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 #define NAMEI_RA_CHUNKS  2
  68 #define NAMEI_RA_BLOCKS  4
  69 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  70 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  71
  72 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
  73                                         struct inode *dir,
  74                                         const char *name, int namelen,
  75                                         unsigned long offset,
  76                                         struct ocfs2_dir_entry **res_dir);
  77
  78 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
  79                               struct inode *dir,
  80                               struct ocfs2_dir_entry *de_del,
  81                               struct buffer_head *bh);
  82
  83 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
  84                              struct inode *dir,
  85                              const char *name, int namelen,
  86                              struct inode *inode, u64 blkno,
  87                              struct buffer_head *parent_fe_bh,
  88                              struct buffer_head *insert_bh);
  89
  90 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  91                               struct inode *dir,
  92                               struct dentry *dentry, int mode,
  93                               dev_t dev,
  94                               struct buffer_head **new_fe_bh,
  95                               struct buffer_head *parent_fe_bh,
  96                               struct ocfs2_journal_handle *handle,
  97                               struct inode **ret_inode,
  98                               struct ocfs2_alloc_context *inode_ac);
  99
 100 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 101                               struct ocfs2_journal_handle *handle,
 102                               struct inode *parent,
 103                               struct inode *inode,
 104                               struct buffer_head *fe_bh,
 105                               struct ocfs2_alloc_context *data_ac);
 106
 107 static int ocfs2_double_lock(struct ocfs2_super *osb,
 108                              struct ocfs2_journal_handle *handle,
 109                              struct buffer_head **bh1,
 110                              struct inode *inode1,
 111                              struct buffer_head **bh2,
 112                              struct inode *inode2);
 113
 114 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 115                                     struct inode **ret_orphan_dir,
 116                                     struct inode *inode,
 117                                     char *name,
 118                                     struct buffer_head **de_bh);
 119
 120 static int ocfs2_orphan_add(struct ocfs2_super *osb,
 121                             struct ocfs2_journal_handle *handle,
 122                             struct inode *inode,
 123                             struct ocfs2_dinode *fe,
 124                             char *name,
 125                             struct buffer_head *de_bh,
 126                             struct inode *orphan_dir_inode);
 127
 128 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 129                                      struct ocfs2_journal_handle *handle,
 130                                      struct inode *inode,
 131                                      const char *symname);
 132
 133 static inline int ocfs2_add_entry(struct ocfs2_journal_handle *handle,
 134                                   struct dentry *dentry,
 135                                   struct inode *inode, u64 blkno,
 136                                   struct buffer_head *parent_fe_bh,
 137                                   struct buffer_head *insert_bh)
 138 {
 139         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
 140                                  dentry->d_name.name, dentry->d_name.len,
 141                                  inode, blkno, parent_fe_bh, insert_bh);
 142 }
 143
 144 /* An orphan dir name is an 8 byte value, printed as a hex string */
 145 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 146
 147 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 148                                    struct nameidata *nd)
 149 {
 150         int status;
 151         u64 blkno;
 152         struct buffer_head *dirent_bh = NULL;
 153         struct inode *inode = NULL;
 154         struct dentry *ret;
 155         struct ocfs2_dir_entry *dirent;
 156         struct ocfs2_inode_info *oi;
 157
 158         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 159                    dentry->d_name.len, dentry->d_name.name);
 160
 161         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 162                 ret = ERR_PTR(-ENAMETOOLONG);
 163                 goto bail;
 164         }
 165
 166         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 167              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 168
 169         status = ocfs2_meta_lock(dir, NULL, NULL, 0);
 170         if (status < 0) {
 171                 if (status != -ENOENT)
 172                         mlog_errno(status);
 173                 ret = ERR_PTR(status);
 174                 goto bail;
 175         }
 176
 177         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 178                                           dentry->d_name.len, &blkno,
 179                                           dir, &dirent_bh, &dirent);
 180         if (status < 0)
 181                 goto bail_add;
 182
 183         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 184         if (IS_ERR(inode)) {
 185                 mlog(ML_ERROR, "Unable to create inode %llu\n",
 186                      (unsigned long long)blkno);
 187                 ret = ERR_PTR(-EACCES);
 188                 goto bail_unlock;
 189         }
 190
 191         oi = OCFS2_I(inode);
 192         /* Clear any orphaned state... If we were able to look up the
 193          * inode from a directory, it certainly can't be orphaned. We
 194          * might have the bad state from a node which intended to
 195          * orphan this inode but crashed before it could commit the
 196          * unlink. */
 197         spin_lock(&oi->ip_lock);
 198         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 199         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 200         spin_unlock(&oi->ip_lock);
 201
 202 bail_add:
 203         dentry->d_op = &ocfs2_dentry_ops;
 204         ret = d_splice_alias(inode, dentry);
 205
 206         if (inode) {
 207                 /*
 208                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 209                  * dentry, it will d_move() it on top of ourse. The
 210                  * return value will indicate this however, so in
 211                  * those cases, we switch them around for the locking
 212                  * code.
 213                  *
 214                  * NOTE: This dentry already has ->d_op set from
 215                  * ocfs2_get_parent() and ocfs2_get_dentry()
 216                  */
 217                 if (ret)
 218                         dentry = ret;
 219
 220                 status = ocfs2_dentry_attach_lock(dentry, inode,
 221                                                   OCFS2_I(dir)->ip_blkno);
 222                 if (status) {
 223                         mlog_errno(status);
 224                         ret = ERR_PTR(status);
 225                         goto bail_unlock;
 226                 }
 227         }
 228
 229 bail_unlock:
 230         /* Don't drop the cluster lock until *after* the d_add --
 231          * unlink on another node will message us to remove that
 232          * dentry under this lock so otherwise we can race this with
 233          * the vote thread and have a stale dentry. */
 234         ocfs2_meta_unlock(dir, 0);
 235
 236 bail:
 237         if (dirent_bh)
 238                 brelse(dirent_bh);
 239
 240         mlog_exit_ptr(ret);
 241
 242         return ret;
 243 }
 244
 245 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 246                               struct ocfs2_journal_handle *handle,
 247                               struct inode *parent,
 248                               struct inode *inode,
 249                               struct buffer_head *fe_bh,
 250                               struct ocfs2_alloc_context *data_ac)
 251 {
 252         int status;
 253         struct buffer_head *new_bh = NULL;
 254         struct ocfs2_dir_entry *de = NULL;
 255
 256         mlog_entry_void();
 257
 258         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 259                                      data_ac, NULL, &new_bh);
 260         if (status < 0) {
 261                 mlog_errno(status);
 262                 goto bail;
 263         }
 264
 265         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 266
 267         status = ocfs2_journal_access(handle, inode, new_bh,
 268                                       OCFS2_JOURNAL_ACCESS_CREATE);
 269         if (status < 0) {
 270                 mlog_errno(status);
 271                 goto bail;
 272         }
 273         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 274
 275         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 276         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 277         de->name_len = 1;
 278         de->rec_len =
 279                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 280         strcpy(de->name, ".");
 281         ocfs2_set_de_type(de, S_IFDIR);
 282         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 283         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 284         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 285                                   OCFS2_DIR_REC_LEN(1));
 286         de->name_len = 2;
 287         strcpy(de->name, "..");
 288         ocfs2_set_de_type(de, S_IFDIR);
 289
 290         status = ocfs2_journal_dirty(handle, new_bh);
 291         if (status < 0) {
 292                 mlog_errno(status);
 293                 goto bail;
 294         }
 295
 296         i_size_write(inode, inode->i_sb->s_blocksize);
 297         inode->i_nlink = 2;
 298         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
 299         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 300         if (status < 0) {
 301                 mlog_errno(status);
 302                 goto bail;
 303         }
 304
 305         status = 0;
 306 bail:
 307         if (new_bh)
 308                 brelse(new_bh);
 309
 310         mlog_exit(status);
 311         return status;
 312 }
 313
 314 static int ocfs2_mknod(struct inode *dir,
 315                        struct dentry *dentry,
 316                        int mode,
 317                        dev_t dev)
 318 {
 319         int status = 0;
 320         struct buffer_head *parent_fe_bh = NULL;
 321         struct ocfs2_journal_handle *handle = NULL;
 322         struct ocfs2_super *osb;
 323         struct ocfs2_dinode *dirfe;
 324         struct buffer_head *new_fe_bh = NULL;
 325         struct buffer_head *de_bh = NULL;
 326         struct inode *inode = NULL;
 327         struct ocfs2_alloc_context *inode_ac = NULL;
 328         struct ocfs2_alloc_context *data_ac = NULL;
 329
 330         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 331                    (unsigned long)dev, dentry->d_name.len,
 332                    dentry->d_name.name);
 333
 334         /* get our super block */
 335         osb = OCFS2_SB(dir->i_sb);
 336
 337         status = ocfs2_meta_lock(dir, NULL, &parent_fe_bh, 1);
 338         if (status < 0) {
 339                 if (status != -ENOENT)
 340                         mlog_errno(status);
 341                 return status;
 342         }
 343
 344         handle = ocfs2_alloc_handle(osb);
 345         if (handle == NULL) {
 346                 status = -ENOMEM;
 347                 mlog_errno(status);
 348                 goto leave;
 349         }
 350
 351         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 352                 status = -EMLINK;
 353                 goto leave;
 354         }
 355
 356         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 357         if (!dirfe->i_links_count) {
 358                 /* can't make a file in a deleted directory. */
 359                 status = -ENOENT;
 360                 goto leave;
 361         }
 362
 363         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 364                                            dentry->d_name.len);
 365         if (status)
 366                 goto leave;
 367
 368         /* get a spot inside the dir. */
 369         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 370                                               dentry->d_name.name,
 371                                               dentry->d_name.len, &de_bh);
 372         if (status < 0) {
 373                 mlog_errno(status);
 374                 goto leave;
 375         }
 376
 377         /* reserve an inode spot */
 378         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
 379         if (status < 0) {
 380                 if (status != -ENOSPC)
 381                         mlog_errno(status);
 382                 goto leave;
 383         }
 384
 385         /* are we making a directory? If so, reserve a cluster for his
 386          * 1st extent. */
 387         if (S_ISDIR(mode)) {
 388                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
 389                 if (status < 0) {
 390                         if (status != -ENOSPC)
 391                                 mlog_errno(status);
 392                         goto leave;
 393                 }
 394         }
 395
 396         handle = ocfs2_start_trans(osb, handle, OCFS2_MKNOD_CREDITS);
 397         if (IS_ERR(handle)) {
 398                 status = PTR_ERR(handle);
 399                 handle = NULL;
 400                 mlog_errno(status);
 401                 goto leave;
 402         }
 403
 404         /* do the real work now. */
 405         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 406                                     &new_fe_bh, parent_fe_bh, handle,
 407                                     &inode, inode_ac);
 408         if (status < 0) {
 409                 mlog_errno(status);
 410                 goto leave;
 411         }
 412
 413         if (S_ISDIR(mode)) {
 414                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 415                                             new_fe_bh, data_ac);
 416                 if (status < 0) {
 417                         mlog_errno(status);
 418                         goto leave;
 419                 }
 420
 421                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 422                                               OCFS2_JOURNAL_ACCESS_WRITE);
 423                 if (status < 0) {
 424                         mlog_errno(status);
 425                         goto leave;
 426                 }
 427                 le16_add_cpu(&dirfe->i_links_count, 1);
 428                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 429                 if (status < 0) {
 430                         mlog_errno(status);
 431                         goto leave;
 432                 }
 433                 inc_nlink(dir);
 434         }
 435
 436         status = ocfs2_add_entry(handle, dentry, inode,
 437                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 438                                  de_bh);
 439         if (status < 0) {
 440                 mlog_errno(status);
 441                 goto leave;
 442         }
 443
 444         status = ocfs2_dentry_attach_lock(dentry, inode,
 445                                           OCFS2_I(dir)->ip_blkno);
 446         if (status) {
 447                 mlog_errno(status);
 448                 goto leave;
 449         }
 450
 451         insert_inode_hash(inode);
 452         dentry->d_op = &ocfs2_dentry_ops;
 453         d_instantiate(dentry, inode);
 454         status = 0;
 455 leave:
 456         if (handle)
 457                 ocfs2_commit_trans(handle);
 458
 459         ocfs2_meta_unlock(dir, 1);
 460
 461         if (status == -ENOSPC)
 462                 mlog(0, "Disk is full\n");
 463
 464         if (new_fe_bh)
 465                 brelse(new_fe_bh);
 466
 467         if (de_bh)
 468                 brelse(de_bh);
 469
 470         if (parent_fe_bh)
 471                 brelse(parent_fe_bh);
 472
 473         if ((status < 0) && inode)
 474                 iput(inode);
 475
 476         if (inode_ac)
 477                 ocfs2_free_alloc_context(inode_ac);
 478
 479         if (data_ac)
 480                 ocfs2_free_alloc_context(data_ac);
 481
 482         mlog_exit(status);
 483
 484         return status;
 485 }
 486
 487 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 488                               struct inode *dir,
 489                               struct dentry *dentry, int mode,
 490                               dev_t dev,
 491                               struct buffer_head **new_fe_bh,
 492                               struct buffer_head *parent_fe_bh,
 493                               struct ocfs2_journal_handle *handle,
 494                               struct inode **ret_inode,
 495                               struct ocfs2_alloc_context *inode_ac)
 496 {
 497         int status = 0;
 498         struct ocfs2_dinode *fe = NULL;
 499         struct ocfs2_extent_list *fel;
 500         u64 fe_blkno = 0;
 501         u16 suballoc_bit;
 502         struct inode *inode = NULL;
 503
 504         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 505                    (unsigned long)dev, dentry->d_name.len,
 506                    dentry->d_name.name);
 507
 508         *new_fe_bh = NULL;
 509         *ret_inode = NULL;
 510
 511         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 512                                        &fe_blkno);
 513         if (status < 0) {
 514                 mlog_errno(status);
 515                 goto leave;
 516         }
 517
 518         inode = new_inode(dir->i_sb);
 519         if (IS_ERR(inode)) {
 520                 status = PTR_ERR(inode);
 521                 mlog(ML_ERROR, "new_inode failed!\n");
 522                 goto leave;
 523         }
 524
 525         /* populate as many fields early on as possible - many of
 526          * these are used by the support functions here and in
 527          * callers. */
 528         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 529         OCFS2_I(inode)->ip_blkno = fe_blkno;
 530         if (S_ISDIR(mode))
 531                 inode->i_nlink = 2;
 532         else
 533                 inode->i_nlink = 1;
 534         inode->i_mode = mode;
 535         spin_lock(&osb->osb_lock);
 536         inode->i_generation = osb->s_next_generation++;
 537         spin_unlock(&osb->osb_lock);
 538
 539         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 540         if (!*new_fe_bh) {
 541                 status = -EIO;
 542                 mlog_errno(status);
 543                 goto leave;
 544         }
 545         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 546
 547         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 548                                       OCFS2_JOURNAL_ACCESS_CREATE);
 549         if (status < 0) {
 550                 mlog_errno(status);
 551                 goto leave;
 552         }
 553
 554         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 555         memset(fe, 0, osb->sb->s_blocksize);
 556
 557         fe->i_generation = cpu_to_le32(inode->i_generation);
 558         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 559         fe->i_blkno = cpu_to_le64(fe_blkno);
 560         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 561         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 562         fe->i_uid = cpu_to_le32(current->fsuid);
 563         if (dir->i_mode & S_ISGID) {
 564                 fe->i_gid = cpu_to_le32(dir->i_gid);
 565                 if (S_ISDIR(mode))
 566                         mode |= S_ISGID;
 567         } else
 568                 fe->i_gid = cpu_to_le32(current->fsgid);
 569         fe->i_mode = cpu_to_le16(mode);
 570         if (S_ISCHR(mode) || S_ISBLK(mode))
 571                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 572
 573         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 574
 575         fe->i_last_eb_blk = 0;
 576         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 577         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 578         fe->i_atime = fe->i_ctime = fe->i_mtime =
 579                 cpu_to_le64(CURRENT_TIME.tv_sec);
 580         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 581                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 582         fe->i_dtime = 0;
 583
 584         fel = &fe->id2.i_list;
 585         fel->l_tree_depth = 0;
 586         fel->l_next_free_rec = 0;
 587         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 588
 589         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 590         if (status < 0) {
 591                 mlog_errno(status);
 592                 goto leave;
 593         }
 594
 595         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 596                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 597                      "i_blkno=%llu, i_ino=%lu\n",
 598                      (unsigned long long) (*new_fe_bh)->b_blocknr,
 599                      (unsigned long long)fe->i_blkno, inode->i_ino);
 600                 BUG();
 601         }
 602
 603         ocfs2_inode_set_new(osb, inode);
 604         status = ocfs2_create_new_inode_locks(inode);
 605         if (status < 0)
 606                 mlog_errno(status);
 607
 608         status = 0; /* error in ocfs2_create_new_inode_locks is not
 609                      * critical */
 610
 611         *ret_inode = inode;
 612 leave:
 613         if (status < 0) {
 614                 if (*new_fe_bh) {
 615                         brelse(*new_fe_bh);
 616                         *new_fe_bh = NULL;
 617                 }
 618                 if (inode)
 619                         iput(inode);
 620         }
 621
 622         mlog_exit(status);
 623         return status;
 624 }
 625
 626 static int ocfs2_mkdir(struct inode *dir,
 627                        struct dentry *dentry,
 628                        int mode)
 629 {
 630         int ret;
 631
 632         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 633                    dentry->d_name.len, dentry->d_name.name);
 634         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 635         mlog_exit(ret);
 636
 637         return ret;
 638 }
 639
 640 static int ocfs2_create(struct inode *dir,
 641                         struct dentry *dentry,
 642                         int mode,
 643                         struct nameidata *nd)
 644 {
 645         int ret;
 646
 647         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 648                    dentry->d_name.len, dentry->d_name.name);
 649         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 650         mlog_exit(ret);
 651
 652         return ret;
 653 }
 654
 655 static int ocfs2_link(struct dentry *old_dentry,
 656                       struct inode *dir,
 657                       struct dentry *dentry)
 658 {
 659         struct ocfs2_journal_handle *handle;
 660         struct inode *inode = old_dentry->d_inode;
 661         int err;
 662         struct buffer_head *fe_bh = NULL;
 663         struct buffer_head *parent_fe_bh = NULL;
 664         struct buffer_head *de_bh = NULL;
 665         struct ocfs2_dinode *fe = NULL;
 666         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 667
 668         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 669                    old_dentry->d_name.len, old_dentry->d_name.name,
 670                    dentry->d_name.len, dentry->d_name.name);
 671
 672         if (S_ISDIR(inode->i_mode))
 673                 return -EPERM;
 674
 675         err = ocfs2_meta_lock(dir, NULL, &parent_fe_bh, 1);
 676         if (err < 0) {
 677                 if (err != -ENOENT)
 678                         mlog_errno(err);
 679                 return err;
 680         }
 681
 682         if (!dir->i_nlink) {
 683                 err = -ENOENT;
 684                 goto out;
 685         }
 686
 687         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 688                                         dentry->d_name.len);
 689         if (err)
 690                 goto out;
 691
 692         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 693                                            dentry->d_name.name,
 694                                            dentry->d_name.len, &de_bh);
 695         if (err < 0) {
 696                 mlog_errno(err);
 697                 goto out;
 698         }
 699
 700         err = ocfs2_meta_lock(inode, NULL, &fe_bh, 1);
 701         if (err < 0) {
 702                 if (err != -ENOENT)
 703                         mlog_errno(err);
 704                 goto out;
 705         }
 706
 707         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 708         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 709                 err = -EMLINK;
 710                 goto out_unlock_inode;
 711         }
 712
 713         handle = ocfs2_start_trans(osb, NULL, OCFS2_LINK_CREDITS);
 714         if (IS_ERR(handle)) {
 715                 err = PTR_ERR(handle);
 716                 handle = NULL;
 717                 mlog_errno(err);
 718                 goto out_unlock_inode;
 719         }
 720
 721         err = ocfs2_journal_access(handle, inode, fe_bh,
 722                                    OCFS2_JOURNAL_ACCESS_WRITE);
 723         if (err < 0) {
 724                 mlog_errno(err);
 725                 goto out_commit;
 726         }
 727
 728         inc_nlink(inode);
 729         inode->i_ctime = CURRENT_TIME;
 730         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 731         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 732         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 733
 734         err = ocfs2_journal_dirty(handle, fe_bh);
 735         if (err < 0) {
 736                 le16_add_cpu(&fe->i_links_count, -1);
 737                 drop_nlink(inode);
 738                 mlog_errno(err);
 739                 goto out_commit;
 740         }
 741
 742         err = ocfs2_add_entry(handle, dentry, inode,
 743                               OCFS2_I(inode)->ip_blkno,
 744                               parent_fe_bh, de_bh);
 745         if (err) {
 746                 le16_add_cpu(&fe->i_links_count, -1);
 747                 drop_nlink(inode);
 748                 mlog_errno(err);
 749                 goto out_commit;
 750         }
 751
 752         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 753         if (err) {
 754                 mlog_errno(err);
 755                 goto out_commit;
 756         }
 757
 758         atomic_inc(&inode->i_count);
 759         dentry->d_op = &ocfs2_dentry_ops;
 760         d_instantiate(dentry, inode);
 761
 762 out_commit:
 763         ocfs2_commit_trans(handle);
 764 out_unlock_inode:
 765         ocfs2_meta_unlock(inode, 1);
 766
 767 out:
 768         ocfs2_meta_unlock(dir, 1);
 769
 770         if (de_bh)
 771                 brelse(de_bh);
 772         if (fe_bh)
 773                 brelse(fe_bh);
 774         if (parent_fe_bh)
 775                 brelse(parent_fe_bh);
 776
 777         mlog_exit(err);
 778
 779         return err;
 780 }
 781
 782 /*
 783  * Takes and drops an exclusive lock on the given dentry. This will
 784  * force other nodes to drop it.
 785  */
 786 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 787 {
 788         int ret;
 789
 790         ret = ocfs2_dentry_lock(dentry, 1);
 791         if (ret)
 792                 mlog_errno(ret);
 793         else
 794                 ocfs2_dentry_unlock(dentry, 1);
 795
 796         return ret;
 797 }
 798
 799 static inline int inode_is_unlinkable(struct inode *inode)
 800 {
 801         if (S_ISDIR(inode->i_mode)) {
 802                 if (inode->i_nlink == 2)
 803                         return 1;
 804                 return 0;
 805         }
 806
 807         if (inode->i_nlink == 1)
 808                 return 1;
 809         return 0;
 810 }
 811
 812 static int ocfs2_unlink(struct inode *dir,
 813                         struct dentry *dentry)
 814 {
 815         int status;
 816         struct inode *inode = dentry->d_inode;
 817         struct inode *orphan_dir = NULL;
 818         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 819         u64 blkno;
 820         struct ocfs2_dinode *fe = NULL;
 821         struct buffer_head *fe_bh = NULL;
 822         struct buffer_head *parent_node_bh = NULL;
 823         struct ocfs2_journal_handle *handle = NULL;
 824         struct ocfs2_dir_entry *dirent = NULL;
 825         struct buffer_head *dirent_bh = NULL;
 826         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 827         struct buffer_head *orphan_entry_bh = NULL;
 828
 829         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 830                    dentry->d_name.len, dentry->d_name.name);
 831
 832         BUG_ON(dentry->d_parent->d_inode != dir);
 833
 834         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 835
 836         if (inode == osb->root_inode) {
 837                 mlog(0, "Cannot delete the root directory\n");
 838                 status = -EPERM;
 839                 goto leave;
 840         }
 841
 842         handle = ocfs2_alloc_handle(osb);
 843         if (handle == NULL) {
 844                 status = -ENOMEM;
 845                 mlog_errno(status);
 846                 goto leave;
 847         }
 848
 849         status = ocfs2_meta_lock(dir, handle, &parent_node_bh, 1);
 850         if (status < 0) {
 851                 if (status != -ENOENT)
 852                         mlog_errno(status);
 853                 goto leave;
 854         }
 855
 856         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 857                                           dentry->d_name.len, &blkno,
 858                                           dir, &dirent_bh, &dirent);
 859         if (status < 0) {
 860                 if (status != -ENOENT)
 861                         mlog_errno(status);
 862                 goto leave;
 863         }
 864
 865         if (OCFS2_I(inode)->ip_blkno != blkno) {
 866                 status = -ENOENT;
 867
 868                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 869                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 870                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 871                 goto leave;
 872         }
 873
 874         status = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
 875         if (status < 0) {
 876                 if (status != -ENOENT)
 877                         mlog_errno(status);
 878                 goto leave;
 879         }
 880
 881         if (S_ISDIR(inode->i_mode)) {
 882                 if (!ocfs2_empty_dir(inode)) {
 883                         status = -ENOTEMPTY;
 884                         goto leave;
 885                 } else if (inode->i_nlink != 2) {
 886                         status = -ENOTEMPTY;
 887                         goto leave;
 888                 }
 889         }
 890
 891         status = ocfs2_remote_dentry_delete(dentry);
 892         if (status < 0) {
 893                 /* This vote should succeed under all normal
 894                  * circumstances. */
 895                 mlog_errno(status);
 896                 goto leave;
 897         }
 898
 899         if (inode_is_unlinkable(inode)) {
 900                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 901                                                   orphan_name,
 902                                                   &orphan_entry_bh);
 903                 if (status < 0) {
 904                         mlog_errno(status);
 905                         goto leave;
 906                 }
 907         }
 908
 909         handle = ocfs2_start_trans(osb, handle, OCFS2_UNLINK_CREDITS);
 910         if (IS_ERR(handle)) {
 911                 status = PTR_ERR(handle);
 912                 handle = NULL;
 913                 mlog_errno(status);
 914                 goto leave;
 915         }
 916
 917         status = ocfs2_journal_access(handle, inode, fe_bh,
 918                                       OCFS2_JOURNAL_ACCESS_WRITE);
 919         if (status < 0) {
 920                 mlog_errno(status);
 921                 goto leave;
 922         }
 923
 924         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 925
 926         if (inode_is_unlinkable(inode)) {
 927                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 928                                           orphan_entry_bh, orphan_dir);
 929                 if (status < 0) {
 930                         mlog_errno(status);
 931                         goto leave;
 932                 }
 933         }
 934
 935         /* delete the name from the parent dir */
 936         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 937         if (status < 0) {
 938                 mlog_errno(status);
 939                 goto leave;
 940         }
 941
 942         if (S_ISDIR(inode->i_mode))
 943                 drop_nlink(inode);
 944         drop_nlink(inode);
 945         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 946
 947         status = ocfs2_journal_dirty(handle, fe_bh);
 948         if (status < 0) {
 949                 mlog_errno(status);
 950                 goto leave;
 951         }
 952
 953         if (S_ISDIR(inode->i_mode)) {
 954                 drop_nlink(dir);
 955                 status = ocfs2_mark_inode_dirty(handle, dir,
 956                                                 parent_node_bh);
 957                 if (status < 0) {
 958                         mlog_errno(status);
 959                         inc_nlink(dir);
 960                 }
 961         }
 962
 963 leave:
 964         if (handle)
 965                 ocfs2_commit_trans(handle);
 966
 967         if (orphan_dir) {
 968                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 969                 ocfs2_meta_unlock(orphan_dir, 1);
 970                 mutex_unlock(&orphan_dir->i_mutex);
 971                 iput(orphan_dir);
 972         }
 973
 974         if (fe_bh)
 975                 brelse(fe_bh);
 976
 977         if (dirent_bh)
 978                 brelse(dirent_bh);
 979
 980         if (parent_node_bh)
 981                 brelse(parent_node_bh);
 982
 983         if (orphan_entry_bh)
 984                 brelse(orphan_entry_bh);
 985
 986         mlog_exit(status);
 987
 988         return status;
 989 }
 990
 991 /*
 992  * The only place this should be used is rename!
 993  * if they have the same id, then the 1st one is the only one locked.
 994  */
 995 static int ocfs2_double_lock(struct ocfs2_super *osb,
 996                              struct ocfs2_journal_handle *handle,
 997                              struct buffer_head **bh1,
 998                              struct inode *inode1,
 999                              struct buffer_head **bh2,
1000                              struct inode *inode2)
1001 {
1002         int status;
1003         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1004         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1005         struct buffer_head **tmpbh;
1006         struct inode *tmpinode;
1007
1008         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1009                    (unsigned long long)oi1->ip_blkno,
1010                    (unsigned long long)oi2->ip_blkno);
1011
1012         BUG_ON(!handle);
1013
1014         if (*bh1)
1015                 *bh1 = NULL;
1016         if (*bh2)
1017                 *bh2 = NULL;
1018
1019         /* we always want to lock the one with the lower lockid first. */
1020         if (oi1->ip_blkno != oi2->ip_blkno) {
1021                 if (oi1->ip_blkno < oi2->ip_blkno) {
1022                         /* switch id1 and id2 around */
1023                         mlog(0, "switching them around...\n");
1024                         tmpbh = bh2;
1025                         bh2 = bh1;
1026                         bh1 = tmpbh;
1027
1028                         tmpinode = inode2;
1029                         inode2 = inode1;
1030                         inode1 = tmpinode;
1031                 }
1032                 /* lock id2 */
1033                 status = ocfs2_meta_lock(inode2, handle, bh2, 1);
1034                 if (status < 0) {
1035                         if (status != -ENOENT)
1036                                 mlog_errno(status);
1037                         goto bail;
1038                 }
1039         }
1040         /* lock id1 */
1041         status = ocfs2_meta_lock(inode1, handle, bh1, 1);
1042         if (status < 0) {
1043                 if (status != -ENOENT)
1044                         mlog_errno(status);
1045                 goto bail;
1046         }
1047 bail:
1048         mlog_exit(status);
1049         return status;
1050 }
1051
1052 #define PARENT_INO(buffer) \
1053         ((struct ocfs2_dir_entry *) \
1054          ((char *)buffer + \
1055           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1056
1057 static int ocfs2_rename(struct inode *old_dir,
1058                         struct dentry *old_dentry,
1059                         struct inode *new_dir,
1060                         struct dentry *new_dentry)
1061 {
1062         int status = 0, rename_lock = 0;
1063         struct inode *old_inode = old_dentry->d_inode;
1064         struct inode *new_inode = new_dentry->d_inode;
1065         struct inode *orphan_dir = NULL;
1066         struct ocfs2_dinode *newfe = NULL;
1067         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1068         struct buffer_head *orphan_entry_bh = NULL;
1069         struct buffer_head *newfe_bh = NULL;
1070         struct buffer_head *insert_entry_bh = NULL;
1071         struct ocfs2_super *osb = NULL;
1072         u64 newfe_blkno;
1073         struct ocfs2_journal_handle *handle = NULL;
1074         struct buffer_head *old_dir_bh = NULL;
1075         struct buffer_head *new_dir_bh = NULL;
1076         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1077                                                                // and new_dentry
1078         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1079         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1080                                                     // this is the 1st dirent bh
1081         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1082
1083         /* At some point it might be nice to break this function up a
1084          * bit. */
1085
1086         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1087                    old_dir, old_dentry, new_dir, new_dentry,
1088                    old_dentry->d_name.len, old_dentry->d_name.name,
1089                    new_dentry->d_name.len, new_dentry->d_name.name);
1090
1091         osb = OCFS2_SB(old_dir->i_sb);
1092
1093         if (new_inode) {
1094                 if (!igrab(new_inode))
1095                         BUG();
1096         }
1097
1098         /* Assume a directory heirarchy thusly:
1099          * a/b/c
1100          * a/d
1101          * a,b,c, and d are all directories.
1102          *
1103          * from cwd of 'a' on both nodes:
1104          * node1: mv b/c d
1105          * node2: mv d   b/c
1106          *
1107          * And that's why, just like the VFS, we need a file system
1108          * rename lock. */
1109         if (old_dentry != new_dentry) {
1110                 status = ocfs2_rename_lock(osb);
1111                 if (status < 0) {
1112                         mlog_errno(status);
1113                         goto bail;
1114                 }
1115                 rename_lock = 1;
1116         }
1117
1118         handle = ocfs2_alloc_handle(osb);
1119         if (handle == NULL) {
1120                 status = -ENOMEM;
1121                 mlog_errno(status);
1122                 goto bail;
1123         }
1124
1125         /* if old and new are the same, this'll just do one lock. */
1126         status = ocfs2_double_lock(osb, handle,
1127                                   &old_dir_bh, old_dir,
1128                                   &new_dir_bh, new_dir);
1129         if (status < 0) {
1130                 mlog_errno(status);
1131                 goto bail;
1132         }
1133
1134         /* make sure both dirs have bhs
1135          * get an extra ref on old_dir_bh if old==new */
1136         if (!new_dir_bh) {
1137                 if (old_dir_bh) {
1138                         new_dir_bh = old_dir_bh;
1139                         get_bh(new_dir_bh);
1140                 } else {
1141                         mlog(ML_ERROR, "no old_dir_bh!\n");
1142                         status = -EIO;
1143                         goto bail;
1144                 }
1145         }
1146
1147         /*
1148          * Though we don't require an inode meta data update if
1149          * old_inode is not a directory, we lock anyway here to ensure
1150          * the vote thread on other nodes won't have to concurrently
1151          * downconvert the inode and the dentry locks.
1152          */
1153         status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
1154         if (status < 0) {
1155                 if (status != -ENOENT)
1156                         mlog_errno(status);
1157                 goto bail;
1158         }
1159
1160         status = ocfs2_remote_dentry_delete(old_dentry);
1161         if (status < 0) {
1162                 mlog_errno(status);
1163                 goto bail;
1164         }
1165
1166         if (S_ISDIR(old_inode->i_mode)) {
1167                 status = -EIO;
1168                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1169                 if (!old_inode_de_bh)
1170                         goto bail;
1171
1172                 status = -EIO;
1173                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1174                     OCFS2_I(old_dir)->ip_blkno)
1175                         goto bail;
1176                 status = -EMLINK;
1177                 if (!new_inode && new_dir!=old_dir &&
1178                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1179                         goto bail;
1180         }
1181
1182         status = -ENOENT;
1183         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1184                                      old_dentry->d_name.len,
1185                                      old_dir, &old_de);
1186         if (!old_de_bh)
1187                 goto bail;
1188
1189         /*
1190          *  Check for inode number is _not_ due to possible IO errors.
1191          *  We might rmdir the source, keep it as pwd of some process
1192          *  and merrily kill the link to whatever was created under the
1193          *  same name. Goodbye sticky bit ;-<
1194          */
1195         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1196                 goto bail;
1197
1198         /* check if the target already exists (in which case we need
1199          * to delete it */
1200         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1201                                           new_dentry->d_name.len,
1202                                           &newfe_blkno, new_dir, &new_de_bh,
1203                                           &new_de);
1204         /* The only error we allow here is -ENOENT because the new
1205          * file not existing is perfectly valid. */
1206         if ((status < 0) && (status != -ENOENT)) {
1207                 /* If we cannot find the file specified we should just */
1208                 /* return the error... */
1209                 mlog_errno(status);
1210                 goto bail;
1211         }
1212
1213         if (!new_de && new_inode)
1214                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1215                      "directory!", new_inode->i_ino);
1216
1217         /* In case we need to overwrite an existing file, we blow it
1218          * away first */
1219         if (new_de) {
1220                 /* VFS didn't think there existed an inode here, but
1221                  * someone else in the cluster must have raced our
1222                  * rename to create one. Today we error cleanly, in
1223                  * the future we should consider calling iget to build
1224                  * a new struct inode for this entry. */
1225                 if (!new_inode) {
1226                         status = -EACCES;
1227
1228                         mlog(0, "We found an inode for name %.*s but VFS "
1229                              "didn't give us one.\n", new_dentry->d_name.len,
1230                              new_dentry->d_name.name);
1231                         goto bail;
1232                 }
1233
1234                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1235                         status = -EACCES;
1236
1237                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1238                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1239                              (unsigned long long)newfe_blkno,
1240                              OCFS2_I(new_inode)->ip_flags);
1241                         goto bail;
1242                 }
1243
1244                 status = ocfs2_meta_lock(new_inode, handle, &newfe_bh, 1);
1245                 if (status < 0) {
1246                         if (status != -ENOENT)
1247                                 mlog_errno(status);
1248                         goto bail;
1249                 }
1250
1251                 status = ocfs2_remote_dentry_delete(new_dentry);
1252                 if (status < 0) {
1253                         mlog_errno(status);
1254                         goto bail;
1255                 }
1256
1257                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1258
1259                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1260                      "newfebh=%p bhblocknr=%llu\n", new_de,
1261                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1262                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1263
1264                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1265                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1266                                                           new_inode,
1267                                                           orphan_name,
1268                                                           &orphan_entry_bh);
1269                         if (status < 0) {
1270                                 mlog_errno(status);
1271                                 goto bail;
1272                         }
1273                 }
1274         } else {
1275                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1276
1277                 status = ocfs2_check_dir_for_entry(new_dir,
1278                                                    new_dentry->d_name.name,
1279                                                    new_dentry->d_name.len);
1280                 if (status)
1281                         goto bail;
1282
1283                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1284                                                       new_dentry->d_name.name,
1285                                                       new_dentry->d_name.len,
1286                                                       &insert_entry_bh);
1287                 if (status < 0) {
1288                         mlog_errno(status);
1289                         goto bail;
1290                 }
1291         }
1292
1293         handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1294         if (IS_ERR(handle)) {
1295                 status = PTR_ERR(handle);
1296                 handle = NULL;
1297                 mlog_errno(status);
1298                 goto bail;
1299         }
1300
1301         if (new_de) {
1302                 if (S_ISDIR(new_inode->i_mode)) {
1303                         if (!ocfs2_empty_dir(new_inode) ||
1304                             new_inode->i_nlink != 2) {
1305                                 status = -ENOTEMPTY;
1306                                 goto bail;
1307                         }
1308                 }
1309                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1310                                               OCFS2_JOURNAL_ACCESS_WRITE);
1311                 if (status < 0) {
1312                         mlog_errno(status);
1313                         goto bail;
1314                 }
1315
1316                 if (S_ISDIR(new_inode->i_mode) ||
1317                     (newfe->i_links_count == cpu_to_le16(1))){
1318                         status = ocfs2_orphan_add(osb, handle, new_inode,
1319                                                   newfe, orphan_name,
1320                                                   orphan_entry_bh, orphan_dir);
1321                         if (status < 0) {
1322                                 mlog_errno(status);
1323                                 goto bail;
1324                         }
1325                 }
1326
1327                 /* change the dirent to point to the correct inode */
1328                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1329                                               OCFS2_JOURNAL_ACCESS_WRITE);
1330                 if (status < 0) {
1331                         mlog_errno(status);
1332                         goto bail;
1333                 }
1334                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1335                 new_de->file_type = old_de->file_type;
1336                 new_dir->i_version++;
1337                 status = ocfs2_journal_dirty(handle, new_de_bh);
1338                 if (status < 0) {
1339                         mlog_errno(status);
1340                         goto bail;
1341                 }
1342
1343                 if (S_ISDIR(new_inode->i_mode))
1344                         newfe->i_links_count = 0;
1345                 else
1346                         le16_add_cpu(&newfe->i_links_count, -1);
1347
1348                 status = ocfs2_journal_dirty(handle, newfe_bh);
1349                 if (status < 0) {
1350                         mlog_errno(status);
1351                         goto bail;
1352                 }
1353         } else {
1354                 /* if the name was not found in new_dir, add it now */
1355                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1356                                          OCFS2_I(old_inode)->ip_blkno,
1357                                          new_dir_bh, insert_entry_bh);
1358         }
1359
1360         old_inode->i_ctime = CURRENT_TIME;
1361         mark_inode_dirty(old_inode);
1362
1363         /* now that the name has been added to new_dir, remove the old name */
1364         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1365         if (status < 0) {
1366                 mlog_errno(status);
1367                 goto bail;
1368         }
1369
1370         if (new_inode) {
1371                 new_inode->i_nlink--;
1372                 new_inode->i_ctime = CURRENT_TIME;
1373         }
1374         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1375         if (old_inode_de_bh) {
1376                 status = ocfs2_journal_access(handle, old_inode,
1377                                              old_inode_de_bh,
1378                                              OCFS2_JOURNAL_ACCESS_WRITE);
1379                 PARENT_INO(old_inode_de_bh->b_data) =
1380                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1381                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1382                 old_dir->i_nlink--;
1383                 if (new_inode) {
1384                         new_inode->i_nlink--;
1385                 } else {
1386                         inc_nlink(new_dir);
1387                         mark_inode_dirty(new_dir);
1388                 }
1389         }
1390         mark_inode_dirty(old_dir);
1391         if (new_inode)
1392                 mark_inode_dirty(new_inode);
1393
1394         if (old_dir != new_dir)
1395                 if (new_dir_nlink != new_dir->i_nlink) {
1396                         if (!new_dir_bh) {
1397                                 mlog(ML_ERROR, "need to change nlink for new "
1398                                      "dir %llu from %d to %d but bh is NULL\n",
1399                                      (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1400                                      (int)new_dir_nlink, new_dir->i_nlink);
1401                         } else {
1402                                 struct ocfs2_dinode *fe;
1403                                 status = ocfs2_journal_access(handle,
1404                                                               new_dir,
1405                                                               new_dir_bh,
1406                                                               OCFS2_JOURNAL_ACCESS_WRITE);
1407                                 fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1408                                 fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1409                                 status = ocfs2_journal_dirty(handle, new_dir_bh);
1410                         }
1411                 }
1412
1413         if (old_dir_nlink != old_dir->i_nlink) {
1414                 if (!old_dir_bh) {
1415                         mlog(ML_ERROR, "need to change nlink for old dir "
1416                              "%llu from %d to %d but bh is NULL!\n",
1417                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1418                              (int)old_dir_nlink, old_dir->i_nlink);
1419                 } else {
1420                         struct ocfs2_dinode *fe;
1421                         status = ocfs2_journal_access(handle, old_dir,
1422                                                       old_dir_bh,
1423                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1424                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1425                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1426                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1427                 }
1428         }
1429
1430         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1431         status = 0;
1432 bail:
1433         if (rename_lock)
1434                 ocfs2_rename_unlock(osb);
1435
1436         if (handle)
1437                 ocfs2_commit_trans(handle);
1438
1439         if (orphan_dir) {
1440                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1441                 ocfs2_meta_unlock(orphan_dir, 1);
1442                 mutex_unlock(&orphan_dir->i_mutex);
1443                 iput(orphan_dir);
1444         }
1445
1446         if (new_inode)
1447                 sync_mapping_buffers(old_inode->i_mapping);
1448
1449         if (new_inode)
1450                 iput(new_inode);
1451         if (newfe_bh)
1452                 brelse(newfe_bh);
1453         if (old_dir_bh)
1454                 brelse(old_dir_bh);
1455         if (new_dir_bh)
1456                 brelse(new_dir_bh);
1457         if (new_de_bh)
1458                 brelse(new_de_bh);
1459         if (old_de_bh)
1460                 brelse(old_de_bh);
1461         if (old_inode_de_bh)
1462                 brelse(old_inode_de_bh);
1463         if (orphan_entry_bh)
1464                 brelse(orphan_entry_bh);
1465         if (insert_entry_bh)
1466                 brelse(insert_entry_bh);
1467
1468         mlog_exit(status);
1469
1470         return status;
1471 }
1472
1473 /*
1474  * we expect i_size = strlen(symname). Copy symname into the file
1475  * data, including the null terminator.
1476  */
1477 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1478                                      struct ocfs2_journal_handle *handle,
1479                                      struct inode *inode,
1480                                      const char *symname)
1481 {
1482         struct buffer_head **bhs = NULL;
1483         const char *c;
1484         struct super_block *sb = osb->sb;
1485         u64 p_blkno;
1486         int p_blocks;
1487         int virtual, blocks, status, i, bytes_left;
1488
1489         bytes_left = i_size_read(inode) + 1;
1490         /* we can't trust i_blocks because we're actually going to
1491          * write i_size + 1 bytes. */
1492         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1493
1494         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1495                         (unsigned long long)inode->i_blocks,
1496                         i_size_read(inode), blocks);
1497
1498         /* Sanity check -- make sure we're going to fit. */
1499         if (bytes_left >
1500             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1501                 status = -EIO;
1502                 mlog_errno(status);
1503                 goto bail;
1504         }
1505
1506         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1507         if (!bhs) {
1508                 status = -ENOMEM;
1509                 mlog_errno(status);
1510                 goto bail;
1511         }
1512
1513         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1514                                              &p_blocks);
1515         if (status < 0) {
1516                 mlog_errno(status);
1517                 goto bail;
1518         }
1519
1520         /* links can never be larger than one cluster so we know this
1521          * is all going to be contiguous, but do a sanity check
1522          * anyway. */
1523         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1524                 status = -EIO;
1525                 mlog_errno(status);
1526                 goto bail;
1527         }
1528
1529         virtual = 0;
1530         while(bytes_left > 0) {
1531                 c = &symname[virtual * sb->s_blocksize];
1532
1533                 bhs[virtual] = sb_getblk(sb, p_blkno);
1534                 if (!bhs[virtual]) {
1535                         status = -ENOMEM;
1536                         mlog_errno(status);
1537                         goto bail;
1538                 }
1539                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1540
1541                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1542                                               OCFS2_JOURNAL_ACCESS_CREATE);
1543                 if (status < 0) {
1544                         mlog_errno(status);
1545                         goto bail;
1546                 }
1547
1548                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1549
1550                 memcpy(bhs[virtual]->b_data, c,
1551                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1552                        bytes_left);
1553
1554                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1555                 if (status < 0) {
1556                         mlog_errno(status);
1557                         goto bail;
1558                 }
1559
1560                 virtual++;
1561                 p_blkno++;
1562                 bytes_left -= sb->s_blocksize;
1563         }
1564
1565         status = 0;
1566 bail:
1567
1568         if (bhs) {
1569                 for(i = 0; i < blocks; i++)
1570                         if (bhs[i])
1571                                 brelse(bhs[i]);
1572                 kfree(bhs);
1573         }
1574
1575         mlog_exit(status);
1576         return status;
1577 }
1578
1579 static int ocfs2_symlink(struct inode *dir,
1580                          struct dentry *dentry,
1581                          const char *symname)
1582 {
1583         int status, l, credits;
1584         u64 newsize;
1585         struct ocfs2_super *osb = NULL;
1586         struct inode *inode = NULL;
1587         struct super_block *sb;
1588         struct buffer_head *new_fe_bh = NULL;
1589         struct buffer_head *de_bh = NULL;
1590         struct buffer_head *parent_fe_bh = NULL;
1591         struct ocfs2_dinode *fe = NULL;
1592         struct ocfs2_dinode *dirfe;
1593         struct ocfs2_journal_handle *handle = NULL;
1594         struct ocfs2_alloc_context *inode_ac = NULL;
1595         struct ocfs2_alloc_context *data_ac = NULL;
1596
1597         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1598                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1599
1600         sb = dir->i_sb;
1601         osb = OCFS2_SB(sb);
1602
1603         l = strlen(symname) + 1;
1604
1605         credits = ocfs2_calc_symlink_credits(sb);
1606
1607         handle = ocfs2_alloc_handle(osb);
1608         if (handle == NULL) {
1609                 status = -ENOMEM;
1610                 mlog_errno(status);
1611                 goto bail;
1612         }
1613
1614         /* lock the parent directory */
1615         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1616         if (status < 0) {
1617                 if (status != -ENOENT)
1618                         mlog_errno(status);
1619                 goto bail;
1620         }
1621
1622         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1623         if (!dirfe->i_links_count) {
1624                 /* can't make a file in a deleted directory. */
1625                 status = -ENOENT;
1626                 goto bail;
1627         }
1628
1629         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1630                                            dentry->d_name.len);
1631         if (status)
1632                 goto bail;
1633
1634         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1635                                               dentry->d_name.name,
1636                                               dentry->d_name.len, &de_bh);
1637         if (status < 0) {
1638                 mlog_errno(status);
1639                 goto bail;
1640         }
1641
1642         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1643         if (status < 0) {
1644                 if (status != -ENOSPC)
1645                         mlog_errno(status);
1646                 goto bail;
1647         }
1648
1649         /* don't reserve bitmap space for fast symlinks. */
1650         if (l > ocfs2_fast_symlink_chars(sb)) {
1651                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1652                 if (status < 0) {
1653                         if (status != -ENOSPC)
1654                                 mlog_errno(status);
1655                         goto bail;
1656                 }
1657         }
1658
1659         handle = ocfs2_start_trans(osb, handle, credits);
1660         if (IS_ERR(handle)) {
1661                 status = PTR_ERR(handle);
1662                 handle = NULL;
1663                 mlog_errno(status);
1664                 goto bail;
1665         }
1666
1667         status = ocfs2_mknod_locked(osb, dir, dentry,
1668                                     S_IFLNK | S_IRWXUGO, 0,
1669                                     &new_fe_bh, parent_fe_bh, handle,
1670                                     &inode, inode_ac);
1671         if (status < 0) {
1672                 mlog_errno(status);
1673                 goto bail;
1674         }
1675
1676         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1677         inode->i_rdev = 0;
1678         newsize = l - 1;
1679         if (l > ocfs2_fast_symlink_chars(sb)) {
1680                 inode->i_op = &ocfs2_symlink_inode_operations;
1681                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1682                                                     handle, data_ac, NULL,
1683                                                     NULL);
1684                 if (status < 0) {
1685                         if (status != -ENOSPC && status != -EINTR) {
1686                                 mlog(ML_ERROR,
1687                                      "Failed to extend file to %llu\n",
1688                                      (unsigned long long)newsize);
1689                                 mlog_errno(status);
1690                                 status = -ENOSPC;
1691                         }
1692                         goto bail;
1693                 }
1694                 i_size_write(inode, newsize);
1695                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1696         } else {
1697                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1698                 memcpy((char *) fe->id2.i_symlink, symname, l);
1699                 i_size_write(inode, newsize);
1700                 inode->i_blocks = 0;
1701         }
1702
1703         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1704         if (status < 0) {
1705                 mlog_errno(status);
1706                 goto bail;
1707         }
1708
1709         if (!ocfs2_inode_is_fast_symlink(inode)) {
1710                 status = ocfs2_create_symlink_data(osb, handle, inode,
1711                                                    symname);
1712                 if (status < 0) {
1713                         mlog_errno(status);
1714                         goto bail;
1715                 }
1716         }
1717
1718         status = ocfs2_add_entry(handle, dentry, inode,
1719                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1720                                  de_bh);
1721         if (status < 0) {
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1727         if (status) {
1728                 mlog_errno(status);
1729                 goto bail;
1730         }
1731
1732         insert_inode_hash(inode);
1733         dentry->d_op = &ocfs2_dentry_ops;
1734         d_instantiate(dentry, inode);
1735 bail:
1736         if (handle)
1737                 ocfs2_commit_trans(handle);
1738         if (new_fe_bh)
1739                 brelse(new_fe_bh);
1740         if (parent_fe_bh)
1741                 brelse(parent_fe_bh);
1742         if (de_bh)
1743                 brelse(de_bh);
1744         if (inode_ac)
1745                 ocfs2_free_alloc_context(inode_ac);
1746         if (data_ac)
1747                 ocfs2_free_alloc_context(data_ac);
1748         if ((status < 0) && inode)
1749                 iput(inode);
1750
1751         mlog_exit(status);
1752
1753         return status;
1754 }
1755
1756 int ocfs2_check_dir_entry(struct inode * dir,
1757                           struct ocfs2_dir_entry * de,
1758                           struct buffer_head * bh,
1759                           unsigned long offset)
1760 {
1761         const char *error_msg = NULL;
1762         const int rlen = le16_to_cpu(de->rec_len);
1763
1764         if (rlen < OCFS2_DIR_REC_LEN(1))
1765                 error_msg = "rec_len is smaller than minimal";
1766         else if (rlen % 4 != 0)
1767                 error_msg = "rec_len % 4 != 0";
1768         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1769                 error_msg = "rec_len is too small for name_len";
1770         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1771                 error_msg = "directory entry across blocks";
1772
1773         if (error_msg != NULL)
1774                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1775                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1776                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1777                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1778                      de->name_len);
1779         return error_msg == NULL ? 1 : 0;
1780 }
1781
1782 /* we don't always have a dentry for what we want to add, so people
1783  * like orphan dir can call this instead.
1784  *
1785  * If you pass me insert_bh, I'll skip the search of the other dir
1786  * blocks and put the record in there.
1787  */
1788 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1789                              struct inode *dir,
1790                              const char *name, int namelen,
1791                              struct inode *inode, u64 blkno,
1792                              struct buffer_head *parent_fe_bh,
1793                              struct buffer_head *insert_bh)
1794 {
1795         unsigned long offset;
1796         unsigned short rec_len;
1797         struct ocfs2_dir_entry *de, *de1;
1798         struct super_block *sb;
1799         int retval, status;
1800
1801         mlog_entry_void();
1802
1803         sb = dir->i_sb;
1804
1805         if (!namelen)
1806                 return -EINVAL;
1807
1808         rec_len = OCFS2_DIR_REC_LEN(namelen);
1809         offset = 0;
1810         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1811         while (1) {
1812                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1813                 /* These checks should've already been passed by the
1814                  * prepare function, but I guess we can leave them
1815                  * here anyway. */
1816                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1817                         retval = -ENOENT;
1818                         goto bail;
1819                 }
1820                 if (ocfs2_match(namelen, name, de)) {
1821                         retval = -EEXIST;
1822                         goto bail;
1823                 }
1824                 if (((le64_to_cpu(de->inode) == 0) &&
1825                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1826                     (le16_to_cpu(de->rec_len) >=
1827                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1828                         status = ocfs2_journal_access(handle, dir, insert_bh,
1829                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1830                         /* By now the buffer is marked for journaling */
1831                         offset += le16_to_cpu(de->rec_len);
1832                         if (le64_to_cpu(de->inode)) {
1833                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1834                                         OCFS2_DIR_REC_LEN(de->name_len));
1835                                 de1->rec_len =
1836                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1837                                         OCFS2_DIR_REC_LEN(de->name_len));
1838                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1839                                 de = de1;
1840                         }
1841                         de->file_type = OCFS2_FT_UNKNOWN;
1842                         if (blkno) {
1843                                 de->inode = cpu_to_le64(blkno);
1844                                 ocfs2_set_de_type(de, inode->i_mode);
1845                         } else
1846                                 de->inode = 0;
1847                         de->name_len = namelen;
1848                         memcpy(de->name, name, namelen);
1849
1850                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1851                         dir->i_version++;
1852                         status = ocfs2_journal_dirty(handle, insert_bh);
1853                         retval = 0;
1854                         goto bail;
1855                 }
1856                 offset += le16_to_cpu(de->rec_len);
1857                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1858         }
1859
1860         /* when you think about it, the assert above should prevent us
1861          * from ever getting here. */
1862         retval = -ENOSPC;
1863 bail:
1864
1865         mlog_exit(retval);
1866         return retval;
1867 }
1868
1869
1870 /*
1871  * ocfs2_delete_entry deletes a directory entry by merging it with the
1872  * previous entry
1873  */
1874 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1875                               struct inode *dir,
1876                               struct ocfs2_dir_entry *de_del,
1877                               struct buffer_head *bh)
1878 {
1879         struct ocfs2_dir_entry *de, *pde;
1880         int i, status = -ENOENT;
1881
1882         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1883
1884         i = 0;
1885         pde = NULL;
1886         de = (struct ocfs2_dir_entry *) bh->b_data;
1887         while (i < bh->b_size) {
1888                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1889                         status = -EIO;
1890                         mlog_errno(status);
1891                         goto bail;
1892                 }
1893                 if (de == de_del)  {
1894                         status = ocfs2_journal_access(handle, dir, bh,
1895                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1896                         if (status < 0) {
1897                                 status = -EIO;
1898                                 mlog_errno(status);
1899                                 goto bail;
1900                         }
1901                         if (pde)
1902                                 pde->rec_len =
1903                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1904                                                     le16_to_cpu(de->rec_len));
1905                         else
1906                                 de->inode = 0;
1907                         dir->i_version++;
1908                         status = ocfs2_journal_dirty(handle, bh);
1909                         goto bail;
1910                 }
1911                 i += le16_to_cpu(de->rec_len);
1912                 pde = de;
1913                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1914         }
1915 bail:
1916         mlog_exit(status);
1917         return status;
1918 }
1919
1920 /*
1921  * Returns 0 if not found, -1 on failure, and 1 on success
1922  */
1923 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1924                                         struct inode *dir,
1925                                         const char *name, int namelen,
1926                                         unsigned long offset,
1927                                         struct ocfs2_dir_entry **res_dir)
1928 {
1929         struct ocfs2_dir_entry *de;
1930         char *dlimit, *de_buf;
1931         int de_len;
1932         int ret = 0;
1933
1934         mlog_entry_void();
1935
1936         de_buf = bh->b_data;
1937         dlimit = de_buf + dir->i_sb->s_blocksize;
1938
1939         while (de_buf < dlimit) {
1940                 /* this code is executed quadratically often */
1941                 /* do minimal checking `by hand' */
1942
1943                 de = (struct ocfs2_dir_entry *) de_buf;
1944
1945                 if (de_buf + namelen <= dlimit &&
1946                     ocfs2_match(namelen, name, de)) {
1947                         /* found a match - just to be sure, do a full check */
1948                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1949                                 ret = -1;
1950                                 goto bail;
1951                         }
1952                         *res_dir = de;
1953                         ret = 1;
1954                         goto bail;
1955                 }
1956
1957                 /* prevent looping on a bad block */
1958                 de_len = le16_to_cpu(de->rec_len);
1959                 if (de_len <= 0) {
1960                         ret = -1;
1961                         goto bail;
1962                 }
1963
1964                 de_buf += de_len;
1965                 offset += de_len;
1966         }
1967
1968 bail:
1969         mlog_exit(ret);
1970         return ret;
1971 }
1972
1973 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1974                                      struct inode *dir,
1975                                      struct ocfs2_dir_entry **res_dir)
1976 {
1977         struct super_block *sb;
1978         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1979         struct buffer_head *bh, *ret = NULL;
1980         unsigned long start, block, b;
1981         int ra_max = 0;         /* Number of bh's in the readahead
1982                                    buffer, bh_use[] */
1983         int ra_ptr = 0;         /* Current index into readahead
1984                                    buffer */
1985         int num = 0;
1986         int nblocks, i, err;
1987
1988         mlog_entry_void();
1989
1990         *res_dir = NULL;
1991         sb = dir->i_sb;
1992
1993         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1994         start = OCFS2_I(dir)->ip_dir_start_lookup;
1995         if (start >= nblocks)
1996                 start = 0;
1997         block = start;
1998
1999 restart:
2000         do {
2001                 /*
2002                  * We deal with the read-ahead logic here.
2003                  */
2004                 if (ra_ptr >= ra_max) {
2005                         /* Refill the readahead buffer */
2006                         ra_ptr = 0;
2007                         b = block;
2008                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2009                                 /*
2010                                  * Terminate if we reach the end of the
2011                                  * directory and must wrap, or if our
2012                                  * search has finished at this block.
2013                                  */
2014                                 if (b >= nblocks || (num && block == start)) {
2015                                         bh_use[ra_max] = NULL;
2016                                         break;
2017                                 }
2018                                 num++;
2019
2020                                 bh = ocfs2_bread(dir, b++, &err, 1);
2021                                 bh_use[ra_max] = bh;
2022                         }
2023                 }
2024                 if ((bh = bh_use[ra_ptr++]) == NULL)
2025                         goto next;
2026                 wait_on_buffer(bh);
2027                 if (!buffer_uptodate(bh)) {
2028                         /* read error, skip block & hope for the best */
2029                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2030                                     "offset %lu\n",
2031                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2032                                     block);
2033                         brelse(bh);
2034                         goto next;
2035                 }
2036                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2037                                           block << sb->s_blocksize_bits,
2038                                           res_dir);
2039                 if (i == 1) {
2040                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2041                         ret = bh;
2042                         goto cleanup_and_exit;
2043                 } else {
2044                         brelse(bh);
2045                         if (i < 0)
2046                                 goto cleanup_and_exit;
2047                 }
2048         next:
2049                 if (++block >= nblocks)
2050                         block = 0;
2051         } while (block != start);
2052
2053         /*
2054          * If the directory has grown while we were searching, then
2055          * search the last part of the directory before giving up.
2056          */
2057         block = nblocks;
2058         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2059         if (block < nblocks) {
2060                 start = 0;
2061                 goto restart;
2062         }
2063
2064 cleanup_and_exit:
2065         /* Clean up the read-ahead blocks */
2066         for (; ra_ptr < ra_max; ra_ptr++)
2067                 brelse(bh_use[ra_ptr]);
2068
2069         mlog_exit_ptr(ret);
2070         return ret;
2071 }
2072
2073 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2074 {
2075         int status, namelen;
2076
2077         mlog_entry_void();
2078
2079         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2080                            (long long)blkno);
2081         if (namelen <= 0) {
2082                 if (namelen)
2083                         status = namelen;
2084                 else
2085                         status = -EINVAL;
2086                 mlog_errno(status);
2087                 goto bail;
2088         }
2089         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2090                 status = -EINVAL;
2091                 mlog_errno(status);
2092                 goto bail;
2093         }
2094
2095         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2096              namelen);
2097
2098         status = 0;
2099 bail:
2100         mlog_exit(status);
2101         return status;
2102 }
2103
2104 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2105                                     struct inode **ret_orphan_dir,
2106                                     struct inode *inode,
2107                                     char *name,
2108                                     struct buffer_head **de_bh)
2109 {
2110         struct inode *orphan_dir_inode;
2111         struct buffer_head *orphan_dir_bh = NULL;
2112         int status = 0;
2113
2114         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2115         if (status < 0) {
2116                 mlog_errno(status);
2117                 return status;
2118         }
2119
2120         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2121                                                        ORPHAN_DIR_SYSTEM_INODE,
2122                                                        osb->slot_num);
2123         if (!orphan_dir_inode) {
2124                 status = -ENOENT;
2125                 mlog_errno(status);
2126                 return status;
2127         }
2128
2129         mutex_lock(&orphan_dir_inode->i_mutex);
2130
2131         status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
2132         if (status < 0) {
2133                 mlog_errno(status);
2134                 goto leave;
2135         }
2136
2137         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2138                                               orphan_dir_bh, name,
2139                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2140         if (status < 0) {
2141                 ocfs2_meta_unlock(orphan_dir_inode, 1);
2142
2143                 mlog_errno(status);
2144                 goto leave;
2145         }
2146
2147         *ret_orphan_dir = orphan_dir_inode;
2148
2149 leave:
2150         if (status) {
2151                 mutex_unlock(&orphan_dir_inode->i_mutex);
2152                 iput(orphan_dir_inode);
2153         }
2154
2155         if (orphan_dir_bh)
2156                 brelse(orphan_dir_bh);
2157
2158         mlog_exit(status);
2159         return status;
2160 }
2161
2162 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2163                             struct ocfs2_journal_handle *handle,
2164                             struct inode *inode,
2165                             struct ocfs2_dinode *fe,
2166                             char *name,
2167                             struct buffer_head *de_bh,
2168                             struct inode *orphan_dir_inode)
2169 {
2170         struct buffer_head *orphan_dir_bh = NULL;
2171         int status = 0;
2172         struct ocfs2_dinode *orphan_fe;
2173
2174         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2175
2176         status = ocfs2_read_block(osb,
2177                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2178                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2179                                   orphan_dir_inode);
2180         if (status < 0) {
2181                 mlog_errno(status);
2182                 goto leave;
2183         }
2184
2185         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2186                                       OCFS2_JOURNAL_ACCESS_WRITE);
2187         if (status < 0) {
2188                 mlog_errno(status);
2189                 goto leave;
2190         }
2191
2192         /* we're a cluster, and nlink can change on disk from
2193          * underneath us... */
2194         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2195         if (S_ISDIR(inode->i_mode))
2196                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2197         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2198
2199         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2200         if (status < 0) {
2201                 mlog_errno(status);
2202                 goto leave;
2203         }
2204
2205         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2206                                    OCFS2_ORPHAN_NAMELEN, inode,
2207                                    OCFS2_I(inode)->ip_blkno,
2208                                    orphan_dir_bh, de_bh);
2209         if (status < 0) {
2210                 mlog_errno(status);
2211                 goto leave;
2212         }
2213
2214         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2215
2216         /* Record which orphan dir our inode now resides
2217          * in. delete_inode will use this to determine which orphan
2218          * dir to lock. */
2219         spin_lock(&OCFS2_I(inode)->ip_lock);
2220         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2221         spin_unlock(&OCFS2_I(inode)->ip_lock);
2222
2223         mlog(0, "Inode %llu orphaned in slot %d\n",
2224              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2225
2226 leave:
2227         if (orphan_dir_bh)
2228                 brelse(orphan_dir_bh);
2229
2230         mlog_exit(status);
2231         return status;
2232 }
2233
2234 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2235 int ocfs2_orphan_del(struct ocfs2_super *osb,
2236                      struct ocfs2_journal_handle *handle,
2237                      struct inode *orphan_dir_inode,
2238                      struct inode *inode,
2239                      struct buffer_head *orphan_dir_bh)
2240 {
2241         char name[OCFS2_ORPHAN_NAMELEN + 1];
2242         struct ocfs2_dinode *orphan_fe;
2243         int status = 0;
2244         struct buffer_head *target_de_bh = NULL;
2245         struct ocfs2_dir_entry *target_de = NULL;
2246
2247         mlog_entry_void();
2248
2249         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2250         if (status < 0) {
2251                 mlog_errno(status);
2252                 goto leave;
2253         }
2254
2255         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2256              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2257              OCFS2_ORPHAN_NAMELEN);
2258
2259         /* find it's spot in the orphan directory */
2260         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2261                                         orphan_dir_inode, &target_de);
2262         if (!target_de_bh) {
2263                 status = -ENOENT;
2264                 mlog_errno(status);
2265                 goto leave;
2266         }
2267
2268         /* remove it from the orphan directory */
2269         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2270                                     target_de_bh);
2271         if (status < 0) {
2272                 mlog_errno(status);
2273                 goto leave;
2274         }
2275
2276         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2277                                       OCFS2_JOURNAL_ACCESS_WRITE);
2278         if (status < 0) {
2279                 mlog_errno(status);
2280                 goto leave;
2281         }
2282
2283         /* do the i_nlink dance! :) */
2284         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2285         if (S_ISDIR(inode->i_mode))
2286                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2287         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2288
2289         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2290         if (status < 0) {
2291                 mlog_errno(status);
2292                 goto leave;
2293         }
2294
2295 leave:
2296         if (target_de_bh)
2297                 brelse(target_de_bh);
2298
2299         mlog_exit(status);
2300         return status;
2301 }
2302
2303 struct inode_operations ocfs2_dir_iops = {
2304         .create         = ocfs2_create,
2305         .lookup         = ocfs2_lookup,
2306         .link           = ocfs2_link,
2307         .unlink         = ocfs2_unlink,
2308         .rmdir          = ocfs2_unlink,
2309         .symlink        = ocfs2_symlink,
2310         .mkdir          = ocfs2_mkdir,
2311         .mknod          = ocfs2_mknod,
2312         .rename         = ocfs2_rename,
2313         .setattr        = ocfs2_setattr,
2314         .getattr        = ocfs2_getattr,
2315 };