir);
 806
 807         BUG_ON(dentry->d_parent->d_inode != dir);
 808
 809         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 810
 811         if (inode == osb->root_inode) {
 812                 mlog(0, "Cannot delete the root directory\n");
 813                 return -EPERM;
 814         }
 815
 816         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 817                                          OI_LS_PARENT);
 818         if (status < 0) {
 819                 if (status != -ENOENT)
 820                         mlog_errno(status);
 821                 return status;
 822         }
 823
 824         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 825                                           dentry->d_name.len, &blkno, dir,
 826                                           &lookup);
 827         if (status < 0) {
 828                 if (status != -ENOENT)
 829                         mlog_errno(status);
 830                 goto leave;
 831         }
 832
 833         if (OCFS2_I(inode)->ip_blkno != blkno) {
 834                 status = -ENOENT;
 835
 836                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 837                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 838                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 839                 goto leave;
 840         }
 841
 842         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 843         if (status < 0) {
 844                 if (status != -ENOENT)
 845                         mlog_errno(status);
 846                 goto leave;
 847         }
 848         child_locked = 1;
 849
 850         if (S_ISDIR(inode->i_mode)) {
 851                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 852                         status = -ENOTEMPTY;
 853                         goto leave;
 854                 }
 855         }
 856
 857         status = ocfs2_remote_dentry_delete(dentry);
 858         if (status < 0) {
 859                 /* This remote delete should succeed under all normal
 860                  * circumstances. */
 861                 mlog_errno(status);
 862                 goto leave;
 863         }
 864
 865         if (inode_is_unlinkable(inode)) {
 866                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 867                                                   OCFS2_I(inode)->ip_blkno,
 868                                                   orphan_name, &orphan_insert);
 869                 if (status < 0) {
 870                         mlog_errno(status);
 871                         goto leave;
 872                 }
 873         }
 874
 875         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 876         if (IS_ERR(handle)) {
 877                 status = PTR_ERR(handle);
 878                 handle = NULL;
 879                 mlog_errno(status);
 880                 goto leave;
 881         }
 882
 883         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 884                                          OCFS2_JOURNAL_ACCESS_WRITE);
 885         if (status < 0) {
 886                 mlog_errno(status);
 887                 goto leave;
 888         }
 889
 890         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 891
 892         if (inode_is_unlinkable(inode)) {
 893                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 894                                           &orphan_insert, orphan_dir);
 895                 if (status < 0) {
 896                         mlog_errno(status);
 897                         goto leave;
 898                 }
 899         }
 900
 901         /* delete the name from the parent dir */
 902         status = ocfs2_delete_entry(handle, dir, &lookup);
 903         if (status < 0) {
 904                 mlog_errno(status);
 905                 goto leave;
 906         }
 907
 908         if (S_ISDIR(inode->i_mode))
 909                 drop_nlink(inode);
 910         drop_nlink(inode);
 911         ocfs2_set_links_count(fe, inode->i_nlink);
 912         ocfs2_journal_dirty(handle, fe_bh);
 913
 914         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 915         if (S_ISDIR(inode->i_mode))
 916                 drop_nlink(dir);
 917
 918         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 919         if (status < 0) {
 920                 mlog_errno(status);
 921                 if (S_ISDIR(inode->i_mode))
 922                         inc_nlink(dir);
 923         }
 924
 925 leave:
 926         if (handle)
 927                 ocfs2_commit_trans(osb, handle);
 928
 929         if (child_locked)
 930                 ocfs2_inode_unlock(inode, 1);
 931
 932         ocfs2_inode_unlock(dir, 1);
 933
 934         if (orphan_dir) {
 935                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 936                 ocfs2_inode_unlock(orphan_dir, 1);
 937                 mutex_unlock(&orphan_dir->i_mutex);
 938                 iput(orphan_dir);
 939         }
 940
 941         brelse(fe_bh);
 942         brelse(parent_node_bh);
 943
 944         ocfs2_free_dir_lookup_result(&orphan_insert);
 945         ocfs2_free_dir_lookup_result(&lookup);
 946
 947         mlog_exit(status);
 948
 949         return status;
 950 }
 951
 952 /*
 953  * The only place this should be used is rename!
 954  * if they have the same id, then the 1st one is the only one locked.
 955  */
 956 static int ocfs2_double_lock(struct ocfs2_super *osb,
 957                              struct buffer_head **bh1,
 958                              struct inode *inode1,
 959                              struct buffer_head **bh2,
 960                              struct inode *inode2)
 961 {
 962         int status;
 963         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 964         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 965         struct buffer_head **tmpbh;
 966         struct inode *tmpinode;
 967
 968         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 969                    (unsigned long long)oi1->ip_blkno,
 970                    (unsigned long long)oi2->ip_blkno);
 971
 972         if (*bh1)
 973                 *bh1 = NULL;
 974         if (*bh2)
 975                 *bh2 = NULL;
 976
 977         /* we always want to lock the one with the lower lockid first. */
 978         if (oi1->ip_blkno != oi2->ip_blkno) {
 979                 if (oi1->ip_blkno < oi2->ip_blkno) {
 980                         /* switch id1 and id2 around */
 981                         mlog(0, "switching them around...\n");
 982                         tmpbh = bh2;
 983                         bh2 = bh1;
 984                         bh1 = tmpbh;
 985
 986                         tmpinode = inode2;
 987                         inode2 = inode1;
 988                         inode1 = tmpinode;
 989                 }
 990                 /* lock id2 */
 991                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 992                                                  OI_LS_RENAME1);
 993                 if (status < 0) {
 994                         if (status != -ENOENT)
 995                                 mlog_errno(status);
 996                         goto bail;
 997                 }
 998         }
 999
1000         /* lock id1 */
1001         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1002         if (status < 0) {
1003                 /*
1004                  * An error return must mean that no cluster locks
1005                  * were held on function exit.
1006                  */
1007                 if (oi1->ip_blkno != oi2->ip_blkno)
1008                         ocfs2_inode_unlock(inode2, 1);
1009
1010                 if (status != -ENOENT)
1011                         mlog_errno(status);
1012         }
1013
1014 bail:
1015         mlog_exit(status);
1016         return status;
1017 }
1018
1019 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1020 {
1021         ocfs2_inode_unlock(inode1, 1);
1022
1023         if (inode1 != inode2)
1024                 ocfs2_inode_unlock(inode2, 1);
1025 }
1026
1027 static int ocfs2_rename(struct inode *old_dir,
1028                         struct dentry *old_dentry,
1029                         struct inode *new_dir,
1030                         struct dentry *new_dentry)
1031 {
1032         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1033         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1034         struct inode *old_inode = old_dentry->d_inode;
1035         struct inode *new_inode = new_dentry->d_inode;
1036         struct inode *orphan_dir = NULL;
1037         struct ocfs2_dinode *newfe = NULL;
1038         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1039         struct buffer_head *newfe_bh = NULL;
1040         struct buffer_head *old_inode_bh = NULL;
1041         struct ocfs2_super *osb = NULL;
1042         u64 newfe_blkno, old_de_ino;
1043         handle_t *handle = NULL;
1044         struct buffer_head *old_dir_bh = NULL;
1045         struct buffer_head *new_dir_bh = NULL;
1046         nlink_t old_dir_nlink = old_dir->i_nlink;
1047         struct ocfs2_dinode *old_di;
1048         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1049         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1050         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1051         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1052         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1053
1054         /* At some point it might be nice to break this function up a
1055          * bit. */
1056
1057         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1058                    old_dir, old_dentry, new_dir, new_dentry,
1059                    old_dentry->d_name.len, old_dentry->d_name.name,
1060                    new_dentry->d_name.len, new_dentry->d_name.name);
1061
1062         dquot_initialize(old_dir);
1063         dquot_initialize(new_dir);
1064
1065         osb = OCFS2_SB(old_dir->i_sb);
1066
1067         if (new_inode) {
1068                 if (!igrab(new_inode))
1069                         BUG();
1070         }
1071
1072         /* Assume a directory hierarchy thusly:
1073          * a/b/c
1074          * a/d
1075          * a,b,c, and d are all directories.
1076          *
1077          * from cwd of 'a' on both nodes:
1078          * node1: mv b/c d
1079          * node2: mv d   b/c
1080          *
1081          * And that's why, just like the VFS, we need a file system
1082          * rename lock. */
1083         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1084                 status = ocfs2_rename_lock(osb);
1085                 if (status < 0) {
1086                         mlog_errno(status);
1087                         goto bail;
1088                 }
1089                 rename_lock = 1;
1090         }
1091
1092         /* if old and new are the same, this'll just do one lock. */
1093         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1094                                    &new_dir_bh, new_dir);
1095         if (status < 0) {
1096                 mlog_errno(status);
1097                 goto bail;
1098         }
1099         parents_locked = 1;
1100
1101         /* make sure both dirs have bhs
1102          * get an extra ref on old_dir_bh if old==new */
1103         if (!new_dir_bh) {
1104                 if (old_dir_bh) {
1105                         new_dir_bh = old_dir_bh;
1106                         get_bh(new_dir_bh);
1107                 } else {
1108                         mlog(ML_ERROR, "no old_dir_bh!\n");
1109                         status = -EIO;
1110                         goto bail;
1111                 }
1112         }
1113
1114         /*
1115          * Aside from allowing a meta data update, the locking here
1116          * also ensures that the downconvert thread on other nodes
1117          * won't have to concurrently downconvert the inode and the
1118          * dentry locks.
1119          */
1120         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1121                                          OI_LS_PARENT);
1122         if (status < 0) {
1123                 if (status != -ENOENT)
1124                         mlog_errno(status);
1125                 goto bail;
1126         }
1127         old_child_locked = 1;
1128
1129         status = ocfs2_remote_dentry_delete(old_dentry);
1130         if (status < 0) {
1131                 mlog_errno(status);
1132                 goto bail;
1133         }
1134
1135         if (S_ISDIR(old_inode->i_mode)) {
1136                 u64 old_inode_parent;
1137
1138                 update_dot_dot = 1;
1139                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1140                                                   old_inode,
1141                                                   &old_inode_dot_dot_res);
1142                 if (status) {
1143                         status = -EIO;
1144                         goto bail;
1145                 }
1146
1147                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1148                         status = -EIO;
1149                         goto bail;
1150                 }
1151
1152                 if (!new_inode && new_dir != old_dir &&
1153                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1154                         status = -EMLINK;
1155                         goto bail;
1156                 }
1157         }
1158
1159         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1160                                             old_dentry->d_name.len,
1161                                             &old_de_ino);
1162         if (status) {
1163                 status = -ENOENT;
1164                 goto bail;
1165         }
1166
1167         /*
1168          *  Check for inode number is _not_ due to possible IO errors.
1169          *  We might rmdir the source, keep it as pwd of some process
1170          *  and merrily kill the link to whatever was created under the
1171          *  same name. Goodbye sticky bit ;-<
1172          */
1173         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1174                 status = -ENOENT;
1175                 goto bail;
1176         }
1177
1178         /* check if the target already exists (in which case we need
1179          * to delete it */
1180         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1181                                           new_dentry->d_name.len,
1182                                           &newfe_blkno, new_dir,
1183                                           &target_lookup_res);
1184         /* The only error we allow here is -ENOENT because the new
1185          * file not existing is perfectly valid. */
1186         if ((status < 0) && (status != -ENOENT)) {
1187                 /* If we cannot find the file specified we should just */
1188                 /* return the error... */
1189                 mlog_errno(status);
1190                 goto bail;
1191         }
1192         if (status == 0)
1193                 target_exists = 1;
1194
1195         if (!target_exists && new_inode) {
1196                 /*
1197                  * Target was unlinked by another node while we were
1198                  * waiting to get to ocfs2_rename(). There isn't
1199                  * anything we can do here to help the situation, so
1200                  * bubble up the appropriate error.
1201                  */
1202                 status = -ENOENT;
1203                 goto bail;
1204         }
1205
1206         /* In case we need to overwrite an existing file, we blow it
1207          * away first */
1208         if (target_exists) {
1209                 /* VFS didn't think there existed an inode here, but
1210                  * someone else in the cluster must have raced our
1211                  * rename to create one. Today we error cleanly, in
1212                  * the future we should consider calling iget to build
1213                  * a new struct inode for this entry. */
1214                 if (!new_inode) {
1215                         status = -EACCES;
1216
1217                         mlog(0, "We found an inode for name %.*s but VFS "
1218                              "didn't give us one.\n", new_dentry->d_name.len,
1219                              new_dentry->d_name.name);
1220                         goto bail;
1221                 }
1222
1223                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1224                         status = -EACCES;
1225
1226                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1227                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1228                              (unsigned long long)newfe_blkno,
1229                              OCFS2_I(new_inode)->ip_flags);
1230                         goto bail;
1231                 }
1232
1233                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1234                 if (status < 0) {
1235                         if (status != -ENOENT)
1236                                 mlog_errno(status);
1237                         goto bail;
1238                 }
1239                 new_child_locked = 1;
1240
1241                 status = ocfs2_remote_dentry_delete(new_dentry);
1242                 if (status < 0) {
1243                         mlog_errno(status);
1244                         goto bail;
1245                 }
1246
1247                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1248
1249                 mlog(0, "aha rename over existing... new_blkno=%llu "
1250                      "newfebh=%p bhblocknr=%llu\n",
1251                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1252                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1253
1254                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1255                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1256                                                 OCFS2_I(new_inode)->ip_blkno,
1257                                                 orphan_name, &orphan_insert);
1258                         if (status < 0) {
1259                                 mlog_errno(status);
1260                                 goto bail;
1261                         }
1262                 }
1263         } else {
1264                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1265
1266                 status = ocfs2_check_dir_for_entry(new_dir,
1267                                                    new_dentry->d_name.name,
1268                                                    new_dentry->d_name.len);
1269                 if (status)
1270                         goto bail;
1271
1272                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1273                                                       new_dentry->d_name.name,
1274                                                       new_dentry->d_name.len,
1275                                                       &target_insert);
1276                 if (status < 0) {
1277                         mlog_errno(status);
1278                         goto bail;
1279                 }
1280         }
1281
1282         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1283         if (IS_ERR(handle)) {
1284                 status = PTR_ERR(handle);
1285                 handle = NULL;
1286                 mlog_errno(status);
1287                 goto bail;
1288         }
1289
1290         if (target_exists) {
1291                 if (S_ISDIR(new_inode->i_mode)) {
1292                         if (new_inode->i_nlink != 2 ||
1293                             !ocfs2_empty_dir(new_inode)) {
1294                                 status = -ENOTEMPTY;
1295                                 goto bail;
1296                         }
1297                 }
1298                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1299                                                  newfe_bh,
1300                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1301                 if (status < 0) {
1302                         mlog_errno(status);
1303                         goto bail;
1304                 }
1305
1306                 if (S_ISDIR(new_inode->i_mode) ||
1307                     (ocfs2_read_links_count(newfe) == 1)) {
1308                         status = ocfs2_orphan_add(osb, handle, new_inode,
1309                                                   newfe_bh, orphan_name,
1310                                                   &orphan_insert, orphan_dir);
1311                         if (status < 0) {
1312                                 mlog_errno(status);
1313                                 goto bail;
1314                         }
1315                 }
1316
1317                 /* change the dirent to point to the correct inode */
1318                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1319                                             old_inode);
1320                 if (status < 0) {
1321                         mlog_errno(status);
1322                         goto bail;
1323                 }
1324                 new_dir->i_version++;
1325
1326                 if (S_ISDIR(new_inode->i_mode))
1327                         ocfs2_set_links_count(newfe, 0);
1328                 else
1329                         ocfs2_add_links_count(newfe, -1);
1330                 ocfs2_journal_dirty(handle, newfe_bh);
1331         } else {
1332                 /* if the name was not found in new_dir, add it now */
1333                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1334                                          OCFS2_I(old_inode)->ip_blkno,
1335                                          new_dir_bh, &target_insert);
1336         }
1337
1338         old_inode->i_ctime = CURRENT_TIME;
1339         mark_inode_dirty(old_inode);
1340
1341         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1342                                          old_inode_bh,
1343                                          OCFS2_JOURNAL_ACCESS_WRITE);
1344         if (status >= 0) {
1345                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1346
1347                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1348                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1349                 ocfs2_journal_dirty(handle, old_inode_bh);
1350         } else
1351                 mlog_errno(status);
1352
1353         /*
1354          * Now that the name has been added to new_dir, remove the old name.
1355          *
1356          * We don't keep any directory entry context around until now
1357          * because the insert might have changed the type of directory
1358          * we're dealing with.
1359          */
1360         status = ocfs2_find_entry(old_dentry->d_name.name,
1361                                   old_dentry->d_name.len, old_dir,
1362                                   &old_entry_lookup);
1363         if (status)
1364                 goto bail;
1365
1366         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1367         if (status < 0) {
1368                 mlog_errno(status);
1369                 goto bail;
1370         }
1371
1372         if (new_inode) {
1373                 new_inode->i_nlink--;
1374                 new_inode->i_ctime = CURRENT_TIME;
1375         }
1376         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1377
1378         if (update_dot_dot) {
1379                 status = ocfs2_update_entry(old_inode, handle,
1380                                             &old_inode_dot_dot_res, new_dir);
1381                 old_dir->i_nlink--;
1382                 if (new_inode) {
1383                         new_inode->i_nlink--;
1384                 } else {
1385                         inc_nlink(new_dir);
1386                         mark_inode_dirty(new_dir);
1387                 }
1388         }
1389         mark_inode_dirty(old_dir);
1390         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1391         if (new_inode) {
1392                 mark_inode_dirty(new_inode);
1393                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1394         }
1395
1396         if (old_dir != new_dir) {
1397                 /* Keep the same times on both directories.*/
1398                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1399
1400                 /*
1401                  * This will also pick up the i_nlink change from the
1402                  * block above.
1403                  */
1404                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1405         }
1406
1407         if (old_dir_nlink != old_dir->i_nlink) {
1408                 if (!old_dir_bh) {
1409                         mlog(ML_ERROR, "need to change nlink for old dir "
1410                              "%llu from %d to %d but bh is NULL!\n",
1411                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1412                              (int)old_dir_nlink, old_dir->i_nlink);
1413                 } else {
1414                         struct ocfs2_dinode *fe;
1415                         status = ocfs2_journal_access_di(handle,
1416                                                          INODE_CACHE(old_dir),
1417                                                          old_dir_bh,
1418                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1419                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1420                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1421                         ocfs2_journal_dirty(handle, old_dir_bh);
1422                 }
1423         }
1424         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1425         status = 0;
1426 bail:
1427         if (rename_lock)
1428                 ocfs2_rename_unlock(osb);
1429
1430         if (handle)
1431                 ocfs2_commit_trans(osb, handle);
1432
1433         if (parents_locked)
1434                 ocfs2_double_unlock(old_dir, new_dir);
1435
1436         if (old_child_locked)
1437                 ocfs2_inode_unlock(old_inode, 1);
1438
1439         if (new_child_locked)
1440                 ocfs2_inode_unlock(new_inode, 1);
1441
1442         if (orphan_dir) {
1443                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1444                 ocfs2_inode_unlock(orphan_dir, 1);
1445                 mutex_unlock(&orphan_dir->i_mutex);
1446                 iput(orphan_dir);
1447         }
1448
1449         if (new_inode)
1450                 sync_mapping_buffers(old_inode->i_mapping);
1451
1452         if (new_inode)
1453                 iput(new_inode);
1454
1455         ocfs2_free_dir_lookup_result(&target_lookup_res);
1456         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1457         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1458         ocfs2_free_dir_lookup_result(&orphan_insert);
1459         ocfs2_free_dir_lookup_result(&target_insert);
1460
1461         brelse(newfe_bh);
1462         brelse(old_inode_bh);
1463         brelse(old_dir_bh);
1464         brelse(new_dir_bh);
1465
1466         mlog_exit(status);
1467
1468         return status;
1469 }
1470
1471 /*
1472  * we expect i_size = strlen(symname). Copy symname into the file
1473  * data, including the null terminator.
1474  */
1475 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1476                                      handle_t *handle,
1477                                      struct inode *inode,
1478                                      const char *symname)
1479 {
1480         struct buffer_head **bhs = NULL;
1481         const char *c;
1482         struct super_block *sb = osb->sb;
1483         u64 p_blkno, p_blocks;
1484         int virtual, blocks, status, i, bytes_left;
1485
1486         bytes_left = i_size_read(inode) + 1;
1487         /* we can't trust i_blocks because we're actually going to
1488          * write i_size + 1 bytes. */
1489         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1490
1491         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1492                         (unsigned long long)inode->i_blocks,
1493                         i_size_read(inode), blocks);
1494
1495         /* Sanity check -- make sure we're going to fit. */
1496         if (bytes_left >
1497             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1498                 status = -EIO;
1499                 mlog_errno(status);
1500                 goto bail;
1501         }
1502
1503         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1504         if (!bhs) {
1505                 status = -ENOMEM;
1506                 mlog_errno(status);
1507                 goto bail;
1508         }
1509
1510         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1511                                              NULL);
1512         if (status < 0) {
1513                 mlog_errno(status);
1514                 goto bail;
1515         }
1516
1517         /* links can never be larger than one cluster so we know this
1518          * is all going to be contiguous, but do a sanity check
1519          * anyway. */
1520         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1521                 status = -EIO;
1522                 mlog_errno(status);
1523                 goto bail;
1524         }
1525
1526         virtual = 0;
1527         while(bytes_left > 0) {
1528                 c = &symname[virtual * sb->s_blocksize];
1529
1530                 bhs[virtual] = sb_getblk(sb, p_blkno);
1531                 if (!bhs[virtual]) {
1532                         status = -ENOMEM;
1533                         mlog_errno(status);
1534                         goto bail;
1535                 }
1536                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1537                                               bhs[virtual]);
1538
1539                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1540                                               bhs[virtual],
1541                                               OCFS2_JOURNAL_ACCESS_CREATE);
1542                 if (status < 0) {
1543                         mlog_errno(status);
1544                         goto bail;
1545                 }
1546
1547                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1548
1549                 memcpy(bhs[virtual]->b_data, c,
1550                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1551                        bytes_left);
1552
1553                 ocfs2_journal_dirty(handle, bhs[virtual]);
1554
1555                 virtual++;
1556                 p_blkno++;
1557                 bytes_left -= sb->s_blocksize;
1558         }
1559
1560         status = 0;
1561 bail:
1562
1563         if (bhs) {
1564                 for(i = 0; i < blocks; i++)
1565                         brelse(bhs[i]);
1566                 kfree(bhs);
1567         }
1568
1569         mlog_exit(status);
1570         return status;
1571 }
1572
1573 static int ocfs2_symlink(struct inode *dir,
1574                          struct dentry *dentry,
1575                          const char *symname)
1576 {
1577         int status, l, credits;
1578         u64 newsize;
1579         struct ocfs2_super *osb = NULL;
1580         struct inode *inode = NULL;
1581         struct super_block *sb;
1582         struct buffer_head *new_fe_bh = NULL;
1583         struct buffer_head *parent_fe_bh = NULL;
1584         struct ocfs2_dinode *fe = NULL;
1585         struct ocfs2_dinode *dirfe;
1586         handle_t *handle = NULL;
1587         struct ocfs2_alloc_context *inode_ac = NULL;
1588         struct ocfs2_alloc_context *data_ac = NULL;
1589         struct ocfs2_alloc_context *xattr_ac = NULL;
1590         int want_clusters = 0;
1591         int xattr_credits = 0;
1592         struct ocfs2_security_xattr_info si = {
1593                 .enable = 1,
1594         };
1595         int did_quota = 0, did_quota_inode = 0;
1596         struct ocfs2_dir_lookup_result lookup = { NULL, };
1597         sigset_t oldset;
1598         int did_block_signals = 0;
1599
1600         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1601                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1602
1603         dquot_initialize(dir);
1604
1605         sb = dir->i_sb;
1606         osb = OCFS2_SB(sb);
1607
1608         l = strlen(symname) + 1;
1609
1610         credits = ocfs2_calc_symlink_credits(sb);
1611
1612         /* lock the parent directory */
1613         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1614         if (status < 0) {
1615                 if (status != -ENOENT)
1616                         mlog_errno(status);
1617                 return status;
1618         }
1619
1620         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1621         if (!ocfs2_read_links_count(dirfe)) {
1622                 /* can't make a file in a deleted directory. */
1623                 status = -ENOENT;
1624                 goto bail;
1625         }
1626
1627         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1628                                            dentry->d_name.len);
1629         if (status)
1630                 goto bail;
1631
1632         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1633                                               dentry->d_name.name,
1634                                               dentry->d_name.len, &lookup);
1635         if (status < 0) {
1636                 mlog_errno(status);
1637                 goto bail;
1638         }
1639
1640         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1641         if (status < 0) {
1642                 if (status != -ENOSPC)
1643                         mlog_errno(status);
1644                 goto bail;
1645         }
1646
1647         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1648         if (!inode) {
1649                 status = -ENOMEM;
1650                 mlog_errno(status);
1651                 goto bail;
1652         }
1653
1654         /* get security xattr */
1655         status = ocfs2_init_security_get(inode, dir, &si);
1656         if (status) {
1657                 if (status == -EOPNOTSUPP)
1658                         si.enable = 0;
1659                 else {
1660                         mlog_errno(status);
1661                         goto bail;
1662                 }
1663         }
1664
1665         /* calculate meta data/clusters for setting security xattr */
1666         if (si.enable) {
1667                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1668                                                   &xattr_credits, &xattr_ac);
1669                 if (status < 0) {
1670                         mlog_errno(status);
1671                         goto bail;
1672                 }
1673         }
1674
1675         /* don't reserve bitmap space for fast symlinks. */
1676         if (l > ocfs2_fast_symlink_chars(sb))
1677                 want_clusters += 1;
1678
1679         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1680         if (status < 0) {
1681                 if (status != -ENOSPC)
1682                         mlog_errno(status);
1683                 goto bail;
1684         }
1685
1686         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1687         if (IS_ERR(handle)) {
1688                 status = PTR_ERR(handle);
1689                 handle = NULL;
1690                 mlog_errno(status);
1691                 goto bail;
1692         }
1693
1694         /* Starting to change things, restart is no longer possible. */
1695         ocfs2_block_signals(&oldset);
1696         did_block_signals = 1;
1697
1698         status = dquot_alloc_inode(inode);
1699         if (status)
1700                 goto bail;
1701         did_quota_inode = 1;
1702
1703         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1704                    inode->i_mode, dentry->d_name.len,
1705                    dentry->d_name.name);
1706
1707         status = ocfs2_mknod_locked(osb, dir, inode,
1708                                     0, &new_fe_bh, parent_fe_bh, handle,
1709                                     inode_ac);
1710         if (status < 0) {
1711                 mlog_errno(status);
1712                 goto bail;
1713         }
1714
1715         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1716         inode->i_rdev = 0;
1717         newsize = l - 1;
1718         if (l > ocfs2_fast_symlink_chars(sb)) {
1719                 u32 offset = 0;
1720
1721                 inode->i_op = &ocfs2_symlink_inode_operations;
1722                 status = dquot_alloc_space_nodirty(inode,
1723                     ocfs2_clusters_to_bytes(osb->sb, 1));
1724                 if (status)
1725                         goto bail;
1726                 did_quota = 1;
1727                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1728                                               new_fe_bh,
1729                                               handle, data_ac, NULL,
1730                                               NULL);
1731                 if (status < 0) {
1732                         if (status != -ENOSPC && status != -EINTR) {
1733                                 mlog(ML_ERROR,
1734                                      "Failed to extend file to %llu\n",
1735                                      (unsigned long long)newsize);
1736                                 mlog_errno(status);
1737                                 status = -ENOSPC;
1738                         }
1739                         goto bail;
1740                 }
1741                 i_size_write(inode, newsize);
1742                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1743         } else {
1744                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1745                 memcpy((char *) fe->id2.i_symlink, symname, l);
1746                 i_size_write(inode, newsize);
1747                 inode->i_blocks = 0;
1748         }
1749
1750         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1751         if (status < 0) {
1752                 mlog_errno(status);
1753                 goto bail;
1754         }
1755
1756         if (!ocfs2_inode_is_fast_symlink(inode)) {
1757                 status = ocfs2_create_symlink_data(osb, handle, inode,
1758                                                    symname);
1759                 if (status < 0) {
1760                         mlog_errno(status);
1761                         goto bail;
1762                 }
1763         }
1764
1765         if (si.enable) {
1766                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1767                                                  xattr_ac, data_ac);
1768                 if (status < 0) {
1769                         mlog_errno(status);
1770                         goto bail;
1771                 }
1772         }
1773
1774         /*
1775          * Do this before adding the entry to the directory. We add
1776          * also set d_op after success so that ->d_iput() will cleanup
1777          * the dentry lock even if ocfs2_add_entry() fails below.
1778          */
1779         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1780         if (status) {
1781                 mlog_errno(status);
1782                 goto bail;
1783         }
1784         dentry->d_op = &ocfs2_dentry_ops;
1785
1786         status = ocfs2_add_entry(handle, dentry, inode,
1787                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1788                                  &lookup);
1789         if (status < 0) {
1790                 mlog_errno(status);
1791                 goto bail;
1792         }
1793
1794         insert_inode_hash(inode);
1795         d_instantiate(dentry, inode);
1796 bail:
1797         if (status < 0 && did_quota)
1798                 dquot_free_space_nodirty(inode,
1799                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1800         if (status < 0 && did_quota_inode)
1801                 dquot_free_inode(inode);
1802         if (handle)
1803                 ocfs2_commit_trans(osb, handle);
1804
1805         ocfs2_inode_unlock(dir, 1);
1806         if (did_block_signals)
1807                 ocfs2_unblock_signals(&oldset);
1808
1809         brelse(new_fe_bh);
1810         brelse(parent_fe_bh);
1811         kfree(si.name);
1812         kfree(si.value);
1813         ocfs2_free_dir_lookup_result(&lookup);
1814         if (inode_ac)
1815                 ocfs2_free_alloc_context(inode_ac);
1816         if (data_ac)
1817                 ocfs2_free_alloc_context(data_ac);
1818         if (xattr_ac)
1819                 ocfs2_free_alloc_context(xattr_ac);
1820         if ((status < 0) && inode) {
1821                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1822                 clear_nlink(inode);
1823                 iput(inode);
1824         }
1825
1826         mlog_exit(status);
1827
1828         return status;
1829 }
1830
1831 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1832 {
1833         int status, namelen;
1834
1835         mlog_entry_void();
1836
1837         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1838                            (long long)blkno);
1839         if (namelen <= 0) {
1840                 if (namelen)
1841                         status = namelen;
1842                 else
1843                         status = -EINVAL;
1844                 mlog_errno(status);
1845                 goto bail;
1846         }
1847         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1848                 status = -EINVAL;
1849                 mlog_errno(status);
1850                 goto bail;
1851         }
1852
1853         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1854              namelen);
1855
1856         status = 0;
1857 bail:
1858         mlog_exit(status);
1859         return status;
1860 }
1861
1862 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1863                                     struct inode **ret_orphan_dir,
1864                                     u64 blkno,
1865                                     char *name,
1866                                     struct ocfs2_dir_lookup_result *lookup)
1867 {
1868         struct inode *orphan_dir_inode;
1869         struct buffer_head *orphan_dir_bh = NULL;
1870         int status = 0;
1871
1872         status = ocfs2_blkno_stringify(blkno, name);
1873         if (status < 0) {
1874                 mlog_errno(status);
1875                 return status;
1876         }
1877
1878         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1879                                                        ORPHAN_DIR_SYSTEM_INODE,
1880                                                        osb->slot_num);
1881         if (!orphan_dir_inode) {
1882                 status = -ENOENT;
1883                 mlog_errno(status);
1884                 return status;
1885         }
1886
1887         mutex_lock(&orphan_dir_inode->i_mutex);
1888
1889         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1890         if (status < 0) {
1891                 mlog_errno(status);
1892                 goto leave;
1893         }
1894
1895         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1896                                               orphan_dir_bh, name,
1897                                               OCFS2_ORPHAN_NAMELEN, lookup);
1898         if (status < 0) {
1899                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1900
1901                 mlog_errno(status);
1902                 goto leave;
1903         }
1904
1905         *ret_orphan_dir = orphan_dir_inode;
1906
1907 leave:
1908         if (status) {
1909                 mutex_unlock(&orphan_dir_inode->i_mutex);
1910                 iput(orphan_dir_inode);
1911         }
1912
1913         brelse(orphan_dir_bh);
1914
1915         mlog_exit(status);
1916         return status;
1917 }
1918
1919 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1920                             handle_t *handle,
1921                             struct inode *inode,
1922                             struct buffer_head *fe_bh,
1923                             char *name,
1924                             struct ocfs2_dir_lookup_result *lookup,
1925                             struct inode *orphan_dir_inode)
1926 {
1927         struct buffer_head *orphan_dir_bh = NULL;
1928         int status = 0;
1929         struct ocfs2_dinode *orphan_fe;
1930         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1931
1932         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1933
1934         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1935         if (status < 0) {
1936                 mlog_errno(status);
1937                 goto leave;
1938         }
1939
1940         status = ocfs2_journal_access_di(handle,
1941                                          INODE_CACHE(orphan_dir_inode),
1942                                          orphan_dir_bh,
1943                                          OCFS2_JOURNAL_ACCESS_WRITE);
1944         if (status < 0) {
1945                 mlog_errno(status);
1946                 goto leave;
1947         }
1948
1949         /* we're a cluster, and nlink can change on disk from
1950          * underneath us... */
1951         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1952         if (S_ISDIR(inode->i_mode))
1953                 ocfs2_add_links_count(orphan_fe, 1);
1954         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1955         ocfs2_journal_dirty(handle, orphan_dir_bh);
1956
1957         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1958                                    OCFS2_ORPHAN_NAMELEN, inode,
1959                                    OCFS2_I(inode)->ip_blkno,
1960                                    orphan_dir_bh, lookup);
1961         if (status < 0) {
1962                 mlog_errno(status);
1963                 goto leave;
1964         }
1965
1966         /*
1967          * We're going to journal the change of i_flags and i_orphaned_slot.
1968          * It's safe anyway, though some callers may duplicate the journaling.
1969          * Journaling within the func just make the logic look more
1970          * straightforward.
1971          */
1972         status = ocfs2_journal_access_di(handle,
1973                                          INODE_CACHE(inode),
1974                                          fe_bh,
1975                                          OCFS2_JOURNAL_ACCESS_WRITE);
1976         if (status < 0) {
1977                 mlog_errno(status);
1978                 goto leave;
1979         }
1980
1981         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1982         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
1983
1984         /* Record which orphan dir our inode now resides
1985          * in. delete_inode will use this to determine which orphan
1986          * dir to lock. */
1987         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1988
1989         ocfs2_journal_dirty(handle, fe_bh);
1990
1991         mlog(0, "Inode %llu orphaned in slot %d\n",
1992              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1993
1994 leave:
1995         brelse(orphan_dir_bh);
1996
1997         mlog_exit(status);
1998         return status;
1999 }
2000
2001 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2002 int ocfs2_orphan_del(struct ocfs2_super *osb,
2003                      handle_t *handle,
2004                      struct inode *orphan_dir_inode,
2005                      struct inode *inode,
2006                      struct buffer_head *orphan_dir_bh)
2007 {
2008         char name[OCFS2_ORPHAN_NAMELEN + 1];
2009         struct ocfs2_dinode *orphan_fe;
2010         int status = 0;
2011         struct ocfs2_dir_lookup_result lookup = { NULL, };
2012
2013         mlog_entry_void();
2014
2015         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2016         if (status < 0) {
2017                 mlog_errno(status);
2018                 goto leave;
2019         }
2020
2021         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2022              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2023              OCFS2_ORPHAN_NAMELEN);
2024
2025         /* find it's spot in the orphan directory */
2026         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2027                                   &lookup);
2028         if (status) {
2029                 mlog_errno(status);
2030                 goto leave;
2031         }
2032
2033         /* remove it from the orphan directory */
2034         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2035         if (status < 0) {
2036                 mlog_errno(status);
2037                 goto leave;
2038         }
2039
2040         status = ocfs2_journal_access_di(handle,
2041                                          INODE_CACHE(orphan_dir_inode),
2042                                          orphan_dir_bh,
2043                                          OCFS2_JOURNAL_ACCESS_WRITE);
2044         if (status < 0) {
2045                 mlog_errno(status);
2046                 goto leave;
2047         }
2048
2049         /* do the i_nlink dance! :) */
2050         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2051         if (S_ISDIR(inode->i_mode))
2052                 ocfs2_add_links_count(orphan_fe, -1);
2053         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2054         ocfs2_journal_dirty(handle, orphan_dir_bh);
2055
2056 leave:
2057         ocfs2_free_dir_lookup_result(&lookup);
2058
2059         mlog_exit(status);
2060         return status;
2061 }
2062
2063 int ocfs2_create_inode_in_orphan(struct inode *dir,
2064                                  int mode,
2065                                  struct inode **new_inode)
2066 {
2067         int status, did_quota_inode = 0;
2068         struct inode *inode = NULL;
2069         struct inode *orphan_dir = NULL;
2070         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2071         struct ocfs2_dinode *di = NULL;
2072         handle_t *handle = NULL;
2073         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2074         struct buffer_head *parent_di_bh = NULL;
2075         struct buffer_head *new_di_bh = NULL;
2076         struct ocfs2_alloc_context *inode_ac = NULL;
2077         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2078
2079         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2080         if (status < 0) {
2081                 if (status != -ENOENT)
2082                         mlog_errno(status);
2083                 return status;
2084         }
2085
2086         /*
2087          * We give the orphan dir the root blkno to fake an orphan name,
2088          * and allocate enough space for our insertion.
2089          */
2090         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
2091                                           osb->root_blkno,
2092                                           orphan_name, &orphan_insert);
2093         if (status < 0) {
2094                 mlog_errno(status);
2095                 goto leave;
2096         }
2097
2098         /* reserve an inode spot */
2099         status = ocfs2_reserve_new_inode(osb, &inode_ac);
2100         if (status < 0) {
2101                 if (status != -ENOSPC)
2102                         mlog_errno(status);
2103                 goto leave;
2104         }
2105
2106         inode = ocfs2_get_init_inode(dir, mode);
2107         if (!inode) {
2108                 status = -ENOMEM;
2109                 mlog_errno(status);
2110                 goto leave;
2111         }
2112
2113         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2114         if (IS_ERR(handle)) {
2115                 status = PTR_ERR(handle);
2116                 handle = NULL;
2117                 mlog_errno(status);
2118                 goto leave;
2119         }
2120
2121         status = dquot_alloc_inode(inode);
2122         if (status)
2123                 goto leave;
2124         did_quota_inode = 1;
2125
2126         inode->i_nlink = 0;
2127         /* do the real work now. */
2128         status = ocfs2_mknod_locked(osb, dir, inode,
2129                                     0, &new_di_bh, parent_di_bh, handle,
2130                                     inode_ac);
2131         if (status < 0) {
2132                 mlog_errno(status);
2133                 goto leave;
2134         }
2135
2136         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
2137         if (status < 0) {
2138                 mlog_errno(status);
2139                 goto leave;
2140         }
2141
2142         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2143         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2144                                   &orphan_insert, orphan_dir);
2145         if (status < 0) {
2146                 mlog_errno(status);
2147                 goto leave;
2148         }
2149
2150         /* get open lock so that only nodes can't remove it from orphan dir. */
2151         status = ocfs2_open_lock(inode);
2152         if (status < 0)
2153                 mlog_errno(status);
2154
2155         insert_inode_hash(inode);
2156 leave:
2157         if (status < 0 && did_quota_inode)
2158                 dquot_free_inode(inode);
2159         if (handle)
2160                 ocfs2_commit_trans(osb, handle);
2161
2162         if (orphan_dir) {
2163                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2164                 ocfs2_inode_unlock(orphan_dir, 1);
2165                 mutex_unlock(&orphan_dir->i_mutex);
2166                 iput(orphan_dir);
2167         }
2168
2169         if (status == -ENOSPC)
2170                 mlog(0, "Disk is full\n");
2171
2172         if ((status < 0) && inode) {
2173                 clear_nlink(inode);
2174                 iput(inode);
2175         }
2176
2177         if (inode_ac)
2178                 ocfs2_free_alloc_context(inode_ac);
2179
2180         brelse(new_di_bh);
2181
2182         if (!status)
2183                 *new_inode = inode;
2184
2185         ocfs2_free_dir_lookup_result(&orphan_insert);
2186
2187         ocfs2_inode_unlock(dir, 1);
2188         brelse(parent_di_bh);
2189         return status;
2190 }
2191
2192 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2193                                    struct inode *inode,
2194                                    struct dentry *dentry)
2195 {
2196         int status = 0;
2197         struct buffer_head *parent_di_bh = NULL;
2198         handle_t *handle = NULL;
2199         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2200         struct ocfs2_dinode *dir_di, *di;
2201         struct inode *orphan_dir_inode = NULL;
2202         struct buffer_head *orphan_dir_bh = NULL;
2203         struct buffer_head *di_bh = NULL;
2204         struct ocfs2_dir_lookup_result lookup = { NULL, };
2205
2206         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2207                    dentry->d_name.len, dentry->d_name.name);
2208
2209         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2210         if (status < 0) {
2211                 if (status != -ENOENT)
2212                         mlog_errno(status);
2213                 return status;
2214         }
2215
2216         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2217         if (!dir_di->i_links_count) {
2218                 /* can't make a file in a deleted directory. */
2219                 status = -ENOENT;
2220                 goto leave;
2221         }
2222
2223         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2224                                            dentry->d_name.len);
2225         if (status)
2226                 goto leave;
2227
2228         /* get a spot inside the dir. */
2229         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2230                                               dentry->d_name.name,
2231                                               dentry->d_name.len, &lookup);
2232         if (status < 0) {
2233                 mlog_errno(status);
2234                 goto leave;
2235         }
2236
2237         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2238                                                        ORPHAN_DIR_SYSTEM_INODE,
2239                                                        osb->slot_num);
2240         if (!orphan_dir_inode) {
2241                 status = -EEXIST;
2242                 mlog_errno(status);
2243                 goto leave;
2244         }
2245
2246         mutex_lock(&orphan_dir_inode->i_mutex);
2247
2248         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2249         if (status < 0) {
2250                 mlog_errno(status);
2251                 mutex_unlock(&orphan_dir_inode->i_mutex);
2252                 iput(orphan_dir_inode);
2253                 goto leave;
2254         }
2255
2256         status = ocfs2_read_inode_block(inode, &di_bh);
2257         if (status < 0) {
2258                 mlog_errno(status);
2259                 goto orphan_unlock;
2260         }
2261
2262         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2263         if (IS_ERR(handle)) {
2264                 status = PTR_ERR(handle);
2265                 handle = NULL;
2266                 mlog_errno(status);
2267                 goto orphan_unlock;
2268         }
2269
2270         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2271                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2272         if (status < 0) {
2273                 mlog_errno(status);
2274                 goto out_commit;
2275         }
2276
2277         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2278                                   orphan_dir_bh);
2279         if (status < 0) {
2280                 mlog_errno(status);
2281                 goto out_commit;
2282         }
2283
2284         di = (struct ocfs2_dinode *)di_bh->b_data;
2285         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2286         di->i_orphaned_slot = 0;
2287         inode->i_nlink = 1;
2288         ocfs2_set_links_count(di, inode->i_nlink);
2289         ocfs2_journal_dirty(handle, di_bh);
2290
2291         status = ocfs2_add_entry(handle, dentry, inode,
2292                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2293                                  &lookup);
2294         if (status < 0) {
2295                 mlog_errno(status);
2296                 goto out_commit;
2297         }
2298
2299         status = ocfs2_dentry_attach_lock(dentry, inode,
2300                                           OCFS2_I(dir)->ip_blkno);
2301         if (status) {
2302                 mlog_errno(status);
2303                 goto out_commit;
2304         }
2305
2306         dentry->d_op = &ocfs2_dentry_ops;
2307         d_instantiate(dentry, inode);
2308         status = 0;
2309 out_commit:
2310         ocfs2_commit_trans(osb, handle);
2311 orphan_unlock:
2312         ocfs2_inode_unlock(orphan_dir_inode, 1);
2313         mutex_unlock(&orphan_dir_inode->i_mutex);
2314         iput(orphan_dir_inode);
2315 leave:
2316
2317         ocfs2_inode_unlock(dir, 1);
2318
2319         brelse(di_bh);
2320         brelse(parent_di_bh);
2321         brelse(orphan_dir_bh);
2322
2323         ocfs2_free_dir_lookup_result(&lookup);
2324
2325         mlog_exit(status);
2326
2327         return status;
2328 }
2329
2330 const struct inode_operations ocfs2_dir_iops = {
2331         .create         = ocfs2_create,
2332         .lookup         = ocfs2_lookup,
2333         .link           = ocfs2_link,
2334         .unlink         = ocfs2_unlink,
2335         .rmdir          = ocfs2_unlink,
2336         .symlink        = ocfs2_symlink,
2337         .mkdir          = ocfs2_mkdir,
2338         .mknod          = ocfs2_mknod,
2339         .rename         = ocfs2_rename,
2340         .setattr        = ocfs2_setattr,
2341         .getattr        = ocfs2_getattr,
2342         .permission     = ocfs2_permission,
2343         .setxattr       = generic_setxattr,
2344         .getxattr       = generic_getxattr,
2345         .listxattr      = ocfs2_listxattr,
2346         .removexattr    = generic_removexattr,
2347         .fiemap         = ocfs2_fiemap,
2348 };