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

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/quotaops.h>
  44
  45 #define MLOG_MASK_PREFIX ML_NAMEI
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 #include "alloc.h"
  51 #include "dcache.h"
  52 #include "dir.h"
  53 #include "dlmglue.h"
  54 #include "extent_map.h"
  55 #include "file.h"
  56 #include "inode.h"
  57 #include "journal.h"
  58 #include "namei.h"
  59 #include "suballoc.h"
  60 #include "super.h"
  61 #include "symlink.h"
  62 #include "sysfile.h"
  63 #include "uptodate.h"
  64 #include "xattr.h"
  65 #include "acl.h"
  66
  67 #include "buffer_head_io.h"
  68
  69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  70                               struct inode *dir,
  71                               struct inode *inode,
  72                               dev_t dev,
  73                               struct buffer_head **new_fe_bh,
  74                               struct buffer_head *parent_fe_bh,
  75                               handle_t *handle,
  76                               struct ocfs2_alloc_context *inode_ac);
  77
  78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  79                                     struct inode **ret_orphan_dir,
  80                                     u64 blkno,
  81                                     char *name,
  82                                     struct ocfs2_dir_lookup_result *lookup);
  83
  84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  85                             handle_t *handle,
  86                             struct inode *inode,
  87                             struct buffer_head *fe_bh,
  88                             char *name,
  89                             struct ocfs2_dir_lookup_result *lookup,
  90                             struct inode *orphan_dir_inode);
  91
  92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  93                                      handle_t *handle,
  94                                      struct inode *inode,
  95                                      const char *symname);
  96
  97 /* An orphan dir name is an 8 byte value, printed as a hex string */
  98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  99
 100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 101                                    struct nameidata *nd)
 102 {
 103         int status;
 104         u64 blkno;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_inode_info *oi;
 108
 109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 110                    dentry->d_name.len, dentry->d_name.name);
 111
 112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 113                 ret = ERR_PTR(-ENAMETOOLONG);
 114                 goto bail;
 115         }
 116
 117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 119
 120         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 121         if (status < 0) {
 122                 if (status != -ENOENT)
 123                         mlog_errno(status);
 124                 ret = ERR_PTR(status);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 129                                             dentry->d_name.len, &blkno);
 130         if (status < 0)
 131                 goto bail_add;
 132
 133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 134         if (IS_ERR(inode)) {
 135                 ret = ERR_PTR(-EACCES);
 136                 goto bail_unlock;
 137         }
 138
 139         oi = OCFS2_I(inode);
 140         /* Clear any orphaned state... If we were able to look up the
 141          * inode from a directory, it certainly can't be orphaned. We
 142          * might have the bad state from a node which intended to
 143          * orphan this inode but crashed before it could commit the
 144          * unlink. */
 145         spin_lock(&oi->ip_lock);
 146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 147         spin_unlock(&oi->ip_lock);
 148
 149 bail_add:
 150         dentry->d_op = &ocfs2_dentry_ops;
 151         ret = d_splice_alias(inode, dentry);
 152
 153         if (inode) {
 154                 /*
 155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 156                  * dentry, it will d_move() it on top of ourse. The
 157                  * return value will indicate this however, so in
 158                  * those cases, we switch them around for the locking
 159                  * code.
 160                  *
 161                  * NOTE: This dentry already has ->d_op set from
 162                  * ocfs2_get_parent() and ocfs2_get_dentry()
 163                  */
 164                 if (ret)
 165                         dentry = ret;
 166
 167                 status = ocfs2_dentry_attach_lock(dentry, inode,
 168                                                   OCFS2_I(dir)->ip_blkno);
 169                 if (status) {
 170                         mlog_errno(status);
 171                         ret = ERR_PTR(status);
 172                         goto bail_unlock;
 173                 }
 174         }
 175
 176 bail_unlock:
 177         /* Don't drop the cluster lock until *after* the d_add --
 178          * unlink on another node will message us to remove that
 179          * dentry under this lock so otherwise we can race this with
 180          * the downconvert thread and have a stale dentry. */
 181         ocfs2_inode_unlock(dir, 0);
 182
 183 bail:
 184
 185         mlog_exit_ptr(ret);
 186
 187         return ret;
 188 }
 189
 190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
 191 {
 192         struct inode *inode;
 193
 194         inode = new_inode(dir->i_sb);
 195         if (!inode) {
 196                 mlog(ML_ERROR, "new_inode failed!\n");
 197                 return NULL;
 198         }
 199
 200         /* populate as many fields early on as possible - many of
 201          * these are used by the support functions here and in
 202          * callers. */
 203         if (S_ISDIR(mode))
 204                 inode->i_nlink = 2;
 205         else
 206                 inode->i_nlink = 1;
 207         inode_init_owner(inode, dir, mode);
 208         dquot_initialize(inode);
 209         return inode;
 210 }
 211
 212 static int ocfs2_mknod(struct inode *dir,
 213                        struct dentry *dentry,
 214                        int mode,
 215                        dev_t dev)
 216 {
 217         int status = 0;
 218         struct buffer_head *parent_fe_bh = NULL;
 219         handle_t *handle = NULL;
 220         struct ocfs2_super *osb;
 221         struct ocfs2_dinode *dirfe;
 222         struct buffer_head *new_fe_bh = NULL;
 223         struct inode *inode = NULL;
 224         struct ocfs2_alloc_context *inode_ac = NULL;
 225         struct ocfs2_alloc_context *data_ac = NULL;
 226         struct ocfs2_alloc_context *meta_ac = NULL;
 227         int want_clusters = 0;
 228         int want_meta = 0;
 229         int xattr_credits = 0;
 230         struct ocfs2_security_xattr_info si = {
 231                 .enable = 1,
 232         };
 233         int did_quota_inode = 0;
 234         struct ocfs2_dir_lookup_result lookup = { NULL, };
 235         sigset_t oldset;
 236         int did_block_signals = 0;
 237
 238         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 239                    (unsigned long)dev, dentry->d_name.len,
 240                    dentry->d_name.name);
 241
 242         dquot_initialize(dir);
 243
 244         /* get our super block */
 245         osb = OCFS2_SB(dir->i_sb);
 246
 247         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 248         if (status < 0) {
 249                 if (status != -ENOENT)
 250                         mlog_errno(status);
 251                 return status;
 252         }
 253
 254         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 255                 status = -EMLINK;
 256                 goto leave;
 257         }
 258
 259         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 260         if (!ocfs2_read_links_count(dirfe)) {
 261                 /* can't make a file in a deleted directory. */
 262                 status = -ENOENT;
 263                 goto leave;
 264         }
 265
 266         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 267                                            dentry->d_name.len);
 268         if (status)
 269                 goto leave;
 270
 271         /* get a spot inside the dir. */
 272         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 273                                               dentry->d_name.name,
 274                                               dentry->d_name.len, &lookup);
 275         if (status < 0) {
 276                 mlog_errno(status);
 277                 goto leave;
 278         }
 279
 280         /* reserve an inode spot */
 281         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 282         if (status < 0) {
 283                 if (status != -ENOSPC)
 284                         mlog_errno(status);
 285                 goto leave;
 286         }
 287
 288         inode = ocfs2_get_init_inode(dir, mode);
 289         if (!inode) {
 290                 status = -ENOMEM;
 291                 mlog_errno(status);
 292                 goto leave;
 293         }
 294
 295         /* get security xattr */
 296         status = ocfs2_init_security_get(inode, dir, &si);
 297         if (status) {
 298                 if (status == -EOPNOTSUPP)
 299                         si.enable = 0;
 300                 else {
 301                         mlog_errno(status);
 302                         goto leave;
 303                 }
 304         }
 305
 306         /* calculate meta data/clusters for setting security and acl xattr */
 307         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 308                                        &si, &want_clusters,
 309                                        &xattr_credits, &want_meta);
 310         if (status < 0) {
 311                 mlog_errno(status);
 312                 goto leave;
 313         }
 314
 315         /* Reserve a cluster if creating an extent based directory. */
 316         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 317                 want_clusters += 1;
 318
 319                 /* Dir indexing requires extra space as well */
 320                 if (ocfs2_supports_indexed_dirs(osb))
 321                         want_meta++;
 322         }
 323
 324         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 325         if (status < 0) {
 326                 if (status != -ENOSPC)
 327                         mlog_errno(status);
 328                 goto leave;
 329         }
 330
 331         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 332         if (status < 0) {
 333                 if (status != -ENOSPC)
 334                         mlog_errno(status);
 335                 goto leave;
 336         }
 337
 338         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 339                                                             S_ISDIR(mode),
 340                                                             xattr_credits));
 341         if (IS_ERR(handle)) {
 342                 status = PTR_ERR(handle);
 343                 handle = NULL;
 344                 mlog_errno(status);
 345                 goto leave;
 346         }
 347
 348         /* Starting to change things, restart is no longer possible. */
 349         ocfs2_block_signals(&oldset);
 350         did_block_signals = 1;
 351
 352         status = dquot_alloc_inode(inode);
 353         if (status)
 354                 goto leave;
 355         did_quota_inode = 1;
 356
 357         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
 358                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
 359                    dentry->d_name.name);
 360
 361         /* do the real work now. */
 362         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 363                                     &new_fe_bh, parent_fe_bh, handle,
 364                                     inode_ac);
 365         if (status < 0) {
 366                 mlog_errno(status);
 367                 goto leave;
 368         }
 369
 370         if (S_ISDIR(mode)) {
 371                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 372                                             new_fe_bh, data_ac, meta_ac);
 373                 if (status < 0) {
 374                         mlog_errno(status);
 375                         goto leave;
 376                 }
 377
 378                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 379                                                  parent_fe_bh,
 380                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 381                 if (status < 0) {
 382                         mlog_errno(status);
 383                         goto leave;
 384                 }
 385                 ocfs2_add_links_count(dirfe, 1);
 386                 ocfs2_journal_dirty(handle, parent_fe_bh);
 387                 inc_nlink(dir);
 388         }
 389
 390         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 391                                 meta_ac, data_ac);
 392         if (status < 0) {
 393                 mlog_errno(status);
 394                 goto leave;
 395         }
 396
 397         if (si.enable) {
 398                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 399                                                  meta_ac, data_ac);
 400                 if (status < 0) {
 401                         mlog_errno(status);
 402                         goto leave;
 403                 }
 404         }
 405
 406         /*
 407          * Do this before adding the entry to the directory. We add
 408          * also set d_op after success so that ->d_iput() will cleanup
 409          * the dentry lock even if ocfs2_add_entry() fails below.
 410          */
 411         status = ocfs2_dentry_attach_lock(dentry, inode,
 412                                           OCFS2_I(dir)->ip_blkno);
 413         if (status) {
 414                 mlog_errno(status);
 415                 goto leave;
 416         }
 417         dentry->d_op = &ocfs2_dentry_ops;
 418
 419         status = ocfs2_add_entry(handle, dentry, inode,
 420                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 421                                  &lookup);
 422         if (status < 0) {
 423                 mlog_errno(status);
 424                 goto leave;
 425         }
 426
 427         insert_inode_hash(inode);
 428         d_instantiate(dentry, inode);
 429         status = 0;
 430 leave:
 431         if (status < 0 && did_quota_inode)
 432                 dquot_free_inode(inode);
 433         if (handle)
 434                 ocfs2_commit_trans(osb, handle);
 435
 436         ocfs2_inode_unlock(dir, 1);
 437         if (did_block_signals)
 438                 ocfs2_unblock_signals(&oldset);
 439
 440         if (status == -ENOSPC)
 441                 mlog(0, "Disk is full\n");
 442
 443         brelse(new_fe_bh);
 444         brelse(parent_fe_bh);
 445         kfree(si.name);
 446         kfree(si.value);
 447
 448         ocfs2_free_dir_lookup_result(&lookup);
 449
 450         if (inode_ac)
 451                 ocfs2_free_alloc_context(inode_ac);
 452
 453         if (data_ac)
 454                 ocfs2_free_alloc_context(data_ac);
 455
 456         if (meta_ac)
 457                 ocfs2_free_alloc_context(meta_ac);
 458
 459         /*
 460          * We should call iput after the i_mutex of the bitmap been
 461          * unlocked in ocfs2_free_alloc_context, or the
 462          * ocfs2_delete_inode will mutex_lock again.
 463          */
 464         if ((status < 0) && inode) {
 465                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 466                 clear_nlink(inode);
 467                 iput(inode);
 468         }
 469
 470         mlog_exit(status);
 471
 472         return status;
 473 }
 474
 475 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 476                               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 {
 484         int status = 0;
 485         struct ocfs2_dinode *fe = NULL;
 486         struct ocfs2_extent_list *fel;
 487         u64 suballoc_loc, fe_blkno = 0;
 488         u16 suballoc_bit;
 489         u16 feat;
 490
 491         *new_fe_bh = NULL;
 492
 493         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 494                                        inode_ac, &suballoc_loc,
 495                                        &suballoc_bit, &fe_blkno);
 496         if (status < 0) {
 497                 mlog_errno(status);
 498                 goto leave;
 499         }
 500
 501         /* populate as many fields early on as possible - many of
 502          * these are used by the support functions here and in
 503          * callers. */
 504         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 505         OCFS2_I(inode)->ip_blkno = fe_blkno;
 506         spin_lock(&osb->osb_lock);
 507         inode->i_generation = osb->s_next_generation++;
 508         spin_unlock(&osb->osb_lock);
 509
 510         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 511         if (!*new_fe_bh) {
 512                 status = -EIO;
 513                 mlog_errno(status);
 514                 goto leave;
 515         }
 516         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 517
 518         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 519                                          *new_fe_bh,
 520                                          OCFS2_JOURNAL_ACCESS_CREATE);
 521         if (status < 0) {
 522                 mlog_errno(status);
 523                 goto leave;
 524         }
 525
 526         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 527         memset(fe, 0, osb->sb->s_blocksize);
 528
 529         fe->i_generation = cpu_to_le32(inode->i_generation);
 530         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 531         fe->i_blkno = cpu_to_le64(fe_blkno);
 532         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 533         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 534         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 535         fe->i_uid = cpu_to_le32(inode->i_uid);
 536         fe->i_gid = cpu_to_le32(inode->i_gid);
 537         fe->i_mode = cpu_to_le16(inode->i_mode);
 538         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 539                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 540
 541         ocfs2_set_links_count(fe, inode->i_nlink);
 542
 543         fe->i_last_eb_blk = 0;
 544         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 545         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 546         fe->i_atime = fe->i_ctime = fe->i_mtime =
 547                 cpu_to_le64(CURRENT_TIME.tv_sec);
 548         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 549                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 550         fe->i_dtime = 0;
 551
 552         /*
 553          * If supported, directories start with inline data. If inline
 554          * isn't supported, but indexing is, we start them as indexed.
 555          */
 556         feat = le16_to_cpu(fe->i_dyn_features);
 557         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 558                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 559
 560                 fe->id2.i_data.id_count = cpu_to_le16(
 561                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 562         } else {
 563                 fel = &fe->id2.i_list;
 564                 fel->l_tree_depth = 0;
 565                 fel->l_next_free_rec = 0;
 566                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 567         }
 568
 569         ocfs2_journal_dirty(handle, *new_fe_bh);
 570
 571         ocfs2_populate_inode(inode, fe, 1);
 572         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 573         if (!ocfs2_mount_local(osb)) {
 574                 status = ocfs2_create_new_inode_locks(inode);
 575                 if (status < 0)
 576                         mlog_errno(status);
 577         }
 578
 579         status = 0; /* error in ocfs2_create_new_inode_locks is not
 580                      * critical */
 581
 582 leave:
 583         if (status < 0) {
 584                 if (*new_fe_bh) {
 585                         brelse(*new_fe_bh);
 586                         *new_fe_bh = NULL;
 587                 }
 588         }
 589
 590         mlog_exit(status);
 591         return status;
 592 }
 593
 594 static int ocfs2_mkdir(struct inode *dir,
 595                        struct dentry *dentry,
 596                        int mode)
 597 {
 598         int ret;
 599
 600         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 601                    dentry->d_name.len, dentry->d_name.name);
 602         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 603         mlog_exit(ret);
 604
 605         return ret;
 606 }
 607
 608 static int ocfs2_create(struct inode *dir,
 609                         struct dentry *dentry,
 610                         int mode,
 611                         struct nameidata *nd)
 612 {
 613         int ret;
 614
 615         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 616                    dentry->d_name.len, dentry->d_name.name);
 617         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 618         mlog_exit(ret);
 619
 620         return ret;
 621 }
 622
 623 static int ocfs2_link(struct dentry *old_dentry,
 624                       struct inode *dir,
 625                       struct dentry *dentry)
 626 {
 627         handle_t *handle;
 628         struct inode *inode = old_dentry->d_inode;
 629         int err;
 630         struct buffer_head *fe_bh = NULL;
 631         struct buffer_head *parent_fe_bh = NULL;
 632         struct ocfs2_dinode *fe = NULL;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634         struct ocfs2_dir_lookup_result lookup = { NULL, };
 635         sigset_t oldset;
 636
 637         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 638                    old_dentry->d_name.len, old_dentry->d_name.name,
 639                    dentry->d_name.len, dentry->d_name.name);
 640
 641         if (S_ISDIR(inode->i_mode))
 642                 return -EPERM;
 643
 644         dquot_initialize(dir);
 645
 646         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 647         if (err < 0) {
 648                 if (err != -ENOENT)
 649                         mlog_errno(err);
 650                 return err;
 651         }
 652
 653         if (!dir->i_nlink) {
 654                 err = -ENOENT;
 655                 goto out;
 656         }
 657
 658         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 659                                         dentry->d_name.len);
 660         if (err)
 661                 goto out;
 662
 663         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 664                                            dentry->d_name.name,
 665                                            dentry->d_name.len, &lookup);
 666         if (err < 0) {
 667                 mlog_errno(err);
 668                 goto out;
 669         }
 670
 671         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 672         if (err < 0) {
 673                 if (err != -ENOENT)
 674                         mlog_errno(err);
 675                 goto out;
 676         }
 677
 678         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 679         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 680                 err = -EMLINK;
 681                 goto out_unlock_inode;
 682         }
 683
 684         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 685         if (IS_ERR(handle)) {
 686                 err = PTR_ERR(handle);
 687                 handle = NULL;
 688                 mlog_errno(err);
 689                 goto out_unlock_inode;
 690         }
 691
 692         /* Starting to change things, restart is no longer possible. */
 693         ocfs2_block_signals(&oldset);
 694
 695         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 696                                       OCFS2_JOURNAL_ACCESS_WRITE);
 697         if (err < 0) {
 698                 mlog_errno(err);
 699                 goto out_commit;
 700         }
 701
 702         inc_nlink(inode);
 703         inode->i_ctime = CURRENT_TIME;
 704         ocfs2_set_links_count(fe, inode->i_nlink);
 705         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 706         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 707         ocfs2_journal_dirty(handle, fe_bh);
 708
 709         err = ocfs2_add_entry(handle, dentry, inode,
 710                               OCFS2_I(inode)->ip_blkno,
 711                               parent_fe_bh, &lookup);
 712         if (err) {
 713                 ocfs2_add_links_count(fe, -1);
 714                 drop_nlink(inode);
 715                 mlog_errno(err);
 716                 goto out_commit;
 717         }
 718
 719         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 720         if (err) {
 721                 mlog_errno(err);
 722                 goto out_commit;
 723         }
 724
 725         atomic_inc(&inode->i_count);
 726         dentry->d_op = &ocfs2_dentry_ops;
 727         d_instantiate(dentry, inode);
 728
 729 out_commit:
 730         ocfs2_commit_trans(osb, handle);
 731         ocfs2_unblock_signals(&oldset);
 732 out_unlock_inode:
 733         ocfs2_inode_unlock(inode, 1);
 734
 735 out:
 736         ocfs2_inode_unlock(dir, 1);
 737
 738         brelse(fe_bh);
 739         brelse(parent_fe_bh);
 740
 741         ocfs2_free_dir_lookup_result(&lookup);
 742
 743         mlog_exit(err);
 744
 745         return err;
 746 }
 747
 748 /*
 749  * Takes and drops an exclusive lock on the given dentry. This will
 750  * force other nodes to drop it.
 751  */
 752 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 753 {
 754         int ret;
 755
 756         ret = ocfs2_dentry_lock(dentry, 1);
 757         if (ret)
 758                 mlog_errno(ret);
 759         else
 760                 ocfs2_dentry_unlock(dentry, 1);
 761
 762         return ret;
 763 }
 764
 765 static inline int inode_is_unlinkable(struct inode *inode)
 766 {
 767         if (S_ISDIR(inode->i_mode)) {
 768                 if (inode->i_nlink == 2)
 769                         return 1;
 770                 return 0;
 771         }
 772
 773         if (inode->i_nlink == 1)
 774                 return 1;
 775         return 0;
 776 }
 777
 778 static int ocfs2_unlink(struct inode *dir,
 779                         struct dentry *dentry)
 780 {
 781         int status;
 782         int child_locked = 0;
 783         struct inode *inode = dentry->d_inode;
 784         struct inode *orphan_dir = NULL;
 785         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 786         u64 blkno;
 787         struct ocfs2_dinode *fe = NULL;
 788         struct buffer_head *fe_bh = NULL;
 789         struct buffer_head *parent_node_bh = NULL;
 790         handle_t *handle = NULL;
 791         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 792         struct ocfs2_dir_lookup_result lookup = { NULL, };
 793         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 794
 795         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 796                    dentry->d_name.len, dentry->d_name.name);
 797
 798         dquot_initialize(dir);
 799
 800         BUG_ON(dentry->d_parent->d_inode != dir);
 801
 802         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 803
 804         if (inode == osb->root_inode) {
 805                 mlog(0, "Cannot delete the root directory\n");
 806                 return -EPERM;
 807         }
 808
 809         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 810                                          OI_LS_PARENT);
 811         if (status < 0) {
 812                 if (status != -ENOENT)
 813                         mlog_errno(status);
 814                 return status;
 815         }
 816
 817         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 818                                           dentry->d_name.len, &blkno, dir,
 819                                           &lookup);
 820         if (status < 0) {
 821                 if (status != -ENOENT)
 822                         mlog_errno(status);
 823                 goto leave;
 824         }
 825
 826         if (OCFS2_I(inode)->ip_blkno != blkno) {
 827                 status = -ENOENT;
 828
 829                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 830                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 831                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 832                 goto leave;
 833         }
 834
 835         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 836         if (status < 0) {
 837                 if (status != -ENOENT)
 838                         mlog_errno(status);
 839                 goto leave;
 840         }
 841         child_locked = 1;
 842
 843         if (S_ISDIR(inode->i_mode)) {
 844                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 845                         status = -ENOTEMPTY;
 846                         goto leave;
 847                 }
 848         }
 849
 850         status = ocfs2_remote_dentry_delete(dentry);
 851         if (status < 0) {
 852                 /* This remote delete should succeed under all normal
 853                  * circumstances. */
 854                 mlog_errno(status);
 855                 goto leave;
 856         }
 857
 858         if (inode_is_unlinkable(inode)) {
 859                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 860                                                   OCFS2_I(inode)->ip_blkno,
 861                                                   orphan_name, &orphan_insert);
 862                 if (status < 0) {
 863                         mlog_errno(status);
 864                         goto leave;
 865                 }
 866         }
 867
 868         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 869         if (IS_ERR(handle)) {
 870                 status = PTR_ERR(handle);
 871                 handle = NULL;
 872                 mlog_errno(status);
 873                 goto leave;
 874         }
 875
 876         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 877                                          OCFS2_JOURNAL_ACCESS_WRITE);
 878         if (status < 0) {
 879                 mlog_errno(status);
 880                 goto leave;
 881         }
 882
 883         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 884
 885         if (inode_is_unlinkable(inode)) {
 886                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 887                                           &orphan_insert, orphan_dir);
 888                 if (status < 0) {
 889                         mlog_errno(status);
 890                         goto leave;
 891                 }
 892         }
 893
 894         /* delete the name from the parent dir */
 895         status = ocfs2_delete_entry(handle, dir, &lookup);
 896         if (status < 0) {
 897                 mlog_errno(status);
 898                 goto leave;
 899         }
 900
 901         if (S_ISDIR(inode->i_mode))
 902                 drop_nlink(inode);
 903         drop_nlink(inode);
 904         ocfs2_set_links_count(fe, inode->i_nlink);
 905         ocfs2_journal_dirty(handle, fe_bh);
 906
 907         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 908         if (S_ISDIR(inode->i_mode))
 909                 drop_nlink(dir);
 910
 911         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 912         if (status < 0) {
 913                 mlog_errno(status);
 914                 if (S_ISDIR(inode->i_mode))
 915                         inc_nlink(dir);
 916         }
 917
 918 leave:
 919         if (handle)
 920                 ocfs2_commit_trans(osb, handle);
 921
 922         if (child_locked)
 923                 ocfs2_inode_unlock(inode, 1);
 924
 925         ocfs2_inode_unlock(dir, 1);
 926
 927         if (orphan_dir) {
 928                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 929                 ocfs2_inode_unlock(orphan_dir, 1);
 930                 mutex_unlock(&orphan_dir->i_mutex);
 931                 iput(orphan_dir);
 932         }
 933
 934         brelse(fe_bh);
 935         brelse(parent_node_bh);
 936
 937         ocfs2_free_dir_lookup_result(&orphan_insert);
 938         ocfs2_free_dir_lookup_result(&lookup);
 939
 940         mlog_exit(status);
 941
 942         return status;
 943 }
 944
 945 /*
 946  * The only place this should be used is rename!
 947  * if they have the same id, then the 1st one is the only one locked.
 948  */
 949 static int ocfs2_double_lock(struct ocfs2_super *osb,
 950                              struct buffer_head **bh1,
 951                              struct inode *inode1,
 952                              struct buffer_head **bh2,
 953                              struct inode *inode2)
 954 {
 955         int status;
 956         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 957         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 958         struct buffer_head **tmpbh;
 959         struct inode *tmpinode;
 960
 961         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 962                    (unsigned long long)oi1->ip_blkno,
 963                    (unsigned long long)oi2->ip_blkno);
 964
 965         if (*bh1)
 966                 *bh1 = NULL;
 967         if (*bh2)
 968                 *bh2 = NULL;
 969
 970         /* we always want to lock the one with the lower lockid first. */
 971         if (oi1->ip_blkno != oi2->ip_blkno) {
 972                 if (oi1->ip_blkno < oi2->ip_blkno) {
 973                         /* switch id1 and id2 around */
 974                         mlog(0, "switching them around...\n");
 975                         tmpbh = bh2;
 976                         bh2 = bh1;
 977                         bh1 = tmpbh;
 978
 979                         tmpinode = inode2;
 980                         inode2 = inode1;
 981                         inode1 = tmpinode;
 982                 }
 983                 /* lock id2 */
 984                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 985                                                  OI_LS_RENAME1);
 986                 if (status < 0) {
 987                         if (status != -ENOENT)
 988                                 mlog_errno(status);
 989                         goto bail;
 990                 }
 991         }
 992
 993         /* lock id1 */
 994         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
 995         if (status < 0) {
 996                 /*
 997                  * An error return must mean that no cluster locks
 998                  * were held on function exit.
 999                  */
1000                 if (oi1->ip_blkno != oi2->ip_blkno)
1001                         ocfs2_inode_unlock(inode2, 1);
1002
1003                 if (status != -ENOENT)
1004                         mlog_errno(status);
1005         }
1006
1007 bail:
1008         mlog_exit(status);
1009         return status;
1010 }
1011
1012 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1013 {
1014         ocfs2_inode_unlock(inode1, 1);
1015
1016         if (inode1 != inode2)
1017                 ocfs2_inode_unlock(inode2, 1);
1018 }
1019
1020 static int ocfs2_rename(struct inode *old_dir,
1021                         struct dentry *old_dentry,
1022                         struct inode *new_dir,
1023                         struct dentry *new_dentry)
1024 {
1025         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1026         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1027         struct inode *old_inode = old_dentry->d_inode;
1028         struct inode *new_inode = new_dentry->d_inode;
1029         struct inode *orphan_dir = NULL;
1030         struct ocfs2_dinode *newfe = NULL;
1031         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1032         struct buffer_head *newfe_bh = NULL;
1033         struct buffer_head *old_inode_bh = NULL;
1034         struct ocfs2_super *osb = NULL;
1035         u64 newfe_blkno, old_de_ino;
1036         handle_t *handle = NULL;
1037         struct buffer_head *old_dir_bh = NULL;
1038         struct buffer_head *new_dir_bh = NULL;
1039         nlink_t old_dir_nlink = old_dir->i_nlink;
1040         struct ocfs2_dinode *old_di;
1041         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1042         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1043         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1044         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1045         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1046
1047         /* At some point it might be nice to break this function up a
1048          * bit. */
1049
1050         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1051                    old_dir, old_dentry, new_dir, new_dentry,
1052                    old_dentry->d_name.len, old_dentry->d_name.name,
1053                    new_dentry->d_name.len, new_dentry->d_name.name);
1054
1055         dquot_initialize(old_dir);
1056         dquot_initialize(new_dir);
1057
1058         osb = OCFS2_SB(old_dir->i_sb);
1059
1060         if (new_inode) {
1061                 if (!igrab(new_inode))
1062                         BUG();
1063         }
1064
1065         /* Assume a directory hierarchy thusly:
1066          * a/b/c
1067          * a/d
1068          * a,b,c, and d are all directories.
1069          *
1070          * from cwd of 'a' on both nodes:
1071          * node1: mv b/c d
1072          * node2: mv d   b/c
1073          *
1074          * And that's why, just like the VFS, we need a file system
1075          * rename lock. */
1076         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1077                 status = ocfs2_rename_lock(osb);
1078                 if (status < 0) {
1079                         mlog_errno(status);
1080                         goto bail;
1081                 }
1082                 rename_lock = 1;
1083         }
1084
1085         /* if old and new are the same, this'll just do one lock. */
1086         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1087                                    &new_dir_bh, new_dir);
1088         if (status < 0) {
1089                 mlog_errno(status);
1090                 goto bail;
1091         }
1092         parents_locked = 1;
1093
1094         /* make sure both dirs have bhs
1095          * get an extra ref on old_dir_bh if old==new */
1096         if (!new_dir_bh) {
1097                 if (old_dir_bh) {
1098                         new_dir_bh = old_dir_bh;
1099                         get_bh(new_dir_bh);
1100                 } else {
1101                         mlog(ML_ERROR, "no old_dir_bh!\n");
1102                         status = -EIO;
1103                         goto bail;
1104                 }
1105         }
1106
1107         /*
1108          * Aside from allowing a meta data update, the locking here
1109          * also ensures that the downconvert thread on other nodes
1110          * won't have to concurrently downconvert the inode and the
1111          * dentry locks.
1112          */
1113         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1114                                          OI_LS_PARENT);
1115         if (status < 0) {
1116                 if (status != -ENOENT)
1117                         mlog_errno(status);
1118                 goto bail;
1119         }
1120         old_child_locked = 1;
1121
1122         status = ocfs2_remote_dentry_delete(old_dentry);
1123         if (status < 0) {
1124                 mlog_errno(status);
1125                 goto bail;
1126         }
1127
1128         if (S_ISDIR(old_inode->i_mode)) {
1129                 u64 old_inode_parent;
1130
1131                 update_dot_dot = 1;
1132                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1133                                                   old_inode,
1134                                                   &old_inode_dot_dot_res);
1135                 if (status) {
1136                         status = -EIO;
1137                         goto bail;
1138                 }
1139
1140                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1141                         status = -EIO;
1142                         goto bail;
1143                 }
1144
1145                 if (!new_inode && new_dir != old_dir &&
1146                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1147                         status = -EMLINK;
1148                         goto bail;
1149                 }
1150         }
1151
1152         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1153                                             old_dentry->d_name.len,
1154                                             &old_de_ino);
1155         if (status) {
1156                 status = -ENOENT;
1157                 goto bail;
1158         }
1159
1160         /*
1161          *  Check for inode number is _not_ due to possible IO errors.
1162          *  We might rmdir the source, keep it as pwd of some process
1163          *  and merrily kill the link to whatever was created under the
1164          *  same name. Goodbye sticky bit ;-<
1165          */
1166         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1167                 status = -ENOENT;
1168                 goto bail;
1169         }
1170
1171         /* check if the target already exists (in which case we need
1172          * to delete it */
1173         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1174                                           new_dentry->d_name.len,
1175                                           &newfe_blkno, new_dir,
1176                                           &target_lookup_res);
1177         /* The only error we allow here is -ENOENT because the new
1178          * file not existing is perfectly valid. */
1179         if ((status < 0) && (status != -ENOENT)) {
1180                 /* If we cannot find the file specified we should just */
1181                 /* return the error... */
1182                 mlog_errno(status);
1183                 goto bail;
1184         }
1185         if (status == 0)
1186                 target_exists = 1;
1187
1188         if (!target_exists && new_inode) {
1189                 /*
1190                  * Target was unlinked by another node while we were
1191                  * waiting to get to ocfs2_rename(). There isn't
1192                  * anything we can do here to help the situation, so
1193                  * bubble up the appropriate error.
1194                  */
1195                 status = -ENOENT;
1196                 goto bail;
1197         }
1198
1199         /* In case we need to overwrite an existing file, we blow it
1200          * away first */
1201         if (target_exists) {
1202                 /* VFS didn't think there existed an inode here, but
1203                  * someone else in the cluster must have raced our
1204                  * rename to create one. Today we error cleanly, in
1205                  * the future we should consider calling iget to build
1206                  * a new struct inode for this entry. */
1207                 if (!new_inode) {
1208                         status = -EACCES;
1209
1210                         mlog(0, "We found an inode for name %.*s but VFS "
1211                              "didn't give us one.\n", new_dentry->d_name.len,
1212                              new_dentry->d_name.name);
1213                         goto bail;
1214                 }
1215
1216                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1217                         status = -EACCES;
1218
1219                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1220                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1221                              (unsigned long long)newfe_blkno,
1222                              OCFS2_I(new_inode)->ip_flags);
1223                         goto bail;
1224                 }
1225
1226                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1227                 if (status < 0) {
1228                         if (status != -ENOENT)
1229                                 mlog_errno(status);
1230                         goto bail;
1231                 }
1232                 new_child_locked = 1;
1233
1234                 status = ocfs2_remote_dentry_delete(new_dentry);
1235                 if (status < 0) {
1236                         mlog_errno(status);
1237                         goto bail;
1238                 }
1239
1240                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1241
1242                 mlog(0, "aha rename over existing... new_blkno=%llu "
1243                      "newfebh=%p bhblocknr=%llu\n",
1244                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1245                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1246
1247                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1248                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1249                                                 OCFS2_I(new_inode)->ip_blkno,
1250                                                 orphan_name, &orphan_insert);
1251                         if (status < 0) {
1252                                 mlog_errno(status);
1253                                 goto bail;
1254                         }
1255                 }
1256         } else {
1257                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1258
1259                 status = ocfs2_check_dir_for_entry(new_dir,
1260                                                    new_dentry->d_name.name,
1261                                                    new_dentry->d_name.len);
1262                 if (status)
1263                         goto bail;
1264
1265                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1266                                                       new_dentry->d_name.name,
1267                                                       new_dentry->d_name.len,
1268                                                       &target_insert);
1269                 if (status < 0) {
1270                         mlog_errno(status);
1271                         goto bail;
1272                 }
1273         }
1274
1275         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1276         if (IS_ERR(handle)) {
1277                 status = PTR_ERR(handle);
1278                 handle = NULL;
1279                 mlog_errno(status);
1280                 goto bail;
1281         }
1282
1283         if (target_exists) {
1284                 if (S_ISDIR(new_inode->i_mode)) {
1285                         if (new_inode->i_nlink != 2 ||
1286                             !ocfs2_empty_dir(new_inode)) {
1287                                 status = -ENOTEMPTY;
1288                                 goto bail;
1289                         }
1290                 }
1291                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1292                                                  newfe_bh,
1293                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1294                 if (status < 0) {
1295                         mlog_errno(status);
1296                         goto bail;
1297                 }
1298
1299                 if (S_ISDIR(new_inode->i_mode) ||
1300                     (ocfs2_read_links_count(newfe) == 1)) {
1301                         status = ocfs2_orphan_add(osb, handle, new_inode,
1302                                                   newfe_bh, orphan_name,
1303                                                   &orphan_insert, orphan_dir);
1304                         if (status < 0) {
1305                                 mlog_errno(status);
1306                                 goto bail;
1307                         }
1308                 }
1309
1310                 /* change the dirent to point to the correct inode */
1311                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1312                                             old_inode);
1313                 if (status < 0) {
1314                         mlog_errno(status);
1315                         goto bail;
1316                 }
1317                 new_dir->i_version++;
1318
1319                 if (S_ISDIR(new_inode->i_mode))
1320                         ocfs2_set_links_count(newfe, 0);
1321                 else
1322                         ocfs2_add_links_count(newfe, -1);
1323                 ocfs2_journal_dirty(handle, newfe_bh);
1324         } else {
1325                 /* if the name was not found in new_dir, add it now */
1326                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1327                                          OCFS2_I(old_inode)->ip_blkno,
1328                                          new_dir_bh, &target_insert);
1329         }
1330
1331         old_inode->i_ctime = CURRENT_TIME;
1332         mark_inode_dirty(old_inode);
1333
1334         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1335                                          old_inode_bh,
1336                                          OCFS2_JOURNAL_ACCESS_WRITE);
1337         if (status >= 0) {
1338                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1339
1340                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1341                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1342                 ocfs2_journal_dirty(handle, old_inode_bh);
1343         } else
1344                 mlog_errno(status);
1345
1346         /*
1347          * Now that the name has been added to new_dir, remove the old name.
1348          *
1349          * We don't keep any directory entry context around until now
1350          * because the insert might have changed the type of directory
1351          * we're dealing with.
1352          */
1353         status = ocfs2_find_entry(old_dentry->d_name.name,
1354                                   old_dentry->d_name.len, old_dir,
1355                                   &old_entry_lookup);
1356         if (status)
1357                 goto bail;
1358
1359         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1360         if (status < 0) {
1361                 mlog_errno(status);
1362                 goto bail;
1363         }
1364
1365         if (new_inode) {
1366                 new_inode->i_nlink--;
1367                 new_inode->i_ctime = CURRENT_TIME;
1368         }
1369         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1370
1371         if (update_dot_dot) {
1372                 status = ocfs2_update_entry(old_inode, handle,
1373                                             &old_inode_dot_dot_res, new_dir);
1374                 old_dir->i_nlink--;
1375                 if (new_inode) {
1376                         new_inode->i_nlink--;
1377                 } else {
1378                         inc_nlink(new_dir);
1379                         mark_inode_dirty(new_dir);
1380                 }
1381         }
1382         mark_inode_dirty(old_dir);
1383         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1384         if (new_inode) {
1385                 mark_inode_dirty(new_inode);
1386                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1387         }
1388
1389         if (old_dir != new_dir) {
1390                 /* Keep the same times on both directories.*/
1391                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1392
1393                 /*
1394                  * This will also pick up the i_nlink change from the
1395                  * block above.
1396                  */
1397                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1398         }
1399
1400         if (old_dir_nlink != old_dir->i_nlink) {
1401                 if (!old_dir_bh) {
1402                         mlog(ML_ERROR, "need to change nlink for old dir "
1403                              "%llu from %d to %d but bh is NULL!\n",
1404                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1405                              (int)old_dir_nlink, old_dir->i_nlink);
1406                 } else {
1407                         struct ocfs2_dinode *fe;
1408                         status = ocfs2_journal_access_di(handle,
1409                                                          INODE_CACHE(old_dir),
1410                                                          old_dir_bh,
1411                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1412                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1413                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1414                         ocfs2_journal_dirty(handle, old_dir_bh);
1415                 }
1416         }
1417         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1418         status = 0;
1419 bail:
1420         if (rename_lock)
1421                 ocfs2_rename_unlock(osb);
1422
1423         if (handle)
1424                 ocfs2_commit_trans(osb, handle);
1425
1426         if (parents_locked)
1427                 ocfs2_double_unlock(old_dir, new_dir);
1428
1429         if (old_child_locked)
1430                 ocfs2_inode_unlock(old_inode, 1);
1431
1432         if (new_child_locked)
1433                 ocfs2_inode_unlock(new_inode, 1);
1434
1435         if (orphan_dir) {
1436                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1437                 ocfs2_inode_unlock(orphan_dir, 1);
1438                 mutex_unlock(&orphan_dir->i_mutex);
1439                 iput(orphan_dir);
1440         }
1441
1442         if (new_inode)
1443                 sync_mapping_buffers(old_inode->i_mapping);
1444
1445         if (new_inode)
1446                 iput(new_inode);
1447
1448         ocfs2_free_dir_lookup_result(&target_lookup_res);
1449         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1450         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1451         ocfs2_free_dir_lookup_result(&orphan_insert);
1452         ocfs2_free_dir_lookup_result(&target_insert);
1453
1454         brelse(newfe_bh);
1455         brelse(old_inode_bh);
1456         brelse(old_dir_bh);
1457         brelse(new_dir_bh);
1458
1459         mlog_exit(status);
1460
1461         return status;
1462 }
1463
1464 /*
1465  * we expect i_size = strlen(symname). Copy symname into the file
1466  * data, including the null terminator.
1467  */
1468 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1469                                      handle_t *handle,
1470                                      struct inode *inode,
1471                                      const char *symname)
1472 {
1473         struct buffer_head **bhs = NULL;
1474         const char *c;
1475         struct super_block *sb = osb->sb;
1476         u64 p_blkno, p_blocks;
1477         int virtual, blocks, status, i, bytes_left;
1478
1479         bytes_left = i_size_read(inode) + 1;
1480         /* we can't trust i_blocks because we're actually going to
1481          * write i_size + 1 bytes. */
1482         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1483
1484         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1485                         (unsigned long long)inode->i_blocks,
1486                         i_size_read(inode), blocks);
1487
1488         /* Sanity check -- make sure we're going to fit. */
1489         if (bytes_left >
1490             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1491                 status = -EIO;
1492                 mlog_errno(status);
1493                 goto bail;
1494         }
1495
1496         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1497         if (!bhs) {
1498                 status = -ENOMEM;
1499                 mlog_errno(status);
1500                 goto bail;
1501         }
1502
1503         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1504                                              NULL);
1505         if (status < 0) {
1506                 mlog_errno(status);
1507                 goto bail;
1508         }
1509
1510         /* links can never be larger than one cluster so we know this
1511          * is all going to be contiguous, but do a sanity check
1512          * anyway. */
1513         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1514                 status = -EIO;
1515                 mlog_errno(status);
1516                 goto bail;
1517         }
1518
1519         virtual = 0;
1520         while(bytes_left > 0) {
1521                 c = &symname[virtual * sb->s_blocksize];
1522
1523                 bhs[virtual] = sb_getblk(sb, p_blkno);
1524                 if (!bhs[virtual]) {
1525                         status = -ENOMEM;
1526                         mlog_errno(status);
1527                         goto bail;
1528                 }
1529                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1530                                               bhs[virtual]);
1531
1532                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1533                                               bhs[virtual],
1534                                               OCFS2_JOURNAL_ACCESS_CREATE);
1535                 if (status < 0) {
1536                         mlog_errno(status);
1537                         goto bail;
1538                 }
1539
1540                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1541
1542                 memcpy(bhs[virtual]->b_data, c,
1543                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1544                        bytes_left);
1545
1546                 ocfs2_journal_dirty(handle, bhs[virtual]);
1547
1548                 virtual++;
1549                 p_blkno++;
1550                 bytes_left -= sb->s_blocksize;
1551         }
1552
1553         status = 0;
1554 bail:
1555
1556         if (bhs) {
1557                 for(i = 0; i < blocks; i++)
1558                         brelse(bhs[i]);
1559                 kfree(bhs);
1560         }
1561
1562         mlog_exit(status);
1563         return status;
1564 }
1565
1566 static int ocfs2_symlink(struct inode *dir,
1567                          struct dentry *dentry,
1568                          const char *symname)
1569 {
1570         int status, l, credits;
1571         u64 newsize;
1572         struct ocfs2_super *osb = NULL;
1573         struct inode *inode = NULL;
1574         struct super_block *sb;
1575         struct buffer_head *new_fe_bh = NULL;
1576         struct buffer_head *parent_fe_bh = NULL;
1577         struct ocfs2_dinode *fe = NULL;
1578         struct ocfs2_dinode *dirfe;
1579         handle_t *handle = NULL;
1580         struct ocfs2_alloc_context *inode_ac = NULL;
1581         struct ocfs2_alloc_context *data_ac = NULL;
1582         struct ocfs2_alloc_context *xattr_ac = NULL;
1583         int want_clusters = 0;
1584         int xattr_credits = 0;
1585         struct ocfs2_security_xattr_info si = {
1586                 .enable = 1,
1587         };
1588         int did_quota = 0, did_quota_inode = 0;
1589         struct ocfs2_dir_lookup_result lookup = { NULL, };
1590         sigset_t oldset;
1591         int did_block_signals = 0;
1592
1593         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1594                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1595
1596         dquot_initialize(dir);
1597
1598         sb = dir->i_sb;
1599         osb = OCFS2_SB(sb);
1600
1601         l = strlen(symname) + 1;
1602
1603         credits = ocfs2_calc_symlink_credits(sb);
1604
1605         /* lock the parent directory */
1606         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1607         if (status < 0) {
1608                 if (status != -ENOENT)
1609                         mlog_errno(status);
1610                 return status;
1611         }
1612
1613         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1614         if (!ocfs2_read_links_count(dirfe)) {
1615                 /* can't make a file in a deleted directory. */
1616                 status = -ENOENT;
1617                 goto bail;
1618         }
1619
1620         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1621                                            dentry->d_name.len);
1622         if (status)
1623                 goto bail;
1624
1625         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1626                                               dentry->d_name.name,
1627                                               dentry->d_name.len, &lookup);
1628         if (status < 0) {
1629                 mlog_errno(status);
1630                 goto bail;
1631         }
1632
1633         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1634         if (status < 0) {
1635                 if (status != -ENOSPC)
1636                         mlog_errno(status);
1637                 goto bail;
1638         }
1639
1640         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1641         if (!inode) {
1642                 status = -ENOMEM;
1643                 mlog_errno(status);
1644                 goto bail;
1645         }
1646
1647         /* get security xattr */
1648         status = ocfs2_init_security_get(inode, dir, &si);
1649         if (status) {
1650                 if (status == -EOPNOTSUPP)
1651                         si.enable = 0;
1652                 else {
1653                         mlog_errno(status);
1654                         goto bail;
1655                 }
1656         }
1657
1658         /* calculate meta data/clusters for setting security xattr */
1659         if (si.enable) {
1660                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1661                                                   &xattr_credits, &xattr_ac);
1662                 if (status < 0) {
1663                         mlog_errno(status);
1664                         goto bail;
1665                 }
1666         }
1667
1668         /* don't reserve bitmap space for fast symlinks. */
1669         if (l > ocfs2_fast_symlink_chars(sb))
1670                 want_clusters += 1;
1671
1672         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1673         if (status < 0) {
1674                 if (status != -ENOSPC)
1675                         mlog_errno(status);
1676                 goto bail;
1677         }
1678
1679         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1680         if (IS_ERR(handle)) {
1681                 status = PTR_ERR(handle);
1682                 handle = NULL;
1683                 mlog_errno(status);
1684                 goto bail;
1685         }
1686
1687         /* Starting to change things, restart is no longer possible. */
1688         ocfs2_block_signals(&oldset);
1689         did_block_signals = 1;
1690
1691         status = dquot_alloc_inode(inode);
1692         if (status)
1693                 goto bail;
1694         did_quota_inode = 1;
1695
1696         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1697                    inode->i_mode, dentry->d_name.len,
1698                    dentry->d_name.name);
1699
1700         status = ocfs2_mknod_locked(osb, dir, inode,
1701                                     0, &new_fe_bh, parent_fe_bh, handle,
1702                                     inode_ac);
1703         if (status < 0) {
1704                 mlog_errno(status);
1705                 goto bail;
1706         }
1707
1708         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1709         inode->i_rdev = 0;
1710         newsize = l - 1;
1711         if (l > ocfs2_fast_symlink_chars(sb)) {
1712                 u32 offset = 0;
1713
1714                 inode->i_op = &ocfs2_symlink_inode_operations;
1715                 status = dquot_alloc_space_nodirty(inode,
1716                     ocfs2_clusters_to_bytes(osb->sb, 1));
1717                 if (status)
1718                         goto bail;
1719                 did_quota = 1;
1720                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1721                                               new_fe_bh,
1722                                               handle, data_ac, NULL,
1723                                               NULL);
1724                 if (status < 0) {
1725                         if (status != -ENOSPC && status != -EINTR) {
1726                                 mlog(ML_ERROR,
1727                                      "Failed to extend file to %llu\n",
1728                                      (unsigned long long)newsize);
1729                                 mlog_errno(status);
1730                                 status = -ENOSPC;
1731                         }
1732                         goto bail;
1733                 }
1734                 i_size_write(inode, newsize);
1735                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1736         } else {
1737                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1738                 memcpy((char *) fe->id2.i_symlink, symname, l);
1739                 i_size_write(inode, newsize);
1740                 inode->i_blocks = 0;
1741         }
1742
1743         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1744         if (status < 0) {
1745                 mlog_errno(status);
1746                 goto bail;
1747         }
1748
1749         if (!ocfs2_inode_is_fast_symlink(inode)) {
1750                 status = ocfs2_create_symlink_data(osb, handle, inode,
1751                                                    symname);
1752                 if (status < 0) {
1753                         mlog_errno(status);
1754                         goto bail;
1755                 }
1756         }
1757
1758         if (si.enable) {
1759                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1760                                                  xattr_ac, data_ac);
1761                 if (status < 0) {
1762                         mlog_errno(status);
1763                         goto bail;
1764                 }
1765         }
1766
1767         /*
1768          * Do this before adding the entry to the directory. We add
1769          * also set d_op after success so that ->d_iput() will cleanup
1770          * the dentry lock even if ocfs2_add_entry() fails below.
1771          */
1772         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1773         if (status) {
1774                 mlog_errno(status);
1775                 goto bail;
1776         }
1777         dentry->d_op = &ocfs2_dentry_ops;
1778
1779         status = ocfs2_add_entry(handle, dentry, inode,
1780                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1781                                  &lookup);
1782         if (status < 0) {
1783                 mlog_errno(status);
1784                 goto bail;
1785         }
1786
1787         insert_inode_hash(inode);
1788         d_instantiate(dentry, inode);
1789 bail:
1790         if (status < 0 && did_quota)
1791                 dquot_free_space_nodirty(inode,
1792                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1793         if (status < 0 && did_quota_inode)
1794                 dquot_free_inode(inode);
1795         if (handle)
1796                 ocfs2_commit_trans(osb, handle);
1797
1798         ocfs2_inode_unlock(dir, 1);
1799         if (did_block_signals)
1800                 ocfs2_unblock_signals(&oldset);
1801
1802         brelse(new_fe_bh);
1803         brelse(parent_fe_bh);
1804         kfree(si.name);
1805         kfree(si.value);
1806         ocfs2_free_dir_lookup_result(&lookup);
1807         if (inode_ac)
1808                 ocfs2_free_alloc_context(inode_ac);
1809         if (data_ac)
1810                 ocfs2_free_alloc_context(data_ac);
1811         if (xattr_ac)
1812                 ocfs2_free_alloc_context(xattr_ac);
1813         if ((status < 0) && inode) {
1814                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1815                 clear_nlink(inode);
1816                 iput(inode);
1817         }
1818
1819         mlog_exit(status);
1820
1821         return status;
1822 }
1823
1824 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1825 {
1826         int status, namelen;
1827
1828         mlog_entry_void();
1829
1830         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1831                            (long long)blkno);
1832         if (namelen <= 0) {
1833                 if (namelen)
1834                         status = namelen;
1835                 else
1836                         status = -EINVAL;
1837                 mlog_errno(status);
1838                 goto bail;
1839         }
1840         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1841                 status = -EINVAL;
1842                 mlog_errno(status);
1843                 goto bail;
1844         }
1845
1846         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1847              namelen);
1848
1849         status = 0;
1850 bail:
1851         mlog_exit(status);
1852         return status;
1853 }
1854
1855 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1856                                     struct inode **ret_orphan_dir,
1857                                     u64 blkno,
1858                                     char *name,
1859                                     struct ocfs2_dir_lookup_result *lookup)
1860 {
1861         struct inode *orphan_dir_inode;
1862         struct buffer_head *orphan_dir_bh = NULL;
1863         int status = 0;
1864
1865         status = ocfs2_blkno_stringify(blkno, name);
1866         if (status < 0) {
1867                 mlog_errno(status);
1868                 return status;
1869         }
1870
1871         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1872                                                        ORPHAN_DIR_SYSTEM_INODE,
1873                                                        osb->slot_num);
1874         if (!orphan_dir_inode) {
1875                 status = -ENOENT;
1876                 mlog_errno(status);
1877                 return status;
1878         }
1879
1880         mutex_lock(&orphan_dir_inode->i_mutex);
1881
1882         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1883         if (status < 0) {
1884                 mlog_errno(status);
1885                 goto leave;
1886         }
1887
1888         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1889                                               orphan_dir_bh, name,
1890                                               OCFS2_ORPHAN_NAMELEN, lookup);
1891         if (status < 0) {
1892                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1893
1894                 mlog_errno(status);
1895                 goto leave;
1896         }
1897
1898         *ret_orphan_dir = orphan_dir_inode;
1899
1900 leave:
1901         if (status) {
1902                 mutex_unlock(&orphan_dir_inode->i_mutex);
1903                 iput(orphan_dir_inode);
1904         }
1905
1906         brelse(orphan_dir_bh);
1907
1908         mlog_exit(status);
1909         return status;
1910 }
1911
1912 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1913                             handle_t *handle,
1914                             struct inode *inode,
1915                             struct buffer_head *fe_bh,
1916                             char *name,
1917                             struct ocfs2_dir_lookup_result *lookup,
1918                             struct inode *orphan_dir_inode)
1919 {
1920         struct buffer_head *orphan_dir_bh = NULL;
1921         int status = 0;
1922         struct ocfs2_dinode *orphan_fe;
1923         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1924
1925         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1926
1927         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1928         if (status < 0) {
1929                 mlog_errno(status);
1930                 goto leave;
1931         }
1932
1933         status = ocfs2_journal_access_di(handle,
1934                                          INODE_CACHE(orphan_dir_inode),
1935                                          orphan_dir_bh,
1936                                          OCFS2_JOURNAL_ACCESS_WRITE);
1937         if (status < 0) {
1938                 mlog_errno(status);
1939                 goto leave;
1940         }
1941
1942         /* we're a cluster, and nlink can change on disk from
1943          * underneath us... */
1944         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1945         if (S_ISDIR(inode->i_mode))
1946                 ocfs2_add_links_count(orphan_fe, 1);
1947         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1948         ocfs2_journal_dirty(handle, orphan_dir_bh);
1949
1950         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1951                                    OCFS2_ORPHAN_NAMELEN, inode,
1952                                    OCFS2_I(inode)->ip_blkno,
1953                                    orphan_dir_bh, lookup);
1954         if (status < 0) {
1955                 mlog_errno(status);
1956                 goto leave;
1957         }
1958
1959         /*
1960          * We're going to journal the change of i_flags and i_orphaned_slot.
1961          * It's safe anyway, though some callers may duplicate the journaling.
1962          * Journaling within the func just make the logic look more
1963          * straightforward.
1964          */
1965         status = ocfs2_journal_access_di(handle,
1966                                          INODE_CACHE(inode),
1967                                          fe_bh,
1968                                          OCFS2_JOURNAL_ACCESS_WRITE);
1969         if (status < 0) {
1970                 mlog_errno(status);
1971                 goto leave;
1972         }
1973
1974         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1975         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
1976
1977         /* Record which orphan dir our inode now resides
1978          * in. delete_inode will use this to determine which orphan
1979          * dir to lock. */
1980         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1981
1982         ocfs2_journal_dirty(handle, fe_bh);
1983
1984         mlog(0, "Inode %llu orphaned in slot %d\n",
1985              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1986
1987 leave:
1988         brelse(orphan_dir_bh);
1989
1990         mlog_exit(status);
1991         return status;
1992 }
1993
1994 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1995 int ocfs2_orphan_del(struct ocfs2_super *osb,
1996                      handle_t *handle,
1997                      struct inode *orphan_dir_inode,
1998                      struct inode *inode,
1999                      struct buffer_head *orphan_dir_bh)
2000 {
2001         char name[OCFS2_ORPHAN_NAMELEN + 1];
2002         struct ocfs2_dinode *orphan_fe;
2003         int status = 0;
2004         struct ocfs2_dir_lookup_result lookup = { NULL, };
2005
2006         mlog_entry_void();
2007
2008         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2009         if (status < 0) {
2010                 mlog_errno(status);
2011                 goto leave;
2012         }
2013
2014         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2015              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2016              OCFS2_ORPHAN_NAMELEN);
2017
2018         /* find it's spot in the orphan directory */
2019         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2020                                   &lookup);
2021         if (status) {
2022                 mlog_errno(status);
2023                 goto leave;
2024         }
2025
2026         /* remove it from the orphan directory */
2027         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2028         if (status < 0) {
2029                 mlog_errno(status);
2030                 goto leave;
2031         }
2032
2033         status = ocfs2_journal_access_di(handle,
2034                                          INODE_CACHE(orphan_dir_inode),
2035                                          orphan_dir_bh,
2036                                          OCFS2_JOURNAL_ACCESS_WRITE);
2037         if (status < 0) {
2038                 mlog_errno(status);
2039                 goto leave;
2040         }
2041
2042         /* do the i_nlink dance! :) */
2043         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2044         if (S_ISDIR(inode->i_mode))
2045                 ocfs2_add_links_count(orphan_fe, -1);
2046         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2047         ocfs2_journal_dirty(handle, orphan_dir_bh);
2048
2049 leave:
2050         ocfs2_free_dir_lookup_result(&lookup);
2051
2052         mlog_exit(status);
2053         return status;
2054 }
2055
2056 int ocfs2_create_inode_in_orphan(struct inode *dir,
2057                                  int mode,
2058                                  struct inode **new_inode)
2059 {
2060         int status, did_quota_inode = 0;
2061         struct inode *inode = NULL;
2062         struct inode *orphan_dir = NULL;
2063         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2064         struct ocfs2_dinode *di = NULL;
2065         handle_t *handle = NULL;
2066         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2067         struct buffer_head *parent_di_bh = NULL;
2068         struct buffer_head *new_di_bh = NULL;
2069         struct ocfs2_alloc_context *inode_ac = NULL;
2070         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2071
2072         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2073         if (status < 0) {
2074                 if (status != -ENOENT)
2075                         mlog_errno(status);
2076                 return status;
2077         }
2078
2079         /*
2080          * We give the orphan dir the root blkno to fake an orphan name,
2081          * and allocate enough space for our insertion.
2082          */
2083         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
2084                                           osb->root_blkno,
2085                                           orphan_name, &orphan_insert);
2086         if (status < 0) {
2087                 mlog_errno(status);
2088                 goto leave;
2089         }
2090
2091         /* reserve an inode spot */
2092         status = ocfs2_reserve_new_inode(osb, &inode_ac);
2093         if (status < 0) {
2094                 if (status != -ENOSPC)
2095                         mlog_errno(status);
2096                 goto leave;
2097         }
2098
2099         inode = ocfs2_get_init_inode(dir, mode);
2100         if (!inode) {
2101                 status = -ENOMEM;
2102                 mlog_errno(status);
2103                 goto leave;
2104         }
2105
2106         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2107         if (IS_ERR(handle)) {
2108                 status = PTR_ERR(handle);
2109                 handle = NULL;
2110                 mlog_errno(status);
2111                 goto leave;
2112         }
2113
2114         status = dquot_alloc_inode(inode);
2115         if (status)
2116                 goto leave;
2117         did_quota_inode = 1;
2118
2119         inode->i_nlink = 0;
2120         /* do the real work now. */
2121         status = ocfs2_mknod_locked(osb, dir, inode,
2122                                     0, &new_di_bh, parent_di_bh, handle,
2123                                     inode_ac);
2124         if (status < 0) {
2125                 mlog_errno(status);
2126                 goto leave;
2127         }
2128
2129         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
2130         if (status < 0) {
2131                 mlog_errno(status);
2132                 goto leave;
2133         }
2134
2135         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2136         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2137                                   &orphan_insert, orphan_dir);
2138         if (status < 0) {
2139                 mlog_errno(status);
2140                 goto leave;
2141         }
2142
2143         /* get open lock so that only nodes can't remove it from orphan dir. */
2144         status = ocfs2_open_lock(inode);
2145         if (status < 0)
2146                 mlog_errno(status);
2147
2148         insert_inode_hash(inode);
2149 leave:
2150         if (status < 0 && did_quota_inode)
2151                 dquot_free_inode(inode);
2152         if (handle)
2153                 ocfs2_commit_trans(osb, handle);
2154
2155         if (orphan_dir) {
2156                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2157                 ocfs2_inode_unlock(orphan_dir, 1);
2158                 mutex_unlock(&orphan_dir->i_mutex);
2159                 iput(orphan_dir);
2160         }
2161
2162         if (status == -ENOSPC)
2163                 mlog(0, "Disk is full\n");
2164
2165         if ((status < 0) && inode) {
2166                 clear_nlink(inode);
2167                 iput(inode);
2168         }
2169
2170         if (inode_ac)
2171                 ocfs2_free_alloc_context(inode_ac);
2172
2173         brelse(new_di_bh);
2174
2175         if (!status)
2176                 *new_inode = inode;
2177
2178         ocfs2_free_dir_lookup_result(&orphan_insert);
2179
2180         ocfs2_inode_unlock(dir, 1);
2181         brelse(parent_di_bh);
2182         return status;
2183 }
2184
2185 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2186                                    struct inode *inode,
2187                                    struct dentry *dentry)
2188 {
2189         int status = 0;
2190         struct buffer_head *parent_di_bh = NULL;
2191         handle_t *handle = NULL;
2192         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2193         struct ocfs2_dinode *dir_di, *di;
2194         struct inode *orphan_dir_inode = NULL;
2195         struct buffer_head *orphan_dir_bh = NULL;
2196         struct buffer_head *di_bh = NULL;
2197         struct ocfs2_dir_lookup_result lookup = { NULL, };
2198
2199         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2200                    dentry->d_name.len, dentry->d_name.name);
2201
2202         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2203         if (status < 0) {
2204                 if (status != -ENOENT)
2205                         mlog_errno(status);
2206                 return status;
2207         }
2208
2209         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2210         if (!dir_di->i_links_count) {
2211                 /* can't make a file in a deleted directory. */
2212                 status = -ENOENT;
2213                 goto leave;
2214         }
2215
2216         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2217                                            dentry->d_name.len);
2218         if (status)
2219                 goto leave;
2220
2221         /* get a spot inside the dir. */
2222         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2223                                               dentry->d_name.name,
2224                                               dentry->d_name.len, &lookup);
2225         if (status < 0) {
2226                 mlog_errno(status);
2227                 goto leave;
2228         }
2229
2230         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2231                                                        ORPHAN_DIR_SYSTEM_INODE,
2232                                                        osb->slot_num);
2233         if (!orphan_dir_inode) {
2234                 status = -EEXIST;
2235                 mlog_errno(status);
2236                 goto leave;
2237         }
2238
2239         mutex_lock(&orphan_dir_inode->i_mutex);
2240
2241         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2242         if (status < 0) {
2243                 mlog_errno(status);
2244                 mutex_unlock(&orphan_dir_inode->i_mutex);
2245                 iput(orphan_dir_inode);
2246                 goto leave;
2247         }
2248
2249         status = ocfs2_read_inode_block(inode, &di_bh);
2250         if (status < 0) {
2251                 mlog_errno(status);
2252                 goto orphan_unlock;
2253         }
2254
2255         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2256         if (IS_ERR(handle)) {
2257                 status = PTR_ERR(handle);
2258                 handle = NULL;
2259                 mlog_errno(status);
2260                 goto orphan_unlock;
2261         }
2262
2263         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2264                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2265         if (status < 0) {
2266                 mlog_errno(status);
2267                 goto out_commit;
2268         }
2269
2270         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2271                                   orphan_dir_bh);
2272         if (status < 0) {
2273                 mlog_errno(status);
2274                 goto out_commit;
2275         }
2276
2277         di = (struct ocfs2_dinode *)di_bh->b_data;
2278         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2279         di->i_orphaned_slot = 0;
2280         inode->i_nlink = 1;
2281         ocfs2_set_links_count(di, inode->i_nlink);
2282         ocfs2_journal_dirty(handle, di_bh);
2283
2284         status = ocfs2_add_entry(handle, dentry, inode,
2285                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2286                                  &lookup);
2287         if (status < 0) {
2288                 mlog_errno(status);
2289                 goto out_commit;
2290         }
2291
2292         status = ocfs2_dentry_attach_lock(dentry, inode,
2293                                           OCFS2_I(dir)->ip_blkno);
2294         if (status) {
2295                 mlog_errno(status);
2296                 goto out_commit;
2297         }
2298
2299         dentry->d_op = &ocfs2_dentry_ops;
2300         d_instantiate(dentry, inode);
2301         status = 0;
2302 out_commit:
2303         ocfs2_commit_trans(osb, handle);
2304 orphan_unlock:
2305         ocfs2_inode_unlock(orphan_dir_inode, 1);
2306         mutex_unlock(&orphan_dir_inode->i_mutex);
2307         iput(orphan_dir_inode);
2308 leave:
2309
2310         ocfs2_inode_unlock(dir, 1);
2311
2312         brelse(di_bh);
2313         brelse(parent_di_bh);
2314         brelse(orphan_dir_bh);
2315
2316         ocfs2_free_dir_lookup_result(&lookup);
2317
2318         mlog_exit(status);
2319
2320         return status;
2321 }
2322
2323 const struct inode_operations ocfs2_dir_iops = {
2324         .create         = ocfs2_create,
2325         .lookup         = ocfs2_lookup,
2326         .link           = ocfs2_link,
2327         .unlink         = ocfs2_unlink,
2328         .rmdir          = ocfs2_unlink,
2329         .symlink        = ocfs2_symlink,
2330         .mkdir          = ocfs2_mkdir,
2331         .mknod          = ocfs2_mknod,
2332         .rename         = ocfs2_rename,
2333         .setattr        = ocfs2_setattr,
2334         .getattr        = ocfs2_getattr,
2335         .permission     = ocfs2_permission,
2336         .setxattr       = generic_setxattr,
2337         .getxattr       = generic_getxattr,
2338         .listxattr      = ocfs2_listxattr,
2339         .removexattr    = generic_removexattr,
2340         .fiemap         = ocfs2_fiemap,
2341 };