fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/quotaops.h>
  44
  45 #define MLOG_MASK_PREFIX ML_NAMEI
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 #include "alloc.h"
  51 #include "dcache.h"
  52 #include "dir.h"
  53 #include "dlmglue.h"
  54 #include "extent_map.h"
  55 #include "file.h"
  56 #include "inode.h"
  57 #include "journal.h"
  58 #include "namei.h"
  59 #include "suballoc.h"
  60 #include "super.h"
  61 #include "symlink.h"
  62 #include "sysfile.h"
  63 #include "uptodate.h"
  64 #include "xattr.h"
  65 #include "acl.h"
  66
  67 #include "buffer_head_io.h"
  68
  69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  70                               struct inode *dir,
  71                               struct inode *inode,
  72                               dev_t dev,
  73                               struct buffer_head **new_fe_bh,
  74                               struct buffer_head *parent_fe_bh,
  75                               handle_t *handle,
  76                               struct ocfs2_alloc_context *inode_ac);
  77
  78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  79                                     struct inode **ret_orphan_dir,
  80                                     u64 blkno,
  81                                     char *name,
  82                                     struct ocfs2_dir_lookup_result *lookup);
  83
  84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  85                             handle_t *handle,
  86                             struct inode *inode,
  87                             struct buffer_head *fe_bh,
  88                             char *name,
  89                             struct ocfs2_dir_lookup_result *lookup,
  90                             struct inode *orphan_dir_inode);
  91
  92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  93                                      handle_t *handle,
  94                                      struct inode *inode,
  95                                      const char *symname);
  96
  97 /* An orphan dir name is an 8 byte value, printed as a hex string */
  98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  99
 100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 101                                    struct nameidata *nd)
 102 {
 103         int status;
 104         u64 blkno;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_inode_info *oi;
 108
 109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 110                    dentry->d_name.len, dentry->d_name.name);
 111
 112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 113                 ret = ERR_PTR(-ENAMETOOLONG);
 114                 goto bail;
 115         }
 116
 117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
 118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
 119
 120         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 121         if (status < 0) {
 122                 if (status != -ENOENT)
 123                         mlog_errno(status);
 124                 ret = ERR_PTR(status);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 129                                             dentry->d_name.len, &blkno);
 130         if (status < 0)
 131                 goto bail_add;
 132
 133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 134         if (IS_ERR(inode)) {
 135                 ret = ERR_PTR(-EACCES);
 136                 goto bail_unlock;
 137         }
 138
 139         oi = OCFS2_I(inode);
 140         /* Clear any orphaned state... If we were able to look up the
 141          * inode from a directory, it certainly can't be orphaned. We
 142          * might have the bad state from a node which intended to
 143          * orphan this inode but crashed before it could commit the
 144          * unlink. */
 145         spin_lock(&oi->ip_lock);
 146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 147         spin_unlock(&oi->ip_lock);
 148
 149 bail_add:
 150         dentry->d_op = &ocfs2_dentry_ops;
 151         ret = d_splice_alias(inode, dentry);
 152
 153         if (inode) {
 154                 /*
 155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 156                  * dentry, it will d_move() it on top of ourse. The
 157                  * return value will indicate this however, so in
 158                  * those cases, we switch them around for the locking
 159                  * code.
 160                  *
 161                  * NOTE: This dentry already has ->d_op set from
 162                  * ocfs2_get_parent() and ocfs2_get_dentry()
 163                  */
 164                 if (ret)
 165                         dentry = ret;
 166
 167                 status = ocfs2_dentry_attach_lock(dentry, inode,
 168                                                   OCFS2_I(dir)->ip_blkno);
 169                 if (status) {
 170                         mlog_errno(status);
 171                         ret = ERR_PTR(status);
 172                         goto bail_unlock;
 173                 }
 174         }
 175
 176 bail_unlock:
 177         /* Don't drop the cluster lock until *after* the d_add --
 178          * unlink on another node will message us to remove that
 179          * dentry under this lock so otherwise we can race this with
 180          * the downconvert thread and have a stale dentry. */
 181         ocfs2_inode_unlock(dir, 0);
 182
 183 bail:
 184
 185         mlog_exit_ptr(ret);
 186
 187         return ret;
 188 }
 189
 190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
 191 {
 192         struct inode *inode;
 193
 194         inode = new_inode(dir->i_sb);
 195         if (!inode) {
 196                 mlog(ML_ERROR, "new_inode failed!\n");
 197                 return NULL;
 198         }
 199
 200         /* populate as many fields early on as possible - many of
 201          * these are used by the support functions here and in
 202          * callers. */
 203         if (S_ISDIR(mode))
 204                 inode->i_nlink = 2;
 205         else
 206                 inode->i_nlink = 1;
 207         inode->i_uid = current_fsuid();
 208         if (dir->i_mode & S_ISGID) {
 209                 inode->i_gid = dir->i_gid;
 210                 if (S_ISDIR(mode))
 211                         mode |= S_ISGID;
 212         } else
 213                 inode->i_gid = current_fsgid();
 214         inode->i_mode = mode;
 215         dquot_initialize(inode);
 216         return inode;
 217 }
 218
 219 static int ocfs2_mknod(struct inode *dir,
 220                        struct dentry *dentry,
 221                        int mode,
 222                        dev_t dev)
 223 {
 224         int status = 0;
 225         struct buffer_head *parent_fe_bh = NULL;
 226         handle_t *handle = NULL;
 227         struct ocfs2_super *osb;
 228         struct ocfs2_dinode *dirfe;
 229         struct buffer_head *new_fe_bh = NULL;
 230         struct inode *inode = NULL;
 231         struct ocfs2_alloc_context *inode_ac = NULL;
 232         struct ocfs2_alloc_context *data_ac = NULL;
 233         struct ocfs2_alloc_context *meta_ac = NULL;
 234         int want_clusters = 0;
 235         int want_meta = 0;
 236         int xattr_credits = 0;
 237         struct ocfs2_security_xattr_info si = {
 238                 .enable = 1,
 239         };
 240         int did_quota_inode = 0;
 241         struct ocfs2_dir_lookup_result lookup = { NULL, };
 242
 243         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 244                    (unsigned long)dev, dentry->d_name.len,
 245                    dentry->d_name.name);
 246
 247         dquot_initialize(dir);
 248
 249         /* get our super block */
 250         osb = OCFS2_SB(dir->i_sb);
 251
 252         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 253         if (status < 0) {
 254                 if (status != -ENOENT)
 255                         mlog_errno(status);
 256                 return status;
 257         }
 258
 259         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 260                 status = -EMLINK;
 261                 goto leave;
 262         }
 263
 264         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 265         if (!ocfs2_read_links_count(dirfe)) {
 266                 /* can't make a file in a deleted directory. */
 267                 status = -ENOENT;
 268                 goto leave;
 269         }
 270
 271         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 272                                            dentry->d_name.len);
 273         if (status)
 274                 goto leave;
 275
 276         /* get a spot inside the dir. */
 277         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 278                                               dentry->d_name.name,
 279                                               dentry->d_name.len, &lookup);
 280         if (status < 0) {
 281                 mlog_errno(status);
 282                 goto leave;
 283         }
 284
 285         /* reserve an inode spot */
 286         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 287         if (status < 0) {
 288                 if (status != -ENOSPC)
 289                         mlog_errno(status);
 290                 goto leave;
 291         }
 292
 293         inode = ocfs2_get_init_inode(dir, mode);
 294         if (!inode) {
 295                 status = -ENOMEM;
 296                 mlog_errno(status);
 297                 goto leave;
 298         }
 299
 300         /* get security xattr */
 301         status = ocfs2_init_security_get(inode, dir, &si);
 302         if (status) {
 303                 if (status == -EOPNOTSUPP)
 304                         si.enable = 0;
 305                 else {
 306                         mlog_errno(status);
 307                         goto leave;
 308                 }
 309         }
 310
 311         /* calculate meta data/clusters for setting security and acl xattr */
 312         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 313                                        &si, &want_clusters,
 314                                        &xattr_credits, &want_meta);
 315         if (status < 0) {
 316                 mlog_errno(status);
 317                 goto leave;
 318         }
 319
 320         /* Reserve a cluster if creating an extent based directory. */
 321         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 322                 want_clusters += 1;
 323
 324                 /* Dir indexing requires extra space as well */
 325                 if (ocfs2_supports_indexed_dirs(osb))
 326                         want_meta++;
 327         }
 328
 329         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 330         if (status < 0) {
 331                 if (status != -ENOSPC)
 332                         mlog_errno(status);
 333                 goto leave;
 334         }
 335
 336         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 337         if (status < 0) {
 338                 if (status != -ENOSPC)
 339                         mlog_errno(status);
 340                 goto leave;
 341         }
 342
 343         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 344                                                             S_ISDIR(mode),
 345                                                             xattr_credits));
 346         if (IS_ERR(handle)) {
 347                 status = PTR_ERR(handle);
 348                 handle = NULL;
 349                 mlog_errno(status);
 350                 goto leave;
 351         }
 352
 353         status = dquot_alloc_inode(inode);
 354         if (status)
 355                 goto leave;
 356         did_quota_inode = 1;
 357
 358         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
 359                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
 360                    dentry->d_name.name);
 361
 362         /* do the real work now. */
 363         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 364                                     &new_fe_bh, parent_fe_bh, handle,
 365                                     inode_ac);
 366         if (status < 0) {
 367                 mlog_errno(status);
 368                 goto leave;
 369         }
 370
 371         if (S_ISDIR(mode)) {
 372                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 373                                             new_fe_bh, data_ac, meta_ac);
 374                 if (status < 0) {
 375                         mlog_errno(status);
 376                         goto leave;
 377                 }
 378
 379                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 380                                                  parent_fe_bh,
 381                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 382                 if (status < 0) {
 383                         mlog_errno(status);
 384                         goto leave;
 385                 }
 386                 ocfs2_add_links_count(dirfe, 1);
 387                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
 388                 if (status < 0) {
 389                         mlog_errno(status);
 390                         goto leave;
 391                 }
 392                 inc_nlink(dir);
 393         }
 394
 395         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 396                                 meta_ac, data_ac);
 397         if (status < 0) {
 398                 mlog_errno(status);
 399                 goto leave;
 400         }
 401
 402         if (si.enable) {
 403                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 404                                                  meta_ac, data_ac);
 405                 if (status < 0) {
 406                         mlog_errno(status);
 407                         goto leave;
 408                 }
 409         }
 410
 411         status = ocfs2_add_entry(handle, dentry, inode,
 412                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 413                                  &lookup);
 414         if (status < 0) {
 415                 mlog_errno(status);
 416                 goto leave;
 417         }
 418
 419         status = ocfs2_dentry_attach_lock(dentry, inode,
 420                                           OCFS2_I(dir)->ip_blkno);
 421         if (status) {
 422                 mlog_errno(status);
 423                 goto leave;
 424         }
 425
 426         insert_inode_hash(inode);
 427         dentry->d_op = &ocfs2_dentry_ops;
 428         d_instantiate(dentry, inode);
 429         status = 0;
 430 leave:
 431         if (status < 0 && did_quota_inode)
 432                 dquot_free_inode(inode);
 433         if (handle)
 434                 ocfs2_commit_trans(osb, handle);
 435
 436         ocfs2_inode_unlock(dir, 1);
 437
 438         if (status == -ENOSPC)
 439                 mlog(0, "Disk is full\n");
 440
 441         brelse(new_fe_bh);
 442         brelse(parent_fe_bh);
 443         kfree(si.name);
 444         kfree(si.value);
 445
 446         ocfs2_free_dir_lookup_result(&lookup);
 447
 448         if (inode_ac)
 449                 ocfs2_free_alloc_context(inode_ac);
 450
 451         if (data_ac)
 452                 ocfs2_free_alloc_context(data_ac);
 453
 454         if (meta_ac)
 455                 ocfs2_free_alloc_context(meta_ac);
 456
 457         /*
 458          * We should call iput after the i_mutex of the bitmap been
 459          * unlocked in ocfs2_free_alloc_context, or the
 460          * ocfs2_delete_inode will mutex_lock again.
 461          */
 462         if ((status < 0) && inode) {
 463                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 464                 clear_nlink(inode);
 465                 iput(inode);
 466         }
 467
 468         mlog_exit(status);
 469
 470         return status;
 471 }
 472
 473 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 474                               struct inode *dir,
 475                               struct inode *inode,
 476                               dev_t dev,
 477                               struct buffer_head **new_fe_bh,
 478                               struct buffer_head *parent_fe_bh,
 479                               handle_t *handle,
 480                               struct ocfs2_alloc_context *inode_ac)
 481 {
 482         int status = 0;
 483         struct ocfs2_dinode *fe = NULL;
 484         struct ocfs2_extent_list *fel;
 485         u64 fe_blkno = 0;
 486         u16 suballoc_bit;
 487         u16 feat;
 488
 489         *new_fe_bh = NULL;
 490
 491         status = ocfs2_claim_new_inode(osb, handle, dir, parent_fe_bh,
 492                                        inode_ac, &suballoc_bit, &fe_blkno);
 493         if (status < 0) {
 494                 mlog_errno(status);
 495                 goto leave;
 496         }
 497
 498         /* populate as many fields early on as possible - many of
 499          * these are used by the support functions here and in
 500          * callers. */
 501         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 502         OCFS2_I(inode)->ip_blkno = fe_blkno;
 503         spin_lock(&osb->osb_lock);
 504         inode->i_generation = osb->s_next_generation++;
 505         spin_unlock(&osb->osb_lock);
 506
 507         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 508         if (!*new_fe_bh) {
 509                 status = -EIO;
 510                 mlog_errno(status);
 511                 goto leave;
 512         }
 513         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 514
 515         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 516                                          *new_fe_bh,
 517                                          OCFS2_JOURNAL_ACCESS_CREATE);
 518         if (status < 0) {
 519                 mlog_errno(status);
 520                 goto leave;
 521         }
 522
 523         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 524         memset(fe, 0, osb->sb->s_blocksize);
 525
 526         fe->i_generation = cpu_to_le32(inode->i_generation);
 527         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 528         fe->i_blkno = cpu_to_le64(fe_blkno);
 529         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 530         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 531         fe->i_uid = cpu_to_le32(inode->i_uid);
 532         fe->i_gid = cpu_to_le32(inode->i_gid);
 533         fe->i_mode = cpu_to_le16(inode->i_mode);
 534         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 535                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 536
 537         ocfs2_set_links_count(fe, inode->i_nlink);
 538
 539         fe->i_last_eb_blk = 0;
 540         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 541         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 542         fe->i_atime = fe->i_ctime = fe->i_mtime =
 543                 cpu_to_le64(CURRENT_TIME.tv_sec);
 544         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 545                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 546         fe->i_dtime = 0;
 547
 548         /*
 549          * If supported, directories start with inline data. If inline
 550          * isn't supported, but indexing is, we start them as indexed.
 551          */
 552         feat = le16_to_cpu(fe->i_dyn_features);
 553         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 554                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 555
 556                 fe->id2.i_data.id_count = cpu_to_le16(
 557                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 558         } else {
 559                 fel = &fe->id2.i_list;
 560                 fel->l_tree_depth = 0;
 561                 fel->l_next_free_rec = 0;
 562                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 563         }
 564
 565         status = ocfs2_journal_dirty(handle, *new_fe_bh);
 566         if (status < 0) {
 567                 mlog_errno(status);
 568                 goto leave;
 569         }
 570
 571         ocfs2_populate_inode(inode, fe, 1);
 572         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 573         if (!ocfs2_mount_local(osb)) {
 574                 status = ocfs2_create_new_inode_locks(inode);
 575                 if (status < 0)
 576                         mlog_errno(status);
 577         }
 578
 579         status = 0; /* error in ocfs2_create_new_inode_locks is not
 580                      * critical */
 581
 582 leave:
 583         if (status < 0) {
 584                 if (*new_fe_bh) {
 585                         brelse(*new_fe_bh);
 586                         *new_fe_bh = NULL;
 587                 }
 588         }
 589
 590         mlog_exit(status);
 591         return status;
 592 }
 593
 594 static int ocfs2_mkdir(struct inode *dir,
 595                        struct dentry *dentry,
 596                        int mode)
 597 {
 598         int ret;
 599
 600         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 601                    dentry->d_name.len, dentry->d_name.name);
 602         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 603         mlog_exit(ret);
 604
 605         return ret;
 606 }
 607
 608 static int ocfs2_create(struct inode *dir,
 609                         struct dentry *dentry,
 610                         int mode,
 611                         struct nameidata *nd)
 612 {
 613         int ret;
 614
 615         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 616                    dentry->d_name.len, dentry->d_name.name);
 617         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 618         mlog_exit(ret);
 619
 620         return ret;
 621 }
 622
 623 static int ocfs2_link(struct dentry *old_dentry,
 624                       struct inode *dir,
 625                       struct dentry *dentry)
 626 {
 627         handle_t *handle;
 628         struct inode *inode = old_dentry->d_inode;
 629         int err;
 630         struct buffer_head *fe_bh = NULL;
 631         struct buffer_head *parent_fe_bh = NULL;
 632         struct ocfs2_dinode *fe = NULL;
 633         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 634         struct ocfs2_dir_lookup_result lookup = { NULL, };
 635
 636         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 637                    old_dentry->d_name.len, old_dentry->d_name.name,
 638                    dentry->d_name.len, dentry->d_name.name);
 639
 640         if (S_ISDIR(inode->i_mode))
 641                 return -EPERM;
 642
 643         dquot_initialize(dir);
 644
 645         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 646         if (err < 0) {
 647                 if (err != -ENOENT)
 648                         mlog_errno(err);
 649                 return err;
 650         }
 651
 652         if (!dir->i_nlink) {
 653                 err = -ENOENT;
 654                 goto out;
 655         }
 656
 657         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 658                                         dentry->d_name.len);
 659         if (err)
 660                 goto out;
 661
 662         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 663                                            dentry->d_name.name,
 664                                            dentry->d_name.len, &lookup);
 665         if (err < 0) {
 666                 mlog_errno(err);
 667                 goto out;
 668         }
 669
 670         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 671         if (err < 0) {
 672                 if (err != -ENOENT)
 673                         mlog_errno(err);
 674                 goto out;
 675         }
 676
 677         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 678         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 679                 err = -EMLINK;
 680                 goto out_unlock_inode;
 681         }
 682
 683         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 684         if (IS_ERR(handle)) {
 685                 err = PTR_ERR(handle);
 686                 handle = NULL;
 687                 mlog_errno(err);
 688                 goto out_unlock_inode;
 689         }
 690
 691         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 692                                       OCFS2_JOURNAL_ACCESS_WRITE);
 693         if (err < 0) {
 694                 mlog_errno(err);
 695                 goto out_commit;
 696         }
 697
 698         inc_nlink(inode);
 699         inode->i_ctime = CURRENT_TIME;
 700         ocfs2_set_links_count(fe, inode->i_nlink);
 701         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 702         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 703
 704         err = ocfs2_journal_dirty(handle, fe_bh);
 705         if (err < 0) {
 706                 ocfs2_add_links_count(fe, -1);
 707                 drop_nlink(inode);
 708                 mlog_errno(err);
 709                 goto out_commit;
 710         }
 711
 712         err = ocfs2_add_entry(handle, dentry, inode,
 713                               OCFS2_I(inode)->ip_blkno,
 714                               parent_fe_bh, &lookup);
 715         if (err) {
 716                 ocfs2_add_links_count(fe, -1);
 717                 drop_nlink(inode);
 718                 mlog_errno(err);
 719                 goto out_commit;
 720         }
 721
 722         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 723         if (err) {
 724                 mlog_errno(err);
 725                 goto out_commit;
 726         }
 727
 728         atomic_inc(&inode->i_count);
 729         dentry->d_op = &ocfs2_dentry_ops;
 730         d_instantiate(dentry, inode);
 731
 732 out_commit:
 733         ocfs2_commit_trans(osb, handle);
 734 out_unlock_inode:
 735         ocfs2_inode_unlock(inode, 1);
 736
 737 out:
 738         ocfs2_inode_unlock(dir, 1);
 739
 740         brelse(fe_bh);
 741         brelse(parent_fe_bh);
 742
 743         ocfs2_free_dir_lookup_result(&lookup);
 744
 745         mlog_exit(err);
 746
 747         return err;
 748 }
 749
 750 /*
 751  * Takes and drops an exclusive lock on the given dentry. This will
 752  * force other nodes to drop it.
 753  */
 754 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 755 {
 756         int ret;
 757
 758         ret = ocfs2_dentry_lock(dentry, 1);
 759         if (ret)
 760                 mlog_errno(ret);
 761         else
 762                 ocfs2_dentry_unlock(dentry, 1);
 763
 764         return ret;
 765 }
 766
 767 static inline int inode_is_unlinkable(struct inode *inode)
 768 {
 769         if (S_ISDIR(inode->i_mode)) {
 770                 if (inode->i_nlink == 2)
 771                         return 1;
 772                 return 0;
 773         }
 774
 775         if (inode->i_nlink == 1)
 776                 return 1;
 777         return 0;
 778 }
 779
 780 static int ocfs2_unlink(struct inode *dir,
 781                         struct dentry *dentry)
 782 {
 783         int status;
 784         int child_locked = 0;
 785         struct inode *inode = dentry->d_inode;
 786         struct inode *orphan_dir = NULL;
 787         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 788         u64 blkno;
 789         struct ocfs2_dinode *fe = NULL;
 790         struct buffer_head *fe_bh = NULL;
 791         struct buffer_head *parent_node_bh = NULL;
 792         handle_t *handle = NULL;
 793         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 794         struct ocfs2_dir_lookup_result lookup = { NULL, };
 795         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 796
 797         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 798                    dentry->d_name.len, dentry->d_name.name);
 799
 800         dquot_initialize(dir);
 801
 802         BUG_ON(dentry->d_parent->d_inode != dir);
 803
 804         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 805
 806         if (inode == osb->root_inode) {
 807                 mlog(0, "Cannot delete the root directory\n");
 808                 return -EPERM;
 809         }
 810
 811         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 812                                          OI_LS_PARENT);
 813         if (status < 0) {
 814                 if (status != -ENOENT)
 815                         mlog_errno(status);
 816                 return status;
 817         }
 818
 819         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 820                                           dentry->d_name.len, &blkno, dir,
 821                                           &lookup);
 822         if (status < 0) {
 823                 if (status != -ENOENT)
 824                         mlog_errno(status);
 825                 goto leave;
 826         }
 827
 828         if (OCFS2_I(inode)->ip_blkno != blkno) {
 829                 status = -ENOENT;
 830
 831                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 832                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 833                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 834                 goto leave;
 835         }
 836
 837         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 838         if (status < 0) {
 839                 if (status != -ENOENT)
 840                         mlog_errno(status);
 841                 goto leave;
 842         }
 843         child_locked = 1;
 844
 845         if (S_ISDIR(inode->i_mode)) {
 846                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 847                         status = -ENOTEMPTY;
 848                         goto leave;
 849                 }
 850         }
 851
 852         status = ocfs2_remote_dentry_delete(dentry);
 853         if (status < 0) {
 854                 /* This remote delete should succeed under all normal
 855                  * circumstances. */
 856                 mlog_errno(status);
 857                 goto leave;
 858         }
 859
 860         if (inode_is_unlinkable(inode)) {
 861                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 862                                                   OCFS2_I(inode)->ip_blkno,
 863                                                   orphan_name, &orphan_insert);
 864                 if (status < 0) {
 865                         mlog_errno(status);
 866                         goto leave;
 867                 }
 868         }
 869
 870         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 871         if (IS_ERR(handle)) {
 872                 status = PTR_ERR(handle);
 873                 handle = NULL;
 874                 mlog_errno(status);
 875                 goto leave;
 876         }
 877
 878         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 879                                          OCFS2_JOURNAL_ACCESS_WRITE);
 880         if (status < 0) {
 881                 mlog_errno(status);
 882                 goto leave;
 883         }
 884
 885         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 886
 887         if (inode_is_unlinkable(inode)) {
 888                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 889                                           &orphan_insert, orphan_dir);
 890                 if (status < 0) {
 891                         mlog_errno(status);
 892                         goto leave;
 893                 }
 894         }
 895
 896         /* delete the name from the parent dir */
 897         status = ocfs2_delete_entry(handle, dir, &lookup);
 898         if (status < 0) {
 899                 mlog_errno(status);
 900                 goto leave;
 901         }
 902
 903         if (S_ISDIR(inode->i_mode))
 904                 drop_nlink(inode);
 905         drop_nlink(inode);
 906         ocfs2_set_links_count(fe, inode->i_nlink);
 907
 908         status = ocfs2_journal_dirty(handle, fe_bh);
 909         if (status < 0) {
 910                 mlog_errno(status);
 911                 goto leave;
 912         }
 913
 914         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 915         if (S_ISDIR(inode->i_mode))
 916                 drop_nlink(dir);
 917
 918         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 919         if (status < 0) {
 920                 mlog_errno(status);
 921                 if (S_ISDIR(inode->i_mode))
 922                         inc_nlink(dir);
 923         }
 924
 925 leave:
 926         if (handle)
 927                 ocfs2_commit_trans(osb, handle);
 928
 929         if (child_locked)
 930                 ocfs2_inode_unlock(inode, 1);
 931
 932         ocfs2_inode_unlock(dir, 1);
 933
 934         if (orphan_dir) {
 935                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 936                 ocfs2_inode_unlock(orphan_dir, 1);
 937                 mutex_unlock(&orphan_dir->i_mutex);
 938                 iput(orphan_dir);
 939         }
 940
 941         brelse(fe_bh);
 942         brelse(parent_node_bh);
 943
 944         ocfs2_free_dir_lookup_result(&orphan_insert);
 945         ocfs2_free_dir_lookup_result(&lookup);
 946
 947         mlog_exit(status);
 948
 949         return status;
 950 }
 951
 952 /*
 953  * The only place this should be used is rename!
 954  * if they have the same id, then the 1st one is the only one locked.
 955  */
 956 static int ocfs2_double_lock(struct ocfs2_super *osb,
 957                              struct buffer_head **bh1,
 958                              struct inode *inode1,
 959                              struct buffer_head **bh2,
 960                              struct inode *inode2)
 961 {
 962         int status;
 963         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 964         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 965         struct buffer_head **tmpbh;
 966         struct inode *tmpinode;
 967
 968         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 969                    (unsigned long long)oi1->ip_blkno,
 970                    (unsigned long long)oi2->ip_blkno);
 971
 972         if (*bh1)
 973                 *bh1 = NULL;
 974         if (*bh2)
 975                 *bh2 = NULL;
 976
 977         /* we always want to lock the one with the lower lockid first. */
 978         if (oi1->ip_blkno != oi2->ip_blkno) {
 979                 if (oi1->ip_blkno < oi2->ip_blkno) {
 980                         /* switch id1 and id2 around */
 981                         mlog(0, "switching them around...\n");
 982                         tmpbh = bh2;
 983                         bh2 = bh1;
 984                         bh1 = tmpbh;
 985
 986                         tmpinode = inode2;
 987                         inode2 = inode1;
 988                         inode1 = tmpinode;
 989                 }
 990                 /* lock id2 */
 991                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 992                                                  OI_LS_RENAME1);
 993                 if (status < 0) {
 994                         if (status != -ENOENT)
 995                                 mlog_errno(status);
 996                         goto bail;
 997                 }
 998         }
 999
1000         /* lock id1 */
1001         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1002         if (status < 0) {
1003                 /*
1004                  * An error return must mean that no cluster locks
1005                  * were held on function exit.
1006                  */
1007                 if (oi1->ip_blkno != oi2->ip_blkno)
1008                         ocfs2_inode_unlock(inode2, 1);
1009
1010                 if (status != -ENOENT)
1011                         mlog_errno(status);
1012         }
1013
1014 bail:
1015         mlog_exit(status);
1016         return status;
1017 }
1018
1019 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1020 {
1021         ocfs2_inode_unlock(inode1, 1);
1022
1023         if (inode1 != inode2)
1024                 ocfs2_inode_unlock(inode2, 1);
1025 }
1026
1027 static int ocfs2_rename(struct inode *old_dir,
1028                         struct dentry *old_dentry,
1029                         struct inode *new_dir,
1030                         struct dentry *new_dentry)
1031 {
1032         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1033         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1034         struct inode *old_inode = old_dentry->d_inode;
1035         struct inode *new_inode = new_dentry->d_inode;
1036         struct inode *orphan_dir = NULL;
1037         struct ocfs2_dinode *newfe = NULL;
1038         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1039         struct buffer_head *newfe_bh = NULL;
1040         struct buffer_head *old_inode_bh = NULL;
1041         struct ocfs2_super *osb = NULL;
1042         u64 newfe_blkno, old_de_ino;
1043         handle_t *handle = NULL;
1044         struct buffer_head *old_dir_bh = NULL;
1045         struct buffer_head *new_dir_bh = NULL;
1046         nlink_t old_dir_nlink = old_dir->i_nlink;
1047         struct ocfs2_dinode *old_di;
1048         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1049         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1050         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1051         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1052         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1053
1054         /* At some point it might be nice to break this function up a
1055          * bit. */
1056
1057         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1058                    old_dir, old_dentry, new_dir, new_dentry,
1059                    old_dentry->d_name.len, old_dentry->d_name.name,
1060                    new_dentry->d_name.len, new_dentry->d_name.name);
1061
1062         dquot_initialize(old_dir);
1063         dquot_initialize(new_dir);
1064
1065         osb = OCFS2_SB(old_dir->i_sb);
1066
1067         if (new_inode) {
1068                 if (!igrab(new_inode))
1069                         BUG();
1070         }
1071
1072         /* Assume a directory hierarchy thusly:
1073          * a/b/c
1074          * a/d
1075          * a,b,c, and d are all directories.
1076          *
1077          * from cwd of 'a' on both nodes:
1078          * node1: mv b/c d
1079          * node2: mv d   b/c
1080          *
1081          * And that's why, just like the VFS, we need a file system
1082          * rename lock. */
1083         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1084                 status = ocfs2_rename_lock(osb);
1085                 if (status < 0) {
1086                         mlog_errno(status);
1087                         goto bail;
1088                 }
1089                 rename_lock = 1;
1090         }
1091
1092         /* if old and new are the same, this'll just do one lock. */
1093         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1094                                    &new_dir_bh, new_dir);
1095         if (status < 0) {
1096                 mlog_errno(status);
1097                 goto bail;
1098         }
1099         parents_locked = 1;
1100
1101         /* make sure both dirs have bhs
1102          * get an extra ref on old_dir_bh if old==new */
1103         if (!new_dir_bh) {
1104                 if (old_dir_bh) {
1105                         new_dir_bh = old_dir_bh;
1106                         get_bh(new_dir_bh);
1107                 } else {
1108                         mlog(ML_ERROR, "no old_dir_bh!\n");
1109                         status = -EIO;
1110                         goto bail;
1111                 }
1112         }
1113
1114         /*
1115          * Aside from allowing a meta data update, the locking here
1116          * also ensures that the downconvert thread on other nodes
1117          * won't have to concurrently downconvert the inode and the
1118          * dentry locks.
1119          */
1120         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1121                                          OI_LS_PARENT);
1122         if (status < 0) {
1123                 if (status != -ENOENT)
1124                         mlog_errno(status);
1125                 goto bail;
1126         }
1127         old_child_locked = 1;
1128
1129         status = ocfs2_remote_dentry_delete(old_dentry);
1130         if (status < 0) {
1131                 mlog_errno(status);
1132                 goto bail;
1133         }
1134
1135         if (S_ISDIR(old_inode->i_mode)) {
1136                 u64 old_inode_parent;
1137
1138                 update_dot_dot = 1;
1139                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1140                                                   old_inode,
1141                                                   &old_inode_dot_dot_res);
1142                 if (status) {
1143                         status = -EIO;
1144                         goto bail;
1145                 }
1146
1147                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1148                         status = -EIO;
1149                         goto bail;
1150                 }
1151
1152                 if (!new_inode && new_dir != old_dir &&
1153                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1154                         status = -EMLINK;
1155                         goto bail;
1156                 }
1157         }
1158
1159         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1160                                             old_dentry->d_name.len,
1161                                             &old_de_ino);
1162         if (status) {
1163                 status = -ENOENT;
1164                 goto bail;
1165         }
1166
1167         /*
1168          *  Check for inode number is _not_ due to possible IO errors.
1169          *  We might rmdir the source, keep it as pwd of some process
1170          *  and merrily kill the link to whatever was created under the
1171          *  same name. Goodbye sticky bit ;-<
1172          */
1173         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1174                 status = -ENOENT;
1175                 goto bail;
1176         }
1177
1178         /* check if the target already exists (in which case we need
1179          * to delete it */
1180         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1181                                           new_dentry->d_name.len,
1182                                           &newfe_blkno, new_dir,
1183                                           &target_lookup_res);
1184         /* The only error we allow here is -ENOENT because the new
1185          * file not existing is perfectly valid. */
1186         if ((status < 0) && (status != -ENOENT)) {
1187                 /* If we cannot find the file specified we should just */
1188                 /* return the error... */
1189                 mlog_errno(status);
1190                 goto bail;
1191         }
1192         if (status == 0)
1193                 target_exists = 1;
1194
1195         if (!target_exists && new_inode) {
1196                 /*
1197                  * Target was unlinked by another node while we were
1198                  * waiting to get to ocfs2_rename(). There isn't
1199                  * anything we can do here to help the situation, so
1200                  * bubble up the appropriate error.
1201                  */
1202                 status = -ENOENT;
1203                 goto bail;
1204         }
1205
1206         /* In case we need to overwrite an existing file, we blow it
1207          * away first */
1208         if (target_exists) {
1209                 /* VFS didn't think there existed an inode here, but
1210                  * someone else in the cluster must have raced our
1211                  * rename to create one. Today we error cleanly, in
1212                  * the future we should consider calling iget to build
1213                  * a new struct inode for this entry. */
1214                 if (!new_inode) {
1215                         status = -EACCES;
1216
1217                         mlog(0, "We found an inode for name %.*s but VFS "
1218                              "didn't give us one.\n", new_dentry->d_name.len,
1219                              new_dentry->d_name.name);
1220                         goto bail;
1221                 }
1222
1223                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1224                         status = -EACCES;
1225
1226                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1227                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1228                              (unsigned long long)newfe_blkno,
1229                              OCFS2_I(new_inode)->ip_flags);
1230                         goto bail;
1231                 }
1232
1233                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1234                 if (status < 0) {
1235                         if (status != -ENOENT)
1236                                 mlog_errno(status);
1237                         goto bail;
1238                 }
1239                 new_child_locked = 1;
1240
1241                 status = ocfs2_remote_dentry_delete(new_dentry);
1242                 if (status < 0) {
1243                         mlog_errno(status);
1244                         goto bail;
1245                 }
1246
1247                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1248
1249                 mlog(0, "aha rename over existing... new_blkno=%llu "
1250                      "newfebh=%p bhblocknr=%llu\n",
1251                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1252                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1253
1254                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1255                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1256                                                 OCFS2_I(new_inode)->ip_blkno,
1257                                                 orphan_name, &orphan_insert);
1258                         if (status < 0) {
1259                                 mlog_errno(status);
1260                                 goto bail;
1261                         }
1262                 }
1263         } else {
1264                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1265
1266                 status = ocfs2_check_dir_for_entry(new_dir,
1267                                                    new_dentry->d_name.name,
1268                                                    new_dentry->d_name.len);
1269                 if (status)
1270                         goto bail;
1271
1272                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1273                                                       new_dentry->d_name.name,
1274                                                       new_dentry->d_name.len,
1275                                                       &target_insert);
1276                 if (status < 0) {
1277                         mlog_errno(status);
1278                         goto bail;
1279                 }
1280         }
1281
1282         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1283         if (IS_ERR(handle)) {
1284                 status = PTR_ERR(handle);
1285                 handle = NULL;
1286                 mlog_errno(status);
1287                 goto bail;
1288         }
1289
1290         if (target_exists) {
1291                 if (S_ISDIR(new_inode->i_mode)) {
1292                         if (new_inode->i_nlink != 2 ||
1293                             !ocfs2_empty_dir(new_inode)) {
1294                                 status = -ENOTEMPTY;
1295                                 goto bail;
1296                         }
1297                 }
1298                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1299                                                  newfe_bh,
1300                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1301                 if (status < 0) {
1302                         mlog_errno(status);
1303                         goto bail;
1304                 }
1305
1306                 if (S_ISDIR(new_inode->i_mode) ||
1307                     (ocfs2_read_links_count(newfe) == 1)) {
1308                         status = ocfs2_orphan_add(osb, handle, new_inode,
1309                                                   newfe_bh, orphan_name,
1310                                                   &orphan_insert, orphan_dir);
1311                         if (status < 0) {
1312                                 mlog_errno(status);
1313                                 goto bail;
1314                         }
1315                 }
1316
1317                 /* change the dirent to point to the correct inode */
1318                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1319                                             old_inode);
1320                 if (status < 0) {
1321                         mlog_errno(status);
1322                         goto bail;
1323                 }
1324                 new_dir->i_version++;
1325
1326                 if (S_ISDIR(new_inode->i_mode))
1327                         ocfs2_set_links_count(newfe, 0);
1328                 else
1329                         ocfs2_add_links_count(newfe, -1);
1330
1331                 status = ocfs2_journal_dirty(handle, newfe_bh);
1332                 if (status < 0) {
1333                         mlog_errno(status);
1334                         goto bail;
1335                 }
1336         } else {
1337                 /* if the name was not found in new_dir, add it now */
1338                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1339                                          OCFS2_I(old_inode)->ip_blkno,
1340                                          new_dir_bh, &target_insert);
1341         }
1342
1343         old_inode->i_ctime = CURRENT_TIME;
1344         mark_inode_dirty(old_inode);
1345
1346         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1347                                          old_inode_bh,
1348                                          OCFS2_JOURNAL_ACCESS_WRITE);
1349         if (status >= 0) {
1350                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1351
1352                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1353                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1354
1355                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1356                 if (status < 0)
1357                         mlog_errno(status);
1358         } else
1359                 mlog_errno(status);
1360
1361         /*
1362          * Now that the name has been added to new_dir, remove the old name.
1363          *
1364          * We don't keep any directory entry context around until now
1365          * because the insert might have changed the type of directory
1366          * we're dealing with.
1367          */
1368         status = ocfs2_find_entry(old_dentry->d_name.name,
1369                                   old_dentry->d_name.len, old_dir,
1370                                   &old_entry_lookup);
1371         if (status)
1372                 goto bail;
1373
1374         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1375         if (status < 0) {
1376                 mlog_errno(status);
1377                 goto bail;
1378         }
1379
1380         if (new_inode) {
1381                 new_inode->i_nlink--;
1382                 new_inode->i_ctime = CURRENT_TIME;
1383         }
1384         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1385
1386         if (update_dot_dot) {
1387                 status = ocfs2_update_entry(old_inode, handle,
1388                                             &old_inode_dot_dot_res, new_dir);
1389                 old_dir->i_nlink--;
1390                 if (new_inode) {
1391                         new_inode->i_nlink--;
1392                 } else {
1393                         inc_nlink(new_dir);
1394                         mark_inode_dirty(new_dir);
1395                 }
1396         }
1397         mark_inode_dirty(old_dir);
1398         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1399         if (new_inode) {
1400                 mark_inode_dirty(new_inode);
1401                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1402         }
1403
1404         if (old_dir != new_dir) {
1405                 /* Keep the same times on both directories.*/
1406                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1407
1408                 /*
1409                  * This will also pick up the i_nlink change from the
1410                  * block above.
1411                  */
1412                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1413         }
1414
1415         if (old_dir_nlink != old_dir->i_nlink) {
1416                 if (!old_dir_bh) {
1417                         mlog(ML_ERROR, "need to change nlink for old dir "
1418                              "%llu from %d to %d but bh is NULL!\n",
1419                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1420                              (int)old_dir_nlink, old_dir->i_nlink);
1421                 } else {
1422                         struct ocfs2_dinode *fe;
1423                         status = ocfs2_journal_access_di(handle,
1424                                                          INODE_CACHE(old_dir),
1425                                                          old_dir_bh,
1426                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1427                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1428                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1429                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1430                 }
1431         }
1432         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1433         status = 0;
1434 bail:
1435         if (rename_lock)
1436                 ocfs2_rename_unlock(osb);
1437
1438         if (handle)
1439                 ocfs2_commit_trans(osb, handle);
1440
1441         if (parents_locked)
1442                 ocfs2_double_unlock(old_dir, new_dir);
1443
1444         if (old_child_locked)
1445                 ocfs2_inode_unlock(old_inode, 1);
1446
1447         if (new_child_locked)
1448                 ocfs2_inode_unlock(new_inode, 1);
1449
1450         if (orphan_dir) {
1451                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1452                 ocfs2_inode_unlock(orphan_dir, 1);
1453                 mutex_unlock(&orphan_dir->i_mutex);
1454                 iput(orphan_dir);
1455         }
1456
1457         if (new_inode)
1458                 sync_mapping_buffers(old_inode->i_mapping);
1459
1460         if (new_inode)
1461                 iput(new_inode);
1462
1463         ocfs2_free_dir_lookup_result(&target_lookup_res);
1464         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1465         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1466         ocfs2_free_dir_lookup_result(&orphan_insert);
1467         ocfs2_free_dir_lookup_result(&target_insert);
1468
1469         brelse(newfe_bh);
1470         brelse(old_inode_bh);
1471         brelse(old_dir_bh);
1472         brelse(new_dir_bh);
1473
1474         mlog_exit(status);
1475
1476         return status;
1477 }
1478
1479 /*
1480  * we expect i_size = strlen(symname). Copy symname into the file
1481  * data, including the null terminator.
1482  */
1483 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1484                                      handle_t *handle,
1485                                      struct inode *inode,
1486                                      const char *symname)
1487 {
1488         struct buffer_head **bhs = NULL;
1489         const char *c;
1490         struct super_block *sb = osb->sb;
1491         u64 p_blkno, p_blocks;
1492         int virtual, blocks, status, i, bytes_left;
1493
1494         bytes_left = i_size_read(inode) + 1;
1495         /* we can't trust i_blocks because we're actually going to
1496          * write i_size + 1 bytes. */
1497         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1498
1499         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1500                         (unsigned long long)inode->i_blocks,
1501                         i_size_read(inode), blocks);
1502
1503         /* Sanity check -- make sure we're going to fit. */
1504         if (bytes_left >
1505             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1506                 status = -EIO;
1507                 mlog_errno(status);
1508                 goto bail;
1509         }
1510
1511         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1512         if (!bhs) {
1513                 status = -ENOMEM;
1514                 mlog_errno(status);
1515                 goto bail;
1516         }
1517
1518         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1519                                              NULL);
1520         if (status < 0) {
1521                 mlog_errno(status);
1522                 goto bail;
1523         }
1524
1525         /* links can never be larger than one cluster so we know this
1526          * is all going to be contiguous, but do a sanity check
1527          * anyway. */
1528         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1529                 status = -EIO;
1530                 mlog_errno(status);
1531                 goto bail;
1532         }
1533
1534         virtual = 0;
1535         while(bytes_left > 0) {
1536                 c = &symname[virtual * sb->s_blocksize];
1537
1538                 bhs[virtual] = sb_getblk(sb, p_blkno);
1539                 if (!bhs[virtual]) {
1540                         status = -ENOMEM;
1541                         mlog_errno(status);
1542                         goto bail;
1543                 }
1544                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1545                                               bhs[virtual]);
1546
1547                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1548                                               bhs[virtual],
1549                                               OCFS2_JOURNAL_ACCESS_CREATE);
1550                 if (status < 0) {
1551                         mlog_errno(status);
1552                         goto bail;
1553                 }
1554
1555                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1556
1557                 memcpy(bhs[virtual]->b_data, c,
1558                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1559                        bytes_left);
1560
1561                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1562                 if (status < 0) {
1563                         mlog_errno(status);
1564                         goto bail;
1565                 }
1566
1567                 virtual++;
1568                 p_blkno++;
1569                 bytes_left -= sb->s_blocksize;
1570         }
1571
1572         status = 0;
1573 bail:
1574
1575         if (bhs) {
1576                 for(i = 0; i < blocks; i++)
1577                         brelse(bhs[i]);
1578                 kfree(bhs);
1579         }
1580
1581         mlog_exit(status);
1582         return status;
1583 }
1584
1585 static int ocfs2_symlink(struct inode *dir,
1586                          struct dentry *dentry,
1587                          const char *symname)
1588 {
1589         int status, l, credits;
1590         u64 newsize;
1591         struct ocfs2_super *osb = NULL;
1592         struct inode *inode = NULL;
1593         struct super_block *sb;
1594         struct buffer_head *new_fe_bh = NULL;
1595         struct buffer_head *parent_fe_bh = NULL;
1596         struct ocfs2_dinode *fe = NULL;
1597         struct ocfs2_dinode *dirfe;
1598         handle_t *handle = NULL;
1599         struct ocfs2_alloc_context *inode_ac = NULL;
1600         struct ocfs2_alloc_context *data_ac = NULL;
1601         struct ocfs2_alloc_context *xattr_ac = NULL;
1602         int want_clusters = 0;
1603         int xattr_credits = 0;
1604         struct ocfs2_security_xattr_info si = {
1605                 .enable = 1,
1606         };
1607         int did_quota = 0, did_quota_inode = 0;
1608         struct ocfs2_dir_lookup_result lookup = { NULL, };
1609
1610         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1611                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1612
1613         dquot_initialize(dir);
1614
1615         sb = dir->i_sb;
1616         osb = OCFS2_SB(sb);
1617
1618         l = strlen(symname) + 1;
1619
1620         credits = ocfs2_calc_symlink_credits(sb);
1621
1622         /* lock the parent directory */
1623         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1624         if (status < 0) {
1625                 if (status != -ENOENT)
1626                         mlog_errno(status);
1627                 return status;
1628         }
1629
1630         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1631         if (!ocfs2_read_links_count(dirfe)) {
1632                 /* can't make a file in a deleted directory. */
1633                 status = -ENOENT;
1634                 goto bail;
1635         }
1636
1637         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1638                                            dentry->d_name.len);
1639         if (status)
1640                 goto bail;
1641
1642         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1643                                               dentry->d_name.name,
1644                                               dentry->d_name.len, &lookup);
1645         if (status < 0) {
1646                 mlog_errno(status);
1647                 goto bail;
1648         }
1649
1650         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1651         if (status < 0) {
1652                 if (status != -ENOSPC)
1653                         mlog_errno(status);
1654                 goto bail;
1655         }
1656
1657         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1658         if (!inode) {
1659                 status = -ENOMEM;
1660                 mlog_errno(status);
1661                 goto bail;
1662         }
1663
1664         /* get security xattr */
1665         status = ocfs2_init_security_get(inode, dir, &si);
1666         if (status) {
1667                 if (status == -EOPNOTSUPP)
1668                         si.enable = 0;
1669                 else {
1670                         mlog_errno(status);
1671                         goto bail;
1672                 }
1673         }
1674
1675         /* calculate meta data/clusters for setting security xattr */
1676         if (si.enable) {
1677                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1678                                                   &xattr_credits, &xattr_ac);
1679                 if (status < 0) {
1680                         mlog_errno(status);
1681                         goto bail;
1682                 }
1683         }
1684
1685         /* don't reserve bitmap space for fast symlinks. */
1686         if (l > ocfs2_fast_symlink_chars(sb))
1687                 want_clusters += 1;
1688
1689         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1690         if (status < 0) {
1691                 if (status != -ENOSPC)
1692                         mlog_errno(status);
1693                 goto bail;
1694         }
1695
1696         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1697         if (IS_ERR(handle)) {
1698                 status = PTR_ERR(handle);
1699                 handle = NULL;
1700                 mlog_errno(status);
1701                 goto bail;
1702         }
1703
1704         status = dquot_alloc_inode(inode);
1705         if (status)
1706                 goto bail;
1707         did_quota_inode = 1;
1708
1709         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1710                    inode->i_mode, dentry->d_name.len,
1711                    dentry->d_name.name);
1712
1713         status = ocfs2_mknod_locked(osb, dir, inode,
1714                                     0, &new_fe_bh, parent_fe_bh, handle,
1715                                     inode_ac);
1716         if (status < 0) {
1717                 mlog_errno(status);
1718                 goto bail;
1719         }
1720
1721         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1722         inode->i_rdev = 0;
1723         newsize = l - 1;
1724         if (l > ocfs2_fast_symlink_chars(sb)) {
1725                 u32 offset = 0;
1726
1727                 inode->i_op = &ocfs2_symlink_inode_operations;
1728                 status = dquot_alloc_space_nodirty(inode,
1729                     ocfs2_clusters_to_bytes(osb->sb, 1));
1730                 if (status)
1731                         goto bail;
1732                 did_quota = 1;
1733                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1734                                               new_fe_bh,
1735                                               handle, data_ac, NULL,
1736                                               NULL);
1737                 if (status < 0) {
1738                         if (status != -ENOSPC && status != -EINTR) {
1739                                 mlog(ML_ERROR,
1740                                      "Failed to extend file to %llu\n",
1741                                      (unsigned long long)newsize);
1742                                 mlog_errno(status);
1743                                 status = -ENOSPC;
1744                         }
1745                         goto bail;
1746                 }
1747                 i_size_write(inode, newsize);
1748                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1749         } else {
1750                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1751                 memcpy((char *) fe->id2.i_symlink, symname, l);
1752                 i_size_write(inode, newsize);
1753                 inode->i_blocks = 0;
1754         }
1755
1756         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1757         if (status < 0) {
1758                 mlog_errno(status);
1759                 goto bail;
1760         }
1761
1762         if (!ocfs2_inode_is_fast_symlink(inode)) {
1763                 status = ocfs2_create_symlink_data(osb, handle, inode,
1764                                                    symname);
1765                 if (status < 0) {
1766                         mlog_errno(status);
1767                         goto bail;
1768                 }
1769         }
1770
1771         if (si.enable) {
1772                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1773                                                  xattr_ac, data_ac);
1774                 if (status < 0) {
1775                         mlog_errno(status);
1776                         goto bail;
1777                 }
1778         }
1779
1780         status = ocfs2_add_entry(handle, dentry, inode,
1781                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1782                                  &lookup);
1783         if (status < 0) {
1784                 mlog_errno(status);
1785                 goto bail;
1786         }
1787
1788         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1789         if (status) {
1790                 mlog_errno(status);
1791                 goto bail;
1792         }
1793
1794         insert_inode_hash(inode);
1795         dentry->d_op = &ocfs2_dentry_ops;
1796         d_instantiate(dentry, inode);
1797 bail:
1798         if (status < 0 && did_quota)
1799                 dquot_free_space_nodirty(inode,
1800                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1801         if (status < 0 && did_quota_inode)
1802                 dquot_free_inode(inode);
1803         if (handle)
1804                 ocfs2_commit_trans(osb, handle);
1805
1806         ocfs2_inode_unlock(dir, 1);
1807
1808         brelse(new_fe_bh);
1809         brelse(parent_fe_bh);
1810         kfree(si.name);
1811         kfree(si.value);
1812         ocfs2_free_dir_lookup_result(&lookup);
1813         if (inode_ac)
1814                 ocfs2_free_alloc_context(inode_ac);
1815         if (data_ac)
1816                 ocfs2_free_alloc_context(data_ac);
1817         if (xattr_ac)
1818                 ocfs2_free_alloc_context(xattr_ac);
1819         if ((status < 0) && inode) {
1820                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1821                 clear_nlink(inode);
1822                 iput(inode);
1823         }
1824
1825         mlog_exit(status);
1826
1827         return status;
1828 }
1829
1830 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1831 {
1832         int status, namelen;
1833
1834         mlog_entry_void();
1835
1836         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1837                            (long long)blkno);
1838         if (namelen <= 0) {
1839                 if (namelen)
1840                         status = namelen;
1841                 else
1842                         status = -EINVAL;
1843                 mlog_errno(status);
1844                 goto bail;
1845         }
1846         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1847                 status = -EINVAL;
1848                 mlog_errno(status);
1849                 goto bail;
1850         }
1851
1852         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1853              namelen);
1854
1855         status = 0;
1856 bail:
1857         mlog_exit(status);
1858         return status;
1859 }
1860
1861 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1862                                     struct inode **ret_orphan_dir,
1863                                     u64 blkno,
1864                                     char *name,
1865                                     struct ocfs2_dir_lookup_result *lookup)
1866 {
1867         struct inode *orphan_dir_inode;
1868         struct buffer_head *orphan_dir_bh = NULL;
1869         int status = 0;
1870
1871         status = ocfs2_blkno_stringify(blkno, name);
1872         if (status < 0) {
1873                 mlog_errno(status);
1874                 return status;
1875         }
1876
1877         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1878                                                        ORPHAN_DIR_SYSTEM_INODE,
1879                                                        osb->slot_num);
1880         if (!orphan_dir_inode) {
1881                 status = -ENOENT;
1882                 mlog_errno(status);
1883                 return status;
1884         }
1885
1886         mutex_lock(&orphan_dir_inode->i_mutex);
1887
1888         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1889         if (status < 0) {
1890                 mlog_errno(status);
1891                 goto leave;
1892         }
1893
1894         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1895                                               orphan_dir_bh, name,
1896                                               OCFS2_ORPHAN_NAMELEN, lookup);
1897         if (status < 0) {
1898                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1899
1900                 mlog_errno(status);
1901                 goto leave;
1902         }
1903
1904         *ret_orphan_dir = orphan_dir_inode;
1905
1906 leave:
1907         if (status) {
1908                 mutex_unlock(&orphan_dir_inode->i_mutex);
1909                 iput(orphan_dir_inode);
1910         }
1911
1912         brelse(orphan_dir_bh);
1913
1914         mlog_exit(status);
1915         return status;
1916 }
1917
1918 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1919                             handle_t *handle,
1920                             struct inode *inode,
1921                             struct buffer_head *fe_bh,
1922                             char *name,
1923                             struct ocfs2_dir_lookup_result *lookup,
1924                             struct inode *orphan_dir_inode)
1925 {
1926         struct buffer_head *orphan_dir_bh = NULL;
1927         int status = 0;
1928         struct ocfs2_dinode *orphan_fe;
1929         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1930
1931         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1932
1933         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1934         if (status < 0) {
1935                 mlog_errno(status);
1936                 goto leave;
1937         }
1938
1939         status = ocfs2_journal_access_di(handle,
1940                                          INODE_CACHE(orphan_dir_inode),
1941                                          orphan_dir_bh,
1942                                          OCFS2_JOURNAL_ACCESS_WRITE);
1943         if (status < 0) {
1944                 mlog_errno(status);
1945                 goto leave;
1946         }
1947
1948         /* we're a cluster, and nlink can change on disk from
1949          * underneath us... */
1950         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1951         if (S_ISDIR(inode->i_mode))
1952                 ocfs2_add_links_count(orphan_fe, 1);
1953         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
1954
1955         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1956         if (status < 0) {
1957                 mlog_errno(status);
1958                 goto leave;
1959         }
1960
1961         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1962                                    OCFS2_ORPHAN_NAMELEN, inode,
1963                                    OCFS2_I(inode)->ip_blkno,
1964                                    orphan_dir_bh, lookup);
1965         if (status < 0) {
1966                 mlog_errno(status);
1967                 goto leave;
1968         }
1969
1970         /*
1971          * We're going to journal the change of i_flags and i_orphaned_slot.
1972          * It's safe anyway, though some callers may duplicate the journaling.
1973          * Journaling within the func just make the logic look more
1974          * straightforward.
1975          */
1976         status = ocfs2_journal_access_di(handle,
1977                                          INODE_CACHE(inode),
1978                                          fe_bh,
1979                                          OCFS2_JOURNAL_ACCESS_WRITE);
1980         if (status < 0) {
1981                 mlog_errno(status);
1982                 goto leave;
1983         }
1984
1985         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1986         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
1987
1988         /* Record which orphan dir our inode now resides
1989          * in. delete_inode will use this to determine which orphan
1990          * dir to lock. */
1991         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1992
1993         ocfs2_journal_dirty(handle, fe_bh);
1994
1995         mlog(0, "Inode %llu orphaned in slot %d\n",
1996              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1997
1998 leave:
1999         brelse(orphan_dir_bh);
2000
2001         mlog_exit(status);
2002         return status;
2003 }
2004
2005 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2006 int ocfs2_orphan_del(struct ocfs2_super *osb,
2007                      handle_t *handle,
2008                      struct inode *orphan_dir_inode,
2009                      struct inode *inode,
2010                      struct buffer_head *orphan_dir_bh)
2011 {
2012         char name[OCFS2_ORPHAN_NAMELEN + 1];
2013         struct ocfs2_dinode *orphan_fe;
2014         int status = 0;
2015         struct ocfs2_dir_lookup_result lookup = { NULL, };
2016
2017         mlog_entry_void();
2018
2019         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2020         if (status < 0) {
2021                 mlog_errno(status);
2022                 goto leave;
2023         }
2024
2025         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2026              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2027              OCFS2_ORPHAN_NAMELEN);
2028
2029         /* find it's spot in the orphan directory */
2030         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2031                                   &lookup);
2032         if (status) {
2033                 mlog_errno(status);
2034                 goto leave;
2035         }
2036
2037         /* remove it from the orphan directory */
2038         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2039         if (status < 0) {
2040                 mlog_errno(status);
2041                 goto leave;
2042         }
2043
2044         status = ocfs2_journal_access_di(handle,
2045                                          INODE_CACHE(orphan_dir_inode),
2046                                          orphan_dir_bh,
2047                                          OCFS2_JOURNAL_ACCESS_WRITE);
2048         if (status < 0) {
2049                 mlog_errno(status);
2050                 goto leave;
2051         }
2052
2053         /* do the i_nlink dance! :) */
2054         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2055         if (S_ISDIR(inode->i_mode))
2056                 ocfs2_add_links_count(orphan_fe, -1);
2057         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2058
2059         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2060         if (status < 0) {
2061                 mlog_errno(status);
2062                 goto leave;
2063         }
2064
2065 leave:
2066         ocfs2_free_dir_lookup_result(&lookup);
2067
2068         mlog_exit(status);
2069         return status;
2070 }
2071
2072 int ocfs2_create_inode_in_orphan(struct inode *dir,
2073                                  int mode,
2074                                  struct inode **new_inode)
2075 {
2076         int status, did_quota_inode = 0;
2077         struct inode *inode = NULL;
2078         struct inode *orphan_dir = NULL;
2079         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2080         struct ocfs2_dinode *di = NULL;
2081         handle_t *handle = NULL;
2082         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2083         struct buffer_head *parent_di_bh = NULL;
2084         struct buffer_head *new_di_bh = NULL;
2085         struct ocfs2_alloc_context *inode_ac = NULL;
2086         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2087
2088         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2089         if (status < 0) {
2090                 if (status != -ENOENT)
2091                         mlog_errno(status);
2092                 return status;
2093         }
2094
2095         /*
2096          * We give the orphan dir the root blkno to fake an orphan name,
2097          * and allocate enough space for our insertion.
2098          */
2099         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
2100                                           osb->root_blkno,
2101                                           orphan_name, &orphan_insert);
2102         if (status < 0) {
2103                 mlog_errno(status);
2104                 goto leave;
2105         }
2106
2107         /* reserve an inode spot */
2108         status = ocfs2_reserve_new_inode(osb, &inode_ac);
2109         if (status < 0) {
2110                 if (status != -ENOSPC)
2111                         mlog_errno(status);
2112                 goto leave;
2113         }
2114
2115         inode = ocfs2_get_init_inode(dir, mode);
2116         if (!inode) {
2117                 status = -ENOMEM;
2118                 mlog_errno(status);
2119                 goto leave;
2120         }
2121
2122         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2123         if (IS_ERR(handle)) {
2124                 status = PTR_ERR(handle);
2125                 handle = NULL;
2126                 mlog_errno(status);
2127                 goto leave;
2128         }
2129
2130         status = dquot_alloc_inode(inode);
2131         if (status)
2132                 goto leave;
2133         did_quota_inode = 1;
2134
2135         inode->i_nlink = 0;
2136         /* do the real work now. */
2137         status = ocfs2_mknod_locked(osb, dir, inode,
2138                                     0, &new_di_bh, parent_di_bh, handle,
2139                                     inode_ac);
2140         if (status < 0) {
2141                 mlog_errno(status);
2142                 goto leave;
2143         }
2144
2145         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, orphan_name);
2146         if (status < 0) {
2147                 mlog_errno(status);
2148                 goto leave;
2149         }
2150
2151         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2152         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2153                                   &orphan_insert, orphan_dir);
2154         if (status < 0) {
2155                 mlog_errno(status);
2156                 goto leave;
2157         }
2158
2159         /* get open lock so that only nodes can't remove it from orphan dir. */
2160         status = ocfs2_open_lock(inode);
2161         if (status < 0)
2162                 mlog_errno(status);
2163
2164         insert_inode_hash(inode);
2165 leave:
2166         if (status < 0 && did_quota_inode)
2167                 dquot_free_inode(inode);
2168         if (handle)
2169                 ocfs2_commit_trans(osb, handle);
2170
2171         if (orphan_dir) {
2172                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2173                 ocfs2_inode_unlock(orphan_dir, 1);
2174                 mutex_unlock(&orphan_dir->i_mutex);
2175                 iput(orphan_dir);
2176         }
2177
2178         if (status == -ENOSPC)
2179                 mlog(0, "Disk is full\n");
2180
2181         if ((status < 0) && inode) {
2182                 clear_nlink(inode);
2183                 iput(inode);
2184         }
2185
2186         if (inode_ac)
2187                 ocfs2_free_alloc_context(inode_ac);
2188
2189         brelse(new_di_bh);
2190
2191         if (!status)
2192                 *new_inode = inode;
2193
2194         ocfs2_free_dir_lookup_result(&orphan_insert);
2195
2196         ocfs2_inode_unlock(dir, 1);
2197         brelse(parent_di_bh);
2198         return status;
2199 }
2200
2201 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2202                                    struct inode *inode,
2203                                    struct dentry *dentry)
2204 {
2205         int status = 0;
2206         struct buffer_head *parent_di_bh = NULL;
2207         handle_t *handle = NULL;
2208         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2209         struct ocfs2_dinode *dir_di, *di;
2210         struct inode *orphan_dir_inode = NULL;
2211         struct buffer_head *orphan_dir_bh = NULL;
2212         struct buffer_head *di_bh = NULL;
2213         struct ocfs2_dir_lookup_result lookup = { NULL, };
2214
2215         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2216                    dentry->d_name.len, dentry->d_name.name);
2217
2218         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2219         if (status < 0) {
2220                 if (status != -ENOENT)
2221                         mlog_errno(status);
2222                 return status;
2223         }
2224
2225         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2226         if (!dir_di->i_links_count) {
2227                 /* can't make a file in a deleted directory. */
2228                 status = -ENOENT;
2229                 goto leave;
2230         }
2231
2232         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2233                                            dentry->d_name.len);
2234         if (status)
2235                 goto leave;
2236
2237         /* get a spot inside the dir. */
2238         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2239                                               dentry->d_name.name,
2240                                               dentry->d_name.len, &lookup);
2241         if (status < 0) {
2242                 mlog_errno(status);
2243                 goto leave;
2244         }
2245
2246         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2247                                                        ORPHAN_DIR_SYSTEM_INODE,
2248                                                        osb->slot_num);
2249         if (!orphan_dir_inode) {
2250                 status = -EEXIST;
2251                 mlog_errno(status);
2252                 goto leave;
2253         }
2254
2255         mutex_lock(&orphan_dir_inode->i_mutex);
2256
2257         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2258         if (status < 0) {
2259                 mlog_errno(status);
2260                 mutex_unlock(&orphan_dir_inode->i_mutex);
2261                 iput(orphan_dir_inode);
2262                 goto leave;
2263         }
2264
2265         status = ocfs2_read_inode_block(inode, &di_bh);
2266         if (status < 0) {
2267                 mlog_errno(status);
2268                 goto orphan_unlock;
2269         }
2270
2271         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2272         if (IS_ERR(handle)) {
2273                 status = PTR_ERR(handle);
2274                 handle = NULL;
2275                 mlog_errno(status);
2276                 goto orphan_unlock;
2277         }
2278
2279         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2280                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2281         if (status < 0) {
2282                 mlog_errno(status);
2283                 goto out_commit;
2284         }
2285
2286         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2287                                   orphan_dir_bh);
2288         if (status < 0) {
2289                 mlog_errno(status);
2290                 goto out_commit;
2291         }
2292
2293         di = (struct ocfs2_dinode *)di_bh->b_data;
2294         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2295         di->i_orphaned_slot = 0;
2296         inode->i_nlink = 1;
2297         ocfs2_set_links_count(di, inode->i_nlink);
2298         ocfs2_journal_dirty(handle, di_bh);
2299
2300         status = ocfs2_add_entry(handle, dentry, inode,
2301                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2302                                  &lookup);
2303         if (status < 0) {
2304                 mlog_errno(status);
2305                 goto out_commit;
2306         }
2307
2308         status = ocfs2_dentry_attach_lock(dentry, inode,
2309                                           OCFS2_I(dir)->ip_blkno);
2310         if (status) {
2311                 mlog_errno(status);
2312                 goto out_commit;
2313         }
2314
2315         dentry->d_op = &ocfs2_dentry_ops;
2316         d_instantiate(dentry, inode);
2317         status = 0;
2318 out_commit:
2319         ocfs2_commit_trans(osb, handle);
2320 orphan_unlock:
2321         ocfs2_inode_unlock(orphan_dir_inode, 1);
2322         mutex_unlock(&orphan_dir_inode->i_mutex);
2323         iput(orphan_dir_inode);
2324 leave:
2325
2326         ocfs2_inode_unlock(dir, 1);
2327
2328         brelse(di_bh);
2329         brelse(parent_di_bh);
2330         brelse(orphan_dir_bh);
2331
2332         ocfs2_free_dir_lookup_result(&lookup);
2333
2334         mlog_exit(status);
2335
2336         return status;
2337 }
2338
2339 const struct inode_operations ocfs2_dir_iops = {
2340         .create         = ocfs2_create,
2341         .lookup         = ocfs2_lookup,
2342         .link           = ocfs2_link,
2343         .unlink         = ocfs2_unlink,
2344         .rmdir          = ocfs2_unlink,
2345         .symlink        = ocfs2_symlink,
2346         .mkdir          = ocfs2_mkdir,
2347         .mknod          = ocfs2_mknod,
2348         .rename         = ocfs2_rename,
2349         .setattr        = ocfs2_setattr,
2350         .getattr        = ocfs2_getattr,
2351         .permission     = ocfs2_permission,
2352         .setxattr       = generic_setxattr,
2353         .getxattr       = generic_getxattr,
2354         .listxattr      = ocfs2_listxattr,
2355         .removexattr    = generic_removexattr,
2356         .fiemap         = ocfs2_fiemap,
2357 };