Pileus Git - ~andy/linux/blob - 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 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "xattr.h"
  64 #include "acl.h"
  65 #include "ocfs2_trace.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                                    unsigned int flags)
 102 {
 103         int status;
 104         u64 blkno;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_inode_info *oi;
 108
 109         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 110                            dentry->d_name.name,
 111                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 112
 113         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 114                 ret = ERR_PTR(-ENAMETOOLONG);
 115                 goto bail;
 116         }
 117
 118         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 119         if (status < 0) {
 120                 if (status != -ENOENT)
 121                         mlog_errno(status);
 122                 ret = ERR_PTR(status);
 123                 goto bail;
 124         }
 125
 126         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 127                                             dentry->d_name.len, &blkno);
 128         if (status < 0)
 129                 goto bail_add;
 130
 131         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 132         if (IS_ERR(inode)) {
 133                 ret = ERR_PTR(-EACCES);
 134                 goto bail_unlock;
 135         }
 136
 137         oi = OCFS2_I(inode);
 138         /* Clear any orphaned state... If we were able to look up the
 139          * inode from a directory, it certainly can't be orphaned. We
 140          * might have the bad state from a node which intended to
 141          * orphan this inode but crashed before it could commit the
 142          * unlink. */
 143         spin_lock(&oi->ip_lock);
 144         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 145         spin_unlock(&oi->ip_lock);
 146
 147 bail_add:
 148         ret = d_splice_alias(inode, dentry);
 149
 150         if (inode) {
 151                 /*
 152                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 153                  * dentry, it will d_move() it on top of ourse. The
 154                  * return value will indicate this however, so in
 155                  * those cases, we switch them around for the locking
 156                  * code.
 157                  *
 158                  * NOTE: This dentry already has ->d_op set from
 159                  * ocfs2_get_parent() and ocfs2_get_dentry()
 160                  */
 161                 if (ret)
 162                         dentry = ret;
 163
 164                 status = ocfs2_dentry_attach_lock(dentry, inode,
 165                                                   OCFS2_I(dir)->ip_blkno);
 166                 if (status) {
 167                         mlog_errno(status);
 168                         ret = ERR_PTR(status);
 169                         goto bail_unlock;
 170                 }
 171         } else
 172                 ocfs2_dentry_attach_gen(dentry);
 173
 174 bail_unlock:
 175         /* Don't drop the cluster lock until *after* the d_add --
 176          * unlink on another node will message us to remove that
 177          * dentry under this lock so otherwise we can race this with
 178          * the downconvert thread and have a stale dentry. */
 179         ocfs2_inode_unlock(dir, 0);
 180
 181 bail:
 182
 183         trace_ocfs2_lookup_ret(ret);
 184
 185         return ret;
 186 }
 187
 188 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 189 {
 190         struct inode *inode;
 191
 192         inode = new_inode(dir->i_sb);
 193         if (!inode) {
 194                 mlog(ML_ERROR, "new_inode failed!\n");
 195                 return NULL;
 196         }
 197
 198         /* populate as many fields early on as possible - many of
 199          * these are used by the support functions here and in
 200          * callers. */
 201         if (S_ISDIR(mode))
 202                 set_nlink(inode, 2);
 203         inode_init_owner(inode, dir, mode);
 204         dquot_initialize(inode);
 205         return inode;
 206 }
 207
 208 static int ocfs2_mknod(struct inode *dir,
 209                        struct dentry *dentry,
 210                        umode_t mode,
 211                        dev_t dev)
 212 {
 213         int status = 0;
 214         struct buffer_head *parent_fe_bh = NULL;
 215         handle_t *handle = NULL;
 216         struct ocfs2_super *osb;
 217         struct ocfs2_dinode *dirfe;
 218         struct buffer_head *new_fe_bh = NULL;
 219         struct inode *inode = NULL;
 220         struct ocfs2_alloc_context *inode_ac = NULL;
 221         struct ocfs2_alloc_context *data_ac = NULL;
 222         struct ocfs2_alloc_context *meta_ac = NULL;
 223         int want_clusters = 0;
 224         int want_meta = 0;
 225         int xattr_credits = 0;
 226         struct ocfs2_security_xattr_info si = {
 227                 .enable = 1,
 228         };
 229         int did_quota_inode = 0;
 230         struct ocfs2_dir_lookup_result lookup = { NULL, };
 231         sigset_t oldset;
 232         int did_block_signals = 0;
 233
 234         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 235                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 236                           (unsigned long)dev, mode);
 237
 238         dquot_initialize(dir);
 239
 240         /* get our super block */
 241         osb = OCFS2_SB(dir->i_sb);
 242
 243         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 244         if (status < 0) {
 245                 if (status != -ENOENT)
 246                         mlog_errno(status);
 247                 return status;
 248         }
 249
 250         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 251                 status = -EMLINK;
 252                 goto leave;
 253         }
 254
 255         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 256         if (!ocfs2_read_links_count(dirfe)) {
 257                 /* can't make a file in a deleted directory. */
 258                 status = -ENOENT;
 259                 goto leave;
 260         }
 261
 262         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 263                                            dentry->d_name.len);
 264         if (status)
 265                 goto leave;
 266
 267         /* get a spot inside the dir. */
 268         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 269                                               dentry->d_name.name,
 270                                               dentry->d_name.len, &lookup);
 271         if (status < 0) {
 272                 mlog_errno(status);
 273                 goto leave;
 274         }
 275
 276         /* reserve an inode spot */
 277         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 278         if (status < 0) {
 279                 if (status != -ENOSPC)
 280                         mlog_errno(status);
 281                 goto leave;
 282         }
 283
 284         inode = ocfs2_get_init_inode(dir, mode);
 285         if (!inode) {
 286                 status = -ENOMEM;
 287                 mlog_errno(status);
 288                 goto leave;
 289         }
 290
 291         /* get security xattr */
 292         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 293         if (status) {
 294                 if (status == -EOPNOTSUPP)
 295                         si.enable = 0;
 296                 else {
 297                         mlog_errno(status);
 298                         goto leave;
 299                 }
 300         }
 301
 302         /* calculate meta data/clusters for setting security and acl xattr */
 303         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 304                                        &si, &want_clusters,
 305                                        &xattr_credits, &want_meta);
 306         if (status < 0) {
 307                 mlog_errno(status);
 308                 goto leave;
 309         }
 310
 311         /* Reserve a cluster if creating an extent based directory. */
 312         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 313                 want_clusters += 1;
 314
 315                 /* Dir indexing requires extra space as well */
 316                 if (ocfs2_supports_indexed_dirs(osb))
 317                         want_meta++;
 318         }
 319
 320         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 321         if (status < 0) {
 322                 if (status != -ENOSPC)
 323                         mlog_errno(status);
 324                 goto leave;
 325         }
 326
 327         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 328         if (status < 0) {
 329                 if (status != -ENOSPC)
 330                         mlog_errno(status);
 331                 goto leave;
 332         }
 333
 334         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 335                                                             S_ISDIR(mode),
 336                                                             xattr_credits));
 337         if (IS_ERR(handle)) {
 338                 status = PTR_ERR(handle);
 339                 handle = NULL;
 340                 mlog_errno(status);
 341                 goto leave;
 342         }
 343
 344         /* Starting to change things, restart is no longer possible. */
 345         ocfs2_block_signals(&oldset);
 346         did_block_signals = 1;
 347
 348         status = dquot_alloc_inode(inode);
 349         if (status)
 350                 goto leave;
 351         did_quota_inode = 1;
 352
 353         /* do the real work now. */
 354         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 355                                     &new_fe_bh, parent_fe_bh, handle,
 356                                     inode_ac);
 357         if (status < 0) {
 358                 mlog_errno(status);
 359                 goto leave;
 360         }
 361
 362         if (S_ISDIR(mode)) {
 363                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 364                                             new_fe_bh, data_ac, meta_ac);
 365                 if (status < 0) {
 366                         mlog_errno(status);
 367                         goto leave;
 368                 }
 369
 370                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 371                                                  parent_fe_bh,
 372                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 373                 if (status < 0) {
 374                         mlog_errno(status);
 375                         goto leave;
 376                 }
 377                 ocfs2_add_links_count(dirfe, 1);
 378                 ocfs2_journal_dirty(handle, parent_fe_bh);
 379                 inc_nlink(dir);
 380         }
 381
 382         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 383                                 meta_ac, data_ac);
 384         if (status < 0) {
 385                 mlog_errno(status);
 386                 goto leave;
 387         }
 388
 389         if (si.enable) {
 390                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 391                                                  meta_ac, data_ac);
 392                 if (status < 0) {
 393                         mlog_errno(status);
 394                         goto leave;
 395                 }
 396         }
 397
 398         /*
 399          * Do this before adding the entry to the directory. We add
 400          * also set d_op after success so that ->d_iput() will cleanup
 401          * the dentry lock even if ocfs2_add_entry() fails below.
 402          */
 403         status = ocfs2_dentry_attach_lock(dentry, inode,
 404                                           OCFS2_I(dir)->ip_blkno);
 405         if (status) {
 406                 mlog_errno(status);
 407                 goto leave;
 408         }
 409
 410         status = ocfs2_add_entry(handle, dentry, inode,
 411                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 412                                  &lookup);
 413         if (status < 0) {
 414                 mlog_errno(status);
 415                 goto leave;
 416         }
 417
 418         insert_inode_hash(inode);
 419         d_instantiate(dentry, inode);
 420         status = 0;
 421 leave:
 422         if (status < 0 && did_quota_inode)
 423                 dquot_free_inode(inode);
 424         if (handle)
 425                 ocfs2_commit_trans(osb, handle);
 426
 427         ocfs2_inode_unlock(dir, 1);
 428         if (did_block_signals)
 429                 ocfs2_unblock_signals(&oldset);
 430
 431         brelse(new_fe_bh);
 432         brelse(parent_fe_bh);
 433         kfree(si.name);
 434         kfree(si.value);
 435
 436         ocfs2_free_dir_lookup_result(&lookup);
 437
 438         if (inode_ac)
 439                 ocfs2_free_alloc_context(inode_ac);
 440
 441         if (data_ac)
 442                 ocfs2_free_alloc_context(data_ac);
 443
 444         if (meta_ac)
 445                 ocfs2_free_alloc_context(meta_ac);
 446
 447         /*
 448          * We should call iput after the i_mutex of the bitmap been
 449          * unlocked in ocfs2_free_alloc_context, or the
 450          * ocfs2_delete_inode will mutex_lock again.
 451          */
 452         if ((status < 0) && inode) {
 453                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 454                 clear_nlink(inode);
 455                 iput(inode);
 456         }
 457
 458         if (status)
 459                 mlog_errno(status);
 460
 461         return status;
 462 }
 463
 464 static int __ocfs2_mknod_locked(struct inode *dir,
 465                                 struct inode *inode,
 466                                 dev_t dev,
 467                                 struct buffer_head **new_fe_bh,
 468                                 struct buffer_head *parent_fe_bh,
 469                                 handle_t *handle,
 470                                 struct ocfs2_alloc_context *inode_ac,
 471                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 472 {
 473         int status = 0;
 474         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 475         struct ocfs2_dinode *fe = NULL;
 476         struct ocfs2_extent_list *fel;
 477         u16 feat;
 478
 479         *new_fe_bh = NULL;
 480
 481         /* populate as many fields early on as possible - many of
 482          * these are used by the support functions here and in
 483          * callers. */
 484         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 485         OCFS2_I(inode)->ip_blkno = fe_blkno;
 486         spin_lock(&osb->osb_lock);
 487         inode->i_generation = osb->s_next_generation++;
 488         spin_unlock(&osb->osb_lock);
 489
 490         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 491         if (!*new_fe_bh) {
 492                 status = -ENOMEM;
 493                 mlog_errno(status);
 494                 goto leave;
 495         }
 496         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 497
 498         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 499                                          *new_fe_bh,
 500                                          OCFS2_JOURNAL_ACCESS_CREATE);
 501         if (status < 0) {
 502                 mlog_errno(status);
 503                 goto leave;
 504         }
 505
 506         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 507         memset(fe, 0, osb->sb->s_blocksize);
 508
 509         fe->i_generation = cpu_to_le32(inode->i_generation);
 510         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 511         fe->i_blkno = cpu_to_le64(fe_blkno);
 512         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 513         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 514         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 515         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 516         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 517         fe->i_mode = cpu_to_le16(inode->i_mode);
 518         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 519                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 520
 521         ocfs2_set_links_count(fe, inode->i_nlink);
 522
 523         fe->i_last_eb_blk = 0;
 524         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 525         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 526         fe->i_atime = fe->i_ctime = fe->i_mtime =
 527                 cpu_to_le64(CURRENT_TIME.tv_sec);
 528         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 529                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 530         fe->i_dtime = 0;
 531
 532         /*
 533          * If supported, directories start with inline data. If inline
 534          * isn't supported, but indexing is, we start them as indexed.
 535          */
 536         feat = le16_to_cpu(fe->i_dyn_features);
 537         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 538                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 539
 540                 fe->id2.i_data.id_count = cpu_to_le16(
 541                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 542         } else {
 543                 fel = &fe->id2.i_list;
 544                 fel->l_tree_depth = 0;
 545                 fel->l_next_free_rec = 0;
 546                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 547         }
 548
 549         ocfs2_journal_dirty(handle, *new_fe_bh);
 550
 551         ocfs2_populate_inode(inode, fe, 1);
 552         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 553         if (!ocfs2_mount_local(osb)) {
 554                 status = ocfs2_create_new_inode_locks(inode);
 555                 if (status < 0)
 556                         mlog_errno(status);
 557         }
 558
 559         status = 0; /* error in ocfs2_create_new_inode_locks is not
 560                      * critical */
 561
 562 leave:
 563         if (status < 0) {
 564                 if (*new_fe_bh) {
 565                         brelse(*new_fe_bh);
 566                         *new_fe_bh = NULL;
 567                 }
 568         }
 569
 570         if (status)
 571                 mlog_errno(status);
 572         return status;
 573 }
 574
 575 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 576                               struct inode *dir,
 577                               struct inode *inode,
 578                               dev_t dev,
 579                               struct buffer_head **new_fe_bh,
 580                               struct buffer_head *parent_fe_bh,
 581                               handle_t *handle,
 582                               struct ocfs2_alloc_context *inode_ac)
 583 {
 584         int status = 0;
 585         u64 suballoc_loc, fe_blkno = 0;
 586         u16 suballoc_bit;
 587
 588         *new_fe_bh = NULL;
 589
 590         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 591                                        inode_ac, &suballoc_loc,
 592                                        &suballoc_bit, &fe_blkno);
 593         if (status < 0) {
 594                 mlog_errno(status);
 595                 return status;
 596         }
 597
 598         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 599                                     parent_fe_bh, handle, inode_ac,
 600                                     fe_blkno, suballoc_loc, suballoc_bit);
 601 }
 602
 603 static int ocfs2_mkdir(struct inode *dir,
 604                        struct dentry *dentry,
 605                        umode_t mode)
 606 {
 607         int ret;
 608
 609         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 610                           OCFS2_I(dir)->ip_blkno, mode);
 611         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 612         if (ret)
 613                 mlog_errno(ret);
 614
 615         return ret;
 616 }
 617
 618 static int ocfs2_create(struct inode *dir,
 619                         struct dentry *dentry,
 620                         umode_t mode,
 621                         bool excl)
 622 {
 623         int ret;
 624
 625         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 626                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 627         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 628         if (ret)
 629                 mlog_errno(ret);
 630
 631         return ret;
 632 }
 633
 634 static int ocfs2_link(struct dentry *old_dentry,
 635                       struct inode *dir,
 636                       struct dentry *dentry)
 637 {
 638         handle_t *handle;
 639         struct inode *inode = old_dentry->d_inode;
 640         int err;
 641         struct buffer_head *fe_bh = NULL;
 642         struct buffer_head *parent_fe_bh = NULL;
 643         struct ocfs2_dinode *fe = NULL;
 644         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 645         struct ocfs2_dir_lookup_result lookup = { NULL, };
 646         sigset_t oldset;
 647
 648         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 649                          old_dentry->d_name.len, old_dentry->d_name.name,
 650                          dentry->d_name.len, dentry->d_name.name);
 651
 652         if (S_ISDIR(inode->i_mode))
 653                 return -EPERM;
 654
 655         dquot_initialize(dir);
 656
 657         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 658         if (err < 0) {
 659                 if (err != -ENOENT)
 660                         mlog_errno(err);
 661                 return err;
 662         }
 663
 664         if (!dir->i_nlink) {
 665                 err = -ENOENT;
 666                 goto out;
 667         }
 668
 669         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 670                                         dentry->d_name.len);
 671         if (err)
 672                 goto out;
 673
 674         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 675                                            dentry->d_name.name,
 676                                            dentry->d_name.len, &lookup);
 677         if (err < 0) {
 678                 mlog_errno(err);
 679                 goto out;
 680         }
 681
 682         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 683         if (err < 0) {
 684                 if (err != -ENOENT)
 685                         mlog_errno(err);
 686                 goto out;
 687         }
 688
 689         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 690         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 691                 err = -EMLINK;
 692                 goto out_unlock_inode;
 693         }
 694
 695         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 696         if (IS_ERR(handle)) {
 697                 err = PTR_ERR(handle);
 698                 handle = NULL;
 699                 mlog_errno(err);
 700                 goto out_unlock_inode;
 701         }
 702
 703         /* Starting to change things, restart is no longer possible. */
 704         ocfs2_block_signals(&oldset);
 705
 706         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 707                                       OCFS2_JOURNAL_ACCESS_WRITE);
 708         if (err < 0) {
 709                 mlog_errno(err);
 710                 goto out_commit;
 711         }
 712
 713         inc_nlink(inode);
 714         inode->i_ctime = CURRENT_TIME;
 715         ocfs2_set_links_count(fe, inode->i_nlink);
 716         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 717         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 718         ocfs2_journal_dirty(handle, fe_bh);
 719
 720         err = ocfs2_add_entry(handle, dentry, inode,
 721                               OCFS2_I(inode)->ip_blkno,
 722                               parent_fe_bh, &lookup);
 723         if (err) {
 724                 ocfs2_add_links_count(fe, -1);
 725                 drop_nlink(inode);
 726                 mlog_errno(err);
 727                 goto out_commit;
 728         }
 729
 730         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 731         if (err) {
 732                 mlog_errno(err);
 733                 goto out_commit;
 734         }
 735
 736         ihold(inode);
 737         d_instantiate(dentry, inode);
 738
 739 out_commit:
 740         ocfs2_commit_trans(osb, handle);
 741         ocfs2_unblock_signals(&oldset);
 742 out_unlock_inode:
 743         ocfs2_inode_unlock(inode, 1);
 744
 745 out:
 746         ocfs2_inode_unlock(dir, 1);
 747
 748         brelse(fe_bh);
 749         brelse(parent_fe_bh);
 750
 751         ocfs2_free_dir_lookup_result(&lookup);
 752
 753         if (err)
 754                 mlog_errno(err);
 755
 756         return err;
 757 }
 758
 759 /*
 760  * Takes and drops an exclusive lock on the given dentry. This will
 761  * force other nodes to drop it.
 762  */
 763 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 764 {
 765         int ret;
 766
 767         ret = ocfs2_dentry_lock(dentry, 1);
 768         if (ret)
 769                 mlog_errno(ret);
 770         else
 771                 ocfs2_dentry_unlock(dentry, 1);
 772
 773         return ret;
 774 }
 775
 776 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 777 {
 778         if (S_ISDIR(inode->i_mode)) {
 779                 if (inode->i_nlink == 2)
 780                         return 1;
 781                 return 0;
 782         }
 783
 784         if (inode->i_nlink == 1)
 785                 return 1;
 786         return 0;
 787 }
 788
 789 static int ocfs2_unlink(struct inode *dir,
 790                         struct dentry *dentry)
 791 {
 792         int status;
 793         int child_locked = 0;
 794         bool is_unlinkable = false;
 795         struct inode *inode = dentry->d_inode;
 796         struct inode *orphan_dir = NULL;
 797         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 798         u64 blkno;
 799         struct ocfs2_dinode *fe = NULL;
 800         struct buffer_head *fe_bh = NULL;
 801         struct buffer_head *parent_node_bh = NULL;
 802         handle_t *handle = NULL;
 803         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 804         struct ocfs2_dir_lookup_result lookup = { NULL, };
 805         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 806
 807         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 808                            dentry->d_name.name,
 809                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 810                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 811
 812         dquot_initialize(dir);
 813
 814         BUG_ON(dentry->d_parent->d_inode != dir);
 815
 816         if (inode == osb->root_inode)
 817                 return -EPERM;
 818
 819         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 820                                          OI_LS_PARENT);
 821         if (status < 0) {
 822                 if (status != -ENOENT)
 823                         mlog_errno(status);
 824                 return status;
 825         }
 826
 827         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 828                                           dentry->d_name.len, &blkno, dir,
 829                                           &lookup);
 830         if (status < 0) {
 831                 if (status != -ENOENT)
 832                         mlog_errno(status);
 833                 goto leave;
 834         }
 835
 836         if (OCFS2_I(inode)->ip_blkno != blkno) {
 837                 status = -ENOENT;
 838
 839                 trace_ocfs2_unlink_noent(
 840                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 841                                 (unsigned long long)blkno,
 842                                 OCFS2_I(inode)->ip_flags);
 843                 goto leave;
 844         }
 845
 846         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 847         if (status < 0) {
 848                 if (status != -ENOENT)
 849                         mlog_errno(status);
 850                 goto leave;
 851         }
 852         child_locked = 1;
 853
 854         if (S_ISDIR(inode->i_mode)) {
 855                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 856                         status = -ENOTEMPTY;
 857                         goto leave;
 858                 }
 859         }
 860
 861         status = ocfs2_remote_dentry_delete(dentry);
 862         if (status < 0) {
 863                 /* This remote delete should succeed under all normal
 864                  * circumstances. */
 865                 mlog_errno(status);
 866                 goto leave;
 867         }
 868
 869         if (ocfs2_inode_is_unlinkable(inode)) {
 870                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 871                                                   OCFS2_I(inode)->ip_blkno,
 872                                                   orphan_name, &orphan_insert);
 873                 if (status < 0) {
 874                         mlog_errno(status);
 875                         goto leave;
 876                 }
 877                 is_unlinkable = true;
 878         }
 879
 880         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 881         if (IS_ERR(handle)) {
 882                 status = PTR_ERR(handle);
 883                 handle = NULL;
 884                 mlog_errno(status);
 885                 goto leave;
 886         }
 887
 888         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 889                                          OCFS2_JOURNAL_ACCESS_WRITE);
 890         if (status < 0) {
 891                 mlog_errno(status);
 892                 goto leave;
 893         }
 894
 895         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 896
 897         /* delete the name from the parent dir */
 898         status = ocfs2_delete_entry(handle, dir, &lookup);
 899         if (status < 0) {
 900                 mlog_errno(status);
 901                 goto leave;
 902         }
 903
 904         if (S_ISDIR(inode->i_mode))
 905                 drop_nlink(inode);
 906         drop_nlink(inode);
 907         ocfs2_set_links_count(fe, inode->i_nlink);
 908         ocfs2_journal_dirty(handle, fe_bh);
 909
 910         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 911         if (S_ISDIR(inode->i_mode))
 912                 drop_nlink(dir);
 913
 914         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 915         if (status < 0) {
 916                 mlog_errno(status);
 917                 if (S_ISDIR(inode->i_mode))
 918                         inc_nlink(dir);
 919                 goto leave;
 920         }
 921
 922         if (is_unlinkable) {
 923                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
 924                                 orphan_name, &orphan_insert, orphan_dir);
 925                 if (status < 0)
 926                         mlog_errno(status);
 927         }
 928
 929 leave:
 930         if (handle)
 931                 ocfs2_commit_trans(osb, handle);
 932
 933         if (child_locked)
 934                 ocfs2_inode_unlock(inode, 1);
 935
 936         ocfs2_inode_unlock(dir, 1);
 937
 938         if (orphan_dir) {
 939                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 940                 ocfs2_inode_unlock(orphan_dir, 1);
 941                 mutex_unlock(&orphan_dir->i_mutex);
 942                 iput(orphan_dir);
 943         }
 944
 945         brelse(fe_bh);
 946         brelse(parent_node_bh);
 947
 948         ocfs2_free_dir_lookup_result(&orphan_insert);
 949         ocfs2_free_dir_lookup_result(&lookup);
 950
 951         if (status && (status != -ENOTEMPTY))
 952                 mlog_errno(status);
 953
 954         return status;
 955 }
 956
 957 /*
 958  * The only place this should be used is rename!
 959  * if they have the same id, then the 1st one is the only one locked.
 960  */
 961 static int ocfs2_double_lock(struct ocfs2_super *osb,
 962                              struct buffer_head **bh1,
 963                              struct inode *inode1,
 964                              struct buffer_head **bh2,
 965                              struct inode *inode2)
 966 {
 967         int status;
 968         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 969         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 970         struct buffer_head **tmpbh;
 971         struct inode *tmpinode;
 972
 973         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
 974                                 (unsigned long long)oi2->ip_blkno);
 975
 976         if (*bh1)
 977                 *bh1 = NULL;
 978         if (*bh2)
 979                 *bh2 = NULL;
 980
 981         /* we always want to lock the one with the lower lockid first. */
 982         if (oi1->ip_blkno != oi2->ip_blkno) {
 983                 if (oi1->ip_blkno < oi2->ip_blkno) {
 984                         /* switch id1 and id2 around */
 985                         tmpbh = bh2;
 986                         bh2 = bh1;
 987                         bh1 = tmpbh;
 988
 989                         tmpinode = inode2;
 990                         inode2 = inode1;
 991                         inode1 = tmpinode;
 992                 }
 993                 /* lock id2 */
 994                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 995                                                  OI_LS_RENAME1);
 996                 if (status < 0) {
 997                         if (status != -ENOENT)
 998                                 mlog_errno(status);
 999                         goto bail;
1000                 }
1001         }
1002
1003         /* lock id1 */
1004         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1005         if (status < 0) {
1006                 /*
1007                  * An error return must mean that no cluster locks
1008                  * were held on function exit.
1009                  */
1010                 if (oi1->ip_blkno != oi2->ip_blkno) {
1011                         ocfs2_inode_unlock(inode2, 1);
1012                         brelse(*bh2);
1013                         *bh2 = NULL;
1014                 }
1015
1016                 if (status != -ENOENT)
1017                         mlog_errno(status);
1018         }
1019
1020         trace_ocfs2_double_lock_end(
1021                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1022                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1023
1024 bail:
1025         if (status)
1026                 mlog_errno(status);
1027         return status;
1028 }
1029
1030 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1031 {
1032         ocfs2_inode_unlock(inode1, 1);
1033
1034         if (inode1 != inode2)
1035                 ocfs2_inode_unlock(inode2, 1);
1036 }
1037
1038 static int ocfs2_rename(struct inode *old_dir,
1039                         struct dentry *old_dentry,
1040                         struct inode *new_dir,
1041                         struct dentry *new_dentry)
1042 {
1043         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1044         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1045         struct inode *old_inode = old_dentry->d_inode;
1046         struct inode *new_inode = new_dentry->d_inode;
1047         struct inode *orphan_dir = NULL;
1048         struct ocfs2_dinode *newfe = NULL;
1049         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1050         struct buffer_head *newfe_bh = NULL;
1051         struct buffer_head *old_inode_bh = NULL;
1052         struct ocfs2_super *osb = NULL;
1053         u64 newfe_blkno, old_de_ino;
1054         handle_t *handle = NULL;
1055         struct buffer_head *old_dir_bh = NULL;
1056         struct buffer_head *new_dir_bh = NULL;
1057         u32 old_dir_nlink = old_dir->i_nlink;
1058         struct ocfs2_dinode *old_di;
1059         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1060         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1061         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1062         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1063         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1064
1065         /* At some point it might be nice to break this function up a
1066          * bit. */
1067
1068         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1069                            old_dentry->d_name.len, old_dentry->d_name.name,
1070                            new_dentry->d_name.len, new_dentry->d_name.name);
1071
1072         dquot_initialize(old_dir);
1073         dquot_initialize(new_dir);
1074
1075         osb = OCFS2_SB(old_dir->i_sb);
1076
1077         if (new_inode) {
1078                 if (!igrab(new_inode))
1079                         BUG();
1080         }
1081
1082         /* Assume a directory hierarchy thusly:
1083          * a/b/c
1084          * a/d
1085          * a,b,c, and d are all directories.
1086          *
1087          * from cwd of 'a' on both nodes:
1088          * node1: mv b/c d
1089          * node2: mv d   b/c
1090          *
1091          * And that's why, just like the VFS, we need a file system
1092          * rename lock. */
1093         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1094                 status = ocfs2_rename_lock(osb);
1095                 if (status < 0) {
1096                         mlog_errno(status);
1097                         goto bail;
1098                 }
1099                 rename_lock = 1;
1100         }
1101
1102         /* if old and new are the same, this'll just do one lock. */
1103         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1104                                    &new_dir_bh, new_dir);
1105         if (status < 0) {
1106                 mlog_errno(status);
1107                 goto bail;
1108         }
1109         parents_locked = 1;
1110
1111         /* make sure both dirs have bhs
1112          * get an extra ref on old_dir_bh if old==new */
1113         if (!new_dir_bh) {
1114                 if (old_dir_bh) {
1115                         new_dir_bh = old_dir_bh;
1116                         get_bh(new_dir_bh);
1117                 } else {
1118                         mlog(ML_ERROR, "no old_dir_bh!\n");
1119                         status = -EIO;
1120                         goto bail;
1121                 }
1122         }
1123
1124         /*
1125          * Aside from allowing a meta data update, the locking here
1126          * also ensures that the downconvert thread on other nodes
1127          * won't have to concurrently downconvert the inode and the
1128          * dentry locks.
1129          */
1130         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1131                                          OI_LS_PARENT);
1132         if (status < 0) {
1133                 if (status != -ENOENT)
1134                         mlog_errno(status);
1135                 goto bail;
1136         }
1137         old_child_locked = 1;
1138
1139         status = ocfs2_remote_dentry_delete(old_dentry);
1140         if (status < 0) {
1141                 mlog_errno(status);
1142                 goto bail;
1143         }
1144
1145         if (S_ISDIR(old_inode->i_mode)) {
1146                 u64 old_inode_parent;
1147
1148                 update_dot_dot = 1;
1149                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1150                                                   old_inode,
1151                                                   &old_inode_dot_dot_res);
1152                 if (status) {
1153                         status = -EIO;
1154                         goto bail;
1155                 }
1156
1157                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1158                         status = -EIO;
1159                         goto bail;
1160                 }
1161
1162                 if (!new_inode && new_dir != old_dir &&
1163                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1164                         status = -EMLINK;
1165                         goto bail;
1166                 }
1167         }
1168
1169         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1170                                             old_dentry->d_name.len,
1171                                             &old_de_ino);
1172         if (status) {
1173                 status = -ENOENT;
1174                 goto bail;
1175         }
1176
1177         /*
1178          *  Check for inode number is _not_ due to possible IO errors.
1179          *  We might rmdir the source, keep it as pwd of some process
1180          *  and merrily kill the link to whatever was created under the
1181          *  same name. Goodbye sticky bit ;-<
1182          */
1183         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1184                 status = -ENOENT;
1185                 goto bail;
1186         }
1187
1188         /* check if the target already exists (in which case we need
1189          * to delete it */
1190         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1191                                           new_dentry->d_name.len,
1192                                           &newfe_blkno, new_dir,
1193                                           &target_lookup_res);
1194         /* The only error we allow here is -ENOENT because the new
1195          * file not existing is perfectly valid. */
1196         if ((status < 0) && (status != -ENOENT)) {
1197                 /* If we cannot find the file specified we should just */
1198                 /* return the error... */
1199                 mlog_errno(status);
1200                 goto bail;
1201         }
1202         if (status == 0)
1203                 target_exists = 1;
1204
1205         if (!target_exists && new_inode) {
1206                 /*
1207                  * Target was unlinked by another node while we were
1208                  * waiting to get to ocfs2_rename(). There isn't
1209                  * anything we can do here to help the situation, so
1210                  * bubble up the appropriate error.
1211                  */
1212                 status = -ENOENT;
1213                 goto bail;
1214         }
1215
1216         /* In case we need to overwrite an existing file, we blow it
1217          * away first */
1218         if (target_exists) {
1219                 /* VFS didn't think there existed an inode here, but
1220                  * someone else in the cluster must have raced our
1221                  * rename to create one. Today we error cleanly, in
1222                  * the future we should consider calling iget to build
1223                  * a new struct inode for this entry. */
1224                 if (!new_inode) {
1225                         status = -EACCES;
1226
1227                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1228                                                 new_dentry->d_name.name);
1229                         goto bail;
1230                 }
1231
1232                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1233                         status = -EACCES;
1234
1235                         trace_ocfs2_rename_disagree(
1236                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1237                              (unsigned long long)newfe_blkno,
1238                              OCFS2_I(new_inode)->ip_flags);
1239                         goto bail;
1240                 }
1241
1242                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1243                 if (status < 0) {
1244                         if (status != -ENOENT)
1245                                 mlog_errno(status);
1246                         goto bail;
1247                 }
1248                 new_child_locked = 1;
1249
1250                 status = ocfs2_remote_dentry_delete(new_dentry);
1251                 if (status < 0) {
1252                         mlog_errno(status);
1253                         goto bail;
1254                 }
1255
1256                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1257
1258                 trace_ocfs2_rename_over_existing(
1259                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1260                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1261
1262                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1263                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1264                                                 OCFS2_I(new_inode)->ip_blkno,
1265                                                 orphan_name, &orphan_insert);
1266                         if (status < 0) {
1267                                 mlog_errno(status);
1268                                 goto bail;
1269                         }
1270                 }
1271         } else {
1272                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1273
1274                 status = ocfs2_check_dir_for_entry(new_dir,
1275                                                    new_dentry->d_name.name,
1276                                                    new_dentry->d_name.len);
1277                 if (status)
1278                         goto bail;
1279
1280                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1281                                                       new_dentry->d_name.name,
1282                                                       new_dentry->d_name.len,
1283                                                       &target_insert);
1284                 if (status < 0) {
1285                         mlog_errno(status);
1286                         goto bail;
1287                 }
1288         }
1289
1290         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1291         if (IS_ERR(handle)) {
1292                 status = PTR_ERR(handle);
1293                 handle = NULL;
1294                 mlog_errno(status);
1295                 goto bail;
1296         }
1297
1298         if (target_exists) {
1299                 if (S_ISDIR(new_inode->i_mode)) {
1300                         if (new_inode->i_nlink != 2 ||
1301                             !ocfs2_empty_dir(new_inode)) {
1302                                 status = -ENOTEMPTY;
1303                                 goto bail;
1304                         }
1305                 }
1306                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1307                                                  newfe_bh,
1308                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1309                 if (status < 0) {
1310                         mlog_errno(status);
1311                         goto bail;
1312                 }
1313
1314                 if (S_ISDIR(new_inode->i_mode) ||
1315                     (ocfs2_read_links_count(newfe) == 1)) {
1316                         status = ocfs2_orphan_add(osb, handle, new_inode,
1317                                                   newfe_bh, orphan_name,
1318                                                   &orphan_insert, orphan_dir);
1319                         if (status < 0) {
1320                                 mlog_errno(status);
1321                                 goto bail;
1322                         }
1323                 }
1324
1325                 /* change the dirent to point to the correct inode */
1326                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1327                                             old_inode);
1328                 if (status < 0) {
1329                         mlog_errno(status);
1330                         goto bail;
1331                 }
1332                 new_dir->i_version++;
1333
1334                 if (S_ISDIR(new_inode->i_mode))
1335                         ocfs2_set_links_count(newfe, 0);
1336                 else
1337                         ocfs2_add_links_count(newfe, -1);
1338                 ocfs2_journal_dirty(handle, newfe_bh);
1339         } else {
1340                 /* if the name was not found in new_dir, add it now */
1341                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1342                                          OCFS2_I(old_inode)->ip_blkno,
1343                                          new_dir_bh, &target_insert);
1344         }
1345
1346         old_inode->i_ctime = CURRENT_TIME;
1347         mark_inode_dirty(old_inode);
1348
1349         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1350                                          old_inode_bh,
1351                                          OCFS2_JOURNAL_ACCESS_WRITE);
1352         if (status >= 0) {
1353                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1354
1355                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1356                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1357                 ocfs2_journal_dirty(handle, old_inode_bh);
1358         } else
1359                 mlog_errno(status);
1360
1361         /*
1362          * Now that the name has been added to new_dir, remove the old name.
1363          *
1364          * We don't keep any directory entry context around until now
1365          * because the insert might have changed the type of directory
1366          * we're dealing with.
1367          */
1368         status = ocfs2_find_entry(old_dentry->d_name.name,
1369                                   old_dentry->d_name.len, old_dir,
1370                                   &old_entry_lookup);
1371         if (status)
1372                 goto bail;
1373
1374         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1375         if (status < 0) {
1376                 mlog_errno(status);
1377                 goto bail;
1378         }
1379
1380         if (new_inode) {
1381                 drop_nlink(new_inode);
1382                 new_inode->i_ctime = CURRENT_TIME;
1383         }
1384         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1385
1386         if (update_dot_dot) {
1387                 status = ocfs2_update_entry(old_inode, handle,
1388                                             &old_inode_dot_dot_res, new_dir);
1389                 drop_nlink(old_dir);
1390                 if (new_inode) {
1391                         drop_nlink(new_inode);
1392                 } else {
1393                         inc_nlink(new_dir);
1394                         mark_inode_dirty(new_dir);
1395                 }
1396         }
1397         mark_inode_dirty(old_dir);
1398         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1399         if (new_inode) {
1400                 mark_inode_dirty(new_inode);
1401                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1402         }
1403
1404         if (old_dir != new_dir) {
1405                 /* Keep the same times on both directories.*/
1406                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1407
1408                 /*
1409                  * This will also pick up the i_nlink change from the
1410                  * block above.
1411                  */
1412                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1413         }
1414
1415         if (old_dir_nlink != old_dir->i_nlink) {
1416                 if (!old_dir_bh) {
1417                         mlog(ML_ERROR, "need to change nlink for old dir "
1418                              "%llu from %d to %d but bh is NULL!\n",
1419                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1420                              (int)old_dir_nlink, old_dir->i_nlink);
1421                 } else {
1422                         struct ocfs2_dinode *fe;
1423                         status = ocfs2_journal_access_di(handle,
1424                                                          INODE_CACHE(old_dir),
1425                                                          old_dir_bh,
1426                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1427                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1428                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1429                         ocfs2_journal_dirty(handle, old_dir_bh);
1430                 }
1431         }
1432         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1433         status = 0;
1434 bail:
1435         if (rename_lock)
1436                 ocfs2_rename_unlock(osb);
1437
1438         if (handle)
1439                 ocfs2_commit_trans(osb, handle);
1440
1441         if (parents_locked)
1442                 ocfs2_double_unlock(old_dir, new_dir);
1443
1444         if (old_child_locked)
1445                 ocfs2_inode_unlock(old_inode, 1);
1446
1447         if (new_child_locked)
1448                 ocfs2_inode_unlock(new_inode, 1);
1449
1450         if (orphan_dir) {
1451                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1452                 ocfs2_inode_unlock(orphan_dir, 1);
1453                 mutex_unlock(&orphan_dir->i_mutex);
1454                 iput(orphan_dir);
1455         }
1456
1457         if (new_inode)
1458                 sync_mapping_buffers(old_inode->i_mapping);
1459
1460         if (new_inode)
1461                 iput(new_inode);
1462
1463         ocfs2_free_dir_lookup_result(&target_lookup_res);
1464         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1465         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1466         ocfs2_free_dir_lookup_result(&orphan_insert);
1467         ocfs2_free_dir_lookup_result(&target_insert);
1468
1469         brelse(newfe_bh);
1470         brelse(old_inode_bh);
1471         brelse(old_dir_bh);
1472         brelse(new_dir_bh);
1473
1474         if (status)
1475                 mlog_errno(status);
1476
1477         return status;
1478 }
1479
1480 /*
1481  * we expect i_size = strlen(symname). Copy symname into the file
1482  * data, including the null terminator.
1483  */
1484 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1485                                      handle_t *handle,
1486                                      struct inode *inode,
1487                                      const char *symname)
1488 {
1489         struct buffer_head **bhs = NULL;
1490         const char *c;
1491         struct super_block *sb = osb->sb;
1492         u64 p_blkno, p_blocks;
1493         int virtual, blocks, status, i, bytes_left;
1494
1495         bytes_left = i_size_read(inode) + 1;
1496         /* we can't trust i_blocks because we're actually going to
1497          * write i_size + 1 bytes. */
1498         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1499
1500         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1501                                         i_size_read(inode), blocks);
1502
1503         /* Sanity check -- make sure we're going to fit. */
1504         if (bytes_left >
1505             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1506                 status = -EIO;
1507                 mlog_errno(status);
1508                 goto bail;
1509         }
1510
1511         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1512         if (!bhs) {
1513                 status = -ENOMEM;
1514                 mlog_errno(status);
1515                 goto bail;
1516         }
1517
1518         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1519                                              NULL);
1520         if (status < 0) {
1521                 mlog_errno(status);
1522                 goto bail;
1523         }
1524
1525         /* links can never be larger than one cluster so we know this
1526          * is all going to be contiguous, but do a sanity check
1527          * anyway. */
1528         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1529                 status = -EIO;
1530                 mlog_errno(status);
1531                 goto bail;
1532         }
1533
1534         virtual = 0;
1535         while(bytes_left > 0) {
1536                 c = &symname[virtual * sb->s_blocksize];
1537
1538                 bhs[virtual] = sb_getblk(sb, p_blkno);
1539                 if (!bhs[virtual]) {
1540                         status = -ENOMEM;
1541                         mlog_errno(status);
1542                         goto bail;
1543                 }
1544                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1545                                               bhs[virtual]);
1546
1547                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1548                                               bhs[virtual],
1549                                               OCFS2_JOURNAL_ACCESS_CREATE);
1550                 if (status < 0) {
1551                         mlog_errno(status);
1552                         goto bail;
1553                 }
1554
1555                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1556
1557                 memcpy(bhs[virtual]->b_data, c,
1558                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1559                        bytes_left);
1560
1561                 ocfs2_journal_dirty(handle, bhs[virtual]);
1562
1563                 virtual++;
1564                 p_blkno++;
1565                 bytes_left -= sb->s_blocksize;
1566         }
1567
1568         status = 0;
1569 bail:
1570
1571         if (bhs) {
1572                 for(i = 0; i < blocks; i++)
1573                         brelse(bhs[i]);
1574                 kfree(bhs);
1575         }
1576
1577         if (status)
1578                 mlog_errno(status);
1579         return status;
1580 }
1581
1582 static int ocfs2_symlink(struct inode *dir,
1583                          struct dentry *dentry,
1584                          const char *symname)
1585 {
1586         int status, l, credits;
1587         u64 newsize;
1588         struct ocfs2_super *osb = NULL;
1589         struct inode *inode = NULL;
1590         struct super_block *sb;
1591         struct buffer_head *new_fe_bh = NULL;
1592         struct buffer_head *parent_fe_bh = NULL;
1593         struct ocfs2_dinode *fe = NULL;
1594         struct ocfs2_dinode *dirfe;
1595         handle_t *handle = NULL;
1596         struct ocfs2_alloc_context *inode_ac = NULL;
1597         struct ocfs2_alloc_context *data_ac = NULL;
1598         struct ocfs2_alloc_context *xattr_ac = NULL;
1599         int want_clusters = 0;
1600         int xattr_credits = 0;
1601         struct ocfs2_security_xattr_info si = {
1602                 .enable = 1,
1603         };
1604         int did_quota = 0, did_quota_inode = 0;
1605         struct ocfs2_dir_lookup_result lookup = { NULL, };
1606         sigset_t oldset;
1607         int did_block_signals = 0;
1608
1609         trace_ocfs2_symlink_begin(dir, dentry, symname,
1610                                   dentry->d_name.len, dentry->d_name.name);
1611
1612         dquot_initialize(dir);
1613
1614         sb = dir->i_sb;
1615         osb = OCFS2_SB(sb);
1616
1617         l = strlen(symname) + 1;
1618
1619         credits = ocfs2_calc_symlink_credits(sb);
1620
1621         /* lock the parent directory */
1622         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1623         if (status < 0) {
1624                 if (status != -ENOENT)
1625                         mlog_errno(status);
1626                 return status;
1627         }
1628
1629         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1630         if (!ocfs2_read_links_count(dirfe)) {
1631                 /* can't make a file in a deleted directory. */
1632                 status = -ENOENT;
1633                 goto bail;
1634         }
1635
1636         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1637                                            dentry->d_name.len);
1638         if (status)
1639                 goto bail;
1640
1641         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1642                                               dentry->d_name.name,
1643                                               dentry->d_name.len, &lookup);
1644         if (status < 0) {
1645                 mlog_errno(status);
1646                 goto bail;
1647         }
1648
1649         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1650         if (status < 0) {
1651                 if (status != -ENOSPC)
1652                         mlog_errno(status);
1653                 goto bail;
1654         }
1655
1656         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1657         if (!inode) {
1658                 status = -ENOMEM;
1659                 mlog_errno(status);
1660                 goto bail;
1661         }
1662
1663         /* get security xattr */
1664         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1665         if (status) {
1666                 if (status == -EOPNOTSUPP)
1667                         si.enable = 0;
1668                 else {
1669                         mlog_errno(status);
1670                         goto bail;
1671                 }
1672         }
1673
1674         /* calculate meta data/clusters for setting security xattr */
1675         if (si.enable) {
1676                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1677                                                   &xattr_credits, &xattr_ac);
1678                 if (status < 0) {
1679                         mlog_errno(status);
1680                         goto bail;
1681                 }
1682         }
1683
1684         /* don't reserve bitmap space for fast symlinks. */
1685         if (l > ocfs2_fast_symlink_chars(sb))
1686                 want_clusters += 1;
1687
1688         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1689         if (status < 0) {
1690                 if (status != -ENOSPC)
1691                         mlog_errno(status);
1692                 goto bail;
1693         }
1694
1695         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1696         if (IS_ERR(handle)) {
1697                 status = PTR_ERR(handle);
1698                 handle = NULL;
1699                 mlog_errno(status);
1700                 goto bail;
1701         }
1702
1703         /* Starting to change things, restart is no longer possible. */
1704         ocfs2_block_signals(&oldset);
1705         did_block_signals = 1;
1706
1707         status = dquot_alloc_inode(inode);
1708         if (status)
1709                 goto bail;
1710         did_quota_inode = 1;
1711
1712         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1713                                    dentry->d_name.name,
1714                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1715                                    inode->i_mode);
1716
1717         status = ocfs2_mknod_locked(osb, dir, inode,
1718                                     0, &new_fe_bh, parent_fe_bh, handle,
1719                                     inode_ac);
1720         if (status < 0) {
1721                 mlog_errno(status);
1722                 goto bail;
1723         }
1724
1725         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1726         inode->i_rdev = 0;
1727         newsize = l - 1;
1728         inode->i_op = &ocfs2_symlink_inode_operations;
1729         if (l > ocfs2_fast_symlink_chars(sb)) {
1730                 u32 offset = 0;
1731
1732                 status = dquot_alloc_space_nodirty(inode,
1733                     ocfs2_clusters_to_bytes(osb->sb, 1));
1734                 if (status)
1735                         goto bail;
1736                 did_quota = 1;
1737                 inode->i_mapping->a_ops = &ocfs2_aops;
1738                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1739                                               new_fe_bh,
1740                                               handle, data_ac, NULL,
1741                                               NULL);
1742                 if (status < 0) {
1743                         if (status != -ENOSPC && status != -EINTR) {
1744                                 mlog(ML_ERROR,
1745                                      "Failed to extend file to %llu\n",
1746                                      (unsigned long long)newsize);
1747                                 mlog_errno(status);
1748                                 status = -ENOSPC;
1749                         }
1750                         goto bail;
1751                 }
1752                 i_size_write(inode, newsize);
1753                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1754         } else {
1755                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1756                 memcpy((char *) fe->id2.i_symlink, symname, l);
1757                 i_size_write(inode, newsize);
1758                 inode->i_blocks = 0;
1759         }
1760
1761         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1762         if (status < 0) {
1763                 mlog_errno(status);
1764                 goto bail;
1765         }
1766
1767         if (!ocfs2_inode_is_fast_symlink(inode)) {
1768                 status = ocfs2_create_symlink_data(osb, handle, inode,
1769                                                    symname);
1770                 if (status < 0) {
1771                         mlog_errno(status);
1772                         goto bail;
1773                 }
1774         }
1775
1776         if (si.enable) {
1777                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1778                                                  xattr_ac, data_ac);
1779                 if (status < 0) {
1780                         mlog_errno(status);
1781                         goto bail;
1782                 }
1783         }
1784
1785         /*
1786          * Do this before adding the entry to the directory. We add
1787          * also set d_op after success so that ->d_iput() will cleanup
1788          * the dentry lock even if ocfs2_add_entry() fails below.
1789          */
1790         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1791         if (status) {
1792                 mlog_errno(status);
1793                 goto bail;
1794         }
1795
1796         status = ocfs2_add_entry(handle, dentry, inode,
1797                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1798                                  &lookup);
1799         if (status < 0) {
1800                 mlog_errno(status);
1801                 goto bail;
1802         }
1803
1804         insert_inode_hash(inode);
1805         d_instantiate(dentry, inode);
1806 bail:
1807         if (status < 0 && did_quota)
1808                 dquot_free_space_nodirty(inode,
1809                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1810         if (status < 0 && did_quota_inode)
1811                 dquot_free_inode(inode);
1812         if (handle)
1813                 ocfs2_commit_trans(osb, handle);
1814
1815         ocfs2_inode_unlock(dir, 1);
1816         if (did_block_signals)
1817                 ocfs2_unblock_signals(&oldset);
1818
1819         brelse(new_fe_bh);
1820         brelse(parent_fe_bh);
1821         kfree(si.name);
1822         kfree(si.value);
1823         ocfs2_free_dir_lookup_result(&lookup);
1824         if (inode_ac)
1825                 ocfs2_free_alloc_context(inode_ac);
1826         if (data_ac)
1827                 ocfs2_free_alloc_context(data_ac);
1828         if (xattr_ac)
1829                 ocfs2_free_alloc_context(xattr_ac);
1830         if ((status < 0) && inode) {
1831                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1832                 clear_nlink(inode);
1833                 iput(inode);
1834         }
1835
1836         if (status)
1837                 mlog_errno(status);
1838
1839         return status;
1840 }
1841
1842 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1843 {
1844         int status, namelen;
1845
1846         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1847                            (long long)blkno);
1848         if (namelen <= 0) {
1849                 if (namelen)
1850                         status = namelen;
1851                 else
1852                         status = -EINVAL;
1853                 mlog_errno(status);
1854                 goto bail;
1855         }
1856         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1857                 status = -EINVAL;
1858                 mlog_errno(status);
1859                 goto bail;
1860         }
1861
1862         trace_ocfs2_blkno_stringify(blkno, name, namelen);
1863
1864         status = 0;
1865 bail:
1866         if (status < 0)
1867                 mlog_errno(status);
1868         return status;
1869 }
1870
1871 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1872                                         struct inode **ret_orphan_dir,
1873                                         struct buffer_head **ret_orphan_dir_bh)
1874 {
1875         struct inode *orphan_dir_inode;
1876         struct buffer_head *orphan_dir_bh = NULL;
1877         int ret = 0;
1878
1879         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1880                                                        ORPHAN_DIR_SYSTEM_INODE,
1881                                                        osb->slot_num);
1882         if (!orphan_dir_inode) {
1883                 ret = -ENOENT;
1884                 mlog_errno(ret);
1885                 return ret;
1886         }
1887
1888         mutex_lock(&orphan_dir_inode->i_mutex);
1889
1890         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1891         if (ret < 0) {
1892                 mutex_unlock(&orphan_dir_inode->i_mutex);
1893                 iput(orphan_dir_inode);
1894
1895                 mlog_errno(ret);
1896                 return ret;
1897         }
1898
1899         *ret_orphan_dir = orphan_dir_inode;
1900         *ret_orphan_dir_bh = orphan_dir_bh;
1901
1902         return 0;
1903 }
1904
1905 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1906                                       struct buffer_head *orphan_dir_bh,
1907                                       u64 blkno,
1908                                       char *name,
1909                                       struct ocfs2_dir_lookup_result *lookup)
1910 {
1911         int ret;
1912         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1913
1914         ret = ocfs2_blkno_stringify(blkno, name);
1915         if (ret < 0) {
1916                 mlog_errno(ret);
1917                 return ret;
1918         }
1919
1920         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1921                                            orphan_dir_bh, name,
1922                                            OCFS2_ORPHAN_NAMELEN, lookup);
1923         if (ret < 0) {
1924                 mlog_errno(ret);
1925                 return ret;
1926         }
1927
1928         return 0;
1929 }
1930
1931 /**
1932  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1933  * insertion of an orphan.
1934  * @osb: ocfs2 file system
1935  * @ret_orphan_dir: Orphan dir inode - returned locked!
1936  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1937  * @lookup: dir lookup result, to be passed back into functions like
1938  *          ocfs2_orphan_add
1939  *
1940  * Returns zero on success and the ret_orphan_dir, name and lookup
1941  * fields will be populated.
1942  *
1943  * Returns non-zero on failure.
1944  */
1945 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1946                                     struct inode **ret_orphan_dir,
1947                                     u64 blkno,
1948                                     char *name,
1949                                     struct ocfs2_dir_lookup_result *lookup)
1950 {
1951         struct inode *orphan_dir_inode = NULL;
1952         struct buffer_head *orphan_dir_bh = NULL;
1953         int ret = 0;
1954
1955         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1956                                            &orphan_dir_bh);
1957         if (ret < 0) {
1958                 mlog_errno(ret);
1959                 return ret;
1960         }
1961
1962         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1963                                          blkno, name, lookup);
1964         if (ret < 0) {
1965                 mlog_errno(ret);
1966                 goto out;
1967         }
1968
1969         *ret_orphan_dir = orphan_dir_inode;
1970
1971 out:
1972         brelse(orphan_dir_bh);
1973
1974         if (ret) {
1975                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1976                 mutex_unlock(&orphan_dir_inode->i_mutex);
1977                 iput(orphan_dir_inode);
1978         }
1979
1980         if (ret)
1981                 mlog_errno(ret);
1982         return ret;
1983 }
1984
1985 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1986                             handle_t *handle,
1987                             struct inode *inode,
1988                             struct buffer_head *fe_bh,
1989                             char *name,
1990                             struct ocfs2_dir_lookup_result *lookup,
1991                             struct inode *orphan_dir_inode)
1992 {
1993         struct buffer_head *orphan_dir_bh = NULL;
1994         int status = 0;
1995         struct ocfs2_dinode *orphan_fe;
1996         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1997
1998         trace_ocfs2_orphan_add_begin(
1999                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2000
2001         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2002         if (status < 0) {
2003                 mlog_errno(status);
2004                 goto leave;
2005         }
2006
2007         status = ocfs2_journal_access_di(handle,
2008                                          INODE_CACHE(orphan_dir_inode),
2009                                          orphan_dir_bh,
2010                                          OCFS2_JOURNAL_ACCESS_WRITE);
2011         if (status < 0) {
2012                 mlog_errno(status);
2013                 goto leave;
2014         }
2015
2016         /*
2017          * We're going to journal the change of i_flags and i_orphaned_slot.
2018          * It's safe anyway, though some callers may duplicate the journaling.
2019          * Journaling within the func just make the logic look more
2020          * straightforward.
2021          */
2022         status = ocfs2_journal_access_di(handle,
2023                                          INODE_CACHE(inode),
2024                                          fe_bh,
2025                                          OCFS2_JOURNAL_ACCESS_WRITE);
2026         if (status < 0) {
2027                 mlog_errno(status);
2028                 goto leave;
2029         }
2030
2031         /* we're a cluster, and nlink can change on disk from
2032          * underneath us... */
2033         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2034         if (S_ISDIR(inode->i_mode))
2035                 ocfs2_add_links_count(orphan_fe, 1);
2036         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2037         ocfs2_journal_dirty(handle, orphan_dir_bh);
2038
2039         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2040                                    OCFS2_ORPHAN_NAMELEN, inode,
2041                                    OCFS2_I(inode)->ip_blkno,
2042                                    orphan_dir_bh, lookup);
2043         if (status < 0) {
2044                 mlog_errno(status);
2045                 goto rollback;
2046         }
2047
2048         fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2049         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2050
2051         /* Record which orphan dir our inode now resides
2052          * in. delete_inode will use this to determine which orphan
2053          * dir to lock. */
2054         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2055
2056         ocfs2_journal_dirty(handle, fe_bh);
2057
2058         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2059                                    osb->slot_num);
2060
2061 rollback:
2062         if (status < 0) {
2063                 if (S_ISDIR(inode->i_mode))
2064                         ocfs2_add_links_count(orphan_fe, -1);
2065                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2066         }
2067
2068 leave:
2069         brelse(orphan_dir_bh);
2070
2071         return status;
2072 }
2073
2074 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2075 int ocfs2_orphan_del(struct ocfs2_super *osb,
2076                      handle_t *handle,
2077                      struct inode *orphan_dir_inode,
2078                      struct inode *inode,
2079                      struct buffer_head *orphan_dir_bh)
2080 {
2081         char name[OCFS2_ORPHAN_NAMELEN + 1];
2082         struct ocfs2_dinode *orphan_fe;
2083         int status = 0;
2084         struct ocfs2_dir_lookup_result lookup = { NULL, };
2085
2086         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2087         if (status < 0) {
2088                 mlog_errno(status);
2089                 goto leave;
2090         }
2091
2092         trace_ocfs2_orphan_del(
2093              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2094              name, OCFS2_ORPHAN_NAMELEN);
2095
2096         /* find it's spot in the orphan directory */
2097         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2098                                   &lookup);
2099         if (status) {
2100                 mlog_errno(status);
2101                 goto leave;
2102         }
2103
2104         /* remove it from the orphan directory */
2105         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2106         if (status < 0) {
2107                 mlog_errno(status);
2108                 goto leave;
2109         }
2110
2111         status = ocfs2_journal_access_di(handle,
2112                                          INODE_CACHE(orphan_dir_inode),
2113                                          orphan_dir_bh,
2114                                          OCFS2_JOURNAL_ACCESS_WRITE);
2115         if (status < 0) {
2116                 mlog_errno(status);
2117                 goto leave;
2118         }
2119
2120         /* do the i_nlink dance! :) */
2121         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2122         if (S_ISDIR(inode->i_mode))
2123                 ocfs2_add_links_count(orphan_fe, -1);
2124         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2125         ocfs2_journal_dirty(handle, orphan_dir_bh);
2126
2127 leave:
2128         ocfs2_free_dir_lookup_result(&lookup);
2129
2130         if (status)
2131                 mlog_errno(status);
2132         return status;
2133 }
2134
2135 /**
2136  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2137  * allocated file. This is different from the typical 'add to orphan dir'
2138  * operation in that the inode does not yet exist. This is a problem because
2139  * the orphan dir stringifies the inode block number to come up with it's
2140  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2141  * problem. This function works around it by calling deeper into the orphan
2142  * and suballoc code than other callers. Use this only by necessity.
2143  * @dir: The directory which this inode will ultimately wind up under - not the
2144  * orphan dir!
2145  * @dir_bh: buffer_head the @dir inode block
2146  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2147  * with the string to be used for orphan dirent. Pass back to the orphan dir
2148  * code.
2149  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2150  * dir code.
2151  * @ret_di_blkno: block number where the new inode will be allocated.
2152  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2153  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2154  *
2155  * Returns zero on success and the ret_orphan_dir, name and lookup
2156  * fields will be populated.
2157  *
2158  * Returns non-zero on failure.
2159  */
2160 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2161                                         struct buffer_head *dir_bh,
2162                                         char *orphan_name,
2163                                         struct inode **ret_orphan_dir,
2164                                         u64 *ret_di_blkno,
2165                                         struct ocfs2_dir_lookup_result *orphan_insert,
2166                                         struct ocfs2_alloc_context **ret_inode_ac)
2167 {
2168         int ret;
2169         u64 di_blkno;
2170         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2171         struct inode *orphan_dir = NULL;
2172         struct buffer_head *orphan_dir_bh = NULL;
2173         struct ocfs2_alloc_context *inode_ac = NULL;
2174
2175         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2176         if (ret < 0) {
2177                 mlog_errno(ret);
2178                 return ret;
2179         }
2180
2181         /* reserve an inode spot */
2182         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2183         if (ret < 0) {
2184                 if (ret != -ENOSPC)
2185                         mlog_errno(ret);
2186                 goto out;
2187         }
2188
2189         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2190                                        &di_blkno);
2191         if (ret) {
2192                 mlog_errno(ret);
2193                 goto out;
2194         }
2195
2196         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2197                                          di_blkno, orphan_name, orphan_insert);
2198         if (ret < 0) {
2199                 mlog_errno(ret);
2200                 goto out;
2201         }
2202
2203 out:
2204         if (ret == 0) {
2205                 *ret_orphan_dir = orphan_dir;
2206                 *ret_di_blkno = di_blkno;
2207                 *ret_inode_ac = inode_ac;
2208                 /*
2209                  * orphan_name and orphan_insert are already up to
2210                  * date via prepare_orphan_dir
2211                  */
2212         } else {
2213                 /* Unroll reserve_new_inode* */
2214                 if (inode_ac)
2215                         ocfs2_free_alloc_context(inode_ac);
2216
2217                 /* Unroll orphan dir locking */
2218                 mutex_unlock(&orphan_dir->i_mutex);
2219                 ocfs2_inode_unlock(orphan_dir, 1);
2220                 iput(orphan_dir);
2221         }
2222
2223         brelse(orphan_dir_bh);
2224
2225         return ret;
2226 }
2227
2228 int ocfs2_create_inode_in_orphan(struct inode *dir,
2229                                  int mode,
2230                                  struct inode **new_inode)
2231 {
2232         int status, did_quota_inode = 0;
2233         struct inode *inode = NULL;
2234         struct inode *orphan_dir = NULL;
2235         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2236         struct ocfs2_dinode *di = NULL;
2237         handle_t *handle = NULL;
2238         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2239         struct buffer_head *parent_di_bh = NULL;
2240         struct buffer_head *new_di_bh = NULL;
2241         struct ocfs2_alloc_context *inode_ac = NULL;
2242         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2243         u64 uninitialized_var(di_blkno), suballoc_loc;
2244         u16 suballoc_bit;
2245
2246         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2247         if (status < 0) {
2248                 if (status != -ENOENT)
2249                         mlog_errno(status);
2250                 return status;
2251         }
2252
2253         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2254                                               orphan_name, &orphan_dir,
2255                                               &di_blkno, &orphan_insert, &inode_ac);
2256         if (status < 0) {
2257                 if (status != -ENOSPC)
2258                         mlog_errno(status);
2259                 goto leave;
2260         }
2261
2262         inode = ocfs2_get_init_inode(dir, mode);
2263         if (!inode) {
2264                 status = -ENOMEM;
2265                 mlog_errno(status);
2266                 goto leave;
2267         }
2268
2269         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2270         if (IS_ERR(handle)) {
2271                 status = PTR_ERR(handle);
2272                 handle = NULL;
2273                 mlog_errno(status);
2274                 goto leave;
2275         }
2276
2277         status = dquot_alloc_inode(inode);
2278         if (status)
2279                 goto leave;
2280         did_quota_inode = 1;
2281
2282         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2283                                               &suballoc_loc,
2284                                               &suballoc_bit, di_blkno);
2285         if (status < 0) {
2286                 mlog_errno(status);
2287                 goto leave;
2288         }
2289
2290         clear_nlink(inode);
2291         /* do the real work now. */
2292         status = __ocfs2_mknod_locked(dir, inode,
2293                                       0, &new_di_bh, parent_di_bh, handle,
2294                                       inode_ac, di_blkno, suballoc_loc,
2295                                       suballoc_bit);
2296         if (status < 0) {
2297                 mlog_errno(status);
2298                 goto leave;
2299         }
2300
2301         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2302         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2303                                   &orphan_insert, orphan_dir);
2304         if (status < 0) {
2305                 mlog_errno(status);
2306                 goto leave;
2307         }
2308
2309         /* get open lock so that only nodes can't remove it from orphan dir. */
2310         status = ocfs2_open_lock(inode);
2311         if (status < 0)
2312                 mlog_errno(status);
2313
2314         insert_inode_hash(inode);
2315 leave:
2316         if (status < 0 && did_quota_inode)
2317                 dquot_free_inode(inode);
2318         if (handle)
2319                 ocfs2_commit_trans(osb, handle);
2320
2321         if (orphan_dir) {
2322                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2323                 ocfs2_inode_unlock(orphan_dir, 1);
2324                 mutex_unlock(&orphan_dir->i_mutex);
2325                 iput(orphan_dir);
2326         }
2327
2328         if ((status < 0) && inode) {
2329                 clear_nlink(inode);
2330                 iput(inode);
2331         }
2332
2333         if (inode_ac)
2334                 ocfs2_free_alloc_context(inode_ac);
2335
2336         brelse(new_di_bh);
2337
2338         if (!status)
2339                 *new_inode = inode;
2340
2341         ocfs2_free_dir_lookup_result(&orphan_insert);
2342
2343         ocfs2_inode_unlock(dir, 1);
2344         brelse(parent_di_bh);
2345         return status;
2346 }
2347
2348 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2349                                    struct inode *inode,
2350                                    struct dentry *dentry)
2351 {
2352         int status = 0;
2353         struct buffer_head *parent_di_bh = NULL;
2354         handle_t *handle = NULL;
2355         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2356         struct ocfs2_dinode *dir_di, *di;
2357         struct inode *orphan_dir_inode = NULL;
2358         struct buffer_head *orphan_dir_bh = NULL;
2359         struct buffer_head *di_bh = NULL;
2360         struct ocfs2_dir_lookup_result lookup = { NULL, };
2361
2362         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2363                                 dentry->d_name.len, dentry->d_name.name,
2364                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2365                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2366
2367         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2368         if (status < 0) {
2369                 if (status != -ENOENT)
2370                         mlog_errno(status);
2371                 return status;
2372         }
2373
2374         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2375         if (!dir_di->i_links_count) {
2376                 /* can't make a file in a deleted directory. */
2377                 status = -ENOENT;
2378                 goto leave;
2379         }
2380
2381         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2382                                            dentry->d_name.len);
2383         if (status)
2384                 goto leave;
2385
2386         /* get a spot inside the dir. */
2387         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2388                                               dentry->d_name.name,
2389                                               dentry->d_name.len, &lookup);
2390         if (status < 0) {
2391                 mlog_errno(status);
2392                 goto leave;
2393         }
2394
2395         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2396                                                        ORPHAN_DIR_SYSTEM_INODE,
2397                                                        osb->slot_num);
2398         if (!orphan_dir_inode) {
2399                 status = -EEXIST;
2400                 mlog_errno(status);
2401                 goto leave;
2402         }
2403
2404         mutex_lock(&orphan_dir_inode->i_mutex);
2405
2406         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2407         if (status < 0) {
2408                 mlog_errno(status);
2409                 mutex_unlock(&orphan_dir_inode->i_mutex);
2410                 iput(orphan_dir_inode);
2411                 goto leave;
2412         }
2413
2414         status = ocfs2_read_inode_block(inode, &di_bh);
2415         if (status < 0) {
2416                 mlog_errno(status);
2417                 goto orphan_unlock;
2418         }
2419
2420         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2421         if (IS_ERR(handle)) {
2422                 status = PTR_ERR(handle);
2423                 handle = NULL;
2424                 mlog_errno(status);
2425                 goto orphan_unlock;
2426         }
2427
2428         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2429                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2430         if (status < 0) {
2431                 mlog_errno(status);
2432                 goto out_commit;
2433         }
2434
2435         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2436                                   orphan_dir_bh);
2437         if (status < 0) {
2438                 mlog_errno(status);
2439                 goto out_commit;
2440         }
2441
2442         di = (struct ocfs2_dinode *)di_bh->b_data;
2443         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2444         di->i_orphaned_slot = 0;
2445         set_nlink(inode, 1);
2446         ocfs2_set_links_count(di, inode->i_nlink);
2447         ocfs2_journal_dirty(handle, di_bh);
2448
2449         status = ocfs2_add_entry(handle, dentry, inode,
2450                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2451                                  &lookup);
2452         if (status < 0) {
2453                 mlog_errno(status);
2454                 goto out_commit;
2455         }
2456
2457         status = ocfs2_dentry_attach_lock(dentry, inode,
2458                                           OCFS2_I(dir)->ip_blkno);
2459         if (status) {
2460                 mlog_errno(status);
2461                 goto out_commit;
2462         }
2463
2464         d_instantiate(dentry, inode);
2465         status = 0;
2466 out_commit:
2467         ocfs2_commit_trans(osb, handle);
2468 orphan_unlock:
2469         ocfs2_inode_unlock(orphan_dir_inode, 1);
2470         mutex_unlock(&orphan_dir_inode->i_mutex);
2471         iput(orphan_dir_inode);
2472 leave:
2473
2474         ocfs2_inode_unlock(dir, 1);
2475
2476         brelse(di_bh);
2477         brelse(parent_di_bh);
2478         brelse(orphan_dir_bh);
2479
2480         ocfs2_free_dir_lookup_result(&lookup);
2481
2482         if (status)
2483                 mlog_errno(status);
2484
2485         return status;
2486 }
2487
2488 const struct inode_operations ocfs2_dir_iops = {
2489         .create         = ocfs2_create,
2490         .lookup         = ocfs2_lookup,
2491         .link           = ocfs2_link,
2492         .unlink         = ocfs2_unlink,
2493         .rmdir          = ocfs2_unlink,
2494         .symlink        = ocfs2_symlink,
2495         .mkdir          = ocfs2_mkdir,
2496         .mknod          = ocfs2_mknod,
2497         .rename         = ocfs2_rename,
2498         .setattr        = ocfs2_setattr,
2499         .getattr        = ocfs2_getattr,
2500         .permission     = ocfs2_permission,
2501         .setxattr       = generic_setxattr,
2502         .getxattr       = generic_getxattr,
2503         .listxattr      = ocfs2_listxattr,
2504         .removexattr    = generic_removexattr,
2505         .fiemap         = ocfs2_fiemap,
2506         .get_acl        = ocfs2_iop_get_acl,
2507 };