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