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
  44 #define MLOG_MASK_PREFIX ML_NAMEI
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "xattr.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  68                               struct inode *dir,
  69                               struct dentry *dentry, int mode,
  70                               dev_t dev,
  71                               struct buffer_head **new_fe_bh,
  72                               struct buffer_head *parent_fe_bh,
  73                               handle_t *handle,
  74                               struct inode **ret_inode,
  75                               struct ocfs2_alloc_context *inode_ac);
  76
  77 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  78                                     struct inode **ret_orphan_dir,
  79                                     struct inode *inode,
  80                                     char *name,
  81                                     struct buffer_head **de_bh);
  82
  83 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  84                             handle_t *handle,
  85                             struct inode *inode,
  86                             struct ocfs2_dinode *fe,
  87                             char *name,
  88                             struct buffer_head *de_bh,
  89                             struct inode *orphan_dir_inode);
  90
  91 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  92                                      handle_t *handle,
  93                                      struct inode *inode,
  94                                      const char *symname);
  95
  96 /* An orphan dir name is an 8 byte value, printed as a hex string */
  97 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  98
  99 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 100                                    struct nameidata *nd)
 101 {
 102         int status;
 103         u64 blkno;
 104         struct inode *inode = NULL;
 105         struct dentry *ret;
 106         struct ocfs2_inode_info *oi;
 107
 108         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 109                    dentry->d_name.len, dentry->d_name.name);
 110
 111         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 112                 ret = ERR_PTR(-ENAMETOOLONG);
 113                 goto bail;
 114         }
 115
 116         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 117              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 118
 119         status = ocfs2_inode_lock(dir, NULL, 0);
 120         if (status < 0) {
 121                 if (status != -ENOENT)
 122                         mlog_errno(status);
 123                 ret = ERR_PTR(status);
 124                 goto bail;
 125         }
 126
 127         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 128                                             dentry->d_name.len, &blkno);
 129         if (status < 0)
 130                 goto bail_add;
 131
 132         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 133         if (IS_ERR(inode)) {
 134                 ret = ERR_PTR(-EACCES);
 135                 goto bail_unlock;
 136         }
 137
 138         oi = OCFS2_I(inode);
 139         /* Clear any orphaned state... If we were able to look up the
 140          * inode from a directory, it certainly can't be orphaned. We
 141          * might have the bad state from a node which intended to
 142          * orphan this inode but crashed before it could commit the
 143          * unlink. */
 144         spin_lock(&oi->ip_lock);
 145         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 146         spin_unlock(&oi->ip_lock);
 147
 148 bail_add:
 149         dentry->d_op = &ocfs2_dentry_ops;
 150         ret = d_splice_alias(inode, dentry);
 151
 152         if (inode) {
 153                 /*
 154                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 155                  * dentry, it will d_move() it on top of ourse. The
 156                  * return value will indicate this however, so in
 157                  * those cases, we switch them around for the locking
 158                  * code.
 159                  *
 160                  * NOTE: This dentry already has ->d_op set from
 161                  * ocfs2_get_parent() and ocfs2_get_dentry()
 162                  */
 163                 if (ret)
 164                         dentry = ret;
 165
 166                 status = ocfs2_dentry_attach_lock(dentry, inode,
 167                                                   OCFS2_I(dir)->ip_blkno);
 168                 if (status) {
 169                         mlog_errno(status);
 170                         ret = ERR_PTR(status);
 171                         goto bail_unlock;
 172                 }
 173         }
 174
 175 bail_unlock:
 176         /* Don't drop the cluster lock until *after* the d_add --
 177          * unlink on another node will message us to remove that
 178          * dentry under this lock so otherwise we can race this with
 179          * the downconvert thread and have a stale dentry. */
 180         ocfs2_inode_unlock(dir, 0);
 181
 182 bail:
 183
 184         mlog_exit_ptr(ret);
 185
 186         return ret;
 187 }
 188
 189 static int ocfs2_mknod(struct inode *dir,
 190                        struct dentry *dentry,
 191                        int mode,
 192                        dev_t dev)
 193 {
 194         int status = 0;
 195         struct buffer_head *parent_fe_bh = NULL;
 196         handle_t *handle = NULL;
 197         struct ocfs2_super *osb;
 198         struct ocfs2_dinode *dirfe;
 199         struct buffer_head *new_fe_bh = NULL;
 200         struct buffer_head *de_bh = NULL;
 201         struct inode *inode = NULL;
 202         struct ocfs2_alloc_context *inode_ac = NULL;
 203         struct ocfs2_alloc_context *data_ac = NULL;
 204
 205         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 206                    (unsigned long)dev, dentry->d_name.len,
 207                    dentry->d_name.name);
 208
 209         /* get our super block */
 210         osb = OCFS2_SB(dir->i_sb);
 211
 212         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 213         if (status < 0) {
 214                 if (status != -ENOENT)
 215                         mlog_errno(status);
 216                 return status;
 217         }
 218
 219         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 220                 status = -EMLINK;
 221                 goto leave;
 222         }
 223
 224         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 225         if (!dirfe->i_links_count) {
 226                 /* can't make a file in a deleted directory. */
 227                 status = -ENOENT;
 228                 goto leave;
 229         }
 230
 231         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 232                                            dentry->d_name.len);
 233         if (status)
 234                 goto leave;
 235
 236         /* get a spot inside the dir. */
 237         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 238                                               dentry->d_name.name,
 239                                               dentry->d_name.len, &de_bh);
 240         if (status < 0) {
 241                 mlog_errno(status);
 242                 goto leave;
 243         }
 244
 245         /* reserve an inode spot */
 246         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 247         if (status < 0) {
 248                 if (status != -ENOSPC)
 249                         mlog_errno(status);
 250                 goto leave;
 251         }
 252
 253         /* Reserve a cluster if creating an extent based directory. */
 254         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 255                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 256                 if (status < 0) {
 257                         if (status != -ENOSPC)
 258                                 mlog_errno(status);
 259                         goto leave;
 260                 }
 261         }
 262
 263         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
 264         if (IS_ERR(handle)) {
 265                 status = PTR_ERR(handle);
 266                 handle = NULL;
 267                 mlog_errno(status);
 268                 goto leave;
 269         }
 270
 271         /* do the real work now. */
 272         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 273                                     &new_fe_bh, parent_fe_bh, handle,
 274                                     &inode, inode_ac);
 275         if (status < 0) {
 276                 mlog_errno(status);
 277                 goto leave;
 278         }
 279
 280         if (S_ISDIR(mode)) {
 281                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 282                                             new_fe_bh, data_ac);
 283                 if (status < 0) {
 284                         mlog_errno(status);
 285                         goto leave;
 286                 }
 287
 288                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 289                                               OCFS2_JOURNAL_ACCESS_WRITE);
 290                 if (status < 0) {
 291                         mlog_errno(status);
 292                         goto leave;
 293                 }
 294                 le16_add_cpu(&dirfe->i_links_count, 1);
 295                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 296                 if (status < 0) {
 297                         mlog_errno(status);
 298                         goto leave;
 299                 }
 300                 inc_nlink(dir);
 301         }
 302
 303         status = ocfs2_add_entry(handle, dentry, inode,
 304                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 305                                  de_bh);
 306         if (status < 0) {
 307                 mlog_errno(status);
 308                 goto leave;
 309         }
 310
 311         status = ocfs2_dentry_attach_lock(dentry, inode,
 312                                           OCFS2_I(dir)->ip_blkno);
 313         if (status) {
 314                 mlog_errno(status);
 315                 goto leave;
 316         }
 317
 318         insert_inode_hash(inode);
 319         dentry->d_op = &ocfs2_dentry_ops;
 320         d_instantiate(dentry, inode);
 321         status = 0;
 322 leave:
 323         if (handle)
 324                 ocfs2_commit_trans(osb, handle);
 325
 326         ocfs2_inode_unlock(dir, 1);
 327
 328         if (status == -ENOSPC)
 329                 mlog(0, "Disk is full\n");
 330
 331         brelse(new_fe_bh);
 332         brelse(de_bh);
 333         brelse(parent_fe_bh);
 334
 335         if ((status < 0) && inode)
 336                 iput(inode);
 337
 338         if (inode_ac)
 339                 ocfs2_free_alloc_context(inode_ac);
 340
 341         if (data_ac)
 342                 ocfs2_free_alloc_context(data_ac);
 343
 344         mlog_exit(status);
 345
 346         return status;
 347 }
 348
 349 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 350                               struct inode *dir,
 351                               struct dentry *dentry, int mode,
 352                               dev_t dev,
 353                               struct buffer_head **new_fe_bh,
 354                               struct buffer_head *parent_fe_bh,
 355                               handle_t *handle,
 356                               struct inode **ret_inode,
 357                               struct ocfs2_alloc_context *inode_ac)
 358 {
 359         int status = 0;
 360         struct ocfs2_dinode *fe = NULL;
 361         struct ocfs2_extent_list *fel;
 362         u64 fe_blkno = 0;
 363         u16 suballoc_bit;
 364         struct inode *inode = NULL;
 365
 366         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 367                    (unsigned long)dev, dentry->d_name.len,
 368                    dentry->d_name.name);
 369
 370         *new_fe_bh = NULL;
 371         *ret_inode = NULL;
 372
 373         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 374                                        &fe_blkno);
 375         if (status < 0) {
 376                 mlog_errno(status);
 377                 goto leave;
 378         }
 379
 380         inode = new_inode(dir->i_sb);
 381         if (IS_ERR(inode)) {
 382                 status = PTR_ERR(inode);
 383                 mlog(ML_ERROR, "new_inode failed!\n");
 384                 goto leave;
 385         }
 386
 387         /* populate as many fields early on as possible - many of
 388          * these are used by the support functions here and in
 389          * callers. */
 390         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 391         OCFS2_I(inode)->ip_blkno = fe_blkno;
 392         if (S_ISDIR(mode))
 393                 inode->i_nlink = 2;
 394         else
 395                 inode->i_nlink = 1;
 396         inode->i_mode = mode;
 397         spin_lock(&osb->osb_lock);
 398         inode->i_generation = osb->s_next_generation++;
 399         spin_unlock(&osb->osb_lock);
 400
 401         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 402         if (!*new_fe_bh) {
 403                 status = -EIO;
 404                 mlog_errno(status);
 405                 goto leave;
 406         }
 407         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 408
 409         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 410                                       OCFS2_JOURNAL_ACCESS_CREATE);
 411         if (status < 0) {
 412                 mlog_errno(status);
 413                 goto leave;
 414         }
 415
 416         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 417         memset(fe, 0, osb->sb->s_blocksize);
 418
 419         fe->i_generation = cpu_to_le32(inode->i_generation);
 420         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 421         fe->i_blkno = cpu_to_le64(fe_blkno);
 422         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 423         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 424         fe->i_uid = cpu_to_le32(current->fsuid);
 425         if (dir->i_mode & S_ISGID) {
 426                 fe->i_gid = cpu_to_le32(dir->i_gid);
 427                 if (S_ISDIR(mode))
 428                         mode |= S_ISGID;
 429         } else
 430                 fe->i_gid = cpu_to_le32(current->fsgid);
 431         fe->i_mode = cpu_to_le16(mode);
 432         if (S_ISCHR(mode) || S_ISBLK(mode))
 433                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 434
 435         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 436
 437         fe->i_last_eb_blk = 0;
 438         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 439         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 440         fe->i_atime = fe->i_ctime = fe->i_mtime =
 441                 cpu_to_le64(CURRENT_TIME.tv_sec);
 442         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 443                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 444         fe->i_dtime = 0;
 445
 446         /*
 447          * If supported, directories start with inline data.
 448          */
 449         if (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) {
 450                 u16 feat = le16_to_cpu(fe->i_dyn_features);
 451
 452                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 453
 454                 fe->id2.i_data.id_count = cpu_to_le16(ocfs2_max_inline_data(osb->sb));
 455         } else {
 456                 fel = &fe->id2.i_list;
 457                 fel->l_tree_depth = 0;
 458                 fel->l_next_free_rec = 0;
 459                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 460         }
 461
 462         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 463         if (status < 0) {
 464                 mlog_errno(status);
 465                 goto leave;
 466         }
 467
 468         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 469                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 470                      "i_blkno=%llu, i_ino=%lu\n",
 471                      (unsigned long long)(*new_fe_bh)->b_blocknr,
 472                      (unsigned long long)le64_to_cpu(fe->i_blkno),
 473                      inode->i_ino);
 474                 BUG();
 475         }
 476
 477         ocfs2_inode_set_new(osb, inode);
 478         if (!ocfs2_mount_local(osb)) {
 479                 status = ocfs2_create_new_inode_locks(inode);
 480                 if (status < 0)
 481                         mlog_errno(status);
 482         }
 483
 484         status = 0; /* error in ocfs2_create_new_inode_locks is not
 485                      * critical */
 486
 487         *ret_inode = inode;
 488 leave:
 489         if (status < 0) {
 490                 if (*new_fe_bh) {
 491                         brelse(*new_fe_bh);
 492                         *new_fe_bh = NULL;
 493                 }
 494                 if (inode)
 495                         iput(inode);
 496         }
 497
 498         mlog_exit(status);
 499         return status;
 500 }
 501
 502 static int ocfs2_mkdir(struct inode *dir,
 503                        struct dentry *dentry,
 504                        int mode)
 505 {
 506         int ret;
 507
 508         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 509                    dentry->d_name.len, dentry->d_name.name);
 510         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 511         mlog_exit(ret);
 512
 513         return ret;
 514 }
 515
 516 static int ocfs2_create(struct inode *dir,
 517                         struct dentry *dentry,
 518                         int mode,
 519                         struct nameidata *nd)
 520 {
 521         int ret;
 522
 523         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 524                    dentry->d_name.len, dentry->d_name.name);
 525         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 526         mlog_exit(ret);
 527
 528         return ret;
 529 }
 530
 531 static int ocfs2_link(struct dentry *old_dentry,
 532                       struct inode *dir,
 533                       struct dentry *dentry)
 534 {
 535         handle_t *handle;
 536         struct inode *inode = old_dentry->d_inode;
 537         int err;
 538         struct buffer_head *fe_bh = NULL;
 539         struct buffer_head *parent_fe_bh = NULL;
 540         struct buffer_head *de_bh = NULL;
 541         struct ocfs2_dinode *fe = NULL;
 542         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 543
 544         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 545                    old_dentry->d_name.len, old_dentry->d_name.name,
 546                    dentry->d_name.len, dentry->d_name.name);
 547
 548         if (S_ISDIR(inode->i_mode))
 549                 return -EPERM;
 550
 551         err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 552         if (err < 0) {
 553                 if (err != -ENOENT)
 554                         mlog_errno(err);
 555                 return err;
 556         }
 557
 558         if (!dir->i_nlink) {
 559                 err = -ENOENT;
 560                 goto out;
 561         }
 562
 563         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 564                                         dentry->d_name.len);
 565         if (err)
 566                 goto out;
 567
 568         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 569                                            dentry->d_name.name,
 570                                            dentry->d_name.len, &de_bh);
 571         if (err < 0) {
 572                 mlog_errno(err);
 573                 goto out;
 574         }
 575
 576         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 577         if (err < 0) {
 578                 if (err != -ENOENT)
 579                         mlog_errno(err);
 580                 goto out;
 581         }
 582
 583         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 584         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 585                 err = -EMLINK;
 586                 goto out_unlock_inode;
 587         }
 588
 589         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
 590         if (IS_ERR(handle)) {
 591                 err = PTR_ERR(handle);
 592                 handle = NULL;
 593                 mlog_errno(err);
 594                 goto out_unlock_inode;
 595         }
 596
 597         err = ocfs2_journal_access(handle, inode, fe_bh,
 598                                    OCFS2_JOURNAL_ACCESS_WRITE);
 599         if (err < 0) {
 600                 mlog_errno(err);
 601                 goto out_commit;
 602         }
 603
 604         inc_nlink(inode);
 605         inode->i_ctime = CURRENT_TIME;
 606         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 607         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 608         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 609
 610         err = ocfs2_journal_dirty(handle, fe_bh);
 611         if (err < 0) {
 612                 le16_add_cpu(&fe->i_links_count, -1);
 613                 drop_nlink(inode);
 614                 mlog_errno(err);
 615                 goto out_commit;
 616         }
 617
 618         err = ocfs2_add_entry(handle, dentry, inode,
 619                               OCFS2_I(inode)->ip_blkno,
 620                               parent_fe_bh, de_bh);
 621         if (err) {
 622                 le16_add_cpu(&fe->i_links_count, -1);
 623                 drop_nlink(inode);
 624                 mlog_errno(err);
 625                 goto out_commit;
 626         }
 627
 628         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 629         if (err) {
 630                 mlog_errno(err);
 631                 goto out_commit;
 632         }
 633
 634         atomic_inc(&inode->i_count);
 635         dentry->d_op = &ocfs2_dentry_ops;
 636         d_instantiate(dentry, inode);
 637
 638 out_commit:
 639         ocfs2_commit_trans(osb, handle);
 640 out_unlock_inode:
 641         ocfs2_inode_unlock(inode, 1);
 642
 643 out:
 644         ocfs2_inode_unlock(dir, 1);
 645
 646         brelse(de_bh);
 647         brelse(fe_bh);
 648         brelse(parent_fe_bh);
 649
 650         mlog_exit(err);
 651
 652         return err;
 653 }
 654
 655 /*
 656  * Takes and drops an exclusive lock on the given dentry. This will
 657  * force other nodes to drop it.
 658  */
 659 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 660 {
 661         int ret;
 662
 663         ret = ocfs2_dentry_lock(dentry, 1);
 664         if (ret)
 665                 mlog_errno(ret);
 666         else
 667                 ocfs2_dentry_unlock(dentry, 1);
 668
 669         return ret;
 670 }
 671
 672 static inline int inode_is_unlinkable(struct inode *inode)
 673 {
 674         if (S_ISDIR(inode->i_mode)) {
 675                 if (inode->i_nlink == 2)
 676                         return 1;
 677                 return 0;
 678         }
 679
 680         if (inode->i_nlink == 1)
 681                 return 1;
 682         return 0;
 683 }
 684
 685 static int ocfs2_unlink(struct inode *dir,
 686                         struct dentry *dentry)
 687 {
 688         int status;
 689         int child_locked = 0;
 690         struct inode *inode = dentry->d_inode;
 691         struct inode *orphan_dir = NULL;
 692         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 693         u64 blkno;
 694         struct ocfs2_dinode *fe = NULL;
 695         struct buffer_head *fe_bh = NULL;
 696         struct buffer_head *parent_node_bh = NULL;
 697         handle_t *handle = NULL;
 698         struct ocfs2_dir_entry *dirent = NULL;
 699         struct buffer_head *dirent_bh = NULL;
 700         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 701         struct buffer_head *orphan_entry_bh = NULL;
 702
 703         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 704                    dentry->d_name.len, dentry->d_name.name);
 705
 706         BUG_ON(dentry->d_parent->d_inode != dir);
 707
 708         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 709
 710         if (inode == osb->root_inode) {
 711                 mlog(0, "Cannot delete the root directory\n");
 712                 return -EPERM;
 713         }
 714
 715         status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
 716         if (status < 0) {
 717                 if (status != -ENOENT)
 718                         mlog_errno(status);
 719                 return status;
 720         }
 721
 722         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 723                                           dentry->d_name.len, &blkno,
 724                                           dir, &dirent_bh, &dirent);
 725         if (status < 0) {
 726                 if (status != -ENOENT)
 727                         mlog_errno(status);
 728                 goto leave;
 729         }
 730
 731         if (OCFS2_I(inode)->ip_blkno != blkno) {
 732                 status = -ENOENT;
 733
 734                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 735                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 736                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 737                 goto leave;
 738         }
 739
 740         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 741         if (status < 0) {
 742                 if (status != -ENOENT)
 743                         mlog_errno(status);
 744                 goto leave;
 745         }
 746         child_locked = 1;
 747
 748         if (S_ISDIR(inode->i_mode)) {
 749                 if (!ocfs2_empty_dir(inode)) {
 750                         status = -ENOTEMPTY;
 751                         goto leave;
 752                 } else if (inode->i_nlink != 2) {
 753                         status = -ENOTEMPTY;
 754                         goto leave;
 755                 }
 756         }
 757
 758         status = ocfs2_remote_dentry_delete(dentry);
 759         if (status < 0) {
 760                 /* This remote delete should succeed under all normal
 761                  * circumstances. */
 762                 mlog_errno(status);
 763                 goto leave;
 764         }
 765
 766         if (inode_is_unlinkable(inode)) {
 767                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 768                                                   orphan_name,
 769                                                   &orphan_entry_bh);
 770                 if (status < 0) {
 771                         mlog_errno(status);
 772                         goto leave;
 773                 }
 774         }
 775
 776         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
 777         if (IS_ERR(handle)) {
 778                 status = PTR_ERR(handle);
 779                 handle = NULL;
 780                 mlog_errno(status);
 781                 goto leave;
 782         }
 783
 784         status = ocfs2_journal_access(handle, inode, fe_bh,
 785                                       OCFS2_JOURNAL_ACCESS_WRITE);
 786         if (status < 0) {
 787                 mlog_errno(status);
 788                 goto leave;
 789         }
 790
 791         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 792
 793         if (inode_is_unlinkable(inode)) {
 794                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 795                                           orphan_entry_bh, orphan_dir);
 796                 if (status < 0) {
 797                         mlog_errno(status);
 798                         goto leave;
 799                 }
 800         }
 801
 802         /* delete the name from the parent dir */
 803         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 804         if (status < 0) {
 805                 mlog_errno(status);
 806                 goto leave;
 807         }
 808
 809         if (S_ISDIR(inode->i_mode))
 810                 drop_nlink(inode);
 811         drop_nlink(inode);
 812         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 813
 814         status = ocfs2_journal_dirty(handle, fe_bh);
 815         if (status < 0) {
 816                 mlog_errno(status);
 817                 goto leave;
 818         }
 819
 820         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 821         if (S_ISDIR(inode->i_mode))
 822                 drop_nlink(dir);
 823
 824         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 825         if (status < 0) {
 826                 mlog_errno(status);
 827                 if (S_ISDIR(inode->i_mode))
 828                         inc_nlink(dir);
 829         }
 830
 831 leave:
 832         if (handle)
 833                 ocfs2_commit_trans(osb, handle);
 834
 835         if (child_locked)
 836                 ocfs2_inode_unlock(inode, 1);
 837
 838         ocfs2_inode_unlock(dir, 1);
 839
 840         if (orphan_dir) {
 841                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 842                 ocfs2_inode_unlock(orphan_dir, 1);
 843                 mutex_unlock(&orphan_dir->i_mutex);
 844                 iput(orphan_dir);
 845         }
 846
 847         brelse(fe_bh);
 848         brelse(dirent_bh);
 849         brelse(parent_node_bh);
 850         brelse(orphan_entry_bh);
 851
 852         mlog_exit(status);
 853
 854         return status;
 855 }
 856
 857 /*
 858  * The only place this should be used is rename!
 859  * if they have the same id, then the 1st one is the only one locked.
 860  */
 861 static int ocfs2_double_lock(struct ocfs2_super *osb,
 862                              struct buffer_head **bh1,
 863                              struct inode *inode1,
 864                              struct buffer_head **bh2,
 865                              struct inode *inode2)
 866 {
 867         int status;
 868         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 869         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 870         struct buffer_head **tmpbh;
 871         struct inode *tmpinode;
 872
 873         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 874                    (unsigned long long)oi1->ip_blkno,
 875                    (unsigned long long)oi2->ip_blkno);
 876
 877         if (*bh1)
 878                 *bh1 = NULL;
 879         if (*bh2)
 880                 *bh2 = NULL;
 881
 882         /* we always want to lock the one with the lower lockid first. */
 883         if (oi1->ip_blkno != oi2->ip_blkno) {
 884                 if (oi1->ip_blkno < oi2->ip_blkno) {
 885                         /* switch id1 and id2 around */
 886                         mlog(0, "switching them around...\n");
 887                         tmpbh = bh2;
 888                         bh2 = bh1;
 889                         bh1 = tmpbh;
 890
 891                         tmpinode = inode2;
 892                         inode2 = inode1;
 893                         inode1 = tmpinode;
 894                 }
 895                 /* lock id2 */
 896                 status = ocfs2_inode_lock(inode2, bh2, 1);
 897                 if (status < 0) {
 898                         if (status != -ENOENT)
 899                                 mlog_errno(status);
 900                         goto bail;
 901                 }
 902         }
 903
 904         /* lock id1 */
 905         status = ocfs2_inode_lock(inode1, bh1, 1);
 906         if (status < 0) {
 907                 /*
 908                  * An error return must mean that no cluster locks
 909                  * were held on function exit.
 910                  */
 911                 if (oi1->ip_blkno != oi2->ip_blkno)
 912                         ocfs2_inode_unlock(inode2, 1);
 913
 914                 if (status != -ENOENT)
 915                         mlog_errno(status);
 916         }
 917
 918 bail:
 919         mlog_exit(status);
 920         return status;
 921 }
 922
 923 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
 924 {
 925         ocfs2_inode_unlock(inode1, 1);
 926
 927         if (inode1 != inode2)
 928                 ocfs2_inode_unlock(inode2, 1);
 929 }
 930
 931 static int ocfs2_rename(struct inode *old_dir,
 932                         struct dentry *old_dentry,
 933                         struct inode *new_dir,
 934                         struct dentry *new_dentry)
 935 {
 936         int status = 0, rename_lock = 0, parents_locked = 0;
 937         int old_child_locked = 0, new_child_locked = 0;
 938         struct inode *old_inode = old_dentry->d_inode;
 939         struct inode *new_inode = new_dentry->d_inode;
 940         struct inode *orphan_dir = NULL;
 941         struct ocfs2_dinode *newfe = NULL;
 942         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 943         struct buffer_head *orphan_entry_bh = NULL;
 944         struct buffer_head *newfe_bh = NULL;
 945         struct buffer_head *old_inode_bh = NULL;
 946         struct buffer_head *insert_entry_bh = NULL;
 947         struct ocfs2_super *osb = NULL;
 948         u64 newfe_blkno, old_de_ino;
 949         handle_t *handle = NULL;
 950         struct buffer_head *old_dir_bh = NULL;
 951         struct buffer_head *new_dir_bh = NULL;
 952         struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
 953                 *new_de = NULL;
 954         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
 955         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
 956                                                     // this is the 1st dirent bh
 957         nlink_t old_dir_nlink = old_dir->i_nlink;
 958         struct ocfs2_dinode *old_di;
 959
 960         /* At some point it might be nice to break this function up a
 961          * bit. */
 962
 963         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
 964                    old_dir, old_dentry, new_dir, new_dentry,
 965                    old_dentry->d_name.len, old_dentry->d_name.name,
 966                    new_dentry->d_name.len, new_dentry->d_name.name);
 967
 968         osb = OCFS2_SB(old_dir->i_sb);
 969
 970         if (new_inode) {
 971                 if (!igrab(new_inode))
 972                         BUG();
 973         }
 974
 975         /* Assume a directory hierarchy thusly:
 976          * a/b/c
 977          * a/d
 978          * a,b,c, and d are all directories.
 979          *
 980          * from cwd of 'a' on both nodes:
 981          * node1: mv b/c d
 982          * node2: mv d   b/c
 983          *
 984          * And that's why, just like the VFS, we need a file system
 985          * rename lock. */
 986         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
 987                 status = ocfs2_rename_lock(osb);
 988                 if (status < 0) {
 989                         mlog_errno(status);
 990                         goto bail;
 991                 }
 992                 rename_lock = 1;
 993         }
 994
 995         /* if old and new are the same, this'll just do one lock. */
 996         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 997                                    &new_dir_bh, new_dir);
 998         if (status < 0) {
 999                 mlog_errno(status);
1000                 goto bail;
1001         }
1002         parents_locked = 1;
1003
1004         /* make sure both dirs have bhs
1005          * get an extra ref on old_dir_bh if old==new */
1006         if (!new_dir_bh) {
1007                 if (old_dir_bh) {
1008                         new_dir_bh = old_dir_bh;
1009                         get_bh(new_dir_bh);
1010                 } else {
1011                         mlog(ML_ERROR, "no old_dir_bh!\n");
1012                         status = -EIO;
1013                         goto bail;
1014                 }
1015         }
1016
1017         /*
1018          * Aside from allowing a meta data update, the locking here
1019          * also ensures that the downconvert thread on other nodes
1020          * won't have to concurrently downconvert the inode and the
1021          * dentry locks.
1022          */
1023         status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1024         if (status < 0) {
1025                 if (status != -ENOENT)
1026                         mlog_errno(status);
1027                 goto bail;
1028         }
1029         old_child_locked = 1;
1030
1031         status = ocfs2_remote_dentry_delete(old_dentry);
1032         if (status < 0) {
1033                 mlog_errno(status);
1034                 goto bail;
1035         }
1036
1037         if (S_ISDIR(old_inode->i_mode)) {
1038                 u64 old_inode_parent;
1039
1040                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1041                                                   old_inode, &old_inode_de_bh,
1042                                                   &old_inode_dot_dot_de);
1043                 if (status) {
1044                         status = -EIO;
1045                         goto bail;
1046                 }
1047
1048                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1049                         status = -EIO;
1050                         goto bail;
1051                 }
1052
1053                 if (!new_inode && new_dir != old_dir &&
1054                     new_dir->i_nlink >= OCFS2_LINK_MAX) {
1055                         status = -EMLINK;
1056                         goto bail;
1057                 }
1058         }
1059
1060         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1061                                             old_dentry->d_name.len,
1062                                             &old_de_ino);
1063         if (status) {
1064                 status = -ENOENT;
1065                 goto bail;
1066         }
1067
1068         /*
1069          *  Check for inode number is _not_ due to possible IO errors.
1070          *  We might rmdir the source, keep it as pwd of some process
1071          *  and merrily kill the link to whatever was created under the
1072          *  same name. Goodbye sticky bit ;-<
1073          */
1074         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1075                 status = -ENOENT;
1076                 goto bail;
1077         }
1078
1079         /* check if the target already exists (in which case we need
1080          * to delete it */
1081         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1082                                           new_dentry->d_name.len,
1083                                           &newfe_blkno, new_dir, &new_de_bh,
1084                                           &new_de);
1085         /* The only error we allow here is -ENOENT because the new
1086          * file not existing is perfectly valid. */
1087         if ((status < 0) && (status != -ENOENT)) {
1088                 /* If we cannot find the file specified we should just */
1089                 /* return the error... */
1090                 mlog_errno(status);
1091                 goto bail;
1092         }
1093
1094         if (!new_de && new_inode) {
1095                 /*
1096                  * Target was unlinked by another node while we were
1097                  * waiting to get to ocfs2_rename(). There isn't
1098                  * anything we can do here to help the situation, so
1099                  * bubble up the appropriate error.
1100                  */
1101                 status = -ENOENT;
1102                 goto bail;
1103         }
1104
1105         /* In case we need to overwrite an existing file, we blow it
1106          * away first */
1107         if (new_de) {
1108                 /* VFS didn't think there existed an inode here, but
1109                  * someone else in the cluster must have raced our
1110                  * rename to create one. Today we error cleanly, in
1111                  * the future we should consider calling iget to build
1112                  * a new struct inode for this entry. */
1113                 if (!new_inode) {
1114                         status = -EACCES;
1115
1116                         mlog(0, "We found an inode for name %.*s but VFS "
1117                              "didn't give us one.\n", new_dentry->d_name.len,
1118                              new_dentry->d_name.name);
1119                         goto bail;
1120                 }
1121
1122                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1123                         status = -EACCES;
1124
1125                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1126                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1127                              (unsigned long long)newfe_blkno,
1128                              OCFS2_I(new_inode)->ip_flags);
1129                         goto bail;
1130                 }
1131
1132                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1133                 if (status < 0) {
1134                         if (status != -ENOENT)
1135                                 mlog_errno(status);
1136                         goto bail;
1137                 }
1138                 new_child_locked = 1;
1139
1140                 status = ocfs2_remote_dentry_delete(new_dentry);
1141                 if (status < 0) {
1142                         mlog_errno(status);
1143                         goto bail;
1144                 }
1145
1146                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1147
1148                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1149                      "newfebh=%p bhblocknr=%llu\n", new_de,
1150                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1151                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1152
1153                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1154                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1155                                                           new_inode,
1156                                                           orphan_name,
1157                                                           &orphan_entry_bh);
1158                         if (status < 0) {
1159                                 mlog_errno(status);
1160                                 goto bail;
1161                         }
1162                 }
1163         } else {
1164                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1165
1166                 status = ocfs2_check_dir_for_entry(new_dir,
1167                                                    new_dentry->d_name.name,
1168                                                    new_dentry->d_name.len);
1169                 if (status)
1170                         goto bail;
1171
1172                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1173                                                       new_dentry->d_name.name,
1174                                                       new_dentry->d_name.len,
1175                                                       &insert_entry_bh);
1176                 if (status < 0) {
1177                         mlog_errno(status);
1178                         goto bail;
1179                 }
1180         }
1181
1182         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1183         if (IS_ERR(handle)) {
1184                 status = PTR_ERR(handle);
1185                 handle = NULL;
1186                 mlog_errno(status);
1187                 goto bail;
1188         }
1189
1190         if (new_de) {
1191                 if (S_ISDIR(new_inode->i_mode)) {
1192                         if (!ocfs2_empty_dir(new_inode) ||
1193                             new_inode->i_nlink != 2) {
1194                                 status = -ENOTEMPTY;
1195                                 goto bail;
1196                         }
1197                 }
1198                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1199                                               OCFS2_JOURNAL_ACCESS_WRITE);
1200                 if (status < 0) {
1201                         mlog_errno(status);
1202                         goto bail;
1203                 }
1204
1205                 if (S_ISDIR(new_inode->i_mode) ||
1206                     (newfe->i_links_count == cpu_to_le16(1))){
1207                         status = ocfs2_orphan_add(osb, handle, new_inode,
1208                                                   newfe, orphan_name,
1209                                                   orphan_entry_bh, orphan_dir);
1210                         if (status < 0) {
1211                                 mlog_errno(status);
1212                                 goto bail;
1213                         }
1214                 }
1215
1216                 /* change the dirent to point to the correct inode */
1217                 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1218                                             new_de, old_inode);
1219                 if (status < 0) {
1220                         mlog_errno(status);
1221                         goto bail;
1222                 }
1223                 new_dir->i_version++;
1224
1225                 if (S_ISDIR(new_inode->i_mode))
1226                         newfe->i_links_count = 0;
1227                 else
1228                         le16_add_cpu(&newfe->i_links_count, -1);
1229
1230                 status = ocfs2_journal_dirty(handle, newfe_bh);
1231                 if (status < 0) {
1232                         mlog_errno(status);
1233                         goto bail;
1234                 }
1235         } else {
1236                 /* if the name was not found in new_dir, add it now */
1237                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1238                                          OCFS2_I(old_inode)->ip_blkno,
1239                                          new_dir_bh, insert_entry_bh);
1240         }
1241
1242         old_inode->i_ctime = CURRENT_TIME;
1243         mark_inode_dirty(old_inode);
1244
1245         status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1246                                       OCFS2_JOURNAL_ACCESS_WRITE);
1247         if (status >= 0) {
1248                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1249
1250                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1251                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1252
1253                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1254                 if (status < 0)
1255                         mlog_errno(status);
1256         } else
1257                 mlog_errno(status);
1258
1259         /*
1260          * Now that the name has been added to new_dir, remove the old name.
1261          *
1262          * We don't keep any directory entry context around until now
1263          * because the insert might have changed the type of directory
1264          * we're dealing with.
1265          */
1266         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1267                                      old_dentry->d_name.len,
1268                                      old_dir, &old_de);
1269         if (!old_de_bh) {
1270                 status = -EIO;
1271                 goto bail;
1272         }
1273
1274         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1275         if (status < 0) {
1276                 mlog_errno(status);
1277                 goto bail;
1278         }
1279
1280         if (new_inode) {
1281                 new_inode->i_nlink--;
1282                 new_inode->i_ctime = CURRENT_TIME;
1283         }
1284         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1285         if (old_inode_de_bh) {
1286                 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1287                                             old_inode_dot_dot_de, new_dir);
1288                 old_dir->i_nlink--;
1289                 if (new_inode) {
1290                         new_inode->i_nlink--;
1291                 } else {
1292                         inc_nlink(new_dir);
1293                         mark_inode_dirty(new_dir);
1294                 }
1295         }
1296         mark_inode_dirty(old_dir);
1297         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1298         if (new_inode) {
1299                 mark_inode_dirty(new_inode);
1300                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1301         }
1302
1303         if (old_dir != new_dir) {
1304                 /* Keep the same times on both directories.*/
1305                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1306
1307                 /*
1308                  * This will also pick up the i_nlink change from the
1309                  * block above.
1310                  */
1311                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1312         }
1313
1314         if (old_dir_nlink != old_dir->i_nlink) {
1315                 if (!old_dir_bh) {
1316                         mlog(ML_ERROR, "need to change nlink for old dir "
1317                              "%llu from %d to %d but bh is NULL!\n",
1318                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1319                              (int)old_dir_nlink, old_dir->i_nlink);
1320                 } else {
1321                         struct ocfs2_dinode *fe;
1322                         status = ocfs2_journal_access(handle, old_dir,
1323                                                       old_dir_bh,
1324                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1325                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1326                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1327                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1328                 }
1329         }
1330
1331         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1332         status = 0;
1333 bail:
1334         if (rename_lock)
1335                 ocfs2_rename_unlock(osb);
1336
1337         if (handle)
1338                 ocfs2_commit_trans(osb, handle);
1339
1340         if (parents_locked)
1341                 ocfs2_double_unlock(old_dir, new_dir);
1342
1343         if (old_child_locked)
1344                 ocfs2_inode_unlock(old_inode, 1);
1345
1346         if (new_child_locked)
1347                 ocfs2_inode_unlock(new_inode, 1);
1348
1349         if (orphan_dir) {
1350                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1351                 ocfs2_inode_unlock(orphan_dir, 1);
1352                 mutex_unlock(&orphan_dir->i_mutex);
1353                 iput(orphan_dir);
1354         }
1355
1356         if (new_inode)
1357                 sync_mapping_buffers(old_inode->i_mapping);
1358
1359         if (new_inode)
1360                 iput(new_inode);
1361         brelse(newfe_bh);
1362         brelse(old_inode_bh);
1363         brelse(old_dir_bh);
1364         brelse(new_dir_bh);
1365         brelse(new_de_bh);
1366         brelse(old_de_bh);
1367         brelse(old_inode_de_bh);
1368         brelse(orphan_entry_bh);
1369         brelse(insert_entry_bh);
1370
1371         mlog_exit(status);
1372
1373         return status;
1374 }
1375
1376 /*
1377  * we expect i_size = strlen(symname). Copy symname into the file
1378  * data, including the null terminator.
1379  */
1380 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1381                                      handle_t *handle,
1382                                      struct inode *inode,
1383                                      const char *symname)
1384 {
1385         struct buffer_head **bhs = NULL;
1386         const char *c;
1387         struct super_block *sb = osb->sb;
1388         u64 p_blkno, p_blocks;
1389         int virtual, blocks, status, i, bytes_left;
1390
1391         bytes_left = i_size_read(inode) + 1;
1392         /* we can't trust i_blocks because we're actually going to
1393          * write i_size + 1 bytes. */
1394         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1395
1396         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1397                         (unsigned long long)inode->i_blocks,
1398                         i_size_read(inode), blocks);
1399
1400         /* Sanity check -- make sure we're going to fit. */
1401         if (bytes_left >
1402             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1403                 status = -EIO;
1404                 mlog_errno(status);
1405                 goto bail;
1406         }
1407
1408         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1409         if (!bhs) {
1410                 status = -ENOMEM;
1411                 mlog_errno(status);
1412                 goto bail;
1413         }
1414
1415         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1416                                              NULL);
1417         if (status < 0) {
1418                 mlog_errno(status);
1419                 goto bail;
1420         }
1421
1422         /* links can never be larger than one cluster so we know this
1423          * is all going to be contiguous, but do a sanity check
1424          * anyway. */
1425         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1426                 status = -EIO;
1427                 mlog_errno(status);
1428                 goto bail;
1429         }
1430
1431         virtual = 0;
1432         while(bytes_left > 0) {
1433                 c = &symname[virtual * sb->s_blocksize];
1434
1435                 bhs[virtual] = sb_getblk(sb, p_blkno);
1436                 if (!bhs[virtual]) {
1437                         status = -ENOMEM;
1438                         mlog_errno(status);
1439                         goto bail;
1440                 }
1441                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1442
1443                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1444                                               OCFS2_JOURNAL_ACCESS_CREATE);
1445                 if (status < 0) {
1446                         mlog_errno(status);
1447                         goto bail;
1448                 }
1449
1450                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1451
1452                 memcpy(bhs[virtual]->b_data, c,
1453                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1454                        bytes_left);
1455
1456                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1457                 if (status < 0) {
1458                         mlog_errno(status);
1459                         goto bail;
1460                 }
1461
1462                 virtual++;
1463                 p_blkno++;
1464                 bytes_left -= sb->s_blocksize;
1465         }
1466
1467         status = 0;
1468 bail:
1469
1470         if (bhs) {
1471                 for(i = 0; i < blocks; i++)
1472                         brelse(bhs[i]);
1473                 kfree(bhs);
1474         }
1475
1476         mlog_exit(status);
1477         return status;
1478 }
1479
1480 static int ocfs2_symlink(struct inode *dir,
1481                          struct dentry *dentry,
1482                          const char *symname)
1483 {
1484         int status, l, credits;
1485         u64 newsize;
1486         struct ocfs2_super *osb = NULL;
1487         struct inode *inode = NULL;
1488         struct super_block *sb;
1489         struct buffer_head *new_fe_bh = NULL;
1490         struct buffer_head *de_bh = NULL;
1491         struct buffer_head *parent_fe_bh = NULL;
1492         struct ocfs2_dinode *fe = NULL;
1493         struct ocfs2_dinode *dirfe;
1494         handle_t *handle = NULL;
1495         struct ocfs2_alloc_context *inode_ac = NULL;
1496         struct ocfs2_alloc_context *data_ac = NULL;
1497
1498         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1499                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1500
1501         sb = dir->i_sb;
1502         osb = OCFS2_SB(sb);
1503
1504         l = strlen(symname) + 1;
1505
1506         credits = ocfs2_calc_symlink_credits(sb);
1507
1508         /* lock the parent directory */
1509         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1510         if (status < 0) {
1511                 if (status != -ENOENT)
1512                         mlog_errno(status);
1513                 return status;
1514         }
1515
1516         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1517         if (!dirfe->i_links_count) {
1518                 /* can't make a file in a deleted directory. */
1519                 status = -ENOENT;
1520                 goto bail;
1521         }
1522
1523         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1524                                            dentry->d_name.len);
1525         if (status)
1526                 goto bail;
1527
1528         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1529                                               dentry->d_name.name,
1530                                               dentry->d_name.len, &de_bh);
1531         if (status < 0) {
1532                 mlog_errno(status);
1533                 goto bail;
1534         }
1535
1536         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1537         if (status < 0) {
1538                 if (status != -ENOSPC)
1539                         mlog_errno(status);
1540                 goto bail;
1541         }
1542
1543         /* don't reserve bitmap space for fast symlinks. */
1544         if (l > ocfs2_fast_symlink_chars(sb)) {
1545                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1546                 if (status < 0) {
1547                         if (status != -ENOSPC)
1548                                 mlog_errno(status);
1549                         goto bail;
1550                 }
1551         }
1552
1553         handle = ocfs2_start_trans(osb, credits);
1554         if (IS_ERR(handle)) {
1555                 status = PTR_ERR(handle);
1556                 handle = NULL;
1557                 mlog_errno(status);
1558                 goto bail;
1559         }
1560
1561         status = ocfs2_mknod_locked(osb, dir, dentry,
1562                                     S_IFLNK | S_IRWXUGO, 0,
1563                                     &new_fe_bh, parent_fe_bh, handle,
1564                                     &inode, inode_ac);
1565         if (status < 0) {
1566                 mlog_errno(status);
1567                 goto bail;
1568         }
1569
1570         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1571         inode->i_rdev = 0;
1572         newsize = l - 1;
1573         if (l > ocfs2_fast_symlink_chars(sb)) {
1574                 u32 offset = 0;
1575
1576                 inode->i_op = &ocfs2_symlink_inode_operations;
1577                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1578                                               new_fe_bh,
1579                                               handle, data_ac, NULL,
1580                                               NULL);
1581                 if (status < 0) {
1582                         if (status != -ENOSPC && status != -EINTR) {
1583                                 mlog(ML_ERROR,
1584                                      "Failed to extend file to %llu\n",
1585                                      (unsigned long long)newsize);
1586                                 mlog_errno(status);
1587                                 status = -ENOSPC;
1588                         }
1589                         goto bail;
1590                 }
1591                 i_size_write(inode, newsize);
1592                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1593         } else {
1594                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1595                 memcpy((char *) fe->id2.i_symlink, symname, l);
1596                 i_size_write(inode, newsize);
1597                 inode->i_blocks = 0;
1598         }
1599
1600         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1601         if (status < 0) {
1602                 mlog_errno(status);
1603                 goto bail;
1604         }
1605
1606         if (!ocfs2_inode_is_fast_symlink(inode)) {
1607                 status = ocfs2_create_symlink_data(osb, handle, inode,
1608                                                    symname);
1609                 if (status < 0) {
1610                         mlog_errno(status);
1611                         goto bail;
1612                 }
1613         }
1614
1615         status = ocfs2_add_entry(handle, dentry, inode,
1616                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1617                                  de_bh);
1618         if (status < 0) {
1619                 mlog_errno(status);
1620                 goto bail;
1621         }
1622
1623         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1624         if (status) {
1625                 mlog_errno(status);
1626                 goto bail;
1627         }
1628
1629         insert_inode_hash(inode);
1630         dentry->d_op = &ocfs2_dentry_ops;
1631         d_instantiate(dentry, inode);
1632 bail:
1633         if (handle)
1634                 ocfs2_commit_trans(osb, handle);
1635
1636         ocfs2_inode_unlock(dir, 1);
1637
1638         brelse(new_fe_bh);
1639         brelse(parent_fe_bh);
1640         brelse(de_bh);
1641         if (inode_ac)
1642                 ocfs2_free_alloc_context(inode_ac);
1643         if (data_ac)
1644                 ocfs2_free_alloc_context(data_ac);
1645         if ((status < 0) && inode)
1646                 iput(inode);
1647
1648         mlog_exit(status);
1649
1650         return status;
1651 }
1652
1653 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1654 {
1655         int status, namelen;
1656
1657         mlog_entry_void();
1658
1659         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1660                            (long long)blkno);
1661         if (namelen <= 0) {
1662                 if (namelen)
1663                         status = namelen;
1664                 else
1665                         status = -EINVAL;
1666                 mlog_errno(status);
1667                 goto bail;
1668         }
1669         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1670                 status = -EINVAL;
1671                 mlog_errno(status);
1672                 goto bail;
1673         }
1674
1675         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1676              namelen);
1677
1678         status = 0;
1679 bail:
1680         mlog_exit(status);
1681         return status;
1682 }
1683
1684 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1685                                     struct inode **ret_orphan_dir,
1686                                     struct inode *inode,
1687                                     char *name,
1688                                     struct buffer_head **de_bh)
1689 {
1690         struct inode *orphan_dir_inode;
1691         struct buffer_head *orphan_dir_bh = NULL;
1692         int status = 0;
1693
1694         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1695         if (status < 0) {
1696                 mlog_errno(status);
1697                 return status;
1698         }
1699
1700         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1701                                                        ORPHAN_DIR_SYSTEM_INODE,
1702                                                        osb->slot_num);
1703         if (!orphan_dir_inode) {
1704                 status = -ENOENT;
1705                 mlog_errno(status);
1706                 return status;
1707         }
1708
1709         mutex_lock(&orphan_dir_inode->i_mutex);
1710
1711         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1712         if (status < 0) {
1713                 mlog_errno(status);
1714                 goto leave;
1715         }
1716
1717         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1718                                               orphan_dir_bh, name,
1719                                               OCFS2_ORPHAN_NAMELEN, de_bh);
1720         if (status < 0) {
1721                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1722
1723                 mlog_errno(status);
1724                 goto leave;
1725         }
1726
1727         *ret_orphan_dir = orphan_dir_inode;
1728
1729 leave:
1730         if (status) {
1731                 mutex_unlock(&orphan_dir_inode->i_mutex);
1732                 iput(orphan_dir_inode);
1733         }
1734
1735         brelse(orphan_dir_bh);
1736
1737         mlog_exit(status);
1738         return status;
1739 }
1740
1741 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1742                             handle_t *handle,
1743                             struct inode *inode,
1744                             struct ocfs2_dinode *fe,
1745                             char *name,
1746                             struct buffer_head *de_bh,
1747                             struct inode *orphan_dir_inode)
1748 {
1749         struct buffer_head *orphan_dir_bh = NULL;
1750         int status = 0;
1751         struct ocfs2_dinode *orphan_fe;
1752
1753         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1754
1755         status = ocfs2_read_block(orphan_dir_inode,
1756                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
1757                                   &orphan_dir_bh);
1758         if (status < 0) {
1759                 mlog_errno(status);
1760                 goto leave;
1761         }
1762
1763         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1764                                       OCFS2_JOURNAL_ACCESS_WRITE);
1765         if (status < 0) {
1766                 mlog_errno(status);
1767                 goto leave;
1768         }
1769
1770         /* we're a cluster, and nlink can change on disk from
1771          * underneath us... */
1772         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1773         if (S_ISDIR(inode->i_mode))
1774                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1775         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1776
1777         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1778         if (status < 0) {
1779                 mlog_errno(status);
1780                 goto leave;
1781         }
1782
1783         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1784                                    OCFS2_ORPHAN_NAMELEN, inode,
1785                                    OCFS2_I(inode)->ip_blkno,
1786                                    orphan_dir_bh, de_bh);
1787         if (status < 0) {
1788                 mlog_errno(status);
1789                 goto leave;
1790         }
1791
1792         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1793
1794         /* Record which orphan dir our inode now resides
1795          * in. delete_inode will use this to determine which orphan
1796          * dir to lock. */
1797         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1798
1799         mlog(0, "Inode %llu orphaned in slot %d\n",
1800              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1801
1802 leave:
1803         brelse(orphan_dir_bh);
1804
1805         mlog_exit(status);
1806         return status;
1807 }
1808
1809 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1810 int ocfs2_orphan_del(struct ocfs2_super *osb,
1811                      handle_t *handle,
1812                      struct inode *orphan_dir_inode,
1813                      struct inode *inode,
1814                      struct buffer_head *orphan_dir_bh)
1815 {
1816         char name[OCFS2_ORPHAN_NAMELEN + 1];
1817         struct ocfs2_dinode *orphan_fe;
1818         int status = 0;
1819         struct buffer_head *target_de_bh = NULL;
1820         struct ocfs2_dir_entry *target_de = NULL;
1821
1822         mlog_entry_void();
1823
1824         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1825         if (status < 0) {
1826                 mlog_errno(status);
1827                 goto leave;
1828         }
1829
1830         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1831              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1832              OCFS2_ORPHAN_NAMELEN);
1833
1834         /* find it's spot in the orphan directory */
1835         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1836                                         orphan_dir_inode, &target_de);
1837         if (!target_de_bh) {
1838                 status = -ENOENT;
1839                 mlog_errno(status);
1840                 goto leave;
1841         }
1842
1843         /* remove it from the orphan directory */
1844         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1845                                     target_de_bh);
1846         if (status < 0) {
1847                 mlog_errno(status);
1848                 goto leave;
1849         }
1850
1851         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
1852                                       OCFS2_JOURNAL_ACCESS_WRITE);
1853         if (status < 0) {
1854                 mlog_errno(status);
1855                 goto leave;
1856         }
1857
1858         /* do the i_nlink dance! :) */
1859         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1860         if (S_ISDIR(inode->i_mode))
1861                 le16_add_cpu(&orphan_fe->i_links_count, -1);
1862         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1863
1864         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1865         if (status < 0) {
1866                 mlog_errno(status);
1867                 goto leave;
1868         }
1869
1870 leave:
1871         brelse(target_de_bh);
1872
1873         mlog_exit(status);
1874         return status;
1875 }
1876
1877 const struct inode_operations ocfs2_dir_iops = {
1878         .create         = ocfs2_create,
1879         .lookup         = ocfs2_lookup,
1880         .link           = ocfs2_link,
1881         .unlink         = ocfs2_unlink,
1882         .rmdir          = ocfs2_unlink,
1883         .symlink        = ocfs2_symlink,
1884         .mkdir          = ocfs2_mkdir,
1885         .mknod          = ocfs2_mknod,
1886         .rename         = ocfs2_rename,
1887         .setattr        = ocfs2_setattr,
1888         .getattr        = ocfs2_getattr,
1889         .permission     = ocfs2_permission,
1890         .setxattr       = generic_setxattr,
1891         .getxattr       = generic_getxattr,
1892         .listxattr      = ocfs2_listxattr,
1893         .removexattr    = generic_removexattr,
1894 };