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         d_set_d_op(dentry, &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         } else
 175                 ocfs2_dentry_attach_gen(dentry);
 176
 177 bail_unlock:
 178         /* Don't drop the cluster lock until *after* the d_add --
 179          * unlink on another node will message us to remove that
 180          * dentry under this lock so otherwise we can race this with
 181          * the downconvert thread and have a stale dentry. */
 182         ocfs2_inode_unlock(dir, 0);
 183
 184 bail:
 185
 186         mlog_exit_ptr(ret);
 187
 188         return ret;
 189 }
 190
 191 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
 192 {
 193         struct inode *inode;
 194
 195         inode = new_inode(dir->i_sb);
 196         if (!inode) {
 197                 mlog(ML_ERROR, "new_inode failed!\n");
 198                 return NULL;
 199         }
 200
 201         /* populate as many fields early on as possible - many of
 202          * these are used by the support functions here and in
 203          * callers. */
 204         if (S_ISDIR(mode))
 205                 inode->i_nlink = 2;
 206         else
 207                 inode->i_nlink = 1;
 208         inode_init_owner(inode, dir, mode);
 209         dquot_initialize(inode);
 210         return inode;
 211 }
 212
 213 static int ocfs2_mknod(struct inode *dir,
 214                        struct dentry *dentry,
 215                        int mode,
 216                        dev_t dev)
 217 {
 218         int status = 0;
 219         struct buffer_head *parent_fe_bh = NULL;
 220         handle_t *handle = NULL;
 221         struct ocfs2_super *osb;
 222         struct ocfs2_dinode *dirfe;
 223         struct buffer_head *new_fe_bh = NULL;
 224         struct inode *inode = NULL;
 225         struct ocfs2_alloc_context *inode_ac = NULL;
 226         struct ocfs2_alloc_context *data_ac = NULL;
 227         struct ocfs2_alloc_context *meta_ac = NULL;
 228         int want_clusters = 0;
 229         int want_meta = 0;
 230         int xattr_credits = 0;
 231         struct ocfs2_security_xattr_info si = {
 232                 .enable = 1,
 233         };
 234         int did_quota_inode = 0;
 235         struct ocfs2_dir_lookup_result lookup = { NULL, };
 236         sigset_t oldset;
 237         int did_block_signals = 0;
 238
 239         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 240                    (unsigned long)dev, dentry->d_name.len,
 241                    dentry->d_name.name);
 242
 243         dquot_initialize(dir);
 244
 245         /* get our super block */
 246         osb = OCFS2_SB(dir->i_sb);
 247
 248         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 249         if (status < 0) {
 250                 if (status != -ENOENT)
 251                         mlog_errno(status);
 252                 return status;
 253         }
 254
 255         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 256                 status = -EMLINK;
 257                 goto leave;
 258         }
 259
 260         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 261         if (!ocfs2_read_links_count(dirfe)) {
 262                 /* can't make a file in a deleted directory. */
 263                 status = -ENOENT;
 264                 goto leave;
 265         }
 266
 267         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 268                                            dentry->d_name.len);
 269         if (status)
 270                 goto leave;
 271
 272         /* get a spot inside the dir. */
 273         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 274                                               dentry->d_name.name,
 275                                               dentry->d_name.len, &lookup);
 276         if (status < 0) {
 277                 mlog_errno(status);
 278                 goto leave;
 279         }
 280
 281         /* reserve an inode spot */
 282         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 283         if (status < 0) {
 284                 if (status != -ENOSPC)
 285                         mlog_errno(status);
 286                 goto leave;
 287         }
 288
 289         inode = ocfs2_get_init_inode(dir, mode);
 290         if (!inode) {
 291                 status = -ENOMEM;
 292                 mlog_errno(status);
 293                 goto leave;
 294         }
 295
 296         /* get security xattr */
 297         status = ocfs2_init_security_get(inode, dir, &si);
 298         if (status) {
 299                 if (status == -EOPNOTSUPP)
 300                         si.enable = 0;
 301                 else {
 302                         mlog_errno(status);
 303                         goto leave;
 304                 }
 305         }
 306
 307         /* calculate meta data/clusters for setting security and acl xattr */
 308         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 309                                        &si, &want_clusters,
 310                                        &xattr_credits, &want_meta);
 311         if (status < 0) {
 312                 mlog_errno(status);
 313                 goto leave;
 314         }
 315
 316         /* Reserve a cluster if creating an extent based directory. */
 317         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 318                 want_clusters += 1;
 319
 320                 /* Dir indexing requires extra space as well */
 321                 if (ocfs2_supports_indexed_dirs(osb))
 322                         want_meta++;
 323         }
 324
 325         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 326         if (status < 0) {
 327                 if (status != -ENOSPC)
 328                         mlog_errno(status);
 329                 goto leave;
 330         }
 331
 332         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 333         if (status < 0) {
 334                 if (status != -ENOSPC)
 335                         mlog_errno(status);
 336                 goto leave;
 337         }
 338
 339         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 340                                                             S_ISDIR(mode),
 341                                                             xattr_credits));
 342         if (IS_ERR(handle)) {
 343                 status = PTR_ERR(handle);
 344                 handle = NULL;
 345                 mlog_errno(status);
 346                 goto leave;
 347         }
 348
 349         /* Starting to change things, restart is no longer possible. */
 350         ocfs2_block_signals(&oldset);
 351         did_block_signals = 1;
 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         /*
 408          * Do this before adding the entry to the directory. We add
 409          * also set d_op after success so that ->d_iput() will cleanup
 410          * the dentry lock even if ocfs2_add_entry() fails below.
 411          */
 412         status = ocfs2_dentry_attach_lock(dentry, inode,
 413                                           OCFS2_I(dir)->ip_blkno);
 414         if (status) {
 415                 mlog_errno(status);
 416                 goto leave;
 417         }
 418         d_set_d_op(dentry, &ocfs2_dentry_ops);
 419
 420         status = ocfs2_add_entry(handle, dentry, inode,
 421                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 422                                  &lookup);
 423         if (status < 0) {
 424                 mlog_errno(status);
 425                 goto leave;
 426         }
 427
 428         insert_inode_hash(inode);
 429         d_instantiate(dentry, inode);
 430         status = 0;
 431 leave:
 432         if (status < 0 && did_quota_inode)
 433                 dquot_free_inode(inode);
 434         if (handle)
 435                 ocfs2_commit_trans(osb, handle);
 436
 437         ocfs2_inode_unlock(dir, 1);
 438         if (did_block_signals)
 439                 ocfs2_unblock_signals(&oldset);
 440
 441         if (status == -ENOSPC)
 442                 mlog(0, "Disk is full\n");
 443
 444         brelse(new_fe_bh);
 445         brelse(parent_fe_bh);
 446         kfree(si.name);
 447         kfree(si.value);
 448
 449         ocfs2_free_dir_lookup_result(&lookup);
 450
 451         if (inode_ac)
 452                 ocfs2_free_alloc_context(inode_ac);
 453
 454         if (data_ac)
 455                 ocfs2_free_alloc_context(data_ac);
 456
 457         if (meta_ac)
 458                 ocfs2_free_alloc_context(meta_ac);
 459
 460         /*
 461          * We should call iput after the i_mutex of the bitmap been
 462          * unlocked in ocfs2_free_alloc_context, or the
 463          * ocfs2_delete_inode will mutex_lock again.
 464          */
 465         if ((status < 0) && inode) {
 466                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 467                 clear_nlink(inode);
 468                 iput(inode);
 469         }
 470
 471         mlog_exit(status);
 472
 473         return status;
 474 }
 475
 476 static int __ocfs2_mknod_locked(struct inode *dir,
 477                                 struct inode *inode,
 478                                 dev_t dev,
 479                                 struct buffer_head **new_fe_bh,
 480                                 struct buffer_head *parent_fe_bh,
 481                                 handle_t *handle,
 482                                 struct ocfs2_alloc_context *inode_ac,
 483                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 484 {
 485         int status = 0;
 486         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 487         struct ocfs2_dinode *fe = NULL;
 488         struct ocfs2_extent_list *fel;
 489         u16 feat;
 490
 491         *new_fe_bh = NULL;
 492
 493         /* populate as many fields early on as possible - many of
 494          * these are used by the support functions here and in
 495          * callers. */
 496         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 497         OCFS2_I(inode)->ip_blkno = fe_blkno;
 498         spin_lock(&osb->osb_lock);
 499         inode->i_generation = osb->s_next_generation++;
 500         spin_unlock(&osb->osb_lock);
 501
 502         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 503         if (!*new_fe_bh) {
 504                 status = -EIO;
 505                 mlog_errno(status);
 506                 goto leave;
 507         }
 508         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 509
 510         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 511                                          *new_fe_bh,
 512                                          OCFS2_JOURNAL_ACCESS_CREATE);
 513         if (status < 0) {
 514                 mlog_errno(status);
 515                 goto leave;
 516         }
 517
 518         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 519         memset(fe, 0, osb->sb->s_blocksize);
 520
 521         fe->i_generation = cpu_to_le32(inode->i_generation);
 522         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 523         fe->i_blkno = cpu_to_le64(fe_blkno);
 524         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 525         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 526         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 527         fe->i_uid = cpu_to_le32(inode->i_uid);
 528         fe->i_gid = cpu_to_le32(inode->i_gid);
 529         fe->i_mode = cpu_to_le16(inode->i_mode);
 530         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 531                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 532
 533         ocfs2_set_links_count(fe, inode->i_nlink);
 534
 535         fe->i_last_eb_blk = 0;
 536         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 537         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 538         fe->i_atime = fe->i_ctime = fe->i_mtime =
 539                 cpu_to_le64(CURRENT_TIME.tv_sec);
 540         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 541                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 542         fe->i_dtime = 0;
 543
 544         /*
 545          * If supported, directories start with inline data. If inline
 546          * isn't supported, but indexing is, we start them as indexed.
 547          */
 548         feat = le16_to_cpu(fe->i_dyn_features);
 549         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 550                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 551
 552                 fe->id2.i_data.id_count = cpu_to_le16(
 553                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 554         } else {
 555                 fel = &fe->id2.i_list;
 556                 fel->l_tree_depth = 0;
 557                 fel->l_next_free_rec = 0;
 558                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 559         }
 560
 561         ocfs2_journal_dirty(handle, *new_fe_bh);
 562
 563         ocfs2_populate_inode(inode, fe, 1);
 564         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 565         if (!ocfs2_mount_local(osb)) {
 566                 status = ocfs2_create_new_inode_locks(inode);
 567                 if (status < 0)
 568                         mlog_errno(status);
 569         }
 570
 571         status = 0; /* error in ocfs2_create_new_inode_locks is not
 572                      * critical */
 573
 574 leave:
 575         if (status < 0) {
 576                 if (*new_fe_bh) {
 577                         brelse(*new_fe_bh);
 578                         *new_fe_bh = NULL;
 579                 }
 580         }
 581
 582         mlog_exit(status);
 583         return status;
 584 }
 585
 586 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 587                               struct inode *dir,
 588                               struct inode *inode,
 589                               dev_t dev,
 590                               struct buffer_head **new_fe_bh,
 591                               struct buffer_head *parent_fe_bh,
 592                               handle_t *handle,
 593                               struct ocfs2_alloc_context *inode_ac)
 594 {
 595         int status = 0;
 596         u64 suballoc_loc, fe_blkno = 0;
 597         u16 suballoc_bit;
 598
 599         *new_fe_bh = NULL;
 600
 601         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 602                                        inode_ac, &suballoc_loc,
 603                                        &suballoc_bit, &fe_blkno);
 604         if (status < 0) {
 605                 mlog_errno(status);
 606                 return status;
 607         }
 608
 609         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 610                                     parent_fe_bh, handle, inode_ac,
 611                                     fe_blkno, suballoc_loc, suballoc_bit);
 612 }
 613
 614 static int ocfs2_mkdir(struct inode *dir,
 615                        struct dentry *dentry,
 616                        int mode)
 617 {
 618         int ret;
 619
 620         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 621                    dentry->d_name.len, dentry->d_name.name);
 622         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 623         mlog_exit(ret);
 624
 625         return ret;
 626 }
 627
 628 static int ocfs2_create(struct inode *dir,
 629                         struct dentry *dentry,
 630                         int mode,
 631                         struct nameidata *nd)
 632 {
 633         int ret;
 634
 635         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 636                    dentry->d_name.len, dentry->d_name.name);
 637         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 638         mlog_exit(ret);
 639
 640         return ret;
 641 }
 642
 643 static int ocfs2_link(struct dentry *old_dentry,
 644                       struct inode *dir,
 645                       struct dentry *dentry)
 646 {
 647         handle_t *handle;
 648         struct inode *inode = old_dentry->d_inode;
 649         int err;
 650         struct buffer_head *fe_bh = NULL;
 651         struct buffer_head *parent_fe_bh = NULL;
 652         struct ocfs2_dinode *fe = NULL;
 653         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 654         struct ocfs2_dir_lookup_result lookup = { NULL, };
 655         sigset_t oldset;
 656
 657         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 658                    old_dentry->d_name.len, old_dentry->d_name.name,
 659                    dentry->d_name.len, dentry->d_name.name);
 660
 661         if (S_ISDIR(inode->i_mode))
 662                 return -EPERM;
 663
 664         dquot_initialize(dir);
 665
 666         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 667         if (err < 0) {
 668                 if (err != -ENOENT)
 669                         mlog_errno(err);
 670                 return err;
 671         }
 672
 673         if (!dir->i_nlink) {
 674                 err = -ENOENT;
 675                 goto out;
 676         }
 677
 678         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 679                                         dentry->d_name.len);
 680         if (err)
 681                 goto out;
 682
 683         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 684                                            dentry->d_name.name,
 685                                            dentry->d_name.len, &lookup);
 686         if (err < 0) {
 687                 mlog_errno(err);
 688                 goto out;
 689         }
 690
 691         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 692         if (err < 0) {
 693                 if (err != -ENOENT)
 694                         mlog_errno(err);
 695                 goto out;
 696         }
 697
 698         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 699         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 700                 err = -EMLINK;
 701                 goto out_unlock_inode;
 702         }
 703
 704         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 705         if (IS_ERR(handle)) {
 706                 err = PTR_ERR(handle);
 707                 handle = NULL;
 708                 mlog_errno(err);
 709                 goto out_unlock_inode;
 710         }
 711
 712         /* Starting to change things, restart is no longer possible. */
 713         ocfs2_block_signals(&oldset);
 714
 715         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 716                                       OCFS2_JOURNAL_ACCESS_WRITE);
 717         if (err < 0) {
 718                 mlog_errno(err);
 719                 goto out_commit;
 720         }
 721
 722         inc_nlink(inode);
 723         inode->i_ctime = CURRENT_TIME;
 724         ocfs2_set_links_count(fe, inode->i_nlink);
 725         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 726         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 727         ocfs2_journal_dirty(handle, fe_bh);
 728
 729         err = ocfs2_add_entry(handle, dentry, inode,
 730                               OCFS2_I(inode)->ip_blkno,
 731                               parent_fe_bh, &lookup);
 732         if (err) {
 733                 ocfs2_add_links_count(fe, -1);
 734                 drop_nlink(inode);
 735                 mlog_errno(err);
 736                 goto out_commit;
 737         }
 738
 739         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 740         if (err) {
 741                 mlog_errno(err);
 742                 goto out_commit;
 743         }
 744
 745         ihold(inode);
 746         d_set_d_op(dentry, &ocfs2_dentry_ops);
 747         d_instantiate(dentry, inode);
 748
 749 out_commit:
 750         ocfs2_commit_trans(osb, handle);
 751         ocfs2_unblock_signals(&oldset);
 752 out_unlock_inode:
 753         ocfs2_inode_unlock(inode, 1);
 754
 755 out:
 756         ocfs2_inode_unlock(dir, 1);
 757
 758         brelse(fe_bh);
 759         brelse(parent_fe_bh);
 760
 761         ocfs2_free_dir_lookup_result(&lookup);
 762
 763         mlog_exit(err);
 764
 765         return err;
 766 }
 767
 768 /*
 769  * Takes and drops an exclusive lock on the given dentry. This will
 770  * force other nodes to drop it.
 771  */
 772 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 773 {
 774         int ret;
 775
 776         ret = ocfs2_dentry_lock(dentry, 1);
 777         if (ret)
 778                 mlog_errno(ret);
 779         else
 780                 ocfs2_dentry_unlock(dentry, 1);
 781
 782         return ret;
 783 }
 784
 785 static inline int inode_is_unlinkable(struct inode *inode)
 786 {
 787         if (S_ISDIR(inode->i_mode)) {
 788                 if (inode->i_nlink == 2)
 789                         return 1;
 790                 return 0;
 791         }
 792
 793         if (inode->i_nlink == 1)
 794                 return 1;
 795         return 0;
 796 }
 797
 798 static int ocfs2_unlink(struct inode *dir,
 799                         struct dentry *dentry)
 800 {
 801         int status;
 802         int child_locked = 0;
 803         struct inode *inode = dentry->d_inode;
 804         struct inode *orphan_dir = NULL;
 805         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 806         u64 blkno;
 807         struct ocfs2_dinode *fe = NULL;
 808         struct buffer_head *fe_bh = NULL;
 809         struct buffer_head *parent_node_bh = NULL;
 810         handle_t *handle = NULL;
 811         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 812         struct ocfs2_dir_lookup_result lookup = { NULL, };
 813         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 814
 815         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 816                    dentry->d_name.len, dentry->d_name.name);
 817
 818         dquot_initialize(dir);
 819
 820         BUG_ON(dentry->d_parent->d_inode != dir);
 821
 822         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 823
 824         if (inode == osb->root_inode) {
 825                 mlog(0, "Cannot delete the root directory\n");
 826                 return -EPERM;
 827         }
 828
 829         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 830                                          OI_LS_PARENT);
 831         if (status < 0) {
 832                 if (status != -ENOENT)
 833                         mlog_errno(status);
 834                 return status;
 835         }
 836
 837         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 838                                           dentry->d_name.len, &blkno, dir,
 839                                           &lookup);
 840         if (status < 0) {
 841                 if (status != -ENOENT)
 842                         mlog_errno(status);
 843                 goto leave;
 844         }
 845
 846         if (OCFS2_I(inode)->ip_blkno != blkno) {
 847                 status = -ENOENT;
 848
 849                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 850                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 851                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 852                 goto leave;
 853         }
 854
 855         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 856         if (status < 0) {
 857                 if (status != -ENOENT)
 858                         mlog_errno(status);
 859                 goto leave;
 860         }
 861         child_locked = 1;
 862
 863         if (S_ISDIR(inode->i_mode)) {
 864                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 865                         status = -ENOTEMPTY;
 866                         goto leave;
 867                 }
 868         }
 869
 870         status = ocfs2_remote_dentry_delete(dentry);
 871         if (status < 0) {
 872                 /* This remote delete should succeed under all normal
 873                  * circumstances. */
 874                 mlog_errno(status);
 875                 goto leave;
 876         }
 877
 878         if (inode_is_unlinkable(inode)) {
 879                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 880                                                   OCFS2_I(inode)->ip_blkno,
 881                                                   orphan_name, &orphan_insert);
 882                 if (status < 0) {
 883                         mlog_errno(status);
 884                         goto leave;
 885                 }
 886         }
 887
 888         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 889         if (IS_ERR(handle)) {
 890                 status = PTR_ERR(handle);
 891                 handle = NULL;
 892                 mlog_errno(status);
 893                 goto leave;
 894         }
 895
 896         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 897                                          OCFS2_JOURNAL_ACCESS_WRITE);
 898         if (status < 0) {
 899                 mlog_errno(status);
 900                 goto leave;
 901         }
 902
 903         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 904
 905         if (inode_is_unlinkable(inode)) {
 906                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 907                                           &orphan_insert, orphan_dir);
 908                 if (status < 0) {
 909                         mlog_errno(status);
 910                         goto leave;
 911                 }
 912         }
 913
 914         /* delete the name from the parent dir */
 915         status = ocfs2_delete_entry(handle, dir, &lookup);
 916         if (status < 0) {
 917                 mlog_errno(status);
 918                 goto leave;
 919         }
 920
 921         if (S_ISDIR(inode->i_mode))
 922                 drop_nlink(inode);
 923         drop_nlink(inode);
 924         ocfs2_set_links_count(fe, inode->i_nlink);
 925         ocfs2_journal_dirty(handle, fe_bh);
 926
 927         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 928         if (S_ISDIR(inode->i_mode))
 929                 drop_nlink(dir);
 930
 931         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 932         if (status < 0) {
 933                 mlog_errno(status);
 934                 if (S_ISDIR(inode->i_mode))
 935                         inc_nlink(dir);
 936         }
 937
 938 leave:
 939         if (handle)
 940                 ocfs2_commit_trans(osb, handle);
 941
 942         if (child_locked)
 943                 ocfs2_inode_unlock(inode, 1);
 944
 945         ocfs2_inode_unlock(dir, 1);
 946
 947         if (orphan_dir) {
 948                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 949                 ocfs2_inode_unlock(orphan_dir, 1);
 950                 mutex_unlock(&orphan_dir->i_mutex);
 951                 iput(orphan_dir);
 952         }
 953
 954         brelse(fe_bh);
 955         brelse(parent_node_bh);
 956
 957         ocfs2_free_dir_lookup_result(&orphan_insert);
 958         ocfs2_free_dir_lookup_result(&lookup);
 959
 960         mlog_exit(status);
 961
 962         return status;
 963 }
 964
 965 /*
 966  * The only place this should be used is rename!
 967  * if they have the same id, then the 1st one is the only one locked.
 968  */
 969 static int ocfs2_double_lock(struct ocfs2_super *osb,
 970                              struct buffer_head **bh1,
 971                              struct inode *inode1,
 972                              struct buffer_head **bh2,
 973                              struct inode *inode2)
 974 {
 975         int status;
 976         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 977         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 978         struct buffer_head **tmpbh;
 979         struct inode *tmpinode;
 980
 981         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 982                    (unsigned long long)oi1->ip_blkno,
 983                    (unsigned long long)oi2->ip_blkno);
 984
 985         if (*bh1)
 986                 *bh1 = NULL;
 987         if (*bh2)
 988                 *bh2 = NULL;
 989
 990         /* we always want to lock the one with the lower lockid first. */
 991         if (oi1->ip_blkno != oi2->ip_blkno) {
 992                 if (oi1->ip_blkno < oi2->ip_blkno) {
 993                         /* switch id1 and id2 around */
 994                         mlog(0, "switching them around...\n");
 995                         tmpbh = bh2;
 996                         bh2 = bh1;
 997                         bh1 = tmpbh;
 998
 999                         tmpinode = inode2;
1000                         inode2 = inode1;
1001                         inode1 = tmpinode;
1002                 }
1003                 /* lock id2 */
1004                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1005                                                  OI_LS_RENAME1);
1006                 if (status < 0) {
1007                         if (status != -ENOENT)
1008                                 mlog_errno(status);
1009                         goto bail;
1010                 }
1011         }
1012
1013         /* lock id1 */
1014         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1015         if (status < 0) {
1016                 /*
1017                  * An error return must mean that no cluster locks
1018                  * were held on function exit.
1019                  */
1020                 if (oi1->ip_blkno != oi2->ip_blkno) {
1021                         ocfs2_inode_unlock(inode2, 1);
1022                         brelse(*bh2);
1023                         *bh2 = NULL;
1024                 }
1025
1026                 if (status != -ENOENT)
1027                         mlog_errno(status);
1028         }
1029
1030 bail:
1031         mlog_exit(status);
1032         return status;
1033 }
1034
1035 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1036 {
1037         ocfs2_inode_unlock(inode1, 1);
1038
1039         if (inode1 != inode2)
1040                 ocfs2_inode_unlock(inode2, 1);
1041 }
1042
1043 static int ocfs2_rename(struct inode *old_dir,
1044                         struct dentry *old_dentry,
1045                         struct inode *new_dir,
1046                         struct dentry *new_dentry)
1047 {
1048         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1049         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1050         struct inode *old_inode = old_dentry->d_inode;
1051         struct inode *new_inode = new_dentry->d_inode;
1052         struct inode *orphan_dir = NULL;
1053         struct ocfs2_dinode *newfe = NULL;
1054         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1055         struct buffer_head *newfe_bh = NULL;
1056         struct buffer_head *old_inode_bh = NULL;
1057         struct ocfs2_super *osb = NULL;
1058         u64 newfe_blkno, old_de_ino;
1059         handle_t *handle = NULL;
1060         struct buffer_head *old_dir_bh = NULL;
1061         struct buffer_head *new_dir_bh = NULL;
1062         nlink_t old_dir_nlink = old_dir->i_nlink;
1063         struct ocfs2_dinode *old_di;
1064         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1065         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1066         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1067         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1068         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1069
1070         /* At some point it might be nice to break this function up a
1071          * bit. */
1072
1073         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1074                    old_dir, old_dentry, new_dir, new_dentry,
1075                    old_dentry->d_name.len, old_dentry->d_name.name,
1076                    new_dentry->d_name.len, new_dentry->d_name.name);
1077
1078         dquot_initialize(old_dir);
1079         dquot_initialize(new_dir);
1080
1081         osb = OCFS2_SB(old_dir->i_sb);
1082
1083         if (new_inode) {
1084                 if (!igrab(new_inode))
1085                         BUG();
1086         }
1087
1088         /* Assume a directory hierarchy thusly:
1089          * a/b/c
1090          * a/d
1091          * a,b,c, and d are all directories.
1092          *
1093          * from cwd of 'a' on both nodes:
1094          * node1: mv b/c d
1095          * node2: mv d   b/c
1096          *
1097          * And that's why, just like the VFS, we need a file system
1098          * rename lock. */
1099         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1100                 status = ocfs2_rename_lock(osb);
1101                 if (status < 0) {
1102                         mlog_errno(status);
1103                         goto bail;
1104                 }
1105                 rename_lock = 1;
1106         }
1107
1108         /* if old and new are the same, this'll just do one lock. */
1109         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1110                                    &new_dir_bh, new_dir);
1111         if (status < 0) {
1112                 mlog_errno(status);
1113                 goto bail;
1114         }
1115         parents_locked = 1;
1116
1117         /* make sure both dirs have bhs
1118          * get an extra ref on old_dir_bh if old==new */
1119         if (!new_dir_bh) {
1120                 if (old_dir_bh) {
1121                         new_dir_bh = old_dir_bh;
1122                         get_bh(new_dir_bh);
1123                 } else {
1124                         mlog(ML_ERROR, "no old_dir_bh!\n");
1125                         status = -EIO;
1126                         goto bail;
1127                 }
1128         }
1129
1130         /*
1131          * Aside from allowing a meta data update, the locking here
1132          * also ensures that the downconvert thread on other nodes
1133          * won't have to concurrently downconvert the inode and the
1134          * dentry locks.
1135          */
1136         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1137                                          OI_LS_PARENT);
1138         if (status < 0) {
1139                 if (status != -ENOENT)
1140                         mlog_errno(status);
1141                 goto bail;
1142         }
1143         old_child_locked = 1;
1144
1145         status = ocfs2_remote_dentry_delete(old_dentry);
1146         if (status < 0) {
1147                 mlog_errno(status);
1148                 goto bail;
1149         }
1150
1151         if (S_ISDIR(old_inode->i_mode)) {
1152                 u64 old_inode_parent;
1153
1154                 update_dot_dot = 1;
1155                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1156                                                   old_inode,
1157                                                   &old_inode_dot_dot_res);
1158                 if (status) {
1159                         status = -EIO;
1160                         goto bail;
1161                 }
1162
1163                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1164                         status = -EIO;
1165                         goto bail;
1166                 }
1167
1168                 if (!new_inode && new_dir != old_dir &&
1169                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1170                         status = -EMLINK;
1171                         goto bail;
1172                 }
1173         }
1174
1175         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1176                                             old_dentry->d_name.len,
1177                                             &old_de_ino);
1178         if (status) {
1179                 status = -ENOENT;
1180                 goto bail;
1181         }
1182
1183         /*
1184          *  Check for inode number is _not_ due to possible IO errors.
1185          *  We might rmdir the source, keep it as pwd of some process
1186          *  and merrily kill the link to whatever was created under the
1187          *  same name. Goodbye sticky bit ;-<
1188          */
1189         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1190                 status = -ENOENT;
1191                 goto bail;
1192         }
1193
1194         /* check if the target already exists (in which case we need
1195          * to delete it */
1196         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1197                                           new_dentry->d_name.len,
1198                                           &newfe_blkno, new_dir,
1199                                           &target_lookup_res);
1200         /* The only error we allow here is -ENOENT because the new
1201          * file not existing is perfectly valid. */
1202         if ((status < 0) && (status != -ENOENT)) {
1203                 /* If we cannot find the file specified we should just */
1204                 /* return the error... */
1205                 mlog_errno(status);
1206                 goto bail;
1207         }
1208         if (status == 0)
1209                 target_exists = 1;
1210
1211         if (!target_exists && new_inode) {
1212                 /*
1213                  * Target was unlinked by another node while we were
1214                  * waiting to get to ocfs2_rename(). There isn't
1215                  * anything we can do here to help the situation, so
1216                  * bubble up the appropriate error.
1217                  */
1218                 status = -ENOENT;
1219                 goto bail;
1220         }
1221
1222         /* In case we need to overwrite an existing file, we blow it
1223          * away first */
1224         if (target_exists) {
1225                 /* VFS didn't think there existed an inode here, but
1226                  * someone else in the cluster must have raced our
1227                  * rename to create one. Today we error cleanly, in
1228                  * the future we should consider calling iget to build
1229                  * a new struct inode for this entry. */
1230                 if (!new_inode) {
1231                         status = -EACCES;
1232
1233                         mlog(0, "We found an inode for name %.*s but VFS "
1234                              "didn't give us one.\n", new_dentry->d_name.len,
1235                              new_dentry->d_name.name);
1236                         goto bail;
1237                 }
1238
1239                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1240                         status = -EACCES;
1241
1242                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1243                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1244                              (unsigned long long)newfe_blkno,
1245                              OCFS2_I(new_inode)->ip_flags);
1246                         goto bail;
1247                 }
1248
1249                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1250                 if (status < 0) {
1251                         if (status != -ENOENT)
1252                                 mlog_errno(status);
1253                         goto bail;
1254                 }
1255                 new_child_locked = 1;
1256
1257                 status = ocfs2_remote_dentry_delete(new_dentry);
1258                 if (status < 0) {
1259                         mlog_errno(status);
1260                         goto bail;
1261                 }
1262
1263                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1264
1265                 mlog(0, "aha rename over existing... new_blkno=%llu "
1266                      "newfebh=%p bhblocknr=%llu\n",
1267                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1268                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1269
1270                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1271                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1272                                                 OCFS2_I(new_inode)->ip_blkno,
1273                                                 orphan_name, &orphan_insert);
1274                         if (status < 0) {
1275                                 mlog_errno(status);
1276                                 goto bail;
1277                         }
1278                 }
1279         } else {
1280                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1281
1282                 status = ocfs2_check_dir_for_entry(new_dir,
1283                                                    new_dentry->d_name.name,
1284                                                    new_dentry->d_name.len);
1285                 if (status)
1286                         goto bail;
1287
1288                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1289                                                       new_dentry->d_name.name,
1290                                                       new_dentry->d_name.len,
1291                                                       &target_insert);
1292                 if (status < 0) {
1293                         mlog_errno(status);
1294                         goto bail;
1295                 }
1296         }
1297
1298         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1299         if (IS_ERR(handle)) {
1300                 status = PTR_ERR(handle);
1301                 handle = NULL;
1302                 mlog_errno(status);
1303                 goto bail;
1304         }
1305
1306         if (target_exists) {
1307                 if (S_ISDIR(new_inode->i_mode)) {
1308                         if (new_inode->i_nlink != 2 ||
1309                             !ocfs2_empty_dir(new_inode)) {
1310                                 status = -ENOTEMPTY;
1311                                 goto bail;
1312                         }
1313                 }
1314                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1315                                                  newfe_bh,
1316                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1317                 if (status < 0) {
1318                         mlog_errno(status);
1319                         goto bail;
1320                 }
1321
1322                 if (S_ISDIR(new_inode->i_mode) ||
1323                     (ocfs2_read_links_count(newfe) == 1)) {
1324                         status = ocfs2_orphan_add(osb, handle, new_inode,
1325                                                   newfe_bh, orphan_name,
1326                                                   &orphan_insert, orphan_dir);
1327                         if (status < 0) {
1328                                 mlog_errno(status);
1329                                 goto bail;
1330                         }
1331                 }
1332
1333                 /* change the dirent to point to the correct inode */
1334                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1335                                             old_inode);
1336                 if (status < 0) {
1337                         mlog_errno(status);
1338                         goto bail;
1339                 }
1340                 new_dir->i_version++;
1341
1342                 if (S_ISDIR(new_inode->i_mode))
1343                         ocfs2_set_links_count(newfe, 0);
1344                 else
1345                         ocfs2_add_links_count(newfe, -1);
1346                 ocfs2_journal_dirty(handle, newfe_bh);
1347         } else {
1348                 /* if the name was not found in new_dir, add it now */
1349                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1350                                          OCFS2_I(old_inode)->ip_blkno,
1351                                          new_dir_bh, &target_insert);
1352         }
1353
1354         old_inode->i_ctime = CURRENT_TIME;
1355         mark_inode_dirty(old_inode);
1356
1357         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1358                                          old_inode_bh,
1359                                          OCFS2_JOURNAL_ACCESS_WRITE);
1360         if (status >= 0) {
1361                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1362
1363                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1364                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1365                 ocfs2_journal_dirty(handle, old_inode_bh);
1366         } else
1367                 mlog_errno(status);
1368
1369         /*
1370          * Now that the name has been added to new_dir, remove the old name.
1371          *
1372          * We don't keep any directory entry context around until now
1373          * because the insert might have changed the type of directory
1374          * we're dealing with.
1375          */
1376         status = ocfs2_find_entry(old_dentry->d_name.name,
1377                                   old_dentry->d_name.len, old_dir,
1378                                   &old_entry_lookup);
1379         if (status)
1380                 goto bail;
1381
1382         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1383         if (status < 0) {
1384                 mlog_errno(status);
1385                 goto bail;
1386         }
1387
1388         if (new_inode) {
1389                 new_inode->i_nlink--;
1390                 new_inode->i_ctime = CURRENT_TIME;
1391         }
1392         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1393
1394         if (update_dot_dot) {
1395                 status = ocfs2_update_entry(old_inode, handle,
1396                                             &old_inode_dot_dot_res, new_dir);
1397                 old_dir->i_nlink--;
1398                 if (new_inode) {
1399                         new_inode->i_nlink--;
1400                 } else {
1401                         inc_nlink(new_dir);
1402                         mark_inode_dirty(new_dir);
1403                 }
1404         }
1405         mark_inode_dirty(old_dir);
1406         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1407         if (new_inode) {
1408                 mark_inode_dirty(new_inode);
1409                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1410         }
1411
1412         if (old_dir != new_dir) {
1413                 /* Keep the same times on both directories.*/
1414                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1415
1416                 /*
1417                  * This will also pick up the i_nlink change from the
1418                  * block above.
1419                  */
1420                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1421         }
1422
1423         if (old_dir_nlink != old_dir->i_nlink) {
1424                 if (!old_dir_bh) {
1425                         mlog(ML_ERROR, "need to change nlink for old dir "
1426                              "%llu from %d to %d but bh is NULL!\n",
1427                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1428                              (int)old_dir_nlink, old_dir->i_nlink);
1429                 } else {
1430                         struct ocfs2_dinode *fe;
1431                         status = ocfs2_journal_access_di(handle,
1432                                                          INODE_CACHE(old_dir),
1433                                                          old_dir_bh,
1434                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1435                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1436                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1437                         ocfs2_journal_dirty(handle, old_dir_bh);
1438                 }
1439         }
1440         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1441         status = 0;
1442 bail:
1443         if (rename_lock)
1444                 ocfs2_rename_unlock(osb);
1445
1446         if (handle)
1447                 ocfs2_commit_trans(osb, handle);
1448
1449         if (parents_locked)
1450                 ocfs2_double_unlock(old_dir, new_dir);
1451
1452         if (old_child_locked)
1453                 ocfs2_inode_unlock(old_inode, 1);
1454
1455         if (new_child_locked)
1456                 ocfs2_inode_unlock(new_inode, 1);
1457
1458         if (orphan_dir) {
1459                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1460                 ocfs2_inode_unlock(orphan_dir, 1);
1461                 mutex_unlock(&orphan_dir->i_mutex);
1462                 iput(orphan_dir);
1463         }
1464
1465         if (new_inode)
1466                 sync_mapping_buffers(old_inode->i_mapping);
1467
1468         if (new_inode)
1469                 iput(new_inode);
1470
1471         ocfs2_free_dir_lookup_result(&target_lookup_res);
1472         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1473         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1474         ocfs2_free_dir_lookup_result(&orphan_insert);
1475         ocfs2_free_dir_lookup_result(&target_insert);
1476
1477         brelse(newfe_bh);
1478         brelse(old_inode_bh);
1479         brelse(old_dir_bh);
1480         brelse(new_dir_bh);
1481
1482         mlog_exit(status);
1483
1484         return status;
1485 }
1486
1487 /*
1488  * we expect i_size = strlen(symname). Copy symname into the file
1489  * data, including the null terminator.
1490  */
1491 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1492                                      handle_t *handle,
1493                                      struct inode *inode,
1494                                      const char *symname)
1495 {
1496         struct buffer_head **bhs = NULL;
1497         const char *c;
1498         struct super_block *sb = osb->sb;
1499         u64 p_blkno, p_blocks;
1500         int virtual, blocks, status, i, bytes_left;
1501
1502         bytes_left = i_size_read(inode) + 1;
1503         /* we can't trust i_blocks because we're actually going to
1504          * write i_size + 1 bytes. */
1505         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1506
1507         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1508                         (unsigned long long)inode->i_blocks,
1509                         i_size_read(inode), blocks);
1510
1511         /* Sanity check -- make sure we're going to fit. */
1512         if (bytes_left >
1513             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1514                 status = -EIO;
1515                 mlog_errno(status);
1516                 goto bail;
1517         }
1518
1519         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1520         if (!bhs) {
1521                 status = -ENOMEM;
1522                 mlog_errno(status);
1523                 goto bail;
1524         }
1525
1526         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1527                                              NULL);
1528         if (status < 0) {
1529                 mlog_errno(status);
1530                 goto bail;
1531         }
1532
1533         /* links can never be larger than one cluster so we know this
1534          * is all going to be contiguous, but do a sanity check
1535          * anyway. */
1536         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1537                 status = -EIO;
1538                 mlog_errno(status);
1539                 goto bail;
1540         }
1541
1542         virtual = 0;
1543         while(bytes_left > 0) {
1544                 c = &symname[virtual * sb->s_blocksize];
1545
1546                 bhs[virtual] = sb_getblk(sb, p_blkno);
1547                 if (!bhs[virtual]) {
1548                         status = -ENOMEM;
1549                         mlog_errno(status);
1550                         goto bail;
1551                 }
1552                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1553                                               bhs[virtual]);
1554
1555                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1556                                               bhs[virtual],
1557                                               OCFS2_JOURNAL_ACCESS_CREATE);
1558                 if (status < 0) {
1559                         mlog_errno(status);
1560                         goto bail;
1561                 }
1562
1563                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1564
1565                 memcpy(bhs[virtual]->b_data, c,
1566                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1567                        bytes_left);
1568
1569                 ocfs2_journal_dirty(handle, bhs[virtual]);
1570
1571                 virtual++;
1572                 p_blkno++;
1573                 bytes_left -= sb->s_blocksize;
1574         }
1575
1576         status = 0;
1577 bail:
1578
1579         if (bhs) {
1580                 for(i = 0; i < blocks; i++)
1581                         brelse(bhs[i]);
1582                 kfree(bhs);
1583         }
1584
1585         mlog_exit(status);
1586         return status;
1587 }
1588
1589 static int ocfs2_symlink(struct inode *dir,
1590                          struct dentry *dentry,
1591                          const char *symname)
1592 {
1593         int status, l, credits;
1594         u64 newsize;
1595         struct ocfs2_super *osb = NULL;
1596         struct inode *inode = NULL;
1597         struct super_block *sb;
1598         struct buffer_head *new_fe_bh = NULL;
1599         struct buffer_head *parent_fe_bh = NULL;
1600         struct ocfs2_dinode *fe = NULL;
1601         struct ocfs2_dinode *dirfe;
1602         handle_t *handle = NULL;
1603         struct ocfs2_alloc_context *inode_ac = NULL;
1604         struct ocfs2_alloc_context *data_ac = NULL;
1605         struct ocfs2_alloc_context *xattr_ac = NULL;
1606         int want_clusters = 0;
1607         int xattr_credits = 0;
1608         struct ocfs2_security_xattr_info si = {
1609                 .enable = 1,
1610         };
1611         int did_quota = 0, did_quota_inode = 0;
1612         struct ocfs2_dir_lookup_result lookup = { NULL, };
1613         sigset_t oldset;
1614         int did_block_signals = 0;
1615
1616         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1617                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1618
1619         dquot_initialize(dir);
1620
1621         sb = dir->i_sb;
1622         osb = OCFS2_SB(sb);
1623
1624         l = strlen(symname) + 1;
1625
1626         credits = ocfs2_calc_symlink_credits(sb);
1627
1628         /* lock the parent directory */
1629         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1630         if (status < 0) {
1631                 if (status != -ENOENT)
1632                         mlog_errno(status);
1633                 return status;
1634         }
1635
1636         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1637         if (!ocfs2_read_links_count(dirfe)) {
1638                 /* can't make a file in a deleted directory. */
1639                 status = -ENOENT;
1640                 goto bail;
1641         }
1642
1643         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1644                                            dentry->d_name.len);
1645         if (status)
1646                 goto bail;
1647
1648         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1649                                               dentry->d_name.name,
1650                                               dentry->d_name.len, &lookup);
1651         if (status < 0) {
1652                 mlog_errno(status);
1653                 goto bail;
1654         }
1655
1656         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1657         if (status < 0) {
1658                 if (status != -ENOSPC)
1659                         mlog_errno(status);
1660                 goto bail;
1661         }
1662
1663         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1664         if (!inode) {
1665                 status = -ENOMEM;
1666                 mlog_errno(status);
1667                 goto bail;
1668         }
1669
1670         /* get security xattr */
1671         status = ocfs2_init_security_get(inode, dir, &si);
1672         if (status) {
1673                 if (status == -EOPNOTSUPP)
1674                         si.enable = 0;
1675                 else {
1676                         mlog_errno(status);
1677                         goto bail;
1678                 }
1679         }
1680
1681         /* calculate meta data/clusters for setting security xattr */
1682         if (si.enable) {
1683                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1684                                                   &xattr_credits, &xattr_ac);
1685                 if (status < 0) {
1686                         mlog_errno(status);
1687                         goto bail;
1688                 }
1689         }
1690
1691         /* don't reserve bitmap space for fast symlinks. */
1692         if (l > ocfs2_fast_symlink_chars(sb))
1693                 want_clusters += 1;
1694
1695         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1696         if (status < 0) {
1697                 if (status != -ENOSPC)
1698                         mlog_errno(status);
1699                 goto bail;
1700         }
1701
1702         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1703         if (IS_ERR(handle)) {
1704                 status = PTR_ERR(handle);
1705                 handle = NULL;
1706                 mlog_errno(status);
1707                 goto bail;
1708         }
1709
1710         /* Starting to change things, restart is no longer possible. */
1711         ocfs2_block_signals(&oldset);
1712         did_block_signals = 1;
1713
1714         status = dquot_alloc_inode(inode);
1715         if (status)
1716                 goto bail;
1717         did_quota_inode = 1;
1718
1719         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1720                    inode->i_mode, dentry->d_name.len,
1721                    dentry->d_name.name);
1722
1723         status = ocfs2_mknod_locked(osb, dir, inode,
1724                                     0, &new_fe_bh, parent_fe_bh, handle,
1725                                     inode_ac);
1726         if (status < 0) {
1727                 mlog_errno(status);
1728                 goto bail;
1729         }
1730
1731         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1732         inode->i_rdev = 0;
1733         newsize = l - 1;
1734         if (l > ocfs2_fast_symlink_chars(sb)) {
1735                 u32 offset = 0;
1736
1737                 inode->i_op = &ocfs2_symlink_inode_operations;
1738                 status = dquot_alloc_space_nodirty(inode,
1739                     ocfs2_clusters_to_bytes(osb->sb, 1));
1740                 if (status)
1741                         goto bail;
1742                 did_quota = 1;
1743                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1744                                               new_fe_bh,
1745                                               handle, data_ac, NULL,
1746                                               NULL);
1747                 if (status < 0) {
1748                         if (status != -ENOSPC && status != -EINTR) {
1749                                 mlog(ML_ERROR,
1750                                      "Failed to extend file to %llu\n",
1751                                      (unsigned long long)newsize);
1752                                 mlog_errno(status);
1753                                 status = -ENOSPC;
1754                         }
1755                         goto bail;
1756                 }
1757                 i_size_write(inode, newsize);
1758                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1759         } else {
1760                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1761                 memcpy((char *) fe->id2.i_symlink, symname, l);
1762                 i_size_write(inode, newsize);
1763                 inode->i_blocks = 0;
1764         }
1765
1766         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1767         if (status < 0) {
1768                 mlog_errno(status);
1769                 goto bail;
1770         }
1771
1772         if (!ocfs2_inode_is_fast_symlink(inode)) {
1773                 status = ocfs2_create_symlink_data(osb, handle, inode,
1774                                                    symname);
1775                 if (status < 0) {
1776                         mlog_errno(status);
1777                         goto bail;
1778                 }
1779         }
1780
1781         if (si.enable) {
1782                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1783                                                  xattr_ac, data_ac);
1784                 if (status < 0) {
1785                         mlog_errno(status);
1786                         goto bail;
1787                 }
1788         }
1789
1790         /*
1791          * Do this before adding the entry to the directory. We add
1792          * also set d_op after success so that ->d_iput() will cleanup
1793          * the dentry lock even if ocfs2_add_entry() fails below.
1794          */
1795         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1796         if (status) {
1797                 mlog_errno(status);
1798                 goto bail;
1799         }
1800         d_set_d_op(dentry, &ocfs2_dentry_ops);
1801
1802         status = ocfs2_add_entry(handle, dentry, inode,
1803                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1804                                  &lookup);
1805         if (status < 0) {
1806                 mlog_errno(status);
1807                 goto bail;
1808         }
1809
1810         insert_inode_hash(inode);
1811         d_instantiate(dentry, inode);
1812 bail:
1813         if (status < 0 && did_quota)
1814                 dquot_free_space_nodirty(inode,
1815                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1816         if (status < 0 && did_quota_inode)
1817                 dquot_free_inode(inode);
1818         if (handle)
1819                 ocfs2_commit_trans(osb, handle);
1820
1821         ocfs2_inode_unlock(dir, 1);
1822         if (did_block_signals)
1823                 ocfs2_unblock_signals(&oldset);
1824
1825         brelse(new_fe_bh);
1826         brelse(parent_fe_bh);
1827         kfree(si.name);
1828         kfree(si.value);
1829         ocfs2_free_dir_lookup_result(&lookup);
1830         if (inode_ac)
1831                 ocfs2_free_alloc_context(inode_ac);
1832         if (data_ac)
1833                 ocfs2_free_alloc_context(data_ac);
1834         if (xattr_ac)
1835                 ocfs2_free_alloc_context(xattr_ac);
1836         if ((status < 0) && inode) {
1837                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1838                 clear_nlink(inode);
1839                 iput(inode);
1840         }
1841
1842         mlog_exit(status);
1843
1844         return status;
1845 }
1846
1847 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1848 {
1849         int status, namelen;
1850
1851         mlog_entry_void();
1852
1853         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1854                            (long long)blkno);
1855         if (namelen <= 0) {
1856                 if (namelen)
1857                         status = namelen;
1858                 else
1859                         status = -EINVAL;
1860                 mlog_errno(status);
1861                 goto bail;
1862         }
1863         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1864                 status = -EINVAL;
1865                 mlog_errno(status);
1866                 goto bail;
1867         }
1868
1869         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1870              namelen);
1871
1872         status = 0;
1873 bail:
1874         mlog_exit(status);
1875         return status;
1876 }
1877
1878 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1879                                         struct inode **ret_orphan_dir,
1880                                         struct buffer_head **ret_orphan_dir_bh)
1881 {
1882         struct inode *orphan_dir_inode;
1883         struct buffer_head *orphan_dir_bh = NULL;
1884         int ret = 0;
1885
1886         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1887                                                        ORPHAN_DIR_SYSTEM_INODE,
1888                                                        osb->slot_num);
1889         if (!orphan_dir_inode) {
1890                 ret = -ENOENT;
1891                 mlog_errno(ret);
1892                 return ret;
1893         }
1894
1895         mutex_lock(&orphan_dir_inode->i_mutex);
1896
1897         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1898         if (ret < 0) {
1899                 mutex_unlock(&orphan_dir_inode->i_mutex);
1900                 iput(orphan_dir_inode);
1901
1902                 mlog_errno(ret);
1903                 return ret;
1904         }
1905
1906         *ret_orphan_dir = orphan_dir_inode;
1907         *ret_orphan_dir_bh = orphan_dir_bh;
1908
1909         return 0;
1910 }
1911
1912 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1913                                       struct buffer_head *orphan_dir_bh,
1914                                       u64 blkno,
1915                                       char *name,
1916                                       struct ocfs2_dir_lookup_result *lookup)
1917 {
1918         int ret;
1919         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1920
1921         ret = ocfs2_blkno_stringify(blkno, name);
1922         if (ret < 0) {
1923                 mlog_errno(ret);
1924                 return ret;
1925         }
1926
1927         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1928                                            orphan_dir_bh, name,
1929                                            OCFS2_ORPHAN_NAMELEN, lookup);
1930         if (ret < 0) {
1931                 mlog_errno(ret);
1932                 return ret;
1933         }
1934
1935         return 0;
1936 }
1937
1938 /**
1939  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1940  * insertion of an orphan.
1941  * @osb: ocfs2 file system
1942  * @ret_orphan_dir: Orphan dir inode - returned locked!
1943  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1944  * @lookup: dir lookup result, to be passed back into functions like
1945  *          ocfs2_orphan_add
1946  *
1947  * Returns zero on success and the ret_orphan_dir, name and lookup
1948  * fields will be populated.
1949  *
1950  * Returns non-zero on failure.
1951  */
1952 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1953                                     struct inode **ret_orphan_dir,
1954                                     u64 blkno,
1955                                     char *name,
1956                                     struct ocfs2_dir_lookup_result *lookup)
1957 {
1958         struct inode *orphan_dir_inode = NULL;
1959         struct buffer_head *orphan_dir_bh = NULL;
1960         int ret = 0;
1961
1962         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1963                                            &orphan_dir_bh);
1964         if (ret < 0) {
1965                 mlog_errno(ret);
1966                 return ret;
1967         }
1968
1969         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1970                                          blkno, name, lookup);
1971         if (ret < 0) {
1972                 mlog_errno(ret);
1973                 goto out;
1974         }
1975
1976         *ret_orphan_dir = orphan_dir_inode;
1977
1978 out:
1979         brelse(orphan_dir_bh);
1980
1981         if (ret) {
1982                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1983                 mutex_unlock(&orphan_dir_inode->i_mutex);
1984                 iput(orphan_dir_inode);
1985         }
1986
1987         mlog_exit(ret);
1988         return ret;
1989 }
1990
1991 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1992                             handle_t *handle,
1993                             struct inode *inode,
1994                             struct buffer_head *fe_bh,
1995                             char *name,
1996                             struct ocfs2_dir_lookup_result *lookup,
1997                             struct inode *orphan_dir_inode)
1998 {
1999         struct buffer_head *orphan_dir_bh = NULL;
2000         int status = 0;
2001         struct ocfs2_dinode *orphan_fe;
2002         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2003
2004         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2005
2006         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2007         if (status < 0) {
2008                 mlog_errno(status);
2009                 goto leave;
2010         }
2011
2012         status = ocfs2_journal_access_di(handle,
2013                                          INODE_CACHE(orphan_dir_inode),
2014                                          orphan_dir_bh,
2015                                          OCFS2_JOURNAL_ACCESS_WRITE);
2016         if (status < 0) {
2017                 mlog_errno(status);
2018                 goto leave;
2019         }
2020
2021         /* we're a cluster, and nlink can change on disk from
2022          * underneath us... */
2023         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2024         if (S_ISDIR(inode->i_mode))
2025                 ocfs2_add_links_count(orphan_fe, 1);
2026         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2027         ocfs2_journal_dirty(handle, orphan_dir_bh);
2028
2029         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2030                                    OCFS2_ORPHAN_NAMELEN, inode,
2031                                    OCFS2_I(inode)->ip_blkno,
2032                                    orphan_dir_bh, lookup);
2033         if (status < 0) {
2034                 mlog_errno(status);
2035                 goto leave;
2036         }
2037
2038         /*
2039          * We're going to journal the change of i_flags and i_orphaned_slot.
2040          * It's safe anyway, though some callers may duplicate the journaling.
2041          * Journaling within the func just make the logic look more
2042          * straightforward.
2043          */
2044         status = ocfs2_journal_access_di(handle,
2045                                          INODE_CACHE(inode),
2046                                          fe_bh,
2047                                          OCFS2_JOURNAL_ACCESS_WRITE);
2048         if (status < 0) {
2049                 mlog_errno(status);
2050                 goto leave;
2051         }
2052
2053         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2054         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2055
2056         /* Record which orphan dir our inode now resides
2057          * in. delete_inode will use this to determine which orphan
2058          * dir to lock. */
2059         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2060
2061         ocfs2_journal_dirty(handle, fe_bh);
2062
2063         mlog(0, "Inode %llu orphaned in slot %d\n",
2064              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2065
2066 leave:
2067         brelse(orphan_dir_bh);
2068
2069         mlog_exit(status);
2070         return status;
2071 }
2072
2073 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2074 int ocfs2_orphan_del(struct ocfs2_super *osb,
2075                      handle_t *handle,
2076                      struct inode *orphan_dir_inode,
2077                      struct inode *inode,
2078                      struct buffer_head *orphan_dir_bh)
2079 {
2080         char name[OCFS2_ORPHAN_NAMELEN + 1];
2081         struct ocfs2_dinode *orphan_fe;
2082         int status = 0;
2083         struct ocfs2_dir_lookup_result lookup = { NULL, };
2084
2085         mlog_entry_void();
2086
2087         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2088         if (status < 0) {
2089                 mlog_errno(status);
2090                 goto leave;
2091         }
2092
2093         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2094              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2095              OCFS2_ORPHAN_NAMELEN);
2096
2097         /* find it's spot in the orphan directory */
2098         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2099                                   &lookup);
2100         if (status) {
2101                 mlog_errno(status);
2102                 goto leave;
2103         }
2104
2105         /* remove it from the orphan directory */
2106         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2107         if (status < 0) {
2108                 mlog_errno(status);
2109                 goto leave;
2110         }
2111
2112         status = ocfs2_journal_access_di(handle,
2113                                          INODE_CACHE(orphan_dir_inode),
2114                                          orphan_dir_bh,
2115                                          OCFS2_JOURNAL_ACCESS_WRITE);
2116         if (status < 0) {
2117                 mlog_errno(status);
2118                 goto leave;
2119         }
2120
2121         /* do the i_nlink dance! :) */
2122         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2123         if (S_ISDIR(inode->i_mode))
2124                 ocfs2_add_links_count(orphan_fe, -1);
2125         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2126         ocfs2_journal_dirty(handle, orphan_dir_bh);
2127
2128 leave:
2129         ocfs2_free_dir_lookup_result(&lookup);
2130
2131         mlog_exit(status);
2132         return status;
2133 }
2134
2135 /**
2136  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to recieve a newly
2137  * allocated file. This is different from the typical 'add to orphan dir'
2138  * operation in that the inode does not yet exist. This is a problem because
2139  * the orphan dir stringifies the inode block number to come up with it's
2140  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2141  * problem. This function works around it by calling deeper into the orphan
2142  * and suballoc code than other callers. Use this only by necessity.
2143  * @dir: The directory which this inode will ultimately wind up under - not the
2144  * orphan dir!
2145  * @dir_bh: buffer_head the @dir inode block
2146  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2147  * with the string to be used for orphan dirent. Pass back to the orphan dir
2148  * code.
2149  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2150  * dir code.
2151  * @ret_di_blkno: block number where the new inode will be allocated.
2152  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2153  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2154  *
2155  * Returns zero on success and the ret_orphan_dir, name and lookup
2156  * fields will be populated.
2157  *
2158  * Returns non-zero on failure.
2159  */
2160 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2161                                         struct buffer_head *dir_bh,
2162                                         char *orphan_name,
2163                                         struct inode **ret_orphan_dir,
2164                                         u64 *ret_di_blkno,
2165                                         struct ocfs2_dir_lookup_result *orphan_insert,
2166                                         struct ocfs2_alloc_context **ret_inode_ac)
2167 {
2168         int ret;
2169         u64 di_blkno;
2170         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2171         struct inode *orphan_dir = NULL;
2172         struct buffer_head *orphan_dir_bh = NULL;
2173         struct ocfs2_alloc_context *inode_ac = NULL;
2174
2175         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2176         if (ret < 0) {
2177                 mlog_errno(ret);
2178                 return ret;
2179         }
2180
2181         /* reserve an inode spot */
2182         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2183         if (ret < 0) {
2184                 if (ret != -ENOSPC)
2185                         mlog_errno(ret);
2186                 goto out;
2187         }
2188
2189         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2190                                        &di_blkno);
2191         if (ret) {
2192                 mlog_errno(ret);
2193                 goto out;
2194         }
2195
2196         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2197                                          di_blkno, orphan_name, orphan_insert);
2198         if (ret < 0) {
2199                 mlog_errno(ret);
2200                 goto out;
2201         }
2202
2203 out:
2204         if (ret == 0) {
2205                 *ret_orphan_dir = orphan_dir;
2206                 *ret_di_blkno = di_blkno;
2207                 *ret_inode_ac = inode_ac;
2208                 /*
2209                  * orphan_name and orphan_insert are already up to
2210                  * date via prepare_orphan_dir
2211                  */
2212         } else {
2213                 /* Unroll reserve_new_inode* */
2214                 if (inode_ac)
2215                         ocfs2_free_alloc_context(inode_ac);
2216
2217                 /* Unroll orphan dir locking */
2218                 mutex_unlock(&orphan_dir->i_mutex);
2219                 ocfs2_inode_unlock(orphan_dir, 1);
2220                 iput(orphan_dir);
2221         }
2222
2223         brelse(orphan_dir_bh);
2224
2225         return 0;
2226 }
2227
2228 int ocfs2_create_inode_in_orphan(struct inode *dir,
2229                                  int mode,
2230                                  struct inode **new_inode)
2231 {
2232         int status, did_quota_inode = 0;
2233         struct inode *inode = NULL;
2234         struct inode *orphan_dir = NULL;
2235         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2236         struct ocfs2_dinode *di = NULL;
2237         handle_t *handle = NULL;
2238         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2239         struct buffer_head *parent_di_bh = NULL;
2240         struct buffer_head *new_di_bh = NULL;
2241         struct ocfs2_alloc_context *inode_ac = NULL;
2242         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2243         u64 uninitialized_var(di_blkno), suballoc_loc;
2244         u16 suballoc_bit;
2245
2246         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2247         if (status < 0) {
2248                 if (status != -ENOENT)
2249                         mlog_errno(status);
2250                 return status;
2251         }
2252
2253         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2254                                               orphan_name, &orphan_dir,
2255                                               &di_blkno, &orphan_insert, &inode_ac);
2256         if (status < 0) {
2257                 if (status != -ENOSPC)
2258                         mlog_errno(status);
2259                 goto leave;
2260         }
2261
2262         inode = ocfs2_get_init_inode(dir, mode);
2263         if (!inode) {
2264                 status = -ENOMEM;
2265                 mlog_errno(status);
2266                 goto leave;
2267         }
2268
2269         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2270         if (IS_ERR(handle)) {
2271                 status = PTR_ERR(handle);
2272                 handle = NULL;
2273                 mlog_errno(status);
2274                 goto leave;
2275         }
2276
2277         status = dquot_alloc_inode(inode);
2278         if (status)
2279                 goto leave;
2280         did_quota_inode = 1;
2281
2282         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2283                                               &suballoc_loc,
2284                                               &suballoc_bit, di_blkno);
2285         if (status < 0) {
2286                 mlog_errno(status);
2287                 goto leave;
2288         }
2289
2290         inode->i_nlink = 0;
2291         /* do the real work now. */
2292         status = __ocfs2_mknod_locked(dir, inode,
2293                                       0, &new_di_bh, parent_di_bh, handle,
2294                                       inode_ac, di_blkno, suballoc_loc,
2295                                       suballoc_bit);
2296         if (status < 0) {
2297                 mlog_errno(status);
2298                 goto leave;
2299         }
2300
2301         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2302         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2303                                   &orphan_insert, orphan_dir);
2304         if (status < 0) {
2305                 mlog_errno(status);
2306                 goto leave;
2307         }
2308
2309         /* get open lock so that only nodes can't remove it from orphan dir. */
2310         status = ocfs2_open_lock(inode);
2311         if (status < 0)
2312                 mlog_errno(status);
2313
2314         insert_inode_hash(inode);
2315 leave:
2316         if (status < 0 && did_quota_inode)
2317                 dquot_free_inode(inode);
2318         if (handle)
2319                 ocfs2_commit_trans(osb, handle);
2320
2321         if (orphan_dir) {
2322                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2323                 ocfs2_inode_unlock(orphan_dir, 1);
2324                 mutex_unlock(&orphan_dir->i_mutex);
2325                 iput(orphan_dir);
2326         }
2327
2328         if (status == -ENOSPC)
2329                 mlog(0, "Disk is full\n");
2330
2331         if ((status < 0) && inode) {
2332                 clear_nlink(inode);
2333                 iput(inode);
2334         }
2335
2336         if (inode_ac)
2337                 ocfs2_free_alloc_context(inode_ac);
2338
2339         brelse(new_di_bh);
2340
2341         if (!status)
2342                 *new_inode = inode;
2343
2344         ocfs2_free_dir_lookup_result(&orphan_insert);
2345
2346         ocfs2_inode_unlock(dir, 1);
2347         brelse(parent_di_bh);
2348         return status;
2349 }
2350
2351 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2352                                    struct inode *inode,
2353                                    struct dentry *dentry)
2354 {
2355         int status = 0;
2356         struct buffer_head *parent_di_bh = NULL;
2357         handle_t *handle = NULL;
2358         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2359         struct ocfs2_dinode *dir_di, *di;
2360         struct inode *orphan_dir_inode = NULL;
2361         struct buffer_head *orphan_dir_bh = NULL;
2362         struct buffer_head *di_bh = NULL;
2363         struct ocfs2_dir_lookup_result lookup = { NULL, };
2364
2365         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2366                    dentry->d_name.len, dentry->d_name.name);
2367
2368         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2369         if (status < 0) {
2370                 if (status != -ENOENT)
2371                         mlog_errno(status);
2372                 return status;
2373         }
2374
2375         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2376         if (!dir_di->i_links_count) {
2377                 /* can't make a file in a deleted directory. */
2378                 status = -ENOENT;
2379                 goto leave;
2380         }
2381
2382         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2383                                            dentry->d_name.len);
2384         if (status)
2385                 goto leave;
2386
2387         /* get a spot inside the dir. */
2388         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2389                                               dentry->d_name.name,
2390                                               dentry->d_name.len, &lookup);
2391         if (status < 0) {
2392                 mlog_errno(status);
2393                 goto leave;
2394         }
2395
2396         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2397                                                        ORPHAN_DIR_SYSTEM_INODE,
2398                                                        osb->slot_num);
2399         if (!orphan_dir_inode) {
2400                 status = -EEXIST;
2401                 mlog_errno(status);
2402                 goto leave;
2403         }
2404
2405         mutex_lock(&orphan_dir_inode->i_mutex);
2406
2407         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2408         if (status < 0) {
2409                 mlog_errno(status);
2410                 mutex_unlock(&orphan_dir_inode->i_mutex);
2411                 iput(orphan_dir_inode);
2412                 goto leave;
2413         }
2414
2415         status = ocfs2_read_inode_block(inode, &di_bh);
2416         if (status < 0) {
2417                 mlog_errno(status);
2418                 goto orphan_unlock;
2419         }
2420
2421         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2422         if (IS_ERR(handle)) {
2423                 status = PTR_ERR(handle);
2424                 handle = NULL;
2425                 mlog_errno(status);
2426                 goto orphan_unlock;
2427         }
2428
2429         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2430                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2431         if (status < 0) {
2432                 mlog_errno(status);
2433                 goto out_commit;
2434         }
2435
2436         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2437                                   orphan_dir_bh);
2438         if (status < 0) {
2439                 mlog_errno(status);
2440                 goto out_commit;
2441         }
2442
2443         di = (struct ocfs2_dinode *)di_bh->b_data;
2444         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2445         di->i_orphaned_slot = 0;
2446         inode->i_nlink = 1;
2447         ocfs2_set_links_count(di, inode->i_nlink);
2448         ocfs2_journal_dirty(handle, di_bh);
2449
2450         status = ocfs2_add_entry(handle, dentry, inode,
2451                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2452                                  &lookup);
2453         if (status < 0) {
2454                 mlog_errno(status);
2455                 goto out_commit;
2456         }
2457
2458         status = ocfs2_dentry_attach_lock(dentry, inode,
2459                                           OCFS2_I(dir)->ip_blkno);
2460         if (status) {
2461                 mlog_errno(status);
2462                 goto out_commit;
2463         }
2464
2465         d_set_d_op(dentry, &ocfs2_dentry_ops);
2466         d_instantiate(dentry, inode);
2467         status = 0;
2468 out_commit:
2469         ocfs2_commit_trans(osb, handle);
2470 orphan_unlock:
2471         ocfs2_inode_unlock(orphan_dir_inode, 1);
2472         mutex_unlock(&orphan_dir_inode->i_mutex);
2473         iput(orphan_dir_inode);
2474 leave:
2475
2476         ocfs2_inode_unlock(dir, 1);
2477
2478         brelse(di_bh);
2479         brelse(parent_di_bh);
2480         brelse(orphan_dir_bh);
2481
2482         ocfs2_free_dir_lookup_result(&lookup);
2483
2484         mlog_exit(status);
2485
2486         return status;
2487 }
2488
2489 const struct inode_operations ocfs2_dir_iops = {
2490         .create         = ocfs2_create,
2491         .lookup         = ocfs2_lookup,
2492         .link           = ocfs2_link,
2493         .unlink         = ocfs2_unlink,
2494         .rmdir          = ocfs2_unlink,
2495         .symlink        = ocfs2_symlink,
2496         .mkdir          = ocfs2_mkdir,
2497         .mknod          = ocfs2_mknod,
2498         .rename         = ocfs2_rename,
2499         .setattr        = ocfs2_setattr,
2500         .getattr        = ocfs2_getattr,
2501         .permission     = ocfs2_permission,
2502         .setxattr       = generic_setxattr,
2503         .getxattr       = generic_getxattr,
2504         .listxattr      = ocfs2_listxattr,
2505         .removexattr    = generic_removexattr,
2506         .fiemap         = ocfs2_fiemap,
2507 };