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_init_owner(inode, dir, mode);
 208         dquot_initialize(inode);
 209         return inode;
 210 }
 211
 212 static int ocfs2_mknod(struct inode *dir,
 213                        struct dentry *dentry,
 214                        int mode,
 215                        dev_t dev)
 216 {
 217         int status = 0;
 218         struct buffer_head *parent_fe_bh = NULL;
 219         handle_t *handle = NULL;
 220         struct ocfs2_super *osb;
 221         struct ocfs2_dinode *dirfe;
 222         struct buffer_head *new_fe_bh = NULL;
 223         struct inode *inode = NULL;
 224         struct ocfs2_alloc_context *inode_ac = NULL;
 225         struct ocfs2_alloc_context *data_ac = NULL;
 226         struct ocfs2_alloc_context *meta_ac = NULL;
 227         int want_clusters = 0;
 228         int want_meta = 0;
 229         int xattr_credits = 0;
 230         struct ocfs2_security_xattr_info si = {
 231                 .enable = 1,
 232         };
 233         int did_quota_inode = 0;
 234         struct ocfs2_dir_lookup_result lookup = { NULL, };
 235         sigset_t oldset;
 236         int did_block_signals = 0;
 237
 238         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
 239                    (unsigned long)dev, dentry->d_name.len,
 240                    dentry->d_name.name);
 241
 242         dquot_initialize(dir);
 243
 244         /* get our super block */
 245         osb = OCFS2_SB(dir->i_sb);
 246
 247         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 248         if (status < 0) {
 249                 if (status != -ENOENT)
 250                         mlog_errno(status);
 251                 return status;
 252         }
 253
 254         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 255                 status = -EMLINK;
 256                 goto leave;
 257         }
 258
 259         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 260         if (!ocfs2_read_links_count(dirfe)) {
 261                 /* can't make a file in a deleted directory. */
 262                 status = -ENOENT;
 263                 goto leave;
 264         }
 265
 266         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 267                                            dentry->d_name.len);
 268         if (status)
 269                 goto leave;
 270
 271         /* get a spot inside the dir. */
 272         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 273                                               dentry->d_name.name,
 274                                               dentry->d_name.len, &lookup);
 275         if (status < 0) {
 276                 mlog_errno(status);
 277                 goto leave;
 278         }
 279
 280         /* reserve an inode spot */
 281         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 282         if (status < 0) {
 283                 if (status != -ENOSPC)
 284                         mlog_errno(status);
 285                 goto leave;
 286         }
 287
 288         inode = ocfs2_get_init_inode(dir, mode);
 289         if (!inode) {
 290                 status = -ENOMEM;
 291                 mlog_errno(status);
 292                 goto leave;
 293         }
 294
 295         /* get security xattr */
 296         status = ocfs2_init_security_get(inode, dir, &si);
 297         if (status) {
 298                 if (status == -EOPNOTSUPP)
 299                         si.enable = 0;
 300                 else {
 301                         mlog_errno(status);
 302                         goto leave;
 303                 }
 304         }
 305
 306         /* calculate meta data/clusters for setting security and acl xattr */
 307         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 308                                        &si, &want_clusters,
 309                                        &xattr_credits, &want_meta);
 310         if (status < 0) {
 311                 mlog_errno(status);
 312                 goto leave;
 313         }
 314
 315         /* Reserve a cluster if creating an extent based directory. */
 316         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 317                 want_clusters += 1;
 318
 319                 /* Dir indexing requires extra space as well */
 320                 if (ocfs2_supports_indexed_dirs(osb))
 321                         want_meta++;
 322         }
 323
 324         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 325         if (status < 0) {
 326                 if (status != -ENOSPC)
 327                         mlog_errno(status);
 328                 goto leave;
 329         }
 330
 331         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 332         if (status < 0) {
 333                 if (status != -ENOSPC)
 334                         mlog_errno(status);
 335                 goto leave;
 336         }
 337
 338         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 339                                                             S_ISDIR(mode),
 340                                                             xattr_credits));
 341         if (IS_ERR(handle)) {
 342                 status = PTR_ERR(handle);
 343                 handle = NULL;
 344                 mlog_errno(status);
 345                 goto leave;
 346         }
 347
 348         /* Starting to change things, restart is no longer possible. */
 349         ocfs2_block_signals(&oldset);
 350         did_block_signals = 1;
 351
 352         status = dquot_alloc_inode(inode);
 353         if (status)
 354                 goto leave;
 355         did_quota_inode = 1;
 356
 357         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
 358                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
 359                    dentry->d_name.name);
 360
 361         /* do the real work now. */
 362         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 363                                     &new_fe_bh, parent_fe_bh, handle,
 364                                     inode_ac);
 365         if (status < 0) {
 366                 mlog_errno(status);
 367                 goto leave;
 368         }
 369
 370         if (S_ISDIR(mode)) {
 371                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 372                                             new_fe_bh, data_ac, meta_ac);
 373                 if (status < 0) {
 374                         mlog_errno(status);
 375                         goto leave;
 376                 }
 377
 378                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 379                                                  parent_fe_bh,
 380                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 381                 if (status < 0) {
 382                         mlog_errno(status);
 383                         goto leave;
 384                 }
 385                 ocfs2_add_links_count(dirfe, 1);
 386                 ocfs2_journal_dirty(handle, parent_fe_bh);
 387                 inc_nlink(dir);
 388         }
 389
 390         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 391                                 meta_ac, data_ac);
 392         if (status < 0) {
 393                 mlog_errno(status);
 394                 goto leave;
 395         }
 396
 397         if (si.enable) {
 398                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 399                                                  meta_ac, data_ac);
 400                 if (status < 0) {
 401                         mlog_errno(status);
 402                         goto leave;
 403                 }
 404         }
 405
 406         /*
 407          * Do this before adding the entry to the directory. We add
 408          * also set d_op after success so that ->d_iput() will cleanup
 409          * the dentry lock even if ocfs2_add_entry() fails below.
 410          */
 411         status = ocfs2_dentry_attach_lock(dentry, inode,
 412                                           OCFS2_I(dir)->ip_blkno);
 413         if (status) {
 414                 mlog_errno(status);
 415                 goto leave;
 416         }
 417         dentry->d_op = &ocfs2_dentry_ops;
 418
 419         status = ocfs2_add_entry(handle, dentry, inode,
 420                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 421                                  &lookup);
 422         if (status < 0) {
 423                 mlog_errno(status);
 424                 goto leave;
 425         }
 426
 427         insert_inode_hash(inode);
 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         if (did_block_signals)
 438                 ocfs2_unblock_signals(&oldset);
 439
 440         if (status == -ENOSPC)
 441                 mlog(0, "Disk is full\n");
 442
 443         brelse(new_fe_bh);
 444         brelse(parent_fe_bh);
 445         kfree(si.name);
 446         kfree(si.value);
 447
 448         ocfs2_free_dir_lookup_result(&lookup);
 449
 450         if (inode_ac)
 451                 ocfs2_free_alloc_context(inode_ac);
 452
 453         if (data_ac)
 454                 ocfs2_free_alloc_context(data_ac);
 455
 456         if (meta_ac)
 457                 ocfs2_free_alloc_context(meta_ac);
 458
 459         /*
 460          * We should call iput after the i_mutex of the bitmap been
 461          * unlocked in ocfs2_free_alloc_context, or the
 462          * ocfs2_delete_inode will mutex_lock again.
 463          */
 464         if ((status < 0) && inode) {
 465                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 466                 clear_nlink(inode);
 467                 iput(inode);
 468         }
 469
 470         mlog_exit(status);
 471
 472         return status;
 473 }
 474
 475 static int __ocfs2_mknod_locked(struct inode *dir,
 476                                 struct inode *inode,
 477                                 dev_t dev,
 478                                 struct buffer_head **new_fe_bh,
 479                                 struct buffer_head *parent_fe_bh,
 480                                 handle_t *handle,
 481                                 struct ocfs2_alloc_context *inode_ac,
 482                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 483 {
 484         int status = 0;
 485         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 486         struct ocfs2_dinode *fe = NULL;
 487         struct ocfs2_extent_list *fel;
 488         u16 feat;
 489
 490         *new_fe_bh = NULL;
 491
 492         /* populate as many fields early on as possible - many of
 493          * these are used by the support functions here and in
 494          * callers. */
 495         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 496         OCFS2_I(inode)->ip_blkno = fe_blkno;
 497         spin_lock(&osb->osb_lock);
 498         inode->i_generation = osb->s_next_generation++;
 499         spin_unlock(&osb->osb_lock);
 500
 501         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 502         if (!*new_fe_bh) {
 503                 status = -EIO;
 504                 mlog_errno(status);
 505                 goto leave;
 506         }
 507         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 508
 509         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 510                                          *new_fe_bh,
 511                                          OCFS2_JOURNAL_ACCESS_CREATE);
 512         if (status < 0) {
 513                 mlog_errno(status);
 514                 goto leave;
 515         }
 516
 517         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 518         memset(fe, 0, osb->sb->s_blocksize);
 519
 520         fe->i_generation = cpu_to_le32(inode->i_generation);
 521         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 522         fe->i_blkno = cpu_to_le64(fe_blkno);
 523         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 524         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 525         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 526         fe->i_uid = cpu_to_le32(inode->i_uid);
 527         fe->i_gid = cpu_to_le32(inode->i_gid);
 528         fe->i_mode = cpu_to_le16(inode->i_mode);
 529         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 530                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 531
 532         ocfs2_set_links_count(fe, inode->i_nlink);
 533
 534         fe->i_last_eb_blk = 0;
 535         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 536         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 537         fe->i_atime = fe->i_ctime = fe->i_mtime =
 538                 cpu_to_le64(CURRENT_TIME.tv_sec);
 539         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 540                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 541         fe->i_dtime = 0;
 542
 543         /*
 544          * If supported, directories start with inline data. If inline
 545          * isn't supported, but indexing is, we start them as indexed.
 546          */
 547         feat = le16_to_cpu(fe->i_dyn_features);
 548         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 549                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 550
 551                 fe->id2.i_data.id_count = cpu_to_le16(
 552                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 553         } else {
 554                 fel = &fe->id2.i_list;
 555                 fel->l_tree_depth = 0;
 556                 fel->l_next_free_rec = 0;
 557                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 558         }
 559
 560         ocfs2_journal_dirty(handle, *new_fe_bh);
 561
 562         ocfs2_populate_inode(inode, fe, 1);
 563         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 564         if (!ocfs2_mount_local(osb)) {
 565                 status = ocfs2_create_new_inode_locks(inode);
 566                 if (status < 0)
 567                         mlog_errno(status);
 568         }
 569
 570         status = 0; /* error in ocfs2_create_new_inode_locks is not
 571                      * critical */
 572
 573 leave:
 574         if (status < 0) {
 575                 if (*new_fe_bh) {
 576                         brelse(*new_fe_bh);
 577                         *new_fe_bh = NULL;
 578                 }
 579         }
 580
 581         mlog_exit(status);
 582         return status;
 583 }
 584
 585 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 586                               struct inode *dir,
 587                               struct inode *inode,
 588                               dev_t dev,
 589                               struct buffer_head **new_fe_bh,
 590                               struct buffer_head *parent_fe_bh,
 591                               handle_t *handle,
 592                               struct ocfs2_alloc_context *inode_ac)
 593 {
 594         int status = 0;
 595         u64 suballoc_loc, fe_blkno = 0;
 596         u16 suballoc_bit;
 597
 598         *new_fe_bh = NULL;
 599
 600         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 601                                        inode_ac, &suballoc_loc,
 602                                        &suballoc_bit, &fe_blkno);
 603         if (status < 0) {
 604                 mlog_errno(status);
 605                 return status;
 606         }
 607
 608         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 609                                     parent_fe_bh, handle, inode_ac,
 610                                     fe_blkno, suballoc_loc, suballoc_bit);
 611 }
 612
 613 static int ocfs2_mkdir(struct inode *dir,
 614                        struct dentry *dentry,
 615                        int mode)
 616 {
 617         int ret;
 618
 619         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 620                    dentry->d_name.len, dentry->d_name.name);
 621         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 622         mlog_exit(ret);
 623
 624         return ret;
 625 }
 626
 627 static int ocfs2_create(struct inode *dir,
 628                         struct dentry *dentry,
 629                         int mode,
 630                         struct nameidata *nd)
 631 {
 632         int ret;
 633
 634         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
 635                    dentry->d_name.len, dentry->d_name.name);
 636         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 637         mlog_exit(ret);
 638
 639         return ret;
 640 }
 641
 642 static int ocfs2_link(struct dentry *old_dentry,
 643                       struct inode *dir,
 644                       struct dentry *dentry)
 645 {
 646         handle_t *handle;
 647         struct inode *inode = old_dentry->d_inode;
 648         int err;
 649         struct buffer_head *fe_bh = NULL;
 650         struct buffer_head *parent_fe_bh = NULL;
 651         struct ocfs2_dinode *fe = NULL;
 652         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 653         struct ocfs2_dir_lookup_result lookup = { NULL, };
 654         sigset_t oldset;
 655
 656         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
 657                    old_dentry->d_name.len, old_dentry->d_name.name,
 658                    dentry->d_name.len, dentry->d_name.name);
 659
 660         if (S_ISDIR(inode->i_mode))
 661                 return -EPERM;
 662
 663         dquot_initialize(dir);
 664
 665         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 666         if (err < 0) {
 667                 if (err != -ENOENT)
 668                         mlog_errno(err);
 669                 return err;
 670         }
 671
 672         if (!dir->i_nlink) {
 673                 err = -ENOENT;
 674                 goto out;
 675         }
 676
 677         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 678                                         dentry->d_name.len);
 679         if (err)
 680                 goto out;
 681
 682         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 683                                            dentry->d_name.name,
 684                                            dentry->d_name.len, &lookup);
 685         if (err < 0) {
 686                 mlog_errno(err);
 687                 goto out;
 688         }
 689
 690         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 691         if (err < 0) {
 692                 if (err != -ENOENT)
 693                         mlog_errno(err);
 694                 goto out;
 695         }
 696
 697         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 698         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 699                 err = -EMLINK;
 700                 goto out_unlock_inode;
 701         }
 702
 703         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 704         if (IS_ERR(handle)) {
 705                 err = PTR_ERR(handle);
 706                 handle = NULL;
 707                 mlog_errno(err);
 708                 goto out_unlock_inode;
 709         }
 710
 711         /* Starting to change things, restart is no longer possible. */
 712         ocfs2_block_signals(&oldset);
 713
 714         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 715                                       OCFS2_JOURNAL_ACCESS_WRITE);
 716         if (err < 0) {
 717                 mlog_errno(err);
 718                 goto out_commit;
 719         }
 720
 721         inc_nlink(inode);
 722         inode->i_ctime = CURRENT_TIME;
 723         ocfs2_set_links_count(fe, inode->i_nlink);
 724         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 725         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 726         ocfs2_journal_dirty(handle, fe_bh);
 727
 728         err = ocfs2_add_entry(handle, dentry, inode,
 729                               OCFS2_I(inode)->ip_blkno,
 730                               parent_fe_bh, &lookup);
 731         if (err) {
 732                 ocfs2_add_links_count(fe, -1);
 733                 drop_nlink(inode);
 734                 mlog_errno(err);
 735                 goto out_commit;
 736         }
 737
 738         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 739         if (err) {
 740                 mlog_errno(err);
 741                 goto out_commit;
 742         }
 743
 744         atomic_inc(&inode->i_count);
 745         dentry->d_op = &ocfs2_dentry_ops;
 746         d_instantiate(dentry, inode);
 747
 748 out_commit:
 749         ocfs2_commit_trans(osb, handle);
 750         ocfs2_unblock_signals(&oldset);
 751 out_unlock_inode:
 752         ocfs2_inode_unlock(inode, 1);
 753
 754 out:
 755         ocfs2_inode_unlock(dir, 1);
 756
 757         brelse(fe_bh);
 758         brelse(parent_fe_bh);
 759
 760         ocfs2_free_dir_lookup_result(&lookup);
 761
 762         mlog_exit(err);
 763
 764         return err;
 765 }
 766
 767 /*
 768  * Takes and drops an exclusive lock on the given dentry. This will
 769  * force other nodes to drop it.
 770  */
 771 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 772 {
 773         int ret;
 774
 775         ret = ocfs2_dentry_lock(dentry, 1);
 776         if (ret)
 777                 mlog_errno(ret);
 778         else
 779                 ocfs2_dentry_unlock(dentry, 1);
 780
 781         return ret;
 782 }
 783
 784 static inline int inode_is_unlinkable(struct inode *inode)
 785 {
 786         if (S_ISDIR(inode->i_mode)) {
 787                 if (inode->i_nlink == 2)
 788                         return 1;
 789                 return 0;
 790         }
 791
 792         if (inode->i_nlink == 1)
 793                 return 1;
 794         return 0;
 795 }
 796
 797 static int ocfs2_unlink(struct inode *dir,
 798                         struct dentry *dentry)
 799 {
 800         int status;
 801         int child_locked = 0;
 802         struct inode *inode = dentry->d_inode;
 803         struct inode *orphan_dir = NULL;
 804         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 805         u64 blkno;
 806         struct ocfs2_dinode *fe = NULL;
 807         struct buffer_head *fe_bh = NULL;
 808         struct buffer_head *parent_node_bh = NULL;
 809         handle_t *handle = NULL;
 810         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 811         struct ocfs2_dir_lookup_result lookup = { NULL, };
 812         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 813
 814         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
 815                    dentry->d_name.len, dentry->d_name.name);
 816
 817         dquot_initialize(dir);
 818
 819         BUG_ON(dentry->d_parent->d_inode != dir);
 820
 821         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
 822
 823         if (inode == osb->root_inode) {
 824                 mlog(0, "Cannot delete the root directory\n");
 825                 return -EPERM;
 826         }
 827
 828         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 829                                          OI_LS_PARENT);
 830         if (status < 0) {
 831                 if (status != -ENOENT)
 832                         mlog_errno(status);
 833                 return status;
 834         }
 835
 836         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 837                                           dentry->d_name.len, &blkno, dir,
 838                                           &lookup);
 839         if (status < 0) {
 840                 if (status != -ENOENT)
 841                         mlog_errno(status);
 842                 goto leave;
 843         }
 844
 845         if (OCFS2_I(inode)->ip_blkno != blkno) {
 846                 status = -ENOENT;
 847
 848                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
 849                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
 850                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
 851                 goto leave;
 852         }
 853
 854         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 855         if (status < 0) {
 856                 if (status != -ENOENT)
 857                         mlog_errno(status);
 858                 goto leave;
 859         }
 860         child_locked = 1;
 861
 862         if (S_ISDIR(inode->i_mode)) {
 863                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 864                         status = -ENOTEMPTY;
 865                         goto leave;
 866                 }
 867         }
 868
 869         status = ocfs2_remote_dentry_delete(dentry);
 870         if (status < 0) {
 871                 /* This remote delete should succeed under all normal
 872                  * circumstances. */
 873                 mlog_errno(status);
 874                 goto leave;
 875         }
 876
 877         if (inode_is_unlinkable(inode)) {
 878                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 879                                                   OCFS2_I(inode)->ip_blkno,
 880                                                   orphan_name, &orphan_insert);
 881                 if (status < 0) {
 882                         mlog_errno(status);
 883                         goto leave;
 884                 }
 885         }
 886
 887         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 888         if (IS_ERR(handle)) {
 889                 status = PTR_ERR(handle);
 890                 handle = NULL;
 891                 mlog_errno(status);
 892                 goto leave;
 893         }
 894
 895         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 896                                          OCFS2_JOURNAL_ACCESS_WRITE);
 897         if (status < 0) {
 898                 mlog_errno(status);
 899                 goto leave;
 900         }
 901
 902         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 903
 904         if (inode_is_unlinkable(inode)) {
 905                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 906                                           &orphan_insert, orphan_dir);
 907                 if (status < 0) {
 908                         mlog_errno(status);
 909                         goto leave;
 910                 }
 911         }
 912
 913         /* delete the name from the parent dir */
 914         status = ocfs2_delete_entry(handle, dir, &lookup);
 915         if (status < 0) {
 916                 mlog_errno(status);
 917                 goto leave;
 918         }
 919
 920         if (S_ISDIR(inode->i_mode))
 921                 drop_nlink(inode);
 922         drop_nlink(inode);
 923         ocfs2_set_links_count(fe, inode->i_nlink);
 924         ocfs2_journal_dirty(handle, fe_bh);
 925
 926         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 927         if (S_ISDIR(inode->i_mode))
 928                 drop_nlink(dir);
 929
 930         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 931         if (status < 0) {
 932                 mlog_errno(status);
 933                 if (S_ISDIR(inode->i_mode))
 934                         inc_nlink(dir);
 935         }
 936
 937 leave:
 938         if (handle)
 939                 ocfs2_commit_trans(osb, handle);
 940
 941         if (child_locked)
 942                 ocfs2_inode_unlock(inode, 1);
 943
 944         ocfs2_inode_unlock(dir, 1);
 945
 946         if (orphan_dir) {
 947                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 948                 ocfs2_inode_unlock(orphan_dir, 1);
 949                 mutex_unlock(&orphan_dir->i_mutex);
 950                 iput(orphan_dir);
 951         }
 952
 953         brelse(fe_bh);
 954         brelse(parent_node_bh);
 955
 956         ocfs2_free_dir_lookup_result(&orphan_insert);
 957         ocfs2_free_dir_lookup_result(&lookup);
 958
 959         mlog_exit(status);
 960
 961         return status;
 962 }
 963
 964 /*
 965  * The only place this should be used is rename!
 966  * if they have the same id, then the 1st one is the only one locked.
 967  */
 968 static int ocfs2_double_lock(struct ocfs2_super *osb,
 969                              struct buffer_head **bh1,
 970                              struct inode *inode1,
 971                              struct buffer_head **bh2,
 972                              struct inode *inode2)
 973 {
 974         int status;
 975         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 976         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 977         struct buffer_head **tmpbh;
 978         struct inode *tmpinode;
 979
 980         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
 981                    (unsigned long long)oi1->ip_blkno,
 982                    (unsigned long long)oi2->ip_blkno);
 983
 984         if (*bh1)
 985                 *bh1 = NULL;
 986         if (*bh2)
 987                 *bh2 = NULL;
 988
 989         /* we always want to lock the one with the lower lockid first. */
 990         if (oi1->ip_blkno != oi2->ip_blkno) {
 991                 if (oi1->ip_blkno < oi2->ip_blkno) {
 992                         /* switch id1 and id2 around */
 993                         mlog(0, "switching them around...\n");
 994                         tmpbh = bh2;
 995                         bh2 = bh1;
 996                         bh1 = tmpbh;
 997
 998                         tmpinode = inode2;
 999                         inode2 = inode1;
1000                         inode1 = tmpinode;
1001                 }
1002                 /* lock id2 */
1003                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1004                                                  OI_LS_RENAME1);
1005                 if (status < 0) {
1006                         if (status != -ENOENT)
1007                                 mlog_errno(status);
1008                         goto bail;
1009                 }
1010         }
1011
1012         /* lock id1 */
1013         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1014         if (status < 0) {
1015                 /*
1016                  * An error return must mean that no cluster locks
1017                  * were held on function exit.
1018                  */
1019                 if (oi1->ip_blkno != oi2->ip_blkno)
1020                         ocfs2_inode_unlock(inode2, 1);
1021
1022                 if (status != -ENOENT)
1023                         mlog_errno(status);
1024         }
1025
1026 bail:
1027         mlog_exit(status);
1028         return status;
1029 }
1030
1031 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1032 {
1033         ocfs2_inode_unlock(inode1, 1);
1034
1035         if (inode1 != inode2)
1036                 ocfs2_inode_unlock(inode2, 1);
1037 }
1038
1039 static int ocfs2_rename(struct inode *old_dir,
1040                         struct dentry *old_dentry,
1041                         struct inode *new_dir,
1042                         struct dentry *new_dentry)
1043 {
1044         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1045         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1046         struct inode *old_inode = old_dentry->d_inode;
1047         struct inode *new_inode = new_dentry->d_inode;
1048         struct inode *orphan_dir = NULL;
1049         struct ocfs2_dinode *newfe = NULL;
1050         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1051         struct buffer_head *newfe_bh = NULL;
1052         struct buffer_head *old_inode_bh = NULL;
1053         struct ocfs2_super *osb = NULL;
1054         u64 newfe_blkno, old_de_ino;
1055         handle_t *handle = NULL;
1056         struct buffer_head *old_dir_bh = NULL;
1057         struct buffer_head *new_dir_bh = NULL;
1058         nlink_t old_dir_nlink = old_dir->i_nlink;
1059         struct ocfs2_dinode *old_di;
1060         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1061         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1062         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1063         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1064         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1065
1066         /* At some point it might be nice to break this function up a
1067          * bit. */
1068
1069         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1070                    old_dir, old_dentry, new_dir, new_dentry,
1071                    old_dentry->d_name.len, old_dentry->d_name.name,
1072                    new_dentry->d_name.len, new_dentry->d_name.name);
1073
1074         dquot_initialize(old_dir);
1075         dquot_initialize(new_dir);
1076
1077         osb = OCFS2_SB(old_dir->i_sb);
1078
1079         if (new_inode) {
1080                 if (!igrab(new_inode))
1081                         BUG();
1082         }
1083
1084         /* Assume a directory hierarchy thusly:
1085          * a/b/c
1086          * a/d
1087          * a,b,c, and d are all directories.
1088          *
1089          * from cwd of 'a' on both nodes:
1090          * node1: mv b/c d
1091          * node2: mv d   b/c
1092          *
1093          * And that's why, just like the VFS, we need a file system
1094          * rename lock. */
1095         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1096                 status = ocfs2_rename_lock(osb);
1097                 if (status < 0) {
1098                         mlog_errno(status);
1099                         goto bail;
1100                 }
1101                 rename_lock = 1;
1102         }
1103
1104         /* if old and new are the same, this'll just do one lock. */
1105         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1106                                    &new_dir_bh, new_dir);
1107         if (status < 0) {
1108                 mlog_errno(status);
1109                 goto bail;
1110         }
1111         parents_locked = 1;
1112
1113         /* make sure both dirs have bhs
1114          * get an extra ref on old_dir_bh if old==new */
1115         if (!new_dir_bh) {
1116                 if (old_dir_bh) {
1117                         new_dir_bh = old_dir_bh;
1118                         get_bh(new_dir_bh);
1119                 } else {
1120                         mlog(ML_ERROR, "no old_dir_bh!\n");
1121                         status = -EIO;
1122                         goto bail;
1123                 }
1124         }
1125
1126         /*
1127          * Aside from allowing a meta data update, the locking here
1128          * also ensures that the downconvert thread on other nodes
1129          * won't have to concurrently downconvert the inode and the
1130          * dentry locks.
1131          */
1132         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1133                                          OI_LS_PARENT);
1134         if (status < 0) {
1135                 if (status != -ENOENT)
1136                         mlog_errno(status);
1137                 goto bail;
1138         }
1139         old_child_locked = 1;
1140
1141         status = ocfs2_remote_dentry_delete(old_dentry);
1142         if (status < 0) {
1143                 mlog_errno(status);
1144                 goto bail;
1145         }
1146
1147         if (S_ISDIR(old_inode->i_mode)) {
1148                 u64 old_inode_parent;
1149
1150                 update_dot_dot = 1;
1151                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1152                                                   old_inode,
1153                                                   &old_inode_dot_dot_res);
1154                 if (status) {
1155                         status = -EIO;
1156                         goto bail;
1157                 }
1158
1159                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1160                         status = -EIO;
1161                         goto bail;
1162                 }
1163
1164                 if (!new_inode && new_dir != old_dir &&
1165                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1166                         status = -EMLINK;
1167                         goto bail;
1168                 }
1169         }
1170
1171         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1172                                             old_dentry->d_name.len,
1173                                             &old_de_ino);
1174         if (status) {
1175                 status = -ENOENT;
1176                 goto bail;
1177         }
1178
1179         /*
1180          *  Check for inode number is _not_ due to possible IO errors.
1181          *  We might rmdir the source, keep it as pwd of some process
1182          *  and merrily kill the link to whatever was created under the
1183          *  same name. Goodbye sticky bit ;-<
1184          */
1185         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1186                 status = -ENOENT;
1187                 goto bail;
1188         }
1189
1190         /* check if the target already exists (in which case we need
1191          * to delete it */
1192         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1193                                           new_dentry->d_name.len,
1194                                           &newfe_blkno, new_dir,
1195                                           &target_lookup_res);
1196         /* The only error we allow here is -ENOENT because the new
1197          * file not existing is perfectly valid. */
1198         if ((status < 0) && (status != -ENOENT)) {
1199                 /* If we cannot find the file specified we should just */
1200                 /* return the error... */
1201                 mlog_errno(status);
1202                 goto bail;
1203         }
1204         if (status == 0)
1205                 target_exists = 1;
1206
1207         if (!target_exists && new_inode) {
1208                 /*
1209                  * Target was unlinked by another node while we were
1210                  * waiting to get to ocfs2_rename(). There isn't
1211                  * anything we can do here to help the situation, so
1212                  * bubble up the appropriate error.
1213                  */
1214                 status = -ENOENT;
1215                 goto bail;
1216         }
1217
1218         /* In case we need to overwrite an existing file, we blow it
1219          * away first */
1220         if (target_exists) {
1221                 /* VFS didn't think there existed an inode here, but
1222                  * someone else in the cluster must have raced our
1223                  * rename to create one. Today we error cleanly, in
1224                  * the future we should consider calling iget to build
1225                  * a new struct inode for this entry. */
1226                 if (!new_inode) {
1227                         status = -EACCES;
1228
1229                         mlog(0, "We found an inode for name %.*s but VFS "
1230                              "didn't give us one.\n", new_dentry->d_name.len,
1231                              new_dentry->d_name.name);
1232                         goto bail;
1233                 }
1234
1235                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1236                         status = -EACCES;
1237
1238                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1239                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1240                              (unsigned long long)newfe_blkno,
1241                              OCFS2_I(new_inode)->ip_flags);
1242                         goto bail;
1243                 }
1244
1245                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1246                 if (status < 0) {
1247                         if (status != -ENOENT)
1248                                 mlog_errno(status);
1249                         goto bail;
1250                 }
1251                 new_child_locked = 1;
1252
1253                 status = ocfs2_remote_dentry_delete(new_dentry);
1254                 if (status < 0) {
1255                         mlog_errno(status);
1256                         goto bail;
1257                 }
1258
1259                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1260
1261                 mlog(0, "aha rename over existing... new_blkno=%llu "
1262                      "newfebh=%p bhblocknr=%llu\n",
1263                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1264                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1265
1266                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1267                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1268                                                 OCFS2_I(new_inode)->ip_blkno,
1269                                                 orphan_name, &orphan_insert);
1270                         if (status < 0) {
1271                                 mlog_errno(status);
1272                                 goto bail;
1273                         }
1274                 }
1275         } else {
1276                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1277
1278                 status = ocfs2_check_dir_for_entry(new_dir,
1279                                                    new_dentry->d_name.name,
1280                                                    new_dentry->d_name.len);
1281                 if (status)
1282                         goto bail;
1283
1284                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1285                                                       new_dentry->d_name.name,
1286                                                       new_dentry->d_name.len,
1287                                                       &target_insert);
1288                 if (status < 0) {
1289                         mlog_errno(status);
1290                         goto bail;
1291                 }
1292         }
1293
1294         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1295         if (IS_ERR(handle)) {
1296                 status = PTR_ERR(handle);
1297                 handle = NULL;
1298                 mlog_errno(status);
1299                 goto bail;
1300         }
1301
1302         if (target_exists) {
1303                 if (S_ISDIR(new_inode->i_mode)) {
1304                         if (new_inode->i_nlink != 2 ||
1305                             !ocfs2_empty_dir(new_inode)) {
1306                                 status = -ENOTEMPTY;
1307                                 goto bail;
1308                         }
1309                 }
1310                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1311                                                  newfe_bh,
1312                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1313                 if (status < 0) {
1314                         mlog_errno(status);
1315                         goto bail;
1316                 }
1317
1318                 if (S_ISDIR(new_inode->i_mode) ||
1319                     (ocfs2_read_links_count(newfe) == 1)) {
1320                         status = ocfs2_orphan_add(osb, handle, new_inode,
1321                                                   newfe_bh, orphan_name,
1322                                                   &orphan_insert, orphan_dir);
1323                         if (status < 0) {
1324                                 mlog_errno(status);
1325                                 goto bail;
1326                         }
1327                 }
1328
1329                 /* change the dirent to point to the correct inode */
1330                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1331                                             old_inode);
1332                 if (status < 0) {
1333                         mlog_errno(status);
1334                         goto bail;
1335                 }
1336                 new_dir->i_version++;
1337
1338                 if (S_ISDIR(new_inode->i_mode))
1339                         ocfs2_set_links_count(newfe, 0);
1340                 else
1341                         ocfs2_add_links_count(newfe, -1);
1342                 ocfs2_journal_dirty(handle, newfe_bh);
1343         } else {
1344                 /* if the name was not found in new_dir, add it now */
1345                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1346                                          OCFS2_I(old_inode)->ip_blkno,
1347                                          new_dir_bh, &target_insert);
1348         }
1349
1350         old_inode->i_ctime = CURRENT_TIME;
1351         mark_inode_dirty(old_inode);
1352
1353         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1354                                          old_inode_bh,
1355                                          OCFS2_JOURNAL_ACCESS_WRITE);
1356         if (status >= 0) {
1357                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1358
1359                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1360                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1361                 ocfs2_journal_dirty(handle, old_inode_bh);
1362         } else
1363                 mlog_errno(status);
1364
1365         /*
1366          * Now that the name has been added to new_dir, remove the old name.
1367          *
1368          * We don't keep any directory entry context around until now
1369          * because the insert might have changed the type of directory
1370          * we're dealing with.
1371          */
1372         status = ocfs2_find_entry(old_dentry->d_name.name,
1373                                   old_dentry->d_name.len, old_dir,
1374                                   &old_entry_lookup);
1375         if (status)
1376                 goto bail;
1377
1378         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1379         if (status < 0) {
1380                 mlog_errno(status);
1381                 goto bail;
1382         }
1383
1384         if (new_inode) {
1385                 new_inode->i_nlink--;
1386                 new_inode->i_ctime = CURRENT_TIME;
1387         }
1388         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1389
1390         if (update_dot_dot) {
1391                 status = ocfs2_update_entry(old_inode, handle,
1392                                             &old_inode_dot_dot_res, new_dir);
1393                 old_dir->i_nlink--;
1394                 if (new_inode) {
1395                         new_inode->i_nlink--;
1396                 } else {
1397                         inc_nlink(new_dir);
1398                         mark_inode_dirty(new_dir);
1399                 }
1400         }
1401         mark_inode_dirty(old_dir);
1402         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1403         if (new_inode) {
1404                 mark_inode_dirty(new_inode);
1405                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1406         }
1407
1408         if (old_dir != new_dir) {
1409                 /* Keep the same times on both directories.*/
1410                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1411
1412                 /*
1413                  * This will also pick up the i_nlink change from the
1414                  * block above.
1415                  */
1416                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1417         }
1418
1419         if (old_dir_nlink != old_dir->i_nlink) {
1420                 if (!old_dir_bh) {
1421                         mlog(ML_ERROR, "need to change nlink for old dir "
1422                              "%llu from %d to %d but bh is NULL!\n",
1423                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1424                              (int)old_dir_nlink, old_dir->i_nlink);
1425                 } else {
1426                         struct ocfs2_dinode *fe;
1427                         status = ocfs2_journal_access_di(handle,
1428                                                          INODE_CACHE(old_dir),
1429                                                          old_dir_bh,
1430                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1431                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1432                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1433                         ocfs2_journal_dirty(handle, old_dir_bh);
1434                 }
1435         }
1436         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1437         status = 0;
1438 bail:
1439         if (rename_lock)
1440                 ocfs2_rename_unlock(osb);
1441
1442         if (handle)
1443                 ocfs2_commit_trans(osb, handle);
1444
1445         if (parents_locked)
1446                 ocfs2_double_unlock(old_dir, new_dir);
1447
1448         if (old_child_locked)
1449                 ocfs2_inode_unlock(old_inode, 1);
1450
1451         if (new_child_locked)
1452                 ocfs2_inode_unlock(new_inode, 1);
1453
1454         if (orphan_dir) {
1455                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1456                 ocfs2_inode_unlock(orphan_dir, 1);
1457                 mutex_unlock(&orphan_dir->i_mutex);
1458                 iput(orphan_dir);
1459         }
1460
1461         if (new_inode)
1462                 sync_mapping_buffers(old_inode->i_mapping);
1463
1464         if (new_inode)
1465                 iput(new_inode);
1466
1467         ocfs2_free_dir_lookup_result(&target_lookup_res);
1468         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1469         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1470         ocfs2_free_dir_lookup_result(&orphan_insert);
1471         ocfs2_free_dir_lookup_result(&target_insert);
1472
1473         brelse(newfe_bh);
1474         brelse(old_inode_bh);
1475         brelse(old_dir_bh);
1476         brelse(new_dir_bh);
1477
1478         mlog_exit(status);
1479
1480         return status;
1481 }
1482
1483 /*
1484  * we expect i_size = strlen(symname). Copy symname into the file
1485  * data, including the null terminator.
1486  */
1487 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1488                                      handle_t *handle,
1489                                      struct inode *inode,
1490                                      const char *symname)
1491 {
1492         struct buffer_head **bhs = NULL;
1493         const char *c;
1494         struct super_block *sb = osb->sb;
1495         u64 p_blkno, p_blocks;
1496         int virtual, blocks, status, i, bytes_left;
1497
1498         bytes_left = i_size_read(inode) + 1;
1499         /* we can't trust i_blocks because we're actually going to
1500          * write i_size + 1 bytes. */
1501         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1502
1503         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1504                         (unsigned long long)inode->i_blocks,
1505                         i_size_read(inode), blocks);
1506
1507         /* Sanity check -- make sure we're going to fit. */
1508         if (bytes_left >
1509             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1510                 status = -EIO;
1511                 mlog_errno(status);
1512                 goto bail;
1513         }
1514
1515         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1516         if (!bhs) {
1517                 status = -ENOMEM;
1518                 mlog_errno(status);
1519                 goto bail;
1520         }
1521
1522         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1523                                              NULL);
1524         if (status < 0) {
1525                 mlog_errno(status);
1526                 goto bail;
1527         }
1528
1529         /* links can never be larger than one cluster so we know this
1530          * is all going to be contiguous, but do a sanity check
1531          * anyway. */
1532         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1533                 status = -EIO;
1534                 mlog_errno(status);
1535                 goto bail;
1536         }
1537
1538         virtual = 0;
1539         while(bytes_left > 0) {
1540                 c = &symname[virtual * sb->s_blocksize];
1541
1542                 bhs[virtual] = sb_getblk(sb, p_blkno);
1543                 if (!bhs[virtual]) {
1544                         status = -ENOMEM;
1545                         mlog_errno(status);
1546                         goto bail;
1547                 }
1548                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1549                                               bhs[virtual]);
1550
1551                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1552                                               bhs[virtual],
1553                                               OCFS2_JOURNAL_ACCESS_CREATE);
1554                 if (status < 0) {
1555                         mlog_errno(status);
1556                         goto bail;
1557                 }
1558
1559                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1560
1561                 memcpy(bhs[virtual]->b_data, c,
1562                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1563                        bytes_left);
1564
1565                 ocfs2_journal_dirty(handle, bhs[virtual]);
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         sigset_t oldset;
1610         int did_block_signals = 0;
1611
1612         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1613                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1614
1615         dquot_initialize(dir);
1616
1617         sb = dir->i_sb;
1618         osb = OCFS2_SB(sb);
1619
1620         l = strlen(symname) + 1;
1621
1622         credits = ocfs2_calc_symlink_credits(sb);
1623
1624         /* lock the parent directory */
1625         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1626         if (status < 0) {
1627                 if (status != -ENOENT)
1628                         mlog_errno(status);
1629                 return status;
1630         }
1631
1632         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1633         if (!ocfs2_read_links_count(dirfe)) {
1634                 /* can't make a file in a deleted directory. */
1635                 status = -ENOENT;
1636                 goto bail;
1637         }
1638
1639         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1640                                            dentry->d_name.len);
1641         if (status)
1642                 goto bail;
1643
1644         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1645                                               dentry->d_name.name,
1646                                               dentry->d_name.len, &lookup);
1647         if (status < 0) {
1648                 mlog_errno(status);
1649                 goto bail;
1650         }
1651
1652         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1653         if (status < 0) {
1654                 if (status != -ENOSPC)
1655                         mlog_errno(status);
1656                 goto bail;
1657         }
1658
1659         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1660         if (!inode) {
1661                 status = -ENOMEM;
1662                 mlog_errno(status);
1663                 goto bail;
1664         }
1665
1666         /* get security xattr */
1667         status = ocfs2_init_security_get(inode, dir, &si);
1668         if (status) {
1669                 if (status == -EOPNOTSUPP)
1670                         si.enable = 0;
1671                 else {
1672                         mlog_errno(status);
1673                         goto bail;
1674                 }
1675         }
1676
1677         /* calculate meta data/clusters for setting security xattr */
1678         if (si.enable) {
1679                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1680                                                   &xattr_credits, &xattr_ac);
1681                 if (status < 0) {
1682                         mlog_errno(status);
1683                         goto bail;
1684                 }
1685         }
1686
1687         /* don't reserve bitmap space for fast symlinks. */
1688         if (l > ocfs2_fast_symlink_chars(sb))
1689                 want_clusters += 1;
1690
1691         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1692         if (status < 0) {
1693                 if (status != -ENOSPC)
1694                         mlog_errno(status);
1695                 goto bail;
1696         }
1697
1698         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1699         if (IS_ERR(handle)) {
1700                 status = PTR_ERR(handle);
1701                 handle = NULL;
1702                 mlog_errno(status);
1703                 goto bail;
1704         }
1705
1706         /* Starting to change things, restart is no longer possible. */
1707         ocfs2_block_signals(&oldset);
1708         did_block_signals = 1;
1709
1710         status = dquot_alloc_inode(inode);
1711         if (status)
1712                 goto bail;
1713         did_quota_inode = 1;
1714
1715         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1716                    inode->i_mode, dentry->d_name.len,
1717                    dentry->d_name.name);
1718
1719         status = ocfs2_mknod_locked(osb, dir, inode,
1720                                     0, &new_fe_bh, parent_fe_bh, handle,
1721                                     inode_ac);
1722         if (status < 0) {
1723                 mlog_errno(status);
1724                 goto bail;
1725         }
1726
1727         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1728         inode->i_rdev = 0;
1729         newsize = l - 1;
1730         if (l > ocfs2_fast_symlink_chars(sb)) {
1731                 u32 offset = 0;
1732
1733                 inode->i_op = &ocfs2_symlink_inode_operations;
1734                 status = dquot_alloc_space_nodirty(inode,
1735                     ocfs2_clusters_to_bytes(osb->sb, 1));
1736                 if (status)
1737                         goto bail;
1738                 did_quota = 1;
1739                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1740                                               new_fe_bh,
1741                                               handle, data_ac, NULL,
1742                                               NULL);
1743                 if (status < 0) {
1744                         if (status != -ENOSPC && status != -EINTR) {
1745                                 mlog(ML_ERROR,
1746                                      "Failed to extend file to %llu\n",
1747                                      (unsigned long long)newsize);
1748                                 mlog_errno(status);
1749                                 status = -ENOSPC;
1750                         }
1751                         goto bail;
1752                 }
1753                 i_size_write(inode, newsize);
1754                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1755         } else {
1756                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1757                 memcpy((char *) fe->id2.i_symlink, symname, l);
1758                 i_size_write(inode, newsize);
1759                 inode->i_blocks = 0;
1760         }
1761
1762         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1763         if (status < 0) {
1764                 mlog_errno(status);
1765                 goto bail;
1766         }
1767
1768         if (!ocfs2_inode_is_fast_symlink(inode)) {
1769                 status = ocfs2_create_symlink_data(osb, handle, inode,
1770                                                    symname);
1771                 if (status < 0) {
1772                         mlog_errno(status);
1773                         goto bail;
1774                 }
1775         }
1776
1777         if (si.enable) {
1778                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1779                                                  xattr_ac, data_ac);
1780                 if (status < 0) {
1781                         mlog_errno(status);
1782                         goto bail;
1783                 }
1784         }
1785
1786         /*
1787          * Do this before adding the entry to the directory. We add
1788          * also set d_op after success so that ->d_iput() will cleanup
1789          * the dentry lock even if ocfs2_add_entry() fails below.
1790          */
1791         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1792         if (status) {
1793                 mlog_errno(status);
1794                 goto bail;
1795         }
1796         dentry->d_op = &ocfs2_dentry_ops;
1797
1798         status = ocfs2_add_entry(handle, dentry, inode,
1799                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1800                                  &lookup);
1801         if (status < 0) {
1802                 mlog_errno(status);
1803                 goto bail;
1804         }
1805
1806         insert_inode_hash(inode);
1807         d_instantiate(dentry, inode);
1808 bail:
1809         if (status < 0 && did_quota)
1810                 dquot_free_space_nodirty(inode,
1811                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1812         if (status < 0 && did_quota_inode)
1813                 dquot_free_inode(inode);
1814         if (handle)
1815                 ocfs2_commit_trans(osb, handle);
1816
1817         ocfs2_inode_unlock(dir, 1);
1818         if (did_block_signals)
1819                 ocfs2_unblock_signals(&oldset);
1820
1821         brelse(new_fe_bh);
1822         brelse(parent_fe_bh);
1823         kfree(si.name);
1824         kfree(si.value);
1825         ocfs2_free_dir_lookup_result(&lookup);
1826         if (inode_ac)
1827                 ocfs2_free_alloc_context(inode_ac);
1828         if (data_ac)
1829                 ocfs2_free_alloc_context(data_ac);
1830         if (xattr_ac)
1831                 ocfs2_free_alloc_context(xattr_ac);
1832         if ((status < 0) && inode) {
1833                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1834                 clear_nlink(inode);
1835                 iput(inode);
1836         }
1837
1838         mlog_exit(status);
1839
1840         return status;
1841 }
1842
1843 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1844 {
1845         int status, namelen;
1846
1847         mlog_entry_void();
1848
1849         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1850                            (long long)blkno);
1851         if (namelen <= 0) {
1852                 if (namelen)
1853                         status = namelen;
1854                 else
1855                         status = -EINVAL;
1856                 mlog_errno(status);
1857                 goto bail;
1858         }
1859         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1860                 status = -EINVAL;
1861                 mlog_errno(status);
1862                 goto bail;
1863         }
1864
1865         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1866              namelen);
1867
1868         status = 0;
1869 bail:
1870         mlog_exit(status);
1871         return status;
1872 }
1873
1874 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1875                                         struct inode **ret_orphan_dir,
1876                                         struct buffer_head **ret_orphan_dir_bh)
1877 {
1878         struct inode *orphan_dir_inode;
1879         struct buffer_head *orphan_dir_bh = NULL;
1880         int ret = 0;
1881
1882         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1883                                                        ORPHAN_DIR_SYSTEM_INODE,
1884                                                        osb->slot_num);
1885         if (!orphan_dir_inode) {
1886                 ret = -ENOENT;
1887                 mlog_errno(ret);
1888                 return ret;
1889         }
1890
1891         mutex_lock(&orphan_dir_inode->i_mutex);
1892
1893         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1894         if (ret < 0) {
1895                 mutex_unlock(&orphan_dir_inode->i_mutex);
1896                 iput(orphan_dir_inode);
1897
1898                 mlog_errno(ret);
1899                 return ret;
1900         }
1901
1902         *ret_orphan_dir = orphan_dir_inode;
1903         *ret_orphan_dir_bh = orphan_dir_bh;
1904
1905         return 0;
1906 }
1907
1908 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1909                                       struct buffer_head *orphan_dir_bh,
1910                                       u64 blkno,
1911                                       char *name,
1912                                       struct ocfs2_dir_lookup_result *lookup)
1913 {
1914         int ret;
1915         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1916
1917         ret = ocfs2_blkno_stringify(blkno, name);
1918         if (ret < 0) {
1919                 mlog_errno(ret);
1920                 return ret;
1921         }
1922
1923         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1924                                            orphan_dir_bh, name,
1925                                            OCFS2_ORPHAN_NAMELEN, lookup);
1926         if (ret < 0) {
1927                 mlog_errno(ret);
1928                 return ret;
1929         }
1930
1931         return 0;
1932 }
1933
1934 /**
1935  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1936  * insertion of an orphan.
1937  * @osb: ocfs2 file system
1938  * @ret_orphan_dir: Orphan dir inode - returned locked!
1939  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1940  * @lookup: dir lookup result, to be passed back into functions like
1941  *          ocfs2_orphan_add
1942  *
1943  * Returns zero on success and the ret_orphan_dir, name and lookup
1944  * fields will be populated.
1945  *
1946  * Returns non-zero on failure.
1947  */
1948 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1949                                     struct inode **ret_orphan_dir,
1950                                     u64 blkno,
1951                                     char *name,
1952                                     struct ocfs2_dir_lookup_result *lookup)
1953 {
1954         struct inode *orphan_dir_inode = NULL;
1955         struct buffer_head *orphan_dir_bh = NULL;
1956         int ret = 0;
1957
1958         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1959                                            &orphan_dir_bh);
1960         if (ret < 0) {
1961                 mlog_errno(ret);
1962                 return ret;
1963         }
1964
1965         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1966                                          blkno, name, lookup);
1967         if (ret < 0) {
1968                 mlog_errno(ret);
1969                 goto out;
1970         }
1971
1972         *ret_orphan_dir = orphan_dir_inode;
1973
1974 out:
1975         brelse(orphan_dir_bh);
1976
1977         if (ret) {
1978                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1979                 mutex_unlock(&orphan_dir_inode->i_mutex);
1980                 iput(orphan_dir_inode);
1981         }
1982
1983         mlog_exit(ret);
1984         return ret;
1985 }
1986
1987 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1988                             handle_t *handle,
1989                             struct inode *inode,
1990                             struct buffer_head *fe_bh,
1991                             char *name,
1992                             struct ocfs2_dir_lookup_result *lookup,
1993                             struct inode *orphan_dir_inode)
1994 {
1995         struct buffer_head *orphan_dir_bh = NULL;
1996         int status = 0;
1997         struct ocfs2_dinode *orphan_fe;
1998         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1999
2000         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2001
2002         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2003         if (status < 0) {
2004                 mlog_errno(status);
2005                 goto leave;
2006         }
2007
2008         status = ocfs2_journal_access_di(handle,
2009                                          INODE_CACHE(orphan_dir_inode),
2010                                          orphan_dir_bh,
2011                                          OCFS2_JOURNAL_ACCESS_WRITE);
2012         if (status < 0) {
2013                 mlog_errno(status);
2014                 goto leave;
2015         }
2016
2017         /* we're a cluster, and nlink can change on disk from
2018          * underneath us... */
2019         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2020         if (S_ISDIR(inode->i_mode))
2021                 ocfs2_add_links_count(orphan_fe, 1);
2022         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2023         ocfs2_journal_dirty(handle, orphan_dir_bh);
2024
2025         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2026                                    OCFS2_ORPHAN_NAMELEN, inode,
2027                                    OCFS2_I(inode)->ip_blkno,
2028                                    orphan_dir_bh, lookup);
2029         if (status < 0) {
2030                 mlog_errno(status);
2031                 goto leave;
2032         }
2033
2034         /*
2035          * We're going to journal the change of i_flags and i_orphaned_slot.
2036          * It's safe anyway, though some callers may duplicate the journaling.
2037          * Journaling within the func just make the logic look more
2038          * straightforward.
2039          */
2040         status = ocfs2_journal_access_di(handle,
2041                                          INODE_CACHE(inode),
2042                                          fe_bh,
2043                                          OCFS2_JOURNAL_ACCESS_WRITE);
2044         if (status < 0) {
2045                 mlog_errno(status);
2046                 goto leave;
2047         }
2048
2049         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2050         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2051
2052         /* Record which orphan dir our inode now resides
2053          * in. delete_inode will use this to determine which orphan
2054          * dir to lock. */
2055         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2056
2057         ocfs2_journal_dirty(handle, fe_bh);
2058
2059         mlog(0, "Inode %llu orphaned in slot %d\n",
2060              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2061
2062 leave:
2063         brelse(orphan_dir_bh);
2064
2065         mlog_exit(status);
2066         return status;
2067 }
2068
2069 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2070 int ocfs2_orphan_del(struct ocfs2_super *osb,
2071                      handle_t *handle,
2072                      struct inode *orphan_dir_inode,
2073                      struct inode *inode,
2074                      struct buffer_head *orphan_dir_bh)
2075 {
2076         char name[OCFS2_ORPHAN_NAMELEN + 1];
2077         struct ocfs2_dinode *orphan_fe;
2078         int status = 0;
2079         struct ocfs2_dir_lookup_result lookup = { NULL, };
2080
2081         mlog_entry_void();
2082
2083         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2084         if (status < 0) {
2085                 mlog_errno(status);
2086                 goto leave;
2087         }
2088
2089         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2090              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2091              OCFS2_ORPHAN_NAMELEN);
2092
2093         /* find it's spot in the orphan directory */
2094         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2095                                   &lookup);
2096         if (status) {
2097                 mlog_errno(status);
2098                 goto leave;
2099         }
2100
2101         /* remove it from the orphan directory */
2102         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2103         if (status < 0) {
2104                 mlog_errno(status);
2105                 goto leave;
2106         }
2107
2108         status = ocfs2_journal_access_di(handle,
2109                                          INODE_CACHE(orphan_dir_inode),
2110                                          orphan_dir_bh,
2111                                          OCFS2_JOURNAL_ACCESS_WRITE);
2112         if (status < 0) {
2113                 mlog_errno(status);
2114                 goto leave;
2115         }
2116
2117         /* do the i_nlink dance! :) */
2118         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2119         if (S_ISDIR(inode->i_mode))
2120                 ocfs2_add_links_count(orphan_fe, -1);
2121         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2122         ocfs2_journal_dirty(handle, orphan_dir_bh);
2123
2124 leave:
2125         ocfs2_free_dir_lookup_result(&lookup);
2126
2127         mlog_exit(status);
2128         return status;
2129 }
2130
2131 /**
2132  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to recieve a newly
2133  * allocated file. This is different from the typical 'add to orphan dir'
2134  * operation in that the inode does not yet exist. This is a problem because
2135  * the orphan dir stringifies the inode block number to come up with it's
2136  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2137  * problem. This function works around it by calling deeper into the orphan
2138  * and suballoc code than other callers. Use this only by necessity.
2139  * @dir: The directory which this inode will ultimately wind up under - not the
2140  * orphan dir!
2141  * @dir_bh: buffer_head the @dir inode block
2142  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2143  * with the string to be used for orphan dirent. Pass back to the orphan dir
2144  * code.
2145  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2146  * dir code.
2147  * @ret_di_blkno: block number where the new inode will be allocated.
2148  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2149  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2150  *
2151  * Returns zero on success and the ret_orphan_dir, name and lookup
2152  * fields will be populated.
2153  *
2154  * Returns non-zero on failure.
2155  */
2156 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2157                                         struct buffer_head *dir_bh,
2158                                         char *orphan_name,
2159                                         struct inode **ret_orphan_dir,
2160                                         u64 *ret_di_blkno,
2161                                         struct ocfs2_dir_lookup_result *orphan_insert,
2162                                         struct ocfs2_alloc_context **ret_inode_ac)
2163 {
2164         int ret;
2165         u64 di_blkno;
2166         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2167         struct inode *orphan_dir = NULL;
2168         struct buffer_head *orphan_dir_bh = NULL;
2169         struct ocfs2_alloc_context *inode_ac = NULL;
2170
2171         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2172         if (ret < 0) {
2173                 mlog_errno(ret);
2174                 return ret;
2175         }
2176
2177         /* reserve an inode spot */
2178         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2179         if (ret < 0) {
2180                 if (ret != -ENOSPC)
2181                         mlog_errno(ret);
2182                 goto out;
2183         }
2184
2185         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2186                                        &di_blkno);
2187         if (ret) {
2188                 mlog_errno(ret);
2189                 goto out;
2190         }
2191
2192         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2193                                          di_blkno, orphan_name, orphan_insert);
2194         if (ret < 0) {
2195                 mlog_errno(ret);
2196                 goto out;
2197         }
2198
2199 out:
2200         if (ret == 0) {
2201                 *ret_orphan_dir = orphan_dir;
2202                 *ret_di_blkno = di_blkno;
2203                 *ret_inode_ac = inode_ac;
2204                 /*
2205                  * orphan_name and orphan_insert are already up to
2206                  * date via prepare_orphan_dir
2207                  */
2208         } else {
2209                 /* Unroll reserve_new_inode* */
2210                 if (inode_ac)
2211                         ocfs2_free_alloc_context(inode_ac);
2212
2213                 /* Unroll orphan dir locking */
2214                 mutex_unlock(&orphan_dir->i_mutex);
2215                 ocfs2_inode_unlock(orphan_dir, 1);
2216                 iput(orphan_dir);
2217         }
2218
2219         brelse(orphan_dir_bh);
2220
2221         return 0;
2222 }
2223
2224 int ocfs2_create_inode_in_orphan(struct inode *dir,
2225                                  int mode,
2226                                  struct inode **new_inode)
2227 {
2228         int status, did_quota_inode = 0;
2229         struct inode *inode = NULL;
2230         struct inode *orphan_dir = NULL;
2231         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2232         struct ocfs2_dinode *di = NULL;
2233         handle_t *handle = NULL;
2234         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2235         struct buffer_head *parent_di_bh = NULL;
2236         struct buffer_head *new_di_bh = NULL;
2237         struct ocfs2_alloc_context *inode_ac = NULL;
2238         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2239         u64 uninitialized_var(di_blkno), suballoc_loc;
2240         u16 suballoc_bit;
2241
2242         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2243         if (status < 0) {
2244                 if (status != -ENOENT)
2245                         mlog_errno(status);
2246                 return status;
2247         }
2248
2249         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2250                                               orphan_name, &orphan_dir,
2251                                               &di_blkno, &orphan_insert, &inode_ac);
2252         if (status < 0) {
2253                 if (status != -ENOSPC)
2254                         mlog_errno(status);
2255                 goto leave;
2256         }
2257
2258         inode = ocfs2_get_init_inode(dir, mode);
2259         if (!inode) {
2260                 status = -ENOMEM;
2261                 mlog_errno(status);
2262                 goto leave;
2263         }
2264
2265         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2266         if (IS_ERR(handle)) {
2267                 status = PTR_ERR(handle);
2268                 handle = NULL;
2269                 mlog_errno(status);
2270                 goto leave;
2271         }
2272
2273         status = dquot_alloc_inode(inode);
2274         if (status)
2275                 goto leave;
2276         did_quota_inode = 1;
2277
2278         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2279                                               &suballoc_loc,
2280                                               &suballoc_bit, di_blkno);
2281         if (status < 0) {
2282                 mlog_errno(status);
2283                 goto leave;
2284         }
2285
2286         inode->i_nlink = 0;
2287         /* do the real work now. */
2288         status = __ocfs2_mknod_locked(dir, inode,
2289                                       0, &new_di_bh, parent_di_bh, handle,
2290                                       inode_ac, di_blkno, suballoc_loc,
2291                                       suballoc_bit);
2292         if (status < 0) {
2293                 mlog_errno(status);
2294                 goto leave;
2295         }
2296
2297         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2298         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2299                                   &orphan_insert, orphan_dir);
2300         if (status < 0) {
2301                 mlog_errno(status);
2302                 goto leave;
2303         }
2304
2305         /* get open lock so that only nodes can't remove it from orphan dir. */
2306         status = ocfs2_open_lock(inode);
2307         if (status < 0)
2308                 mlog_errno(status);
2309
2310         insert_inode_hash(inode);
2311 leave:
2312         if (status < 0 && did_quota_inode)
2313                 dquot_free_inode(inode);
2314         if (handle)
2315                 ocfs2_commit_trans(osb, handle);
2316
2317         if (orphan_dir) {
2318                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2319                 ocfs2_inode_unlock(orphan_dir, 1);
2320                 mutex_unlock(&orphan_dir->i_mutex);
2321                 iput(orphan_dir);
2322         }
2323
2324         if (status == -ENOSPC)
2325                 mlog(0, "Disk is full\n");
2326
2327         if ((status < 0) && inode) {
2328                 clear_nlink(inode);
2329                 iput(inode);
2330         }
2331
2332         if (inode_ac)
2333                 ocfs2_free_alloc_context(inode_ac);
2334
2335         brelse(new_di_bh);
2336
2337         if (!status)
2338                 *new_inode = inode;
2339
2340         ocfs2_free_dir_lookup_result(&orphan_insert);
2341
2342         ocfs2_inode_unlock(dir, 1);
2343         brelse(parent_di_bh);
2344         return status;
2345 }
2346
2347 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2348                                    struct inode *inode,
2349                                    struct dentry *dentry)
2350 {
2351         int status = 0;
2352         struct buffer_head *parent_di_bh = NULL;
2353         handle_t *handle = NULL;
2354         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2355         struct ocfs2_dinode *dir_di, *di;
2356         struct inode *orphan_dir_inode = NULL;
2357         struct buffer_head *orphan_dir_bh = NULL;
2358         struct buffer_head *di_bh = NULL;
2359         struct ocfs2_dir_lookup_result lookup = { NULL, };
2360
2361         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2362                    dentry->d_name.len, dentry->d_name.name);
2363
2364         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2365         if (status < 0) {
2366                 if (status != -ENOENT)
2367                         mlog_errno(status);
2368                 return status;
2369         }
2370
2371         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2372         if (!dir_di->i_links_count) {
2373                 /* can't make a file in a deleted directory. */
2374                 status = -ENOENT;
2375                 goto leave;
2376         }
2377
2378         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2379                                            dentry->d_name.len);
2380         if (status)
2381                 goto leave;
2382
2383         /* get a spot inside the dir. */
2384         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2385                                               dentry->d_name.name,
2386                                               dentry->d_name.len, &lookup);
2387         if (status < 0) {
2388                 mlog_errno(status);
2389                 goto leave;
2390         }
2391
2392         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2393                                                        ORPHAN_DIR_SYSTEM_INODE,
2394                                                        osb->slot_num);
2395         if (!orphan_dir_inode) {
2396                 status = -EEXIST;
2397                 mlog_errno(status);
2398                 goto leave;
2399         }
2400
2401         mutex_lock(&orphan_dir_inode->i_mutex);
2402
2403         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2404         if (status < 0) {
2405                 mlog_errno(status);
2406                 mutex_unlock(&orphan_dir_inode->i_mutex);
2407                 iput(orphan_dir_inode);
2408                 goto leave;
2409         }
2410
2411         status = ocfs2_read_inode_block(inode, &di_bh);
2412         if (status < 0) {
2413                 mlog_errno(status);
2414                 goto orphan_unlock;
2415         }
2416
2417         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2418         if (IS_ERR(handle)) {
2419                 status = PTR_ERR(handle);
2420                 handle = NULL;
2421                 mlog_errno(status);
2422                 goto orphan_unlock;
2423         }
2424
2425         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2426                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2427         if (status < 0) {
2428                 mlog_errno(status);
2429                 goto out_commit;
2430         }
2431
2432         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2433                                   orphan_dir_bh);
2434         if (status < 0) {
2435                 mlog_errno(status);
2436                 goto out_commit;
2437         }
2438
2439         di = (struct ocfs2_dinode *)di_bh->b_data;
2440         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2441         di->i_orphaned_slot = 0;
2442         inode->i_nlink = 1;
2443         ocfs2_set_links_count(di, inode->i_nlink);
2444         ocfs2_journal_dirty(handle, di_bh);
2445
2446         status = ocfs2_add_entry(handle, dentry, inode,
2447                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2448                                  &lookup);
2449         if (status < 0) {
2450                 mlog_errno(status);
2451                 goto out_commit;
2452         }
2453
2454         status = ocfs2_dentry_attach_lock(dentry, inode,
2455                                           OCFS2_I(dir)->ip_blkno);
2456         if (status) {
2457                 mlog_errno(status);
2458                 goto out_commit;
2459         }
2460
2461         dentry->d_op = &ocfs2_dentry_ops;
2462         d_instantiate(dentry, inode);
2463         status = 0;
2464 out_commit:
2465         ocfs2_commit_trans(osb, handle);
2466 orphan_unlock:
2467         ocfs2_inode_unlock(orphan_dir_inode, 1);
2468         mutex_unlock(&orphan_dir_inode->i_mutex);
2469         iput(orphan_dir_inode);
2470 leave:
2471
2472         ocfs2_inode_unlock(dir, 1);
2473
2474         brelse(di_bh);
2475         brelse(parent_di_bh);
2476         brelse(orphan_dir_bh);
2477
2478         ocfs2_free_dir_lookup_result(&lookup);
2479
2480         mlog_exit(status);
2481
2482         return status;
2483 }
2484
2485 const struct inode_operations ocfs2_dir_iops = {
2486         .create         = ocfs2_create,
2487         .lookup         = ocfs2_lookup,
2488         .link           = ocfs2_link,
2489         .unlink         = ocfs2_unlink,
2490         .rmdir          = ocfs2_unlink,
2491         .symlink        = ocfs2_symlink,
2492         .mkdir          = ocfs2_mkdir,
2493         .mknod          = ocfs2_mknod,
2494         .rename         = ocfs2_rename,
2495         .setattr        = ocfs2_setattr,
2496         .getattr        = ocfs2_getattr,
2497         .permission     = ocfs2_permission,
2498         .setxattr       = generic_setxattr,
2499         .getxattr       = generic_getxattr,
2500         .listxattr      = ocfs2_listxattr,
2501         .removexattr    = generic_removexattr,
2502         .fiemap         = ocfs2_fiemap,
2503 };