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

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43
  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 "vote.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 buffer_head *dirent_bh = NULL;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_dir_entry *dirent;
 108         struct ocfs2_inode_info *oi;
 109
 110         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 111                    dentry->d_name.len, dentry->d_name.name);
 112
 113         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 114                 ret = ERR_PTR(-ENAMETOOLONG);
 115                 goto bail;
 116         }
 117
 118         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 119              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 120
 121         status = ocfs2_meta_lock(dir, NULL, 0);
 122         if (status < 0) {
 123                 if (status != -ENOENT)
 124                         mlog_errno(status);
 125                 ret = ERR_PTR(status);
 126                 goto bail;
 127         }
 128
 129         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 130                                           dentry->d_name.len, &blkno,
 131                                           dir, &dirent_bh, &dirent);
 132         if (status < 0)
 133                 goto bail_add;
 134
 135         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 136         if (IS_ERR(inode)) {
 137                 ret = ERR_PTR(-EACCES);
 138                 goto bail_unlock;
 139         }
 140
 141         oi = OCFS2_I(inode);
 142         /* Clear any orphaned state... If we were able to look up the
 143          * inode from a directory, it certainly can't be orphaned. We
 144          * might have the bad state from a node which intended to
 145          * orphan this inode but crashed before it could commit the
 146          * unlink. */
 147         spin_lock(&oi->ip_lock);
 148         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 149         spin_unlock(&oi->ip_lock);
 150
 151 bail_add:
 152         dentry->d_op = &ocfs2_dentry_ops;
 153         ret = d_splice_alias(inode, dentry);
 154
 155         if (inode) {
 156                 /*
 157                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 158                  * dentry, it will d_move() it on top of ourse. The
 159                  * return value will indicate this however, so in
 160                  * those cases, we switch them around for the locking
 161                  * code.
 162                  *
 163                  * NOTE: This dentry already has ->d_op set from
 164                  * ocfs2_get_parent() and ocfs2_get_dentry()
 165                  */
 166                 if (ret)
 167                         dentry = ret;
 168
 169                 status = ocfs2_dentry_attach_lock(dentry, inode,
 170                                                   OCFS2_I(dir)->ip_blkno);
 171                 if (status) {
 172                         mlog_errno(status);
 173                         ret = ERR_PTR(status);
 174                         goto bail_unlock;
 175                 }
 176         }
 177
 178 bail_unlock:
 179         /* Don't drop the cluster lock until *after* the d_add --
 180          * unlink on another node will message us to remove that
 181          * dentry under this lock so otherwise we can race this with
 182          * the vote thread and have a stale dentry. */
 183         ocfs2_meta_unlock(dir, 0);
 184
 185 bail:
 186         if (dirent_bh)
 187                 brelse(dirent_bh);
 188
 189         mlog_exit_ptr(ret);
 190
 191         return ret;
 192 }
 193
 194 static int ocfs2_mknod(struct inode *dir,
 195                        struct dentry *dentry,
 196                        int mode,
 197                        dev_t dev)
 198 {
 199         int status = 0;
 200         struct buffer_head *parent_fe_bh = NULL;
 201         handle_t *handle = NULL;
 202         struct ocfs2_super *osb;
 203         struct ocfs2_dinode *dirfe;
 204         struct buffer_head *new_fe_bh = NULL;
 205         struct buffer_head *de_bh = NULL;
 206         struct inode *inode = NULL;
 207         struct ocfs2_alloc_context *inode_ac = NULL;
 208         struct ocfs2_alloc_context *data_ac = NULL;
 209
 210         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 211                    (unsigned long)dev, dentry->d_name.len,
 212                    dentry->d_name.name);
 213
 214         /* get our super block */
 215         osb = OCFS2_SB(dir->i_sb);
 216
 217         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 218         if (status < 0) {
 219                 if (status != -ENOENT)
 220                         mlog_errno(status);
 221                 return status;
 222         }
 223
 224         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 225                 status = -EMLINK;
 226                 goto leave;
 227         }
 228
 229         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 230         if (!dirfe->i_links_count) {
 231                 /* can't make a file in a deleted directory. */
 232                 status = -ENOENT;
 233                 goto leave;
 234         }
 235
 236         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 237                                            dentry->d_name.len);
 238         if (status)
 239                 goto leave;
 240
 241         /* get a spot inside the dir. */
 242         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 243                                               dentry->d_name.name,
 244                                               dentry->d_name.len, &de_bh);
 245         if (status < 0) {
 246                 mlog_errno(status);
 247                 goto leave;
 248         }
 249
 250         /* reserve an inode spot */
 251         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 252         if (status < 0) {
 253                 if (status != -ENOSPC)
 254                         mlog_errno(status);
 255                 goto leave;
 256         }
 257
 258         /* are we making a directory? If so, reserve a cluster for his
 259          * 1st extent. */
 260         if (S_ISDIR(mode)) {
 261                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 262                 if (status < 0) {
 263                         if (status != -ENOSPC)
 264                                 mlog_errno(status);
 265                         goto leave;
 266                 }
 267         }
 268
 269         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
 270         if (IS_ERR(handle)) {
 271                 status = PTR_ERR(handle);
 272                 handle = NULL;
 273                 mlog_errno(status);
 274                 goto leave;
 275         }
 276
 277         /* do the real work now. */
 278         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 279                                     &new_fe_bh, parent_fe_bh, handle,
 280                                     &inode, inode_ac);
 281         if (status < 0) {
 282                 mlog_errno(status);
 283                 goto leave;
 284         }
 285
 286         if (S_ISDIR(mode)) {
 287                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 288                                             new_fe_bh, data_ac);
 289                 if (status < 0) {
 290                         mlog_errno(status);
 291                         goto leave;
 292                 }
 293
 294                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 295                                               OCFS2_JOURNAL_ACCESS_WRITE);
 296                 if (status < 0) {
 297                         mlog_errno(status);
 298                         goto leave;
 299                 }
 300                 le16_add_cpu(&dirfe->i_links_count, 1);
 301                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 302                 if (status < 0) {
 303                         mlog_errno(status);
 304                         goto leave;
 305                 }
 306                 inc_nlink(dir);
 307         }
 308
 309         status = ocfs2_add_entry(handle, dentry, inode,
 310                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 311                                  de_bh);
 312         if (status < 0) {
 313                 mlog_errno(status);
 314                 goto leave;
 315         }
 316
 317         status = ocfs2_dentry_attach_lock(dentry, inode,
 318                                           OCFS2_I(dir)->ip_blkno);
 319         if (status) {
 320                 mlog_errno(status);
 321                 goto leave;
 322         }
 323
 324         insert_inode_hash(inode);
 325         dentry->d_op = &ocfs2_dentry_ops;
 326         d_instantiate(dentry, inode);
 327         status = 0;
 328 leave:
 329         if (handle)
 330                 ocfs2_commit_trans(osb, handle);
 331
 332         ocfs2_meta_unlock(dir, 1);
 333
 334         if (status == -ENOSPC)
 335                 mlog(0, "Disk is full\n");
 336
 337         if (new_fe_bh)
 338                 brelse(new_fe_bh);
 339
 340         if (de_bh)
 341                 brelse(de_bh);
 342
 343         if (parent_fe_bh)
 344                 brelse(parent_fe_bh);
 345
 346         if ((status < 0) && inode)
 347                 iput(inode);
 348
 349         if (inode_ac)
 350                 ocfs2_free_alloc_context(inode_ac);
 351
 352         if (data_ac)
 353                 ocfs2_free_alloc_context(data_ac);
 354
 355         mlog_exit(status);
 356
 357         return status;
 358 }
 359
 360 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 361                               struct inode *dir,
 362                               struct dentry *dentry, int mode,
 363                               dev_t dev,
 364                               struct buffer_head **new_fe_bh,
 365                               struct buffer_head *parent_fe_bh,
 366                               handle_t *handle,
 367                               struct inode **ret_inode,
 368                               struct ocfs2_alloc_context *inode_ac)
 369 {
 370         int status = 0;
 371         struct ocfs2_dinode *fe = NULL;
 372         struct ocfs2_extent_list *fel;
 373         u64 fe_blkno = 0;
 374         u16 suballoc_bit;
 375         struct inode *inode = NULL;
 376
 377         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 378                    (unsigned long)dev, dentry->d_name.len,
 379                    dentry->d_name.name);
 380
 381         *new_fe_bh = NULL;
 382         *ret_inode = NULL;
 383
 384         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 385                                        &fe_blkno);
 386         if (status < 0) {
 387                 mlog_errno(status);
 388                 goto leave;
 389         }
 390
 391         inode = new_inode(dir->i_sb);
 392         if (IS_ERR(inode)) {
 393                 status = PTR_ERR(inode);
 394                 mlog(ML_ERROR, "new_inode failed!\n");
 395                 goto leave;
 396         }
 397
 398         /* populate as many fields early on as possible - many of
 399          * these are used by the support functions here and in
 400          * callers. */
 401         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 402         OCFS2_I(inode)->ip_blkno = fe_blkno;
 403         if (S_ISDIR(mode))
 404                 inode->i_nlink = 2;
 405         else
 406                 inode->i_nlink = 1;
 407         inode->i_mode = mode;
 408         spin_lock(&osb->osb_lock);
 409         inode->i_generation = osb->s_next_generation++;
 410         spin_unlock(&osb->osb_lock);
 411
 412         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 413         if (!*new_fe_bh) {
 414                 status = -EIO;
 415                 mlog_errno(status);
 416                 goto leave;
 417         }
 418         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 419
 420         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 421                                       OCFS2_JOURNAL_ACCESS_CREATE);
 422         if (status < 0) {
 423                 mlog_errno(status);
 424                 goto leave;
 425         }
 426
 427         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 428         memset(fe, 0, osb->sb->s_blocksize);
 429
 430         fe->i_generation = cpu_to_le32(inode->i_generation);
 431         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 432         fe->i_blkno = cpu_to_le64(fe_blkno);
 433         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 434         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 435         fe->i_uid = cpu_to_le32(current->fsuid);
 436         if (dir->i_mode & S_ISGID) {
 437                 fe->i_gid = cpu_to_le32(dir->i_gid);
 438                 if (S_ISDIR(mode))
 439                         mode |= S_ISGID;
 440         } else
 441                 fe->i_gid = cpu_to_le32(current->fsgid);
 442         fe->i_mode = cpu_to_le16(mode);
 443         if (S_ISCHR(mode) || S_ISBLK(mode))
 444                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 445
 446         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 447
 448         fe->i_last_eb_blk = 0;
 449         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 450         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 451         fe->i_atime = fe->i_ctime = fe->i_mtime =
 452                 cpu_to_le64(CURRENT_TIME.tv_sec);
 453         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 454                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 455         fe->i_dtime = 0;
 456
 457         fel = &fe->id2.i_list;
 458         fel->l_tree_depth = 0;
 459         fel->l_next_free_rec = 0;
 460         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 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_meta_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_meta_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_meta_unlock(inode, 1);
 642
 643 out:
 644         ocfs2_meta_unlock(dir, 1);
 645
 646         if (de_bh)
 647                 brelse(de_bh);
 648         if (fe_bh)
 649                 brelse(fe_bh);
 650         if (parent_fe_bh)
 651                 brelse(parent_fe_bh);
 652
 653         mlog_exit(err);
 654
 655         return err;
 656 }
 657
 658 /*
 659  * Takes and drops an exclusive lock on the given dentry. This will
 660  * force other nodes to drop it.
 661  */
 662 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 663 {
 664         int ret;
 665
 666         ret = ocfs2_dentry_lock(dentry, 1);
 667         if (ret)
 668                 mlog_errno(ret);
 669         else
 670                 ocfs2_dentry_unlock(dentry, 1);
 671
 672         return ret;
 673 }
 674
 675 static inline int inode_is_unlinkable(struct inode *inode)
 676 {
 677         if (S_ISDIR(inode->i_mode)) {
 678                 if (inode->i_nlink == 2)
 679                         return 1;
 680                 return 0;
 681         }
 682
 683         if (inode->i_nlink == 1)
 684                 return 1;
 685         return 0;
 686 }
 687
 688 static int ocfs2_unlink(struct inode *dir,
 689                         struct dentry *dentry)
 690 {
 691         int status;
 692         int child_locked = 0;
 693         struct inode *inode = dentry->d_inode;
 694         struct inode *orphan_dir = NULL;
 695         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 696         u64 blkno;
 697         struct ocfs2_dinode *fe = NULL;
 698         struct buffer_head *fe_bh = NULL;
 699         struct buffer_head *parent_node_bh = NULL;
 700         handle_t *handle = NULL;
 701         struct ocfs2_dir_entry *dirent = NULL;
 702         struct buffer_head *dirent_bh = NULL;
 703         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 704         struct buffer_head *orphan_entry_bh = NULL;
 705
 706         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 707                    dentry->d_name.len, dentry->d_name.name);
 708
 709         BUG_ON(dentry->d_parent->d_inode != dir);
 710
 711         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 712
 713         if (inode == osb->root_inode) {
 714                 mlog(0, "Cannot delete the root directory\n");
 715                 return -EPERM;
 716         }
 717
 718         status = ocfs2_meta_lock(dir, &parent_node_bh, 1);
 719         if (status < 0) {
 720                 if (status != -ENOENT)
 721                         mlog_errno(status);
 722                 return status;
 723         }
 724
 725         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 726                                           dentry->d_name.len, &blkno,
 727                                           dir, &dirent_bh, &dirent);
 728         if (status < 0) {
 729                 if (status != -ENOENT)
 730                         mlog_errno(status);
 731                 goto leave;
 732         }
 733
 734         if (OCFS2_I(inode)->ip_blkno != blkno) {
 735                 status = -ENOENT;
 736
 737                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 738                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 739                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 740                 goto leave;
 741         }
 742
 743         status = ocfs2_meta_lock(inode, &fe_bh, 1);
 744         if (status < 0) {
 745                 if (status != -ENOENT)
 746                         mlog_errno(status);
 747                 goto leave;
 748         }
 749         child_locked = 1;
 750
 751         if (S_ISDIR(inode->i_mode)) {
 752                 if (!ocfs2_empty_dir(inode)) {
 753                         status = -ENOTEMPTY;
 754                         goto leave;
 755                 } else if (inode->i_nlink != 2) {
 756                         status = -ENOTEMPTY;
 757                         goto leave;
 758                 }
 759         }
 760
 761         status = ocfs2_remote_dentry_delete(dentry);
 762         if (status < 0) {
 763                 /* This vote should succeed under all normal
 764                  * circumstances. */
 765                 mlog_errno(status);
 766                 goto leave;
 767         }
 768
 769         if (inode_is_unlinkable(inode)) {
 770                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 771                                                   orphan_name,
 772                                                   &orphan_entry_bh);
 773                 if (status < 0) {
 774                         mlog_errno(status);
 775                         goto leave;
 776                 }
 777         }
 778
 779         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
 780         if (IS_ERR(handle)) {
 781                 status = PTR_ERR(handle);
 782                 handle = NULL;
 783                 mlog_errno(status);
 784                 goto leave;
 785         }
 786
 787         status = ocfs2_journal_access(handle, inode, fe_bh,
 788                                       OCFS2_JOURNAL_ACCESS_WRITE);
 789         if (status < 0) {
 790                 mlog_errno(status);
 791                 goto leave;
 792         }
 793
 794         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 795
 796         if (inode_is_unlinkable(inode)) {
 797                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 798                                           orphan_entry_bh, orphan_dir);
 799                 if (status < 0) {
 800                         mlog_errno(status);
 801                         goto leave;
 802                 }
 803         }
 804
 805         /* delete the name from the parent dir */
 806         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 807         if (status < 0) {
 808                 mlog_errno(status);
 809                 goto leave;
 810         }
 811
 812         if (S_ISDIR(inode->i_mode))
 813                 drop_nlink(inode);
 814         drop_nlink(inode);
 815         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 816
 817         status = ocfs2_journal_dirty(handle, fe_bh);
 818         if (status < 0) {
 819                 mlog_errno(status);
 820                 goto leave;
 821         }
 822
 823         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 824         if (S_ISDIR(inode->i_mode))
 825                 drop_nlink(dir);
 826
 827         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 828         if (status < 0) {
 829                 mlog_errno(status);
 830                 if (S_ISDIR(inode->i_mode))
 831                         inc_nlink(dir);
 832         }
 833
 834 leave:
 835         if (handle)
 836                 ocfs2_commit_trans(osb, handle);
 837
 838         if (child_locked)
 839                 ocfs2_meta_unlock(inode, 1);
 840
 841         ocfs2_meta_unlock(dir, 1);
 842
 843         if (orphan_dir) {
 844                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 845                 ocfs2_meta_unlock(orphan_dir, 1);
 846                 mutex_unlock(&orphan_dir->i_mutex);
 847                 iput(orphan_dir);
 848         }
 849
 850         if (fe_bh)
 851                 brelse(fe_bh);
 852
 853         if (dirent_bh)
 854                 brelse(dirent_bh);
 855
 856         if (parent_node_bh)
 857                 brelse(parent_node_bh);
 858
 859         if (orphan_entry_bh)
 860                 brelse(orphan_entry_bh);
 861
 862         mlog_exit(status);
 863
 864         return status;
 865 }
 866
 867 /*
 868  * The only place this should be used is rename!
 869  * if they have the same id, then the 1st one is the only one locked.
 870  */
 871 static int ocfs2_double_lock(struct ocfs2_super *osb,
 872                              struct buffer_head **bh1,
 873                              struct inode *inode1,
 874                              struct buffer_head **bh2,
 875                              struct inode *inode2)
 876 {
 877         int status;
 878         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 879         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 880         struct buffer_head **tmpbh;
 881         struct inode *tmpinode;
 882
 883         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 884                    (unsigned long long)oi1->ip_blkno,
 885                    (unsigned long long)oi2->ip_blkno);
 886
 887         if (*bh1)
 888                 *bh1 = NULL;
 889         if (*bh2)
 890                 *bh2 = NULL;
 891
 892         /* we always want to lock the one with the lower lockid first. */
 893         if (oi1->ip_blkno != oi2->ip_blkno) {
 894                 if (oi1->ip_blkno < oi2->ip_blkno) {
 895                         /* switch id1 and id2 around */
 896                         mlog(0, "switching them around...\n");
 897                         tmpbh = bh2;
 898                         bh2 = bh1;
 899                         bh1 = tmpbh;
 900
 901                         tmpinode = inode2;
 902                         inode2 = inode1;
 903                         inode1 = tmpinode;
 904                 }
 905                 /* lock id2 */
 906                 status = ocfs2_meta_lock(inode2, bh2, 1);
 907                 if (status < 0) {
 908                         if (status != -ENOENT)
 909                                 mlog_errno(status);
 910                         goto bail;
 911                 }
 912         }
 913
 914         /* lock id1 */
 915         status = ocfs2_meta_lock(inode1, bh1, 1);
 916         if (status < 0) {
 917                 /*
 918                  * An error return must mean that no cluster locks
 919                  * were held on function exit.
 920                  */
 921                 if (oi1->ip_blkno != oi2->ip_blkno)
 922                         ocfs2_meta_unlock(inode2, 1);
 923
 924                 if (status != -ENOENT)
 925                         mlog_errno(status);
 926         }
 927
 928 bail:
 929         mlog_exit(status);
 930         return status;
 931 }
 932
 933 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
 934 {
 935         ocfs2_meta_unlock(inode1, 1);
 936
 937         if (inode1 != inode2)
 938                 ocfs2_meta_unlock(inode2, 1);
 939 }
 940
 941 #define PARENT_INO(buffer) \
 942         ((struct ocfs2_dir_entry *) \
 943          ((char *)buffer + \
 944           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
 945
 946 static int ocfs2_rename(struct inode *old_dir,
 947                         struct dentry *old_dentry,
 948                         struct inode *new_dir,
 949                         struct dentry *new_dentry)
 950 {
 951         int status = 0, rename_lock = 0, parents_locked = 0;
 952         int old_child_locked = 0, new_child_locked = 0;
 953         struct inode *old_inode = old_dentry->d_inode;
 954         struct inode *new_inode = new_dentry->d_inode;
 955         struct inode *orphan_dir = NULL;
 956         struct ocfs2_dinode *newfe = NULL;
 957         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 958         struct buffer_head *orphan_entry_bh = NULL;
 959         struct buffer_head *newfe_bh = NULL;
 960         struct buffer_head *old_inode_bh = NULL;
 961         struct buffer_head *insert_entry_bh = NULL;
 962         struct ocfs2_super *osb = NULL;
 963         u64 newfe_blkno;
 964         handle_t *handle = NULL;
 965         struct buffer_head *old_dir_bh = NULL;
 966         struct buffer_head *new_dir_bh = NULL;
 967         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
 968                                                                // and new_dentry
 969         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
 970         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
 971                                                     // this is the 1st dirent bh
 972         nlink_t old_dir_nlink = old_dir->i_nlink;
 973         struct ocfs2_dinode *old_di;
 974
 975         /* At some point it might be nice to break this function up a
 976          * bit. */
 977
 978         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
 979                    old_dir, old_dentry, new_dir, new_dentry,
 980                    old_dentry->d_name.len, old_dentry->d_name.name,
 981                    new_dentry->d_name.len, new_dentry->d_name.name);
 982
 983         osb = OCFS2_SB(old_dir->i_sb);
 984
 985         if (new_inode) {
 986                 if (!igrab(new_inode))
 987                         BUG();
 988         }
 989
 990         /* Assume a directory hierarchy thusly:
 991          * a/b/c
 992          * a/d
 993          * a,b,c, and d are all directories.
 994          *
 995          * from cwd of 'a' on both nodes:
 996          * node1: mv b/c d
 997          * node2: mv d   b/c
 998          *
 999          * And that's why, just like the VFS, we need a file system
1000          * rename lock. */
1001         if (old_dentry != new_dentry) {
1002                 status = ocfs2_rename_lock(osb);
1003                 if (status < 0) {
1004                         mlog_errno(status);
1005                         goto bail;
1006                 }
1007                 rename_lock = 1;
1008         }
1009
1010         /* if old and new are the same, this'll just do one lock. */
1011         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1012                                    &new_dir_bh, new_dir);
1013         if (status < 0) {
1014                 mlog_errno(status);
1015                 goto bail;
1016         }
1017         parents_locked = 1;
1018
1019         /* make sure both dirs have bhs
1020          * get an extra ref on old_dir_bh if old==new */
1021         if (!new_dir_bh) {
1022                 if (old_dir_bh) {
1023                         new_dir_bh = old_dir_bh;
1024                         get_bh(new_dir_bh);
1025                 } else {
1026                         mlog(ML_ERROR, "no old_dir_bh!\n");
1027                         status = -EIO;
1028                         goto bail;
1029                 }
1030         }
1031
1032         /*
1033          * Aside from allowing a meta data update, the locking here
1034          * also ensures that the vote thread on other nodes won't have
1035          * to concurrently downconvert the inode and the dentry locks.
1036          */
1037         status = ocfs2_meta_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                 status = -EIO;
1053                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1054                 if (!old_inode_de_bh)
1055                         goto bail;
1056
1057                 status = -EIO;
1058                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1059                     OCFS2_I(old_dir)->ip_blkno)
1060                         goto bail;
1061                 status = -EMLINK;
1062                 if (!new_inode && new_dir!=old_dir &&
1063                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1064                         goto bail;
1065         }
1066
1067         status = -ENOENT;
1068         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1069                                      old_dentry->d_name.len,
1070                                      old_dir, &old_de);
1071         if (!old_de_bh)
1072                 goto bail;
1073
1074         /*
1075          *  Check for inode number is _not_ due to possible IO errors.
1076          *  We might rmdir the source, keep it as pwd of some process
1077          *  and merrily kill the link to whatever was created under the
1078          *  same name. Goodbye sticky bit ;-<
1079          */
1080         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1081                 goto bail;
1082
1083         /* check if the target already exists (in which case we need
1084          * to delete it */
1085         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1086                                           new_dentry->d_name.len,
1087                                           &newfe_blkno, new_dir, &new_de_bh,
1088                                           &new_de);
1089         /* The only error we allow here is -ENOENT because the new
1090          * file not existing is perfectly valid. */
1091         if ((status < 0) && (status != -ENOENT)) {
1092                 /* If we cannot find the file specified we should just */
1093                 /* return the error... */
1094                 mlog_errno(status);
1095                 goto bail;
1096         }
1097
1098         if (!new_de && new_inode)
1099                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1100                      "directory!", new_inode->i_ino);
1101
1102         /* In case we need to overwrite an existing file, we blow it
1103          * away first */
1104         if (new_de) {
1105                 /* VFS didn't think there existed an inode here, but
1106                  * someone else in the cluster must have raced our
1107                  * rename to create one. Today we error cleanly, in
1108                  * the future we should consider calling iget to build
1109                  * a new struct inode for this entry. */
1110                 if (!new_inode) {
1111                         status = -EACCES;
1112
1113                         mlog(0, "We found an inode for name %.*s but VFS "
1114                              "didn't give us one.\n", new_dentry->d_name.len,
1115                              new_dentry->d_name.name);
1116                         goto bail;
1117                 }
1118
1119                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1120                         status = -EACCES;
1121
1122                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1123                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1124                              (unsigned long long)newfe_blkno,
1125                              OCFS2_I(new_inode)->ip_flags);
1126                         goto bail;
1127                 }
1128
1129                 status = ocfs2_meta_lock(new_inode, &newfe_bh, 1);
1130                 if (status < 0) {
1131                         if (status != -ENOENT)
1132                                 mlog_errno(status);
1133                         goto bail;
1134                 }
1135                 new_child_locked = 1;
1136
1137                 status = ocfs2_remote_dentry_delete(new_dentry);
1138                 if (status < 0) {
1139                         mlog_errno(status);
1140                         goto bail;
1141                 }
1142
1143                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1144
1145                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1146                      "newfebh=%p bhblocknr=%llu\n", new_de,
1147                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1148                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1149
1150                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1151                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1152                                                           new_inode,
1153                                                           orphan_name,
1154                                                           &orphan_entry_bh);
1155                         if (status < 0) {
1156                                 mlog_errno(status);
1157                                 goto bail;
1158                         }
1159                 }
1160         } else {
1161                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1162
1163                 status = ocfs2_check_dir_for_entry(new_dir,
1164                                                    new_dentry->d_name.name,
1165                                                    new_dentry->d_name.len);
1166                 if (status)
1167                         goto bail;
1168
1169                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1170                                                       new_dentry->d_name.name,
1171                                                       new_dentry->d_name.len,
1172                                                       &insert_entry_bh);
1173                 if (status < 0) {
1174                         mlog_errno(status);
1175                         goto bail;
1176                 }
1177         }
1178
1179         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1180         if (IS_ERR(handle)) {
1181                 status = PTR_ERR(handle);
1182                 handle = NULL;
1183                 mlog_errno(status);
1184                 goto bail;
1185         }
1186
1187         if (new_de) {
1188                 if (S_ISDIR(new_inode->i_mode)) {
1189                         if (!ocfs2_empty_dir(new_inode) ||
1190                             new_inode->i_nlink != 2) {
1191                                 status = -ENOTEMPTY;
1192                                 goto bail;
1193                         }
1194                 }
1195                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1196                                               OCFS2_JOURNAL_ACCESS_WRITE);
1197                 if (status < 0) {
1198                         mlog_errno(status);
1199                         goto bail;
1200                 }
1201
1202                 if (S_ISDIR(new_inode->i_mode) ||
1203                     (newfe->i_links_count == cpu_to_le16(1))){
1204                         status = ocfs2_orphan_add(osb, handle, new_inode,
1205                                                   newfe, orphan_name,
1206                                                   orphan_entry_bh, orphan_dir);
1207                         if (status < 0) {
1208                                 mlog_errno(status);
1209                                 goto bail;
1210                         }
1211                 }
1212
1213                 /* change the dirent to point to the correct inode */
1214                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1215                                               OCFS2_JOURNAL_ACCESS_WRITE);
1216                 if (status < 0) {
1217                         mlog_errno(status);
1218                         goto bail;
1219                 }
1220                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1221                 new_de->file_type = old_de->file_type;
1222                 new_dir->i_version++;
1223                 status = ocfs2_journal_dirty(handle, new_de_bh);
1224                 if (status < 0) {
1225                         mlog_errno(status);
1226                         goto bail;
1227                 }
1228
1229                 if (S_ISDIR(new_inode->i_mode))
1230                         newfe->i_links_count = 0;
1231                 else
1232                         le16_add_cpu(&newfe->i_links_count, -1);
1233
1234                 status = ocfs2_journal_dirty(handle, newfe_bh);
1235                 if (status < 0) {
1236                         mlog_errno(status);
1237                         goto bail;
1238                 }
1239         } else {
1240                 /* if the name was not found in new_dir, add it now */
1241                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1242                                          OCFS2_I(old_inode)->ip_blkno,
1243                                          new_dir_bh, insert_entry_bh);
1244         }
1245
1246         old_inode->i_ctime = CURRENT_TIME;
1247         mark_inode_dirty(old_inode);
1248
1249         status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1250                                       OCFS2_JOURNAL_ACCESS_WRITE);
1251         if (status >= 0) {
1252                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1253
1254                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1255                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1256
1257                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1258                 if (status < 0)
1259                         mlog_errno(status);
1260         } else
1261                 mlog_errno(status);
1262
1263         /* now that the name has been added to new_dir, remove the old name */
1264         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1265         if (status < 0) {
1266                 mlog_errno(status);
1267                 goto bail;
1268         }
1269
1270         if (new_inode) {
1271                 new_inode->i_nlink--;
1272                 new_inode->i_ctime = CURRENT_TIME;
1273         }
1274         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1275         if (old_inode_de_bh) {
1276                 status = ocfs2_journal_access(handle, old_inode,
1277                                              old_inode_de_bh,
1278                                              OCFS2_JOURNAL_ACCESS_WRITE);
1279                 PARENT_INO(old_inode_de_bh->b_data) =
1280                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1281                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1282                 old_dir->i_nlink--;
1283                 if (new_inode) {
1284                         new_inode->i_nlink--;
1285                 } else {
1286                         inc_nlink(new_dir);
1287                         mark_inode_dirty(new_dir);
1288                 }
1289         }
1290         mark_inode_dirty(old_dir);
1291         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1292         if (new_inode) {
1293                 mark_inode_dirty(new_inode);
1294                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1295         }
1296
1297         if (old_dir != new_dir) {
1298                 /* Keep the same times on both directories.*/
1299                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1300
1301                 /*
1302                  * This will also pick up the i_nlink change from the
1303                  * block above.
1304                  */
1305                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1306         }
1307
1308         if (old_dir_nlink != old_dir->i_nlink) {
1309                 if (!old_dir_bh) {
1310                         mlog(ML_ERROR, "need to change nlink for old dir "
1311                              "%llu from %d to %d but bh is NULL!\n",
1312                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1313                              (int)old_dir_nlink, old_dir->i_nlink);
1314                 } else {
1315                         struct ocfs2_dinode *fe;
1316                         status = ocfs2_journal_access(handle, old_dir,
1317                                                       old_dir_bh,
1318                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1319                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1320                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1321                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1322                 }
1323         }
1324
1325         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1326         status = 0;
1327 bail:
1328         if (rename_lock)
1329                 ocfs2_rename_unlock(osb);
1330
1331         if (handle)
1332                 ocfs2_commit_trans(osb, handle);
1333
1334         if (parents_locked)
1335                 ocfs2_double_unlock(old_dir, new_dir);
1336
1337         if (old_child_locked)
1338                 ocfs2_meta_unlock(old_inode, 1);
1339
1340         if (new_child_locked)
1341                 ocfs2_meta_unlock(new_inode, 1);
1342
1343         if (orphan_dir) {
1344                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1345                 ocfs2_meta_unlock(orphan_dir, 1);
1346                 mutex_unlock(&orphan_dir->i_mutex);
1347                 iput(orphan_dir);
1348         }
1349
1350         if (new_inode)
1351                 sync_mapping_buffers(old_inode->i_mapping);
1352
1353         if (new_inode)
1354                 iput(new_inode);
1355         if (newfe_bh)
1356                 brelse(newfe_bh);
1357         if (old_inode_bh)
1358                 brelse(old_inode_bh);
1359         if (old_dir_bh)
1360                 brelse(old_dir_bh);
1361         if (new_dir_bh)
1362                 brelse(new_dir_bh);
1363         if (new_de_bh)
1364                 brelse(new_de_bh);
1365         if (old_de_bh)
1366                 brelse(old_de_bh);
1367         if (old_inode_de_bh)
1368                 brelse(old_inode_de_bh);
1369         if (orphan_entry_bh)
1370                 brelse(orphan_entry_bh);
1371         if (insert_entry_bh)
1372                 brelse(insert_entry_bh);
1373
1374         mlog_exit(status);
1375
1376         return status;
1377 }
1378
1379 /*
1380  * we expect i_size = strlen(symname). Copy symname into the file
1381  * data, including the null terminator.
1382  */
1383 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1384                                      handle_t *handle,
1385                                      struct inode *inode,
1386                                      const char *symname)
1387 {
1388         struct buffer_head **bhs = NULL;
1389         const char *c;
1390         struct super_block *sb = osb->sb;
1391         u64 p_blkno, p_blocks;
1392         int virtual, blocks, status, i, bytes_left;
1393
1394         bytes_left = i_size_read(inode) + 1;
1395         /* we can't trust i_blocks because we're actually going to
1396          * write i_size + 1 bytes. */
1397         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1398
1399         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1400                         (unsigned long long)inode->i_blocks,
1401                         i_size_read(inode), blocks);
1402
1403         /* Sanity check -- make sure we're going to fit. */
1404         if (bytes_left >
1405             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1406                 status = -EIO;
1407                 mlog_errno(status);
1408                 goto bail;
1409         }
1410
1411         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1412         if (!bhs) {
1413                 status = -ENOMEM;
1414                 mlog_errno(status);
1415                 goto bail;
1416         }
1417
1418         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1419                                              NULL);
1420         if (status < 0) {
1421                 mlog_errno(status);
1422                 goto bail;
1423         }
1424
1425         /* links can never be larger than one cluster so we know this
1426          * is all going to be contiguous, but do a sanity check
1427          * anyway. */
1428         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1429                 status = -EIO;
1430                 mlog_errno(status);
1431                 goto bail;
1432         }
1433
1434         virtual = 0;
1435         while(bytes_left > 0) {
1436                 c = &symname[virtual * sb->s_blocksize];
1437
1438                 bhs[virtual] = sb_getblk(sb, p_blkno);
1439                 if (!bhs[virtual]) {
1440                         status = -ENOMEM;
1441                         mlog_errno(status);
1442                         goto bail;
1443                 }
1444                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1445
1446                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1447                                               OCFS2_JOURNAL_ACCESS_CREATE);
1448                 if (status < 0) {
1449                         mlog_errno(status);
1450                         goto bail;
1451                 }
1452
1453                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1454
1455                 memcpy(bhs[virtual]->b_data, c,
1456                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1457                        bytes_left);
1458
1459                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1460                 if (status < 0) {
1461                         mlog_errno(status);
1462                         goto bail;
1463                 }
1464
1465                 virtual++;
1466                 p_blkno++;
1467                 bytes_left -= sb->s_blocksize;
1468         }
1469
1470         status = 0;
1471 bail:
1472
1473         if (bhs) {
1474                 for(i = 0; i < blocks; i++)
1475                         if (bhs[i])
1476                                 brelse(bhs[i]);
1477                 kfree(bhs);
1478         }
1479
1480         mlog_exit(status);
1481         return status;
1482 }
1483
1484 static int ocfs2_symlink(struct inode *dir,
1485                          struct dentry *dentry,
1486                          const char *symname)
1487 {
1488         int status, l, credits;
1489         u64 newsize;
1490         struct ocfs2_super *osb = NULL;
1491         struct inode *inode = NULL;
1492         struct super_block *sb;
1493         struct buffer_head *new_fe_bh = NULL;
1494         struct buffer_head *de_bh = NULL;
1495         struct buffer_head *parent_fe_bh = NULL;
1496         struct ocfs2_dinode *fe = NULL;
1497         struct ocfs2_dinode *dirfe;
1498         handle_t *handle = NULL;
1499         struct ocfs2_alloc_context *inode_ac = NULL;
1500         struct ocfs2_alloc_context *data_ac = NULL;
1501
1502         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1503                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1504
1505         sb = dir->i_sb;
1506         osb = OCFS2_SB(sb);
1507
1508         l = strlen(symname) + 1;
1509
1510         credits = ocfs2_calc_symlink_credits(sb);
1511
1512         /* lock the parent directory */
1513         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
1514         if (status < 0) {
1515                 if (status != -ENOENT)
1516                         mlog_errno(status);
1517                 return status;
1518         }
1519
1520         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1521         if (!dirfe->i_links_count) {
1522                 /* can't make a file in a deleted directory. */
1523                 status = -ENOENT;
1524                 goto bail;
1525         }
1526
1527         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1528                                            dentry->d_name.len);
1529         if (status)
1530                 goto bail;
1531
1532         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1533                                               dentry->d_name.name,
1534                                               dentry->d_name.len, &de_bh);
1535         if (status < 0) {
1536                 mlog_errno(status);
1537                 goto bail;
1538         }
1539
1540         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1541         if (status < 0) {
1542                 if (status != -ENOSPC)
1543                         mlog_errno(status);
1544                 goto bail;
1545         }
1546
1547         /* don't reserve bitmap space for fast symlinks. */
1548         if (l > ocfs2_fast_symlink_chars(sb)) {
1549                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1550                 if (status < 0) {
1551                         if (status != -ENOSPC)
1552                                 mlog_errno(status);
1553                         goto bail;
1554                 }
1555         }
1556
1557         handle = ocfs2_start_trans(osb, credits);
1558         if (IS_ERR(handle)) {
1559                 status = PTR_ERR(handle);
1560                 handle = NULL;
1561                 mlog_errno(status);
1562                 goto bail;
1563         }
1564
1565         status = ocfs2_mknod_locked(osb, dir, dentry,
1566                                     S_IFLNK | S_IRWXUGO, 0,
1567                                     &new_fe_bh, parent_fe_bh, handle,
1568                                     &inode, inode_ac);
1569         if (status < 0) {
1570                 mlog_errno(status);
1571                 goto bail;
1572         }
1573
1574         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1575         inode->i_rdev = 0;
1576         newsize = l - 1;
1577         if (l > ocfs2_fast_symlink_chars(sb)) {
1578                 u32 offset = 0;
1579
1580                 inode->i_op = &ocfs2_symlink_inode_operations;
1581                 status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
1582                                                     new_fe_bh,
1583                                                     handle, data_ac, NULL,
1584                                                     NULL);
1585                 if (status < 0) {
1586                         if (status != -ENOSPC && status != -EINTR) {
1587                                 mlog(ML_ERROR,
1588                                      "Failed to extend file to %llu\n",
1589                                      (unsigned long long)newsize);
1590                                 mlog_errno(status);
1591                                 status = -ENOSPC;
1592                         }
1593                         goto bail;
1594                 }
1595                 i_size_write(inode, newsize);
1596                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1597         } else {
1598                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1599                 memcpy((char *) fe->id2.i_symlink, symname, l);
1600                 i_size_write(inode, newsize);
1601                 inode->i_blocks = 0;
1602         }
1603
1604         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1605         if (status < 0) {
1606                 mlog_errno(status);
1607                 goto bail;
1608         }
1609
1610         if (!ocfs2_inode_is_fast_symlink(inode)) {
1611                 status = ocfs2_create_symlink_data(osb, handle, inode,
1612                                                    symname);
1613                 if (status < 0) {
1614                         mlog_errno(status);
1615                         goto bail;
1616                 }
1617         }
1618
1619         status = ocfs2_add_entry(handle, dentry, inode,
1620                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1621                                  de_bh);
1622         if (status < 0) {
1623                 mlog_errno(status);
1624                 goto bail;
1625         }
1626
1627         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1628         if (status) {
1629                 mlog_errno(status);
1630                 goto bail;
1631         }
1632
1633         insert_inode_hash(inode);
1634         dentry->d_op = &ocfs2_dentry_ops;
1635         d_instantiate(dentry, inode);
1636 bail:
1637         if (handle)
1638                 ocfs2_commit_trans(osb, handle);
1639
1640         ocfs2_meta_unlock(dir, 1);
1641
1642         if (new_fe_bh)
1643                 brelse(new_fe_bh);
1644         if (parent_fe_bh)
1645                 brelse(parent_fe_bh);
1646         if (de_bh)
1647                 brelse(de_bh);
1648         if (inode_ac)
1649                 ocfs2_free_alloc_context(inode_ac);
1650         if (data_ac)
1651                 ocfs2_free_alloc_context(data_ac);
1652         if ((status < 0) && inode)
1653                 iput(inode);
1654
1655         mlog_exit(status);
1656
1657         return status;
1658 }
1659
1660 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1661 {
1662         int status, namelen;
1663
1664         mlog_entry_void();
1665
1666         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1667                            (long long)blkno);
1668         if (namelen <= 0) {
1669                 if (namelen)
1670                         status = namelen;
1671                 else
1672                         status = -EINVAL;
1673                 mlog_errno(status);
1674                 goto bail;
1675         }
1676         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1677                 status = -EINVAL;
1678                 mlog_errno(status);
1679                 goto bail;
1680         }
1681
1682         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1683              namelen);
1684
1685         status = 0;
1686 bail:
1687         mlog_exit(status);
1688         return status;
1689 }
1690
1691 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1692                                     struct inode **ret_orphan_dir,
1693                                     struct inode *inode,
1694                                     char *name,
1695                                     struct buffer_head **de_bh)
1696 {
1697         struct inode *orphan_dir_inode;
1698         struct buffer_head *orphan_dir_bh = NULL;
1699         int status = 0;
1700
1701         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1702         if (status < 0) {
1703                 mlog_errno(status);
1704                 return status;
1705         }
1706
1707         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1708                                                        ORPHAN_DIR_SYSTEM_INODE,
1709                                                        osb->slot_num);
1710         if (!orphan_dir_inode) {
1711                 status = -ENOENT;
1712                 mlog_errno(status);
1713                 return status;
1714         }
1715
1716         mutex_lock(&orphan_dir_inode->i_mutex);
1717
1718         status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1719         if (status < 0) {
1720                 mlog_errno(status);
1721                 goto leave;
1722         }
1723
1724         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1725                                               orphan_dir_bh, name,
1726                                               OCFS2_ORPHAN_NAMELEN, de_bh);
1727         if (status < 0) {
1728                 ocfs2_meta_unlock(orphan_dir_inode, 1);
1729
1730                 mlog_errno(status);
1731                 goto leave;
1732         }
1733
1734         *ret_orphan_dir = orphan_dir_inode;
1735
1736 leave:
1737         if (status) {
1738                 mutex_unlock(&orphan_dir_inode->i_mutex);
1739                 iput(orphan_dir_inode);
1740         }
1741
1742         if (orphan_dir_bh)
1743                 brelse(orphan_dir_bh);
1744
1745         mlog_exit(status);
1746         return status;
1747 }
1748
1749 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1750                             handle_t *handle,
1751                             struct inode *inode,
1752                             struct ocfs2_dinode *fe,
1753                             char *name,
1754                             struct buffer_head *de_bh,
1755                             struct inode *orphan_dir_inode)
1756 {
1757         struct buffer_head *orphan_dir_bh = NULL;
1758         int status = 0;
1759         struct ocfs2_dinode *orphan_fe;
1760
1761         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1762
1763         status = ocfs2_read_block(osb,
1764                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
1765                                   &orphan_dir_bh, OCFS2_BH_CACHED,
1766                                   orphan_dir_inode);
1767         if (status < 0) {
1768                 mlog_errno(status);
1769                 goto leave;
1770         }
1771
1772         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1773                                       OCFS2_JOURNAL_ACCESS_WRITE);
1774         if (status < 0) {
1775                 mlog_errno(status);
1776                 goto leave;
1777         }
1778
1779         /* we're a cluster, and nlink can change on disk from
1780          * underneath us... */
1781         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1782         if (S_ISDIR(inode->i_mode))
1783                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1784         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1785
1786         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1787         if (status < 0) {
1788                 mlog_errno(status);
1789                 goto leave;
1790         }
1791
1792         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1793                                    OCFS2_ORPHAN_NAMELEN, inode,
1794                                    OCFS2_I(inode)->ip_blkno,
1795                                    orphan_dir_bh, de_bh);
1796         if (status < 0) {
1797                 mlog_errno(status);
1798                 goto leave;
1799         }
1800
1801         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1802
1803         /* Record which orphan dir our inode now resides
1804          * in. delete_inode will use this to determine which orphan
1805          * dir to lock. */
1806         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1807
1808         mlog(0, "Inode %llu orphaned in slot %d\n",
1809              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1810
1811 leave:
1812         if (orphan_dir_bh)
1813                 brelse(orphan_dir_bh);
1814
1815         mlog_exit(status);
1816         return status;
1817 }
1818
1819 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1820 int ocfs2_orphan_del(struct ocfs2_super *osb,
1821                      handle_t *handle,
1822                      struct inode *orphan_dir_inode,
1823                      struct inode *inode,
1824                      struct buffer_head *orphan_dir_bh)
1825 {
1826         char name[OCFS2_ORPHAN_NAMELEN + 1];
1827         struct ocfs2_dinode *orphan_fe;
1828         int status = 0;
1829         struct buffer_head *target_de_bh = NULL;
1830         struct ocfs2_dir_entry *target_de = NULL;
1831
1832         mlog_entry_void();
1833
1834         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1835         if (status < 0) {
1836                 mlog_errno(status);
1837                 goto leave;
1838         }
1839
1840         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1841              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1842              OCFS2_ORPHAN_NAMELEN);
1843
1844         /* find it's spot in the orphan directory */
1845         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1846                                         orphan_dir_inode, &target_de);
1847         if (!target_de_bh) {
1848                 status = -ENOENT;
1849                 mlog_errno(status);
1850                 goto leave;
1851         }
1852
1853         /* remove it from the orphan directory */
1854         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1855                                     target_de_bh);
1856         if (status < 0) {
1857                 mlog_errno(status);
1858                 goto leave;
1859         }
1860
1861         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
1862                                       OCFS2_JOURNAL_ACCESS_WRITE);
1863         if (status < 0) {
1864                 mlog_errno(status);
1865                 goto leave;
1866         }
1867
1868         /* do the i_nlink dance! :) */
1869         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1870         if (S_ISDIR(inode->i_mode))
1871                 le16_add_cpu(&orphan_fe->i_links_count, -1);
1872         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1873
1874         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1875         if (status < 0) {
1876                 mlog_errno(status);
1877                 goto leave;
1878         }
1879
1880 leave:
1881         if (target_de_bh)
1882                 brelse(target_de_bh);
1883
1884         mlog_exit(status);
1885         return status;
1886 }
1887
1888 const struct inode_operations ocfs2_dir_iops = {
1889         .create         = ocfs2_create,
1890         .lookup         = ocfs2_lookup,
1891         .link           = ocfs2_link,
1892         .unlink         = ocfs2_unlink,
1893         .rmdir          = ocfs2_unlink,
1894         .symlink        = ocfs2_symlink,
1895         .mkdir          = ocfs2_mkdir,
1896         .mknod          = ocfs2_mknod,
1897         .rename         = ocfs2_rename,
1898         .setattr        = ocfs2_setattr,
1899         .getattr        = ocfs2_getattr,
1900         .permission     = ocfs2_permission,
1901 };