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 "xattr.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_inode_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, 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 downconvert thread and have a stale dentry. */
 180         ocfs2_inode_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_inode_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         /* Reserve a cluster if creating an extent based directory. */
 254         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 255                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
 256                 if (status < 0) {
 257                         if (status != -ENOSPC)
 258                                 mlog_errno(status);
 259                         goto leave;
 260                 }
 261         }
 262
 263         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
 264         if (IS_ERR(handle)) {
 265                 status = PTR_ERR(handle);
 266                 handle = NULL;
 267                 mlog_errno(status);
 268                 goto leave;
 269         }
 270
 271         /* do the real work now. */
 272         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
 273                                     &new_fe_bh, parent_fe_bh, handle,
 274                                     &inode, inode_ac);
 275         if (status < 0) {
 276                 mlog_errno(status);
 277                 goto leave;
 278         }
 279
 280         if (S_ISDIR(mode)) {
 281                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 282                                             new_fe_bh, data_ac);
 283                 if (status < 0) {
 284                         mlog_errno(status);
 285                         goto leave;
 286                 }
 287
 288                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
 289                                               OCFS2_JOURNAL_ACCESS_WRITE);
 290                 if (status < 0) {
 291                         mlog_errno(status);
 292                         goto leave;
 293                 }
 294                 le16_add_cpu(&dirfe->i_links_count, 1);
 295                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 296                 if (status < 0) {
 297                         mlog_errno(status);
 298                         goto leave;
 299                 }
 300                 inc_nlink(dir);
 301         }
 302
 303         status = ocfs2_add_entry(handle, dentry, inode,
 304                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 305                                  de_bh);
 306         if (status < 0) {
 307                 mlog_errno(status);
 308                 goto leave;
 309         }
 310
 311         status = ocfs2_dentry_attach_lock(dentry, inode,
 312                                           OCFS2_I(dir)->ip_blkno);
 313         if (status) {
 314                 mlog_errno(status);
 315                 goto leave;
 316         }
 317
 318         insert_inode_hash(inode);
 319         dentry->d_op = &ocfs2_dentry_ops;
 320         d_instantiate(dentry, inode);
 321         status = 0;
 322 leave:
 323         if (handle)
 324                 ocfs2_commit_trans(osb, handle);
 325
 326         ocfs2_inode_unlock(dir, 1);
 327
 328         if (status == -ENOSPC)
 329                 mlog(0, "Disk is full\n");
 330
 331         brelse(new_fe_bh);
 332         brelse(de_bh);
 333         brelse(parent_fe_bh);
 334
 335         if ((status < 0) && inode)
 336                 iput(inode);
 337
 338         if (inode_ac)
 339                 ocfs2_free_alloc_context(inode_ac);
 340
 341         if (data_ac)
 342                 ocfs2_free_alloc_context(data_ac);
 343
 344         mlog_exit(status);
 345
 346         return status;
 347 }
 348
 349 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 350                               struct inode *dir,
 351                               struct dentry *dentry, int mode,
 352                               dev_t dev,
 353                               struct buffer_head **new_fe_bh,
 354                               struct buffer_head *parent_fe_bh,
 355                               handle_t *handle,
 356                               struct inode **ret_inode,
 357                               struct ocfs2_alloc_context *inode_ac)
 358 {
 359         int status = 0;
 360         struct ocfs2_dinode *fe = NULL;
 361         struct ocfs2_extent_list *fel;
 362         u64 fe_blkno = 0;
 363         u16 suballoc_bit;
 364         struct inode *inode = NULL;
 365
 366         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 367                    (unsigned long)dev, dentry->d_name.len,
 368                    dentry->d_name.name);
 369
 370         *new_fe_bh = NULL;
 371         *ret_inode = NULL;
 372
 373         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
 374                                        &fe_blkno);
 375         if (status < 0) {
 376                 mlog_errno(status);
 377                 goto leave;
 378         }
 379
 380         inode = new_inode(dir->i_sb);
 381         if (!inode) {
 382                 status = -ENOMEM;
 383                 mlog(ML_ERROR, "new_inode failed!\n");
 384                 goto leave;
 385         }
 386
 387         /* populate as many fields early on as possible - many of
 388          * these are used by the support functions here and in
 389          * callers. */
 390         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 391         OCFS2_I(inode)->ip_blkno = fe_blkno;
 392         if (S_ISDIR(mode))
 393                 inode->i_nlink = 2;
 394         else
 395                 inode->i_nlink = 1;
 396         inode->i_mode = mode;
 397         spin_lock(&osb->osb_lock);
 398         inode->i_generation = osb->s_next_generation++;
 399         spin_unlock(&osb->osb_lock);
 400
 401         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 402         if (!*new_fe_bh) {
 403                 status = -EIO;
 404                 mlog_errno(status);
 405                 goto leave;
 406         }
 407         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
 408
 409         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
 410                                       OCFS2_JOURNAL_ACCESS_CREATE);
 411         if (status < 0) {
 412                 mlog_errno(status);
 413                 goto leave;
 414         }
 415
 416         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 417         memset(fe, 0, osb->sb->s_blocksize);
 418
 419         fe->i_generation = cpu_to_le32(inode->i_generation);
 420         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 421         fe->i_blkno = cpu_to_le64(fe_blkno);
 422         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 423         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 424         fe->i_uid = cpu_to_le32(current->fsuid);
 425         if (dir->i_mode & S_ISGID) {
 426                 fe->i_gid = cpu_to_le32(dir->i_gid);
 427                 if (S_ISDIR(mode))
 428                         mode |= S_ISGID;
 429         } else
 430                 fe->i_gid = cpu_to_le32(current->fsgid);
 431         fe->i_mode = cpu_to_le16(mode);
 432         if (S_ISCHR(mode) || S_ISBLK(mode))
 433                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 434
 435         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 436
 437         fe->i_last_eb_blk = 0;
 438         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 439         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 440         fe->i_atime = fe->i_ctime = fe->i_mtime =
 441                 cpu_to_le64(CURRENT_TIME.tv_sec);
 442         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 443                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 444         fe->i_dtime = 0;
 445
 446         /*
 447          * If supported, directories start with inline data.
 448          */
 449         if (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) {
 450                 u16 feat = le16_to_cpu(fe->i_dyn_features);
 451
 452                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 453
 454                 fe->id2.i_data.id_count = cpu_to_le16(ocfs2_max_inline_data(osb->sb));
 455         } else {
 456                 fel = &fe->id2.i_list;
 457                 fel->l_tree_depth = 0;
 458                 fel->l_next_free_rec = 0;
 459                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 460         }
 461
 462         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 463         if (status < 0) {
 464                 mlog_errno(status);
 465                 goto leave;
 466         }
 467
 468         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
 469                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
 470                      "i_blkno=%llu, i_ino=%lu\n",
 471                      (unsigned long long)(*new_fe_bh)->b_blocknr,
 472                      (unsigned long long)le64_to_cpu(fe->i_blkno),
 473                      inode->i_ino);
 474                 BUG();
 475         }
 476
 477         ocfs2_inode_set_new(osb, inode);
 478         if (!ocfs2_mount_local(osb)) {
 479                 status = ocfs2_create_new_inode_locks(inode);
 480                 if (status < 0)
 481                         mlog_errno(status);
 482         }
 483
 484         status = 0; /* error in ocfs2_create_new_inode_locks is not
 485                      * critical */
 486
 487         *ret_inode = inode;
 488 leave:
 489         if (status < 0) {
 490                 if (*new_fe_bh) {
 491                         brelse(*new_fe_bh);
 492                         *new_fe_bh = NULL;
 493                 }
 494                 if (inode) {
 495                         clear_nlink(inode);
 496                         iput(inode);
 497                 }
 498         }
 499
 500         mlog_exit(status);
 501         return status;
 502 }
 503
 504 static int ocfs2_mkdir(struct inode *dir,
 505                        struct dentry *dentry,
 506                        int mode)
 507 {
 508         int ret;
 509
 510         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 511                    dentry->d_name.len, dentry->d_name.name);
 512         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 513         mlog_exit(ret);
 514
 515         return ret;
 516 }
 517
 518 static int ocfs2_create(struct inode *dir,
 519                         struct dentry *dentry,
 520                         int mode,
 521                         struct nameidata *nd)
 522 {
 523         int ret;
 524
 525         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 526                    dentry->d_name.len, dentry->d_name.name);
 527         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 528         mlog_exit(ret);
 529
 530         return ret;
 531 }
 532
 533 static int ocfs2_link(struct dentry *old_dentry,
 534                       struct inode *dir,
 535                       struct dentry *dentry)
 536 {
 537         handle_t *handle;
 538         struct inode *inode = old_dentry->d_inode;
 539         int err;
 540         struct buffer_head *fe_bh = NULL;
 541         struct buffer_head *parent_fe_bh = NULL;
 542         struct buffer_head *de_bh = NULL;
 543         struct ocfs2_dinode *fe = NULL;
 544         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 545
 546         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 547                    old_dentry->d_name.len, old_dentry->d_name.name,
 548                    dentry->d_name.len, dentry->d_name.name);
 549
 550         if (S_ISDIR(inode->i_mode))
 551                 return -EPERM;
 552
 553         err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 554         if (err < 0) {
 555                 if (err != -ENOENT)
 556                         mlog_errno(err);
 557                 return err;
 558         }
 559
 560         if (!dir->i_nlink) {
 561                 err = -ENOENT;
 562                 goto out;
 563         }
 564
 565         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 566                                         dentry->d_name.len);
 567         if (err)
 568                 goto out;
 569
 570         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 571                                            dentry->d_name.name,
 572                                            dentry->d_name.len, &de_bh);
 573         if (err < 0) {
 574                 mlog_errno(err);
 575                 goto out;
 576         }
 577
 578         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 579         if (err < 0) {
 580                 if (err != -ENOENT)
 581                         mlog_errno(err);
 582                 goto out;
 583         }
 584
 585         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 586         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
 587                 err = -EMLINK;
 588                 goto out_unlock_inode;
 589         }
 590
 591         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
 592         if (IS_ERR(handle)) {
 593                 err = PTR_ERR(handle);
 594                 handle = NULL;
 595                 mlog_errno(err);
 596                 goto out_unlock_inode;
 597         }
 598
 599         err = ocfs2_journal_access(handle, inode, fe_bh,
 600                                    OCFS2_JOURNAL_ACCESS_WRITE);
 601         if (err < 0) {
 602                 mlog_errno(err);
 603                 goto out_commit;
 604         }
 605
 606         inc_nlink(inode);
 607         inode->i_ctime = CURRENT_TIME;
 608         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 609         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 610         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 611
 612         err = ocfs2_journal_dirty(handle, fe_bh);
 613         if (err < 0) {
 614                 le16_add_cpu(&fe->i_links_count, -1);
 615                 drop_nlink(inode);
 616                 mlog_errno(err);
 617                 goto out_commit;
 618         }
 619
 620         err = ocfs2_add_entry(handle, dentry, inode,
 621                               OCFS2_I(inode)->ip_blkno,
 622                               parent_fe_bh, de_bh);
 623         if (err) {
 624                 le16_add_cpu(&fe->i_links_count, -1);
 625                 drop_nlink(inode);
 626                 mlog_errno(err);
 627                 goto out_commit;
 628         }
 629
 630         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 631         if (err) {
 632                 mlog_errno(err);
 633                 goto out_commit;
 634         }
 635
 636         atomic_inc(&inode->i_count);
 637         dentry->d_op = &ocfs2_dentry_ops;
 638         d_instantiate(dentry, inode);
 639
 640 out_commit:
 641         ocfs2_commit_trans(osb, handle);
 642 out_unlock_inode:
 643         ocfs2_inode_unlock(inode, 1);
 644
 645 out:
 646         ocfs2_inode_unlock(dir, 1);
 647
 648         brelse(de_bh);
 649         brelse(fe_bh);
 650         brelse(parent_fe_bh);
 651
 652         mlog_exit(err);
 653
 654         return err;
 655 }
 656
 657 /*
 658  * Takes and drops an exclusive lock on the given dentry. This will
 659  * force other nodes to drop it.
 660  */
 661 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 662 {
 663         int ret;
 664
 665         ret = ocfs2_dentry_lock(dentry, 1);
 666         if (ret)
 667                 mlog_errno(ret);
 668         else
 669                 ocfs2_dentry_unlock(dentry, 1);
 670
 671         return ret;
 672 }
 673
 674 static inline int inode_is_unlinkable(struct inode *inode)
 675 {
 676         if (S_ISDIR(inode->i_mode)) {
 677                 if (inode->i_nlink == 2)
 678                         return 1;
 679                 return 0;
 680         }
 681
 682         if (inode->i_nlink == 1)
 683                 return 1;
 684         return 0;
 685 }
 686
 687 static int ocfs2_unlink(struct inode *dir,
 688                         struct dentry *dentry)
 689 {
 690         int status;
 691         int child_locked = 0;
 692         struct inode *inode = dentry->d_inode;
 693         struct inode *orphan_dir = NULL;
 694         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 695         u64 blkno;
 696         struct ocfs2_dinode *fe = NULL;
 697         struct buffer_head *fe_bh = NULL;
 698         struct buffer_head *parent_node_bh = NULL;
 699         handle_t *handle = NULL;
 700         struct ocfs2_dir_entry *dirent = NULL;
 701         struct buffer_head *dirent_bh = NULL;
 702         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 703         struct buffer_head *orphan_entry_bh = NULL;
 704
 705         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 706                    dentry->d_name.len, dentry->d_name.name);
 707
 708         BUG_ON(dentry->d_parent->d_inode != dir);
 709
 710         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 711
 712         if (inode == osb->root_inode) {
 713                 mlog(0, "Cannot delete the root directory\n");
 714                 return -EPERM;
 715         }
 716
 717         status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
 718         if (status < 0) {
 719                 if (status != -ENOENT)
 720                         mlog_errno(status);
 721                 return status;
 722         }
 723
 724         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 725                                           dentry->d_name.len, &blkno,
 726                                           dir, &dirent_bh, &dirent);
 727         if (status < 0) {
 728                 if (status != -ENOENT)
 729                         mlog_errno(status);
 730                 goto leave;
 731         }
 732
 733         if (OCFS2_I(inode)->ip_blkno != blkno) {
 734                 status = -ENOENT;
 735
 736                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 737                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 738                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 739                 goto leave;
 740         }
 741
 742         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 743         if (status < 0) {
 744                 if (status != -ENOENT)
 745                         mlog_errno(status);
 746                 goto leave;
 747         }
 748         child_locked = 1;
 749
 750         if (S_ISDIR(inode->i_mode)) {
 751                 if (!ocfs2_empty_dir(inode)) {
 752                         status = -ENOTEMPTY;
 753                         goto leave;
 754                 } else if (inode->i_nlink != 2) {
 755                         status = -ENOTEMPTY;
 756                         goto leave;
 757                 }
 758         }
 759
 760         status = ocfs2_remote_dentry_delete(dentry);
 761         if (status < 0) {
 762                 /* This remote delete should succeed under all normal
 763                  * circumstances. */
 764                 mlog_errno(status);
 765                 goto leave;
 766         }
 767
 768         if (inode_is_unlinkable(inode)) {
 769                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
 770                                                   orphan_name,
 771                                                   &orphan_entry_bh);
 772                 if (status < 0) {
 773                         mlog_errno(status);
 774                         goto leave;
 775                 }
 776         }
 777
 778         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
 779         if (IS_ERR(handle)) {
 780                 status = PTR_ERR(handle);
 781                 handle = NULL;
 782                 mlog_errno(status);
 783                 goto leave;
 784         }
 785
 786         status = ocfs2_journal_access(handle, inode, fe_bh,
 787                                       OCFS2_JOURNAL_ACCESS_WRITE);
 788         if (status < 0) {
 789                 mlog_errno(status);
 790                 goto leave;
 791         }
 792
 793         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 794
 795         if (inode_is_unlinkable(inode)) {
 796                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
 797                                           orphan_entry_bh, orphan_dir);
 798                 if (status < 0) {
 799                         mlog_errno(status);
 800                         goto leave;
 801                 }
 802         }
 803
 804         /* delete the name from the parent dir */
 805         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
 806         if (status < 0) {
 807                 mlog_errno(status);
 808                 goto leave;
 809         }
 810
 811         if (S_ISDIR(inode->i_mode))
 812                 drop_nlink(inode);
 813         drop_nlink(inode);
 814         fe->i_links_count = cpu_to_le16(inode->i_nlink);
 815
 816         status = ocfs2_journal_dirty(handle, fe_bh);
 817         if (status < 0) {
 818                 mlog_errno(status);
 819                 goto leave;
 820         }
 821
 822         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 823         if (S_ISDIR(inode->i_mode))
 824                 drop_nlink(dir);
 825
 826         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 827         if (status < 0) {
 828                 mlog_errno(status);
 829                 if (S_ISDIR(inode->i_mode))
 830                         inc_nlink(dir);
 831         }
 832
 833 leave:
 834         if (handle)
 835                 ocfs2_commit_trans(osb, handle);
 836
 837         if (child_locked)
 838                 ocfs2_inode_unlock(inode, 1);
 839
 840         ocfs2_inode_unlock(dir, 1);
 841
 842         if (orphan_dir) {
 843                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 844                 ocfs2_inode_unlock(orphan_dir, 1);
 845                 mutex_unlock(&orphan_dir->i_mutex);
 846                 iput(orphan_dir);
 847         }
 848
 849         brelse(fe_bh);
 850         brelse(dirent_bh);
 851         brelse(parent_node_bh);
 852         brelse(orphan_entry_bh);
 853
 854         mlog_exit(status);
 855
 856         return status;
 857 }
 858
 859 /*
 860  * The only place this should be used is rename!
 861  * if they have the same id, then the 1st one is the only one locked.
 862  */
 863 static int ocfs2_double_lock(struct ocfs2_super *osb,
 864                              struct buffer_head **bh1,
 865                              struct inode *inode1,
 866                              struct buffer_head **bh2,
 867                              struct inode *inode2)
 868 {
 869         int status;
 870         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 871         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 872         struct buffer_head **tmpbh;
 873         struct inode *tmpinode;
 874
 875         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 876                    (unsigned long long)oi1->ip_blkno,
 877                    (unsigned long long)oi2->ip_blkno);
 878
 879         if (*bh1)
 880                 *bh1 = NULL;
 881         if (*bh2)
 882                 *bh2 = NULL;
 883
 884         /* we always want to lock the one with the lower lockid first. */
 885         if (oi1->ip_blkno != oi2->ip_blkno) {
 886                 if (oi1->ip_blkno < oi2->ip_blkno) {
 887                         /* switch id1 and id2 around */
 888                         mlog(0, "switching them around...\n");
 889                         tmpbh = bh2;
 890                         bh2 = bh1;
 891                         bh1 = tmpbh;
 892
 893                         tmpinode = inode2;
 894                         inode2 = inode1;
 895                         inode1 = tmpinode;
 896                 }
 897                 /* lock id2 */
 898                 status = ocfs2_inode_lock(inode2, bh2, 1);
 899                 if (status < 0) {
 900                         if (status != -ENOENT)
 901                                 mlog_errno(status);
 902                         goto bail;
 903                 }
 904         }
 905
 906         /* lock id1 */
 907         status = ocfs2_inode_lock(inode1, bh1, 1);
 908         if (status < 0) {
 909                 /*
 910                  * An error return must mean that no cluster locks
 911                  * were held on function exit.
 912                  */
 913                 if (oi1->ip_blkno != oi2->ip_blkno)
 914                         ocfs2_inode_unlock(inode2, 1);
 915
 916                 if (status != -ENOENT)
 917                         mlog_errno(status);
 918         }
 919
 920 bail:
 921         mlog_exit(status);
 922         return status;
 923 }
 924
 925 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
 926 {
 927         ocfs2_inode_unlock(inode1, 1);
 928
 929         if (inode1 != inode2)
 930                 ocfs2_inode_unlock(inode2, 1);
 931 }
 932
 933 static int ocfs2_rename(struct inode *old_dir,
 934                         struct dentry *old_dentry,
 935                         struct inode *new_dir,
 936                         struct dentry *new_dentry)
 937 {
 938         int status = 0, rename_lock = 0, parents_locked = 0;
 939         int old_child_locked = 0, new_child_locked = 0;
 940         struct inode *old_inode = old_dentry->d_inode;
 941         struct inode *new_inode = new_dentry->d_inode;
 942         struct inode *orphan_dir = NULL;
 943         struct ocfs2_dinode *newfe = NULL;
 944         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 945         struct buffer_head *orphan_entry_bh = NULL;
 946         struct buffer_head *newfe_bh = NULL;
 947         struct buffer_head *old_inode_bh = NULL;
 948         struct buffer_head *insert_entry_bh = NULL;
 949         struct ocfs2_super *osb = NULL;
 950         u64 newfe_blkno, old_de_ino;
 951         handle_t *handle = NULL;
 952         struct buffer_head *old_dir_bh = NULL;
 953         struct buffer_head *new_dir_bh = NULL;
 954         struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
 955                 *new_de = NULL;
 956         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
 957         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
 958                                                     // this is the 1st dirent bh
 959         nlink_t old_dir_nlink = old_dir->i_nlink;
 960         struct ocfs2_dinode *old_di;
 961
 962         /* At some point it might be nice to break this function up a
 963          * bit. */
 964
 965         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
 966                    old_dir, old_dentry, new_dir, new_dentry,
 967                    old_dentry->d_name.len, old_dentry->d_name.name,
 968                    new_dentry->d_name.len, new_dentry->d_name.name);
 969
 970         osb = OCFS2_SB(old_dir->i_sb);
 971
 972         if (new_inode) {
 973                 if (!igrab(new_inode))
 974                         BUG();
 975         }
 976
 977         /* Assume a directory hierarchy thusly:
 978          * a/b/c
 979          * a/d
 980          * a,b,c, and d are all directories.
 981          *
 982          * from cwd of 'a' on both nodes:
 983          * node1: mv b/c d
 984          * node2: mv d   b/c
 985          *
 986          * And that's why, just like the VFS, we need a file system
 987          * rename lock. */
 988         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
 989                 status = ocfs2_rename_lock(osb);
 990                 if (status < 0) {
 991                         mlog_errno(status);
 992                         goto bail;
 993                 }
 994                 rename_lock = 1;
 995         }
 996
 997         /* if old and new are the same, this'll just do one lock. */
 998         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 999                                    &new_dir_bh, new_dir);
1000         if (status < 0) {
1001                 mlog_errno(status);
1002                 goto bail;
1003         }
1004         parents_locked = 1;
1005
1006         /* make sure both dirs have bhs
1007          * get an extra ref on old_dir_bh if old==new */
1008         if (!new_dir_bh) {
1009                 if (old_dir_bh) {
1010                         new_dir_bh = old_dir_bh;
1011                         get_bh(new_dir_bh);
1012                 } else {
1013                         mlog(ML_ERROR, "no old_dir_bh!\n");
1014                         status = -EIO;
1015                         goto bail;
1016                 }
1017         }
1018
1019         /*
1020          * Aside from allowing a meta data update, the locking here
1021          * also ensures that the downconvert thread on other nodes
1022          * won't have to concurrently downconvert the inode and the
1023          * dentry locks.
1024          */
1025         status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1026         if (status < 0) {
1027                 if (status != -ENOENT)
1028                         mlog_errno(status);
1029                 goto bail;
1030         }
1031         old_child_locked = 1;
1032
1033         status = ocfs2_remote_dentry_delete(old_dentry);
1034         if (status < 0) {
1035                 mlog_errno(status);
1036                 goto bail;
1037         }
1038
1039         if (S_ISDIR(old_inode->i_mode)) {
1040                 u64 old_inode_parent;
1041
1042                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1043                                                   old_inode, &old_inode_de_bh,
1044                                                   &old_inode_dot_dot_de);
1045                 if (status) {
1046                         status = -EIO;
1047                         goto bail;
1048                 }
1049
1050                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1051                         status = -EIO;
1052                         goto bail;
1053                 }
1054
1055                 if (!new_inode && new_dir != old_dir &&
1056                     new_dir->i_nlink >= OCFS2_LINK_MAX) {
1057                         status = -EMLINK;
1058                         goto bail;
1059                 }
1060         }
1061
1062         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1063                                             old_dentry->d_name.len,
1064                                             &old_de_ino);
1065         if (status) {
1066                 status = -ENOENT;
1067                 goto bail;
1068         }
1069
1070         /*
1071          *  Check for inode number is _not_ due to possible IO errors.
1072          *  We might rmdir the source, keep it as pwd of some process
1073          *  and merrily kill the link to whatever was created under the
1074          *  same name. Goodbye sticky bit ;-<
1075          */
1076         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1077                 status = -ENOENT;
1078                 goto bail;
1079         }
1080
1081         /* check if the target already exists (in which case we need
1082          * to delete it */
1083         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1084                                           new_dentry->d_name.len,
1085                                           &newfe_blkno, new_dir, &new_de_bh,
1086                                           &new_de);
1087         /* The only error we allow here is -ENOENT because the new
1088          * file not existing is perfectly valid. */
1089         if ((status < 0) && (status != -ENOENT)) {
1090                 /* If we cannot find the file specified we should just */
1091                 /* return the error... */
1092                 mlog_errno(status);
1093                 goto bail;
1094         }
1095
1096         if (!new_de && new_inode) {
1097                 /*
1098                  * Target was unlinked by another node while we were
1099                  * waiting to get to ocfs2_rename(). There isn't
1100                  * anything we can do here to help the situation, so
1101                  * bubble up the appropriate error.
1102                  */
1103                 status = -ENOENT;
1104                 goto bail;
1105         }
1106
1107         /* In case we need to overwrite an existing file, we blow it
1108          * away first */
1109         if (new_de) {
1110                 /* VFS didn't think there existed an inode here, but
1111                  * someone else in the cluster must have raced our
1112                  * rename to create one. Today we error cleanly, in
1113                  * the future we should consider calling iget to build
1114                  * a new struct inode for this entry. */
1115                 if (!new_inode) {
1116                         status = -EACCES;
1117
1118                         mlog(0, "We found an inode for name %.*s but VFS "
1119                              "didn't give us one.\n", new_dentry->d_name.len,
1120                              new_dentry->d_name.name);
1121                         goto bail;
1122                 }
1123
1124                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1125                         status = -EACCES;
1126
1127                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1128                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1129                              (unsigned long long)newfe_blkno,
1130                              OCFS2_I(new_inode)->ip_flags);
1131                         goto bail;
1132                 }
1133
1134                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1135                 if (status < 0) {
1136                         if (status != -ENOENT)
1137                                 mlog_errno(status);
1138                         goto bail;
1139                 }
1140                 new_child_locked = 1;
1141
1142                 status = ocfs2_remote_dentry_delete(new_dentry);
1143                 if (status < 0) {
1144                         mlog_errno(status);
1145                         goto bail;
1146                 }
1147
1148                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1149
1150                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1151                      "newfebh=%p bhblocknr=%llu\n", new_de,
1152                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1153                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1154
1155                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1156                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1157                                                           new_inode,
1158                                                           orphan_name,
1159                                                           &orphan_entry_bh);
1160                         if (status < 0) {
1161                                 mlog_errno(status);
1162                                 goto bail;
1163                         }
1164                 }
1165         } else {
1166                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1167
1168                 status = ocfs2_check_dir_for_entry(new_dir,
1169                                                    new_dentry->d_name.name,
1170                                                    new_dentry->d_name.len);
1171                 if (status)
1172                         goto bail;
1173
1174                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1175                                                       new_dentry->d_name.name,
1176                                                       new_dentry->d_name.len,
1177                                                       &insert_entry_bh);
1178                 if (status < 0) {
1179                         mlog_errno(status);
1180                         goto bail;
1181                 }
1182         }
1183
1184         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1185         if (IS_ERR(handle)) {
1186                 status = PTR_ERR(handle);
1187                 handle = NULL;
1188                 mlog_errno(status);
1189                 goto bail;
1190         }
1191
1192         if (new_de) {
1193                 if (S_ISDIR(new_inode->i_mode)) {
1194                         if (!ocfs2_empty_dir(new_inode) ||
1195                             new_inode->i_nlink != 2) {
1196                                 status = -ENOTEMPTY;
1197                                 goto bail;
1198                         }
1199                 }
1200                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1201                                               OCFS2_JOURNAL_ACCESS_WRITE);
1202                 if (status < 0) {
1203                         mlog_errno(status);
1204                         goto bail;
1205                 }
1206
1207                 if (S_ISDIR(new_inode->i_mode) ||
1208                     (newfe->i_links_count == cpu_to_le16(1))){
1209                         status = ocfs2_orphan_add(osb, handle, new_inode,
1210                                                   newfe, orphan_name,
1211                                                   orphan_entry_bh, orphan_dir);
1212                         if (status < 0) {
1213                                 mlog_errno(status);
1214                                 goto bail;
1215                         }
1216                 }
1217
1218                 /* change the dirent to point to the correct inode */
1219                 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1220                                             new_de, old_inode);
1221                 if (status < 0) {
1222                         mlog_errno(status);
1223                         goto bail;
1224                 }
1225                 new_dir->i_version++;
1226
1227                 if (S_ISDIR(new_inode->i_mode))
1228                         newfe->i_links_count = 0;
1229                 else
1230                         le16_add_cpu(&newfe->i_links_count, -1);
1231
1232                 status = ocfs2_journal_dirty(handle, newfe_bh);
1233                 if (status < 0) {
1234                         mlog_errno(status);
1235                         goto bail;
1236                 }
1237         } else {
1238                 /* if the name was not found in new_dir, add it now */
1239                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1240                                          OCFS2_I(old_inode)->ip_blkno,
1241                                          new_dir_bh, insert_entry_bh);
1242         }
1243
1244         old_inode->i_ctime = CURRENT_TIME;
1245         mark_inode_dirty(old_inode);
1246
1247         status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1248                                       OCFS2_JOURNAL_ACCESS_WRITE);
1249         if (status >= 0) {
1250                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1251
1252                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1253                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1254
1255                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1256                 if (status < 0)
1257                         mlog_errno(status);
1258         } else
1259                 mlog_errno(status);
1260
1261         /*
1262          * Now that the name has been added to new_dir, remove the old name.
1263          *
1264          * We don't keep any directory entry context around until now
1265          * because the insert might have changed the type of directory
1266          * we're dealing with.
1267          */
1268         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1269                                      old_dentry->d_name.len,
1270                                      old_dir, &old_de);
1271         if (!old_de_bh) {
1272                 status = -EIO;
1273                 goto bail;
1274         }
1275
1276         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1277         if (status < 0) {
1278                 mlog_errno(status);
1279                 goto bail;
1280         }
1281
1282         if (new_inode) {
1283                 new_inode->i_nlink--;
1284                 new_inode->i_ctime = CURRENT_TIME;
1285         }
1286         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1287         if (old_inode_de_bh) {
1288                 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1289                                             old_inode_dot_dot_de, new_dir);
1290                 old_dir->i_nlink--;
1291                 if (new_inode) {
1292                         new_inode->i_nlink--;
1293                 } else {
1294                         inc_nlink(new_dir);
1295                         mark_inode_dirty(new_dir);
1296                 }
1297         }
1298         mark_inode_dirty(old_dir);
1299         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1300         if (new_inode) {
1301                 mark_inode_dirty(new_inode);
1302                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1303         }
1304
1305         if (old_dir != new_dir) {
1306                 /* Keep the same times on both directories.*/
1307                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1308
1309                 /*
1310                  * This will also pick up the i_nlink change from the
1311                  * block above.
1312                  */
1313                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1314         }
1315
1316         if (old_dir_nlink != old_dir->i_nlink) {
1317                 if (!old_dir_bh) {
1318                         mlog(ML_ERROR, "need to change nlink for old dir "
1319                              "%llu from %d to %d but bh is NULL!\n",
1320                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1321                              (int)old_dir_nlink, old_dir->i_nlink);
1322                 } else {
1323                         struct ocfs2_dinode *fe;
1324                         status = ocfs2_journal_access(handle, old_dir,
1325                                                       old_dir_bh,
1326                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1327                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1328                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1329                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1330                 }
1331         }
1332
1333         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1334         status = 0;
1335 bail:
1336         if (rename_lock)
1337                 ocfs2_rename_unlock(osb);
1338
1339         if (handle)
1340                 ocfs2_commit_trans(osb, handle);
1341
1342         if (parents_locked)
1343                 ocfs2_double_unlock(old_dir, new_dir);
1344
1345         if (old_child_locked)
1346                 ocfs2_inode_unlock(old_inode, 1);
1347
1348         if (new_child_locked)
1349                 ocfs2_inode_unlock(new_inode, 1);
1350
1351         if (orphan_dir) {
1352                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1353                 ocfs2_inode_unlock(orphan_dir, 1);
1354                 mutex_unlock(&orphan_dir->i_mutex);
1355                 iput(orphan_dir);
1356         }
1357
1358         if (new_inode)
1359                 sync_mapping_buffers(old_inode->i_mapping);
1360
1361         if (new_inode)
1362                 iput(new_inode);
1363         brelse(newfe_bh);
1364         brelse(old_inode_bh);
1365         brelse(old_dir_bh);
1366         brelse(new_dir_bh);
1367         brelse(new_de_bh);
1368         brelse(old_de_bh);
1369         brelse(old_inode_de_bh);
1370         brelse(orphan_entry_bh);
1371         brelse(insert_entry_bh);
1372
1373         mlog_exit(status);
1374
1375         return status;
1376 }
1377
1378 /*
1379  * we expect i_size = strlen(symname). Copy symname into the file
1380  * data, including the null terminator.
1381  */
1382 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1383                                      handle_t *handle,
1384                                      struct inode *inode,
1385                                      const char *symname)
1386 {
1387         struct buffer_head **bhs = NULL;
1388         const char *c;
1389         struct super_block *sb = osb->sb;
1390         u64 p_blkno, p_blocks;
1391         int virtual, blocks, status, i, bytes_left;
1392
1393         bytes_left = i_size_read(inode) + 1;
1394         /* we can't trust i_blocks because we're actually going to
1395          * write i_size + 1 bytes. */
1396         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1397
1398         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1399                         (unsigned long long)inode->i_blocks,
1400                         i_size_read(inode), blocks);
1401
1402         /* Sanity check -- make sure we're going to fit. */
1403         if (bytes_left >
1404             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1405                 status = -EIO;
1406                 mlog_errno(status);
1407                 goto bail;
1408         }
1409
1410         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1411         if (!bhs) {
1412                 status = -ENOMEM;
1413                 mlog_errno(status);
1414                 goto bail;
1415         }
1416
1417         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1418                                              NULL);
1419         if (status < 0) {
1420                 mlog_errno(status);
1421                 goto bail;
1422         }
1423
1424         /* links can never be larger than one cluster so we know this
1425          * is all going to be contiguous, but do a sanity check
1426          * anyway. */
1427         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1428                 status = -EIO;
1429                 mlog_errno(status);
1430                 goto bail;
1431         }
1432
1433         virtual = 0;
1434         while(bytes_left > 0) {
1435                 c = &symname[virtual * sb->s_blocksize];
1436
1437                 bhs[virtual] = sb_getblk(sb, p_blkno);
1438                 if (!bhs[virtual]) {
1439                         status = -ENOMEM;
1440                         mlog_errno(status);
1441                         goto bail;
1442                 }
1443                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1444
1445                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1446                                               OCFS2_JOURNAL_ACCESS_CREATE);
1447                 if (status < 0) {
1448                         mlog_errno(status);
1449                         goto bail;
1450                 }
1451
1452                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1453
1454                 memcpy(bhs[virtual]->b_data, c,
1455                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1456                        bytes_left);
1457
1458                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1459                 if (status < 0) {
1460                         mlog_errno(status);
1461                         goto bail;
1462                 }
1463
1464                 virtual++;
1465                 p_blkno++;
1466                 bytes_left -= sb->s_blocksize;
1467         }
1468
1469         status = 0;
1470 bail:
1471
1472         if (bhs) {
1473                 for(i = 0; i < blocks; i++)
1474                         brelse(bhs[i]);
1475                 kfree(bhs);
1476         }
1477
1478         mlog_exit(status);
1479         return status;
1480 }
1481
1482 static int ocfs2_symlink(struct inode *dir,
1483                          struct dentry *dentry,
1484                          const char *symname)
1485 {
1486         int status, l, credits;
1487         u64 newsize;
1488         struct ocfs2_super *osb = NULL;
1489         struct inode *inode = NULL;
1490         struct super_block *sb;
1491         struct buffer_head *new_fe_bh = NULL;
1492         struct buffer_head *de_bh = NULL;
1493         struct buffer_head *parent_fe_bh = NULL;
1494         struct ocfs2_dinode *fe = NULL;
1495         struct ocfs2_dinode *dirfe;
1496         handle_t *handle = NULL;
1497         struct ocfs2_alloc_context *inode_ac = NULL;
1498         struct ocfs2_alloc_context *data_ac = NULL;
1499
1500         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1501                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1502
1503         sb = dir->i_sb;
1504         osb = OCFS2_SB(sb);
1505
1506         l = strlen(symname) + 1;
1507
1508         credits = ocfs2_calc_symlink_credits(sb);
1509
1510         /* lock the parent directory */
1511         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1512         if (status < 0) {
1513                 if (status != -ENOENT)
1514                         mlog_errno(status);
1515                 return status;
1516         }
1517
1518         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1519         if (!dirfe->i_links_count) {
1520                 /* can't make a file in a deleted directory. */
1521                 status = -ENOENT;
1522                 goto bail;
1523         }
1524
1525         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1526                                            dentry->d_name.len);
1527         if (status)
1528                 goto bail;
1529
1530         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1531                                               dentry->d_name.name,
1532                                               dentry->d_name.len, &de_bh);
1533         if (status < 0) {
1534                 mlog_errno(status);
1535                 goto bail;
1536         }
1537
1538         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1539         if (status < 0) {
1540                 if (status != -ENOSPC)
1541                         mlog_errno(status);
1542                 goto bail;
1543         }
1544
1545         /* don't reserve bitmap space for fast symlinks. */
1546         if (l > ocfs2_fast_symlink_chars(sb)) {
1547                 status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1548                 if (status < 0) {
1549                         if (status != -ENOSPC)
1550                                 mlog_errno(status);
1551                         goto bail;
1552                 }
1553         }
1554
1555         handle = ocfs2_start_trans(osb, credits);
1556         if (IS_ERR(handle)) {
1557                 status = PTR_ERR(handle);
1558                 handle = NULL;
1559                 mlog_errno(status);
1560                 goto bail;
1561         }
1562
1563         status = ocfs2_mknod_locked(osb, dir, dentry,
1564                                     S_IFLNK | S_IRWXUGO, 0,
1565                                     &new_fe_bh, parent_fe_bh, handle,
1566                                     &inode, inode_ac);
1567         if (status < 0) {
1568                 mlog_errno(status);
1569                 goto bail;
1570         }
1571
1572         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1573         inode->i_rdev = 0;
1574         newsize = l - 1;
1575         if (l > ocfs2_fast_symlink_chars(sb)) {
1576                 u32 offset = 0;
1577
1578                 inode->i_op = &ocfs2_symlink_inode_operations;
1579                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1580                                               new_fe_bh,
1581                                               handle, data_ac, NULL,
1582                                               NULL);
1583                 if (status < 0) {
1584                         if (status != -ENOSPC && status != -EINTR) {
1585                                 mlog(ML_ERROR,
1586                                      "Failed to extend file to %llu\n",
1587                                      (unsigned long long)newsize);
1588                                 mlog_errno(status);
1589                                 status = -ENOSPC;
1590                         }
1591                         goto bail;
1592                 }
1593                 i_size_write(inode, newsize);
1594                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1595         } else {
1596                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1597                 memcpy((char *) fe->id2.i_symlink, symname, l);
1598                 i_size_write(inode, newsize);
1599                 inode->i_blocks = 0;
1600         }
1601
1602         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1603         if (status < 0) {
1604                 mlog_errno(status);
1605                 goto bail;
1606         }
1607
1608         if (!ocfs2_inode_is_fast_symlink(inode)) {
1609                 status = ocfs2_create_symlink_data(osb, handle, inode,
1610                                                    symname);
1611                 if (status < 0) {
1612                         mlog_errno(status);
1613                         goto bail;
1614                 }
1615         }
1616
1617         status = ocfs2_add_entry(handle, dentry, inode,
1618                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1619                                  de_bh);
1620         if (status < 0) {
1621                 mlog_errno(status);
1622                 goto bail;
1623         }
1624
1625         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1626         if (status) {
1627                 mlog_errno(status);
1628                 goto bail;
1629         }
1630
1631         insert_inode_hash(inode);
1632         dentry->d_op = &ocfs2_dentry_ops;
1633         d_instantiate(dentry, inode);
1634 bail:
1635         if (handle)
1636                 ocfs2_commit_trans(osb, handle);
1637
1638         ocfs2_inode_unlock(dir, 1);
1639
1640         brelse(new_fe_bh);
1641         brelse(parent_fe_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_inode_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_inode_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         brelse(orphan_dir_bh);
1738
1739         mlog_exit(status);
1740         return status;
1741 }
1742
1743 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1744                             handle_t *handle,
1745                             struct inode *inode,
1746                             struct ocfs2_dinode *fe,
1747                             char *name,
1748                             struct buffer_head *de_bh,
1749                             struct inode *orphan_dir_inode)
1750 {
1751         struct buffer_head *orphan_dir_bh = NULL;
1752         int status = 0;
1753         struct ocfs2_dinode *orphan_fe;
1754
1755         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1756
1757         status = ocfs2_read_block(orphan_dir_inode,
1758                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
1759                                   &orphan_dir_bh);
1760         if (status < 0) {
1761                 mlog_errno(status);
1762                 goto leave;
1763         }
1764
1765         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1766                                       OCFS2_JOURNAL_ACCESS_WRITE);
1767         if (status < 0) {
1768                 mlog_errno(status);
1769                 goto leave;
1770         }
1771
1772         /* we're a cluster, and nlink can change on disk from
1773          * underneath us... */
1774         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1775         if (S_ISDIR(inode->i_mode))
1776                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1777         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1778
1779         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1780         if (status < 0) {
1781                 mlog_errno(status);
1782                 goto leave;
1783         }
1784
1785         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1786                                    OCFS2_ORPHAN_NAMELEN, inode,
1787                                    OCFS2_I(inode)->ip_blkno,
1788                                    orphan_dir_bh, de_bh);
1789         if (status < 0) {
1790                 mlog_errno(status);
1791                 goto leave;
1792         }
1793
1794         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1795
1796         /* Record which orphan dir our inode now resides
1797          * in. delete_inode will use this to determine which orphan
1798          * dir to lock. */
1799         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1800
1801         mlog(0, "Inode %llu orphaned in slot %d\n",
1802              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1803
1804 leave:
1805         brelse(orphan_dir_bh);
1806
1807         mlog_exit(status);
1808         return status;
1809 }
1810
1811 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1812 int ocfs2_orphan_del(struct ocfs2_super *osb,
1813                      handle_t *handle,
1814                      struct inode *orphan_dir_inode,
1815                      struct inode *inode,
1816                      struct buffer_head *orphan_dir_bh)
1817 {
1818         char name[OCFS2_ORPHAN_NAMELEN + 1];
1819         struct ocfs2_dinode *orphan_fe;
1820         int status = 0;
1821         struct buffer_head *target_de_bh = NULL;
1822         struct ocfs2_dir_entry *target_de = NULL;
1823
1824         mlog_entry_void();
1825
1826         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1827         if (status < 0) {
1828                 mlog_errno(status);
1829                 goto leave;
1830         }
1831
1832         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1833              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1834              OCFS2_ORPHAN_NAMELEN);
1835
1836         /* find it's spot in the orphan directory */
1837         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1838                                         orphan_dir_inode, &target_de);
1839         if (!target_de_bh) {
1840                 status = -ENOENT;
1841                 mlog_errno(status);
1842                 goto leave;
1843         }
1844
1845         /* remove it from the orphan directory */
1846         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1847                                     target_de_bh);
1848         if (status < 0) {
1849                 mlog_errno(status);
1850                 goto leave;
1851         }
1852
1853         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
1854                                       OCFS2_JOURNAL_ACCESS_WRITE);
1855         if (status < 0) {
1856                 mlog_errno(status);
1857                 goto leave;
1858         }
1859
1860         /* do the i_nlink dance! :) */
1861         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1862         if (S_ISDIR(inode->i_mode))
1863                 le16_add_cpu(&orphan_fe->i_links_count, -1);
1864         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1865
1866         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1867         if (status < 0) {
1868                 mlog_errno(status);
1869                 goto leave;
1870         }
1871
1872 leave:
1873         brelse(target_de_bh);
1874
1875         mlog_exit(status);
1876         return status;
1877 }
1878
1879 const struct inode_operations ocfs2_dir_iops = {
1880         .create         = ocfs2_create,
1881         .lookup         = ocfs2_lookup,
1882         .link           = ocfs2_link,
1883         .unlink         = ocfs2_unlink,
1884         .rmdir          = ocfs2_unlink,
1885         .symlink        = ocfs2_symlink,
1886         .mkdir          = ocfs2_mkdir,
1887         .mknod          = ocfs2_mknod,
1888         .rename         = ocfs2_rename,
1889         .setattr        = ocfs2_setattr,
1890         .getattr        = ocfs2_getattr,
1891         .permission     = ocfs2_permission,
1892         .setxattr       = generic_setxattr,
1893         .getxattr       = generic_getxattr,
1894         .listxattr      = ocfs2_listxattr,
1895         .removexattr    = generic_removexattr,
1896 };