fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/vs_tag.h>
  44
  45 #define MLOG_MASK_PREFIX ML_NAMEI
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 #include "alloc.h"
  51 #include "dcache.h"
  52 #include "dir.h"
  53 #include "dlmglue.h"
  54 #include "extent_map.h"
  55 #include "file.h"
  56 #include "inode.h"
  57 #include "journal.h"
  58 #include "namei.h"
  59 #include "suballoc.h"
  60 #include "super.h"
  61 #include "symlink.h"
  62 #include "sysfile.h"
  63 #include "uptodate.h"
  64 #include "vote.h"
  65
  66 #include "buffer_head_io.h"
  67
  68 #define NAMEI_RA_CHUNKS  2
  69 #define NAMEI_RA_BLOCKS  4
  70 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
  71 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
  72
  73 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
  74                                         struct inode *dir,
  75                                         const char *name, int namelen,
  76                                         unsigned long offset,
  77                                         struct ocfs2_dir_entry **res_dir);
  78
  79 static int ocfs2_delete_entry(handle_t *handle,
  80                               struct inode *dir,
  81                               struct ocfs2_dir_entry *de_del,
  82                               struct buffer_head *bh);
  83
  84 static int __ocfs2_add_entry(handle_t *handle,
  85                              struct inode *dir,
  86                              const char *name, int namelen,
  87                              struct inode *inode, u64 blkno,
  88                              struct buffer_head *parent_fe_bh,
  89                              struct buffer_head *insert_bh);
  90
  91 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  92                               struct inode *dir,
  93                               struct dentry *dentry, int mode,
  94                               dev_t dev,
  95                               struct buffer_head **new_fe_bh,
  96                               struct buffer_head *parent_fe_bh,
  97                               handle_t *handle,
  98                               struct inode **ret_inode,
  99                               struct ocfs2_alloc_context *inode_ac);
 100
 101 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 102                               handle_t *handle,
 103                               struct inode *parent,
 104                               struct inode *inode,
 105                               struct buffer_head *fe_bh,
 106                               struct ocfs2_alloc_context *data_ac);
 107
 108 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
 109                                     struct inode **ret_orphan_dir,
 110                                     struct inode *inode,
 111                                     char *name,
 112                                     struct buffer_head **de_bh);
 113
 114 static int ocfs2_orphan_add(struct ocfs2_super *osb,
 115                             handle_t *handle,
 116                             struct inode *inode,
 117                             struct ocfs2_dinode *fe,
 118                             char *name,
 119                             struct buffer_head *de_bh,
 120                             struct inode *orphan_dir_inode);
 121
 122 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
 123                                      handle_t *handle,
 124                                      struct inode *inode,
 125                                      const char *symname);
 126
 127 static inline int ocfs2_add_entry(handle_t *handle,
 128                                   struct dentry *dentry,
 129                                   struct inode *inode, u64 blkno,
 130                                   struct buffer_head *parent_fe_bh,
 131                                   struct buffer_head *insert_bh)
 132 {
 133         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
 134                                  dentry->d_name.name, dentry->d_name.len,
 135                                  inode, blkno, parent_fe_bh, insert_bh);
 136 }
 137
 138 /* An orphan dir name is an 8 byte value, printed as a hex string */
 139 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 140
 141 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 142                                    struct nameidata *nd)
 143 {
 144         int status;
 145         u64 blkno;
 146         struct buffer_head *dirent_bh = NULL;
 147         struct inode *inode = NULL;
 148         struct dentry *ret;
 149         struct ocfs2_dir_entry *dirent;
 150         struct ocfs2_inode_info *oi;
 151
 152         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 153                    dentry->d_name.len, dentry->d_name.name);
 154
 155         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 156                 ret = ERR_PTR(-ENAMETOOLONG);
 157                 goto bail;
 158         }
 159
 160         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 161              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 162
 163         status = ocfs2_meta_lock(dir, NULL, 0);
 164         if (status < 0) {
 165                 if (status != -ENOENT)
 166                         mlog_errno(status);
 167                 ret = ERR_PTR(status);
 168                 goto bail;
 169         }
 170
 171         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 172                                           dentry->d_name.len, &blkno,
 173                                           dir, &dirent_bh, &dirent);
 174         if (status < 0)
 175                 goto bail_add;
 176
 177         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 178         if (IS_ERR(inode)) {
 179                 mlog(ML_ERROR, "Unable to create inode %llu\n",
 180                      (unsigned long long)blkno);
 181                 ret = ERR_PTR(-EACCES);
 182                 goto bail_unlock;
 183         }
 184
 185         oi = OCFS2_I(inode);
 186         /* Clear any orphaned state... If we were able to look up the
 187          * inode from a directory, it certainly can't be orphaned. We
 188          * might have the bad state from a node which intended to
 189          * orphan this inode but crashed before it could commit the
 190          * unlink. */
 191         spin_lock(&oi->ip_lock);
 192         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 193         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
 194         spin_unlock(&oi->ip_lock);
 195
 196 bail_add:
 197         dentry->d_op = &ocfs2_dentry_ops;
 198         ret = d_splice_alias(inode, dentry);
 199
 200         if (inode) {
 201                 /*
 202                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 203                  * dentry, it will d_move() it on top of ourse. The
 204                  * return value will indicate this however, so in
 205                  * those cases, we switch them around for the locking
 206                  * code.
 207                  *
 208                  * NOTE: This dentry already has ->d_op set from
 209                  * ocfs2_get_parent() and ocfs2_get_dentry()
 210                  */
 211                 if (ret)
 212                         dentry = ret;
 213
 214                 status = ocfs2_dentry_attach_lock(dentry, inode,
 215                                                   OCFS2_I(dir)->ip_blkno);
 216                 if (status) {
 217                         mlog_errno(status);
 218                         ret = ERR_PTR(status);
 219                         goto bail_unlock;
 220                 }
 221         }
 222
 223 bail_unlock:
 224         /* Don't drop the cluster lock until *after* the d_add --
 225          * unlink on another node will message us to remove that
 226          * dentry under this lock so otherwise we can race this with
 227          * the vote thread and have a stale dentry. */
 228         ocfs2_meta_unlock(dir, 0);
 229
 230 bail:
 231         if (dirent_bh)
 232                 brelse(dirent_bh);
 233
 234         mlog_exit_ptr(ret);
 235
 236         return ret;
 237 }
 238
 239 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
 240                               handle_t *handle,
 241                               struct inode *parent,
 242                               struct inode *inode,
 243                               struct buffer_head *fe_bh,
 244                               struct ocfs2_alloc_context *data_ac)
 245 {
 246         int status;
 247         struct buffer_head *new_bh = NULL;
 248         struct ocfs2_dir_entry *de = NULL;
 249
 250         mlog_entry_void();
 251
 252         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
 253                                      data_ac, NULL, &new_bh);
 254         if (status < 0) {
 255                 mlog_errno(status);
 256                 goto bail;
 257         }
 258
 259         ocfs2_set_new_buffer_uptodate(inode, new_bh);
 260
 261         status = ocfs2_journal_access(handle, inode, new_bh,
 262                                       OCFS2_JOURNAL_ACCESS_CREATE);
 263         if (status < 0) {
 264                 mlog_errno(status);
 265                 goto bail;
 266         }
 267         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
 268
 269         de = (struct ocfs2_dir_entry *) new_bh->b_data;
 270         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
 271         de->name_len = 1;
 272         de->rec_len =
 273                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
 274         strcpy(de->name, ".");
 275         ocfs2_set_de_type(de, S_IFDIR);
 276         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
 277         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
 278         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
 279                                   OCFS2_DIR_REC_LEN(1));
 280         de->name_len = 2;
 281         strcpy(de->name, "..");
 282         ocfs2_set_de_type(de, S_IFDIR);
 283
 284         status = ocfs2_journal_dirty(handle, new_bh);
 285         if (status < 0) {
 286                 mlog_errno(status);
 287                 goto bail;
 288         }
 289
 290         i_size_write(inode, inode->i_sb->s_blocksize);
 291         inode->i_nlink = 2;
 292         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
 293         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
 294         if (status < 0) {
 295                 mlog_errno(status);
 296                 goto bail;
 297         }
 298
 299         status = 0;
 300 bail:
 301         if (new_bh)
 302                 brelse(new_bh);
 303
 304         mlog_exit(status);
 305         return status;
 306 }
 307
 308 static int ocfs2_mknod(struct inode *dir,
 309                        struct dentry *dentry,
 310                        int mode,
 311                        dev_t dev)
 312 {
 313         int status = 0;
 314         struct buffer_head *parent_fe_bh = NULL;
 315         handle_t *handle = NULL;
 316         struct ocfs2_super *osb;
 317         struct ocfs2_dinode *dirfe;
 318         struct buffer_head *new_fe_bh = NULL;
 319         struct buffer_head *de_bh = NULL;
 320         struct inode *inode = NULL;
 321         struct ocfs2_alloc_context *inode_ac = NULL;
 322         struct ocfs2_alloc_context *data_ac = NULL;
 323
 324         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 325                    (unsigned long)dev, dentry->d_name.len,
 326                    dentry->d_name.name);
 327
 328         /* get our super block */
 329         osb = OCFS2_SB(dir->i_sb);
 330
 331         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 332         if (status < 0) {
 333                 if (status != -ENOENT)
 334                         mlog_errno(status);
 335                 return status;
 336         }
 337
 338         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 339                 status = -EMLINK;
 340                 goto leave;
 341         }
 342
 343         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 344         if (!dirfe->i_links_count) {
 345                 /* can't make a file in a deleted directory. */
 346                 status = -ENOENT;
 347                 goto leave;
 348         }
 349
 350         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 351                                            dentry->d_name.len);
 352         if (status)
 353                 goto leave;
 354
 355         /* get a spot inside the dir. */
 356         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 357                                               dentry->d_name.name,
 358                                               dentry->d_name.len, &de_bh);
 359         if (status < 0) {
 360                 mlog_errno(status);
 361                 goto leave;
 362         }
 363
 364         /* reserve an inode spot */
 365         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 366         if (status < 0) {
 367                 if (status != -ENOSPC)
 368                         mlog_errno(status);
 369                 goto leave;
 370         }
 371
 372         /* are we making a directory? If so, reserve a cluster for his
 373          * 1st extent. */
 374         if (S_ISDIR(mode)) {
 375                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 376                 if (status < 0) {
 377                         if (status != -ENOSPC)
 378                                 mlog_errno(status);
 379                         goto leave;
 380                 }
 381         }
 382
 383         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
 384         if (IS_ERR(handle)) {
 385                 status = PTR_ERR(handle);
 386                 handle = NULL;
 387                 mlog_errno(status);
 388                 goto leave;
 389         }
 390
 391         /* do the real work now. */
 392         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 393                                     &new_fe_bh, parent_fe_bh, handle,
 394                                     &inode, inode_ac);
 395         if (status < 0) {
 396                 mlog_errno(status);
 397                 goto leave;
 398         }
 399
 400         if (S_ISDIR(mode)) {
 401                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 402                                             new_fe_bh, data_ac);
 403                 if (status < 0) {
 404                         mlog_errno(status);
 405                         goto leave;
 406                 }
 407
 408                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 409                                               OCFS2_JOURNAL_ACCESS_WRITE);
 410                 if (status < 0) {
 411                         mlog_errno(status);
 412                         goto leave;
 413                 }
 414                 le16_add_cpu(&dirfe->i_links_count, 1);
 415                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 416                 if (status < 0) {
 417                         mlog_errno(status);
 418                         goto leave;
 419                 }
 420                 inc_nlink(dir);
 421         }
 422
 423         status = ocfs2_add_entry(handle, dentry, inode,
 424                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 425                                  de_bh);
 426         if (status < 0) {
 427                 mlog_errno(status);
 428                 goto leave;
 429         }
 430
 431         status = ocfs2_dentry_attach_lock(dentry, inode,
 432                                           OCFS2_I(dir)->ip_blkno);
 433         if (status) {
 434                 mlog_errno(status);
 435                 goto leave;
 436         }
 437
 438         insert_inode_hash(inode);
 439         dentry->d_op = &ocfs2_dentry_ops;
 440         d_instantiate(dentry, inode);
 441         status = 0;
 442 leave:
 443         if (handle)
 444                 ocfs2_commit_trans(osb, handle);
 445
 446         ocfs2_meta_unlock(dir, 1);
 447
 448         if (status == -ENOSPC)
 449                 mlog(0, "Disk is full\n");
 450
 451         if (new_fe_bh)
 452                 brelse(new_fe_bh);
 453
 454         if (de_bh)
 455                 brelse(de_bh);
 456
 457         if (parent_fe_bh)
 458                 brelse(parent_fe_bh);
 459
 460         if ((status < 0) && inode)
 461                 iput(inode);
 462
 463         if (inode_ac)
 464                 ocfs2_free_alloc_context(inode_ac);
 465
 466         if (data_ac)
 467                 ocfs2_free_alloc_context(data_ac);
 468
 469         mlog_exit(status);
 470
 471         return status;
 472 }
 473
 474 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 475                               struct inode *dir,
 476                               struct dentry *dentry, int mode,
 477                               dev_t dev,
 478                               struct buffer_head **new_fe_bh,
 479                               struct buffer_head *parent_fe_bh,
 480                               handle_t *handle,
 481                               struct inode **ret_inode,
 482                               struct ocfs2_alloc_context *inode_ac)
 483 {
 484         int status = 0;
 485         struct ocfs2_dinode *fe = NULL;
 486         struct ocfs2_extent_list *fel;
 487         u64 fe_blkno = 0;
 488         u16 suballoc_bit;
 489         struct inode *inode = NULL;
 490         uid_t uid;
 491         gid_t gid;
 492         tag_t tag;
 493
 494         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 495                    (unsigned long)dev, dentry->d_name.len,
 496                    dentry->d_name.name);
 497
 498         *new_fe_bh = NULL;
 499         *ret_inode = NULL;
 500
 501         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 502                                        &fe_blkno);
 503         if (status < 0) {
 504                 mlog_errno(status);
 505                 goto leave;
 506         }
 507
 508         inode = new_inode(dir->i_sb);
 509         if (IS_ERR(inode)) {
 510                 status = PTR_ERR(inode);
 511                 mlog(ML_ERROR, "new_inode failed!\n");
 512                 goto leave;
 513         }
 514
 515         /* populate as many fields early on as possible - many of
 516          * these are used by the support functions here and in
 517          * callers. */
 518         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 519         OCFS2_I(inode)->ip_blkno = fe_blkno;
 520         if (S_ISDIR(mode))
 521                 inode->i_nlink = 2;
 522         else
 523                 inode->i_nlink = 1;
 524         inode->i_mode = mode;
 525         spin_lock(&osb->osb_lock);
 526         inode->i_generation = osb->s_next_generation++;
 527         spin_unlock(&osb->osb_lock);
 528
 529         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 530         if (!*new_fe_bh) {
 531                 status = -EIO;
 532                 mlog_errno(status);
 533                 goto leave;
 534         }
 535         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 536
 537         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 538                                       OCFS2_JOURNAL_ACCESS_CREATE);
 539         if (status < 0) {
 540                 mlog_errno(status);
 541                 goto leave;
 542         }
 543
 544         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 545         memset(fe, 0, osb->sb->s_blocksize);
 546
 547         fe->i_generation = cpu_to_le32(inode->i_generation);
 548         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 549         fe->i_blkno = cpu_to_le64(fe_blkno);
 550         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 551         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 552
 553         tag = dx_current_fstag(osb->sb);
 554         uid = current->fsuid;
 555         if (dir->i_mode & S_ISGID) {
 556                 gid = dir->i_gid;
 557                 if (S_ISDIR(mode))
 558                         mode |= S_ISGID;
 559         } else
 560                 gid = current->fsgid;
 561
 562         fe->i_uid = cpu_to_le32(TAGINO_UID(DX_TAG(inode), uid, tag));
 563         fe->i_gid = cpu_to_le32(TAGINO_GID(DX_TAG(inode), gid, tag));
 564         inode->i_tag = tag;
 565         fe->i_mode = cpu_to_le16(mode);
 566         if (S_ISCHR(mode) || S_ISBLK(mode))
 567                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 568
 569         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 570
 571         fe->i_last_eb_blk = 0;
 572         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 573         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 574         fe->i_atime = fe->i_ctime = fe->i_mtime =
 575                 cpu_to_le64(CURRENT_TIME.tv_sec);
 576         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 577                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 578         fe->i_dtime = 0;
 579
 580         fel = &fe->id2.i_list;
 581         fel->l_tree_depth = 0;
 582         fel->l_next_free_rec = 0;
 583         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 584
 585         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 586         if (status < 0) {
 587                 mlog_errno(status);
 588                 goto leave;
 589         }
 590
 591         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 592                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 593                      "i_blkno=%llu, i_ino=%lu\n",
 594                      (unsigned long long) (*new_fe_bh)->b_blocknr,
 595                      (unsigned long long)fe->i_blkno, inode->i_ino);
 596                 BUG();
 597         }
 598
 599         ocfs2_inode_set_new(osb, inode);
 600         if (!ocfs2_mount_local(osb)) {
 601                 status = ocfs2_create_new_inode_locks(inode);
 602                 if (status < 0)
 603                         mlog_errno(status);
 604         }
 605
 606         status = 0; /* error in ocfs2_create_new_inode_locks is not
 607                      * critical */
 608
 609         *ret_inode = inode;
 610 leave:
 611         if (status < 0) {
 612                 if (*new_fe_bh) {
 613                         brelse(*new_fe_bh);
 614                         *new_fe_bh = NULL;
 615                 }
 616                 if (inode)
 617                         iput(inode);
 618         }
 619
 620         mlog_exit(status);
 621         return status;
 622 }
 623
 624 static int ocfs2_mkdir(struct inode *dir,
 625                        struct dentry *dentry,
 626                        int mode)
 627 {
 628         int ret;
 629
 630         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 631                    dentry->d_name.len, dentry->d_name.name);
 632         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 633         mlog_exit(ret);
 634
 635         return ret;
 636 }
 637
 638 static int ocfs2_create(struct inode *dir,
 639                         struct dentry *dentry,
 640                         int mode,
 641                         struct nameidata *nd)
 642 {
 643         int ret;
 644
 645         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 646                    dentry->d_name.len, dentry->d_name.name);
 647         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 648         mlog_exit(ret);
 649
 650         return ret;
 651 }
 652
 653 static int ocfs2_link(struct dentry *old_dentry,
 654                       struct inode *dir,
 655                       struct dentry *dentry)
 656 {
 657         handle_t *handle;
 658         struct inode *inode = old_dentry->d_inode;
 659         int err;
 660         struct buffer_head *fe_bh = NULL;
 661         struct buffer_head *parent_fe_bh = NULL;
 662         struct buffer_head *de_bh = NULL;
 663         struct ocfs2_dinode *fe = NULL;
 664         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 665
 666         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 667                    old_dentry->d_name.len, old_dentry->d_name.name,
 668                    dentry->d_name.len, dentry->d_name.name);
 669
 670         if (S_ISDIR(inode->i_mode))
 671                 return -EPERM;
 672
 673         err = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 674         if (err < 0) {
 675                 if (err != -ENOENT)
 676                         mlog_errno(err);
 677                 return err;
 678         }
 679
 680         if (!dir->i_nlink) {
 681                 err = -ENOENT;
 682                 goto out;
 683         }
 684
 685         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 686                                         dentry->d_name.len);
 687         if (err)
 688                 goto out;
 689
 690         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 691                                            dentry->d_name.name,
 692                                            dentry->d_name.len, &de_bh);
 693         if (err < 0) {
 694                 mlog_errno(err);
 695                 goto out;
 696         }
 697
 698         err = ocfs2_meta_lock(inode, &fe_bh, 1);
 699         if (err < 0) {
 700                 if (err != -ENOENT)
 701                         mlog_errno(err);
 702                 goto out;
 703         }
 704
 705         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 706         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 707                 err = -EMLINK;
 708                 goto out_unlock_inode;
 709         }
 710
 711         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
 712         if (IS_ERR(handle)) {
 713                 err = PTR_ERR(handle);
 714                 handle = NULL;
 715                 mlog_errno(err);
 716                 goto out_unlock_inode;
 717         }
 718
 719         err = ocfs2_journal_access(handle, inode, fe_bh,
 720                                    OCFS2_JOURNAL_ACCESS_WRITE);
 721         if (err < 0) {
 722                 mlog_errno(err);
 723                 goto out_commit;
 724         }
 725
 726         inc_nlink(inode);
 727         inode->i_ctime = CURRENT_TIME;
 728         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 729         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 730         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 731
 732         err = ocfs2_journal_dirty(handle, fe_bh);
 733         if (err < 0) {
 734                 le16_add_cpu(&fe->i_links_count, -1);
 735                 drop_nlink(inode);
 736                 mlog_errno(err);
 737                 goto out_commit;
 738         }
 739
 740         err = ocfs2_add_entry(handle, dentry, inode,
 741                               OCFS2_I(inode)->ip_blkno,
 742                               parent_fe_bh, de_bh);
 743         if (err) {
 744                 le16_add_cpu(&fe->i_links_count, -1);
 745                 drop_nlink(inode);
 746                 mlog_errno(err);
 747                 goto out_commit;
 748         }
 749
 750         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 751         if (err) {
 752                 mlog_errno(err);
 753                 goto out_commit;
 754         }
 755
 756         atomic_inc(&inode->i_count);
 757         dentry->d_op = &ocfs2_dentry_ops;
 758         d_instantiate(dentry, inode);
 759
 760 out_commit:
 761         ocfs2_commit_trans(osb, handle);
 762 out_unlock_inode:
 763         ocfs2_meta_unlock(inode, 1);
 764
 765 out:
 766         ocfs2_meta_unlock(dir, 1);
 767
 768         if (de_bh)
 769                 brelse(de_bh);
 770         if (fe_bh)
 771                 brelse(fe_bh);
 772         if (parent_fe_bh)
 773                 brelse(parent_fe_bh);
 774
 775         mlog_exit(err);
 776
 777         return err;
 778 }
 779
 780 /*
 781  * Takes and drops an exclusive lock on the given dentry. This will
 782  * force other nodes to drop it.
 783  */
 784 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 785 {
 786         int ret;
 787
 788         ret = ocfs2_dentry_lock(dentry, 1);
 789         if (ret)
 790                 mlog_errno(ret);
 791         else
 792                 ocfs2_dentry_unlock(dentry, 1);
 793
 794         return ret;
 795 }
 796
 797 static inline int inode_is_unlinkable(struct inode *inode)
 798 {
 799         if (S_ISDIR(inode->i_mode)) {
 800                 if (inode->i_nlink == 2)
 801                         return 1;
 802                 return 0;
 803         }
 804
 805         if (inode->i_nlink == 1)
 806                 return 1;
 807         return 0;
 808 }
 809
 810 static int ocfs2_unlink(struct inode *dir,
 811                         struct dentry *dentry)
 812 {
 813         int status;
 814         int child_locked = 0;
 815         struct inode *inode = dentry->d_inode;
 816         struct inode *orphan_dir = NULL;
 817         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 818         u64 blkno;
 819         struct ocfs2_dinode *fe = NULL;
 820         struct buffer_head *fe_bh = NULL;
 821         struct buffer_head *parent_node_bh = NULL;
 822         handle_t *handle = NULL;
 823         struct ocfs2_dir_entry *dirent = NULL;
 824         struct buffer_head *dirent_bh = NULL;
 825         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 826         struct buffer_head *orphan_entry_bh = NULL;
 827
 828         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 829                    dentry->d_name.len, dentry->d_name.name);
 830
 831         BUG_ON(dentry->d_parent->d_inode != dir);
 832
 833         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 834
 835         if (inode == osb->root_inode) {
 836                 mlog(0, "Cannot delete the root directory\n");
 837                 return -EPERM;
 838         }
 839
 840         status = ocfs2_meta_lock(dir, &parent_node_bh, 1);
 841         if (status < 0) {
 842                 if (status != -ENOENT)
 843                         mlog_errno(status);
 844                 return status;
 845         }
 846
 847         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 848                                           dentry->d_name.len, &blkno,
 849                                           dir, &dirent_bh, &dirent);
 850         if (status < 0) {
 851                 if (status != -ENOENT)
 852                         mlog_errno(status);
 853                 goto leave;
 854         }
 855
 856         if (OCFS2_I(inode)->ip_blkno != blkno) {
 857                 status = -ENOENT;
 858
 859                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 860                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 861                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 862                 goto leave;
 863         }
 864
 865         status = ocfs2_meta_lock(inode, &fe_bh, 1);
 866         if (status < 0) {
 867                 if (status != -ENOENT)
 868                         mlog_errno(status);
 869                 goto leave;
 870         }
 871         child_locked = 1;
 872
 873         if (S_ISDIR(inode->i_mode)) {
 874                 if (!ocfs2_empty_dir(inode)) {
 875                         status = -ENOTEMPTY;
 876                         goto leave;
 877                 } else if (inode->i_nlink != 2) {
 878                         status = -ENOTEMPTY;
 879                         goto leave;
 880                 }
 881         }
 882
 883         status = ocfs2_remote_dentry_delete(dentry);
 884         if (status < 0) {
 885                 /* This vote should succeed under all normal
 886                  * circumstances. */
 887                 mlog_errno(status);
 888                 goto leave;
 889         }
 890
 891         if (inode_is_unlinkable(inode)) {
 892                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 893                                                   orphan_name,
 894                                                   &orphan_entry_bh);
 895                 if (status < 0) {
 896                         mlog_errno(status);
 897                         goto leave;
 898                 }
 899         }
 900
 901         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
 902         if (IS_ERR(handle)) {
 903                 status = PTR_ERR(handle);
 904                 handle = NULL;
 905                 mlog_errno(status);
 906                 goto leave;
 907         }
 908
 909         status = ocfs2_journal_access(handle, inode, fe_bh,
 910                                       OCFS2_JOURNAL_ACCESS_WRITE);
 911         if (status < 0) {
 912                 mlog_errno(status);
 913                 goto leave;
 914         }
 915
 916         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 917
 918         if (inode_is_unlinkable(inode)) {
 919                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 920                                           orphan_entry_bh, orphan_dir);
 921                 if (status < 0) {
 922                         mlog_errno(status);
 923                         goto leave;
 924                 }
 925         }
 926
 927         /* delete the name from the parent dir */
 928         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 929         if (status < 0) {
 930                 mlog_errno(status);
 931                 goto leave;
 932         }
 933
 934         if (S_ISDIR(inode->i_mode))
 935                 drop_nlink(inode);
 936         drop_nlink(inode);
 937         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 938
 939         status = ocfs2_journal_dirty(handle, fe_bh);
 940         if (status < 0) {
 941                 mlog_errno(status);
 942                 goto leave;
 943         }
 944
 945         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 946         if (S_ISDIR(inode->i_mode))
 947                 drop_nlink(dir);
 948
 949         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 950         if (status < 0) {
 951                 mlog_errno(status);
 952                 if (S_ISDIR(inode->i_mode))
 953                         inc_nlink(dir);
 954         }
 955
 956 leave:
 957         if (handle)
 958                 ocfs2_commit_trans(osb, handle);
 959
 960         if (child_locked)
 961                 ocfs2_meta_unlock(inode, 1);
 962
 963         ocfs2_meta_unlock(dir, 1);
 964
 965         if (orphan_dir) {
 966                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 967                 ocfs2_meta_unlock(orphan_dir, 1);
 968                 mutex_unlock(&orphan_dir->i_mutex);
 969                 iput(orphan_dir);
 970         }
 971
 972         if (fe_bh)
 973                 brelse(fe_bh);
 974
 975         if (dirent_bh)
 976                 brelse(dirent_bh);
 977
 978         if (parent_node_bh)
 979                 brelse(parent_node_bh);
 980
 981         if (orphan_entry_bh)
 982                 brelse(orphan_entry_bh);
 983
 984         mlog_exit(status);
 985
 986         return status;
 987 }
 988
 989 /*
 990  * The only place this should be used is rename!
 991  * if they have the same id, then the 1st one is the only one locked.
 992  */
 993 static int ocfs2_double_lock(struct ocfs2_super *osb,
 994                              struct buffer_head **bh1,
 995                              struct inode *inode1,
 996                              struct buffer_head **bh2,
 997                              struct inode *inode2)
 998 {
 999         int status;
1000         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1001         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1002         struct buffer_head **tmpbh;
1003         struct inode *tmpinode;
1004
1005         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1006                    (unsigned long long)oi1->ip_blkno,
1007                    (unsigned long long)oi2->ip_blkno);
1008
1009         if (*bh1)
1010                 *bh1 = NULL;
1011         if (*bh2)
1012                 *bh2 = NULL;
1013
1014         /* we always want to lock the one with the lower lockid first. */
1015         if (oi1->ip_blkno != oi2->ip_blkno) {
1016                 if (oi1->ip_blkno < oi2->ip_blkno) {
1017                         /* switch id1 and id2 around */
1018                         mlog(0, "switching them around...\n");
1019                         tmpbh = bh2;
1020                         bh2 = bh1;
1021                         bh1 = tmpbh;
1022
1023                         tmpinode = inode2;
1024                         inode2 = inode1;
1025                         inode1 = tmpinode;
1026                 }
1027                 /* lock id2 */
1028                 status = ocfs2_meta_lock(inode2, bh2, 1);
1029                 if (status < 0) {
1030                         if (status != -ENOENT)
1031                                 mlog_errno(status);
1032                         goto bail;
1033                 }
1034         }
1035
1036         /* lock id1 */
1037         status = ocfs2_meta_lock(inode1, bh1, 1);
1038         if (status < 0) {
1039                 /*
1040                  * An error return must mean that no cluster locks
1041                  * were held on function exit.
1042                  */
1043                 if (oi1->ip_blkno != oi2->ip_blkno)
1044                         ocfs2_meta_unlock(inode2, 1);
1045
1046                 if (status != -ENOENT)
1047                         mlog_errno(status);
1048         }
1049
1050 bail:
1051         mlog_exit(status);
1052         return status;
1053 }
1054
1055 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1056 {
1057         ocfs2_meta_unlock(inode1, 1);
1058
1059         if (inode1 != inode2)
1060                 ocfs2_meta_unlock(inode2, 1);
1061 }
1062
1063 #define PARENT_INO(buffer) \
1064         ((struct ocfs2_dir_entry *) \
1065          ((char *)buffer + \
1066           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1067
1068 static int ocfs2_rename(struct inode *old_dir,
1069                         struct dentry *old_dentry,
1070                         struct inode *new_dir,
1071                         struct dentry *new_dentry)
1072 {
1073         int status = 0, rename_lock = 0, parents_locked = 0;
1074         int old_child_locked = 0, new_child_locked = 0;
1075         struct inode *old_inode = old_dentry->d_inode;
1076         struct inode *new_inode = new_dentry->d_inode;
1077         struct inode *orphan_dir = NULL;
1078         struct ocfs2_dinode *newfe = NULL;
1079         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1080         struct buffer_head *orphan_entry_bh = NULL;
1081         struct buffer_head *newfe_bh = NULL;
1082         struct buffer_head *old_inode_bh = NULL;
1083         struct buffer_head *insert_entry_bh = NULL;
1084         struct ocfs2_super *osb = NULL;
1085         u64 newfe_blkno;
1086         handle_t *handle = NULL;
1087         struct buffer_head *old_dir_bh = NULL;
1088         struct buffer_head *new_dir_bh = NULL;
1089         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1090                                                                // and new_dentry
1091         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1092         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1093                                                     // this is the 1st dirent bh
1094         nlink_t old_dir_nlink = old_dir->i_nlink;
1095
1096         /* At some point it might be nice to break this function up a
1097          * bit. */
1098
1099         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1100                    old_dir, old_dentry, new_dir, new_dentry,
1101                    old_dentry->d_name.len, old_dentry->d_name.name,
1102                    new_dentry->d_name.len, new_dentry->d_name.name);
1103
1104         osb = OCFS2_SB(old_dir->i_sb);
1105
1106         if (new_inode) {
1107                 if (!igrab(new_inode))
1108                         BUG();
1109         }
1110
1111         /* Assume a directory heirarchy thusly:
1112          * a/b/c
1113          * a/d
1114          * a,b,c, and d are all directories.
1115          *
1116          * from cwd of 'a' on both nodes:
1117          * node1: mv b/c d
1118          * node2: mv d   b/c
1119          *
1120          * And that's why, just like the VFS, we need a file system
1121          * rename lock. */
1122         if (old_dentry != new_dentry) {
1123                 status = ocfs2_rename_lock(osb);
1124                 if (status < 0) {
1125                         mlog_errno(status);
1126                         goto bail;
1127                 }
1128                 rename_lock = 1;
1129         }
1130
1131         /* if old and new are the same, this'll just do one lock. */
1132         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1133                                    &new_dir_bh, new_dir);
1134         if (status < 0) {
1135                 mlog_errno(status);
1136                 goto bail;
1137         }
1138         parents_locked = 1;
1139
1140         /* make sure both dirs have bhs
1141          * get an extra ref on old_dir_bh if old==new */
1142         if (!new_dir_bh) {
1143                 if (old_dir_bh) {
1144                         new_dir_bh = old_dir_bh;
1145                         get_bh(new_dir_bh);
1146                 } else {
1147                         mlog(ML_ERROR, "no old_dir_bh!\n");
1148                         status = -EIO;
1149                         goto bail;
1150                 }
1151         }
1152
1153         /*
1154          * Aside from allowing a meta data update, the locking here
1155          * also ensures that the vote thread on other nodes won't have
1156          * to concurrently downconvert the inode and the dentry locks.
1157          */
1158         status = ocfs2_meta_lock(old_inode, &old_inode_bh, 1);
1159         if (status < 0) {
1160                 if (status != -ENOENT)
1161                         mlog_errno(status);
1162                 goto bail;
1163         }
1164         old_child_locked = 1;
1165
1166         status = ocfs2_remote_dentry_delete(old_dentry);
1167         if (status < 0) {
1168                 mlog_errno(status);
1169                 goto bail;
1170         }
1171
1172         if (S_ISDIR(old_inode->i_mode)) {
1173                 status = -EIO;
1174                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1175                 if (!old_inode_de_bh)
1176                         goto bail;
1177
1178                 status = -EIO;
1179                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1180                     OCFS2_I(old_dir)->ip_blkno)
1181                         goto bail;
1182                 status = -EMLINK;
1183                 if (!new_inode && new_dir!=old_dir &&
1184                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1185                         goto bail;
1186         }
1187
1188         status = -ENOENT;
1189         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1190                                      old_dentry->d_name.len,
1191                                      old_dir, &old_de);
1192         if (!old_de_bh)
1193                 goto bail;
1194
1195         /*
1196          *  Check for inode number is _not_ due to possible IO errors.
1197          *  We might rmdir the source, keep it as pwd of some process
1198          *  and merrily kill the link to whatever was created under the
1199          *  same name. Goodbye sticky bit ;-<
1200          */
1201         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1202                 goto bail;
1203
1204         /* check if the target already exists (in which case we need
1205          * to delete it */
1206         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1207                                           new_dentry->d_name.len,
1208                                           &newfe_blkno, new_dir, &new_de_bh,
1209                                           &new_de);
1210         /* The only error we allow here is -ENOENT because the new
1211          * file not existing is perfectly valid. */
1212         if ((status < 0) && (status != -ENOENT)) {
1213                 /* If we cannot find the file specified we should just */
1214                 /* return the error... */
1215                 mlog_errno(status);
1216                 goto bail;
1217         }
1218
1219         if (!new_de && new_inode)
1220                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1221                      "directory!", new_inode->i_ino);
1222
1223         /* In case we need to overwrite an existing file, we blow it
1224          * away first */
1225         if (new_de) {
1226                 /* VFS didn't think there existed an inode here, but
1227                  * someone else in the cluster must have raced our
1228                  * rename to create one. Today we error cleanly, in
1229                  * the future we should consider calling iget to build
1230                  * a new struct inode for this entry. */
1231                 if (!new_inode) {
1232                         status = -EACCES;
1233
1234                         mlog(0, "We found an inode for name %.*s but VFS "
1235                              "didn't give us one.\n", new_dentry->d_name.len,
1236                              new_dentry->d_name.name);
1237                         goto bail;
1238                 }
1239
1240                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1241                         status = -EACCES;
1242
1243                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1244                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1245                              (unsigned long long)newfe_blkno,
1246                              OCFS2_I(new_inode)->ip_flags);
1247                         goto bail;
1248                 }
1249
1250                 status = ocfs2_meta_lock(new_inode, &newfe_bh, 1);
1251                 if (status < 0) {
1252                         if (status != -ENOENT)
1253                                 mlog_errno(status);
1254                         goto bail;
1255                 }
1256                 new_child_locked = 1;
1257
1258                 status = ocfs2_remote_dentry_delete(new_dentry);
1259                 if (status < 0) {
1260                         mlog_errno(status);
1261                         goto bail;
1262                 }
1263
1264                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1265
1266                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1267                      "newfebh=%p bhblocknr=%llu\n", new_de,
1268                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1269                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1270
1271                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1272                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1273                                                           new_inode,
1274                                                           orphan_name,
1275                                                           &orphan_entry_bh);
1276                         if (status < 0) {
1277                                 mlog_errno(status);
1278                                 goto bail;
1279                         }
1280                 }
1281         } else {
1282                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1283
1284                 status = ocfs2_check_dir_for_entry(new_dir,
1285                                                    new_dentry->d_name.name,
1286                                                    new_dentry->d_name.len);
1287                 if (status)
1288                         goto bail;
1289
1290                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1291                                                       new_dentry->d_name.name,
1292                                                       new_dentry->d_name.len,
1293                                                       &insert_entry_bh);
1294                 if (status < 0) {
1295                         mlog_errno(status);
1296                         goto bail;
1297                 }
1298         }
1299
1300         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1301         if (IS_ERR(handle)) {
1302                 status = PTR_ERR(handle);
1303                 handle = NULL;
1304                 mlog_errno(status);
1305                 goto bail;
1306         }
1307
1308         if (new_de) {
1309                 if (S_ISDIR(new_inode->i_mode)) {
1310                         if (!ocfs2_empty_dir(new_inode) ||
1311                             new_inode->i_nlink != 2) {
1312                                 status = -ENOTEMPTY;
1313                                 goto bail;
1314                         }
1315                 }
1316                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1317                                               OCFS2_JOURNAL_ACCESS_WRITE);
1318                 if (status < 0) {
1319                         mlog_errno(status);
1320                         goto bail;
1321                 }
1322
1323                 if (S_ISDIR(new_inode->i_mode) ||
1324                     (newfe->i_links_count == cpu_to_le16(1))){
1325                         status = ocfs2_orphan_add(osb, handle, new_inode,
1326                                                   newfe, orphan_name,
1327                                                   orphan_entry_bh, orphan_dir);
1328                         if (status < 0) {
1329                                 mlog_errno(status);
1330                                 goto bail;
1331                         }
1332                 }
1333
1334                 /* change the dirent to point to the correct inode */
1335                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1336                                               OCFS2_JOURNAL_ACCESS_WRITE);
1337                 if (status < 0) {
1338                         mlog_errno(status);
1339                         goto bail;
1340                 }
1341                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1342                 new_de->file_type = old_de->file_type;
1343                 new_dir->i_version++;
1344                 status = ocfs2_journal_dirty(handle, new_de_bh);
1345                 if (status < 0) {
1346                         mlog_errno(status);
1347                         goto bail;
1348                 }
1349
1350                 if (S_ISDIR(new_inode->i_mode))
1351                         newfe->i_links_count = 0;
1352                 else
1353                         le16_add_cpu(&newfe->i_links_count, -1);
1354
1355                 status = ocfs2_journal_dirty(handle, newfe_bh);
1356                 if (status < 0) {
1357                         mlog_errno(status);
1358                         goto bail;
1359                 }
1360         } else {
1361                 /* if the name was not found in new_dir, add it now */
1362                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1363                                          OCFS2_I(old_inode)->ip_blkno,
1364                                          new_dir_bh, insert_entry_bh);
1365         }
1366
1367         old_inode->i_ctime = CURRENT_TIME;
1368         mark_inode_dirty(old_inode);
1369         ocfs2_mark_inode_dirty(handle, old_inode, old_inode_bh);
1370
1371         /* now that the name has been added to new_dir, remove the old name */
1372         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1373         if (status < 0) {
1374                 mlog_errno(status);
1375                 goto bail;
1376         }
1377
1378         if (new_inode) {
1379                 new_inode->i_nlink--;
1380                 new_inode->i_ctime = CURRENT_TIME;
1381         }
1382         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1383         if (old_inode_de_bh) {
1384                 status = ocfs2_journal_access(handle, old_inode,
1385                                              old_inode_de_bh,
1386                                              OCFS2_JOURNAL_ACCESS_WRITE);
1387                 PARENT_INO(old_inode_de_bh->b_data) =
1388                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1389                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1390                 old_dir->i_nlink--;
1391                 if (new_inode) {
1392                         new_inode->i_nlink--;
1393                 } else {
1394                         inc_nlink(new_dir);
1395                         mark_inode_dirty(new_dir);
1396                 }
1397         }
1398         mark_inode_dirty(old_dir);
1399         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1400         if (new_inode) {
1401                 mark_inode_dirty(new_inode);
1402                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1403         }
1404
1405         if (old_dir != new_dir) {
1406                 /* Keep the same times on both directories.*/
1407                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1408
1409                 /*
1410                  * This will also pick up the i_nlink change from the
1411                  * block above.
1412                  */
1413                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1414         }
1415
1416         if (old_dir_nlink != old_dir->i_nlink) {
1417                 if (!old_dir_bh) {
1418                         mlog(ML_ERROR, "need to change nlink for old dir "
1419                              "%llu from %d to %d but bh is NULL!\n",
1420                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1421                              (int)old_dir_nlink, old_dir->i_nlink);
1422                 } else {
1423                         struct ocfs2_dinode *fe;
1424                         status = ocfs2_journal_access(handle, old_dir,
1425                                                       old_dir_bh,
1426                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1427                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1428                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1429                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1430                 }
1431         }
1432
1433         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1434         status = 0;
1435 bail:
1436         if (rename_lock)
1437                 ocfs2_rename_unlock(osb);
1438
1439         if (handle)
1440                 ocfs2_commit_trans(osb, handle);
1441
1442         if (parents_locked)
1443                 ocfs2_double_unlock(old_dir, new_dir);
1444
1445         if (old_child_locked)
1446                 ocfs2_meta_unlock(old_inode, 1);
1447
1448         if (new_child_locked)
1449                 ocfs2_meta_unlock(new_inode, 1);
1450
1451         if (orphan_dir) {
1452                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1453                 ocfs2_meta_unlock(orphan_dir, 1);
1454                 mutex_unlock(&orphan_dir->i_mutex);
1455                 iput(orphan_dir);
1456         }
1457
1458         if (new_inode)
1459                 sync_mapping_buffers(old_inode->i_mapping);
1460
1461         if (new_inode)
1462                 iput(new_inode);
1463         if (newfe_bh)
1464                 brelse(newfe_bh);
1465         if (old_inode_bh)
1466                 brelse(old_inode_bh);
1467         if (old_dir_bh)
1468                 brelse(old_dir_bh);
1469         if (new_dir_bh)
1470                 brelse(new_dir_bh);
1471         if (new_de_bh)
1472                 brelse(new_de_bh);
1473         if (old_de_bh)
1474                 brelse(old_de_bh);
1475         if (old_inode_de_bh)
1476                 brelse(old_inode_de_bh);
1477         if (orphan_entry_bh)
1478                 brelse(orphan_entry_bh);
1479         if (insert_entry_bh)
1480                 brelse(insert_entry_bh);
1481
1482         mlog_exit(status);
1483
1484         return status;
1485 }
1486
1487 /*
1488  * we expect i_size = strlen(symname). Copy symname into the file
1489  * data, including the null terminator.
1490  */
1491 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1492                                      handle_t *handle,
1493                                      struct inode *inode,
1494                                      const char *symname)
1495 {
1496         struct buffer_head **bhs = NULL;
1497         const char *c;
1498         struct super_block *sb = osb->sb;
1499         u64 p_blkno;
1500         int p_blocks;
1501         int virtual, blocks, status, i, bytes_left;
1502
1503         bytes_left = i_size_read(inode) + 1;
1504         /* we can't trust i_blocks because we're actually going to
1505          * write i_size + 1 bytes. */
1506         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1507
1508         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1509                         (unsigned long long)inode->i_blocks,
1510                         i_size_read(inode), blocks);
1511
1512         /* Sanity check -- make sure we're going to fit. */
1513         if (bytes_left >
1514             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1515                 status = -EIO;
1516                 mlog_errno(status);
1517                 goto bail;
1518         }
1519
1520         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1521         if (!bhs) {
1522                 status = -ENOMEM;
1523                 mlog_errno(status);
1524                 goto bail;
1525         }
1526
1527         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1528                                              &p_blocks);
1529         if (status < 0) {
1530                 mlog_errno(status);
1531                 goto bail;
1532         }
1533
1534         /* links can never be larger than one cluster so we know this
1535          * is all going to be contiguous, but do a sanity check
1536          * anyway. */
1537         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1538                 status = -EIO;
1539                 mlog_errno(status);
1540                 goto bail;
1541         }
1542
1543         virtual = 0;
1544         while(bytes_left > 0) {
1545                 c = &symname[virtual * sb->s_blocksize];
1546
1547                 bhs[virtual] = sb_getblk(sb, p_blkno);
1548                 if (!bhs[virtual]) {
1549                         status = -ENOMEM;
1550                         mlog_errno(status);
1551                         goto bail;
1552                 }
1553                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1554
1555                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1556                                               OCFS2_JOURNAL_ACCESS_CREATE);
1557                 if (status < 0) {
1558                         mlog_errno(status);
1559                         goto bail;
1560                 }
1561
1562                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1563
1564                 memcpy(bhs[virtual]->b_data, c,
1565                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1566                        bytes_left);
1567
1568                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1569                 if (status < 0) {
1570                         mlog_errno(status);
1571                         goto bail;
1572                 }
1573
1574                 virtual++;
1575                 p_blkno++;
1576                 bytes_left -= sb->s_blocksize;
1577         }
1578
1579         status = 0;
1580 bail:
1581
1582         if (bhs) {
1583                 for(i = 0; i < blocks; i++)
1584                         if (bhs[i])
1585                                 brelse(bhs[i]);
1586                 kfree(bhs);
1587         }
1588
1589         mlog_exit(status);
1590         return status;
1591 }
1592
1593 static int ocfs2_symlink(struct inode *dir,
1594                          struct dentry *dentry,
1595                          const char *symname)
1596 {
1597         int status, l, credits;
1598         u64 newsize;
1599         struct ocfs2_super *osb = NULL;
1600         struct inode *inode = NULL;
1601         struct super_block *sb;
1602         struct buffer_head *new_fe_bh = NULL;
1603         struct buffer_head *de_bh = NULL;
1604         struct buffer_head *parent_fe_bh = NULL;
1605         struct ocfs2_dinode *fe = NULL;
1606         struct ocfs2_dinode *dirfe;
1607         handle_t *handle = NULL;
1608         struct ocfs2_alloc_context *inode_ac = NULL;
1609         struct ocfs2_alloc_context *data_ac = NULL;
1610
1611         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1612                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1613
1614         sb = dir->i_sb;
1615         osb = OCFS2_SB(sb);
1616
1617         l = strlen(symname) + 1;
1618
1619         credits = ocfs2_calc_symlink_credits(sb);
1620
1621         /* lock the parent directory */
1622         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
1623         if (status < 0) {
1624                 if (status != -ENOENT)
1625                         mlog_errno(status);
1626                 return status;
1627         }
1628
1629         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1630         if (!dirfe->i_links_count) {
1631                 /* can't make a file in a deleted directory. */
1632                 status = -ENOENT;
1633                 goto bail;
1634         }
1635
1636         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1637                                            dentry->d_name.len);
1638         if (status)
1639                 goto bail;
1640
1641         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1642                                               dentry->d_name.name,
1643                                               dentry->d_name.len, &de_bh);
1644         if (status < 0) {
1645                 mlog_errno(status);
1646                 goto bail;
1647         }
1648
1649         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1650         if (status < 0) {
1651                 if (status != -ENOSPC)
1652                         mlog_errno(status);
1653                 goto bail;
1654         }
1655
1656         /* don't reserve bitmap space for fast symlinks. */
1657         if (l > ocfs2_fast_symlink_chars(sb)) {
1658                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1659                 if (status < 0) {
1660                         if (status != -ENOSPC)
1661                                 mlog_errno(status);
1662                         goto bail;
1663                 }
1664         }
1665
1666         handle = ocfs2_start_trans(osb, credits);
1667         if (IS_ERR(handle)) {
1668                 status = PTR_ERR(handle);
1669                 handle = NULL;
1670                 mlog_errno(status);
1671                 goto bail;
1672         }
1673
1674         status = ocfs2_mknod_locked(osb, dir, dentry,
1675                                     S_IFLNK | S_IRWXUGO, 0,
1676                                     &new_fe_bh, parent_fe_bh, handle,
1677                                     &inode, inode_ac);
1678         if (status < 0) {
1679                 mlog_errno(status);
1680                 goto bail;
1681         }
1682
1683         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1684         inode->i_rdev = 0;
1685         newsize = l - 1;
1686         if (l > ocfs2_fast_symlink_chars(sb)) {
1687                 inode->i_op = &ocfs2_symlink_inode_operations;
1688                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1689                                                     handle, data_ac, NULL,
1690                                                     NULL);
1691                 if (status < 0) {
1692                         if (status != -ENOSPC && status != -EINTR) {
1693                                 mlog(ML_ERROR,
1694                                      "Failed to extend file to %llu\n",
1695                                      (unsigned long long)newsize);
1696                                 mlog_errno(status);
1697                                 status = -ENOSPC;
1698                         }
1699                         goto bail;
1700                 }
1701                 i_size_write(inode, newsize);
1702                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1703         } else {
1704                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1705                 memcpy((char *) fe->id2.i_symlink, symname, l);
1706                 i_size_write(inode, newsize);
1707                 inode->i_blocks = 0;
1708         }
1709
1710         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1711         if (status < 0) {
1712                 mlog_errno(status);
1713                 goto bail;
1714         }
1715
1716         if (!ocfs2_inode_is_fast_symlink(inode)) {
1717                 status = ocfs2_create_symlink_data(osb, handle, inode,
1718                                                    symname);
1719                 if (status < 0) {
1720                         mlog_errno(status);
1721                         goto bail;
1722                 }
1723         }
1724
1725         status = ocfs2_add_entry(handle, dentry, inode,
1726                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1727                                  de_bh);
1728         if (status < 0) {
1729                 mlog_errno(status);
1730                 goto bail;
1731         }
1732
1733         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1734         if (status) {
1735                 mlog_errno(status);
1736                 goto bail;
1737         }
1738
1739         insert_inode_hash(inode);
1740         dentry->d_op = &ocfs2_dentry_ops;
1741         d_instantiate(dentry, inode);
1742 bail:
1743         if (handle)
1744                 ocfs2_commit_trans(osb, handle);
1745
1746         ocfs2_meta_unlock(dir, 1);
1747
1748         if (new_fe_bh)
1749                 brelse(new_fe_bh);
1750         if (parent_fe_bh)
1751                 brelse(parent_fe_bh);
1752         if (de_bh)
1753                 brelse(de_bh);
1754         if (inode_ac)
1755                 ocfs2_free_alloc_context(inode_ac);
1756         if (data_ac)
1757                 ocfs2_free_alloc_context(data_ac);
1758         if ((status < 0) && inode)
1759                 iput(inode);
1760
1761         mlog_exit(status);
1762
1763         return status;
1764 }
1765
1766 int ocfs2_check_dir_entry(struct inode * dir,
1767                           struct ocfs2_dir_entry * de,
1768                           struct buffer_head * bh,
1769                           unsigned long offset)
1770 {
1771         const char *error_msg = NULL;
1772         const int rlen = le16_to_cpu(de->rec_len);
1773
1774         if (rlen < OCFS2_DIR_REC_LEN(1))
1775                 error_msg = "rec_len is smaller than minimal";
1776         else if (rlen % 4 != 0)
1777                 error_msg = "rec_len % 4 != 0";
1778         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1779                 error_msg = "rec_len is too small for name_len";
1780         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1781                 error_msg = "directory entry across blocks";
1782
1783         if (error_msg != NULL)
1784                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1785                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1786                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1787                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1788                      de->name_len);
1789         return error_msg == NULL ? 1 : 0;
1790 }
1791
1792 /* we don't always have a dentry for what we want to add, so people
1793  * like orphan dir can call this instead.
1794  *
1795  * If you pass me insert_bh, I'll skip the search of the other dir
1796  * blocks and put the record in there.
1797  */
1798 static int __ocfs2_add_entry(handle_t *handle,
1799                              struct inode *dir,
1800                              const char *name, int namelen,
1801                              struct inode *inode, u64 blkno,
1802                              struct buffer_head *parent_fe_bh,
1803                              struct buffer_head *insert_bh)
1804 {
1805         unsigned long offset;
1806         unsigned short rec_len;
1807         struct ocfs2_dir_entry *de, *de1;
1808         struct super_block *sb;
1809         int retval, status;
1810
1811         mlog_entry_void();
1812
1813         sb = dir->i_sb;
1814
1815         if (!namelen)
1816                 return -EINVAL;
1817
1818         rec_len = OCFS2_DIR_REC_LEN(namelen);
1819         offset = 0;
1820         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1821         while (1) {
1822                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1823                 /* These checks should've already been passed by the
1824                  * prepare function, but I guess we can leave them
1825                  * here anyway. */
1826                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1827                         retval = -ENOENT;
1828                         goto bail;
1829                 }
1830                 if (ocfs2_match(namelen, name, de)) {
1831                         retval = -EEXIST;
1832                         goto bail;
1833                 }
1834                 if (((le64_to_cpu(de->inode) == 0) &&
1835                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1836                     (le16_to_cpu(de->rec_len) >=
1837                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1838                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1839                         retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1840                         if (retval < 0) {
1841                                 mlog_errno(retval);
1842                                 goto bail;
1843                         }
1844
1845                         status = ocfs2_journal_access(handle, dir, insert_bh,
1846                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1847                         /* By now the buffer is marked for journaling */
1848                         offset += le16_to_cpu(de->rec_len);
1849                         if (le64_to_cpu(de->inode)) {
1850                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1851                                         OCFS2_DIR_REC_LEN(de->name_len));
1852                                 de1->rec_len =
1853                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1854                                         OCFS2_DIR_REC_LEN(de->name_len));
1855                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1856                                 de = de1;
1857                         }
1858                         de->file_type = OCFS2_FT_UNKNOWN;
1859                         if (blkno) {
1860                                 de->inode = cpu_to_le64(blkno);
1861                                 ocfs2_set_de_type(de, inode->i_mode);
1862                         } else
1863                                 de->inode = 0;
1864                         de->name_len = namelen;
1865                         memcpy(de->name, name, namelen);
1866
1867                         dir->i_version++;
1868                         status = ocfs2_journal_dirty(handle, insert_bh);
1869                         retval = 0;
1870                         goto bail;
1871                 }
1872                 offset += le16_to_cpu(de->rec_len);
1873                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1874         }
1875
1876         /* when you think about it, the assert above should prevent us
1877          * from ever getting here. */
1878         retval = -ENOSPC;
1879 bail:
1880
1881         mlog_exit(retval);
1882         return retval;
1883 }
1884
1885
1886 /*
1887  * ocfs2_delete_entry deletes a directory entry by merging it with the
1888  * previous entry
1889  */
1890 static int ocfs2_delete_entry(handle_t *handle,
1891                               struct inode *dir,
1892                               struct ocfs2_dir_entry *de_del,
1893                               struct buffer_head *bh)
1894 {
1895         struct ocfs2_dir_entry *de, *pde;
1896         int i, status = -ENOENT;
1897
1898         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1899
1900         i = 0;
1901         pde = NULL;
1902         de = (struct ocfs2_dir_entry *) bh->b_data;
1903         while (i < bh->b_size) {
1904                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1905                         status = -EIO;
1906                         mlog_errno(status);
1907                         goto bail;
1908                 }
1909                 if (de == de_del)  {
1910                         status = ocfs2_journal_access(handle, dir, bh,
1911                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1912                         if (status < 0) {
1913                                 status = -EIO;
1914                                 mlog_errno(status);
1915                                 goto bail;
1916                         }
1917                         if (pde)
1918                                 pde->rec_len =
1919                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1920                                                     le16_to_cpu(de->rec_len));
1921                         else
1922                                 de->inode = 0;
1923                         dir->i_version++;
1924                         status = ocfs2_journal_dirty(handle, bh);
1925                         goto bail;
1926                 }
1927                 i += le16_to_cpu(de->rec_len);
1928                 pde = de;
1929                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1930         }
1931 bail:
1932         mlog_exit(status);
1933         return status;
1934 }
1935
1936 /*
1937  * Returns 0 if not found, -1 on failure, and 1 on success
1938  */
1939 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1940                                         struct inode *dir,
1941                                         const char *name, int namelen,
1942                                         unsigned long offset,
1943                                         struct ocfs2_dir_entry **res_dir)
1944 {
1945         struct ocfs2_dir_entry *de;
1946         char *dlimit, *de_buf;
1947         int de_len;
1948         int ret = 0;
1949
1950         mlog_entry_void();
1951
1952         de_buf = bh->b_data;
1953         dlimit = de_buf + dir->i_sb->s_blocksize;
1954
1955         while (de_buf < dlimit) {
1956                 /* this code is executed quadratically often */
1957                 /* do minimal checking `by hand' */
1958
1959                 de = (struct ocfs2_dir_entry *) de_buf;
1960
1961                 if (de_buf + namelen <= dlimit &&
1962                     ocfs2_match(namelen, name, de)) {
1963                         /* found a match - just to be sure, do a full check */
1964                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1965                                 ret = -1;
1966                                 goto bail;
1967                         }
1968                         *res_dir = de;
1969                         ret = 1;
1970                         goto bail;
1971                 }
1972
1973                 /* prevent looping on a bad block */
1974                 de_len = le16_to_cpu(de->rec_len);
1975                 if (de_len <= 0) {
1976                         ret = -1;
1977                         goto bail;
1978                 }
1979
1980                 de_buf += de_len;
1981                 offset += de_len;
1982         }
1983
1984 bail:
1985         mlog_exit(ret);
1986         return ret;
1987 }
1988
1989 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1990                                      struct inode *dir,
1991                                      struct ocfs2_dir_entry **res_dir)
1992 {
1993         struct super_block *sb;
1994         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1995         struct buffer_head *bh, *ret = NULL;
1996         unsigned long start, block, b;
1997         int ra_max = 0;         /* Number of bh's in the readahead
1998                                    buffer, bh_use[] */
1999         int ra_ptr = 0;         /* Current index into readahead
2000                                    buffer */
2001         int num = 0;
2002         int nblocks, i, err;
2003
2004         mlog_entry_void();
2005
2006         *res_dir = NULL;
2007         sb = dir->i_sb;
2008
2009         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2010         start = OCFS2_I(dir)->ip_dir_start_lookup;
2011         if (start >= nblocks)
2012                 start = 0;
2013         block = start;
2014
2015 restart:
2016         do {
2017                 /*
2018                  * We deal with the read-ahead logic here.
2019                  */
2020                 if (ra_ptr >= ra_max) {
2021                         /* Refill the readahead buffer */
2022                         ra_ptr = 0;
2023                         b = block;
2024                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2025                                 /*
2026                                  * Terminate if we reach the end of the
2027                                  * directory and must wrap, or if our
2028                                  * search has finished at this block.
2029                                  */
2030                                 if (b >= nblocks || (num && block == start)) {
2031                                         bh_use[ra_max] = NULL;
2032                                         break;
2033                                 }
2034                                 num++;
2035
2036                                 bh = ocfs2_bread(dir, b++, &err, 1);
2037                                 bh_use[ra_max] = bh;
2038                         }
2039                 }
2040                 if ((bh = bh_use[ra_ptr++]) == NULL)
2041                         goto next;
2042                 wait_on_buffer(bh);
2043                 if (!buffer_uptodate(bh)) {
2044                         /* read error, skip block & hope for the best */
2045                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2046                                     "offset %lu\n",
2047                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2048                                     block);
2049                         brelse(bh);
2050                         goto next;
2051                 }
2052                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2053                                           block << sb->s_blocksize_bits,
2054                                           res_dir);
2055                 if (i == 1) {
2056                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2057                         ret = bh;
2058                         goto cleanup_and_exit;
2059                 } else {
2060                         brelse(bh);
2061                         if (i < 0)
2062                                 goto cleanup_and_exit;
2063                 }
2064         next:
2065                 if (++block >= nblocks)
2066                         block = 0;
2067         } while (block != start);
2068
2069         /*
2070          * If the directory has grown while we were searching, then
2071          * search the last part of the directory before giving up.
2072          */
2073         block = nblocks;
2074         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2075         if (block < nblocks) {
2076                 start = 0;
2077                 goto restart;
2078         }
2079
2080 cleanup_and_exit:
2081         /* Clean up the read-ahead blocks */
2082         for (; ra_ptr < ra_max; ra_ptr++)
2083                 brelse(bh_use[ra_ptr]);
2084
2085         mlog_exit_ptr(ret);
2086         return ret;
2087 }
2088
2089 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2090 {
2091         int status, namelen;
2092
2093         mlog_entry_void();
2094
2095         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2096                            (long long)blkno);
2097         if (namelen <= 0) {
2098                 if (namelen)
2099                         status = namelen;
2100                 else
2101                         status = -EINVAL;
2102                 mlog_errno(status);
2103                 goto bail;
2104         }
2105         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2106                 status = -EINVAL;
2107                 mlog_errno(status);
2108                 goto bail;
2109         }
2110
2111         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2112              namelen);
2113
2114         status = 0;
2115 bail:
2116         mlog_exit(status);
2117         return status;
2118 }
2119
2120 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2121                                     struct inode **ret_orphan_dir,
2122                                     struct inode *inode,
2123                                     char *name,
2124                                     struct buffer_head **de_bh)
2125 {
2126         struct inode *orphan_dir_inode;
2127         struct buffer_head *orphan_dir_bh = NULL;
2128         int status = 0;
2129
2130         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2131         if (status < 0) {
2132                 mlog_errno(status);
2133                 return status;
2134         }
2135
2136         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2137                                                        ORPHAN_DIR_SYSTEM_INODE,
2138                                                        osb->slot_num);
2139         if (!orphan_dir_inode) {
2140                 status = -ENOENT;
2141                 mlog_errno(status);
2142                 return status;
2143         }
2144
2145         mutex_lock(&orphan_dir_inode->i_mutex);
2146
2147         status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2148         if (status < 0) {
2149                 mlog_errno(status);
2150                 goto leave;
2151         }
2152
2153         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2154                                               orphan_dir_bh, name,
2155                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2156         if (status < 0) {
2157                 ocfs2_meta_unlock(orphan_dir_inode, 1);
2158
2159                 mlog_errno(status);
2160                 goto leave;
2161         }
2162
2163         *ret_orphan_dir = orphan_dir_inode;
2164
2165 leave:
2166         if (status) {
2167                 mutex_unlock(&orphan_dir_inode->i_mutex);
2168                 iput(orphan_dir_inode);
2169         }
2170
2171         if (orphan_dir_bh)
2172                 brelse(orphan_dir_bh);
2173
2174         mlog_exit(status);
2175         return status;
2176 }
2177
2178 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2179                             handle_t *handle,
2180                             struct inode *inode,
2181                             struct ocfs2_dinode *fe,
2182                             char *name,
2183                             struct buffer_head *de_bh,
2184                             struct inode *orphan_dir_inode)
2185 {
2186         struct buffer_head *orphan_dir_bh = NULL;
2187         int status = 0;
2188         struct ocfs2_dinode *orphan_fe;
2189
2190         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2191
2192         status = ocfs2_read_block(osb,
2193                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2194                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2195                                   orphan_dir_inode);
2196         if (status < 0) {
2197                 mlog_errno(status);
2198                 goto leave;
2199         }
2200
2201         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2202                                       OCFS2_JOURNAL_ACCESS_WRITE);
2203         if (status < 0) {
2204                 mlog_errno(status);
2205                 goto leave;
2206         }
2207
2208         /* we're a cluster, and nlink can change on disk from
2209          * underneath us... */
2210         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2211         if (S_ISDIR(inode->i_mode))
2212                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2213         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2214
2215         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2216         if (status < 0) {
2217                 mlog_errno(status);
2218                 goto leave;
2219         }
2220
2221         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2222                                    OCFS2_ORPHAN_NAMELEN, inode,
2223                                    OCFS2_I(inode)->ip_blkno,
2224                                    orphan_dir_bh, de_bh);
2225         if (status < 0) {
2226                 mlog_errno(status);
2227                 goto leave;
2228         }
2229
2230         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2231
2232         /* Record which orphan dir our inode now resides
2233          * in. delete_inode will use this to determine which orphan
2234          * dir to lock. */
2235         spin_lock(&OCFS2_I(inode)->ip_lock);
2236         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2237         spin_unlock(&OCFS2_I(inode)->ip_lock);
2238
2239         mlog(0, "Inode %llu orphaned in slot %d\n",
2240              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2241
2242 leave:
2243         if (orphan_dir_bh)
2244                 brelse(orphan_dir_bh);
2245
2246         mlog_exit(status);
2247         return status;
2248 }
2249
2250 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2251 int ocfs2_orphan_del(struct ocfs2_super *osb,
2252                      handle_t *handle,
2253                      struct inode *orphan_dir_inode,
2254                      struct inode *inode,
2255                      struct buffer_head *orphan_dir_bh)
2256 {
2257         char name[OCFS2_ORPHAN_NAMELEN + 1];
2258         struct ocfs2_dinode *orphan_fe;
2259         int status = 0;
2260         struct buffer_head *target_de_bh = NULL;
2261         struct ocfs2_dir_entry *target_de = NULL;
2262
2263         mlog_entry_void();
2264
2265         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2266         if (status < 0) {
2267                 mlog_errno(status);
2268                 goto leave;
2269         }
2270
2271         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2272              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2273              OCFS2_ORPHAN_NAMELEN);
2274
2275         /* find it's spot in the orphan directory */
2276         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2277                                         orphan_dir_inode, &target_de);
2278         if (!target_de_bh) {
2279                 status = -ENOENT;
2280                 mlog_errno(status);
2281                 goto leave;
2282         }
2283
2284         /* remove it from the orphan directory */
2285         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2286                                     target_de_bh);
2287         if (status < 0) {
2288                 mlog_errno(status);
2289                 goto leave;
2290         }
2291
2292         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2293                                       OCFS2_JOURNAL_ACCESS_WRITE);
2294         if (status < 0) {
2295                 mlog_errno(status);
2296                 goto leave;
2297         }
2298
2299         /* do the i_nlink dance! :) */
2300         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2301         if (S_ISDIR(inode->i_mode))
2302                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2303         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2304
2305         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2306         if (status < 0) {
2307                 mlog_errno(status);
2308                 goto leave;
2309         }
2310
2311 leave:
2312         if (target_de_bh)
2313                 brelse(target_de_bh);
2314
2315         mlog_exit(status);
2316         return status;
2317 }
2318
2319 struct inode_operations ocfs2_dir_iops = {
2320         .create         = ocfs2_create,
2321         .lookup         = ocfs2_lookup,
2322         .link           = ocfs2_link,
2323         .unlink         = ocfs2_unlink,
2324         .rmdir          = ocfs2_unlink,
2325         .symlink        = ocfs2_symlink,
2326         .mkdir          = ocfs2_mkdir,
2327         .mknod          = ocfs2_mknod,
2328         .rename         = ocfs2_rename,
2329         .setattr        = ocfs2_setattr,
2330         .getattr        = ocfs2_getattr,
2331         .sync_flags     = ocfs2_sync_flags,
2332         .permission     = ocfs2_permission,
2333 };