fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43
  44 #define MLOG_MASK_PREFIX ML_NAMEI
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "vote.h"
  64
  65 #include "buffer_head_io.h"
  66
  67 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  68                               struct inode *dir,
  69                               struct dentry *dentry, int mode,
  70                               dev_t dev,
  71                               struct buffer_head **new_fe_bh,
  72                               struct buffer_head *parent_fe_bh,
  73                               handle_t *handle,
  74                               struct inode **ret_inode,
  75                               struct ocfs2_alloc_context *inode_ac);
  76
  77 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  78                                     struct inode **ret_orphan_dir,
  79                                     struct inode *inode,
  80                                     char *name,
  81                                     struct buffer_head **de_bh);
  82
  83 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  84                             handle_t *handle,
  85                             struct inode *inode,
  86                             struct ocfs2_dinode *fe,
  87                             char *name,
  88                             struct buffer_head *de_bh,
  89                             struct inode *orphan_dir_inode);
  90
  91 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  92                                      handle_t *handle,
  93                                      struct inode *inode,
  94                                      const char *symname);
  95
  96 /* An orphan dir name is an 8 byte value, printed as a hex string */
  97 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  98
  99 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 100                                    struct nameidata *nd)
 101 {
 102         int status;
 103         u64 blkno;
 104         struct inode *inode = NULL;
 105         struct dentry *ret;
 106         struct ocfs2_inode_info *oi;
 107
 108         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 109                    dentry->d_name.len, dentry->d_name.name);
 110
 111         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 112                 ret = ERR_PTR(-ENAMETOOLONG);
 113                 goto bail;
 114         }
 115
 116         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 117              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 118
 119         status = ocfs2_meta_lock(dir, NULL, 0);
 120         if (status < 0) {
 121                 if (status != -ENOENT)
 122                         mlog_errno(status);
 123                 ret = ERR_PTR(status);
 124                 goto bail;
 125         }
 126
 127         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 128                                             dentry->d_name.len, &blkno);
 129         if (status < 0)
 130                 goto bail_add;
 131
 132         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
 133         if (IS_ERR(inode)) {
 134                 ret = ERR_PTR(-EACCES);
 135                 goto bail_unlock;
 136         }
 137
 138         oi = OCFS2_I(inode);
 139         /* Clear any orphaned state... If we were able to look up the
 140          * inode from a directory, it certainly can't be orphaned. We
 141          * might have the bad state from a node which intended to
 142          * orphan this inode but crashed before it could commit the
 143          * unlink. */
 144         spin_lock(&oi->ip_lock);
 145         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 146         spin_unlock(&oi->ip_lock);
 147
 148 bail_add:
 149         dentry->d_op = &ocfs2_dentry_ops;
 150         ret = d_splice_alias(inode, dentry);
 151
 152         if (inode) {
 153                 /*
 154                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 155                  * dentry, it will d_move() it on top of ourse. The
 156                  * return value will indicate this however, so in
 157                  * those cases, we switch them around for the locking
 158                  * code.
 159                  *
 160                  * NOTE: This dentry already has ->d_op set from
 161                  * ocfs2_get_parent() and ocfs2_get_dentry()
 162                  */
 163                 if (ret)
 164                         dentry = ret;
 165
 166                 status = ocfs2_dentry_attach_lock(dentry, inode,
 167                                                   OCFS2_I(dir)->ip_blkno);
 168                 if (status) {
 169                         mlog_errno(status);
 170                         ret = ERR_PTR(status);
 171                         goto bail_unlock;
 172                 }
 173         }
 174
 175 bail_unlock:
 176         /* Don't drop the cluster lock until *after* the d_add --
 177          * unlink on another node will message us to remove that
 178          * dentry under this lock so otherwise we can race this with
 179          * the vote thread and have a stale dentry. */
 180         ocfs2_meta_unlock(dir, 0);
 181
 182 bail:
 183
 184         mlog_exit_ptr(ret);
 185
 186         return ret;
 187 }
 188
 189 static int ocfs2_mknod(struct inode *dir,
 190                        struct dentry *dentry,
 191                        int mode,
 192                        dev_t dev)
 193 {
 194         int status = 0;
 195         struct buffer_head *parent_fe_bh = NULL;
 196         handle_t *handle = NULL;
 197         struct ocfs2_super *osb;
 198         struct ocfs2_dinode *dirfe;
 199         struct buffer_head *new_fe_bh = NULL;
 200         struct buffer_head *de_bh = NULL;
 201         struct inode *inode = NULL;
 202         struct ocfs2_alloc_context *inode_ac = NULL;
 203         struct ocfs2_alloc_context *data_ac = NULL;
 204
 205         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 206                    (unsigned long)dev, dentry->d_name.len,
 207                    dentry->d_name.name);
 208
 209         /* get our super block */
 210         osb = OCFS2_SB(dir->i_sb);
 211
 212         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 213         if (status < 0) {
 214                 if (status != -ENOENT)
 215                         mlog_errno(status);
 216                 return status;
 217         }
 218
 219         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
 220                 status = -EMLINK;
 221                 goto leave;
 222         }
 223
 224         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 225         if (!dirfe->i_links_count) {
 226                 /* can't make a file in a deleted directory. */
 227                 status = -ENOENT;
 228                 goto leave;
 229         }
 230
 231         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 232                                            dentry->d_name.len);
 233         if (status)
 234                 goto leave;
 235
 236         /* get a spot inside the dir. */
 237         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 238                                               dentry->d_name.name,
 239                                               dentry->d_name.len, &de_bh);
 240         if (status < 0) {
 241                 mlog_errno(status);
 242                 goto leave;
 243         }
 244
 245         /* reserve an inode spot */
 246         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 247         if (status < 0) {
 248                 if (status != -ENOSPC)
 249                         mlog_errno(status);
 250                 goto leave;
 251         }
 252
 253         /* are we making a directory? If so, reserve a cluster for his
 254          * 1st extent. */
 255         if (S_ISDIR(mode)) {
 256                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 257                 if (status < 0) {
 258                         if (status != -ENOSPC)
 259                                 mlog_errno(status);
 260                         goto leave;
 261                 }
 262         }
 263
 264         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
 265         if (IS_ERR(handle)) {
 266                 status = PTR_ERR(handle);
 267                 handle = NULL;
 268                 mlog_errno(status);
 269                 goto leave;
 270         }
 271
 272         /* do the real work now. */
 273         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 274                                     &new_fe_bh, parent_fe_bh, handle,
 275                                     &inode, inode_ac);
 276         if (status < 0) {
 277                 mlog_errno(status);
 278                 goto leave;
 279         }
 280
 281         if (S_ISDIR(mode)) {
 282                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 283                                             new_fe_bh, data_ac);
 284                 if (status < 0) {
 285                         mlog_errno(status);
 286                         goto leave;
 287                 }
 288
 289                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 290                                               OCFS2_JOURNAL_ACCESS_WRITE);
 291                 if (status < 0) {
 292                         mlog_errno(status);
 293                         goto leave;
 294                 }
 295                 le16_add_cpu(&dirfe->i_links_count, 1);
 296                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 297                 if (status < 0) {
 298                         mlog_errno(status);
 299                         goto leave;
 300                 }
 301                 inc_nlink(dir);
 302         }
 303
 304         status = ocfs2_add_entry(handle, dentry, inode,
 305                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 306                                  de_bh);
 307         if (status < 0) {
 308                 mlog_errno(status);
 309                 goto leave;
 310         }
 311
 312         status = ocfs2_dentry_attach_lock(dentry, inode,
 313                                           OCFS2_I(dir)->ip_blkno);
 314         if (status) {
 315                 mlog_errno(status);
 316                 goto leave;
 317         }
 318
 319         insert_inode_hash(inode);
 320         dentry->d_op = &ocfs2_dentry_ops;
 321         d_instantiate(dentry, inode);
 322         status = 0;
 323 leave:
 324         if (handle)
 325                 ocfs2_commit_trans(osb, handle);
 326
 327         ocfs2_meta_unlock(dir, 1);
 328
 329         if (status == -ENOSPC)
 330                 mlog(0, "Disk is full\n");
 331
 332         if (new_fe_bh)
 333                 brelse(new_fe_bh);
 334
 335         if (de_bh)
 336                 brelse(de_bh);
 337
 338         if (parent_fe_bh)
 339                 brelse(parent_fe_bh);
 340
 341         if ((status < 0) && inode)
 342                 iput(inode);
 343
 344         if (inode_ac)
 345                 ocfs2_free_alloc_context(inode_ac);
 346
 347         if (data_ac)
 348                 ocfs2_free_alloc_context(data_ac);
 349
 350         mlog_exit(status);
 351
 352         return status;
 353 }
 354
 355 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 356                               struct inode *dir,
 357                               struct dentry *dentry, int mode,
 358                               dev_t dev,
 359                               struct buffer_head **new_fe_bh,
 360                               struct buffer_head *parent_fe_bh,
 361                               handle_t *handle,
 362                               struct inode **ret_inode,
 363                               struct ocfs2_alloc_context *inode_ac)
 364 {
 365         int status = 0;
 366         struct ocfs2_dinode *fe = NULL;
 367         struct ocfs2_extent_list *fel;
 368         u64 fe_blkno = 0;
 369         u16 suballoc_bit;
 370         struct inode *inode = NULL;
 371
 372         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 373                    (unsigned long)dev, dentry->d_name.len,
 374                    dentry->d_name.name);
 375
 376         *new_fe_bh = NULL;
 377         *ret_inode = NULL;
 378
 379         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 380                                        &fe_blkno);
 381         if (status < 0) {
 382                 mlog_errno(status);
 383                 goto leave;
 384         }
 385
 386         inode = new_inode(dir->i_sb);
 387         if (IS_ERR(inode)) {
 388                 status = PTR_ERR(inode);
 389                 mlog(ML_ERROR, "new_inode failed!\n");
 390                 goto leave;
 391         }
 392
 393         /* populate as many fields early on as possible - many of
 394          * these are used by the support functions here and in
 395          * callers. */
 396         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 397         OCFS2_I(inode)->ip_blkno = fe_blkno;
 398         if (S_ISDIR(mode))
 399                 inode->i_nlink = 2;
 400         else
 401                 inode->i_nlink = 1;
 402         inode->i_mode = mode;
 403         spin_lock(&osb->osb_lock);
 404         inode->i_generation = osb->s_next_generation++;
 405         spin_unlock(&osb->osb_lock);
 406
 407         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 408         if (!*new_fe_bh) {
 409                 status = -EIO;
 410                 mlog_errno(status);
 411                 goto leave;
 412         }
 413         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 414
 415         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 416                                       OCFS2_JOURNAL_ACCESS_CREATE);
 417         if (status < 0) {
 418                 mlog_errno(status);
 419                 goto leave;
 420         }
 421
 422         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 423         memset(fe, 0, osb->sb->s_blocksize);
 424
 425         fe->i_generation = cpu_to_le32(inode->i_generation);
 426         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 427         fe->i_blkno = cpu_to_le64(fe_blkno);
 428         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 429         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
 430         fe->i_uid = cpu_to_le32(current->fsuid);
 431         if (dir->i_mode & S_ISGID) {
 432                 fe->i_gid = cpu_to_le32(dir->i_gid);
 433                 if (S_ISDIR(mode))
 434                         mode |= S_ISGID;
 435         } else
 436                 fe->i_gid = cpu_to_le32(current->fsgid);
 437         fe->i_mode = cpu_to_le16(mode);
 438         if (S_ISCHR(mode) || S_ISBLK(mode))
 439                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 440
 441         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 442
 443         fe->i_last_eb_blk = 0;
 444         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 445         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 446         fe->i_atime = fe->i_ctime = fe->i_mtime =
 447                 cpu_to_le64(CURRENT_TIME.tv_sec);
 448         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 449                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 450         fe->i_dtime = 0;
 451
 452         fel = &fe->id2.i_list;
 453         fel->l_tree_depth = 0;
 454         fel->l_next_free_rec = 0;
 455         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 456
 457         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 458         if (status < 0) {
 459                 mlog_errno(status);
 460                 goto leave;
 461         }
 462
 463         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 464                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 465                      "i_blkno=%llu, i_ino=%lu\n",
 466                      (unsigned long long)(*new_fe_bh)->b_blocknr,
 467                      (unsigned long long)le64_to_cpu(fe->i_blkno),
 468                      inode->i_ino);
 469                 BUG();
 470         }
 471
 472         ocfs2_inode_set_new(osb, inode);
 473         if (!ocfs2_mount_local(osb)) {
 474                 status = ocfs2_create_new_inode_locks(inode);
 475                 if (status < 0)
 476                         mlog_errno(status);
 477         }
 478
 479         status = 0; /* error in ocfs2_create_new_inode_locks is not
 480                      * critical */
 481
 482         *ret_inode = inode;
 483 leave:
 484         if (status < 0) {
 485                 if (*new_fe_bh) {
 486                         brelse(*new_fe_bh);
 487                         *new_fe_bh = NULL;
 488                 }
 489                 if (inode)
 490                         iput(inode);
 491         }
 492
 493         mlog_exit(status);
 494         return status;
 495 }
 496
 497 static int ocfs2_mkdir(struct inode *dir,
 498                        struct dentry *dentry,
 499                        int mode)
 500 {
 501         int ret;
 502
 503         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 504                    dentry->d_name.len, dentry->d_name.name);
 505         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 506         mlog_exit(ret);
 507
 508         return ret;
 509 }
 510
 511 static int ocfs2_create(struct inode *dir,
 512                         struct dentry *dentry,
 513                         int mode,
 514                         struct nameidata *nd)
 515 {
 516         int ret;
 517
 518         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 519                    dentry->d_name.len, dentry->d_name.name);
 520         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 521         mlog_exit(ret);
 522
 523         return ret;
 524 }
 525
 526 static int ocfs2_link(struct dentry *old_dentry,
 527                       struct inode *dir,
 528                       struct dentry *dentry)
 529 {
 530         handle_t *handle;
 531         struct inode *inode = old_dentry->d_inode;
 532         int err;
 533         struct buffer_head *fe_bh = NULL;
 534         struct buffer_head *parent_fe_bh = NULL;
 535         struct buffer_head *de_bh = NULL;
 536         struct ocfs2_dinode *fe = NULL;
 537         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 538
 539         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 540                    old_dentry->d_name.len, old_dentry->d_name.name,
 541                    dentry->d_name.len, dentry->d_name.name);
 542
 543         if (S_ISDIR(inode->i_mode))
 544                 return -EPERM;
 545
 546         err = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
 547         if (err < 0) {
 548                 if (err != -ENOENT)
 549                         mlog_errno(err);
 550                 return err;
 551         }
 552
 553         if (!dir->i_nlink) {
 554                 err = -ENOENT;
 555                 goto out;
 556         }
 557
 558         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 559                                         dentry->d_name.len);
 560         if (err)
 561                 goto out;
 562
 563         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 564                                            dentry->d_name.name,
 565                                            dentry->d_name.len, &de_bh);
 566         if (err < 0) {
 567                 mlog_errno(err);
 568                 goto out;
 569         }
 570
 571         err = ocfs2_meta_lock(inode, &fe_bh, 1);
 572         if (err < 0) {
 573                 if (err != -ENOENT)
 574                         mlog_errno(err);
 575                 goto out;
 576         }
 577
 578         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 579         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 580                 err = -EMLINK;
 581                 goto out_unlock_inode;
 582         }
 583
 584         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
 585         if (IS_ERR(handle)) {
 586                 err = PTR_ERR(handle);
 587                 handle = NULL;
 588                 mlog_errno(err);
 589                 goto out_unlock_inode;
 590         }
 591
 592         err = ocfs2_journal_access(handle, inode, fe_bh,
 593                                    OCFS2_JOURNAL_ACCESS_WRITE);
 594         if (err < 0) {
 595                 mlog_errno(err);
 596                 goto out_commit;
 597         }
 598
 599         inc_nlink(inode);
 600         inode->i_ctime = CURRENT_TIME;
 601         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 602         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 603         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 604
 605         err = ocfs2_journal_dirty(handle, fe_bh);
 606         if (err < 0) {
 607                 le16_add_cpu(&fe->i_links_count, -1);
 608                 drop_nlink(inode);
 609                 mlog_errno(err);
 610                 goto out_commit;
 611         }
 612
 613         err = ocfs2_add_entry(handle, dentry, inode,
 614                               OCFS2_I(inode)->ip_blkno,
 615                               parent_fe_bh, de_bh);
 616         if (err) {
 617                 le16_add_cpu(&fe->i_links_count, -1);
 618                 drop_nlink(inode);
 619                 mlog_errno(err);
 620                 goto out_commit;
 621         }
 622
 623         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 624         if (err) {
 625                 mlog_errno(err);
 626                 goto out_commit;
 627         }
 628
 629         atomic_inc(&inode->i_count);
 630         dentry->d_op = &ocfs2_dentry_ops;
 631         d_instantiate(dentry, inode);
 632
 633 out_commit:
 634         ocfs2_commit_trans(osb, handle);
 635 out_unlock_inode:
 636         ocfs2_meta_unlock(inode, 1);
 637
 638 out:
 639         ocfs2_meta_unlock(dir, 1);
 640
 641         if (de_bh)
 642                 brelse(de_bh);
 643         if (fe_bh)
 644                 brelse(fe_bh);
 645         if (parent_fe_bh)
 646                 brelse(parent_fe_bh);
 647
 648         mlog_exit(err);
 649
 650         return err;
 651 }
 652
 653 /*
 654  * Takes and drops an exclusive lock on the given dentry. This will
 655  * force other nodes to drop it.
 656  */
 657 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 658 {
 659         int ret;
 660
 661         ret = ocfs2_dentry_lock(dentry, 1);
 662         if (ret)
 663                 mlog_errno(ret);
 664         else
 665                 ocfs2_dentry_unlock(dentry, 1);
 666
 667         return ret;
 668 }
 669
 670 static inline int inode_is_unlinkable(struct inode *inode)
 671 {
 672         if (S_ISDIR(inode->i_mode)) {
 673                 if (inode->i_nlink == 2)
 674                         return 1;
 675                 return 0;
 676         }
 677
 678         if (inode->i_nlink == 1)
 679                 return 1;
 680         return 0;
 681 }
 682
 683 static int ocfs2_unlink(struct inode *dir,
 684                         struct dentry *dentry)
 685 {
 686         int status;
 687         int child_locked = 0;
 688         struct inode *inode = dentry->d_inode;
 689         struct inode *orphan_dir = NULL;
 690         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 691         u64 blkno;
 692         struct ocfs2_dinode *fe = NULL;
 693         struct buffer_head *fe_bh = NULL;
 694         struct buffer_head *parent_node_bh = NULL;
 695         handle_t *handle = NULL;
 696         struct ocfs2_dir_entry *dirent = NULL;
 697         struct buffer_head *dirent_bh = NULL;
 698         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 699         struct buffer_head *orphan_entry_bh = NULL;
 700
 701         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 702                    dentry->d_name.len, dentry->d_name.name);
 703
 704         BUG_ON(dentry->d_parent->d_inode != dir);
 705
 706         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 707
 708         if (inode == osb->root_inode) {
 709                 mlog(0, "Cannot delete the root directory\n");
 710                 return -EPERM;
 711         }
 712
 713         status = ocfs2_meta_lock(dir, &parent_node_bh, 1);
 714         if (status < 0) {
 715                 if (status != -ENOENT)
 716                         mlog_errno(status);
 717                 return status;
 718         }
 719
 720         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 721                                           dentry->d_name.len, &blkno,
 722                                           dir, &dirent_bh, &dirent);
 723         if (status < 0) {
 724                 if (status != -ENOENT)
 725                         mlog_errno(status);
 726                 goto leave;
 727         }
 728
 729         if (OCFS2_I(inode)->ip_blkno != blkno) {
 730                 status = -ENOENT;
 731
 732                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 733                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 734                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 735                 goto leave;
 736         }
 737
 738         status = ocfs2_meta_lock(inode, &fe_bh, 1);
 739         if (status < 0) {
 740                 if (status != -ENOENT)
 741                         mlog_errno(status);
 742                 goto leave;
 743         }
 744         child_locked = 1;
 745
 746         if (S_ISDIR(inode->i_mode)) {
 747                 if (!ocfs2_empty_dir(inode)) {
 748                         status = -ENOTEMPTY;
 749                         goto leave;
 750                 } else if (inode->i_nlink != 2) {
 751                         status = -ENOTEMPTY;
 752                         goto leave;
 753                 }
 754         }
 755
 756         status = ocfs2_remote_dentry_delete(dentry);
 757         if (status < 0) {
 758                 /* This vote should succeed under all normal
 759                  * circumstances. */
 760                 mlog_errno(status);
 761                 goto leave;
 762         }
 763
 764         if (inode_is_unlinkable(inode)) {
 765                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 766                                                   orphan_name,
 767                                                   &orphan_entry_bh);
 768                 if (status < 0) {
 769                         mlog_errno(status);
 770                         goto leave;
 771                 }
 772         }
 773
 774         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
 775         if (IS_ERR(handle)) {
 776                 status = PTR_ERR(handle);
 777                 handle = NULL;
 778                 mlog_errno(status);
 779                 goto leave;
 780         }
 781
 782         status = ocfs2_journal_access(handle, inode, fe_bh,
 783                                       OCFS2_JOURNAL_ACCESS_WRITE);
 784         if (status < 0) {
 785                 mlog_errno(status);
 786                 goto leave;
 787         }
 788
 789         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 790
 791         if (inode_is_unlinkable(inode)) {
 792                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 793                                           orphan_entry_bh, orphan_dir);
 794                 if (status < 0) {
 795                         mlog_errno(status);
 796                         goto leave;
 797                 }
 798         }
 799
 800         /* delete the name from the parent dir */
 801         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 802         if (status < 0) {
 803                 mlog_errno(status);
 804                 goto leave;
 805         }
 806
 807         if (S_ISDIR(inode->i_mode))
 808                 drop_nlink(inode);
 809         drop_nlink(inode);
 810         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 811
 812         status = ocfs2_journal_dirty(handle, fe_bh);
 813         if (status < 0) {
 814                 mlog_errno(status);
 815                 goto leave;
 816         }
 817
 818         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 819         if (S_ISDIR(inode->i_mode))
 820                 drop_nlink(dir);
 821
 822         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 823         if (status < 0) {
 824                 mlog_errno(status);
 825                 if (S_ISDIR(inode->i_mode))
 826                         inc_nlink(dir);
 827         }
 828
 829 leave:
 830         if (handle)
 831                 ocfs2_commit_trans(osb, handle);
 832
 833         if (child_locked)
 834                 ocfs2_meta_unlock(inode, 1);
 835
 836         ocfs2_meta_unlock(dir, 1);
 837
 838         if (orphan_dir) {
 839                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 840                 ocfs2_meta_unlock(orphan_dir, 1);
 841                 mutex_unlock(&orphan_dir->i_mutex);
 842                 iput(orphan_dir);
 843         }
 844
 845         if (fe_bh)
 846                 brelse(fe_bh);
 847
 848         if (dirent_bh)
 849                 brelse(dirent_bh);
 850
 851         if (parent_node_bh)
 852                 brelse(parent_node_bh);
 853
 854         if (orphan_entry_bh)
 855                 brelse(orphan_entry_bh);
 856
 857         mlog_exit(status);
 858
 859         return status;
 860 }
 861
 862 /*
 863  * The only place this should be used is rename!
 864  * if they have the same id, then the 1st one is the only one locked.
 865  */
 866 static int ocfs2_double_lock(struct ocfs2_super *osb,
 867                              struct buffer_head **bh1,
 868                              struct inode *inode1,
 869                              struct buffer_head **bh2,
 870                              struct inode *inode2)
 871 {
 872         int status;
 873         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 874         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 875         struct buffer_head **tmpbh;
 876         struct inode *tmpinode;
 877
 878         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 879                    (unsigned long long)oi1->ip_blkno,
 880                    (unsigned long long)oi2->ip_blkno);
 881
 882         if (*bh1)
 883                 *bh1 = NULL;
 884         if (*bh2)
 885                 *bh2 = NULL;
 886
 887         /* we always want to lock the one with the lower lockid first. */
 888         if (oi1->ip_blkno != oi2->ip_blkno) {
 889                 if (oi1->ip_blkno < oi2->ip_blkno) {
 890                         /* switch id1 and id2 around */
 891                         mlog(0, "switching them around...\n");
 892                         tmpbh = bh2;
 893                         bh2 = bh1;
 894                         bh1 = tmpbh;
 895
 896                         tmpinode = inode2;
 897                         inode2 = inode1;
 898                         inode1 = tmpinode;
 899                 }
 900                 /* lock id2 */
 901                 status = ocfs2_meta_lock(inode2, bh2, 1);
 902                 if (status < 0) {
 903                         if (status != -ENOENT)
 904                                 mlog_errno(status);
 905                         goto bail;
 906                 }
 907         }
 908
 909         /* lock id1 */
 910         status = ocfs2_meta_lock(inode1, bh1, 1);
 911         if (status < 0) {
 912                 /*
 913                  * An error return must mean that no cluster locks
 914                  * were held on function exit.
 915                  */
 916                 if (oi1->ip_blkno != oi2->ip_blkno)
 917                         ocfs2_meta_unlock(inode2, 1);
 918
 919                 if (status != -ENOENT)
 920                         mlog_errno(status);
 921         }
 922
 923 bail:
 924         mlog_exit(status);
 925         return status;
 926 }
 927
 928 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
 929 {
 930         ocfs2_meta_unlock(inode1, 1);
 931
 932         if (inode1 != inode2)
 933                 ocfs2_meta_unlock(inode2, 1);
 934 }
 935
 936 #define PARENT_INO(buffer) \
 937         ((struct ocfs2_dir_entry *) \
 938          ((char *)buffer + \
 939           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
 940
 941 static int ocfs2_rename(struct inode *old_dir,
 942                         struct dentry *old_dentry,
 943                         struct inode *new_dir,
 944                         struct dentry *new_dentry)
 945 {
 946         int status = 0, rename_lock = 0, parents_locked = 0;
 947         int old_child_locked = 0, new_child_locked = 0;
 948         struct inode *old_inode = old_dentry->d_inode;
 949         struct inode *new_inode = new_dentry->d_inode;
 950         struct inode *orphan_dir = NULL;
 951         struct ocfs2_dinode *newfe = NULL;
 952         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 953         struct buffer_head *orphan_entry_bh = NULL;
 954         struct buffer_head *newfe_bh = NULL;
 955         struct buffer_head *old_inode_bh = NULL;
 956         struct buffer_head *insert_entry_bh = NULL;
 957         struct ocfs2_super *osb = NULL;
 958         u64 newfe_blkno;
 959         handle_t *handle = NULL;
 960         struct buffer_head *old_dir_bh = NULL;
 961         struct buffer_head *new_dir_bh = NULL;
 962         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
 963                                                                // and new_dentry
 964         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
 965         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
 966                                                     // this is the 1st dirent bh
 967         nlink_t old_dir_nlink = old_dir->i_nlink;
 968         struct ocfs2_dinode *old_di;
 969
 970         /* At some point it might be nice to break this function up a
 971          * bit. */
 972
 973         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
 974                    old_dir, old_dentry, new_dir, new_dentry,
 975                    old_dentry->d_name.len, old_dentry->d_name.name,
 976                    new_dentry->d_name.len, new_dentry->d_name.name);
 977
 978         osb = OCFS2_SB(old_dir->i_sb);
 979
 980         if (new_inode) {
 981                 if (!igrab(new_inode))
 982                         BUG();
 983         }
 984
 985         /* Assume a directory hierarchy thusly:
 986          * a/b/c
 987          * a/d
 988          * a,b,c, and d are all directories.
 989          *
 990          * from cwd of 'a' on both nodes:
 991          * node1: mv b/c d
 992          * node2: mv d   b/c
 993          *
 994          * And that's why, just like the VFS, we need a file system
 995          * rename lock. */
 996         if (old_dentry != new_dentry) {
 997                 status = ocfs2_rename_lock(osb);
 998                 if (status < 0) {
 999                         mlog_errno(status);
1000                         goto bail;
1001                 }
1002                 rename_lock = 1;
1003         }
1004
1005         /* if old and new are the same, this'll just do one lock. */
1006         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1007                                    &new_dir_bh, new_dir);
1008         if (status < 0) {
1009                 mlog_errno(status);
1010                 goto bail;
1011         }
1012         parents_locked = 1;
1013
1014         /* make sure both dirs have bhs
1015          * get an extra ref on old_dir_bh if old==new */
1016         if (!new_dir_bh) {
1017                 if (old_dir_bh) {
1018                         new_dir_bh = old_dir_bh;
1019                         get_bh(new_dir_bh);
1020                 } else {
1021                         mlog(ML_ERROR, "no old_dir_bh!\n");
1022                         status = -EIO;
1023                         goto bail;
1024                 }
1025         }
1026
1027         /*
1028          * Aside from allowing a meta data update, the locking here
1029          * also ensures that the vote thread on other nodes won't have
1030          * to concurrently downconvert the inode and the dentry locks.
1031          */
1032         status = ocfs2_meta_lock(old_inode, &old_inode_bh, 1);
1033         if (status < 0) {
1034                 if (status != -ENOENT)
1035                         mlog_errno(status);
1036                 goto bail;
1037         }
1038         old_child_locked = 1;
1039
1040         status = ocfs2_remote_dentry_delete(old_dentry);
1041         if (status < 0) {
1042                 mlog_errno(status);
1043                 goto bail;
1044         }
1045
1046         if (S_ISDIR(old_inode->i_mode)) {
1047                 status = -EIO;
1048                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1049                 if (!old_inode_de_bh)
1050                         goto bail;
1051
1052                 status = -EIO;
1053                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1054                     OCFS2_I(old_dir)->ip_blkno)
1055                         goto bail;
1056                 status = -EMLINK;
1057                 if (!new_inode && new_dir!=old_dir &&
1058                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1059                         goto bail;
1060         }
1061
1062         status = -ENOENT;
1063         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1064                                      old_dentry->d_name.len,
1065                                      old_dir, &old_de);
1066         if (!old_de_bh)
1067                 goto bail;
1068
1069         /*
1070          *  Check for inode number is _not_ due to possible IO errors.
1071          *  We might rmdir the source, keep it as pwd of some process
1072          *  and merrily kill the link to whatever was created under the
1073          *  same name. Goodbye sticky bit ;-<
1074          */
1075         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1076                 goto bail;
1077
1078         /* check if the target already exists (in which case we need
1079          * to delete it */
1080         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1081                                           new_dentry->d_name.len,
1082                                           &newfe_blkno, new_dir, &new_de_bh,
1083                                           &new_de);
1084         /* The only error we allow here is -ENOENT because the new
1085          * file not existing is perfectly valid. */
1086         if ((status < 0) && (status != -ENOENT)) {
1087                 /* If we cannot find the file specified we should just */
1088                 /* return the error... */
1089                 mlog_errno(status);
1090                 goto bail;
1091         }
1092
1093         if (!new_de && new_inode)
1094                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1095                      "directory!", new_inode->i_ino);
1096
1097         /* In case we need to overwrite an existing file, we blow it
1098          * away first */
1099         if (new_de) {
1100                 /* VFS didn't think there existed an inode here, but
1101                  * someone else in the cluster must have raced our
1102                  * rename to create one. Today we error cleanly, in
1103                  * the future we should consider calling iget to build
1104                  * a new struct inode for this entry. */
1105                 if (!new_inode) {
1106                         status = -EACCES;
1107
1108                         mlog(0, "We found an inode for name %.*s but VFS "
1109                              "didn't give us one.\n", new_dentry->d_name.len,
1110                              new_dentry->d_name.name);
1111                         goto bail;
1112                 }
1113
1114                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1115                         status = -EACCES;
1116
1117                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1118                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1119                              (unsigned long long)newfe_blkno,
1120                              OCFS2_I(new_inode)->ip_flags);
1121                         goto bail;
1122                 }
1123
1124                 status = ocfs2_meta_lock(new_inode, &newfe_bh, 1);
1125                 if (status < 0) {
1126                         if (status != -ENOENT)
1127                                 mlog_errno(status);
1128                         goto bail;
1129                 }
1130                 new_child_locked = 1;
1131
1132                 status = ocfs2_remote_dentry_delete(new_dentry);
1133                 if (status < 0) {
1134                         mlog_errno(status);
1135                         goto bail;
1136                 }
1137
1138                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1139
1140                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1141                      "newfebh=%p bhblocknr=%llu\n", new_de,
1142                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1143                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1144
1145                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1146                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1147                                                           new_inode,
1148                                                           orphan_name,
1149                                                           &orphan_entry_bh);
1150                         if (status < 0) {
1151                                 mlog_errno(status);
1152                                 goto bail;
1153                         }
1154                 }
1155         } else {
1156                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1157
1158                 status = ocfs2_check_dir_for_entry(new_dir,
1159                                                    new_dentry->d_name.name,
1160                                                    new_dentry->d_name.len);
1161                 if (status)
1162                         goto bail;
1163
1164                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1165                                                       new_dentry->d_name.name,
1166                                                       new_dentry->d_name.len,
1167                                                       &insert_entry_bh);
1168                 if (status < 0) {
1169                         mlog_errno(status);
1170                         goto bail;
1171                 }
1172         }
1173
1174         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1175         if (IS_ERR(handle)) {
1176                 status = PTR_ERR(handle);
1177                 handle = NULL;
1178                 mlog_errno(status);
1179                 goto bail;
1180         }
1181
1182         if (new_de) {
1183                 if (S_ISDIR(new_inode->i_mode)) {
1184                         if (!ocfs2_empty_dir(new_inode) ||
1185                             new_inode->i_nlink != 2) {
1186                                 status = -ENOTEMPTY;
1187                                 goto bail;
1188                         }
1189                 }
1190                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1191                                               OCFS2_JOURNAL_ACCESS_WRITE);
1192                 if (status < 0) {
1193                         mlog_errno(status);
1194                         goto bail;
1195                 }
1196
1197                 if (S_ISDIR(new_inode->i_mode) ||
1198                     (newfe->i_links_count == cpu_to_le16(1))){
1199                         status = ocfs2_orphan_add(osb, handle, new_inode,
1200                                                   newfe, orphan_name,
1201                                                   orphan_entry_bh, orphan_dir);
1202                         if (status < 0) {
1203                                 mlog_errno(status);
1204                                 goto bail;
1205                         }
1206                 }
1207
1208                 /* change the dirent to point to the correct inode */
1209                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1210                                               OCFS2_JOURNAL_ACCESS_WRITE);
1211                 if (status < 0) {
1212                         mlog_errno(status);
1213                         goto bail;
1214                 }
1215                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1216                 new_de->file_type = old_de->file_type;
1217                 new_dir->i_version++;
1218                 status = ocfs2_journal_dirty(handle, new_de_bh);
1219                 if (status < 0) {
1220                         mlog_errno(status);
1221                         goto bail;
1222                 }
1223
1224                 if (S_ISDIR(new_inode->i_mode))
1225                         newfe->i_links_count = 0;
1226                 else
1227                         le16_add_cpu(&newfe->i_links_count, -1);
1228
1229                 status = ocfs2_journal_dirty(handle, newfe_bh);
1230                 if (status < 0) {
1231                         mlog_errno(status);
1232                         goto bail;
1233                 }
1234         } else {
1235                 /* if the name was not found in new_dir, add it now */
1236                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1237                                          OCFS2_I(old_inode)->ip_blkno,
1238                                          new_dir_bh, insert_entry_bh);
1239         }
1240
1241         old_inode->i_ctime = CURRENT_TIME;
1242         mark_inode_dirty(old_inode);
1243
1244         status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1245                                       OCFS2_JOURNAL_ACCESS_WRITE);
1246         if (status >= 0) {
1247                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1248
1249                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1250                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1251
1252                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1253                 if (status < 0)
1254                         mlog_errno(status);
1255         } else
1256                 mlog_errno(status);
1257
1258         /* now that the name has been added to new_dir, remove the old name */
1259         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1260         if (status < 0) {
1261                 mlog_errno(status);
1262                 goto bail;
1263         }
1264
1265         if (new_inode) {
1266                 new_inode->i_nlink--;
1267                 new_inode->i_ctime = CURRENT_TIME;
1268         }
1269         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1270         if (old_inode_de_bh) {
1271                 status = ocfs2_journal_access(handle, old_inode,
1272                                              old_inode_de_bh,
1273                                              OCFS2_JOURNAL_ACCESS_WRITE);
1274                 PARENT_INO(old_inode_de_bh->b_data) =
1275                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1276                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1277                 old_dir->i_nlink--;
1278                 if (new_inode) {
1279                         new_inode->i_nlink--;
1280                 } else {
1281                         inc_nlink(new_dir);
1282                         mark_inode_dirty(new_dir);
1283                 }
1284         }
1285         mark_inode_dirty(old_dir);
1286         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1287         if (new_inode) {
1288                 mark_inode_dirty(new_inode);
1289                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1290         }
1291
1292         if (old_dir != new_dir) {
1293                 /* Keep the same times on both directories.*/
1294                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1295
1296                 /*
1297                  * This will also pick up the i_nlink change from the
1298                  * block above.
1299                  */
1300                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1301         }
1302
1303         if (old_dir_nlink != old_dir->i_nlink) {
1304                 if (!old_dir_bh) {
1305                         mlog(ML_ERROR, "need to change nlink for old dir "
1306                              "%llu from %d to %d but bh is NULL!\n",
1307                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1308                              (int)old_dir_nlink, old_dir->i_nlink);
1309                 } else {
1310                         struct ocfs2_dinode *fe;
1311                         status = ocfs2_journal_access(handle, old_dir,
1312                                                       old_dir_bh,
1313                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1314                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1315                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1316                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1317                 }
1318         }
1319
1320         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1321         status = 0;
1322 bail:
1323         if (rename_lock)
1324                 ocfs2_rename_unlock(osb);
1325
1326         if (handle)
1327                 ocfs2_commit_trans(osb, handle);
1328
1329         if (parents_locked)
1330                 ocfs2_double_unlock(old_dir, new_dir);
1331
1332         if (old_child_locked)
1333                 ocfs2_meta_unlock(old_inode, 1);
1334
1335         if (new_child_locked)
1336                 ocfs2_meta_unlock(new_inode, 1);
1337
1338         if (orphan_dir) {
1339                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1340                 ocfs2_meta_unlock(orphan_dir, 1);
1341                 mutex_unlock(&orphan_dir->i_mutex);
1342                 iput(orphan_dir);
1343         }
1344
1345         if (new_inode)
1346                 sync_mapping_buffers(old_inode->i_mapping);
1347
1348         if (new_inode)
1349                 iput(new_inode);
1350         if (newfe_bh)
1351                 brelse(newfe_bh);
1352         if (old_inode_bh)
1353                 brelse(old_inode_bh);
1354         if (old_dir_bh)
1355                 brelse(old_dir_bh);
1356         if (new_dir_bh)
1357                 brelse(new_dir_bh);
1358         if (new_de_bh)
1359                 brelse(new_de_bh);
1360         if (old_de_bh)
1361                 brelse(old_de_bh);
1362         if (old_inode_de_bh)
1363                 brelse(old_inode_de_bh);
1364         if (orphan_entry_bh)
1365                 brelse(orphan_entry_bh);
1366         if (insert_entry_bh)
1367                 brelse(insert_entry_bh);
1368
1369         mlog_exit(status);
1370
1371         return status;
1372 }
1373
1374 /*
1375  * we expect i_size = strlen(symname). Copy symname into the file
1376  * data, including the null terminator.
1377  */
1378 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1379                                      handle_t *handle,
1380                                      struct inode *inode,
1381                                      const char *symname)
1382 {
1383         struct buffer_head **bhs = NULL;
1384         const char *c;
1385         struct super_block *sb = osb->sb;
1386         u64 p_blkno, p_blocks;
1387         int virtual, blocks, status, i, bytes_left;
1388
1389         bytes_left = i_size_read(inode) + 1;
1390         /* we can't trust i_blocks because we're actually going to
1391          * write i_size + 1 bytes. */
1392         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1393
1394         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1395                         (unsigned long long)inode->i_blocks,
1396                         i_size_read(inode), blocks);
1397
1398         /* Sanity check -- make sure we're going to fit. */
1399         if (bytes_left >
1400             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1401                 status = -EIO;
1402                 mlog_errno(status);
1403                 goto bail;
1404         }
1405
1406         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1407         if (!bhs) {
1408                 status = -ENOMEM;
1409                 mlog_errno(status);
1410                 goto bail;
1411         }
1412
1413         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1414                                              NULL);
1415         if (status < 0) {
1416                 mlog_errno(status);
1417                 goto bail;
1418         }
1419
1420         /* links can never be larger than one cluster so we know this
1421          * is all going to be contiguous, but do a sanity check
1422          * anyway. */
1423         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1424                 status = -EIO;
1425                 mlog_errno(status);
1426                 goto bail;
1427         }
1428
1429         virtual = 0;
1430         while(bytes_left > 0) {
1431                 c = &symname[virtual * sb->s_blocksize];
1432
1433                 bhs[virtual] = sb_getblk(sb, p_blkno);
1434                 if (!bhs[virtual]) {
1435                         status = -ENOMEM;
1436                         mlog_errno(status);
1437                         goto bail;
1438                 }
1439                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1440
1441                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1442                                               OCFS2_JOURNAL_ACCESS_CREATE);
1443                 if (status < 0) {
1444                         mlog_errno(status);
1445                         goto bail;
1446                 }
1447
1448                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1449
1450                 memcpy(bhs[virtual]->b_data, c,
1451                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1452                        bytes_left);
1453
1454                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1455                 if (status < 0) {
1456                         mlog_errno(status);
1457                         goto bail;
1458                 }
1459
1460                 virtual++;
1461                 p_blkno++;
1462                 bytes_left -= sb->s_blocksize;
1463         }
1464
1465         status = 0;
1466 bail:
1467
1468         if (bhs) {
1469                 for(i = 0; i < blocks; i++)
1470                         if (bhs[i])
1471                                 brelse(bhs[i]);
1472                 kfree(bhs);
1473         }
1474
1475         mlog_exit(status);
1476         return status;
1477 }
1478
1479 static int ocfs2_symlink(struct inode *dir,
1480                          struct dentry *dentry,
1481                          const char *symname)
1482 {
1483         int status, l, credits;
1484         u64 newsize;
1485         struct ocfs2_super *osb = NULL;
1486         struct inode *inode = NULL;
1487         struct super_block *sb;
1488         struct buffer_head *new_fe_bh = NULL;
1489         struct buffer_head *de_bh = NULL;
1490         struct buffer_head *parent_fe_bh = NULL;
1491         struct ocfs2_dinode *fe = NULL;
1492         struct ocfs2_dinode *dirfe;
1493         handle_t *handle = NULL;
1494         struct ocfs2_alloc_context *inode_ac = NULL;
1495         struct ocfs2_alloc_context *data_ac = NULL;
1496
1497         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1498                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1499
1500         sb = dir->i_sb;
1501         osb = OCFS2_SB(sb);
1502
1503         l = strlen(symname) + 1;
1504
1505         credits = ocfs2_calc_symlink_credits(sb);
1506
1507         /* lock the parent directory */
1508         status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
1509         if (status < 0) {
1510                 if (status != -ENOENT)
1511                         mlog_errno(status);
1512                 return status;
1513         }
1514
1515         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1516         if (!dirfe->i_links_count) {
1517                 /* can't make a file in a deleted directory. */
1518                 status = -ENOENT;
1519                 goto bail;
1520         }
1521
1522         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1523                                            dentry->d_name.len);
1524         if (status)
1525                 goto bail;
1526
1527         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1528                                               dentry->d_name.name,
1529                                               dentry->d_name.len, &de_bh);
1530         if (status < 0) {
1531                 mlog_errno(status);
1532                 goto bail;
1533         }
1534
1535         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1536         if (status < 0) {
1537                 if (status != -ENOSPC)
1538                         mlog_errno(status);
1539                 goto bail;
1540         }
1541
1542         /* don't reserve bitmap space for fast symlinks. */
1543         if (l > ocfs2_fast_symlink_chars(sb)) {
1544                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1545                 if (status < 0) {
1546                         if (status != -ENOSPC)
1547                                 mlog_errno(status);
1548                         goto bail;
1549                 }
1550         }
1551
1552         handle = ocfs2_start_trans(osb, credits);
1553         if (IS_ERR(handle)) {
1554                 status = PTR_ERR(handle);
1555                 handle = NULL;
1556                 mlog_errno(status);
1557                 goto bail;
1558         }
1559
1560         status = ocfs2_mknod_locked(osb, dir, dentry,
1561                                     S_IFLNK | S_IRWXUGO, 0,
1562                                     &new_fe_bh, parent_fe_bh, handle,
1563                                     &inode, inode_ac);
1564         if (status < 0) {
1565                 mlog_errno(status);
1566                 goto bail;
1567         }
1568
1569         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1570         inode->i_rdev = 0;
1571         newsize = l - 1;
1572         if (l > ocfs2_fast_symlink_chars(sb)) {
1573                 u32 offset = 0;
1574
1575                 inode->i_op = &ocfs2_symlink_inode_operations;
1576                 status = ocfs2_do_extend_allocation(osb, inode, &offset, 1, 0,
1577                                                     new_fe_bh,
1578                                                     handle, data_ac, NULL,
1579                                                     NULL);
1580                 if (status < 0) {
1581                         if (status != -ENOSPC && status != -EINTR) {
1582                                 mlog(ML_ERROR,
1583                                      "Failed to extend file to %llu\n",
1584                                      (unsigned long long)newsize);
1585                                 mlog_errno(status);
1586                                 status = -ENOSPC;
1587                         }
1588                         goto bail;
1589                 }
1590                 i_size_write(inode, newsize);
1591                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1592         } else {
1593                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1594                 memcpy((char *) fe->id2.i_symlink, symname, l);
1595                 i_size_write(inode, newsize);
1596                 inode->i_blocks = 0;
1597         }
1598
1599         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1600         if (status < 0) {
1601                 mlog_errno(status);
1602                 goto bail;
1603         }
1604
1605         if (!ocfs2_inode_is_fast_symlink(inode)) {
1606                 status = ocfs2_create_symlink_data(osb, handle, inode,
1607                                                    symname);
1608                 if (status < 0) {
1609                         mlog_errno(status);
1610                         goto bail;
1611                 }
1612         }
1613
1614         status = ocfs2_add_entry(handle, dentry, inode,
1615                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1616                                  de_bh);
1617         if (status < 0) {
1618                 mlog_errno(status);
1619                 goto bail;
1620         }
1621
1622         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1623         if (status) {
1624                 mlog_errno(status);
1625                 goto bail;
1626         }
1627
1628         insert_inode_hash(inode);
1629         dentry->d_op = &ocfs2_dentry_ops;
1630         d_instantiate(dentry, inode);
1631 bail:
1632         if (handle)
1633                 ocfs2_commit_trans(osb, handle);
1634
1635         ocfs2_meta_unlock(dir, 1);
1636
1637         if (new_fe_bh)
1638                 brelse(new_fe_bh);
1639         if (parent_fe_bh)
1640                 brelse(parent_fe_bh);
1641         if (de_bh)
1642                 brelse(de_bh);
1643         if (inode_ac)
1644                 ocfs2_free_alloc_context(inode_ac);
1645         if (data_ac)
1646                 ocfs2_free_alloc_context(data_ac);
1647         if ((status < 0) && inode)
1648                 iput(inode);
1649
1650         mlog_exit(status);
1651
1652         return status;
1653 }
1654
1655 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1656 {
1657         int status, namelen;
1658
1659         mlog_entry_void();
1660
1661         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1662                            (long long)blkno);
1663         if (namelen <= 0) {
1664                 if (namelen)
1665                         status = namelen;
1666                 else
1667                         status = -EINVAL;
1668                 mlog_errno(status);
1669                 goto bail;
1670         }
1671         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1672                 status = -EINVAL;
1673                 mlog_errno(status);
1674                 goto bail;
1675         }
1676
1677         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1678              namelen);
1679
1680         status = 0;
1681 bail:
1682         mlog_exit(status);
1683         return status;
1684 }
1685
1686 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1687                                     struct inode **ret_orphan_dir,
1688                                     struct inode *inode,
1689                                     char *name,
1690                                     struct buffer_head **de_bh)
1691 {
1692         struct inode *orphan_dir_inode;
1693         struct buffer_head *orphan_dir_bh = NULL;
1694         int status = 0;
1695
1696         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1697         if (status < 0) {
1698                 mlog_errno(status);
1699                 return status;
1700         }
1701
1702         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1703                                                        ORPHAN_DIR_SYSTEM_INODE,
1704                                                        osb->slot_num);
1705         if (!orphan_dir_inode) {
1706                 status = -ENOENT;
1707                 mlog_errno(status);
1708                 return status;
1709         }
1710
1711         mutex_lock(&orphan_dir_inode->i_mutex);
1712
1713         status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1714         if (status < 0) {
1715                 mlog_errno(status);
1716                 goto leave;
1717         }
1718
1719         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1720                                               orphan_dir_bh, name,
1721                                               OCFS2_ORPHAN_NAMELEN, de_bh);
1722         if (status < 0) {
1723                 ocfs2_meta_unlock(orphan_dir_inode, 1);
1724
1725                 mlog_errno(status);
1726                 goto leave;
1727         }
1728
1729         *ret_orphan_dir = orphan_dir_inode;
1730
1731 leave:
1732         if (status) {
1733                 mutex_unlock(&orphan_dir_inode->i_mutex);
1734                 iput(orphan_dir_inode);
1735         }
1736
1737         if (orphan_dir_bh)
1738                 brelse(orphan_dir_bh);
1739
1740         mlog_exit(status);
1741         return status;
1742 }
1743
1744 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1745                             handle_t *handle,
1746                             struct inode *inode,
1747                             struct ocfs2_dinode *fe,
1748                             char *name,
1749                             struct buffer_head *de_bh,
1750                             struct inode *orphan_dir_inode)
1751 {
1752         struct buffer_head *orphan_dir_bh = NULL;
1753         int status = 0;
1754         struct ocfs2_dinode *orphan_fe;
1755
1756         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1757
1758         status = ocfs2_read_block(osb,
1759                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
1760                                   &orphan_dir_bh, OCFS2_BH_CACHED,
1761                                   orphan_dir_inode);
1762         if (status < 0) {
1763                 mlog_errno(status);
1764                 goto leave;
1765         }
1766
1767         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1768                                       OCFS2_JOURNAL_ACCESS_WRITE);
1769         if (status < 0) {
1770                 mlog_errno(status);
1771                 goto leave;
1772         }
1773
1774         /* we're a cluster, and nlink can change on disk from
1775          * underneath us... */
1776         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1777         if (S_ISDIR(inode->i_mode))
1778                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1779         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1780
1781         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1782         if (status < 0) {
1783                 mlog_errno(status);
1784                 goto leave;
1785         }
1786
1787         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1788                                    OCFS2_ORPHAN_NAMELEN, inode,
1789                                    OCFS2_I(inode)->ip_blkno,
1790                                    orphan_dir_bh, de_bh);
1791         if (status < 0) {
1792                 mlog_errno(status);
1793                 goto leave;
1794         }
1795
1796         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1797
1798         /* Record which orphan dir our inode now resides
1799          * in. delete_inode will use this to determine which orphan
1800          * dir to lock. */
1801         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1802
1803         mlog(0, "Inode %llu orphaned in slot %d\n",
1804              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1805
1806 leave:
1807         if (orphan_dir_bh)
1808                 brelse(orphan_dir_bh);
1809
1810         mlog_exit(status);
1811         return status;
1812 }
1813
1814 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1815 int ocfs2_orphan_del(struct ocfs2_super *osb,
1816                      handle_t *handle,
1817                      struct inode *orphan_dir_inode,
1818                      struct inode *inode,
1819                      struct buffer_head *orphan_dir_bh)
1820 {
1821         char name[OCFS2_ORPHAN_NAMELEN + 1];
1822         struct ocfs2_dinode *orphan_fe;
1823         int status = 0;
1824         struct buffer_head *target_de_bh = NULL;
1825         struct ocfs2_dir_entry *target_de = NULL;
1826
1827         mlog_entry_void();
1828
1829         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1830         if (status < 0) {
1831                 mlog_errno(status);
1832                 goto leave;
1833         }
1834
1835         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1836              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1837              OCFS2_ORPHAN_NAMELEN);
1838
1839         /* find it's spot in the orphan directory */
1840         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1841                                         orphan_dir_inode, &target_de);
1842         if (!target_de_bh) {
1843                 status = -ENOENT;
1844                 mlog_errno(status);
1845                 goto leave;
1846         }
1847
1848         /* remove it from the orphan directory */
1849         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1850                                     target_de_bh);
1851         if (status < 0) {
1852                 mlog_errno(status);
1853                 goto leave;
1854         }
1855
1856         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
1857                                       OCFS2_JOURNAL_ACCESS_WRITE);
1858         if (status < 0) {
1859                 mlog_errno(status);
1860                 goto leave;
1861         }
1862
1863         /* do the i_nlink dance! :) */
1864         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1865         if (S_ISDIR(inode->i_mode))
1866                 le16_add_cpu(&orphan_fe->i_links_count, -1);
1867         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1868
1869         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1870         if (status < 0) {
1871                 mlog_errno(status);
1872                 goto leave;
1873         }
1874
1875 leave:
1876         if (target_de_bh)
1877                 brelse(target_de_bh);
1878
1879         mlog_exit(status);
1880         return status;
1881 }
1882
1883 const struct inode_operations ocfs2_dir_iops = {
1884         .create         = ocfs2_create,
1885         .lookup         = ocfs2_lookup,
1886         .link           = ocfs2_link,
1887         .unlink         = ocfs2_unlink,
1888         .rmdir          = ocfs2_unlink,
1889         .symlink        = ocfs2_symlink,
1890         .mkdir          = ocfs2_mkdir,
1891         .mknod          = ocfs2_mknod,
1892         .rename         = ocfs2_rename,
1893         .setattr        = ocfs2_setattr,
1894         .getattr        = ocfs2_getattr,
1895         .permission     = ocfs2_permission,
1896 };