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 buffer_head *fe_bh,
  88                             char *name,
  89                             struct ocfs2_dir_lookup_result *lookup,
  90                             struct inode *orphan_dir_inode);
  91
  92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  93                                      handle_t *handle,
  94                                      struct inode *inode,
  95                                      const char *symname);
  96
  97 /* An orphan dir name is an 8 byte value, printed as a hex string */
  98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  99
 100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 101                                    struct nameidata *nd)
 102 {
 103         int status;
 104         u64 blkno;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_inode_info *oi;
 108
 109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 110                    dentry->d_name.len, dentry->d_name.name);
 111
 112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 113                 ret = ERR_PTR(-ENAMETOOLONG);
 114                 goto bail;
 115         }
 116
 117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 119
 120         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 121         if (status < 0) {
 122                 if (status != -ENOENT)
 123                         mlog_errno(status);
 124                 ret = ERR_PTR(status);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 129                                             dentry->d_name.len, &blkno);
 130         if (status < 0)
 131                 goto bail_add;
 132
 133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 134         if (IS_ERR(inode)) {
 135                 ret = ERR_PTR(-EACCES);
 136                 goto bail_unlock;
 137         }
 138
 139         oi = OCFS2_I(inode);
 140         /* Clear any orphaned state... If we were able to look up the
 141          * inode from a directory, it certainly can't be orphaned. We
 142          * might have the bad state from a node which intended to
 143          * orphan this inode but crashed before it could commit the
 144          * unlink. */
 145         spin_lock(&oi->ip_lock);
 146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 147         spin_unlock(&oi->ip_lock);
 148
 149 bail_add:
 150         dentry->d_op = &ocfs2_dentry_ops;
 151         ret = d_splice_alias(inode, dentry);
 152
 153         if (inode) {
 154                 /*
 155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 156                  * dentry, it will d_move() it on top of ourse. The
 157                  * return value will indicate this however, so in
 158                  * those cases, we switch them around for the locking
 159                  * code.
 160                  *
 161                  * NOTE: This dentry already has ->d_op set from
 162                  * ocfs2_get_parent() and ocfs2_get_dentry()
 163                  */
 164                 if (ret)
 165                         dentry = ret;
 166
 167                 status = ocfs2_dentry_attach_lock(dentry, inode,
 168                                                   OCFS2_I(dir)->ip_blkno);
 169                 if (status) {
 170                         mlog_errno(status);
 171                         ret = ERR_PTR(status);
 172                         goto bail_unlock;
 173                 }
 174         }
 175
 176 bail_unlock:
 177         /* Don't drop the cluster lock until *after* the d_add --
 178          * unlink on another node will message us to remove that
 179          * dentry under this lock so otherwise we can race this with
 180          * the downconvert thread and have a stale dentry. */
 181         ocfs2_inode_unlock(dir, 0);
 182
 183 bail:
 184
 185         mlog_exit_ptr(ret);
 186
 187         return ret;
 188 }
 189
 190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
 191 {
 192         struct inode *inode;
 193
 194         inode = new_inode(dir->i_sb);
 195         if (!inode) {
 196                 mlog(ML_ERROR, "new_inode failed!\n");
 197                 return NULL;
 198         }
 199
 200         /* populate as many fields early on as possible - many of
 201          * these are used by the support functions here and in
 202          * callers. */
 203         if (S_ISDIR(mode))
 204                 inode->i_nlink = 2;
 205         else
 206                 inode->i_nlink = 1;
 207         inode->i_uid = current_fsuid();
 208         if (dir->i_mode & S_ISGID) {
 209                 inode->i_gid = dir->i_gid;
 210                 if (S_ISDIR(mode))
 211                         mode |= S_ISGID;
 212         } else
 213                 inode->i_gid = current_fsgid();
 214         inode->i_mode = mode;
 215         dquot_initialize(inode);
 216         return inode;
 217 }
 218
 219 static int ocfs2_mknod(struct inode *dir,
 220                        struct dentry *dentry,
 221                        int mode,
 222                        dev_t dev)
 223 {
 224         int status = 0;
 225         struct buffer_head *parent_fe_bh = NULL;
 226         handle_t *handle = NULL;
 227         struct ocfs2_super *osb;
 228         struct ocfs2_dinode *dirfe;
 229         struct buffer_head *new_fe_bh = NULL;
 230         struct inode *inode = NULL;
 231         struct ocfs2_alloc_context *inode_ac = NULL;
 232         struct ocfs2_alloc_context *data_ac = NULL;
 233         struct ocfs2_alloc_context *meta_ac = NULL;
 234         int want_clusters = 0;
 235         int want_meta = 0;
 236         int xattr_credits = 0;
 237         struct ocfs2_security_xattr_info si = {
 238                 .enable = 1,
 239         };
 240         int did_quota_inode = 0;
 241         struct ocfs2_dir_lookup_result lookup = { NULL, };
 242
 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                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 388                 if (status < 0) {
 389                         mlog_errno(status);
 390                         goto leave;
 391                 }
 392                 inc_nlink(dir);
 393         }
 394
 395         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 396                                 meta_ac, data_ac);
 397         if (status < 0) {
 398                 mlog_errno(status);
 399                 goto leave;
 400         }
 401
 402         if (si.enable) {
 403                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 404                                                  meta_ac, data_ac);
 405                 if (status < 0) {
 406                         mlog_errno(status);
 407                         goto leave;
 408                 }
 409         }
 410
 411         status = ocfs2_add_entry(handle, dentry, inode,
 412                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 413                                  &lookup);
 414         if (status < 0) {
 415                 mlog_errno(status);
 416                 goto leave;
 417         }
 418
 419         status = ocfs2_dentry_attach_lock(dentry, inode,
 420                                           OCFS2_I(dir)->ip_blkno);
 421         if (status) {
 422                 mlog_errno(status);
 423                 goto leave;
 424         }
 425
 426         insert_inode_hash(inode);
 427         dentry->d_op = &ocfs2_dentry_ops;
 428         d_instantiate(dentry, inode);
 429         status = 0;
 430 leave:
 431         if (status < 0 && did_quota_inode)
 432                 dquot_free_inode(inode);
 433         if (handle)
 434                 ocfs2_commit_trans(osb, handle);
 435
 436         ocfs2_inode_unlock(dir, 1);
 437
 438         if (status == -ENOSPC)
 439                 mlog(0, "Disk is full\n");
 440
 441         brelse(new_fe_bh);
 442         brelse(parent_fe_bh);
 443         kfree(si.name);
 444         kfree(si.value);
 445
 446         ocfs2_free_dir_lookup_result(&lookup);
 447
 448         if ((status < 0) && inode) {
 449                 clear_nlink(inode);
 450                 iput(inode);
 451         }
 452
 453         if (inode_ac)
 454                 ocfs2_free_alloc_context(inode_ac);
 455
 456         if (data_ac)
 457                 ocfs2_free_alloc_context(data_ac);
 458
 459         if (meta_ac)
 460                 ocfs2_free_alloc_context(meta_ac);
 461
 462         mlog_exit(status);
 463
 464         return status;
 465 }
 466
 467 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 468                               struct inode *dir,
 469                               struct inode *inode,
 470                               dev_t dev,
 471                               struct buffer_head **new_fe_bh,
 472                               struct buffer_head *parent_fe_bh,
 473                               handle_t *handle,
 474                               struct ocfs2_alloc_context *inode_ac)
 475 {
 476         int status = 0;
 477         struct ocfs2_dinode *fe = NULL;
 478         struct ocfs2_extent_list *fel;
 479         u64 fe_blkno = 0;
 480         u16 suballoc_bit;
 481         u16 feat;
 482
 483         *new_fe_bh = NULL;
 484
 485         status = ocfs2_claim_new_inode(osb, handle, dir, parent_fe_bh,
 486                                        inode_ac, &suballoc_bit, &fe_blkno);
 487         if (status < 0) {
 488                 mlog_errno(status);
 489                 goto leave;
 490         }
 491
 492         /* populate as many fields early on as possible - many of
 493          * these are used by the support functions here and in
 494          * callers. */
 495         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 496         OCFS2_I(inode)->ip_blkno = fe_blkno;
 497         spin_lock(&osb->osb_lock);
 498         inode->i_generation = osb->s_next_generation++;
 499         spin_unlock(&osb->osb_lock);
 500
 501         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 502         if (!*new_fe_bh) {
 503                 status = -EIO;
 504                 mlog_errno(status);
 505                 goto leave;
 506         }
 507         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 508
 509         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 510                                          *new_fe_bh,
 511                                          OCFS2_JOURNAL_ACCESS_CREATE);
 512         if (status < 0) {
 513                 mlog_errno(status);
 514                 goto leave;
 515         }
 516
 517         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 518         memset(fe, 0, osb->sb->s_blocksize);
 519
 520         fe->i_generation = cpu_to_le32(inode->i_generation);
 521         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 522         fe->i_blkno = cpu_to_le64(fe_blkno);
 523         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 524         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 525         fe->i_uid = cpu_to_le32(inode->i_uid);
 526         fe->i_gid = cpu_to_le32(inode->i_gid);
 527         fe->i_mode = cpu_to_le16(inode->i_mode);
 528         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 529                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 530
 531         ocfs2_set_links_count(fe, inode->i_nlink);
 532
 533         fe->i_last_eb_blk = 0;
 534         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 535         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 536         fe->i_atime = fe->i_ctime = fe->i_mtime =
 537                 cpu_to_le64(CURRENT_TIME.tv_sec);
 538         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 539                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 540         fe->i_dtime = 0;
 541
 542         /*
 543          * If supported, directories start with inline data. If inline
 544          * isn't supported, but indexing is, we start them as indexed.
 545          */
 546         feat = le16_to_cpu(fe->i_dyn_features);
 547         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 548                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 549
 550                 fe->id2.i_data.id_count = cpu_to_le16(
 551                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 552         } else {
 553                 fel = &fe->id2.i_list;
 554                 fel->l_tree_depth = 0;
 555                 fel->l_next_free_rec = 0;
 556                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 557         }
 558
 559         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 560         if (status < 0) {
 561                 mlog_errno(status);
 562                 goto leave;
 563         }
 564
 565         ocfs2_populate_inode(inode, fe, 1);
 566         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 567         if (!ocfs2_mount_local(osb)) {
 568                 status = ocfs2_create_new_inode_locks(inode);
 569                 if (status < 0)
 570                         mlog_errno(status);
 571         }
 572
 573         status = 0; /* error in ocfs2_create_new_inode_locks is not
 574                      * critical */
 575
 576 leave:
 577         if (status < 0) {
 578                 if (*new_fe_bh) {
 579                         brelse(*new_fe_bh);
 580                         *new_fe_bh = NULL;
 581                 }
 582         }
 583
 584         mlog_exit(status);
 585         return status;
 586 }
 587
 588 static int ocfs2_mkdir(struct inode *dir,
 589                        struct dentry *dentry,
 590                        int mode)
 591 {
 592         int ret;
 593
 594         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 595                    dentry->d_name.len, dentry->d_name.name);
 596         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 597         mlog_exit(ret);
 598
 599         return ret;
 600 }
 601
 602 static int ocfs2_create(struct inode *dir,
 603                         struct dentry *dentry,
 604                         int mode,
 605                         struct nameidata *nd)
 606 {
 607         int ret;
 608
 609         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 610                    dentry->d_name.len, dentry->d_name.name);
 611         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 612         mlog_exit(ret);
 613
 614         return ret;
 615 }
 616
 617 static int ocfs2_link(struct dentry *old_dentry,
 618                       struct inode *dir,
 619                       struct dentry *dentry)
 620 {
 621         handle_t *handle;
 622         struct inode *inode = old_dentry->d_inode;
 623         int err;
 624         struct buffer_head *fe_bh = NULL;
 625         struct buffer_head *parent_fe_bh = NULL;
 626         struct ocfs2_dinode *fe = NULL;
 627         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 628         struct ocfs2_dir_lookup_result lookup = { NULL, };
 629
 630         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 631                    old_dentry->d_name.len, old_dentry->d_name.name,
 632                    dentry->d_name.len, dentry->d_name.name);
 633
 634         if (S_ISDIR(inode->i_mode))
 635                 return -EPERM;
 636
 637         dquot_initialize(dir);
 638
 639         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 640         if (err < 0) {
 641                 if (err != -ENOENT)
 642                         mlog_errno(err);
 643                 return err;
 644         }
 645
 646         if (!dir->i_nlink) {
 647                 err = -ENOENT;
 648                 goto out;
 649         }
 650
 651         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 652                                         dentry->d_name.len);
 653         if (err)
 654                 goto out;
 655
 656         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 657                                            dentry->d_name.name,
 658                                            dentry->d_name.len, &lookup);
 659         if (err < 0) {
 660                 mlog_errno(err);
 661                 goto out;
 662         }
 663
 664         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 665         if (err < 0) {
 666                 if (err != -ENOENT)
 667                         mlog_errno(err);
 668                 goto out;
 669         }
 670
 671         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 672         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 673                 err = -EMLINK;
 674                 goto out_unlock_inode;
 675         }
 676
 677         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 678         if (IS_ERR(handle)) {
 679                 err = PTR_ERR(handle);
 680                 handle = NULL;
 681                 mlog_errno(err);
 682                 goto out_unlock_inode;
 683         }
 684
 685         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 686                                       OCFS2_JOURNAL_ACCESS_WRITE);
 687         if (err < 0) {
 688                 mlog_errno(err);
 689                 goto out_commit;
 690         }
 691
 692         inc_nlink(inode);
 693         inode->i_ctime = CURRENT_TIME;
 694         ocfs2_set_links_count(fe, inode->i_nlink);
 695         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 696         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 697
 698         err = ocfs2_journal_dirty(handle, fe_bh);
 699         if (err < 0) {
 700                 ocfs2_add_links_count(fe, -1);
 701                 drop_nlink(inode);
 702                 mlog_errno(err);
 703                 goto out_commit;
 704         }
 705
 706         err = ocfs2_add_entry(handle, dentry, inode,
 707                               OCFS2_I(inode)->ip_blkno,
 708                               parent_fe_bh, &lookup);
 709         if (err) {
 710                 ocfs2_add_links_count(fe, -1);
 711                 drop_nlink(inode);
 712                 mlog_errno(err);
 713                 goto out_commit;
 714         }
 715
 716         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 717         if (err) {
 718                 mlog_errno(err);
 719                 goto out_commit;
 720         }
 721
 722         atomic_inc(&inode->i_count);
 723         dentry->d_op = &ocfs2_dentry_ops;
 724         d_instantiate(dentry, inode);
 725
 726 out_commit:
 727         ocfs2_commit_trans(osb, handle);
 728 out_unlock_inode:
 729         ocfs2_inode_unlock(inode, 1);
 730
 731 out:
 732         ocfs2_inode_unlock(dir, 1);
 733
 734         brelse(fe_bh);
 735         brelse(parent_fe_bh);
 736
 737         ocfs2_free_dir_lookup_result(&lookup);
 738
 739         mlog_exit(err);
 740
 741         return err;
 742 }
 743
 744 /*
 745  * Takes and drops an exclusive lock on the given dentry. This will
 746  * force other nodes to drop it.
 747  */
 748 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 749 {
 750         int ret;
 751
 752         ret = ocfs2_dentry_lock(dentry, 1);
 753         if (ret)
 754                 mlog_errno(ret);
 755         else
 756                 ocfs2_dentry_unlock(dentry, 1);
 757
 758         return ret;
 759 }
 760
 761 static inline int inode_is_unlinkable(struct inode *inode)
 762 {
 763         if (S_ISDIR(inode->i_mode)) {
 764                 if (inode->i_nlink == 2)
 765                         return 1;
 766                 return 0;
 767         }
 768
 769         if (inode->i_nlink == 1)
 770                 return 1;
 771         return 0;
 772 }
 773
 774 static int ocfs2_unlink(struct inode *dir,
 775                         struct dentry *dentry)
 776 {
 777         int status;
 778         int child_locked = 0;
 779         struct inode *inode = dentry->d_inode;
 780         struct inode *orphan_dir = NULL;
 781         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 782         u64 blkno;
 783         struct ocfs2_dinode *fe = NULL;
 784         struct buffer_head *fe_bh = NULL;
 785         struct buffer_head *parent_node_bh = NULL;
 786         handle_t *handle = NULL;
 787         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 788         struct ocfs2_dir_lookup_result lookup = { NULL, };
 789         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 790
 791         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 792                    dentry->d_name.len, dentry->d_name.name);
 793
 794         dquot_initialize(dir);
 795
 796         BUG_ON(dentry->d_parent->d_inode != dir);
 797
 798         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 799
 800         if (inode == osb->root_inode) {
 801                 mlog(0, "Cannot delete the root directory\n");
 802                 return -EPERM;
 803         }
 804
 805         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 806                                          OI_LS_PARENT);
 807         if (status < 0) {
 808                 if (status != -ENOENT)
 809                         mlog_errno(status);
 810                 return status;
 811         }
 812
 813         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 814                                           dentry->d_name.len, &blkno, dir,
 815                                           &lookup);
 816         if (status < 0) {
 817                 if (status != -ENOENT)
 818                         mlog_errno(status);
 819                 goto leave;
 820         }
 821
 822         if (OCFS2_I(inode)->ip_blkno != blkno) {
 823                 status = -ENOENT;
 824
 825                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 826                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 827                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 828                 goto leave;
 829         }
 830
 831         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 832         if (status < 0) {
 833                 if (status != -ENOENT)
 834                         mlog_errno(status);
 835                 goto leave;
 836         }
 837         child_locked = 1;
 838
 839         if (S_ISDIR(inode->i_mode)) {
 840                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 841                         status = -ENOTEMPTY;
 842                         goto leave;
 843                 }
 844         }
 845
 846         status = ocfs2_remote_dentry_delete(dentry);
 847         if (status < 0) {
 848                 /* This remote delete should succeed under all normal
 849                  * circumstances. */
 850                 mlog_errno(status);
 851                 goto leave;
 852         }
 853
 854         if (inode_is_unlinkable(inode)) {
 855                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 856                                                   OCFS2_I(inode)->ip_blkno,
 857                                                   orphan_name, &orphan_insert);
 858                 if (status < 0) {
 859                         mlog_errno(status);
 860                         goto leave;
 861                 }
 862         }
 863
 864         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 865         if (IS_ERR(handle)) {
 866                 status = PTR_ERR(handle);
 867                 handle = NULL;
 868                 mlog_errno(status);
 869                 goto leave;
 870         }
 871
 872         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 873                                          OCFS2_JOURNAL_ACCESS_WRITE);
 874         if (status < 0) {
 875                 mlog_errno(status);
 876                 goto leave;
 877         }
 878
 879         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 880
 881         if (inode_is_unlinkable(inode)) {
 882                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 883                                           &orphan_insert, orphan_dir);
 884                 if (status < 0) {
 885                         mlog_errno(status);
 886                         goto leave;
 887                 }
 888         }
 889
 890         /* delete the name from the parent dir */
 891         status = ocfs2_delete_entry(handle, dir, &lookup);
 892         if (status < 0) {
 893                 mlog_errno(status);
 894                 goto leave;
 895         }
 896
 897         if (S_ISDIR(inode->i_mode))
 898                 drop_nlink(inode);
 899         drop_nlink(inode);
 900         ocfs2_set_links_count(fe, inode->i_nlink);
 901
 902         status = ocfs2_journal_dirty(handle, fe_bh);
 903         if (status < 0) {
 904                 mlog_errno(status);
 905                 goto leave;
 906         }
 907
 908         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 909         if (S_ISDIR(inode->i_mode))
 910                 drop_nlink(dir);
 911
 912         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 913         if (status < 0) {
 914                 mlog_errno(status);
 915                 if (S_ISDIR(inode->i_mode))
 916                         inc_nlink(dir);
 917         }
 918
 919 leave:
 920         if (handle)
 921                 ocfs2_commit_trans(osb, handle);
 922
 923         if (child_locked)
 924                 ocfs2_inode_unlock(inode, 1);
 925
 926         ocfs2_inode_unlock(dir, 1);
 927
 928         if (orphan_dir) {
 929                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 930                 ocfs2_inode_unlock(orphan_dir, 1);
 931                 mutex_unlock(&orphan_dir->i_mutex);
 932                 iput(orphan_dir);
 933         }
 934
 935         brelse(fe_bh);
 936         brelse(parent_node_bh);
 937
 938         ocfs2_free_dir_lookup_result(&orphan_insert);
 939         ocfs2_free_dir_lookup_result(&lookup);
 940
 941         mlog_exit(status);
 942
 943         return status;
 944 }
 945
 946 /*
 947  * The only place this should be used is rename!
 948  * if they have the same id, then the 1st one is the only one locked.
 949  */
 950 static int ocfs2_double_lock(struct ocfs2_super *osb,
 951                              struct buffer_head **bh1,
 952                              struct inode *inode1,
 953                              struct buffer_head **bh2,
 954                              struct inode *inode2)
 955 {
 956         int status;
 957         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 958         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 959         struct buffer_head **tmpbh;
 960         struct inode *tmpinode;
 961
 962         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 963                    (unsigned long long)oi1->ip_blkno,
 964                    (unsigned long long)oi2->ip_blkno);
 965
 966         if (*bh1)
 967                 *bh1 = NULL;
 968         if (*bh2)
 969                 *bh2 = NULL;
 970
 971         /* we always want to lock the one with the lower lockid first. */
 972         if (oi1->ip_blkno != oi2->ip_blkno) {
 973                 if (oi1->ip_blkno < oi2->ip_blkno) {
 974                         /* switch id1 and id2 around */
 975                         mlog(0, "switching them around...\n");
 976                         tmpbh = bh2;
 977                         bh2 = bh1;
 978                         bh1 = tmpbh;
 979
 980                         tmpinode = inode2;
 981                         inode2 = inode1;
 982                         inode1 = tmpinode;
 983                 }
 984                 /* lock id2 */
 985                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 986                                                  OI_LS_RENAME1);
 987                 if (status < 0) {
 988                         if (status != -ENOENT)
 989                                 mlog_errno(status);
 990                         goto bail;
 991                 }
 992         }
 993
 994         /* lock id1 */
 995         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
 996         if (status < 0) {
 997                 /*
 998                  * An error return must mean that no cluster locks
 999                  * were held on function exit.
1000                  */
1001                 if (oi1->ip_blkno != oi2->ip_blkno)
1002                         ocfs2_inode_unlock(inode2, 1);
1003
1004                 if (status != -ENOENT)
1005                         mlog_errno(status);
1006         }
1007
1008 bail:
1009         mlog_exit(status);
1010         return status;
1011 }
1012
1013 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1014 {
1015         ocfs2_inode_unlock(inode1, 1);
1016
1017         if (inode1 != inode2)
1018                 ocfs2_inode_unlock(inode2, 1);
1019 }
1020
1021 static int ocfs2_rename(struct inode *old_dir,
1022                         struct dentry *old_dentry,
1023                         struct inode *new_dir,
1024                         struct dentry *new_dentry)
1025 {
1026         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1027         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1028         struct inode *old_inode = old_dentry->d_inode;
1029         struct inode *new_inode = new_dentry->d_inode;
1030         struct inode *orphan_dir = NULL;
1031         struct ocfs2_dinode *newfe = NULL;
1032         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1033         struct buffer_head *newfe_bh = NULL;
1034         struct buffer_head *old_inode_bh = NULL;
1035         struct ocfs2_super *osb = NULL;
1036         u64 newfe_blkno, old_de_ino;
1037         handle_t *handle = NULL;
1038         struct buffer_head *old_dir_bh = NULL;
1039         struct buffer_head *new_dir_bh = NULL;
1040         nlink_t old_dir_nlink = old_dir->i_nlink;
1041         struct ocfs2_dinode *old_di;
1042         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1043         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1044         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1045         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1046         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1047
1048         /* At some point it might be nice to break this function up a
1049          * bit. */
1050
1051         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1052                    old_dir, old_dentry, new_dir, new_dentry,
1053                    old_dentry->d_name.len, old_dentry->d_name.name,
1054                    new_dentry->d_name.len, new_dentry->d_name.name);
1055
1056         dquot_initialize(old_dir);
1057         dquot_initialize(new_dir);
1058
1059         osb = OCFS2_SB(old_dir->i_sb);
1060
1061         if (new_inode) {
1062                 if (!igrab(new_inode))
1063                         BUG();
1064         }
1065
1066         /* Assume a directory hierarchy thusly:
1067          * a/b/c
1068          * a/d
1069          * a,b,c, and d are all directories.
1070          *
1071          * from cwd of 'a' on both nodes:
1072          * node1: mv b/c d
1073          * node2: mv d   b/c
1074          *
1075          * And that's why, just like the VFS, we need a file system
1076          * rename lock. */
1077         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1078                 status = ocfs2_rename_lock(osb);
1079                 if (status < 0) {
1080                         mlog_errno(status);
1081                         goto bail;
1082                 }
1083                 rename_lock = 1;
1084         }
1085
1086         /* if old and new are the same, this'll just do one lock. */
1087         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1088                                    &new_dir_bh, new_dir);
1089         if (status < 0) {
1090                 mlog_errno(status);
1091                 goto bail;
1092         }
1093         parents_locked = 1;
1094
1095         /* make sure both dirs have bhs
1096          * get an extra ref on old_dir_bh if old==new */
1097         if (!new_dir_bh) {
1098                 if (old_dir_bh) {
1099                         new_dir_bh = old_dir_bh;
1100                         get_bh(new_dir_bh);
1101                 } else {
1102                         mlog(ML_ERROR, "no old_dir_bh!\n");
1103                         status = -EIO;
1104                         goto bail;
1105                 }
1106         }
1107
1108         /*
1109          * Aside from allowing a meta data update, the locking here
1110          * also ensures that the downconvert thread on other nodes
1111          * won't have to concurrently downconvert the inode and the
1112          * dentry locks.
1113          */
1114         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1115                                          OI_LS_PARENT);
1116         if (status < 0) {
1117                 if (status != -ENOENT)
1118                         mlog_errno(status);
1119                 goto bail;
1120         }
1121         old_child_locked = 1;
1122
1123         status = ocfs2_remote_dentry_delete(old_dentry);
1124         if (status < 0) {
1125                 mlog_errno(status);
1126                 goto bail;
1127         }
1128
1129         if (S_ISDIR(old_inode->i_mode)) {
1130                 u64 old_inode_parent;
1131
1132                 update_dot_dot = 1;
1133                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1134                                                   old_inode,
1135                                                   &old_inode_dot_dot_res);
1136                 if (status) {
1137                         status = -EIO;
1138                         goto bail;
1139                 }
1140
1141                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1142                         status = -EIO;
1143                         goto bail;
1144                 }
1145
1146                 if (!new_inode && new_dir != old_dir &&
1147                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1148                         status = -EMLINK;
1149                         goto bail;
1150                 }
1151         }
1152
1153         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1154                                             old_dentry->d_name.len,
1155                                             &old_de_ino);
1156         if (status) {
1157                 status = -ENOENT;
1158                 goto bail;
1159         }
1160
1161         /*
1162          *  Check for inode number is _not_ due to possible IO errors.
1163          *  We might rmdir the source, keep it as pwd of some process
1164          *  and merrily kill the link to whatever was created under the
1165          *  same name. Goodbye sticky bit ;-<
1166          */
1167         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1168                 status = -ENOENT;
1169                 goto bail;
1170         }
1171
1172         /* check if the target already exists (in which case we need
1173          * to delete it */
1174         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1175                                           new_dentry->d_name.len,
1176                                           &newfe_blkno, new_dir,
1177                                           &target_lookup_res);
1178         /* The only error we allow here is -ENOENT because the new
1179          * file not existing is perfectly valid. */
1180         if ((status < 0) && (status != -ENOENT)) {
1181                 /* If we cannot find the file specified we should just */
1182                 /* return the error... */
1183                 mlog_errno(status);
1184                 goto bail;
1185         }
1186         if (status == 0)
1187                 target_exists = 1;
1188
1189         if (!target_exists && new_inode) {
1190                 /*
1191                  * Target was unlinked by another node while we were
1192                  * waiting to get to ocfs2_rename(). There isn't
1193                  * anything we can do here to help the situation, so
1194                  * bubble up the appropriate error.
1195                  */
1196                 status = -ENOENT;
1197                 goto bail;
1198         }
1199
1200         /* In case we need to overwrite an existing file, we blow it
1201          * away first */
1202         if (target_exists) {
1203                 /* VFS didn't think there existed an inode here, but
1204                  * someone else in the cluster must have raced our
1205                  * rename to create one. Today we error cleanly, in
1206                  * the future we should consider calling iget to build
1207                  * a new struct inode for this entry. */
1208                 if (!new_inode) {
1209                         status = -EACCES;
1210
1211                         mlog(0, "We found an inode for name %.*s but VFS "
1212                              "didn't give us one.\n", new_dentry->d_name.len,
1213                              new_dentry->d_name.name);
1214                         goto bail;
1215                 }
1216
1217                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1218                         status = -EACCES;
1219
1220                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1221                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1222                              (unsigned long long)newfe_blkno,
1223                              OCFS2_I(new_inode)->ip_flags);
1224                         goto bail;
1225                 }
1226
1227                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1228                 if (status < 0) {
1229                         if (status != -ENOENT)
1230                                 mlog_errno(status);
1231                         goto bail;
1232                 }
1233                 new_child_locked = 1;
1234
1235                 status = ocfs2_remote_dentry_delete(new_dentry);
1236                 if (status < 0) {
1237                         mlog_errno(status);
1238                         goto bail;
1239                 }
1240
1241                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1242
1243                 mlog(0, "aha rename over existing... new_blkno=%llu "
1244                      "newfebh=%p bhblocknr=%llu\n",
1245                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1246                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1247
1248                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1249                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1250                                                 OCFS2_I(new_inode)->ip_blkno,
1251                                                 orphan_name, &orphan_insert);
1252                         if (status < 0) {
1253                                 mlog_errno(status);
1254                                 goto bail;
1255                         }
1256                 }
1257         } else {
1258                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1259
1260                 status = ocfs2_check_dir_for_entry(new_dir,
1261                                                    new_dentry->d_name.name,
1262                                                    new_dentry->d_name.len);
1263                 if (status)
1264                         goto bail;
1265
1266                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1267                                                       new_dentry->d_name.name,
1268                                                       new_dentry->d_name.len,
1269                                                       &target_insert);
1270                 if (status < 0) {
1271                         mlog_errno(status);
1272                         goto bail;
1273                 }
1274         }
1275
1276         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1277         if (IS_ERR(handle)) {
1278                 status = PTR_ERR(handle);
1279                 handle = NULL;
1280                 mlog_errno(status);
1281                 goto bail;
1282         }
1283
1284         if (target_exists) {
1285                 if (S_ISDIR(new_inode->i_mode)) {
1286                         if (new_inode->i_nlink != 2 ||
1287                             !ocfs2_empty_dir(new_inode)) {
1288                                 status = -ENOTEMPTY;
1289                                 goto bail;
1290                         }
1291                 }
1292                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1293                                                  newfe_bh,
1294                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1295                 if (status < 0) {
1296                         mlog_errno(status);
1297                         goto bail;
1298                 }
1299
1300                 if (S_ISDIR(new_inode->i_mode) ||
1301                     (ocfs2_read_links_count(newfe) == 1)) {
1302                         status = ocfs2_orphan_add(osb, handle, new_inode,
1303                                                   newfe_bh, orphan_name,
1304                                                   &orphan_insert, orphan_dir);
1305                         if (status < 0) {
1306                                 mlog_errno(status);
1307                                 goto bail;
1308                         }
1309                 }
1310
1311                 /* change the dirent to point to the correct inode */
1312                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1313                                             old_inode);
1314                 if (status < 0) {
1315                         mlog_errno(status);
1316                         goto bail;
1317                 }
1318                 new_dir->i_version++;
1319
1320                 if (S_ISDIR(new_inode->i_mode))
1321                         ocfs2_set_links_count(newfe, 0);
1322                 else
1323                         ocfs2_add_links_count(newfe, -1);
1324
1325                 status = ocfs2_journal_dirty(handle, newfe_bh);
1326                 if (status < 0) {
1327                         mlog_errno(status);
1328                         goto bail;
1329                 }
1330         } else {
1331                 /* if the name was not found in new_dir, add it now */
1332                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1333                                          OCFS2_I(old_inode)->ip_blkno,
1334                                          new_dir_bh, &target_insert);
1335         }
1336
1337         old_inode->i_ctime = CURRENT_TIME;
1338         mark_inode_dirty(old_inode);
1339
1340         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1341                                          old_inode_bh,
1342                                          OCFS2_JOURNAL_ACCESS_WRITE);
1343         if (status >= 0) {
1344                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1345
1346                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1347                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1348
1349                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1350                 if (status < 0)
1351                         mlog_errno(status);
1352         } else
1353                 mlog_errno(status);
1354
1355         /*
1356          * Now that the name has been added to new_dir, remove the old name.
1357          *
1358          * We don't keep any directory entry context around until now
1359          * because the insert might have changed the type of directory
1360          * we're dealing with.
1361          */
1362         status = ocfs2_find_entry(old_dentry->d_name.name,
1363                                   old_dentry->d_name.len, old_dir,
1364                                   &old_entry_lookup);
1365         if (status)
1366                 goto bail;
1367
1368         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1369         if (status < 0) {
1370                 mlog_errno(status);
1371                 goto bail;
1372         }
1373
1374         if (new_inode) {
1375                 new_inode->i_nlink--;
1376                 new_inode->i_ctime = CURRENT_TIME;
1377         }
1378         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1379
1380         if (update_dot_dot) {
1381                 status = ocfs2_update_entry(old_inode, handle,
1382                                             &old_inode_dot_dot_res, new_dir);
1383                 old_dir->i_nlink--;
1384                 if (new_inode) {
1385                         new_inode->i_nlink--;
1386                 } else {
1387                         inc_nlink(new_dir);
1388                         mark_inode_dirty(new_dir);
1389                 }
1390         }
1391         mark_inode_dirty(old_dir);
1392         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1393         if (new_inode) {
1394                 mark_inode_dirty(new_inode);
1395                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1396         }
1397
1398         if (old_dir != new_dir) {
1399                 /* Keep the same times on both directories.*/
1400                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1401
1402                 /*
1403                  * This will also pick up the i_nlink change from the
1404                  * block above.
1405                  */
1406                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1407         }
1408
1409         if (old_dir_nlink != old_dir->i_nlink) {
1410                 if (!old_dir_bh) {
1411                         mlog(ML_ERROR, "need to change nlink for old dir "
1412                              "%llu from %d to %d but bh is NULL!\n",
1413                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1414                              (int)old_dir_nlink, old_dir->i_nlink);
1415                 } else {
1416                         struct ocfs2_dinode *fe;
1417                         status = ocfs2_journal_access_di(handle,
1418                                                          INODE_CACHE(old_dir),
1419                                                          old_dir_bh,
1420                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1421                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1422                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1423                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1424                 }
1425         }
1426         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1427         status = 0;
1428 bail:
1429         if (rename_lock)
1430                 ocfs2_rename_unlock(osb);
1431
1432         if (handle)
1433                 ocfs2_commit_trans(osb, handle);
1434
1435         if (parents_locked)
1436                 ocfs2_double_unlock(old_dir, new_dir);
1437
1438         if (old_child_locked)
1439                 ocfs2_inode_unlock(old_inode, 1);
1440
1441         if (new_child_locked)
1442                 ocfs2_inode_unlock(new_inode, 1);
1443
1444         if (orphan_dir) {
1445                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1446                 ocfs2_inode_unlock(orphan_dir, 1);
1447                 mutex_unlock(&orphan_dir->i_mutex);
1448                 iput(orphan_dir);
1449         }
1450
1451         if (new_inode)
1452                 sync_mapping_buffers(old_inode->i_mapping);
1453
1454         if (new_inode)
1455                 iput(new_inode);
1456
1457         ocfs2_free_dir_lookup_result(&target_lookup_res);
1458         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1459         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1460         ocfs2_free_dir_lookup_result(&orphan_insert);
1461         ocfs2_free_dir_lookup_result(&target_insert);
1462
1463         brelse(newfe_bh);
1464         brelse(old_inode_bh);
1465         brelse(old_dir_bh);
1466         brelse(new_dir_bh);
1467
1468         mlog_exit(status);
1469
1470         return status;
1471 }
1472
1473 /*
1474  * we expect i_size = strlen(symname). Copy symname into the file
1475  * data, including the null terminator.
1476  */
1477 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1478                                      handle_t *handle,
1479                                      struct inode *inode,
1480                                      const char *symname)
1481 {
1482         struct buffer_head **bhs = NULL;
1483         const char *c;
1484         struct super_block *sb = osb->sb;
1485         u64 p_blkno, p_blocks;
1486         int virtual, blocks, status, i, bytes_left;
1487
1488         bytes_left = i_size_read(inode) + 1;
1489         /* we can't trust i_blocks because we're actually going to
1490          * write i_size + 1 bytes. */
1491         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1492
1493         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1494                         (unsigned long long)inode->i_blocks,
1495                         i_size_read(inode), blocks);
1496
1497         /* Sanity check -- make sure we're going to fit. */
1498         if (bytes_left >
1499             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1500                 status = -EIO;
1501                 mlog_errno(status);
1502                 goto bail;
1503         }
1504
1505         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1506         if (!bhs) {
1507                 status = -ENOMEM;
1508                 mlog_errno(status);
1509                 goto bail;
1510         }
1511
1512         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1513                                              NULL);
1514         if (status < 0) {
1515                 mlog_errno(status);
1516                 goto bail;
1517         }
1518
1519         /* links can never be larger than one cluster so we know this
1520          * is all going to be contiguous, but do a sanity check
1521          * anyway. */
1522         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1523                 status = -EIO;
1524                 mlog_errno(status);
1525                 goto bail;
1526         }
1527
1528         virtual = 0;
1529         while(bytes_left > 0) {
1530                 c = &symname[virtual * sb->s_blocksize];
1531
1532                 bhs[virtual] = sb_getblk(sb, p_blkno);
1533                 if (!bhs[virtual]) {
1534                         status = -ENOMEM;
1535                         mlog_errno(status);
1536                         goto bail;
1537                 }
1538                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1539                                               bhs[virtual]);
1540
1541                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1542                                               bhs[virtual],
1543                                               OCFS2_JOURNAL_ACCESS_CREATE);
1544                 if (status < 0) {
1545                         mlog_errno(status);
1546                         goto bail;
1547                 }
1548
1549                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1550
1551                 memcpy(bhs[virtual]->b_data, c,
1552                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1553                        bytes_left);
1554
1555                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1556                 if (status < 0) {
1557                         mlog_errno(status);
1558                         goto bail;
1559                 }
1560
1561                 virtual++;
1562                 p_blkno++;
1563                 bytes_left -= sb->s_blocksize;
1564         }
1565
1566         status = 0;
1567 bail:
1568
1569         if (bhs) {
1570                 for(i = 0; i < blocks; i++)
1571                         brelse(bhs[i]);
1572                 kfree(bhs);
1573         }
1574
1575         mlog_exit(status);
1576         return status;
1577 }
1578
1579 static int ocfs2_symlink(struct inode *dir,
1580                          struct dentry *dentry,
1581                          const char *symname)
1582 {
1583         int status, l, credits;
1584         u64 newsize;
1585         struct ocfs2_super *osb = NULL;
1586         struct inode *inode = NULL;
1587         struct super_block *sb;
1588         struct buffer_head *new_fe_bh = NULL;
1589         struct buffer_head *parent_fe_bh = NULL;
1590         struct ocfs2_dinode *fe = NULL;
1591         struct ocfs2_dinode *dirfe;
1592         handle_t *handle = NULL;
1593         struct ocfs2_alloc_context *inode_ac = NULL;
1594         struct ocfs2_alloc_context *data_ac = NULL;
1595         struct ocfs2_alloc_context *xattr_ac = NULL;
1596         int want_clusters = 0;
1597         int xattr_credits = 0;
1598         struct ocfs2_security_xattr_info si = {
1599                 .enable = 1,
1600         };
1601         int did_quota = 0, did_quota_inode = 0;
1602         struct ocfs2_dir_lookup_result lookup = { NULL, };
1603
1604         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1605                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1606
1607         dquot_initialize(dir);
1608
1609         sb = dir->i_sb;
1610         osb = OCFS2_SB(sb);
1611
1612         l = strlen(symname) + 1;
1613
1614         credits = ocfs2_calc_symlink_credits(sb);
1615
1616         /* lock the parent directory */
1617         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1618         if (status < 0) {
1619                 if (status != -ENOENT)
1620                         mlog_errno(status);
1621                 return status;
1622         }
1623
1624         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1625         if (!ocfs2_read_links_count(dirfe)) {
1626                 /* can't make a file in a deleted directory. */
1627                 status = -ENOENT;
1628                 goto bail;
1629         }
1630
1631         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1632                                            dentry->d_name.len);
1633         if (status)
1634                 goto bail;
1635
1636         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1637                                               dentry->d_name.name,
1638                                               dentry->d_name.len, &lookup);
1639         if (status < 0) {
1640                 mlog_errno(status);
1641                 goto bail;
1642         }
1643
1644         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1645         if (status < 0) {
1646                 if (status != -ENOSPC)
1647                         mlog_errno(status);
1648                 goto bail;
1649         }
1650
1651         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1652         if (!inode) {
1653                 status = -ENOMEM;
1654                 mlog_errno(status);
1655                 goto bail;
1656         }
1657
1658         /* get security xattr */
1659         status = ocfs2_init_security_get(inode, dir, &si);
1660         if (status) {
1661                 if (status == -EOPNOTSUPP)
1662                         si.enable = 0;
1663                 else {
1664                         mlog_errno(status);
1665                         goto bail;
1666                 }
1667         }
1668
1669         /* calculate meta data/clusters for setting security xattr */
1670         if (si.enable) {
1671                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1672                                                   &xattr_credits, &xattr_ac);
1673                 if (status < 0) {
1674                         mlog_errno(status);
1675                         goto bail;
1676                 }
1677         }
1678
1679         /* don't reserve bitmap space for fast symlinks. */
1680         if (l > ocfs2_fast_symlink_chars(sb))
1681                 want_clusters += 1;
1682
1683         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1684         if (status < 0) {
1685                 if (status != -ENOSPC)
1686                         mlog_errno(status);
1687                 goto bail;
1688         }
1689
1690         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1691         if (IS_ERR(handle)) {
1692                 status = PTR_ERR(handle);
1693                 handle = NULL;
1694                 mlog_errno(status);
1695                 goto bail;
1696         }
1697
1698         status = dquot_alloc_inode(inode);
1699         if (status)
1700                 goto bail;
1701         did_quota_inode = 1;
1702
1703         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1704                    inode->i_mode, dentry->d_name.len,
1705                    dentry->d_name.name);
1706
1707         status = ocfs2_mknod_locked(osb, dir, inode,
1708                                     0, &new_fe_bh, parent_fe_bh, handle,
1709                                     inode_ac);
1710         if (status < 0) {
1711                 mlog_errno(status);
1712                 goto bail;
1713         }
1714
1715         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1716         inode->i_rdev = 0;
1717         newsize = l - 1;
1718         if (l > ocfs2_fast_symlink_chars(sb)) {
1719                 u32 offset = 0;
1720
1721                 inode->i_op = &ocfs2_symlink_inode_operations;
1722                 status = dquot_alloc_space_nodirty(inode,
1723                     ocfs2_clusters_to_bytes(osb->sb, 1));
1724                 if (status)
1725                         goto bail;
1726                 did_quota = 1;
1727                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1728                                               new_fe_bh,
1729                                               handle, data_ac, NULL,
1730                                               NULL);
1731                 if (status < 0) {
1732                         if (status != -ENOSPC && status != -EINTR) {
1733                                 mlog(ML_ERROR,
1734                                      "Failed to extend file to %llu\n",
1735                                      (unsigned long long)newsize);
1736                                 mlog_errno(status);
1737                                 status = -ENOSPC;
1738                         }
1739                         goto bail;
1740                 }
1741                 i_size_write(inode, newsize);
1742                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1743         } else {
1744                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1745                 memcpy((char *) fe->id2.i_symlink, symname, l);
1746                 i_size_write(inode, newsize);
1747                 inode->i_blocks = 0;
1748         }
1749
1750         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1751         if (status < 0) {
1752                 mlog_errno(status);
1753                 goto bail;
1754         }
1755
1756         if (!ocfs2_inode_is_fast_symlink(inode)) {
1757                 status = ocfs2_create_symlink_data(osb, handle, inode,
1758                                                    symname);
1759                 if (status < 0) {
1760                         mlog_errno(status);
1761                         goto bail;
1762                 }
1763         }
1764
1765         if (si.enable) {
1766                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1767                                                  xattr_ac, data_ac);
1768                 if (status < 0) {
1769                         mlog_errno(status);
1770                         goto bail;
1771                 }
1772         }
1773
1774         status = ocfs2_add_entry(handle, dentry, inode,
1775                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1776                                  &lookup);
1777         if (status < 0) {
1778                 mlog_errno(status);
1779                 goto bail;
1780         }
1781
1782         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1783         if (status) {
1784                 mlog_errno(status);
1785                 goto bail;
1786         }
1787
1788         insert_inode_hash(inode);
1789         dentry->d_op = &ocfs2_dentry_ops;
1790         d_instantiate(dentry, inode);
1791 bail:
1792         if (status < 0 && did_quota)
1793                 dquot_free_space_nodirty(inode,
1794                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1795         if (status < 0 && did_quota_inode)
1796                 dquot_free_inode(inode);
1797         if (handle)
1798                 ocfs2_commit_trans(osb, handle);
1799
1800         ocfs2_inode_unlock(dir, 1);
1801
1802         brelse(new_fe_bh);
1803         brelse(parent_fe_bh);
1804         kfree(si.name);
1805         kfree(si.value);
1806         ocfs2_free_dir_lookup_result(&lookup);
1807         if (inode_ac)
1808                 ocfs2_free_alloc_context(inode_ac);
1809         if (data_ac)
1810                 ocfs2_free_alloc_context(data_ac);
1811         if (xattr_ac)
1812                 ocfs2_free_alloc_context(xattr_ac);
1813         if ((status < 0) && inode) {
1814                 clear_nlink(inode);
1815                 iput(inode);
1816         }
1817
1818         mlog_exit(status);
1819
1820         return status;
1821 }
1822
1823 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1824 {
1825         int status, namelen;
1826
1827         mlog_entry_void();
1828
1829         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1830                            (long long)blkno);
1831         if (namelen <= 0) {
1832                 if (namelen)
1833                         status = namelen;
1834                 else
1835                         status = -EINVAL;
1836                 mlog_errno(status);
1837                 goto bail;
1838         }
1839         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1840                 status = -EINVAL;
1841                 mlog_errno(status);
1842                 goto bail;
1843         }
1844
1845         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1846              namelen);
1847
1848         status = 0;
1849 bail:
1850         mlog_exit(status);
1851         return status;
1852 }
1853
1854 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1855                                     struct inode **ret_orphan_dir,
1856                                     u64 blkno,
1857                                     char *name,
1858                                     struct ocfs2_dir_lookup_result *lookup)
1859 {
1860         struct inode *orphan_dir_inode;
1861         struct buffer_head *orphan_dir_bh = NULL;
1862         int status = 0;
1863
1864         status = ocfs2_blkno_stringify(blkno, name);
1865         if (status < 0) {
1866                 mlog_errno(status);
1867                 return status;
1868         }
1869
1870         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1871                                                        ORPHAN_DIR_SYSTEM_INODE,
1872                                                        osb->slot_num);
1873         if (!orphan_dir_inode) {
1874                 status = -ENOENT;
1875                 mlog_errno(status);
1876                 return status;
1877         }
1878
1879         mutex_lock(&orphan_dir_inode->i_mutex);
1880
1881         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1882         if (status < 0) {
1883                 mlog_errno(status);
1884                 goto leave;
1885         }
1886
1887         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1888                                               orphan_dir_bh, name,
1889                                               OCFS2_ORPHAN_NAMELEN, lookup);
1890         if (status < 0) {
1891                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1892
1893                 mlog_errno(status);
1894                 goto leave;
1895         }
1896
1897         *ret_orphan_dir = orphan_dir_inode;
1898
1899 leave:
1900         if (status) {
1901                 mutex_unlock(&orphan_dir_inode->i_mutex);
1902                 iput(orphan_dir_inode);
1903         }
1904
1905         brelse(orphan_dir_bh);
1906
1907         mlog_exit(status);
1908         return status;
1909 }
1910
1911 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1912                             handle_t *handle,
1913                             struct inode *inode,
1914                             struct buffer_head *fe_bh,
1915                             char *name,
1916                             struct ocfs2_dir_lookup_result *lookup,
1917                             struct inode *orphan_dir_inode)
1918 {
1919         struct buffer_head *orphan_dir_bh = NULL;
1920         int status = 0;
1921         struct ocfs2_dinode *orphan_fe;
1922         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1923
1924         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1925
1926         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1927         if (status < 0) {
1928                 mlog_errno(status);
1929                 goto leave;
1930         }
1931
1932         status = ocfs2_journal_access_di(handle,
1933                                          INODE_CACHE(orphan_dir_inode),
1934                                          orphan_dir_bh,
1935                                          OCFS2_JOURNAL_ACCESS_WRITE);
1936         if (status < 0) {
1937                 mlog_errno(status);
1938                 goto leave;
1939         }
1940
1941         /* we're a cluster, and nlink can change on disk from
1942          * underneath us... */
1943         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1944         if (S_ISDIR(inode->i_mode))
1945                 ocfs2_add_links_count(orphan_fe, 1);
1946         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1947
1948         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1949         if (status < 0) {
1950                 mlog_errno(status);
1951                 goto leave;
1952         }
1953
1954         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1955                                    OCFS2_ORPHAN_NAMELEN, inode,
1956                                    OCFS2_I(inode)->ip_blkno,
1957                                    orphan_dir_bh, lookup);
1958         if (status < 0) {
1959                 mlog_errno(status);
1960                 goto leave;
1961         }
1962
1963         /*
1964          * We're going to journal the change of i_flags and i_orphaned_slot.
1965          * It's safe anyway, though some callers may duplicate the journaling.
1966          * Journaling within the func just make the logic look more
1967          * straightforward.
1968          */
1969         status = ocfs2_journal_access_di(handle,
1970                                          INODE_CACHE(inode),
1971                                          fe_bh,
1972                                          OCFS2_JOURNAL_ACCESS_WRITE);
1973         if (status < 0) {
1974                 mlog_errno(status);
1975                 goto leave;
1976         }
1977
1978         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1979
1980         /* Record which orphan dir our inode now resides
1981          * in. delete_inode will use this to determine which orphan
1982          * dir to lock. */
1983         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1984
1985         ocfs2_journal_dirty(handle, fe_bh);
1986
1987         mlog(0, "Inode %llu orphaned in slot %d\n",
1988              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1989
1990 leave:
1991         brelse(orphan_dir_bh);
1992
1993         mlog_exit(status);
1994         return status;
1995 }
1996
1997 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1998 int ocfs2_orphan_del(struct ocfs2_super *osb,
1999                      handle_t *handle,
2000                      struct inode *orphan_dir_inode,
2001                      struct inode *inode,
2002                      struct buffer_head *orphan_dir_bh)
2003 {
2004         char name[OCFS2_ORPHAN_NAMELEN + 1];
2005         struct ocfs2_dinode *orphan_fe;
2006         int status = 0;
2007         struct ocfs2_dir_lookup_result lookup = { NULL, };
2008
2009         mlog_entry_void();
2010
2011         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2012         if (status < 0) {
2013                 mlog_errno(status);
2014                 goto leave;
2015         }
2016
2017         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2018              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2019              OCFS2_ORPHAN_NAMELEN);
2020
2021         /* find it's spot in the orphan directory */
2022         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2023                                   &lookup);
2024         if (status) {
2025                 mlog_errno(status);
2026                 goto leave;
2027         }
2028
2029         /* remove it from the orphan directory */
2030         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2031         if (status < 0) {
2032                 mlog_errno(status);
2033                 goto leave;
2034         }
2035
2036         status = ocfs2_journal_access_di(handle,
2037                                          INODE_CACHE(orphan_dir_inode),
2038                                          orphan_dir_bh,
2039                                          OCFS2_JOURNAL_ACCESS_WRITE);
2040         if (status < 0) {
2041                 mlog_errno(status);
2042                 goto leave;
2043         }
2044
2045         /* do the i_nlink dance! :) */
2046         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2047         if (S_ISDIR(inode->i_mode))
2048                 ocfs2_add_links_count(orphan_fe, -1);
2049         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2050
2051         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2052         if (status < 0) {
2053                 mlog_errno(status);
2054                 goto leave;
2055         }
2056
2057 leave:
2058         ocfs2_free_dir_lookup_result(&lookup);
2059
2060         mlog_exit(status);
2061         return status;
2062 }
2063
2064 int ocfs2_create_inode_in_orphan(struct inode *dir,
2065                                  int mode,
2066                                  struct inode **new_inode)
2067 {
2068         int status, did_quota_inode = 0;
2069         struct inode *inode = NULL;
2070         struct inode *orphan_dir = NULL;
2071         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2072         struct ocfs2_dinode *di = NULL;
2073         handle_t *handle = NULL;
2074         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2075         struct buffer_head *parent_di_bh = NULL;
2076         struct buffer_head *new_di_bh = NULL;
2077         struct ocfs2_alloc_context *inode_ac = NULL;
2078         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2079
2080         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2081         if (status < 0) {
2082                 if (status != -ENOENT)
2083                         mlog_errno(status);
2084                 return status;
2085         }
2086
2087         /*
2088          * We give the orphan dir the root blkno to fake an orphan name,
2089          * and allocate enough space for our insertion.
2090          */
2091         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
2092                                           osb->root_blkno,
2093                                           orphan_name, &orphan_insert);
2094         if (status < 0) {
2095                 mlog_errno(status);
2096                 goto leave;
2097         }
2098
2099         /* reserve an inode spot */
2100         status = ocfs2_reserve_new_inode(osb, &inode_ac);
2101         if (status < 0) {
2102                 if (status != -ENOSPC)
2103                         mlog_errno(status);
2104                 goto leave;
2105         }
2106
2107         inode = ocfs2_get_init_inode(dir, mode);
2108         if (!inode) {
2109                 status = -ENOMEM;
2110                 mlog_errno(status);
2111                 goto leave;
2112         }
2113
2114         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2115         if (IS_ERR(handle)) {
2116                 status = PTR_ERR(handle);
2117                 handle = NULL;
2118                 mlog_errno(status);
2119                 goto leave;
2120         }
2121
2122         status = dquot_alloc_inode(inode);
2123         if (status)
2124                 goto leave;
2125         did_quota_inode = 1;
2126
2127         inode->i_nlink = 0;
2128         /* do the real work now. */
2129         status = ocfs2_mknod_locked(osb, dir, inode,
2130                                     0, &new_di_bh, parent_di_bh, handle,
2131                                     inode_ac);
2132         if (status < 0) {
2133                 mlog_errno(status);
2134                 goto leave;
2135         }
2136
2137         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
2138         if (status < 0) {
2139                 mlog_errno(status);
2140                 goto leave;
2141         }
2142
2143         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2144         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2145                                   &orphan_insert, orphan_dir);
2146         if (status < 0) {
2147                 mlog_errno(status);
2148                 goto leave;
2149         }
2150
2151         /* get open lock so that only nodes can't remove it from orphan dir. */
2152         status = ocfs2_open_lock(inode);
2153         if (status < 0)
2154                 mlog_errno(status);
2155
2156         insert_inode_hash(inode);
2157 leave:
2158         if (status < 0 && did_quota_inode)
2159                 dquot_free_inode(inode);
2160         if (handle)
2161                 ocfs2_commit_trans(osb, handle);
2162
2163         if (orphan_dir) {
2164                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2165                 ocfs2_inode_unlock(orphan_dir, 1);
2166                 mutex_unlock(&orphan_dir->i_mutex);
2167                 iput(orphan_dir);
2168         }
2169
2170         if (status == -ENOSPC)
2171                 mlog(0, "Disk is full\n");
2172
2173         if ((status < 0) && inode) {
2174                 clear_nlink(inode);
2175                 iput(inode);
2176         }
2177
2178         if (inode_ac)
2179                 ocfs2_free_alloc_context(inode_ac);
2180
2181         brelse(new_di_bh);
2182
2183         if (!status)
2184                 *new_inode = inode;
2185
2186         ocfs2_free_dir_lookup_result(&orphan_insert);
2187
2188         ocfs2_inode_unlock(dir, 1);
2189         brelse(parent_di_bh);
2190         return status;
2191 }
2192
2193 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2194                                    struct inode *inode,
2195                                    struct dentry *dentry)
2196 {
2197         int status = 0;
2198         struct buffer_head *parent_di_bh = NULL;
2199         handle_t *handle = NULL;
2200         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2201         struct ocfs2_dinode *dir_di, *di;
2202         struct inode *orphan_dir_inode = NULL;
2203         struct buffer_head *orphan_dir_bh = NULL;
2204         struct buffer_head *di_bh = NULL;
2205         struct ocfs2_dir_lookup_result lookup = { NULL, };
2206
2207         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2208                    dentry->d_name.len, dentry->d_name.name);
2209
2210         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2211         if (status < 0) {
2212                 if (status != -ENOENT)
2213                         mlog_errno(status);
2214                 return status;
2215         }
2216
2217         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2218         if (!dir_di->i_links_count) {
2219                 /* can't make a file in a deleted directory. */
2220                 status = -ENOENT;
2221                 goto leave;
2222         }
2223
2224         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2225                                            dentry->d_name.len);
2226         if (status)
2227                 goto leave;
2228
2229         /* get a spot inside the dir. */
2230         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2231                                               dentry->d_name.name,
2232                                               dentry->d_name.len, &lookup);
2233         if (status < 0) {
2234                 mlog_errno(status);
2235                 goto leave;
2236         }
2237
2238         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2239                                                        ORPHAN_DIR_SYSTEM_INODE,
2240                                                        osb->slot_num);
2241         if (!orphan_dir_inode) {
2242                 status = -EEXIST;
2243                 mlog_errno(status);
2244                 goto leave;
2245         }
2246
2247         mutex_lock(&orphan_dir_inode->i_mutex);
2248
2249         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2250         if (status < 0) {
2251                 mlog_errno(status);
2252                 mutex_unlock(&orphan_dir_inode->i_mutex);
2253                 iput(orphan_dir_inode);
2254                 goto leave;
2255         }
2256
2257         status = ocfs2_read_inode_block(inode, &di_bh);
2258         if (status < 0) {
2259                 mlog_errno(status);
2260                 goto orphan_unlock;
2261         }
2262
2263         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2264         if (IS_ERR(handle)) {
2265                 status = PTR_ERR(handle);
2266                 handle = NULL;
2267                 mlog_errno(status);
2268                 goto orphan_unlock;
2269         }
2270
2271         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2272                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2273         if (status < 0) {
2274                 mlog_errno(status);
2275                 goto out_commit;
2276         }
2277
2278         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2279                                   orphan_dir_bh);
2280         if (status < 0) {
2281                 mlog_errno(status);
2282                 goto out_commit;
2283         }
2284
2285         di = (struct ocfs2_dinode *)di_bh->b_data;
2286         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2287         di->i_orphaned_slot = 0;
2288         inode->i_nlink = 1;
2289         ocfs2_set_links_count(di, inode->i_nlink);
2290         ocfs2_journal_dirty(handle, di_bh);
2291
2292         status = ocfs2_add_entry(handle, dentry, inode,
2293                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2294                                  &lookup);
2295         if (status < 0) {
2296                 mlog_errno(status);
2297                 goto out_commit;
2298         }
2299
2300         status = ocfs2_dentry_attach_lock(dentry, inode,
2301                                           OCFS2_I(dir)->ip_blkno);
2302         if (status) {
2303                 mlog_errno(status);
2304                 goto out_commit;
2305         }
2306
2307         dentry->d_op = &ocfs2_dentry_ops;
2308         d_instantiate(dentry, inode);
2309         status = 0;
2310 out_commit:
2311         ocfs2_commit_trans(osb, handle);
2312 orphan_unlock:
2313         ocfs2_inode_unlock(orphan_dir_inode, 1);
2314         mutex_unlock(&orphan_dir_inode->i_mutex);
2315         iput(orphan_dir_inode);
2316 leave:
2317
2318         ocfs2_inode_unlock(dir, 1);
2319
2320         brelse(di_bh);
2321         brelse(parent_di_bh);
2322         brelse(orphan_dir_bh);
2323
2324         ocfs2_free_dir_lookup_result(&lookup);
2325
2326         mlog_exit(status);
2327
2328         return status;
2329 }
2330
2331 const struct inode_operations ocfs2_dir_iops = {
2332         .create         = ocfs2_create,
2333         .lookup         = ocfs2_lookup,
2334         .link           = ocfs2_link,
2335         .unlink         = ocfs2_unlink,
2336         .rmdir          = ocfs2_unlink,
2337         .symlink        = ocfs2_symlink,
2338         .mkdir          = ocfs2_mkdir,
2339         .mknod          = ocfs2_mknod,
2340         .rename         = ocfs2_rename,
2341         .setattr        = ocfs2_setattr,
2342         .getattr        = ocfs2_getattr,
2343         .permission     = ocfs2_permission,
2344         .setxattr       = generic_setxattr,
2345         .getxattr       = generic_getxattr,
2346         .listxattr      = ocfs2_listxattr,
2347         .removexattr    = generic_removexattr,
2348         .fiemap         = ocfs2_fiemap,
2349 };