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

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/quotaops.h>
  44
  45 #define MLOG_MASK_PREFIX ML_NAMEI
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 #include "alloc.h"
  51 #include "dcache.h"
  52 #include "dir.h"
  53 #include "dlmglue.h"
  54 #include "extent_map.h"
  55 #include "file.h"
  56 #include "inode.h"
  57 #include "journal.h"
  58 #include "namei.h"
  59 #include "suballoc.h"
  60 #include "super.h"
  61 #include "symlink.h"
  62 #include "sysfile.h"
  63 #include "uptodate.h"
  64 #include "xattr.h"
  65 #include "acl.h"
  66
  67 #include "buffer_head_io.h"
  68
  69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  70                               struct inode *dir,
  71                               struct inode *inode,
  72                               dev_t dev,
  73                               struct buffer_head **new_fe_bh,
  74                               struct buffer_head *parent_fe_bh,
  75                               handle_t *handle,
  76                               struct ocfs2_alloc_context *inode_ac);
  77
  78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  79                                     struct inode **ret_orphan_dir,
  80                                     u64 blkno,
  81                                     char *name,
  82                                     struct ocfs2_dir_lookup_result *lookup);
  83
  84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  85                             handle_t *handle,
  86                             struct inode *inode,
  87                             struct ocfs2_dinode *fe,
  88                             char *name,
  89                             struct ocfs2_dir_lookup_result *lookup,
  90                             struct inode *orphan_dir_inode);
  91
  92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  93                                      handle_t *handle,
  94                                      struct inode *inode,
  95                                      const char *symname);
  96
  97 /* An orphan dir name is an 8 byte value, printed as a hex string */
  98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  99
 100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 101                                    struct nameidata *nd)
 102 {
 103         int status;
 104         u64 blkno;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_inode_info *oi;
 108
 109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 110                    dentry->d_name.len, dentry->d_name.name);
 111
 112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 113                 ret = ERR_PTR(-ENAMETOOLONG);
 114                 goto bail;
 115         }
 116
 117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 119
 120         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 121         if (status < 0) {
 122                 if (status != -ENOENT)
 123                         mlog_errno(status);
 124                 ret = ERR_PTR(status);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 129                                             dentry->d_name.len, &blkno);
 130         if (status < 0)
 131                 goto bail_add;
 132
 133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 134         if (IS_ERR(inode)) {
 135                 ret = ERR_PTR(-EACCES);
 136                 goto bail_unlock;
 137         }
 138
 139         oi = OCFS2_I(inode);
 140         /* Clear any orphaned state... If we were able to look up the
 141          * inode from a directory, it certainly can't be orphaned. We
 142          * might have the bad state from a node which intended to
 143          * orphan this inode but crashed before it could commit the
 144          * unlink. */
 145         spin_lock(&oi->ip_lock);
 146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 147         spin_unlock(&oi->ip_lock);
 148
 149 bail_add:
 150         dentry->d_op = &ocfs2_dentry_ops;
 151         ret = d_splice_alias(inode, dentry);
 152
 153         if (inode) {
 154                 /*
 155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 156                  * dentry, it will d_move() it on top of ourse. The
 157                  * return value will indicate this however, so in
 158                  * those cases, we switch them around for the locking
 159                  * code.
 160                  *
 161                  * NOTE: This dentry already has ->d_op set from
 162                  * ocfs2_get_parent() and ocfs2_get_dentry()
 163                  */
 164                 if (ret)
 165                         dentry = ret;
 166
 167                 status = ocfs2_dentry_attach_lock(dentry, inode,
 168                                                   OCFS2_I(dir)->ip_blkno);
 169                 if (status) {
 170                         mlog_errno(status);
 171                         ret = ERR_PTR(status);
 172                         goto bail_unlock;
 173                 }
 174         }
 175
 176 bail_unlock:
 177         /* Don't drop the cluster lock until *after* the d_add --
 178          * unlink on another node will message us to remove that
 179          * dentry under this lock so otherwise we can race this with
 180          * the downconvert thread and have a stale dentry. */
 181         ocfs2_inode_unlock(dir, 0);
 182
 183 bail:
 184
 185         mlog_exit_ptr(ret);
 186
 187         return ret;
 188 }
 189
 190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
 191 {
 192         struct inode *inode;
 193
 194         inode = new_inode(dir->i_sb);
 195         if (!inode) {
 196                 mlog(ML_ERROR, "new_inode failed!\n");
 197                 return NULL;
 198         }
 199
 200         /* populate as many fields early on as possible - many of
 201          * these are used by the support functions here and in
 202          * callers. */
 203         if (S_ISDIR(mode))
 204                 inode->i_nlink = 2;
 205         else
 206                 inode->i_nlink = 1;
 207         inode->i_uid = current_fsuid();
 208         if (dir->i_mode & S_ISGID) {
 209                 inode->i_gid = dir->i_gid;
 210                 if (S_ISDIR(mode))
 211                         mode |= S_ISGID;
 212         } else
 213                 inode->i_gid = current_fsgid();
 214         inode->i_mode = mode;
 215         dquot_initialize(inode);
 216         return inode;
 217 }
 218
 219 static int ocfs2_mknod(struct inode *dir,
 220                        struct dentry *dentry,
 221                        int mode,
 222                        dev_t dev)
 223 {
 224         int status = 0;
 225         struct buffer_head *parent_fe_bh = NULL;
 226         handle_t *handle = NULL;
 227         struct ocfs2_super *osb;
 228         struct ocfs2_dinode *dirfe;
 229         struct buffer_head *new_fe_bh = NULL;
 230         struct inode *inode = NULL;
 231         struct ocfs2_alloc_context *inode_ac = NULL;
 232         struct ocfs2_alloc_context *data_ac = NULL;
 233         struct ocfs2_alloc_context *meta_ac = NULL;
 234         int want_clusters = 0;
 235         int want_meta = 0;
 236         int xattr_credits = 0;
 237         struct ocfs2_security_xattr_info si = {
 238                 .enable = 1,
 239         };
 240         int did_quota_inode = 0;
 241         struct ocfs2_dir_lookup_result lookup = { NULL, };
 242
 243         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 244                    (unsigned long)dev, dentry->d_name.len,
 245                    dentry->d_name.name);
 246
 247         dquot_initialize(dir);
 248
 249         /* get our super block */
 250         osb = OCFS2_SB(dir->i_sb);
 251
 252         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 253         if (status < 0) {
 254                 if (status != -ENOENT)
 255                         mlog_errno(status);
 256                 return status;
 257         }
 258
 259         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 260                 status = -EMLINK;
 261                 goto leave;
 262         }
 263
 264         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 265         if (!ocfs2_read_links_count(dirfe)) {
 266                 /* can't make a file in a deleted directory. */
 267                 status = -ENOENT;
 268                 goto leave;
 269         }
 270
 271         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 272                                            dentry->d_name.len);
 273         if (status)
 274                 goto leave;
 275
 276         /* get a spot inside the dir. */
 277         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 278                                               dentry->d_name.name,
 279                                               dentry->d_name.len, &lookup);
 280         if (status < 0) {
 281                 mlog_errno(status);
 282                 goto leave;
 283         }
 284
 285         /* reserve an inode spot */
 286         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 287         if (status < 0) {
 288                 if (status != -ENOSPC)
 289                         mlog_errno(status);
 290                 goto leave;
 291         }
 292
 293         inode = ocfs2_get_init_inode(dir, mode);
 294         if (!inode) {
 295                 status = -ENOMEM;
 296                 mlog_errno(status);
 297                 goto leave;
 298         }
 299
 300         /* get security xattr */
 301         status = ocfs2_init_security_get(inode, dir, &si);
 302         if (status) {
 303                 if (status == -EOPNOTSUPP)
 304                         si.enable = 0;
 305                 else {
 306                         mlog_errno(status);
 307                         goto leave;
 308                 }
 309         }
 310
 311         /* calculate meta data/clusters for setting security and acl xattr */
 312         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 313                                        &si, &want_clusters,
 314                                        &xattr_credits, &want_meta);
 315         if (status < 0) {
 316                 mlog_errno(status);
 317                 goto leave;
 318         }
 319
 320         /* Reserve a cluster if creating an extent based directory. */
 321         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 322                 want_clusters += 1;
 323
 324                 /* Dir indexing requires extra space as well */
 325                 if (ocfs2_supports_indexed_dirs(osb))
 326                         want_meta++;
 327         }
 328
 329         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 330         if (status < 0) {
 331                 if (status != -ENOSPC)
 332                         mlog_errno(status);
 333                 goto leave;
 334         }
 335
 336         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 337         if (status < 0) {
 338                 if (status != -ENOSPC)
 339                         mlog_errno(status);
 340                 goto leave;
 341         }
 342
 343         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 344                                                             S_ISDIR(mode),
 345                                                             xattr_credits));
 346         if (IS_ERR(handle)) {
 347                 status = PTR_ERR(handle);
 348                 handle = NULL;
 349                 mlog_errno(status);
 350                 goto leave;
 351         }
 352
 353         status = dquot_alloc_inode(inode);
 354         if (status)
 355                 goto leave;
 356         did_quota_inode = 1;
 357
 358         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
 359                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
 360                    dentry->d_name.name);
 361
 362         /* do the real work now. */
 363         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 364                                     &new_fe_bh, parent_fe_bh, handle,
 365                                     inode_ac);
 366         if (status < 0) {
 367                 mlog_errno(status);
 368                 goto leave;
 369         }
 370
 371         if (S_ISDIR(mode)) {
 372                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 373                                             new_fe_bh, data_ac, meta_ac);
 374                 if (status < 0) {
 375                         mlog_errno(status);
 376                         goto leave;
 377                 }
 378
 379                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 380                                                  parent_fe_bh,
 381                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 382                 if (status < 0) {
 383                         mlog_errno(status);
 384                         goto leave;
 385                 }
 386                 ocfs2_add_links_count(dirfe, 1);
 387                 ocfs2_journal_dirty(handle, parent_fe_bh);
 388                 inc_nlink(dir);
 389         }
 390
 391         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 392                                 meta_ac, data_ac);
 393         if (status < 0) {
 394                 mlog_errno(status);
 395                 goto leave;
 396         }
 397
 398         if (si.enable) {
 399                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 400                                                  meta_ac, data_ac);
 401                 if (status < 0) {
 402                         mlog_errno(status);
 403                         goto leave;
 404                 }
 405         }
 406
 407         status = ocfs2_add_entry(handle, dentry, inode,
 408                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 409                                  &lookup);
 410         if (status < 0) {
 411                 mlog_errno(status);
 412                 goto leave;
 413         }
 414
 415         status = ocfs2_dentry_attach_lock(dentry, inode,
 416                                           OCFS2_I(dir)->ip_blkno);
 417         if (status) {
 418                 mlog_errno(status);
 419                 goto leave;
 420         }
 421
 422         insert_inode_hash(inode);
 423         dentry->d_op = &ocfs2_dentry_ops;
 424         d_instantiate(dentry, inode);
 425         status = 0;
 426 leave:
 427         if (status < 0 && did_quota_inode)
 428                 dquot_free_inode(inode);
 429         if (handle)
 430                 ocfs2_commit_trans(osb, handle);
 431
 432         ocfs2_inode_unlock(dir, 1);
 433
 434         if (status == -ENOSPC)
 435                 mlog(0, "Disk is full\n");
 436
 437         brelse(new_fe_bh);
 438         brelse(parent_fe_bh);
 439         kfree(si.name);
 440         kfree(si.value);
 441
 442         ocfs2_free_dir_lookup_result(&lookup);
 443
 444         if ((status < 0) && inode) {
 445                 clear_nlink(inode);
 446                 iput(inode);
 447         }
 448
 449         if (inode_ac)
 450                 ocfs2_free_alloc_context(inode_ac);
 451
 452         if (data_ac)
 453                 ocfs2_free_alloc_context(data_ac);
 454
 455         if (meta_ac)
 456                 ocfs2_free_alloc_context(meta_ac);
 457
 458         mlog_exit(status);
 459
 460         return status;
 461 }
 462
 463 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 464                               struct inode *dir,
 465                               struct inode *inode,
 466                               dev_t dev,
 467                               struct buffer_head **new_fe_bh,
 468                               struct buffer_head *parent_fe_bh,
 469                               handle_t *handle,
 470                               struct ocfs2_alloc_context *inode_ac)
 471 {
 472         int status = 0;
 473         struct ocfs2_dinode *fe = NULL;
 474         struct ocfs2_extent_list *fel;
 475         u64 suballoc_loc, fe_blkno = 0;
 476         u16 suballoc_bit;
 477         u16 feat;
 478
 479         *new_fe_bh = NULL;
 480
 481         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 482                                        inode_ac, &suballoc_loc,
 483                                        &suballoc_bit, &fe_blkno);
 484         if (status < 0) {
 485                 mlog_errno(status);
 486                 goto leave;
 487         }
 488
 489         /* populate as many fields early on as possible - many of
 490          * these are used by the support functions here and in
 491          * callers. */
 492         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 493         OCFS2_I(inode)->ip_blkno = fe_blkno;
 494         spin_lock(&osb->osb_lock);
 495         inode->i_generation = osb->s_next_generation++;
 496         spin_unlock(&osb->osb_lock);
 497
 498         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 499         if (!*new_fe_bh) {
 500                 status = -EIO;
 501                 mlog_errno(status);
 502                 goto leave;
 503         }
 504         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 505
 506         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 507                                          *new_fe_bh,
 508                                          OCFS2_JOURNAL_ACCESS_CREATE);
 509         if (status < 0) {
 510                 mlog_errno(status);
 511                 goto leave;
 512         }
 513
 514         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 515         memset(fe, 0, osb->sb->s_blocksize);
 516
 517         fe->i_generation = cpu_to_le32(inode->i_generation);
 518         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 519         fe->i_blkno = cpu_to_le64(fe_blkno);
 520         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 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         ocfs2_journal_dirty(handle, *new_fe_bh);
 558
 559         ocfs2_populate_inode(inode, fe, 1);
 560         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 561         if (!ocfs2_mount_local(osb)) {
 562                 status = ocfs2_create_new_inode_locks(inode);
 563                 if (status < 0)
 564                         mlog_errno(status);
 565         }
 566
 567         status = 0; /* error in ocfs2_create_new_inode_locks is not
 568                      * critical */
 569
 570 leave:
 571         if (status < 0) {
 572                 if (*new_fe_bh) {
 573                         brelse(*new_fe_bh);
 574                         *new_fe_bh = NULL;
 575                 }
 576         }
 577
 578         mlog_exit(status);
 579         return status;
 580 }
 581
 582 static int ocfs2_mkdir(struct inode *dir,
 583                        struct dentry *dentry,
 584                        int mode)
 585 {
 586         int ret;
 587
 588         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 589                    dentry->d_name.len, dentry->d_name.name);
 590         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 591         mlog_exit(ret);
 592
 593         return ret;
 594 }
 595
 596 static int ocfs2_create(struct inode *dir,
 597                         struct dentry *dentry,
 598                         int mode,
 599                         struct nameidata *nd)
 600 {
 601         int ret;
 602
 603         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 604                    dentry->d_name.len, dentry->d_name.name);
 605         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 606         mlog_exit(ret);
 607
 608         return ret;
 609 }
 610
 611 static int ocfs2_link(struct dentry *old_dentry,
 612                       struct inode *dir,
 613                       struct dentry *dentry)
 614 {
 615         handle_t *handle;
 616         struct inode *inode = old_dentry->d_inode;
 617         int err;
 618         struct buffer_head *fe_bh = NULL;
 619         struct buffer_head *parent_fe_bh = NULL;
 620         struct ocfs2_dinode *fe = NULL;
 621         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 622         struct ocfs2_dir_lookup_result lookup = { NULL, };
 623
 624         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 625                    old_dentry->d_name.len, old_dentry->d_name.name,
 626                    dentry->d_name.len, dentry->d_name.name);
 627
 628         if (S_ISDIR(inode->i_mode))
 629                 return -EPERM;
 630
 631         dquot_initialize(dir);
 632
 633         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 634         if (err < 0) {
 635                 if (err != -ENOENT)
 636                         mlog_errno(err);
 637                 return err;
 638         }
 639
 640         if (!dir->i_nlink) {
 641                 err = -ENOENT;
 642                 goto out;
 643         }
 644
 645         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 646                                         dentry->d_name.len);
 647         if (err)
 648                 goto out;
 649
 650         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 651                                            dentry->d_name.name,
 652                                            dentry->d_name.len, &lookup);
 653         if (err < 0) {
 654                 mlog_errno(err);
 655                 goto out;
 656         }
 657
 658         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 659         if (err < 0) {
 660                 if (err != -ENOENT)
 661                         mlog_errno(err);
 662                 goto out;
 663         }
 664
 665         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 666         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 667                 err = -EMLINK;
 668                 goto out_unlock_inode;
 669         }
 670
 671         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 672         if (IS_ERR(handle)) {
 673                 err = PTR_ERR(handle);
 674                 handle = NULL;
 675                 mlog_errno(err);
 676                 goto out_unlock_inode;
 677         }
 678
 679         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 680                                       OCFS2_JOURNAL_ACCESS_WRITE);
 681         if (err < 0) {
 682                 mlog_errno(err);
 683                 goto out_commit;
 684         }
 685
 686         inc_nlink(inode);
 687         inode->i_ctime = CURRENT_TIME;
 688         ocfs2_set_links_count(fe, inode->i_nlink);
 689         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 690         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 691         ocfs2_journal_dirty(handle, fe_bh);
 692
 693         err = ocfs2_add_entry(handle, dentry, inode,
 694                               OCFS2_I(inode)->ip_blkno,
 695                               parent_fe_bh, &lookup);
 696         if (err) {
 697                 ocfs2_add_links_count(fe, -1);
 698                 drop_nlink(inode);
 699                 mlog_errno(err);
 700                 goto out_commit;
 701         }
 702
 703         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 704         if (err) {
 705                 mlog_errno(err);
 706                 goto out_commit;
 707         }
 708
 709         atomic_inc(&inode->i_count);
 710         dentry->d_op = &ocfs2_dentry_ops;
 711         d_instantiate(dentry, inode);
 712
 713 out_commit:
 714         ocfs2_commit_trans(osb, handle);
 715 out_unlock_inode:
 716         ocfs2_inode_unlock(inode, 1);
 717
 718 out:
 719         ocfs2_inode_unlock(dir, 1);
 720
 721         brelse(fe_bh);
 722         brelse(parent_fe_bh);
 723
 724         ocfs2_free_dir_lookup_result(&lookup);
 725
 726         mlog_exit(err);
 727
 728         return err;
 729 }
 730
 731 /*
 732  * Takes and drops an exclusive lock on the given dentry. This will
 733  * force other nodes to drop it.
 734  */
 735 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 736 {
 737         int ret;
 738
 739         ret = ocfs2_dentry_lock(dentry, 1);
 740         if (ret)
 741                 mlog_errno(ret);
 742         else
 743                 ocfs2_dentry_unlock(dentry, 1);
 744
 745         return ret;
 746 }
 747
 748 static inline int inode_is_unlinkable(struct inode *inode)
 749 {
 750         if (S_ISDIR(inode->i_mode)) {
 751                 if (inode->i_nlink == 2)
 752                         return 1;
 753                 return 0;
 754         }
 755
 756         if (inode->i_nlink == 1)
 757                 return 1;
 758         return 0;
 759 }
 760
 761 static int ocfs2_unlink(struct inode *dir,
 762                         struct dentry *dentry)
 763 {
 764         int status;
 765         int child_locked = 0;
 766         struct inode *inode = dentry->d_inode;
 767         struct inode *orphan_dir = NULL;
 768         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 769         u64 blkno;
 770         struct ocfs2_dinode *fe = NULL;
 771         struct buffer_head *fe_bh = NULL;
 772         struct buffer_head *parent_node_bh = NULL;
 773         handle_t *handle = NULL;
 774         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 775         struct ocfs2_dir_lookup_result lookup = { NULL, };
 776         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 777
 778         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 779                    dentry->d_name.len, dentry->d_name.name);
 780
 781         dquot_initialize(dir);
 782
 783         BUG_ON(dentry->d_parent->d_inode != dir);
 784
 785         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 786
 787         if (inode == osb->root_inode) {
 788                 mlog(0, "Cannot delete the root directory\n");
 789                 return -EPERM;
 790         }
 791
 792         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 793                                          OI_LS_PARENT);
 794         if (status < 0) {
 795                 if (status != -ENOENT)
 796                         mlog_errno(status);
 797                 return status;
 798         }
 799
 800         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 801                                           dentry->d_name.len, &blkno, dir,
 802                                           &lookup);
 803         if (status < 0) {
 804                 if (status != -ENOENT)
 805                         mlog_errno(status);
 806                 goto leave;
 807         }
 808
 809         if (OCFS2_I(inode)->ip_blkno != blkno) {
 810                 status = -ENOENT;
 811
 812                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 813                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 814                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 815                 goto leave;
 816         }
 817
 818         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 819         if (status < 0) {
 820                 if (status != -ENOENT)
 821                         mlog_errno(status);
 822                 goto leave;
 823         }
 824         child_locked = 1;
 825
 826         if (S_ISDIR(inode->i_mode)) {
 827                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 828                         status = -ENOTEMPTY;
 829                         goto leave;
 830                 }
 831         }
 832
 833         status = ocfs2_remote_dentry_delete(dentry);
 834         if (status < 0) {
 835                 /* This remote delete should succeed under all normal
 836                  * circumstances. */
 837                 mlog_errno(status);
 838                 goto leave;
 839         }
 840
 841         if (inode_is_unlinkable(inode)) {
 842                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 843                                                   OCFS2_I(inode)->ip_blkno,
 844                                                   orphan_name, &orphan_insert);
 845                 if (status < 0) {
 846                         mlog_errno(status);
 847                         goto leave;
 848                 }
 849         }
 850
 851         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 852         if (IS_ERR(handle)) {
 853                 status = PTR_ERR(handle);
 854                 handle = NULL;
 855                 mlog_errno(status);
 856                 goto leave;
 857         }
 858
 859         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 860                                          OCFS2_JOURNAL_ACCESS_WRITE);
 861         if (status < 0) {
 862                 mlog_errno(status);
 863                 goto leave;
 864         }
 865
 866         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 867
 868         if (inode_is_unlinkable(inode)) {
 869                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 870                                           &orphan_insert, orphan_dir);
 871                 if (status < 0) {
 872                         mlog_errno(status);
 873                         goto leave;
 874                 }
 875         }
 876
 877         /* delete the name from the parent dir */
 878         status = ocfs2_delete_entry(handle, dir, &lookup);
 879         if (status < 0) {
 880                 mlog_errno(status);
 881                 goto leave;
 882         }
 883
 884         if (S_ISDIR(inode->i_mode))
 885                 drop_nlink(inode);
 886         drop_nlink(inode);
 887         ocfs2_set_links_count(fe, inode->i_nlink);
 888         ocfs2_journal_dirty(handle, fe_bh);
 889
 890         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 891         if (S_ISDIR(inode->i_mode))
 892                 drop_nlink(dir);
 893
 894         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 895         if (status < 0) {
 896                 mlog_errno(status);
 897                 if (S_ISDIR(inode->i_mode))
 898                         inc_nlink(dir);
 899         }
 900
 901 leave:
 902         if (handle)
 903                 ocfs2_commit_trans(osb, handle);
 904
 905         if (child_locked)
 906                 ocfs2_inode_unlock(inode, 1);
 907
 908         ocfs2_inode_unlock(dir, 1);
 909
 910         if (orphan_dir) {
 911                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 912                 ocfs2_inode_unlock(orphan_dir, 1);
 913                 mutex_unlock(&orphan_dir->i_mutex);
 914                 iput(orphan_dir);
 915         }
 916
 917         brelse(fe_bh);
 918         brelse(parent_node_bh);
 919
 920         ocfs2_free_dir_lookup_result(&orphan_insert);
 921         ocfs2_free_dir_lookup_result(&lookup);
 922
 923         mlog_exit(status);
 924
 925         return status;
 926 }
 927
 928 /*
 929  * The only place this should be used is rename!
 930  * if they have the same id, then the 1st one is the only one locked.
 931  */
 932 static int ocfs2_double_lock(struct ocfs2_super *osb,
 933                              struct buffer_head **bh1,
 934                              struct inode *inode1,
 935                              struct buffer_head **bh2,
 936                              struct inode *inode2)
 937 {
 938         int status;
 939         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 940         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 941         struct buffer_head **tmpbh;
 942         struct inode *tmpinode;
 943
 944         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 945                    (unsigned long long)oi1->ip_blkno,
 946                    (unsigned long long)oi2->ip_blkno);
 947
 948         if (*bh1)
 949                 *bh1 = NULL;
 950         if (*bh2)
 951                 *bh2 = NULL;
 952
 953         /* we always want to lock the one with the lower lockid first. */
 954         if (oi1->ip_blkno != oi2->ip_blkno) {
 955                 if (oi1->ip_blkno < oi2->ip_blkno) {
 956                         /* switch id1 and id2 around */
 957                         mlog(0, "switching them around...\n");
 958                         tmpbh = bh2;
 959                         bh2 = bh1;
 960                         bh1 = tmpbh;
 961
 962                         tmpinode = inode2;
 963                         inode2 = inode1;
 964                         inode1 = tmpinode;
 965                 }
 966                 /* lock id2 */
 967                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 968                                                  OI_LS_RENAME1);
 969                 if (status < 0) {
 970                         if (status != -ENOENT)
 971                                 mlog_errno(status);
 972                         goto bail;
 973                 }
 974         }
 975
 976         /* lock id1 */
 977         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
 978         if (status < 0) {
 979                 /*
 980                  * An error return must mean that no cluster locks
 981                  * were held on function exit.
 982                  */
 983                 if (oi1->ip_blkno != oi2->ip_blkno)
 984                         ocfs2_inode_unlock(inode2, 1);
 985
 986                 if (status != -ENOENT)
 987                         mlog_errno(status);
 988         }
 989
 990 bail:
 991         mlog_exit(status);
 992         return status;
 993 }
 994
 995 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
 996 {
 997         ocfs2_inode_unlock(inode1, 1);
 998
 999         if (inode1 != inode2)
1000                 ocfs2_inode_unlock(inode2, 1);
1001 }
1002
1003 static int ocfs2_rename(struct inode *old_dir,
1004                         struct dentry *old_dentry,
1005                         struct inode *new_dir,
1006                         struct dentry *new_dentry)
1007 {
1008         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1009         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1010         struct inode *old_inode = old_dentry->d_inode;
1011         struct inode *new_inode = new_dentry->d_inode;
1012         struct inode *orphan_dir = NULL;
1013         struct ocfs2_dinode *newfe = NULL;
1014         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1015         struct buffer_head *newfe_bh = NULL;
1016         struct buffer_head *old_inode_bh = NULL;
1017         struct ocfs2_super *osb = NULL;
1018         u64 newfe_blkno, old_de_ino;
1019         handle_t *handle = NULL;
1020         struct buffer_head *old_dir_bh = NULL;
1021         struct buffer_head *new_dir_bh = NULL;
1022         nlink_t old_dir_nlink = old_dir->i_nlink;
1023         struct ocfs2_dinode *old_di;
1024         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1025         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1026         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1027         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1028         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1029
1030         /* At some point it might be nice to break this function up a
1031          * bit. */
1032
1033         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1034                    old_dir, old_dentry, new_dir, new_dentry,
1035                    old_dentry->d_name.len, old_dentry->d_name.name,
1036                    new_dentry->d_name.len, new_dentry->d_name.name);
1037
1038         dquot_initialize(old_dir);
1039         dquot_initialize(new_dir);
1040
1041         osb = OCFS2_SB(old_dir->i_sb);
1042
1043         if (new_inode) {
1044                 if (!igrab(new_inode))
1045                         BUG();
1046         }
1047
1048         /* Assume a directory hierarchy thusly:
1049          * a/b/c
1050          * a/d
1051          * a,b,c, and d are all directories.
1052          *
1053          * from cwd of 'a' on both nodes:
1054          * node1: mv b/c d
1055          * node2: mv d   b/c
1056          *
1057          * And that's why, just like the VFS, we need a file system
1058          * rename lock. */
1059         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1060                 status = ocfs2_rename_lock(osb);
1061                 if (status < 0) {
1062                         mlog_errno(status);
1063                         goto bail;
1064                 }
1065                 rename_lock = 1;
1066         }
1067
1068         /* if old and new are the same, this'll just do one lock. */
1069         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1070                                    &new_dir_bh, new_dir);
1071         if (status < 0) {
1072                 mlog_errno(status);
1073                 goto bail;
1074         }
1075         parents_locked = 1;
1076
1077         /* make sure both dirs have bhs
1078          * get an extra ref on old_dir_bh if old==new */
1079         if (!new_dir_bh) {
1080                 if (old_dir_bh) {
1081                         new_dir_bh = old_dir_bh;
1082                         get_bh(new_dir_bh);
1083                 } else {
1084                         mlog(ML_ERROR, "no old_dir_bh!\n");
1085                         status = -EIO;
1086                         goto bail;
1087                 }
1088         }
1089
1090         /*
1091          * Aside from allowing a meta data update, the locking here
1092          * also ensures that the downconvert thread on other nodes
1093          * won't have to concurrently downconvert the inode and the
1094          * dentry locks.
1095          */
1096         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1097                                          OI_LS_PARENT);
1098         if (status < 0) {
1099                 if (status != -ENOENT)
1100                         mlog_errno(status);
1101                 goto bail;
1102         }
1103         old_child_locked = 1;
1104
1105         status = ocfs2_remote_dentry_delete(old_dentry);
1106         if (status < 0) {
1107                 mlog_errno(status);
1108                 goto bail;
1109         }
1110
1111         if (S_ISDIR(old_inode->i_mode)) {
1112                 u64 old_inode_parent;
1113
1114                 update_dot_dot = 1;
1115                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1116                                                   old_inode,
1117                                                   &old_inode_dot_dot_res);
1118                 if (status) {
1119                         status = -EIO;
1120                         goto bail;
1121                 }
1122
1123                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1124                         status = -EIO;
1125                         goto bail;
1126                 }
1127
1128                 if (!new_inode && new_dir != old_dir &&
1129                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1130                         status = -EMLINK;
1131                         goto bail;
1132                 }
1133         }
1134
1135         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1136                                             old_dentry->d_name.len,
1137                                             &old_de_ino);
1138         if (status) {
1139                 status = -ENOENT;
1140                 goto bail;
1141         }
1142
1143         /*
1144          *  Check for inode number is _not_ due to possible IO errors.
1145          *  We might rmdir the source, keep it as pwd of some process
1146          *  and merrily kill the link to whatever was created under the
1147          *  same name. Goodbye sticky bit ;-<
1148          */
1149         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1150                 status = -ENOENT;
1151                 goto bail;
1152         }
1153
1154         /* check if the target already exists (in which case we need
1155          * to delete it */
1156         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1157                                           new_dentry->d_name.len,
1158                                           &newfe_blkno, new_dir,
1159                                           &target_lookup_res);
1160         /* The only error we allow here is -ENOENT because the new
1161          * file not existing is perfectly valid. */
1162         if ((status < 0) && (status != -ENOENT)) {
1163                 /* If we cannot find the file specified we should just */
1164                 /* return the error... */
1165                 mlog_errno(status);
1166                 goto bail;
1167         }
1168         if (status == 0)
1169                 target_exists = 1;
1170
1171         if (!target_exists && new_inode) {
1172                 /*
1173                  * Target was unlinked by another node while we were
1174                  * waiting to get to ocfs2_rename(). There isn't
1175                  * anything we can do here to help the situation, so
1176                  * bubble up the appropriate error.
1177                  */
1178                 status = -ENOENT;
1179                 goto bail;
1180         }
1181
1182         /* In case we need to overwrite an existing file, we blow it
1183          * away first */
1184         if (target_exists) {
1185                 /* VFS didn't think there existed an inode here, but
1186                  * someone else in the cluster must have raced our
1187                  * rename to create one. Today we error cleanly, in
1188                  * the future we should consider calling iget to build
1189                  * a new struct inode for this entry. */
1190                 if (!new_inode) {
1191                         status = -EACCES;
1192
1193                         mlog(0, "We found an inode for name %.*s but VFS "
1194                              "didn't give us one.\n", new_dentry->d_name.len,
1195                              new_dentry->d_name.name);
1196                         goto bail;
1197                 }
1198
1199                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1200                         status = -EACCES;
1201
1202                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1203                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1204                              (unsigned long long)newfe_blkno,
1205                              OCFS2_I(new_inode)->ip_flags);
1206                         goto bail;
1207                 }
1208
1209                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1210                 if (status < 0) {
1211                         if (status != -ENOENT)
1212                                 mlog_errno(status);
1213                         goto bail;
1214                 }
1215                 new_child_locked = 1;
1216
1217                 status = ocfs2_remote_dentry_delete(new_dentry);
1218                 if (status < 0) {
1219                         mlog_errno(status);
1220                         goto bail;
1221                 }
1222
1223                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1224
1225                 mlog(0, "aha rename over existing... new_blkno=%llu "
1226                      "newfebh=%p bhblocknr=%llu\n",
1227                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1228                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1229
1230                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1231                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1232                                                 OCFS2_I(new_inode)->ip_blkno,
1233                                                 orphan_name, &orphan_insert);
1234                         if (status < 0) {
1235                                 mlog_errno(status);
1236                                 goto bail;
1237                         }
1238                 }
1239         } else {
1240                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1241
1242                 status = ocfs2_check_dir_for_entry(new_dir,
1243                                                    new_dentry->d_name.name,
1244                                                    new_dentry->d_name.len);
1245                 if (status)
1246                         goto bail;
1247
1248                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1249                                                       new_dentry->d_name.name,
1250                                                       new_dentry->d_name.len,
1251                                                       &target_insert);
1252                 if (status < 0) {
1253                         mlog_errno(status);
1254                         goto bail;
1255                 }
1256         }
1257
1258         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1259         if (IS_ERR(handle)) {
1260                 status = PTR_ERR(handle);
1261                 handle = NULL;
1262                 mlog_errno(status);
1263                 goto bail;
1264         }
1265
1266         if (target_exists) {
1267                 if (S_ISDIR(new_inode->i_mode)) {
1268                         if (new_inode->i_nlink != 2 ||
1269                             !ocfs2_empty_dir(new_inode)) {
1270                                 status = -ENOTEMPTY;
1271                                 goto bail;
1272                         }
1273                 }
1274                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1275                                                  newfe_bh,
1276                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1277                 if (status < 0) {
1278                         mlog_errno(status);
1279                         goto bail;
1280                 }
1281
1282                 if (S_ISDIR(new_inode->i_mode) ||
1283                     (ocfs2_read_links_count(newfe) == 1)) {
1284                         status = ocfs2_orphan_add(osb, handle, new_inode,
1285                                                   newfe, orphan_name,
1286                                                   &orphan_insert, orphan_dir);
1287                         if (status < 0) {
1288                                 mlog_errno(status);
1289                                 goto bail;
1290                         }
1291                 }
1292
1293                 /* change the dirent to point to the correct inode */
1294                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1295                                             old_inode);
1296                 if (status < 0) {
1297                         mlog_errno(status);
1298                         goto bail;
1299                 }
1300                 new_dir->i_version++;
1301
1302                 if (S_ISDIR(new_inode->i_mode))
1303                         ocfs2_set_links_count(newfe, 0);
1304                 else
1305                         ocfs2_add_links_count(newfe, -1);
1306                 ocfs2_journal_dirty(handle, newfe_bh);
1307         } else {
1308                 /* if the name was not found in new_dir, add it now */
1309                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1310                                          OCFS2_I(old_inode)->ip_blkno,
1311                                          new_dir_bh, &target_insert);
1312         }
1313
1314         old_inode->i_ctime = CURRENT_TIME;
1315         mark_inode_dirty(old_inode);
1316
1317         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1318                                          old_inode_bh,
1319                                          OCFS2_JOURNAL_ACCESS_WRITE);
1320         if (status >= 0) {
1321                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1322
1323                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1324                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1325                 ocfs2_journal_dirty(handle, old_inode_bh);
1326         } else
1327                 mlog_errno(status);
1328
1329         /*
1330          * Now that the name has been added to new_dir, remove the old name.
1331          *
1332          * We don't keep any directory entry context around until now
1333          * because the insert might have changed the type of directory
1334          * we're dealing with.
1335          */
1336         status = ocfs2_find_entry(old_dentry->d_name.name,
1337                                   old_dentry->d_name.len, old_dir,
1338                                   &old_entry_lookup);
1339         if (status)
1340                 goto bail;
1341
1342         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1343         if (status < 0) {
1344                 mlog_errno(status);
1345                 goto bail;
1346         }
1347
1348         if (new_inode) {
1349                 new_inode->i_nlink--;
1350                 new_inode->i_ctime = CURRENT_TIME;
1351         }
1352         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1353
1354         if (update_dot_dot) {
1355                 status = ocfs2_update_entry(old_inode, handle,
1356                                             &old_inode_dot_dot_res, new_dir);
1357                 old_dir->i_nlink--;
1358                 if (new_inode) {
1359                         new_inode->i_nlink--;
1360                 } else {
1361                         inc_nlink(new_dir);
1362                         mark_inode_dirty(new_dir);
1363                 }
1364         }
1365         mark_inode_dirty(old_dir);
1366         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1367         if (new_inode) {
1368                 mark_inode_dirty(new_inode);
1369                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1370         }
1371
1372         if (old_dir != new_dir) {
1373                 /* Keep the same times on both directories.*/
1374                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1375
1376                 /*
1377                  * This will also pick up the i_nlink change from the
1378                  * block above.
1379                  */
1380                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1381         }
1382
1383         if (old_dir_nlink != old_dir->i_nlink) {
1384                 if (!old_dir_bh) {
1385                         mlog(ML_ERROR, "need to change nlink for old dir "
1386                              "%llu from %d to %d but bh is NULL!\n",
1387                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1388                              (int)old_dir_nlink, old_dir->i_nlink);
1389                 } else {
1390                         struct ocfs2_dinode *fe;
1391                         status = ocfs2_journal_access_di(handle,
1392                                                          INODE_CACHE(old_dir),
1393                                                          old_dir_bh,
1394                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1395                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1396                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1397                         ocfs2_journal_dirty(handle, old_dir_bh);
1398                 }
1399         }
1400         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1401         status = 0;
1402 bail:
1403         if (rename_lock)
1404                 ocfs2_rename_unlock(osb);
1405
1406         if (handle)
1407                 ocfs2_commit_trans(osb, handle);
1408
1409         if (parents_locked)
1410                 ocfs2_double_unlock(old_dir, new_dir);
1411
1412         if (old_child_locked)
1413                 ocfs2_inode_unlock(old_inode, 1);
1414
1415         if (new_child_locked)
1416                 ocfs2_inode_unlock(new_inode, 1);
1417
1418         if (orphan_dir) {
1419                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1420                 ocfs2_inode_unlock(orphan_dir, 1);
1421                 mutex_unlock(&orphan_dir->i_mutex);
1422                 iput(orphan_dir);
1423         }
1424
1425         if (new_inode)
1426                 sync_mapping_buffers(old_inode->i_mapping);
1427
1428         if (new_inode)
1429                 iput(new_inode);
1430
1431         ocfs2_free_dir_lookup_result(&target_lookup_res);
1432         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1433         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1434         ocfs2_free_dir_lookup_result(&orphan_insert);
1435         ocfs2_free_dir_lookup_result(&target_insert);
1436
1437         brelse(newfe_bh);
1438         brelse(old_inode_bh);
1439         brelse(old_dir_bh);
1440         brelse(new_dir_bh);
1441
1442         mlog_exit(status);
1443
1444         return status;
1445 }
1446
1447 /*
1448  * we expect i_size = strlen(symname). Copy symname into the file
1449  * data, including the null terminator.
1450  */
1451 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1452                                      handle_t *handle,
1453                                      struct inode *inode,
1454                                      const char *symname)
1455 {
1456         struct buffer_head **bhs = NULL;
1457         const char *c;
1458         struct super_block *sb = osb->sb;
1459         u64 p_blkno, p_blocks;
1460         int virtual, blocks, status, i, bytes_left;
1461
1462         bytes_left = i_size_read(inode) + 1;
1463         /* we can't trust i_blocks because we're actually going to
1464          * write i_size + 1 bytes. */
1465         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1466
1467         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1468                         (unsigned long long)inode->i_blocks,
1469                         i_size_read(inode), blocks);
1470
1471         /* Sanity check -- make sure we're going to fit. */
1472         if (bytes_left >
1473             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1474                 status = -EIO;
1475                 mlog_errno(status);
1476                 goto bail;
1477         }
1478
1479         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1480         if (!bhs) {
1481                 status = -ENOMEM;
1482                 mlog_errno(status);
1483                 goto bail;
1484         }
1485
1486         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1487                                              NULL);
1488         if (status < 0) {
1489                 mlog_errno(status);
1490                 goto bail;
1491         }
1492
1493         /* links can never be larger than one cluster so we know this
1494          * is all going to be contiguous, but do a sanity check
1495          * anyway. */
1496         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1497                 status = -EIO;
1498                 mlog_errno(status);
1499                 goto bail;
1500         }
1501
1502         virtual = 0;
1503         while(bytes_left > 0) {
1504                 c = &symname[virtual * sb->s_blocksize];
1505
1506                 bhs[virtual] = sb_getblk(sb, p_blkno);
1507                 if (!bhs[virtual]) {
1508                         status = -ENOMEM;
1509                         mlog_errno(status);
1510                         goto bail;
1511                 }
1512                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1513                                               bhs[virtual]);
1514
1515                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1516                                               bhs[virtual],
1517                                               OCFS2_JOURNAL_ACCESS_CREATE);
1518                 if (status < 0) {
1519                         mlog_errno(status);
1520                         goto bail;
1521                 }
1522
1523                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1524
1525                 memcpy(bhs[virtual]->b_data, c,
1526                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1527                        bytes_left);
1528
1529                 ocfs2_journal_dirty(handle, bhs[virtual]);
1530
1531                 virtual++;
1532                 p_blkno++;
1533                 bytes_left -= sb->s_blocksize;
1534         }
1535
1536         status = 0;
1537 bail:
1538
1539         if (bhs) {
1540                 for(i = 0; i < blocks; i++)
1541                         brelse(bhs[i]);
1542                 kfree(bhs);
1543         }
1544
1545         mlog_exit(status);
1546         return status;
1547 }
1548
1549 static int ocfs2_symlink(struct inode *dir,
1550                          struct dentry *dentry,
1551                          const char *symname)
1552 {
1553         int status, l, credits;
1554         u64 newsize;
1555         struct ocfs2_super *osb = NULL;
1556         struct inode *inode = NULL;
1557         struct super_block *sb;
1558         struct buffer_head *new_fe_bh = NULL;
1559         struct buffer_head *parent_fe_bh = NULL;
1560         struct ocfs2_dinode *fe = NULL;
1561         struct ocfs2_dinode *dirfe;
1562         handle_t *handle = NULL;
1563         struct ocfs2_alloc_context *inode_ac = NULL;
1564         struct ocfs2_alloc_context *data_ac = NULL;
1565         struct ocfs2_alloc_context *xattr_ac = NULL;
1566         int want_clusters = 0;
1567         int xattr_credits = 0;
1568         struct ocfs2_security_xattr_info si = {
1569                 .enable = 1,
1570         };
1571         int did_quota = 0, did_quota_inode = 0;
1572         struct ocfs2_dir_lookup_result lookup = { NULL, };
1573
1574         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1575                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1576
1577         dquot_initialize(dir);
1578
1579         sb = dir->i_sb;
1580         osb = OCFS2_SB(sb);
1581
1582         l = strlen(symname) + 1;
1583
1584         credits = ocfs2_calc_symlink_credits(sb);
1585
1586         /* lock the parent directory */
1587         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1588         if (status < 0) {
1589                 if (status != -ENOENT)
1590                         mlog_errno(status);
1591                 return status;
1592         }
1593
1594         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1595         if (!ocfs2_read_links_count(dirfe)) {
1596                 /* can't make a file in a deleted directory. */
1597                 status = -ENOENT;
1598                 goto bail;
1599         }
1600
1601         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1602                                            dentry->d_name.len);
1603         if (status)
1604                 goto bail;
1605
1606         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1607                                               dentry->d_name.name,
1608                                               dentry->d_name.len, &lookup);
1609         if (status < 0) {
1610                 mlog_errno(status);
1611                 goto bail;
1612         }
1613
1614         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1615         if (status < 0) {
1616                 if (status != -ENOSPC)
1617                         mlog_errno(status);
1618                 goto bail;
1619         }
1620
1621         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1622         if (!inode) {
1623                 status = -ENOMEM;
1624                 mlog_errno(status);
1625                 goto bail;
1626         }
1627
1628         /* get security xattr */
1629         status = ocfs2_init_security_get(inode, dir, &si);
1630         if (status) {
1631                 if (status == -EOPNOTSUPP)
1632                         si.enable = 0;
1633                 else {
1634                         mlog_errno(status);
1635                         goto bail;
1636                 }
1637         }
1638
1639         /* calculate meta data/clusters for setting security xattr */
1640         if (si.enable) {
1641                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1642                                                   &xattr_credits, &xattr_ac);
1643                 if (status < 0) {
1644                         mlog_errno(status);
1645                         goto bail;
1646                 }
1647         }
1648
1649         /* don't reserve bitmap space for fast symlinks. */
1650         if (l > ocfs2_fast_symlink_chars(sb))
1651                 want_clusters += 1;
1652
1653         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1654         if (status < 0) {
1655                 if (status != -ENOSPC)
1656                         mlog_errno(status);
1657                 goto bail;
1658         }
1659
1660         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1661         if (IS_ERR(handle)) {
1662                 status = PTR_ERR(handle);
1663                 handle = NULL;
1664                 mlog_errno(status);
1665                 goto bail;
1666         }
1667
1668         status = dquot_alloc_inode(inode);
1669         if (status)
1670                 goto bail;
1671         did_quota_inode = 1;
1672
1673         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1674                    inode->i_mode, dentry->d_name.len,
1675                    dentry->d_name.name);
1676
1677         status = ocfs2_mknod_locked(osb, dir, inode,
1678                                     0, &new_fe_bh, parent_fe_bh, handle,
1679                                     inode_ac);
1680         if (status < 0) {
1681                 mlog_errno(status);
1682                 goto bail;
1683         }
1684
1685         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1686         inode->i_rdev = 0;
1687         newsize = l - 1;
1688         if (l > ocfs2_fast_symlink_chars(sb)) {
1689                 u32 offset = 0;
1690
1691                 inode->i_op = &ocfs2_symlink_inode_operations;
1692                 status = dquot_alloc_space_nodirty(inode,
1693                     ocfs2_clusters_to_bytes(osb->sb, 1));
1694                 if (status)
1695                         goto bail;
1696                 did_quota = 1;
1697                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1698                                               new_fe_bh,
1699                                               handle, data_ac, NULL,
1700                                               NULL);
1701                 if (status < 0) {
1702                         if (status != -ENOSPC && status != -EINTR) {
1703                                 mlog(ML_ERROR,
1704                                      "Failed to extend file to %llu\n",
1705                                      (unsigned long long)newsize);
1706                                 mlog_errno(status);
1707                                 status = -ENOSPC;
1708                         }
1709                         goto bail;
1710                 }
1711                 i_size_write(inode, newsize);
1712                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1713         } else {
1714                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1715                 memcpy((char *) fe->id2.i_symlink, symname, l);
1716                 i_size_write(inode, newsize);
1717                 inode->i_blocks = 0;
1718         }
1719
1720         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1721         if (status < 0) {
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         if (!ocfs2_inode_is_fast_symlink(inode)) {
1727                 status = ocfs2_create_symlink_data(osb, handle, inode,
1728                                                    symname);
1729                 if (status < 0) {
1730                         mlog_errno(status);
1731                         goto bail;
1732                 }
1733         }
1734
1735         if (si.enable) {
1736                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1737                                                  xattr_ac, data_ac);
1738                 if (status < 0) {
1739                         mlog_errno(status);
1740                         goto bail;
1741                 }
1742         }
1743
1744         status = ocfs2_add_entry(handle, dentry, inode,
1745                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1746                                  &lookup);
1747         if (status < 0) {
1748                 mlog_errno(status);
1749                 goto bail;
1750         }
1751
1752         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1753         if (status) {
1754                 mlog_errno(status);
1755                 goto bail;
1756         }
1757
1758         insert_inode_hash(inode);
1759         dentry->d_op = &ocfs2_dentry_ops;
1760         d_instantiate(dentry, inode);
1761 bail:
1762         if (status < 0 && did_quota)
1763                 dquot_free_space_nodirty(inode,
1764                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1765         if (status < 0 && did_quota_inode)
1766                 dquot_free_inode(inode);
1767         if (handle)
1768                 ocfs2_commit_trans(osb, handle);
1769
1770         ocfs2_inode_unlock(dir, 1);
1771
1772         brelse(new_fe_bh);
1773         brelse(parent_fe_bh);
1774         kfree(si.name);
1775         kfree(si.value);
1776         ocfs2_free_dir_lookup_result(&lookup);
1777         if (inode_ac)
1778                 ocfs2_free_alloc_context(inode_ac);
1779         if (data_ac)
1780                 ocfs2_free_alloc_context(data_ac);
1781         if (xattr_ac)
1782                 ocfs2_free_alloc_context(xattr_ac);
1783         if ((status < 0) && inode) {
1784                 clear_nlink(inode);
1785                 iput(inode);
1786         }
1787
1788         mlog_exit(status);
1789
1790         return status;
1791 }
1792
1793 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1794 {
1795         int status, namelen;
1796
1797         mlog_entry_void();
1798
1799         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1800                            (long long)blkno);
1801         if (namelen <= 0) {
1802                 if (namelen)
1803                         status = namelen;
1804                 else
1805                         status = -EINVAL;
1806                 mlog_errno(status);
1807                 goto bail;
1808         }
1809         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1810                 status = -EINVAL;
1811                 mlog_errno(status);
1812                 goto bail;
1813         }
1814
1815         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1816              namelen);
1817
1818         status = 0;
1819 bail:
1820         mlog_exit(status);
1821         return status;
1822 }
1823
1824 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1825                                     struct inode **ret_orphan_dir,
1826                                     u64 blkno,
1827                                     char *name,
1828                                     struct ocfs2_dir_lookup_result *lookup)
1829 {
1830         struct inode *orphan_dir_inode;
1831         struct buffer_head *orphan_dir_bh = NULL;
1832         int status = 0;
1833
1834         status = ocfs2_blkno_stringify(blkno, name);
1835         if (status < 0) {
1836                 mlog_errno(status);
1837                 return status;
1838         }
1839
1840         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1841                                                        ORPHAN_DIR_SYSTEM_INODE,
1842                                                        osb->slot_num);
1843         if (!orphan_dir_inode) {
1844                 status = -ENOENT;
1845                 mlog_errno(status);
1846                 return status;
1847         }
1848
1849         mutex_lock(&orphan_dir_inode->i_mutex);
1850
1851         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1852         if (status < 0) {
1853                 mlog_errno(status);
1854                 goto leave;
1855         }
1856
1857         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1858                                               orphan_dir_bh, name,
1859                                               OCFS2_ORPHAN_NAMELEN, lookup);
1860         if (status < 0) {
1861                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1862
1863                 mlog_errno(status);
1864                 goto leave;
1865         }
1866
1867         *ret_orphan_dir = orphan_dir_inode;
1868
1869 leave:
1870         if (status) {
1871                 mutex_unlock(&orphan_dir_inode->i_mutex);
1872                 iput(orphan_dir_inode);
1873         }
1874
1875         brelse(orphan_dir_bh);
1876
1877         mlog_exit(status);
1878         return status;
1879 }
1880
1881 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1882                             handle_t *handle,
1883                             struct inode *inode,
1884                             struct ocfs2_dinode *fe,
1885                             char *name,
1886                             struct ocfs2_dir_lookup_result *lookup,
1887                             struct inode *orphan_dir_inode)
1888 {
1889         struct buffer_head *orphan_dir_bh = NULL;
1890         int status = 0;
1891         struct ocfs2_dinode *orphan_fe;
1892
1893         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1894
1895         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1896         if (status < 0) {
1897                 mlog_errno(status);
1898                 goto leave;
1899         }
1900
1901         status = ocfs2_journal_access_di(handle,
1902                                          INODE_CACHE(orphan_dir_inode),
1903                                          orphan_dir_bh,
1904                                          OCFS2_JOURNAL_ACCESS_WRITE);
1905         if (status < 0) {
1906                 mlog_errno(status);
1907                 goto leave;
1908         }
1909
1910         /* we're a cluster, and nlink can change on disk from
1911          * underneath us... */
1912         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1913         if (S_ISDIR(inode->i_mode))
1914                 ocfs2_add_links_count(orphan_fe, 1);
1915         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1916         ocfs2_journal_dirty(handle, orphan_dir_bh);
1917
1918         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1919                                    OCFS2_ORPHAN_NAMELEN, inode,
1920                                    OCFS2_I(inode)->ip_blkno,
1921                                    orphan_dir_bh, lookup);
1922         if (status < 0) {
1923                 mlog_errno(status);
1924                 goto leave;
1925         }
1926
1927         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1928
1929         /* Record which orphan dir our inode now resides
1930          * in. delete_inode will use this to determine which orphan
1931          * dir to lock. */
1932         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1933
1934         mlog(0, "Inode %llu orphaned in slot %d\n",
1935              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1936
1937 leave:
1938         brelse(orphan_dir_bh);
1939
1940         mlog_exit(status);
1941         return status;
1942 }
1943
1944 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1945 int ocfs2_orphan_del(struct ocfs2_super *osb,
1946                      handle_t *handle,
1947                      struct inode *orphan_dir_inode,
1948                      struct inode *inode,
1949                      struct buffer_head *orphan_dir_bh)
1950 {
1951         char name[OCFS2_ORPHAN_NAMELEN + 1];
1952         struct ocfs2_dinode *orphan_fe;
1953         int status = 0;
1954         struct ocfs2_dir_lookup_result lookup = { NULL, };
1955
1956         mlog_entry_void();
1957
1958         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1959         if (status < 0) {
1960                 mlog_errno(status);
1961                 goto leave;
1962         }
1963
1964         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1965              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1966              OCFS2_ORPHAN_NAMELEN);
1967
1968         /* find it's spot in the orphan directory */
1969         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
1970                                   &lookup);
1971         if (status) {
1972                 mlog_errno(status);
1973                 goto leave;
1974         }
1975
1976         /* remove it from the orphan directory */
1977         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
1978         if (status < 0) {
1979                 mlog_errno(status);
1980                 goto leave;
1981         }
1982
1983         status = ocfs2_journal_access_di(handle,
1984                                          INODE_CACHE(orphan_dir_inode),
1985                                          orphan_dir_bh,
1986                                          OCFS2_JOURNAL_ACCESS_WRITE);
1987         if (status < 0) {
1988                 mlog_errno(status);
1989                 goto leave;
1990         }
1991
1992         /* do the i_nlink dance! :) */
1993         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1994         if (S_ISDIR(inode->i_mode))
1995                 ocfs2_add_links_count(orphan_fe, -1);
1996         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1997         ocfs2_journal_dirty(handle, orphan_dir_bh);
1998
1999 leave:
2000         ocfs2_free_dir_lookup_result(&lookup);
2001
2002         mlog_exit(status);
2003         return status;
2004 }
2005
2006 int ocfs2_create_inode_in_orphan(struct inode *dir,
2007                                  int mode,
2008                                  struct inode **new_inode)
2009 {
2010         int status, did_quota_inode = 0;
2011         struct inode *inode = NULL;
2012         struct inode *orphan_dir = NULL;
2013         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2014         struct ocfs2_dinode *di = NULL;
2015         handle_t *handle = NULL;
2016         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2017         struct buffer_head *parent_di_bh = NULL;
2018         struct buffer_head *new_di_bh = NULL;
2019         struct ocfs2_alloc_context *inode_ac = NULL;
2020         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2021
2022         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2023         if (status < 0) {
2024                 if (status != -ENOENT)
2025                         mlog_errno(status);
2026                 return status;
2027         }
2028
2029         /*
2030          * We give the orphan dir the root blkno to fake an orphan name,
2031          * and allocate enough space for our insertion.
2032          */
2033         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
2034                                           osb->root_blkno,
2035                                           orphan_name, &orphan_insert);
2036         if (status < 0) {
2037                 mlog_errno(status);
2038                 goto leave;
2039         }
2040
2041         /* reserve an inode spot */
2042         status = ocfs2_reserve_new_inode(osb, &inode_ac);
2043         if (status < 0) {
2044                 if (status != -ENOSPC)
2045                         mlog_errno(status);
2046                 goto leave;
2047         }
2048
2049         inode = ocfs2_get_init_inode(dir, mode);
2050         if (!inode) {
2051                 status = -ENOMEM;
2052                 mlog_errno(status);
2053                 goto leave;
2054         }
2055
2056         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2057         if (IS_ERR(handle)) {
2058                 status = PTR_ERR(handle);
2059                 handle = NULL;
2060                 mlog_errno(status);
2061                 goto leave;
2062         }
2063
2064         status = dquot_alloc_inode(inode);
2065         if (status)
2066                 goto leave;
2067         did_quota_inode = 1;
2068
2069         inode->i_nlink = 0;
2070         /* do the real work now. */
2071         status = ocfs2_mknod_locked(osb, dir, inode,
2072                                     0, &new_di_bh, parent_di_bh, handle,
2073                                     inode_ac);
2074         if (status < 0) {
2075                 mlog_errno(status);
2076                 goto leave;
2077         }
2078
2079         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
2080         if (status < 0) {
2081                 mlog_errno(status);
2082                 goto leave;
2083         }
2084
2085         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2086         status = ocfs2_orphan_add(osb, handle, inode, di, orphan_name,
2087                                   &orphan_insert, orphan_dir);
2088         if (status < 0) {
2089                 mlog_errno(status);
2090                 goto leave;
2091         }
2092
2093         /* get open lock so that only nodes can't remove it from orphan dir. */
2094         status = ocfs2_open_lock(inode);
2095         if (status < 0)
2096                 mlog_errno(status);
2097
2098         insert_inode_hash(inode);
2099 leave:
2100         if (status < 0 && did_quota_inode)
2101                 dquot_free_inode(inode);
2102         if (handle)
2103                 ocfs2_commit_trans(osb, handle);
2104
2105         if (orphan_dir) {
2106                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2107                 ocfs2_inode_unlock(orphan_dir, 1);
2108                 mutex_unlock(&orphan_dir->i_mutex);
2109                 iput(orphan_dir);
2110         }
2111
2112         if (status == -ENOSPC)
2113                 mlog(0, "Disk is full\n");
2114
2115         if ((status < 0) && inode) {
2116                 clear_nlink(inode);
2117                 iput(inode);
2118         }
2119
2120         if (inode_ac)
2121                 ocfs2_free_alloc_context(inode_ac);
2122
2123         brelse(new_di_bh);
2124
2125         if (!status)
2126                 *new_inode = inode;
2127
2128         ocfs2_free_dir_lookup_result(&orphan_insert);
2129
2130         ocfs2_inode_unlock(dir, 1);
2131         brelse(parent_di_bh);
2132         return status;
2133 }
2134
2135 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2136                                    struct inode *inode,
2137                                    struct dentry *dentry)
2138 {
2139         int status = 0;
2140         struct buffer_head *parent_di_bh = NULL;
2141         handle_t *handle = NULL;
2142         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2143         struct ocfs2_dinode *dir_di, *di;
2144         struct inode *orphan_dir_inode = NULL;
2145         struct buffer_head *orphan_dir_bh = NULL;
2146         struct buffer_head *di_bh = NULL;
2147         struct ocfs2_dir_lookup_result lookup = { NULL, };
2148
2149         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2150                    dentry->d_name.len, dentry->d_name.name);
2151
2152         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2153         if (status < 0) {
2154                 if (status != -ENOENT)
2155                         mlog_errno(status);
2156                 return status;
2157         }
2158
2159         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2160         if (!dir_di->i_links_count) {
2161                 /* can't make a file in a deleted directory. */
2162                 status = -ENOENT;
2163                 goto leave;
2164         }
2165
2166         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2167                                            dentry->d_name.len);
2168         if (status)
2169                 goto leave;
2170
2171         /* get a spot inside the dir. */
2172         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2173                                               dentry->d_name.name,
2174                                               dentry->d_name.len, &lookup);
2175         if (status < 0) {
2176                 mlog_errno(status);
2177                 goto leave;
2178         }
2179
2180         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2181                                                        ORPHAN_DIR_SYSTEM_INODE,
2182                                                        osb->slot_num);
2183         if (!orphan_dir_inode) {
2184                 status = -EEXIST;
2185                 mlog_errno(status);
2186                 goto leave;
2187         }
2188
2189         mutex_lock(&orphan_dir_inode->i_mutex);
2190
2191         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2192         if (status < 0) {
2193                 mlog_errno(status);
2194                 mutex_unlock(&orphan_dir_inode->i_mutex);
2195                 iput(orphan_dir_inode);
2196                 goto leave;
2197         }
2198
2199         status = ocfs2_read_inode_block(inode, &di_bh);
2200         if (status < 0) {
2201                 mlog_errno(status);
2202                 goto orphan_unlock;
2203         }
2204
2205         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2206         if (IS_ERR(handle)) {
2207                 status = PTR_ERR(handle);
2208                 handle = NULL;
2209                 mlog_errno(status);
2210                 goto orphan_unlock;
2211         }
2212
2213         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2214                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2215         if (status < 0) {
2216                 mlog_errno(status);
2217                 goto out_commit;
2218         }
2219
2220         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2221                                   orphan_dir_bh);
2222         if (status < 0) {
2223                 mlog_errno(status);
2224                 goto out_commit;
2225         }
2226
2227         di = (struct ocfs2_dinode *)di_bh->b_data;
2228         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2229         di->i_orphaned_slot = 0;
2230         inode->i_nlink = 1;
2231         ocfs2_set_links_count(di, inode->i_nlink);
2232         ocfs2_journal_dirty(handle, di_bh);
2233
2234         status = ocfs2_add_entry(handle, dentry, inode,
2235                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2236                                  &lookup);
2237         if (status < 0) {
2238                 mlog_errno(status);
2239                 goto out_commit;
2240         }
2241
2242         status = ocfs2_dentry_attach_lock(dentry, inode,
2243                                           OCFS2_I(dir)->ip_blkno);
2244         if (status) {
2245                 mlog_errno(status);
2246                 goto out_commit;
2247         }
2248
2249         dentry->d_op = &ocfs2_dentry_ops;
2250         d_instantiate(dentry, inode);
2251         status = 0;
2252 out_commit:
2253         ocfs2_commit_trans(osb, handle);
2254 orphan_unlock:
2255         ocfs2_inode_unlock(orphan_dir_inode, 1);
2256         mutex_unlock(&orphan_dir_inode->i_mutex);
2257         iput(orphan_dir_inode);
2258 leave:
2259
2260         ocfs2_inode_unlock(dir, 1);
2261
2262         brelse(di_bh);
2263         brelse(parent_di_bh);
2264         brelse(orphan_dir_bh);
2265
2266         ocfs2_free_dir_lookup_result(&lookup);
2267
2268         mlog_exit(status);
2269
2270         return status;
2271 }
2272
2273 const struct inode_operations ocfs2_dir_iops = {
2274         .create         = ocfs2_create,
2275         .lookup         = ocfs2_lookup,
2276         .link           = ocfs2_link,
2277         .unlink         = ocfs2_unlink,
2278         .rmdir          = ocfs2_unlink,
2279         .symlink        = ocfs2_symlink,
2280         .mkdir          = ocfs2_mkdir,
2281         .mknod          = ocfs2_mknod,
2282         .rename         = ocfs2_rename,
2283         .setattr        = ocfs2_setattr,
2284         .getattr        = ocfs2_getattr,
2285         .permission     = ocfs2_permission,
2286         .setxattr       = generic_setxattr,
2287         .getxattr       = generic_getxattr,
2288         .listxattr      = ocfs2_listxattr,
2289         .removexattr    = generic_removexattr,
2290         .fiemap         = ocfs2_fiemap,
2291 };