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
  64 #include "buffer_head_io.h"
  65
  66 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  67                               struct inode *dir,
  68                               struct dentry *dentry, int mode,
  69                               dev_t dev,
  70                               struct buffer_head **new_fe_bh,
  71                               struct buffer_head *parent_fe_bh,
  72                               handle_t *handle,
  73                               struct inode **ret_inode,
  74                               struct ocfs2_alloc_context *inode_ac);
  75
  76 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  77                                     struct inode **ret_orphan_dir,
  78                                     struct inode *inode,
  79                                     char *name,
  80                                     struct buffer_head **de_bh);
  81
  82 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  83                             handle_t *handle,
  84                             struct inode *inode,
  85                             struct ocfs2_dinode *fe,
  86                             char *name,
  87                             struct buffer_head *de_bh,
  88                             struct inode *orphan_dir_inode);
  89
  90 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  91                                      handle_t *handle,
  92                                      struct inode *inode,
  93                                      const char *symname);
  94
  95 /* An orphan dir name is an 8 byte value, printed as a hex string */
  96 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  97
  98 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
  99                                    struct nameidata *nd)
 100 {
 101         int status;
 102         u64 blkno;
 103         struct inode *inode = NULL;
 104         struct dentry *ret;
 105         struct ocfs2_inode_info *oi;
 106
 107         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 108                    dentry->d_name.len, dentry->d_name.name);
 109
 110         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 111                 ret = ERR_PTR(-ENAMETOOLONG);
 112                 goto bail;
 113         }
 114
 115         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 116              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 117
 118         status = ocfs2_inode_lock(dir, NULL, 0);
 119         if (status < 0) {
 120                 if (status != -ENOENT)
 121                         mlog_errno(status);
 122                 ret = ERR_PTR(status);
 123                 goto bail;
 124         }
 125
 126         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 127                                             dentry->d_name.len, &blkno);
 128         if (status < 0)
 129                 goto bail_add;
 130
 131         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 132         if (IS_ERR(inode)) {
 133                 ret = ERR_PTR(-EACCES);
 134                 goto bail_unlock;
 135         }
 136
 137         oi = OCFS2_I(inode);
 138         /* Clear any orphaned state... If we were able to look up the
 139          * inode from a directory, it certainly can't be orphaned. We
 140          * might have the bad state from a node which intended to
 141          * orphan this inode but crashed before it could commit the
 142          * unlink. */
 143         spin_lock(&oi->ip_lock);
 144         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 145         spin_unlock(&oi->ip_lock);
 146
 147 bail_add:
 148         dentry->d_op = &ocfs2_dentry_ops;
 149         ret = d_splice_alias(inode, dentry);
 150
 151         if (inode) {
 152                 /*
 153                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 154                  * dentry, it will d_move() it on top of ourse. The
 155                  * return value will indicate this however, so in
 156                  * those cases, we switch them around for the locking
 157                  * code.
 158                  *
 159                  * NOTE: This dentry already has ->d_op set from
 160                  * ocfs2_get_parent() and ocfs2_get_dentry()
 161                  */
 162                 if (ret)
 163                         dentry = ret;
 164
 165                 status = ocfs2_dentry_attach_lock(dentry, inode,
 166                                                   OCFS2_I(dir)->ip_blkno);
 167                 if (status) {
 168                         mlog_errno(status);
 169                         ret = ERR_PTR(status);
 170                         goto bail_unlock;
 171                 }
 172         }
 173
 174 bail_unlock:
 175         /* Don't drop the cluster lock until *after* the d_add --
 176          * unlink on another node will message us to remove that
 177          * dentry under this lock so otherwise we can race this with
 178          * the downconvert thread and have a stale dentry. */
 179         ocfs2_inode_unlock(dir, 0);
 180
 181 bail:
 182
 183         mlog_exit_ptr(ret);
 184
 185         return ret;
 186 }
 187
 188 static int ocfs2_mknod(struct inode *dir,
 189                        struct dentry *dentry,
 190                        int mode,
 191                        dev_t dev)
 192 {
 193         int status = 0;
 194         struct buffer_head *parent_fe_bh = NULL;
 195         handle_t *handle = NULL;
 196         struct ocfs2_super *osb;
 197         struct ocfs2_dinode *dirfe;
 198         struct buffer_head *new_fe_bh = NULL;
 199         struct buffer_head *de_bh = NULL;
 200         struct inode *inode = NULL;
 201         struct ocfs2_alloc_context *inode_ac = NULL;
 202         struct ocfs2_alloc_context *data_ac = NULL;
 203
 204         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 205                    (unsigned long)dev, dentry->d_name.len,
 206                    dentry->d_name.name);
 207
 208         /* get our super block */
 209         osb = OCFS2_SB(dir->i_sb);
 210
 211         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 212         if (status < 0) {
 213                 if (status != -ENOENT)
 214                         mlog_errno(status);
 215                 return status;
 216         }
 217
 218         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 219                 status = -EMLINK;
 220                 goto leave;
 221         }
 222
 223         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 224         if (!dirfe->i_links_count) {
 225                 /* can't make a file in a deleted directory. */
 226                 status = -ENOENT;
 227                 goto leave;
 228         }
 229
 230         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 231                                            dentry->d_name.len);
 232         if (status)
 233                 goto leave;
 234
 235         /* get a spot inside the dir. */
 236         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 237                                               dentry->d_name.name,
 238                                               dentry->d_name.len, &de_bh);
 239         if (status < 0) {
 240                 mlog_errno(status);
 241                 goto leave;
 242         }
 243
 244         /* reserve an inode spot */
 245         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 246         if (status < 0) {
 247                 if (status != -ENOSPC)
 248                         mlog_errno(status);
 249                 goto leave;
 250         }
 251
 252         /* Reserve a cluster if creating an extent based directory. */
 253         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 254                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 255                 if (status < 0) {
 256                         if (status != -ENOSPC)
 257                                 mlog_errno(status);
 258                         goto leave;
 259                 }
 260         }
 261
 262         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
 263         if (IS_ERR(handle)) {
 264                 status = PTR_ERR(handle);
 265                 handle = NULL;
 266                 mlog_errno(status);
 267                 goto leave;
 268         }
 269
 270         /* do the real work now. */
 271         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 272                                     &new_fe_bh, parent_fe_bh, handle,
 273                                     &inode, inode_ac);
 274         if (status < 0) {
 275                 mlog_errno(status);
 276                 goto leave;
 277         }
 278
 279         if (S_ISDIR(mode)) {
 280                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 281                                             new_fe_bh, data_ac);
 282                 if (status < 0) {
 283                         mlog_errno(status);
 284                         goto leave;
 285                 }
 286
 287                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 288                                               OCFS2_JOURNAL_ACCESS_WRITE);
 289                 if (status < 0) {
 290                         mlog_errno(status);
 291                         goto leave;
 292                 }
 293                 le16_add_cpu(&dirfe->i_links_count, 1);
 294                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 295                 if (status < 0) {
 296                         mlog_errno(status);
 297                         goto leave;
 298                 }
 299                 inc_nlink(dir);
 300         }
 301
 302         status = ocfs2_add_entry(handle, dentry, inode,
 303                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 304                                  de_bh);
 305         if (status < 0) {
 306                 mlog_errno(status);
 307                 goto leave;
 308         }
 309
 310         status = ocfs2_dentry_attach_lock(dentry, inode,
 311                                           OCFS2_I(dir)->ip_blkno);
 312         if (status) {
 313                 mlog_errno(status);
 314                 goto leave;
 315         }
 316
 317         insert_inode_hash(inode);
 318         dentry->d_op = &ocfs2_dentry_ops;
 319         d_instantiate(dentry, inode);
 320         status = 0;
 321 leave:
 322         if (handle)
 323                 ocfs2_commit_trans(osb, handle);
 324
 325         ocfs2_inode_unlock(dir, 1);
 326
 327         if (status == -ENOSPC)
 328                 mlog(0, "Disk is full\n");
 329
 330         if (new_fe_bh)
 331                 brelse(new_fe_bh);
 332
 333         if (de_bh)
 334                 brelse(de_bh);
 335
 336         if (parent_fe_bh)
 337                 brelse(parent_fe_bh);
 338
 339         if ((status < 0) && inode)
 340                 iput(inode);
 341
 342         if (inode_ac)
 343                 ocfs2_free_alloc_context(inode_ac);
 344
 345         if (data_ac)
 346                 ocfs2_free_alloc_context(data_ac);
 347
 348         mlog_exit(status);
 349
 350         return status;
 351 }
 352
 353 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 354                               struct inode *dir,
 355                               struct dentry *dentry, int mode,
 356                               dev_t dev,
 357                               struct buffer_head **new_fe_bh,
 358                               struct buffer_head *parent_fe_bh,
 359                               handle_t *handle,
 360                               struct inode **ret_inode,
 361                               struct ocfs2_alloc_context *inode_ac)
 362 {
 363         int status = 0;
 364         struct ocfs2_dinode *fe = NULL;
 365         struct ocfs2_extent_list *fel;
 366         u64 fe_blkno = 0;
 367         u16 suballoc_bit;
 368         struct inode *inode = NULL;
 369
 370         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 371                    (unsigned long)dev, dentry->d_name.len,
 372                    dentry->d_name.name);
 373
 374         *new_fe_bh = NULL;
 375         *ret_inode = NULL;
 376
 377         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 378                                        &fe_blkno);
 379         if (status < 0) {
 380                 mlog_errno(status);
 381                 goto leave;
 382         }
 383
 384         inode = new_inode(dir->i_sb);
 385         if (IS_ERR(inode)) {
 386                 status = PTR_ERR(inode);
 387                 mlog(ML_ERROR, "new_inode failed!\n");
 388                 goto leave;
 389         }
 390
 391         /* populate as many fields early on as possible - many of
 392          * these are used by the support functions here and in
 393          * callers. */
 394         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 395         OCFS2_I(inode)->ip_blkno = fe_blkno;
 396         if (S_ISDIR(mode))
 397                 inode->i_nlink = 2;
 398         else
 399                 inode->i_nlink = 1;
 400         inode->i_mode = mode;
 401         spin_lock(&osb->osb_lock);
 402         inode->i_generation = osb->s_next_generation++;
 403         spin_unlock(&osb->osb_lock);
 404
 405         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 406         if (!*new_fe_bh) {
 407                 status = -EIO;
 408                 mlog_errno(status);
 409                 goto leave;
 410         }
 411         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 412
 413         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 414                                       OCFS2_JOURNAL_ACCESS_CREATE);
 415         if (status < 0) {
 416                 mlog_errno(status);
 417                 goto leave;
 418         }
 419
 420         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 421         memset(fe, 0, osb->sb->s_blocksize);
 422
 423         fe->i_generation = cpu_to_le32(inode->i_generation);
 424         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 425         fe->i_blkno = cpu_to_le64(fe_blkno);
 426         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 427         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 428         fe->i_uid = cpu_to_le32(current->fsuid);
 429         if (dir->i_mode & S_ISGID) {
 430                 fe->i_gid = cpu_to_le32(dir->i_gid);
 431                 if (S_ISDIR(mode))
 432                         mode |= S_ISGID;
 433         } else
 434                 fe->i_gid = cpu_to_le32(current->fsgid);
 435         fe->i_mode = cpu_to_le16(mode);
 436         if (S_ISCHR(mode) || S_ISBLK(mode))
 437                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 438
 439         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 440
 441         fe->i_last_eb_blk = 0;
 442         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 443         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 444         fe->i_atime = fe->i_ctime = fe->i_mtime =
 445                 cpu_to_le64(CURRENT_TIME.tv_sec);
 446         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 447                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 448         fe->i_dtime = 0;
 449
 450         /*
 451          * If supported, directories start with inline data.
 452          */
 453         if (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) {
 454                 u16 feat = le16_to_cpu(fe->i_dyn_features);
 455
 456                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 457
 458                 fe->id2.i_data.id_count = cpu_to_le16(ocfs2_max_inline_data(osb->sb));
 459         } else {
 460                 fel = &fe->id2.i_list;
 461                 fel->l_tree_depth = 0;
 462                 fel->l_next_free_rec = 0;
 463                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 464         }
 465
 466         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 467         if (status < 0) {
 468                 mlog_errno(status);
 469                 goto leave;
 470         }
 471
 472         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 473                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 474                      "i_blkno=%llu, i_ino=%lu\n",
 475                      (unsigned long long)(*new_fe_bh)->b_blocknr,
 476                      (unsigned long long)le64_to_cpu(fe->i_blkno),
 477                      inode->i_ino);
 478                 BUG();
 479         }
 480
 481         ocfs2_inode_set_new(osb, inode);
 482         if (!ocfs2_mount_local(osb)) {
 483                 status = ocfs2_create_new_inode_locks(inode);
 484                 if (status < 0)
 485                         mlog_errno(status);
 486         }
 487
 488         status = 0; /* error in ocfs2_create_new_inode_locks is not
 489                      * critical */
 490
 491         *ret_inode = inode;
 492 leave:
 493         if (status < 0) {
 494                 if (*new_fe_bh) {
 495                         brelse(*new_fe_bh);
 496                         *new_fe_bh = NULL;
 497                 }
 498                 if (inode)
 499                         iput(inode);
 500         }
 501
 502         mlog_exit(status);
 503         return status;
 504 }
 505
 506 static int ocfs2_mkdir(struct inode *dir,
 507                        struct dentry *dentry,
 508                        int mode)
 509 {
 510         int ret;
 511
 512         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 513                    dentry->d_name.len, dentry->d_name.name);
 514         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 515         mlog_exit(ret);
 516
 517         return ret;
 518 }
 519
 520 static int ocfs2_create(struct inode *dir,
 521                         struct dentry *dentry,
 522                         int mode,
 523                         struct nameidata *nd)
 524 {
 525         int ret;
 526
 527         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 528                    dentry->d_name.len, dentry->d_name.name);
 529         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 530         mlog_exit(ret);
 531
 532         return ret;
 533 }
 534
 535 static int ocfs2_link(struct dentry *old_dentry,
 536                       struct inode *dir,
 537                       struct dentry *dentry)
 538 {
 539         handle_t *handle;
 540         struct inode *inode = old_dentry->d_inode;
 541         int err;
 542         struct buffer_head *fe_bh = NULL;
 543         struct buffer_head *parent_fe_bh = NULL;
 544         struct buffer_head *de_bh = NULL;
 545         struct ocfs2_dinode *fe = NULL;
 546         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 547
 548         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 549                    old_dentry->d_name.len, old_dentry->d_name.name,
 550                    dentry->d_name.len, dentry->d_name.name);
 551
 552         if (S_ISDIR(inode->i_mode))
 553                 return -EPERM;
 554
 555         err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 556         if (err < 0) {
 557                 if (err != -ENOENT)
 558                         mlog_errno(err);
 559                 return err;
 560         }
 561
 562         if (!dir->i_nlink) {
 563                 err = -ENOENT;
 564                 goto out;
 565         }
 566
 567         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 568                                         dentry->d_name.len);
 569         if (err)
 570                 goto out;
 571
 572         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 573                                            dentry->d_name.name,
 574                                            dentry->d_name.len, &de_bh);
 575         if (err < 0) {
 576                 mlog_errno(err);
 577                 goto out;
 578         }
 579
 580         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 581         if (err < 0) {
 582                 if (err != -ENOENT)
 583                         mlog_errno(err);
 584                 goto out;
 585         }
 586
 587         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 588         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 589                 err = -EMLINK;
 590                 goto out_unlock_inode;
 591         }
 592
 593         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
 594         if (IS_ERR(handle)) {
 595                 err = PTR_ERR(handle);
 596                 handle = NULL;
 597                 mlog_errno(err);
 598                 goto out_unlock_inode;
 599         }
 600
 601         err = ocfs2_journal_access(handle, inode, fe_bh,
 602                                    OCFS2_JOURNAL_ACCESS_WRITE);
 603         if (err < 0) {
 604                 mlog_errno(err);
 605                 goto out_commit;
 606         }
 607
 608         inc_nlink(inode);
 609         inode->i_ctime = CURRENT_TIME;
 610         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 611         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 612         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 613
 614         err = ocfs2_journal_dirty(handle, fe_bh);
 615         if (err < 0) {
 616                 le16_add_cpu(&fe->i_links_count, -1);
 617                 drop_nlink(inode);
 618                 mlog_errno(err);
 619                 goto out_commit;
 620         }
 621
 622         err = ocfs2_add_entry(handle, dentry, inode,
 623                               OCFS2_I(inode)->ip_blkno,
 624                               parent_fe_bh, de_bh);
 625         if (err) {
 626                 le16_add_cpu(&fe->i_links_count, -1);
 627                 drop_nlink(inode);
 628                 mlog_errno(err);
 629                 goto out_commit;
 630         }
 631
 632         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 633         if (err) {
 634                 mlog_errno(err);
 635                 goto out_commit;
 636         }
 637
 638         atomic_inc(&inode->i_count);
 639         dentry->d_op = &ocfs2_dentry_ops;
 640         d_instantiate(dentry, inode);
 641
 642 out_commit:
 643         ocfs2_commit_trans(osb, handle);
 644 out_unlock_inode:
 645         ocfs2_inode_unlock(inode, 1);
 646
 647 out:
 648         ocfs2_inode_unlock(dir, 1);
 649
 650         if (de_bh)
 651                 brelse(de_bh);
 652         if (fe_bh)
 653                 brelse(fe_bh);
 654         if (parent_fe_bh)
 655                 brelse(parent_fe_bh);
 656
 657         mlog_exit(err);
 658
 659         return err;
 660 }
 661
 662 /*
 663  * Takes and drops an exclusive lock on the given dentry. This will
 664  * force other nodes to drop it.
 665  */
 666 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 667 {
 668         int ret;
 669
 670         ret = ocfs2_dentry_lock(dentry, 1);
 671         if (ret)
 672                 mlog_errno(ret);
 673         else
 674                 ocfs2_dentry_unlock(dentry, 1);
 675
 676         return ret;
 677 }
 678
 679 static inline int inode_is_unlinkable(struct inode *inode)
 680 {
 681         if (S_ISDIR(inode->i_mode)) {
 682                 if (inode->i_nlink == 2)
 683                         return 1;
 684                 return 0;
 685         }
 686
 687         if (inode->i_nlink == 1)
 688                 return 1;
 689         return 0;
 690 }
 691
 692 static int ocfs2_unlink(struct inode *dir,
 693                         struct dentry *dentry)
 694 {
 695         int status;
 696         int child_locked = 0;
 697         struct inode *inode = dentry->d_inode;
 698         struct inode *orphan_dir = NULL;
 699         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 700         u64 blkno;
 701         struct ocfs2_dinode *fe = NULL;
 702         struct buffer_head *fe_bh = NULL;
 703         struct buffer_head *parent_node_bh = NULL;
 704         handle_t *handle = NULL;
 705         struct ocfs2_dir_entry *dirent = NULL;
 706         struct buffer_head *dirent_bh = NULL;
 707         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 708         struct buffer_head *orphan_entry_bh = NULL;
 709
 710         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 711                    dentry->d_name.len, dentry->d_name.name);
 712
 713         BUG_ON(dentry->d_parent->d_inode != dir);
 714
 715         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 716
 717         if (inode == osb->root_inode) {
 718                 mlog(0, "Cannot delete the root directory\n");
 719                 return -EPERM;
 720         }
 721
 722         status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
 723         if (status < 0) {
 724                 if (status != -ENOENT)
 725                         mlog_errno(status);
 726                 return status;
 727         }
 728
 729         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 730                                           dentry->d_name.len, &blkno,
 731                                           dir, &dirent_bh, &dirent);
 732         if (status < 0) {
 733                 if (status != -ENOENT)
 734                         mlog_errno(status);
 735                 goto leave;
 736         }
 737
 738         if (OCFS2_I(inode)->ip_blkno != blkno) {
 739                 status = -ENOENT;
 740
 741                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 742                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 743                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 744                 goto leave;
 745         }
 746
 747         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 748         if (status < 0) {
 749                 if (status != -ENOENT)
 750                         mlog_errno(status);
 751                 goto leave;
 752         }
 753         child_locked = 1;
 754
 755         if (S_ISDIR(inode->i_mode)) {
 756                 if (!ocfs2_empty_dir(inode)) {
 757                         status = -ENOTEMPTY;
 758                         goto leave;
 759                 } else if (inode->i_nlink != 2) {
 760                         status = -ENOTEMPTY;
 761                         goto leave;
 762                 }
 763         }
 764
 765         status = ocfs2_remote_dentry_delete(dentry);
 766         if (status < 0) {
 767                 /* This remote delete should succeed under all normal
 768                  * circumstances. */
 769                 mlog_errno(status);
 770                 goto leave;
 771         }
 772
 773         if (inode_is_unlinkable(inode)) {
 774                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 775                                                   orphan_name,
 776                                                   &orphan_entry_bh);
 777                 if (status < 0) {
 778                         mlog_errno(status);
 779                         goto leave;
 780                 }
 781         }
 782
 783         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
 784         if (IS_ERR(handle)) {
 785                 status = PTR_ERR(handle);
 786                 handle = NULL;
 787                 mlog_errno(status);
 788                 goto leave;
 789         }
 790
 791         status = ocfs2_journal_access(handle, inode, fe_bh,
 792                                       OCFS2_JOURNAL_ACCESS_WRITE);
 793         if (status < 0) {
 794                 mlog_errno(status);
 795                 goto leave;
 796         }
 797
 798         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 799
 800         if (inode_is_unlinkable(inode)) {
 801                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 802                                           orphan_entry_bh, orphan_dir);
 803                 if (status < 0) {
 804                         mlog_errno(status);
 805                         goto leave;
 806                 }
 807         }
 808
 809         /* delete the name from the parent dir */
 810         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 811         if (status < 0) {
 812                 mlog_errno(status);
 813                 goto leave;
 814         }
 815
 816         if (S_ISDIR(inode->i_mode))
 817                 drop_nlink(inode);
 818         drop_nlink(inode);
 819         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 820
 821         status = ocfs2_journal_dirty(handle, fe_bh);
 822         if (status < 0) {
 823                 mlog_errno(status);
 824                 goto leave;
 825         }
 826
 827         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 828         if (S_ISDIR(inode->i_mode))
 829                 drop_nlink(dir);
 830
 831         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 832         if (status < 0) {
 833                 mlog_errno(status);
 834                 if (S_ISDIR(inode->i_mode))
 835                         inc_nlink(dir);
 836         }
 837
 838 leave:
 839         if (handle)
 840                 ocfs2_commit_trans(osb, handle);
 841
 842         if (child_locked)
 843                 ocfs2_inode_unlock(inode, 1);
 844
 845         ocfs2_inode_unlock(dir, 1);
 846
 847         if (orphan_dir) {
 848                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 849                 ocfs2_inode_unlock(orphan_dir, 1);
 850                 mutex_unlock(&orphan_dir->i_mutex);
 851                 iput(orphan_dir);
 852         }
 853
 854         if (fe_bh)
 855                 brelse(fe_bh);
 856
 857         if (dirent_bh)
 858                 brelse(dirent_bh);
 859
 860         if (parent_node_bh)
 861                 brelse(parent_node_bh);
 862
 863         if (orphan_entry_bh)
 864                 brelse(orphan_entry_bh);
 865
 866         mlog_exit(status);
 867
 868         return status;
 869 }
 870
 871 /*
 872  * The only place this should be used is rename!
 873  * if they have the same id, then the 1st one is the only one locked.
 874  */
 875 static int ocfs2_double_lock(struct ocfs2_super *osb,
 876                              struct buffer_head **bh1,
 877                              struct inode *inode1,
 878                              struct buffer_head **bh2,
 879                              struct inode *inode2)
 880 {
 881         int status;
 882         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 883         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 884         struct buffer_head **tmpbh;
 885         struct inode *tmpinode;
 886
 887         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 888                    (unsigned long long)oi1->ip_blkno,
 889                    (unsigned long long)oi2->ip_blkno);
 890
 891         if (*bh1)
 892                 *bh1 = NULL;
 893         if (*bh2)
 894                 *bh2 = NULL;
 895
 896         /* we always want to lock the one with the lower lockid first. */
 897         if (oi1->ip_blkno != oi2->ip_blkno) {
 898                 if (oi1->ip_blkno < oi2->ip_blkno) {
 899                         /* switch id1 and id2 around */
 900                         mlog(0, "switching them around...\n");
 901                         tmpbh = bh2;
 902                         bh2 = bh1;
 903                         bh1 = tmpbh;
 904
 905                         tmpinode = inode2;
 906                         inode2 = inode1;
 907                         inode1 = tmpinode;
 908                 }
 909                 /* lock id2 */
 910                 status = ocfs2_inode_lock(inode2, bh2, 1);
 911                 if (status < 0) {
 912                         if (status != -ENOENT)
 913                                 mlog_errno(status);
 914                         goto bail;
 915                 }
 916         }
 917
 918         /* lock id1 */
 919         status = ocfs2_inode_lock(inode1, bh1, 1);
 920         if (status < 0) {
 921                 /*
 922                  * An error return must mean that no cluster locks
 923                  * were held on function exit.
 924                  */
 925                 if (oi1->ip_blkno != oi2->ip_blkno)
 926                         ocfs2_inode_unlock(inode2, 1);
 927
 928                 if (status != -ENOENT)
 929                         mlog_errno(status);
 930         }
 931
 932 bail:
 933         mlog_exit(status);
 934         return status;
 935 }
 936
 937 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
 938 {
 939         ocfs2_inode_unlock(inode1, 1);
 940
 941         if (inode1 != inode2)
 942                 ocfs2_inode_unlock(inode2, 1);
 943 }
 944
 945 static int ocfs2_rename(struct inode *old_dir,
 946                         struct dentry *old_dentry,
 947                         struct inode *new_dir,
 948                         struct dentry *new_dentry)
 949 {
 950         int status = 0, rename_lock = 0, parents_locked = 0;
 951         int old_child_locked = 0, new_child_locked = 0;
 952         struct inode *old_inode = old_dentry->d_inode;
 953         struct inode *new_inode = new_dentry->d_inode;
 954         struct inode *orphan_dir = NULL;
 955         struct ocfs2_dinode *newfe = NULL;
 956         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 957         struct buffer_head *orphan_entry_bh = NULL;
 958         struct buffer_head *newfe_bh = NULL;
 959         struct buffer_head *old_inode_bh = NULL;
 960         struct buffer_head *insert_entry_bh = NULL;
 961         struct ocfs2_super *osb = NULL;
 962         u64 newfe_blkno, old_de_ino;
 963         handle_t *handle = NULL;
 964         struct buffer_head *old_dir_bh = NULL;
 965         struct buffer_head *new_dir_bh = NULL;
 966         struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
 967                 *new_de = NULL;
 968         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
 969         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
 970                                                     // this is the 1st dirent bh
 971         nlink_t old_dir_nlink = old_dir->i_nlink;
 972         struct ocfs2_dinode *old_di;
 973
 974         /* At some point it might be nice to break this function up a
 975          * bit. */
 976
 977         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
 978                    old_dir, old_dentry, new_dir, new_dentry,
 979                    old_dentry->d_name.len, old_dentry->d_name.name,
 980                    new_dentry->d_name.len, new_dentry->d_name.name);
 981
 982         osb = OCFS2_SB(old_dir->i_sb);
 983
 984         if (new_inode) {
 985                 if (!igrab(new_inode))
 986                         BUG();
 987         }
 988
 989         /* Assume a directory hierarchy thusly:
 990          * a/b/c
 991          * a/d
 992          * a,b,c, and d are all directories.
 993          *
 994          * from cwd of 'a' on both nodes:
 995          * node1: mv b/c d
 996          * node2: mv d   b/c
 997          *
 998          * And that's why, just like the VFS, we need a file system
 999          * rename lock. */
1000         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1001                 status = ocfs2_rename_lock(osb);
1002                 if (status < 0) {
1003                         mlog_errno(status);
1004                         goto bail;
1005                 }
1006                 rename_lock = 1;
1007         }
1008
1009         /* if old and new are the same, this'll just do one lock. */
1010         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1011                                    &new_dir_bh, new_dir);
1012         if (status < 0) {
1013                 mlog_errno(status);
1014                 goto bail;
1015         }
1016         parents_locked = 1;
1017
1018         /* make sure both dirs have bhs
1019          * get an extra ref on old_dir_bh if old==new */
1020         if (!new_dir_bh) {
1021                 if (old_dir_bh) {
1022                         new_dir_bh = old_dir_bh;
1023                         get_bh(new_dir_bh);
1024                 } else {
1025                         mlog(ML_ERROR, "no old_dir_bh!\n");
1026                         status = -EIO;
1027                         goto bail;
1028                 }
1029         }
1030
1031         /*
1032          * Aside from allowing a meta data update, the locking here
1033          * also ensures that the downconvert thread on other nodes
1034          * won't have to concurrently downconvert the inode and the
1035          * dentry locks.
1036          */
1037         status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1038         if (status < 0) {
1039                 if (status != -ENOENT)
1040                         mlog_errno(status);
1041                 goto bail;
1042         }
1043         old_child_locked = 1;
1044
1045         status = ocfs2_remote_dentry_delete(old_dentry);
1046         if (status < 0) {
1047                 mlog_errno(status);
1048                 goto bail;
1049         }
1050
1051         if (S_ISDIR(old_inode->i_mode)) {
1052                 u64 old_inode_parent;
1053
1054                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1055                                                   old_inode, &old_inode_de_bh,
1056                                                   &old_inode_dot_dot_de);
1057                 if (status) {
1058                         status = -EIO;
1059                         goto bail;
1060                 }
1061
1062                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1063                         status = -EIO;
1064                         goto bail;
1065                 }
1066
1067                 if (!new_inode && new_dir != old_dir &&
1068                     new_dir->i_nlink >= OCFS2_LINK_MAX) {
1069                         status = -EMLINK;
1070                         goto bail;
1071                 }
1072         }
1073
1074         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1075                                             old_dentry->d_name.len,
1076                                             &old_de_ino);
1077         if (status) {
1078                 status = -ENOENT;
1079                 goto bail;
1080         }
1081
1082         /*
1083          *  Check for inode number is _not_ due to possible IO errors.
1084          *  We might rmdir the source, keep it as pwd of some process
1085          *  and merrily kill the link to whatever was created under the
1086          *  same name. Goodbye sticky bit ;-<
1087          */
1088         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1089                 status = -ENOENT;
1090                 goto bail;
1091         }
1092
1093         /* check if the target already exists (in which case we need
1094          * to delete it */
1095         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1096                                           new_dentry->d_name.len,
1097                                           &newfe_blkno, new_dir, &new_de_bh,
1098                                           &new_de);
1099         /* The only error we allow here is -ENOENT because the new
1100          * file not existing is perfectly valid. */
1101         if ((status < 0) && (status != -ENOENT)) {
1102                 /* If we cannot find the file specified we should just */
1103                 /* return the error... */
1104                 mlog_errno(status);
1105                 goto bail;
1106         }
1107
1108         if (!new_de && new_inode) {
1109                 /*
1110                  * Target was unlinked by another node while we were
1111                  * waiting to get to ocfs2_rename(). There isn't
1112                  * anything we can do here to help the situation, so
1113                  * bubble up the appropriate error.
1114                  */
1115                 status = -ENOENT;
1116                 goto bail;
1117         }
1118
1119         /* In case we need to overwrite an existing file, we blow it
1120          * away first */
1121         if (new_de) {
1122                 /* VFS didn't think there existed an inode here, but
1123                  * someone else in the cluster must have raced our
1124                  * rename to create one. Today we error cleanly, in
1125                  * the future we should consider calling iget to build
1126                  * a new struct inode for this entry. */
1127                 if (!new_inode) {
1128                         status = -EACCES;
1129
1130                         mlog(0, "We found an inode for name %.*s but VFS "
1131                              "didn't give us one.\n", new_dentry->d_name.len,
1132                              new_dentry->d_name.name);
1133                         goto bail;
1134                 }
1135
1136                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1137                         status = -EACCES;
1138
1139                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1140                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1141                              (unsigned long long)newfe_blkno,
1142                              OCFS2_I(new_inode)->ip_flags);
1143                         goto bail;
1144                 }
1145
1146                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1147                 if (status < 0) {
1148                         if (status != -ENOENT)
1149                                 mlog_errno(status);
1150                         goto bail;
1151                 }
1152                 new_child_locked = 1;
1153
1154                 status = ocfs2_remote_dentry_delete(new_dentry);
1155                 if (status < 0) {
1156                         mlog_errno(status);
1157                         goto bail;
1158                 }
1159
1160                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1161
1162                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1163                      "newfebh=%p bhblocknr=%llu\n", new_de,
1164                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1165                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1166
1167                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1168                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1169                                                           new_inode,
1170                                                           orphan_name,
1171                                                           &orphan_entry_bh);
1172                         if (status < 0) {
1173                                 mlog_errno(status);
1174                                 goto bail;
1175                         }
1176                 }
1177         } else {
1178                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1179
1180                 status = ocfs2_check_dir_for_entry(new_dir,
1181                                                    new_dentry->d_name.name,
1182                                                    new_dentry->d_name.len);
1183                 if (status)
1184                         goto bail;
1185
1186                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1187                                                       new_dentry->d_name.name,
1188                                                       new_dentry->d_name.len,
1189                                                       &insert_entry_bh);
1190                 if (status < 0) {
1191                         mlog_errno(status);
1192                         goto bail;
1193                 }
1194         }
1195
1196         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1197         if (IS_ERR(handle)) {
1198                 status = PTR_ERR(handle);
1199                 handle = NULL;
1200                 mlog_errno(status);
1201                 goto bail;
1202         }
1203
1204         if (new_de) {
1205                 if (S_ISDIR(new_inode->i_mode)) {
1206                         if (!ocfs2_empty_dir(new_inode) ||
1207                             new_inode->i_nlink != 2) {
1208                                 status = -ENOTEMPTY;
1209                                 goto bail;
1210                         }
1211                 }
1212                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1213                                               OCFS2_JOURNAL_ACCESS_WRITE);
1214                 if (status < 0) {
1215                         mlog_errno(status);
1216                         goto bail;
1217                 }
1218
1219                 if (S_ISDIR(new_inode->i_mode) ||
1220                     (newfe->i_links_count == cpu_to_le16(1))){
1221                         status = ocfs2_orphan_add(osb, handle, new_inode,
1222                                                   newfe, orphan_name,
1223                                                   orphan_entry_bh, orphan_dir);
1224                         if (status < 0) {
1225                                 mlog_errno(status);
1226                                 goto bail;
1227                         }
1228                 }
1229
1230                 /* change the dirent to point to the correct inode */
1231                 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1232                                             new_de, old_inode);
1233                 if (status < 0) {
1234                         mlog_errno(status);
1235                         goto bail;
1236                 }
1237                 new_dir->i_version++;
1238
1239                 if (S_ISDIR(new_inode->i_mode))
1240                         newfe->i_links_count = 0;
1241                 else
1242                         le16_add_cpu(&newfe->i_links_count, -1);
1243
1244                 status = ocfs2_journal_dirty(handle, newfe_bh);
1245                 if (status < 0) {
1246                         mlog_errno(status);
1247                         goto bail;
1248                 }
1249         } else {
1250                 /* if the name was not found in new_dir, add it now */
1251                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1252                                          OCFS2_I(old_inode)->ip_blkno,
1253                                          new_dir_bh, insert_entry_bh);
1254         }
1255
1256         old_inode->i_ctime = CURRENT_TIME;
1257         mark_inode_dirty(old_inode);
1258
1259         status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1260                                       OCFS2_JOURNAL_ACCESS_WRITE);
1261         if (status >= 0) {
1262                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1263
1264                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1265                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1266
1267                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1268                 if (status < 0)
1269                         mlog_errno(status);
1270         } else
1271                 mlog_errno(status);
1272
1273         /*
1274          * Now that the name has been added to new_dir, remove the old name.
1275          *
1276          * We don't keep any directory entry context around until now
1277          * because the insert might have changed the type of directory
1278          * we're dealing with.
1279          */
1280         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1281                                      old_dentry->d_name.len,
1282                                      old_dir, &old_de);
1283         if (!old_de_bh) {
1284                 status = -EIO;
1285                 goto bail;
1286         }
1287
1288         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1289         if (status < 0) {
1290                 mlog_errno(status);
1291                 goto bail;
1292         }
1293
1294         if (new_inode) {
1295                 new_inode->i_nlink--;
1296                 new_inode->i_ctime = CURRENT_TIME;
1297         }
1298         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1299         if (old_inode_de_bh) {
1300                 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1301                                             old_inode_dot_dot_de, new_dir);
1302                 old_dir->i_nlink--;
1303                 if (new_inode) {
1304                         new_inode->i_nlink--;
1305                 } else {
1306                         inc_nlink(new_dir);
1307                         mark_inode_dirty(new_dir);
1308                 }
1309         }
1310         mark_inode_dirty(old_dir);
1311         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1312         if (new_inode) {
1313                 mark_inode_dirty(new_inode);
1314                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1315         }
1316
1317         if (old_dir != new_dir) {
1318                 /* Keep the same times on both directories.*/
1319                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1320
1321                 /*
1322                  * This will also pick up the i_nlink change from the
1323                  * block above.
1324                  */
1325                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1326         }
1327
1328         if (old_dir_nlink != old_dir->i_nlink) {
1329                 if (!old_dir_bh) {
1330                         mlog(ML_ERROR, "need to change nlink for old dir "
1331                              "%llu from %d to %d but bh is NULL!\n",
1332                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1333                              (int)old_dir_nlink, old_dir->i_nlink);
1334                 } else {
1335                         struct ocfs2_dinode *fe;
1336                         status = ocfs2_journal_access(handle, old_dir,
1337                                                       old_dir_bh,
1338                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1339                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1340                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1341                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1342                 }
1343         }
1344
1345         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1346         status = 0;
1347 bail:
1348         if (rename_lock)
1349                 ocfs2_rename_unlock(osb);
1350
1351         if (handle)
1352                 ocfs2_commit_trans(osb, handle);
1353
1354         if (parents_locked)
1355                 ocfs2_double_unlock(old_dir, new_dir);
1356
1357         if (old_child_locked)
1358                 ocfs2_inode_unlock(old_inode, 1);
1359
1360         if (new_child_locked)
1361                 ocfs2_inode_unlock(new_inode, 1);
1362
1363         if (orphan_dir) {
1364                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1365                 ocfs2_inode_unlock(orphan_dir, 1);
1366                 mutex_unlock(&orphan_dir->i_mutex);
1367                 iput(orphan_dir);
1368         }
1369
1370         if (new_inode)
1371                 sync_mapping_buffers(old_inode->i_mapping);
1372
1373         if (new_inode)
1374                 iput(new_inode);
1375         if (newfe_bh)
1376                 brelse(newfe_bh);
1377         if (old_inode_bh)
1378                 brelse(old_inode_bh);
1379         if (old_dir_bh)
1380                 brelse(old_dir_bh);
1381         if (new_dir_bh)
1382                 brelse(new_dir_bh);
1383         if (new_de_bh)
1384                 brelse(new_de_bh);
1385         if (old_de_bh)
1386                 brelse(old_de_bh);
1387         if (old_inode_de_bh)
1388                 brelse(old_inode_de_bh);
1389         if (orphan_entry_bh)
1390                 brelse(orphan_entry_bh);
1391         if (insert_entry_bh)
1392                 brelse(insert_entry_bh);
1393
1394         mlog_exit(status);
1395
1396         return status;
1397 }
1398
1399 /*
1400  * we expect i_size = strlen(symname). Copy symname into the file
1401  * data, including the null terminator.
1402  */
1403 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1404                                      handle_t *handle,
1405                                      struct inode *inode,
1406                                      const char *symname)
1407 {
1408         struct buffer_head **bhs = NULL;
1409         const char *c;
1410         struct super_block *sb = osb->sb;
1411         u64 p_blkno, p_blocks;
1412         int virtual, blocks, status, i, bytes_left;
1413
1414         bytes_left = i_size_read(inode) + 1;
1415         /* we can't trust i_blocks because we're actually going to
1416          * write i_size + 1 bytes. */
1417         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1418
1419         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1420                         (unsigned long long)inode->i_blocks,
1421                         i_size_read(inode), blocks);
1422
1423         /* Sanity check -- make sure we're going to fit. */
1424         if (bytes_left >
1425             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1426                 status = -EIO;
1427                 mlog_errno(status);
1428                 goto bail;
1429         }
1430
1431         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1432         if (!bhs) {
1433                 status = -ENOMEM;
1434                 mlog_errno(status);
1435                 goto bail;
1436         }
1437
1438         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1439                                              NULL);
1440         if (status < 0) {
1441                 mlog_errno(status);
1442                 goto bail;
1443         }
1444
1445         /* links can never be larger than one cluster so we know this
1446          * is all going to be contiguous, but do a sanity check
1447          * anyway. */
1448         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1449                 status = -EIO;
1450                 mlog_errno(status);
1451                 goto bail;
1452         }
1453
1454         virtual = 0;
1455         while(bytes_left > 0) {
1456                 c = &symname[virtual * sb->s_blocksize];
1457
1458                 bhs[virtual] = sb_getblk(sb, p_blkno);
1459                 if (!bhs[virtual]) {
1460                         status = -ENOMEM;
1461                         mlog_errno(status);
1462                         goto bail;
1463                 }
1464                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1465
1466                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1467                                               OCFS2_JOURNAL_ACCESS_CREATE);
1468                 if (status < 0) {
1469                         mlog_errno(status);
1470                         goto bail;
1471                 }
1472
1473                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1474
1475                 memcpy(bhs[virtual]->b_data, c,
1476                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1477                        bytes_left);
1478
1479                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1480                 if (status < 0) {
1481                         mlog_errno(status);
1482                         goto bail;
1483                 }
1484
1485                 virtual++;
1486                 p_blkno++;
1487                 bytes_left -= sb->s_blocksize;
1488         }
1489
1490         status = 0;
1491 bail:
1492
1493         if (bhs) {
1494                 for(i = 0; i < blocks; i++)
1495                         if (bhs[i])
1496                                 brelse(bhs[i]);
1497                 kfree(bhs);
1498         }
1499
1500         mlog_exit(status);
1501         return status;
1502 }
1503
1504 static int ocfs2_symlink(struct inode *dir,
1505                          struct dentry *dentry,
1506                          const char *symname)
1507 {
1508         int status, l, credits;
1509         u64 newsize;
1510         struct ocfs2_super *osb = NULL;
1511         struct inode *inode = NULL;
1512         struct super_block *sb;
1513         struct buffer_head *new_fe_bh = NULL;
1514         struct buffer_head *de_bh = NULL;
1515         struct buffer_head *parent_fe_bh = NULL;
1516         struct ocfs2_dinode *fe = NULL;
1517         struct ocfs2_dinode *dirfe;
1518         handle_t *handle = NULL;
1519         struct ocfs2_alloc_context *inode_ac = NULL;
1520         struct ocfs2_alloc_context *data_ac = NULL;
1521
1522         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1523                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1524
1525         sb = dir->i_sb;
1526         osb = OCFS2_SB(sb);
1527
1528         l = strlen(symname) + 1;
1529
1530         credits = ocfs2_calc_symlink_credits(sb);
1531
1532         /* lock the parent directory */
1533         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1534         if (status < 0) {
1535                 if (status != -ENOENT)
1536                         mlog_errno(status);
1537                 return status;
1538         }
1539
1540         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1541         if (!dirfe->i_links_count) {
1542                 /* can't make a file in a deleted directory. */
1543                 status = -ENOENT;
1544                 goto bail;
1545         }
1546
1547         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1548                                            dentry->d_name.len);
1549         if (status)
1550                 goto bail;
1551
1552         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1553                                               dentry->d_name.name,
1554                                               dentry->d_name.len, &de_bh);
1555         if (status < 0) {
1556                 mlog_errno(status);
1557                 goto bail;
1558         }
1559
1560         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1561         if (status < 0) {
1562                 if (status != -ENOSPC)
1563                         mlog_errno(status);
1564                 goto bail;
1565         }
1566
1567         /* don't reserve bitmap space for fast symlinks. */
1568         if (l > ocfs2_fast_symlink_chars(sb)) {
1569                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1570                 if (status < 0) {
1571                         if (status != -ENOSPC)
1572                                 mlog_errno(status);
1573                         goto bail;
1574                 }
1575         }
1576
1577         handle = ocfs2_start_trans(osb, credits);
1578         if (IS_ERR(handle)) {
1579                 status = PTR_ERR(handle);
1580                 handle = NULL;
1581                 mlog_errno(status);
1582                 goto bail;
1583         }
1584
1585         status = ocfs2_mknod_locked(osb, dir, dentry,
1586                                     S_IFLNK | S_IRWXUGO, 0,
1587                                     &new_fe_bh, parent_fe_bh, handle,
1588                                     &inode, inode_ac);
1589         if (status < 0) {
1590                 mlog_errno(status);
1591                 goto bail;
1592         }
1593
1594         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1595         inode->i_rdev = 0;
1596         newsize = l - 1;
1597         if (l > ocfs2_fast_symlink_chars(sb)) {
1598                 u32 offset = 0;
1599
1600                 inode->i_op = &ocfs2_symlink_inode_operations;
1601                 status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
1602                                                     new_fe_bh,
1603                                                     handle, data_ac, NULL,
1604                                                     NULL);
1605                 if (status < 0) {
1606                         if (status != -ENOSPC && status != -EINTR) {
1607                                 mlog(ML_ERROR,
1608                                      "Failed to extend file to %llu\n",
1609                                      (unsigned long long)newsize);
1610                                 mlog_errno(status);
1611                                 status = -ENOSPC;
1612                         }
1613                         goto bail;
1614                 }
1615                 i_size_write(inode, newsize);
1616                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1617         } else {
1618                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1619                 memcpy((char *) fe->id2.i_symlink, symname, l);
1620                 i_size_write(inode, newsize);
1621                 inode->i_blocks = 0;
1622         }
1623
1624         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1625         if (status < 0) {
1626                 mlog_errno(status);
1627                 goto bail;
1628         }
1629
1630         if (!ocfs2_inode_is_fast_symlink(inode)) {
1631                 status = ocfs2_create_symlink_data(osb, handle, inode,
1632                                                    symname);
1633                 if (status < 0) {
1634                         mlog_errno(status);
1635                         goto bail;
1636                 }
1637         }
1638
1639         status = ocfs2_add_entry(handle, dentry, inode,
1640                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1641                                  de_bh);
1642         if (status < 0) {
1643                 mlog_errno(status);
1644                 goto bail;
1645         }
1646
1647         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1648         if (status) {
1649                 mlog_errno(status);
1650                 goto bail;
1651         }
1652
1653         insert_inode_hash(inode);
1654         dentry->d_op = &ocfs2_dentry_ops;
1655         d_instantiate(dentry, inode);
1656 bail:
1657         if (handle)
1658                 ocfs2_commit_trans(osb, handle);
1659
1660         ocfs2_inode_unlock(dir, 1);
1661
1662         if (new_fe_bh)
1663                 brelse(new_fe_bh);
1664         if (parent_fe_bh)
1665                 brelse(parent_fe_bh);
1666         if (de_bh)
1667                 brelse(de_bh);
1668         if (inode_ac)
1669                 ocfs2_free_alloc_context(inode_ac);
1670         if (data_ac)
1671                 ocfs2_free_alloc_context(data_ac);
1672         if ((status < 0) && inode)
1673                 iput(inode);
1674
1675         mlog_exit(status);
1676
1677         return status;
1678 }
1679
1680 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1681 {
1682         int status, namelen;
1683
1684         mlog_entry_void();
1685
1686         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1687                            (long long)blkno);
1688         if (namelen <= 0) {
1689                 if (namelen)
1690                         status = namelen;
1691                 else
1692                         status = -EINVAL;
1693                 mlog_errno(status);
1694                 goto bail;
1695         }
1696         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1697                 status = -EINVAL;
1698                 mlog_errno(status);
1699                 goto bail;
1700         }
1701
1702         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1703              namelen);
1704
1705         status = 0;
1706 bail:
1707         mlog_exit(status);
1708         return status;
1709 }
1710
1711 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1712                                     struct inode **ret_orphan_dir,
1713                                     struct inode *inode,
1714                                     char *name,
1715                                     struct buffer_head **de_bh)
1716 {
1717         struct inode *orphan_dir_inode;
1718         struct buffer_head *orphan_dir_bh = NULL;
1719         int status = 0;
1720
1721         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1722         if (status < 0) {
1723                 mlog_errno(status);
1724                 return status;
1725         }
1726
1727         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1728                                                        ORPHAN_DIR_SYSTEM_INODE,
1729                                                        osb->slot_num);
1730         if (!orphan_dir_inode) {
1731                 status = -ENOENT;
1732                 mlog_errno(status);
1733                 return status;
1734         }
1735
1736         mutex_lock(&orphan_dir_inode->i_mutex);
1737
1738         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1739         if (status < 0) {
1740                 mlog_errno(status);
1741                 goto leave;
1742         }
1743
1744         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1745                                               orphan_dir_bh, name,
1746                                               OCFS2_ORPHAN_NAMELEN, de_bh);
1747         if (status < 0) {
1748                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1749
1750                 mlog_errno(status);
1751                 goto leave;
1752         }
1753
1754         *ret_orphan_dir = orphan_dir_inode;
1755
1756 leave:
1757         if (status) {
1758                 mutex_unlock(&orphan_dir_inode->i_mutex);
1759                 iput(orphan_dir_inode);
1760         }
1761
1762         if (orphan_dir_bh)
1763                 brelse(orphan_dir_bh);
1764
1765         mlog_exit(status);
1766         return status;
1767 }
1768
1769 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1770                             handle_t *handle,
1771                             struct inode *inode,
1772                             struct ocfs2_dinode *fe,
1773                             char *name,
1774                             struct buffer_head *de_bh,
1775                             struct inode *orphan_dir_inode)
1776 {
1777         struct buffer_head *orphan_dir_bh = NULL;
1778         int status = 0;
1779         struct ocfs2_dinode *orphan_fe;
1780
1781         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1782
1783         status = ocfs2_read_block(osb,
1784                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
1785                                   &orphan_dir_bh, OCFS2_BH_CACHED,
1786                                   orphan_dir_inode);
1787         if (status < 0) {
1788                 mlog_errno(status);
1789                 goto leave;
1790         }
1791
1792         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1793                                       OCFS2_JOURNAL_ACCESS_WRITE);
1794         if (status < 0) {
1795                 mlog_errno(status);
1796                 goto leave;
1797         }
1798
1799         /* we're a cluster, and nlink can change on disk from
1800          * underneath us... */
1801         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1802         if (S_ISDIR(inode->i_mode))
1803                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1804         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1805
1806         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1807         if (status < 0) {
1808                 mlog_errno(status);
1809                 goto leave;
1810         }
1811
1812         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1813                                    OCFS2_ORPHAN_NAMELEN, inode,
1814                                    OCFS2_I(inode)->ip_blkno,
1815                                    orphan_dir_bh, de_bh);
1816         if (status < 0) {
1817                 mlog_errno(status);
1818                 goto leave;
1819         }
1820
1821         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1822
1823         /* Record which orphan dir our inode now resides
1824          * in. delete_inode will use this to determine which orphan
1825          * dir to lock. */
1826         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1827
1828         mlog(0, "Inode %llu orphaned in slot %d\n",
1829              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1830
1831 leave:
1832         if (orphan_dir_bh)
1833                 brelse(orphan_dir_bh);
1834
1835         mlog_exit(status);
1836         return status;
1837 }
1838
1839 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1840 int ocfs2_orphan_del(struct ocfs2_super *osb,
1841                      handle_t *handle,
1842                      struct inode *orphan_dir_inode,
1843                      struct inode *inode,
1844                      struct buffer_head *orphan_dir_bh)
1845 {
1846         char name[OCFS2_ORPHAN_NAMELEN + 1];
1847         struct ocfs2_dinode *orphan_fe;
1848         int status = 0;
1849         struct buffer_head *target_de_bh = NULL;
1850         struct ocfs2_dir_entry *target_de = NULL;
1851
1852         mlog_entry_void();
1853
1854         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1855         if (status < 0) {
1856                 mlog_errno(status);
1857                 goto leave;
1858         }
1859
1860         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1861              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1862              OCFS2_ORPHAN_NAMELEN);
1863
1864         /* find it's spot in the orphan directory */
1865         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1866                                         orphan_dir_inode, &target_de);
1867         if (!target_de_bh) {
1868                 status = -ENOENT;
1869                 mlog_errno(status);
1870                 goto leave;
1871         }
1872
1873         /* remove it from the orphan directory */
1874         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1875                                     target_de_bh);
1876         if (status < 0) {
1877                 mlog_errno(status);
1878                 goto leave;
1879         }
1880
1881         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
1882                                       OCFS2_JOURNAL_ACCESS_WRITE);
1883         if (status < 0) {
1884                 mlog_errno(status);
1885                 goto leave;
1886         }
1887
1888         /* do the i_nlink dance! :) */
1889         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1890         if (S_ISDIR(inode->i_mode))
1891                 le16_add_cpu(&orphan_fe->i_links_count, -1);
1892         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1893
1894         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1895         if (status < 0) {
1896                 mlog_errno(status);
1897                 goto leave;
1898         }
1899
1900 leave:
1901         if (target_de_bh)
1902                 brelse(target_de_bh);
1903
1904         mlog_exit(status);
1905         return status;
1906 }
1907
1908 const struct inode_operations ocfs2_dir_iops = {
1909         .create         = ocfs2_create,
1910         .lookup         = ocfs2_lookup,
1911         .link           = ocfs2_link,
1912         .unlink         = ocfs2_unlink,
1913         .rmdir          = ocfs2_unlink,
1914         .symlink        = ocfs2_symlink,
1915         .mkdir          = ocfs2_mkdir,
1916         .mknod          = ocfs2_mknod,
1917         .rename         = ocfs2_rename,
1918         .setattr        = ocfs2_setattr,
1919         .getattr        = ocfs2_getattr,
1920         .permission     = ocfs2_permission,
1921 };