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                                     bool dio);
  84
  85 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  86                             handle_t *handle,
  87                             struct inode *inode,
  88                             struct buffer_head *fe_bh,
  89                             char *name,
  90                             struct ocfs2_dir_lookup_result *lookup,
  91                             struct inode *orphan_dir_inode,
  92                             bool dio);
  93
  94 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  95                                      handle_t *handle,
  96                                      struct inode *inode,
  97                                      const char *symname);
  98
  99 static int ocfs2_double_lock(struct ocfs2_super *osb,
 100                              struct buffer_head **bh1,
 101                              struct inode *inode1,
 102                              struct buffer_head **bh2,
 103                              struct inode *inode2,
 104                              int rename);
 105
 106 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
 107 /* An orphan dir name is an 8 byte value, printed as a hex string */
 108 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 109
 110 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 111                                    unsigned int flags)
 112 {
 113         int status;
 114         u64 blkno;
 115         struct inode *inode = NULL;
 116         struct dentry *ret;
 117         struct ocfs2_inode_info *oi;
 118
 119         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 120                            dentry->d_name.name,
 121                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 122
 123         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 124                 ret = ERR_PTR(-ENAMETOOLONG);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 129         if (status < 0) {
 130                 if (status != -ENOENT)
 131                         mlog_errno(status);
 132                 ret = ERR_PTR(status);
 133                 goto bail;
 134         }
 135
 136         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 137                                             dentry->d_name.len, &blkno);
 138         if (status < 0)
 139                 goto bail_add;
 140
 141         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 142         if (IS_ERR(inode)) {
 143                 ret = ERR_PTR(-EACCES);
 144                 goto bail_unlock;
 145         }
 146
 147         oi = OCFS2_I(inode);
 148         /* Clear any orphaned state... If we were able to look up the
 149          * inode from a directory, it certainly can't be orphaned. We
 150          * might have the bad state from a node which intended to
 151          * orphan this inode but crashed before it could commit the
 152          * unlink. */
 153         spin_lock(&oi->ip_lock);
 154         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 155         spin_unlock(&oi->ip_lock);
 156
 157 bail_add:
 158         ret = d_splice_alias(inode, dentry);
 159
 160         if (inode) {
 161                 /*
 162                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 163                  * dentry, it will d_move() it on top of ourse. The
 164                  * return value will indicate this however, so in
 165                  * those cases, we switch them around for the locking
 166                  * code.
 167                  *
 168                  * NOTE: This dentry already has ->d_op set from
 169                  * ocfs2_get_parent() and ocfs2_get_dentry()
 170                  */
 171                 if (!IS_ERR_OR_NULL(ret))
 172                         dentry = ret;
 173
 174                 status = ocfs2_dentry_attach_lock(dentry, inode,
 175                                                   OCFS2_I(dir)->ip_blkno);
 176                 if (status) {
 177                         mlog_errno(status);
 178                         ret = ERR_PTR(status);
 179                         goto bail_unlock;
 180                 }
 181         } else
 182                 ocfs2_dentry_attach_gen(dentry);
 183
 184 bail_unlock:
 185         /* Don't drop the cluster lock until *after* the d_add --
 186          * unlink on another node will message us to remove that
 187          * dentry under this lock so otherwise we can race this with
 188          * the downconvert thread and have a stale dentry. */
 189         ocfs2_inode_unlock(dir, 0);
 190
 191 bail:
 192
 193         trace_ocfs2_lookup_ret(ret);
 194
 195         return ret;
 196 }
 197
 198 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 199 {
 200         struct inode *inode;
 201         int status;
 202
 203         inode = new_inode(dir->i_sb);
 204         if (!inode) {
 205                 mlog(ML_ERROR, "new_inode failed!\n");
 206                 return ERR_PTR(-ENOMEM);
 207         }
 208
 209         /* populate as many fields early on as possible - many of
 210          * these are used by the support functions here and in
 211          * callers. */
 212         if (S_ISDIR(mode))
 213                 set_nlink(inode, 2);
 214         inode_init_owner(inode, dir, mode);
 215         status = dquot_initialize(inode);
 216         if (status)
 217                 return ERR_PTR(status);
 218
 219         return inode;
 220 }
 221
 222 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 223                 struct dentry *dentry, struct inode *inode)
 224 {
 225         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 226
 227         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 228         ocfs2_lock_res_free(&dl->dl_lockres);
 229         BUG_ON(dl->dl_count != 1);
 230         spin_lock(&dentry_attach_lock);
 231         dentry->d_fsdata = NULL;
 232         spin_unlock(&dentry_attach_lock);
 233         kfree(dl);
 234         iput(inode);
 235 }
 236
 237 static int ocfs2_mknod(struct inode *dir,
 238                        struct dentry *dentry,
 239                        umode_t mode,
 240                        dev_t dev)
 241 {
 242         int status = 0;
 243         struct buffer_head *parent_fe_bh = NULL;
 244         handle_t *handle = NULL;
 245         struct ocfs2_super *osb;
 246         struct ocfs2_dinode *dirfe;
 247         struct ocfs2_dinode *fe = NULL;
 248         struct buffer_head *new_fe_bh = NULL;
 249         struct inode *inode = NULL;
 250         struct ocfs2_alloc_context *inode_ac = NULL;
 251         struct ocfs2_alloc_context *data_ac = NULL;
 252         struct ocfs2_alloc_context *meta_ac = NULL;
 253         int want_clusters = 0;
 254         int want_meta = 0;
 255         int xattr_credits = 0;
 256         struct ocfs2_security_xattr_info si = {
 257                 .enable = 1,
 258         };
 259         int did_quota_inode = 0;
 260         struct ocfs2_dir_lookup_result lookup = { NULL, };
 261         sigset_t oldset;
 262         int did_block_signals = 0;
 263         struct ocfs2_dentry_lock *dl = NULL;
 264
 265         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 266                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 267                           (unsigned long)dev, mode);
 268
 269         status = dquot_initialize(dir);
 270         if (status) {
 271                 mlog_errno(status);
 272                 return status;
 273         }
 274
 275         /* get our super block */
 276         osb = OCFS2_SB(dir->i_sb);
 277
 278         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 279         if (status < 0) {
 280                 if (status != -ENOENT)
 281                         mlog_errno(status);
 282                 return status;
 283         }
 284
 285         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 286                 status = -EMLINK;
 287                 goto leave;
 288         }
 289
 290         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 291         if (!ocfs2_read_links_count(dirfe)) {
 292                 /* can't make a file in a deleted directory. */
 293                 status = -ENOENT;
 294                 goto leave;
 295         }
 296
 297         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 298                                            dentry->d_name.len);
 299         if (status)
 300                 goto leave;
 301
 302         /* get a spot inside the dir. */
 303         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 304                                               dentry->d_name.name,
 305                                               dentry->d_name.len, &lookup);
 306         if (status < 0) {
 307                 mlog_errno(status);
 308                 goto leave;
 309         }
 310
 311         /* reserve an inode spot */
 312         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 313         if (status < 0) {
 314                 if (status != -ENOSPC)
 315                         mlog_errno(status);
 316                 goto leave;
 317         }
 318
 319         inode = ocfs2_get_init_inode(dir, mode);
 320         if (IS_ERR(inode)) {
 321                 status = PTR_ERR(inode);
 322                 inode = NULL;
 323                 mlog_errno(status);
 324                 goto leave;
 325         }
 326
 327         /* get security xattr */
 328         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 329         if (status) {
 330                 if (status == -EOPNOTSUPP)
 331                         si.enable = 0;
 332                 else {
 333                         mlog_errno(status);
 334                         goto leave;
 335                 }
 336         }
 337
 338         /* calculate meta data/clusters for setting security and acl xattr */
 339         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 340                                        &si, &want_clusters,
 341                                        &xattr_credits, &want_meta);
 342         if (status < 0) {
 343                 mlog_errno(status);
 344                 goto leave;
 345         }
 346
 347         /* Reserve a cluster if creating an extent based directory. */
 348         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 349                 want_clusters += 1;
 350
 351                 /* Dir indexing requires extra space as well */
 352                 if (ocfs2_supports_indexed_dirs(osb))
 353                         want_meta++;
 354         }
 355
 356         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 357         if (status < 0) {
 358                 if (status != -ENOSPC)
 359                         mlog_errno(status);
 360                 goto leave;
 361         }
 362
 363         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 364         if (status < 0) {
 365                 if (status != -ENOSPC)
 366                         mlog_errno(status);
 367                 goto leave;
 368         }
 369
 370         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 371                                                             S_ISDIR(mode),
 372                                                             xattr_credits));
 373         if (IS_ERR(handle)) {
 374                 status = PTR_ERR(handle);
 375                 handle = NULL;
 376                 mlog_errno(status);
 377                 goto leave;
 378         }
 379
 380         /* Starting to change things, restart is no longer possible. */
 381         ocfs2_block_signals(&oldset);
 382         did_block_signals = 1;
 383
 384         status = dquot_alloc_inode(inode);
 385         if (status)
 386                 goto leave;
 387         did_quota_inode = 1;
 388
 389         /* do the real work now. */
 390         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 391                                     &new_fe_bh, parent_fe_bh, handle,
 392                                     inode_ac);
 393         if (status < 0) {
 394                 mlog_errno(status);
 395                 goto leave;
 396         }
 397
 398         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
 399         if (S_ISDIR(mode)) {
 400                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 401                                             new_fe_bh, data_ac, meta_ac);
 402                 if (status < 0) {
 403                         mlog_errno(status);
 404                         goto leave;
 405                 }
 406
 407                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 408                                                  parent_fe_bh,
 409                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 410                 if (status < 0) {
 411                         mlog_errno(status);
 412                         goto leave;
 413                 }
 414                 ocfs2_add_links_count(dirfe, 1);
 415                 ocfs2_journal_dirty(handle, parent_fe_bh);
 416                 inc_nlink(dir);
 417         }
 418
 419         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 420                          meta_ac, data_ac);
 421
 422         if (status < 0) {
 423                 mlog_errno(status);
 424                 goto leave;
 425         }
 426
 427         if (si.enable) {
 428                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 429                                                  meta_ac, data_ac);
 430                 if (status < 0) {
 431                         mlog_errno(status);
 432                         goto leave;
 433                 }
 434         }
 435
 436         /*
 437          * Do this before adding the entry to the directory. We add
 438          * also set d_op after success so that ->d_iput() will cleanup
 439          * the dentry lock even if ocfs2_add_entry() fails below.
 440          */
 441         status = ocfs2_dentry_attach_lock(dentry, inode,
 442                                           OCFS2_I(dir)->ip_blkno);
 443         if (status) {
 444                 mlog_errno(status);
 445                 goto leave;
 446         }
 447
 448         dl = dentry->d_fsdata;
 449
 450         status = ocfs2_add_entry(handle, dentry, inode,
 451                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 452                                  &lookup);
 453         if (status < 0) {
 454                 mlog_errno(status);
 455                 goto leave;
 456         }
 457
 458         insert_inode_hash(inode);
 459         d_instantiate(dentry, inode);
 460         status = 0;
 461 leave:
 462         if (status < 0 && did_quota_inode)
 463                 dquot_free_inode(inode);
 464         if (handle) {
 465                 if (status < 0 && fe)
 466                         ocfs2_set_links_count(fe, 0);
 467                 ocfs2_commit_trans(osb, handle);
 468         }
 469
 470         ocfs2_inode_unlock(dir, 1);
 471         if (did_block_signals)
 472                 ocfs2_unblock_signals(&oldset);
 473
 474         brelse(new_fe_bh);
 475         brelse(parent_fe_bh);
 476         kfree(si.value);
 477
 478         ocfs2_free_dir_lookup_result(&lookup);
 479
 480         if (inode_ac)
 481                 ocfs2_free_alloc_context(inode_ac);
 482
 483         if (data_ac)
 484                 ocfs2_free_alloc_context(data_ac);
 485
 486         if (meta_ac)
 487                 ocfs2_free_alloc_context(meta_ac);
 488
 489         /*
 490          * We should call iput after the i_mutex of the bitmap been
 491          * unlocked in ocfs2_free_alloc_context, or the
 492          * ocfs2_delete_inode will mutex_lock again.
 493          */
 494         if ((status < 0) && inode) {
 495                 if (dl)
 496                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 497
 498                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 499                 clear_nlink(inode);
 500                 iput(inode);
 501         }
 502
 503         if (status)
 504                 mlog_errno(status);
 505
 506         return status;
 507 }
 508
 509 static int __ocfs2_mknod_locked(struct inode *dir,
 510                                 struct inode *inode,
 511                                 dev_t dev,
 512                                 struct buffer_head **new_fe_bh,
 513                                 struct buffer_head *parent_fe_bh,
 514                                 handle_t *handle,
 515                                 struct ocfs2_alloc_context *inode_ac,
 516                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 517 {
 518         int status = 0;
 519         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 520         struct ocfs2_dinode *fe = NULL;
 521         struct ocfs2_extent_list *fel;
 522         u16 feat;
 523         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 524         struct timespec64 ts;
 525
 526         *new_fe_bh = NULL;
 527
 528         /* populate as many fields early on as possible - many of
 529          * these are used by the support functions here and in
 530          * callers. */
 531         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 532         OCFS2_I(inode)->ip_blkno = fe_blkno;
 533         spin_lock(&osb->osb_lock);
 534         inode->i_generation = osb->s_next_generation++;
 535         spin_unlock(&osb->osb_lock);
 536
 537         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 538         if (!*new_fe_bh) {
 539                 status = -ENOMEM;
 540                 mlog_errno(status);
 541                 goto leave;
 542         }
 543         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 544
 545         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 546                                          *new_fe_bh,
 547                                          OCFS2_JOURNAL_ACCESS_CREATE);
 548         if (status < 0) {
 549                 mlog_errno(status);
 550                 goto leave;
 551         }
 552
 553         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 554         memset(fe, 0, osb->sb->s_blocksize);
 555
 556         fe->i_generation = cpu_to_le32(inode->i_generation);
 557         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 558         fe->i_blkno = cpu_to_le64(fe_blkno);
 559         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 560         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 561         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 562         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 563         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 564         fe->i_mode = cpu_to_le16(inode->i_mode);
 565         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 566                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 567
 568         ocfs2_set_links_count(fe, inode->i_nlink);
 569
 570         fe->i_last_eb_blk = 0;
 571         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 572         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 573         ktime_get_real_ts64(&ts);
 574         fe->i_atime = fe->i_ctime = fe->i_mtime =
 575                 cpu_to_le64(ts.tv_sec);
 576         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 577                 cpu_to_le32(ts.tv_nsec);
 578         fe->i_dtime = 0;
 579
 580         /*
 581          * If supported, directories start with inline data. If inline
 582          * isn't supported, but indexing is, we start them as indexed.
 583          */
 584         feat = le16_to_cpu(fe->i_dyn_features);
 585         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 586                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 587
 588                 fe->id2.i_data.id_count = cpu_to_le16(
 589                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 590         } else {
 591                 fel = &fe->id2.i_list;
 592                 fel->l_tree_depth = 0;
 593                 fel->l_next_free_rec = 0;
 594                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 595         }
 596
 597         ocfs2_journal_dirty(handle, *new_fe_bh);
 598
 599         ocfs2_populate_inode(inode, fe, 1);
 600         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 601         if (!ocfs2_mount_local(osb)) {
 602                 status = ocfs2_create_new_inode_locks(inode);
 603                 if (status < 0)
 604                         mlog_errno(status);
 605         }
 606
 607         oi->i_sync_tid = handle->h_transaction->t_tid;
 608         oi->i_datasync_tid = handle->h_transaction->t_tid;
 609
 610 leave:
 611         if (status < 0) {
 612                 if (*new_fe_bh) {
 613                         brelse(*new_fe_bh);
 614                         *new_fe_bh = NULL;
 615                 }
 616         }
 617
 618         if (status)
 619                 mlog_errno(status);
 620         return status;
 621 }
 622
 623 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 624                               struct inode *dir,
 625                               struct inode *inode,
 626                               dev_t dev,
 627                               struct buffer_head **new_fe_bh,
 628                               struct buffer_head *parent_fe_bh,
 629                               handle_t *handle,
 630                               struct ocfs2_alloc_context *inode_ac)
 631 {
 632         int status = 0;
 633         u64 suballoc_loc, fe_blkno = 0;
 634         u16 suballoc_bit;
 635
 636         *new_fe_bh = NULL;
 637
 638         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 639                                        inode_ac, &suballoc_loc,
 640                                        &suballoc_bit, &fe_blkno);
 641         if (status < 0) {
 642                 mlog_errno(status);
 643                 return status;
 644         }
 645
 646         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 647                                     parent_fe_bh, handle, inode_ac,
 648                                     fe_blkno, suballoc_loc, suballoc_bit);
 649 }
 650
 651 static int ocfs2_mkdir(struct inode *dir,
 652                        struct dentry *dentry,
 653                        umode_t mode)
 654 {
 655         int ret;
 656
 657         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 658                           OCFS2_I(dir)->ip_blkno, mode);
 659         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 660         if (ret)
 661                 mlog_errno(ret);
 662
 663         return ret;
 664 }
 665
 666 static int ocfs2_create(struct inode *dir,
 667                         struct dentry *dentry,
 668                         umode_t mode,
 669                         bool excl)
 670 {
 671         int ret;
 672
 673         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 674                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 675         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 676         if (ret)
 677                 mlog_errno(ret);
 678
 679         return ret;
 680 }
 681
 682 static int ocfs2_link(struct dentry *old_dentry,
 683                       struct inode *dir,
 684                       struct dentry *dentry)
 685 {
 686         handle_t *handle;
 687         struct inode *inode = d_inode(old_dentry);
 688         struct inode *old_dir = d_inode(old_dentry->d_parent);
 689         int err;
 690         struct buffer_head *fe_bh = NULL;
 691         struct buffer_head *old_dir_bh = NULL;
 692         struct buffer_head *parent_fe_bh = NULL;
 693         struct ocfs2_dinode *fe = NULL;
 694         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 695         struct ocfs2_dir_lookup_result lookup = { NULL, };
 696         sigset_t oldset;
 697         u64 old_de_ino;
 698
 699         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 700                          old_dentry->d_name.len, old_dentry->d_name.name,
 701                          dentry->d_name.len, dentry->d_name.name);
 702
 703         if (S_ISDIR(inode->i_mode))
 704                 return -EPERM;
 705
 706         err = dquot_initialize(dir);
 707         if (err) {
 708                 mlog_errno(err);
 709                 return err;
 710         }
 711
 712         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 713                         &parent_fe_bh, dir, 0);
 714         if (err < 0) {
 715                 if (err != -ENOENT)
 716                         mlog_errno(err);
 717                 return err;
 718         }
 719
 720         /* make sure both dirs have bhs
 721          * get an extra ref on old_dir_bh if old==new */
 722         if (!parent_fe_bh) {
 723                 if (old_dir_bh) {
 724                         parent_fe_bh = old_dir_bh;
 725                         get_bh(parent_fe_bh);
 726                 } else {
 727                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 728                         err = -EIO;
 729                         goto out;
 730                 }
 731         }
 732
 733         if (!dir->i_nlink) {
 734                 err = -ENOENT;
 735                 goto out;
 736         }
 737
 738         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 739                         old_dentry->d_name.len, &old_de_ino);
 740         if (err) {
 741                 err = -ENOENT;
 742                 goto out;
 743         }
 744
 745         /*
 746          * Check whether another node removed the source inode while we
 747          * were in the vfs.
 748          */
 749         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 750                 err = -ENOENT;
 751                 goto out;
 752         }
 753
 754         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 755                                         dentry->d_name.len);
 756         if (err)
 757                 goto out;
 758
 759         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 760                                            dentry->d_name.name,
 761                                            dentry->d_name.len, &lookup);
 762         if (err < 0) {
 763                 mlog_errno(err);
 764                 goto out;
 765         }
 766
 767         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 768         if (err < 0) {
 769                 if (err != -ENOENT)
 770                         mlog_errno(err);
 771                 goto out;
 772         }
 773
 774         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 775         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 776                 err = -EMLINK;
 777                 goto out_unlock_inode;
 778         }
 779
 780         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 781         if (IS_ERR(handle)) {
 782                 err = PTR_ERR(handle);
 783                 handle = NULL;
 784                 mlog_errno(err);
 785                 goto out_unlock_inode;
 786         }
 787
 788         /* Starting to change things, restart is no longer possible. */
 789         ocfs2_block_signals(&oldset);
 790
 791         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 792                                       OCFS2_JOURNAL_ACCESS_WRITE);
 793         if (err < 0) {
 794                 mlog_errno(err);
 795                 goto out_commit;
 796         }
 797
 798         inc_nlink(inode);
 799         inode->i_ctime = current_time(inode);
 800         ocfs2_set_links_count(fe, inode->i_nlink);
 801         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 802         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 803         ocfs2_journal_dirty(handle, fe_bh);
 804
 805         err = ocfs2_add_entry(handle, dentry, inode,
 806                               OCFS2_I(inode)->ip_blkno,
 807                               parent_fe_bh, &lookup);
 808         if (err) {
 809                 ocfs2_add_links_count(fe, -1);
 810                 drop_nlink(inode);
 811                 mlog_errno(err);
 812                 goto out_commit;
 813         }
 814
 815         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 816         if (err) {
 817                 mlog_errno(err);
 818                 goto out_commit;
 819         }
 820
 821         ihold(inode);
 822         d_instantiate(dentry, inode);
 823
 824 out_commit:
 825         ocfs2_commit_trans(osb, handle);
 826         ocfs2_unblock_signals(&oldset);
 827 out_unlock_inode:
 828         ocfs2_inode_unlock(inode, 1);
 829
 830 out:
 831         ocfs2_double_unlock(old_dir, dir);
 832
 833         brelse(fe_bh);
 834         brelse(parent_fe_bh);
 835         brelse(old_dir_bh);
 836
 837         ocfs2_free_dir_lookup_result(&lookup);
 838
 839         if (err)
 840                 mlog_errno(err);
 841
 842         return err;
 843 }
 844
 845 /*
 846  * Takes and drops an exclusive lock on the given dentry. This will
 847  * force other nodes to drop it.
 848  */
 849 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 850 {
 851         int ret;
 852
 853         ret = ocfs2_dentry_lock(dentry, 1);
 854         if (ret)
 855                 mlog_errno(ret);
 856         else
 857                 ocfs2_dentry_unlock(dentry, 1);
 858
 859         return ret;
 860 }
 861
 862 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 863 {
 864         if (S_ISDIR(inode->i_mode)) {
 865                 if (inode->i_nlink == 2)
 866                         return 1;
 867                 return 0;
 868         }
 869
 870         if (inode->i_nlink == 1)
 871                 return 1;
 872         return 0;
 873 }
 874
 875 static int ocfs2_unlink(struct inode *dir,
 876                         struct dentry *dentry)
 877 {
 878         int status;
 879         int child_locked = 0;
 880         bool is_unlinkable = false;
 881         struct inode *inode = d_inode(dentry);
 882         struct inode *orphan_dir = NULL;
 883         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 884         u64 blkno;
 885         struct ocfs2_dinode *fe = NULL;
 886         struct buffer_head *fe_bh = NULL;
 887         struct buffer_head *parent_node_bh = NULL;
 888         handle_t *handle = NULL;
 889         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 890         struct ocfs2_dir_lookup_result lookup = { NULL, };
 891         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 892
 893         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 894                            dentry->d_name.name,
 895                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 896                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 897
 898         status = dquot_initialize(dir);
 899         if (status) {
 900                 mlog_errno(status);
 901                 return status;
 902         }
 903
 904         BUG_ON(d_inode(dentry->d_parent) != dir);
 905
 906         if (inode == osb->root_inode)
 907                 return -EPERM;
 908
 909         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 910                                          OI_LS_PARENT);
 911         if (status < 0) {
 912                 if (status != -ENOENT)
 913                         mlog_errno(status);
 914                 return status;
 915         }
 916
 917         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 918                                           dentry->d_name.len, &blkno, dir,
 919                                           &lookup);
 920         if (status < 0) {
 921                 if (status != -ENOENT)
 922                         mlog_errno(status);
 923                 goto leave;
 924         }
 925
 926         if (OCFS2_I(inode)->ip_blkno != blkno) {
 927                 status = -ENOENT;
 928
 929                 trace_ocfs2_unlink_noent(
 930                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 931                                 (unsigned long long)blkno,
 932                                 OCFS2_I(inode)->ip_flags);
 933                 goto leave;
 934         }
 935
 936         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 937         if (status < 0) {
 938                 if (status != -ENOENT)
 939                         mlog_errno(status);
 940                 goto leave;
 941         }
 942         child_locked = 1;
 943
 944         if (S_ISDIR(inode->i_mode)) {
 945                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 946                         status = -ENOTEMPTY;
 947                         goto leave;
 948                 }
 949         }
 950
 951         status = ocfs2_remote_dentry_delete(dentry);
 952         if (status < 0) {
 953                 /* This remote delete should succeed under all normal
 954                  * circumstances. */
 955                 mlog_errno(status);
 956                 goto leave;
 957         }
 958
 959         if (ocfs2_inode_is_unlinkable(inode)) {
 960                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 961                                                   OCFS2_I(inode)->ip_blkno,
 962                                                   orphan_name, &orphan_insert,
 963                                                   false);
 964                 if (status < 0) {
 965                         mlog_errno(status);
 966                         goto leave;
 967                 }
 968                 is_unlinkable = true;
 969         }
 970
 971         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 972         if (IS_ERR(handle)) {
 973                 status = PTR_ERR(handle);
 974                 handle = NULL;
 975                 mlog_errno(status);
 976                 goto leave;
 977         }
 978
 979         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 980                                          OCFS2_JOURNAL_ACCESS_WRITE);
 981         if (status < 0) {
 982                 mlog_errno(status);
 983                 goto leave;
 984         }
 985
 986         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 987
 988         /* delete the name from the parent dir */
 989         status = ocfs2_delete_entry(handle, dir, &lookup);
 990         if (status < 0) {
 991                 mlog_errno(status);
 992                 goto leave;
 993         }
 994
 995         if (S_ISDIR(inode->i_mode))
 996                 drop_nlink(inode);
 997         drop_nlink(inode);
 998         ocfs2_set_links_count(fe, inode->i_nlink);
 999         ocfs2_journal_dirty(handle, fe_bh);
1000
1001         dir->i_ctime = dir->i_mtime = current_time(dir);
1002         if (S_ISDIR(inode->i_mode))
1003                 drop_nlink(dir);
1004
1005         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1006         if (status < 0) {
1007                 mlog_errno(status);
1008                 if (S_ISDIR(inode->i_mode))
1009                         inc_nlink(dir);
1010                 goto leave;
1011         }
1012
1013         if (is_unlinkable) {
1014                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1015                                 orphan_name, &orphan_insert, orphan_dir, false);
1016                 if (status < 0)
1017                         mlog_errno(status);
1018         }
1019
1020 leave:
1021         if (handle)
1022                 ocfs2_commit_trans(osb, handle);
1023
1024         if (orphan_dir) {
1025                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1026                 ocfs2_inode_unlock(orphan_dir, 1);
1027                 inode_unlock(orphan_dir);
1028                 iput(orphan_dir);
1029         }
1030
1031         if (child_locked)
1032                 ocfs2_inode_unlock(inode, 1);
1033
1034         ocfs2_inode_unlock(dir, 1);
1035
1036         brelse(fe_bh);
1037         brelse(parent_node_bh);
1038
1039         ocfs2_free_dir_lookup_result(&orphan_insert);
1040         ocfs2_free_dir_lookup_result(&lookup);
1041
1042         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1043                 mlog_errno(status);
1044
1045         return status;
1046 }
1047
1048 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1049                 u64 src_inode_no, u64 dest_inode_no)
1050 {
1051         int ret = 0, i = 0;
1052         u64 parent_inode_no = 0;
1053         u64 child_inode_no = src_inode_no;
1054         struct inode *child_inode;
1055
1056 #define MAX_LOOKUP_TIMES 32
1057         while (1) {
1058                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1059                 if (IS_ERR(child_inode)) {
1060                         ret = PTR_ERR(child_inode);
1061                         break;
1062                 }
1063
1064                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1065                 if (ret < 0) {
1066                         iput(child_inode);
1067                         if (ret != -ENOENT)
1068                                 mlog_errno(ret);
1069                         break;
1070                 }
1071
1072                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1073                                 &parent_inode_no);
1074                 ocfs2_inode_unlock(child_inode, 0);
1075                 iput(child_inode);
1076                 if (ret < 0) {
1077                         ret = -ENOENT;
1078                         break;
1079                 }
1080
1081                 if (parent_inode_no == dest_inode_no) {
1082                         ret = 1;
1083                         break;
1084                 }
1085
1086                 if (parent_inode_no == osb->root_inode->i_ino) {
1087                         ret = 0;
1088                         break;
1089                 }
1090
1091                 child_inode_no = parent_inode_no;
1092
1093                 if (++i >= MAX_LOOKUP_TIMES) {
1094                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1095                                         "may have nested directories, "
1096                                         "src inode: %llu, dest inode: %llu.\n",
1097                                         (unsigned long long)src_inode_no,
1098                                         (unsigned long long)dest_inode_no);
1099                         ret = 0;
1100                         break;
1101                 }
1102         }
1103
1104         return ret;
1105 }
1106
1107 /*
1108  * The only place this should be used is rename and link!
1109  * if they have the same id, then the 1st one is the only one locked.
1110  */
1111 static int ocfs2_double_lock(struct ocfs2_super *osb,
1112                              struct buffer_head **bh1,
1113                              struct inode *inode1,
1114                              struct buffer_head **bh2,
1115                              struct inode *inode2,
1116                              int rename)
1117 {
1118         int status;
1119         int inode1_is_ancestor, inode2_is_ancestor;
1120         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1121         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1122
1123         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1124                                 (unsigned long long)oi2->ip_blkno);
1125
1126         if (*bh1)
1127                 *bh1 = NULL;
1128         if (*bh2)
1129                 *bh2 = NULL;
1130
1131         /* we always want to lock the one with the lower lockid first.
1132          * and if they are nested, we lock ancestor first */
1133         if (oi1->ip_blkno != oi2->ip_blkno) {
1134                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1135                                 oi1->ip_blkno);
1136                 if (inode1_is_ancestor < 0) {
1137                         status = inode1_is_ancestor;
1138                         goto bail;
1139                 }
1140
1141                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1142                                 oi2->ip_blkno);
1143                 if (inode2_is_ancestor < 0) {
1144                         status = inode2_is_ancestor;
1145                         goto bail;
1146                 }
1147
1148                 if ((inode1_is_ancestor == 1) ||
1149                                 (oi1->ip_blkno < oi2->ip_blkno &&
1150                                 inode2_is_ancestor == 0)) {
1151                         /* switch id1 and id2 around */
1152                         swap(bh2, bh1);
1153                         swap(inode2, inode1);
1154                 }
1155                 /* lock id2 */
1156                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1157                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1158                 if (status < 0) {
1159                         if (status != -ENOENT)
1160                                 mlog_errno(status);
1161                         goto bail;
1162                 }
1163         }
1164
1165         /* lock id1 */
1166         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1167                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1168         if (status < 0) {
1169                 /*
1170                  * An error return must mean that no cluster locks
1171                  * were held on function exit.
1172                  */
1173                 if (oi1->ip_blkno != oi2->ip_blkno) {
1174                         ocfs2_inode_unlock(inode2, 1);
1175                         brelse(*bh2);
1176                         *bh2 = NULL;
1177                 }
1178
1179                 if (status != -ENOENT)
1180                         mlog_errno(status);
1181         }
1182
1183         trace_ocfs2_double_lock_end(
1184                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1185                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1186
1187 bail:
1188         if (status)
1189                 mlog_errno(status);
1190         return status;
1191 }
1192
1193 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1194 {
1195         ocfs2_inode_unlock(inode1, 1);
1196
1197         if (inode1 != inode2)
1198                 ocfs2_inode_unlock(inode2, 1);
1199 }
1200
1201 static int ocfs2_rename(struct inode *old_dir,
1202                         struct dentry *old_dentry,
1203                         struct inode *new_dir,
1204                         struct dentry *new_dentry,
1205                         unsigned int flags)
1206 {
1207         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1208         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1209         struct inode *old_inode = d_inode(old_dentry);
1210         struct inode *new_inode = d_inode(new_dentry);
1211         struct inode *orphan_dir = NULL;
1212         struct ocfs2_dinode *newfe = NULL;
1213         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1214         struct buffer_head *newfe_bh = NULL;
1215         struct buffer_head *old_inode_bh = NULL;
1216         struct ocfs2_super *osb = NULL;
1217         u64 newfe_blkno, old_de_ino;
1218         handle_t *handle = NULL;
1219         struct buffer_head *old_dir_bh = NULL;
1220         struct buffer_head *new_dir_bh = NULL;
1221         u32 old_dir_nlink = old_dir->i_nlink;
1222         struct ocfs2_dinode *old_di;
1223         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1224         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1225         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1226         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1227         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1228         bool should_add_orphan = false;
1229
1230         if (flags)
1231                 return -EINVAL;
1232
1233         /* At some point it might be nice to break this function up a
1234          * bit. */
1235
1236         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1237                            old_dentry->d_name.len, old_dentry->d_name.name,
1238                            new_dentry->d_name.len, new_dentry->d_name.name);
1239
1240         status = dquot_initialize(old_dir);
1241         if (status) {
1242                 mlog_errno(status);
1243                 goto bail;
1244         }
1245         status = dquot_initialize(new_dir);
1246         if (status) {
1247                 mlog_errno(status);
1248                 goto bail;
1249         }
1250
1251         osb = OCFS2_SB(old_dir->i_sb);
1252
1253         if (new_inode) {
1254                 if (!igrab(new_inode))
1255                         BUG();
1256         }
1257
1258         /* Assume a directory hierarchy thusly:
1259          * a/b/c
1260          * a/d
1261          * a,b,c, and d are all directories.
1262          *
1263          * from cwd of 'a' on both nodes:
1264          * node1: mv b/c d
1265          * node2: mv d   b/c
1266          *
1267          * And that's why, just like the VFS, we need a file system
1268          * rename lock. */
1269         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1270                 status = ocfs2_rename_lock(osb);
1271                 if (status < 0) {
1272                         mlog_errno(status);
1273                         goto bail;
1274                 }
1275                 rename_lock = 1;
1276
1277                 /* here we cannot guarantee the inodes haven't just been
1278                  * changed, so check if they are nested again */
1279                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1280                                 old_inode->i_ino);
1281                 if (status < 0) {
1282                         mlog_errno(status);
1283                         goto bail;
1284                 } else if (status == 1) {
1285                         status = -EPERM;
1286                         trace_ocfs2_rename_not_permitted(
1287                                         (unsigned long long)old_inode->i_ino,
1288                                         (unsigned long long)new_dir->i_ino);
1289                         goto bail;
1290                 }
1291         }
1292
1293         /* if old and new are the same, this'll just do one lock. */
1294         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1295                                    &new_dir_bh, new_dir, 1);
1296         if (status < 0) {
1297                 mlog_errno(status);
1298                 goto bail;
1299         }
1300         parents_locked = 1;
1301
1302         if (!new_dir->i_nlink) {
1303                 status = -EACCES;
1304                 goto bail;
1305         }
1306
1307         /* make sure both dirs have bhs
1308          * get an extra ref on old_dir_bh if old==new */
1309         if (!new_dir_bh) {
1310                 if (old_dir_bh) {
1311                         new_dir_bh = old_dir_bh;
1312                         get_bh(new_dir_bh);
1313                 } else {
1314                         mlog(ML_ERROR, "no old_dir_bh!\n");
1315                         status = -EIO;
1316                         goto bail;
1317                 }
1318         }
1319
1320         /*
1321          * Aside from allowing a meta data update, the locking here
1322          * also ensures that the downconvert thread on other nodes
1323          * won't have to concurrently downconvert the inode and the
1324          * dentry locks.
1325          */
1326         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1327                                          OI_LS_PARENT);
1328         if (status < 0) {
1329                 if (status != -ENOENT)
1330                         mlog_errno(status);
1331                 goto bail;
1332         }
1333         old_child_locked = 1;
1334
1335         status = ocfs2_remote_dentry_delete(old_dentry);
1336         if (status < 0) {
1337                 mlog_errno(status);
1338                 goto bail;
1339         }
1340
1341         if (S_ISDIR(old_inode->i_mode)) {
1342                 u64 old_inode_parent;
1343
1344                 update_dot_dot = 1;
1345                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1346                                                   old_inode,
1347                                                   &old_inode_dot_dot_res);
1348                 if (status) {
1349                         status = -EIO;
1350                         goto bail;
1351                 }
1352
1353                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1354                         status = -EIO;
1355                         goto bail;
1356                 }
1357
1358                 if (!new_inode && new_dir != old_dir &&
1359                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1360                         status = -EMLINK;
1361                         goto bail;
1362                 }
1363         }
1364
1365         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1366                                             old_dentry->d_name.len,
1367                                             &old_de_ino);
1368         if (status) {
1369                 status = -ENOENT;
1370                 goto bail;
1371         }
1372
1373         /*
1374          *  Check for inode number is _not_ due to possible IO errors.
1375          *  We might rmdir the source, keep it as pwd of some process
1376          *  and merrily kill the link to whatever was created under the
1377          *  same name. Goodbye sticky bit ;-<
1378          */
1379         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1380                 status = -ENOENT;
1381                 goto bail;
1382         }
1383
1384         /* check if the target already exists (in which case we need
1385          * to delete it */
1386         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1387                                           new_dentry->d_name.len,
1388                                           &newfe_blkno, new_dir,
1389                                           &target_lookup_res);
1390         /* The only error we allow here is -ENOENT because the new
1391          * file not existing is perfectly valid. */
1392         if ((status < 0) && (status != -ENOENT)) {
1393                 /* If we cannot find the file specified we should just */
1394                 /* return the error... */
1395                 mlog_errno(status);
1396                 goto bail;
1397         }
1398         if (status == 0)
1399                 target_exists = 1;
1400
1401         if (!target_exists && new_inode) {
1402                 /*
1403                  * Target was unlinked by another node while we were
1404                  * waiting to get to ocfs2_rename(). There isn't
1405                  * anything we can do here to help the situation, so
1406                  * bubble up the appropriate error.
1407                  */
1408                 status = -ENOENT;
1409                 goto bail;
1410         }
1411
1412         /* In case we need to overwrite an existing file, we blow it
1413          * away first */
1414         if (target_exists) {
1415                 /* VFS didn't think there existed an inode here, but
1416                  * someone else in the cluster must have raced our
1417                  * rename to create one. Today we error cleanly, in
1418                  * the future we should consider calling iget to build
1419                  * a new struct inode for this entry. */
1420                 if (!new_inode) {
1421                         status = -EACCES;
1422
1423                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1424                                                 new_dentry->d_name.name);
1425                         goto bail;
1426                 }
1427
1428                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1429                         status = -EACCES;
1430
1431                         trace_ocfs2_rename_disagree(
1432                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1433                              (unsigned long long)newfe_blkno,
1434                              OCFS2_I(new_inode)->ip_flags);
1435                         goto bail;
1436                 }
1437
1438                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1439                 if (status < 0) {
1440                         if (status != -ENOENT)
1441                                 mlog_errno(status);
1442                         goto bail;
1443                 }
1444                 new_child_locked = 1;
1445
1446                 status = ocfs2_remote_dentry_delete(new_dentry);
1447                 if (status < 0) {
1448                         mlog_errno(status);
1449                         goto bail;
1450                 }
1451
1452                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1453
1454                 trace_ocfs2_rename_over_existing(
1455                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1456                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1457
1458                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1459                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1460                                                 OCFS2_I(new_inode)->ip_blkno,
1461                                                 orphan_name, &orphan_insert,
1462                                                 false);
1463                         if (status < 0) {
1464                                 mlog_errno(status);
1465                                 goto bail;
1466                         }
1467                         should_add_orphan = true;
1468                 }
1469         } else {
1470                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1471
1472                 status = ocfs2_check_dir_for_entry(new_dir,
1473                                                    new_dentry->d_name.name,
1474                                                    new_dentry->d_name.len);
1475                 if (status)
1476                         goto bail;
1477
1478                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1479                                                       new_dentry->d_name.name,
1480                                                       new_dentry->d_name.len,
1481                                                       &target_insert);
1482                 if (status < 0) {
1483                         mlog_errno(status);
1484                         goto bail;
1485                 }
1486         }
1487
1488         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1489         if (IS_ERR(handle)) {
1490                 status = PTR_ERR(handle);
1491                 handle = NULL;
1492                 mlog_errno(status);
1493                 goto bail;
1494         }
1495
1496         if (target_exists) {
1497                 if (S_ISDIR(new_inode->i_mode)) {
1498                         if (new_inode->i_nlink != 2 ||
1499                             !ocfs2_empty_dir(new_inode)) {
1500                                 status = -ENOTEMPTY;
1501                                 goto bail;
1502                         }
1503                 }
1504                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1505                                                  newfe_bh,
1506                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1507                 if (status < 0) {
1508                         mlog_errno(status);
1509                         goto bail;
1510                 }
1511
1512                 /* change the dirent to point to the correct inode */
1513                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1514                                             old_inode);
1515                 if (status < 0) {
1516                         mlog_errno(status);
1517                         goto bail;
1518                 }
1519                 new_dir->i_version++;
1520
1521                 if (S_ISDIR(new_inode->i_mode))
1522                         ocfs2_set_links_count(newfe, 0);
1523                 else
1524                         ocfs2_add_links_count(newfe, -1);
1525                 ocfs2_journal_dirty(handle, newfe_bh);
1526                 if (should_add_orphan) {
1527                         status = ocfs2_orphan_add(osb, handle, new_inode,
1528                                         newfe_bh, orphan_name,
1529                                         &orphan_insert, orphan_dir, false);
1530                         if (status < 0) {
1531                                 mlog_errno(status);
1532                                 goto bail;
1533                         }
1534                 }
1535         } else {
1536                 /* if the name was not found in new_dir, add it now */
1537                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1538                                          OCFS2_I(old_inode)->ip_blkno,
1539                                          new_dir_bh, &target_insert);
1540         }
1541
1542         old_inode->i_ctime = current_time(old_inode);
1543         mark_inode_dirty(old_inode);
1544
1545         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1546                                          old_inode_bh,
1547                                          OCFS2_JOURNAL_ACCESS_WRITE);
1548         if (status >= 0) {
1549                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1550
1551                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1552                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1553                 ocfs2_journal_dirty(handle, old_inode_bh);
1554         } else
1555                 mlog_errno(status);
1556
1557         /*
1558          * Now that the name has been added to new_dir, remove the old name.
1559          *
1560          * We don't keep any directory entry context around until now
1561          * because the insert might have changed the type of directory
1562          * we're dealing with.
1563          */
1564         status = ocfs2_find_entry(old_dentry->d_name.name,
1565                                   old_dentry->d_name.len, old_dir,
1566                                   &old_entry_lookup);
1567         if (status) {
1568                 if (!is_journal_aborted(osb->journal->j_journal)) {
1569                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1570                                         "is not deleted.",
1571                                         new_dentry->d_name.len, new_dentry->d_name.name,
1572                                         old_dentry->d_name.len, old_dentry->d_name.name);
1573                 }
1574                 goto bail;
1575         }
1576
1577         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1578         if (status < 0) {
1579                 mlog_errno(status);
1580                 if (!is_journal_aborted(osb->journal->j_journal)) {
1581                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1582                                         "is not deleted.",
1583                                         new_dentry->d_name.len, new_dentry->d_name.name,
1584                                         old_dentry->d_name.len, old_dentry->d_name.name);
1585                 }
1586                 goto bail;
1587         }
1588
1589         if (new_inode) {
1590                 drop_nlink(new_inode);
1591                 new_inode->i_ctime = current_time(new_inode);
1592         }
1593         old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1594
1595         if (update_dot_dot) {
1596                 status = ocfs2_update_entry(old_inode, handle,
1597                                             &old_inode_dot_dot_res, new_dir);
1598                 drop_nlink(old_dir);
1599                 if (new_inode) {
1600                         drop_nlink(new_inode);
1601                 } else {
1602                         inc_nlink(new_dir);
1603                         mark_inode_dirty(new_dir);
1604                 }
1605         }
1606         mark_inode_dirty(old_dir);
1607         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1608         if (new_inode) {
1609                 mark_inode_dirty(new_inode);
1610                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1611         }
1612
1613         if (old_dir != new_dir) {
1614                 /* Keep the same times on both directories.*/
1615                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1616
1617                 /*
1618                  * This will also pick up the i_nlink change from the
1619                  * block above.
1620                  */
1621                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1622         }
1623
1624         if (old_dir_nlink != old_dir->i_nlink) {
1625                 if (!old_dir_bh) {
1626                         mlog(ML_ERROR, "need to change nlink for old dir "
1627                              "%llu from %d to %d but bh is NULL!\n",
1628                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1629                              (int)old_dir_nlink, old_dir->i_nlink);
1630                 } else {
1631                         struct ocfs2_dinode *fe;
1632                         status = ocfs2_journal_access_di(handle,
1633                                                          INODE_CACHE(old_dir),
1634                                                          old_dir_bh,
1635                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1636                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1637                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1638                         ocfs2_journal_dirty(handle, old_dir_bh);
1639                 }
1640         }
1641         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1642         status = 0;
1643 bail:
1644         if (handle)
1645                 ocfs2_commit_trans(osb, handle);
1646
1647         if (orphan_dir) {
1648                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1649                 ocfs2_inode_unlock(orphan_dir, 1);
1650                 inode_unlock(orphan_dir);
1651                 iput(orphan_dir);
1652         }
1653
1654         if (new_child_locked)
1655                 ocfs2_inode_unlock(new_inode, 1);
1656
1657         if (old_child_locked)
1658                 ocfs2_inode_unlock(old_inode, 1);
1659
1660         if (parents_locked)
1661                 ocfs2_double_unlock(old_dir, new_dir);
1662
1663         if (rename_lock)
1664                 ocfs2_rename_unlock(osb);
1665
1666         if (new_inode)
1667                 sync_mapping_buffers(old_inode->i_mapping);
1668
1669         iput(new_inode);
1670
1671         ocfs2_free_dir_lookup_result(&target_lookup_res);
1672         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1673         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1674         ocfs2_free_dir_lookup_result(&orphan_insert);
1675         ocfs2_free_dir_lookup_result(&target_insert);
1676
1677         brelse(newfe_bh);
1678         brelse(old_inode_bh);
1679         brelse(old_dir_bh);
1680         brelse(new_dir_bh);
1681
1682         if (status)
1683                 mlog_errno(status);
1684
1685         return status;
1686 }
1687
1688 /*
1689  * we expect i_size = strlen(symname). Copy symname into the file
1690  * data, including the null terminator.
1691  */
1692 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1693                                      handle_t *handle,
1694                                      struct inode *inode,
1695                                      const char *symname)
1696 {
1697         struct buffer_head **bhs = NULL;
1698         const char *c;
1699         struct super_block *sb = osb->sb;
1700         u64 p_blkno, p_blocks;
1701         int virtual, blocks, status, i, bytes_left;
1702
1703         bytes_left = i_size_read(inode) + 1;
1704         /* we can't trust i_blocks because we're actually going to
1705          * write i_size + 1 bytes. */
1706         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1707
1708         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1709                                         i_size_read(inode), blocks);
1710
1711         /* Sanity check -- make sure we're going to fit. */
1712         if (bytes_left >
1713             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1714                 status = -EIO;
1715                 mlog_errno(status);
1716                 goto bail;
1717         }
1718
1719         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1720         if (!bhs) {
1721                 status = -ENOMEM;
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1727                                              NULL);
1728         if (status < 0) {
1729                 mlog_errno(status);
1730                 goto bail;
1731         }
1732
1733         /* links can never be larger than one cluster so we know this
1734          * is all going to be contiguous, but do a sanity check
1735          * anyway. */
1736         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1737                 status = -EIO;
1738                 mlog_errno(status);
1739                 goto bail;
1740         }
1741
1742         virtual = 0;
1743         while(bytes_left > 0) {
1744                 c = &symname[virtual * sb->s_blocksize];
1745
1746                 bhs[virtual] = sb_getblk(sb, p_blkno);
1747                 if (!bhs[virtual]) {
1748                         status = -ENOMEM;
1749                         mlog_errno(status);
1750                         goto bail;
1751                 }
1752                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1753                                               bhs[virtual]);
1754
1755                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1756                                               bhs[virtual],
1757                                               OCFS2_JOURNAL_ACCESS_CREATE);
1758                 if (status < 0) {
1759                         mlog_errno(status);
1760                         goto bail;
1761                 }
1762
1763                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1764
1765                 memcpy(bhs[virtual]->b_data, c,
1766                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1767                        bytes_left);
1768
1769                 ocfs2_journal_dirty(handle, bhs[virtual]);
1770
1771                 virtual++;
1772                 p_blkno++;
1773                 bytes_left -= sb->s_blocksize;
1774         }
1775
1776         status = 0;
1777 bail:
1778
1779         if (bhs) {
1780                 for(i = 0; i < blocks; i++)
1781                         brelse(bhs[i]);
1782                 kfree(bhs);
1783         }
1784
1785         if (status)
1786                 mlog_errno(status);
1787         return status;
1788 }
1789
1790 static int ocfs2_symlink(struct inode *dir,
1791                          struct dentry *dentry,
1792                          const char *symname)
1793 {
1794         int status, l, credits;
1795         u64 newsize;
1796         struct ocfs2_super *osb = NULL;
1797         struct inode *inode = NULL;
1798         struct super_block *sb;
1799         struct buffer_head *new_fe_bh = NULL;
1800         struct buffer_head *parent_fe_bh = NULL;
1801         struct ocfs2_dinode *fe = NULL;
1802         struct ocfs2_dinode *dirfe;
1803         handle_t *handle = NULL;
1804         struct ocfs2_alloc_context *inode_ac = NULL;
1805         struct ocfs2_alloc_context *data_ac = NULL;
1806         struct ocfs2_alloc_context *xattr_ac = NULL;
1807         int want_clusters = 0;
1808         int xattr_credits = 0;
1809         struct ocfs2_security_xattr_info si = {
1810                 .enable = 1,
1811         };
1812         int did_quota = 0, did_quota_inode = 0;
1813         struct ocfs2_dir_lookup_result lookup = { NULL, };
1814         sigset_t oldset;
1815         int did_block_signals = 0;
1816         struct ocfs2_dentry_lock *dl = NULL;
1817
1818         trace_ocfs2_symlink_begin(dir, dentry, symname,
1819                                   dentry->d_name.len, dentry->d_name.name);
1820
1821         status = dquot_initialize(dir);
1822         if (status) {
1823                 mlog_errno(status);
1824                 goto bail;
1825         }
1826
1827         sb = dir->i_sb;
1828         osb = OCFS2_SB(sb);
1829
1830         l = strlen(symname) + 1;
1831
1832         credits = ocfs2_calc_symlink_credits(sb);
1833
1834         /* lock the parent directory */
1835         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1836         if (status < 0) {
1837                 if (status != -ENOENT)
1838                         mlog_errno(status);
1839                 return status;
1840         }
1841
1842         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1843         if (!ocfs2_read_links_count(dirfe)) {
1844                 /* can't make a file in a deleted directory. */
1845                 status = -ENOENT;
1846                 goto bail;
1847         }
1848
1849         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1850                                            dentry->d_name.len);
1851         if (status)
1852                 goto bail;
1853
1854         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1855                                               dentry->d_name.name,
1856                                               dentry->d_name.len, &lookup);
1857         if (status < 0) {
1858                 mlog_errno(status);
1859                 goto bail;
1860         }
1861
1862         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1863         if (status < 0) {
1864                 if (status != -ENOSPC)
1865                         mlog_errno(status);
1866                 goto bail;
1867         }
1868
1869         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1870         if (IS_ERR(inode)) {
1871                 status = PTR_ERR(inode);
1872                 inode = NULL;
1873                 mlog_errno(status);
1874                 goto bail;
1875         }
1876
1877         /* get security xattr */
1878         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1879         if (status) {
1880                 if (status == -EOPNOTSUPP)
1881                         si.enable = 0;
1882                 else {
1883                         mlog_errno(status);
1884                         goto bail;
1885                 }
1886         }
1887
1888         /* calculate meta data/clusters for setting security xattr */
1889         if (si.enable) {
1890                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1891                                                   &xattr_credits, &xattr_ac);
1892                 if (status < 0) {
1893                         mlog_errno(status);
1894                         goto bail;
1895                 }
1896         }
1897
1898         /* don't reserve bitmap space for fast symlinks. */
1899         if (l > ocfs2_fast_symlink_chars(sb))
1900                 want_clusters += 1;
1901
1902         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1903         if (status < 0) {
1904                 if (status != -ENOSPC)
1905                         mlog_errno(status);
1906                 goto bail;
1907         }
1908
1909         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1910         if (IS_ERR(handle)) {
1911                 status = PTR_ERR(handle);
1912                 handle = NULL;
1913                 mlog_errno(status);
1914                 goto bail;
1915         }
1916
1917         /* Starting to change things, restart is no longer possible. */
1918         ocfs2_block_signals(&oldset);
1919         did_block_signals = 1;
1920
1921         status = dquot_alloc_inode(inode);
1922         if (status)
1923                 goto bail;
1924         did_quota_inode = 1;
1925
1926         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1927                                    dentry->d_name.name,
1928                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1929                                    inode->i_mode);
1930
1931         status = ocfs2_mknod_locked(osb, dir, inode,
1932                                     0, &new_fe_bh, parent_fe_bh, handle,
1933                                     inode_ac);
1934         if (status < 0) {
1935                 mlog_errno(status);
1936                 goto bail;
1937         }
1938
1939         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1940         inode->i_rdev = 0;
1941         newsize = l - 1;
1942         inode->i_op = &ocfs2_symlink_inode_operations;
1943         inode_nohighmem(inode);
1944         if (l > ocfs2_fast_symlink_chars(sb)) {
1945                 u32 offset = 0;
1946
1947                 status = dquot_alloc_space_nodirty(inode,
1948                     ocfs2_clusters_to_bytes(osb->sb, 1));
1949                 if (status)
1950                         goto bail;
1951                 did_quota = 1;
1952                 inode->i_mapping->a_ops = &ocfs2_aops;
1953                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1954                                               new_fe_bh,
1955                                               handle, data_ac, NULL,
1956                                               NULL);
1957                 if (status < 0) {
1958                         if (status != -ENOSPC && status != -EINTR) {
1959                                 mlog(ML_ERROR,
1960                                      "Failed to extend file to %llu\n",
1961                                      (unsigned long long)newsize);
1962                                 mlog_errno(status);
1963                                 status = -ENOSPC;
1964                         }
1965                         goto bail;
1966                 }
1967                 i_size_write(inode, newsize);
1968                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1969         } else {
1970                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1971                 memcpy((char *) fe->id2.i_symlink, symname, l);
1972                 i_size_write(inode, newsize);
1973                 inode->i_blocks = 0;
1974         }
1975
1976         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1977         if (status < 0) {
1978                 mlog_errno(status);
1979                 goto bail;
1980         }
1981
1982         if (!ocfs2_inode_is_fast_symlink(inode)) {
1983                 status = ocfs2_create_symlink_data(osb, handle, inode,
1984                                                    symname);
1985                 if (status < 0) {
1986                         mlog_errno(status);
1987                         goto bail;
1988                 }
1989         }
1990
1991         if (si.enable) {
1992                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1993                                                  xattr_ac, data_ac);
1994                 if (status < 0) {
1995                         mlog_errno(status);
1996                         goto bail;
1997                 }
1998         }
1999
2000         /*
2001          * Do this before adding the entry to the directory. We add
2002          * also set d_op after success so that ->d_iput() will cleanup
2003          * the dentry lock even if ocfs2_add_entry() fails below.
2004          */
2005         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2006         if (status) {
2007                 mlog_errno(status);
2008                 goto bail;
2009         }
2010
2011         dl = dentry->d_fsdata;
2012
2013         status = ocfs2_add_entry(handle, dentry, inode,
2014                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2015                                  &lookup);
2016         if (status < 0) {
2017                 mlog_errno(status);
2018                 goto bail;
2019         }
2020
2021         insert_inode_hash(inode);
2022         d_instantiate(dentry, inode);
2023 bail:
2024         if (status < 0 && did_quota)
2025                 dquot_free_space_nodirty(inode,
2026                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2027         if (status < 0 && did_quota_inode)
2028                 dquot_free_inode(inode);
2029         if (handle) {
2030                 if (status < 0 && fe)
2031                         ocfs2_set_links_count(fe, 0);
2032                 ocfs2_commit_trans(osb, handle);
2033         }
2034
2035         ocfs2_inode_unlock(dir, 1);
2036         if (did_block_signals)
2037                 ocfs2_unblock_signals(&oldset);
2038
2039         brelse(new_fe_bh);
2040         brelse(parent_fe_bh);
2041         kfree(si.value);
2042         ocfs2_free_dir_lookup_result(&lookup);
2043         if (inode_ac)
2044                 ocfs2_free_alloc_context(inode_ac);
2045         if (data_ac)
2046                 ocfs2_free_alloc_context(data_ac);
2047         if (xattr_ac)
2048                 ocfs2_free_alloc_context(xattr_ac);
2049         if ((status < 0) && inode) {
2050                 if (dl)
2051                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2052
2053                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2054                 clear_nlink(inode);
2055                 iput(inode);
2056         }
2057
2058         if (status)
2059                 mlog_errno(status);
2060
2061         return status;
2062 }
2063
2064 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2065 {
2066         int status, namelen;
2067
2068         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2069                            (long long)blkno);
2070         if (namelen <= 0) {
2071                 if (namelen)
2072                         status = namelen;
2073                 else
2074                         status = -EINVAL;
2075                 mlog_errno(status);
2076                 goto bail;
2077         }
2078         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2079                 status = -EINVAL;
2080                 mlog_errno(status);
2081                 goto bail;
2082         }
2083
2084         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2085
2086         status = 0;
2087 bail:
2088         if (status < 0)
2089                 mlog_errno(status);
2090         return status;
2091 }
2092
2093 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2094                                         struct inode **ret_orphan_dir,
2095                                         struct buffer_head **ret_orphan_dir_bh)
2096 {
2097         struct inode *orphan_dir_inode;
2098         struct buffer_head *orphan_dir_bh = NULL;
2099         int ret = 0;
2100
2101         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2102                                                        ORPHAN_DIR_SYSTEM_INODE,
2103                                                        osb->slot_num);
2104         if (!orphan_dir_inode) {
2105                 ret = -ENOENT;
2106                 mlog_errno(ret);
2107                 return ret;
2108         }
2109
2110         inode_lock(orphan_dir_inode);
2111
2112         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2113         if (ret < 0) {
2114                 inode_unlock(orphan_dir_inode);
2115                 iput(orphan_dir_inode);
2116
2117                 mlog_errno(ret);
2118                 return ret;
2119         }
2120
2121         *ret_orphan_dir = orphan_dir_inode;
2122         *ret_orphan_dir_bh = orphan_dir_bh;
2123
2124         return 0;
2125 }
2126
2127 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2128                                       struct buffer_head *orphan_dir_bh,
2129                                       u64 blkno,
2130                                       char *name,
2131                                       struct ocfs2_dir_lookup_result *lookup,
2132                                       bool dio)
2133 {
2134         int ret;
2135         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2136         int namelen = dio ?
2137                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2138                         OCFS2_ORPHAN_NAMELEN;
2139
2140         if (dio) {
2141                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2142                                 OCFS2_DIO_ORPHAN_PREFIX);
2143                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2144                         ret = -EINVAL;
2145                         mlog_errno(ret);
2146                         return ret;
2147                 }
2148
2149                 ret = ocfs2_blkno_stringify(blkno,
2150                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2151         } else
2152                 ret = ocfs2_blkno_stringify(blkno, name);
2153         if (ret < 0) {
2154                 mlog_errno(ret);
2155                 return ret;
2156         }
2157
2158         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2159                                            orphan_dir_bh, name,
2160                                            namelen, lookup);
2161         if (ret < 0) {
2162                 mlog_errno(ret);
2163                 return ret;
2164         }
2165
2166         return 0;
2167 }
2168
2169 /**
2170  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2171  * insertion of an orphan.
2172  * @osb: ocfs2 file system
2173  * @ret_orphan_dir: Orphan dir inode - returned locked!
2174  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2175  * @lookup: dir lookup result, to be passed back into functions like
2176  *          ocfs2_orphan_add
2177  *
2178  * Returns zero on success and the ret_orphan_dir, name and lookup
2179  * fields will be populated.
2180  *
2181  * Returns non-zero on failure.
2182  */
2183 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2184                                     struct inode **ret_orphan_dir,
2185                                     u64 blkno,
2186                                     char *name,
2187                                     struct ocfs2_dir_lookup_result *lookup,
2188                                     bool dio)
2189 {
2190         struct inode *orphan_dir_inode = NULL;
2191         struct buffer_head *orphan_dir_bh = NULL;
2192         int ret = 0;
2193
2194         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2195                                            &orphan_dir_bh);
2196         if (ret < 0) {
2197                 mlog_errno(ret);
2198                 return ret;
2199         }
2200
2201         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2202                                          blkno, name, lookup, dio);
2203         if (ret < 0) {
2204                 mlog_errno(ret);
2205                 goto out;
2206         }
2207
2208         *ret_orphan_dir = orphan_dir_inode;
2209
2210 out:
2211         brelse(orphan_dir_bh);
2212
2213         if (ret) {
2214                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2215                 inode_unlock(orphan_dir_inode);
2216                 iput(orphan_dir_inode);
2217         }
2218
2219         if (ret)
2220                 mlog_errno(ret);
2221         return ret;
2222 }
2223
2224 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2225                             handle_t *handle,
2226                             struct inode *inode,
2227                             struct buffer_head *fe_bh,
2228                             char *name,
2229                             struct ocfs2_dir_lookup_result *lookup,
2230                             struct inode *orphan_dir_inode,
2231                             bool dio)
2232 {
2233         struct buffer_head *orphan_dir_bh = NULL;
2234         int status = 0;
2235         struct ocfs2_dinode *orphan_fe;
2236         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2237         int namelen = dio ?
2238                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2239                         OCFS2_ORPHAN_NAMELEN;
2240
2241         trace_ocfs2_orphan_add_begin(
2242                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2243
2244         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2245         if (status < 0) {
2246                 mlog_errno(status);
2247                 goto leave;
2248         }
2249
2250         status = ocfs2_journal_access_di(handle,
2251                                          INODE_CACHE(orphan_dir_inode),
2252                                          orphan_dir_bh,
2253                                          OCFS2_JOURNAL_ACCESS_WRITE);
2254         if (status < 0) {
2255                 mlog_errno(status);
2256                 goto leave;
2257         }
2258
2259         /*
2260          * We're going to journal the change of i_flags and i_orphaned_slot.
2261          * It's safe anyway, though some callers may duplicate the journaling.
2262          * Journaling within the func just make the logic look more
2263          * straightforward.
2264          */
2265         status = ocfs2_journal_access_di(handle,
2266                                          INODE_CACHE(inode),
2267                                          fe_bh,
2268                                          OCFS2_JOURNAL_ACCESS_WRITE);
2269         if (status < 0) {
2270                 mlog_errno(status);
2271                 goto leave;
2272         }
2273
2274         /* we're a cluster, and nlink can change on disk from
2275          * underneath us... */
2276         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2277         if (S_ISDIR(inode->i_mode))
2278                 ocfs2_add_links_count(orphan_fe, 1);
2279         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2280         ocfs2_journal_dirty(handle, orphan_dir_bh);
2281
2282         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2283                                    namelen, inode,
2284                                    OCFS2_I(inode)->ip_blkno,
2285                                    orphan_dir_bh, lookup);
2286         if (status < 0) {
2287                 mlog_errno(status);
2288                 goto rollback;
2289         }
2290
2291         if (dio) {
2292                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2293                  * slot.
2294                  */
2295                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2296                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2297         } else {
2298                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2299                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2300
2301                 /* Record which orphan dir our inode now resides
2302                  * in. delete_inode will use this to determine which orphan
2303                  * dir to lock. */
2304                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2305         }
2306
2307         ocfs2_journal_dirty(handle, fe_bh);
2308
2309         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2310                                    osb->slot_num);
2311
2312 rollback:
2313         if (status < 0) {
2314                 if (S_ISDIR(inode->i_mode))
2315                         ocfs2_add_links_count(orphan_fe, -1);
2316                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2317         }
2318
2319 leave:
2320         brelse(orphan_dir_bh);
2321
2322         return status;
2323 }
2324
2325 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2326 int ocfs2_orphan_del(struct ocfs2_super *osb,
2327                      handle_t *handle,
2328                      struct inode *orphan_dir_inode,
2329                      struct inode *inode,
2330                      struct buffer_head *orphan_dir_bh,
2331                      bool dio)
2332 {
2333         const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
2334         char name[namelen + 1];
2335         struct ocfs2_dinode *orphan_fe;
2336         int status = 0;
2337         struct ocfs2_dir_lookup_result lookup = { NULL, };
2338
2339         if (dio) {
2340                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2341                                 OCFS2_DIO_ORPHAN_PREFIX);
2342                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2343                         status = -EINVAL;
2344                         mlog_errno(status);
2345                         return status;
2346                 }
2347
2348                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2349                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2350         } else
2351                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2352         if (status < 0) {
2353                 mlog_errno(status);
2354                 goto leave;
2355         }
2356
2357         trace_ocfs2_orphan_del(
2358              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2359              name, strlen(name));
2360
2361         status = ocfs2_journal_access_di(handle,
2362                                          INODE_CACHE(orphan_dir_inode),
2363                                          orphan_dir_bh,
2364                                          OCFS2_JOURNAL_ACCESS_WRITE);
2365         if (status < 0) {
2366                 mlog_errno(status);
2367                 goto leave;
2368         }
2369
2370         /* find it's spot in the orphan directory */
2371         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2372                                   &lookup);
2373         if (status) {
2374                 mlog_errno(status);
2375                 goto leave;
2376         }
2377
2378         /* remove it from the orphan directory */
2379         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2380         if (status < 0) {
2381                 mlog_errno(status);
2382                 goto leave;
2383         }
2384
2385         /* do the i_nlink dance! :) */
2386         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2387         if (S_ISDIR(inode->i_mode))
2388                 ocfs2_add_links_count(orphan_fe, -1);
2389         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2390         ocfs2_journal_dirty(handle, orphan_dir_bh);
2391
2392 leave:
2393         ocfs2_free_dir_lookup_result(&lookup);
2394
2395         if (status)
2396                 mlog_errno(status);
2397         return status;
2398 }
2399
2400 /**
2401  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2402  * allocated file. This is different from the typical 'add to orphan dir'
2403  * operation in that the inode does not yet exist. This is a problem because
2404  * the orphan dir stringifies the inode block number to come up with it's
2405  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2406  * problem. This function works around it by calling deeper into the orphan
2407  * and suballoc code than other callers. Use this only by necessity.
2408  * @dir: The directory which this inode will ultimately wind up under - not the
2409  * orphan dir!
2410  * @dir_bh: buffer_head the @dir inode block
2411  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2412  * with the string to be used for orphan dirent. Pass back to the orphan dir
2413  * code.
2414  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2415  * dir code.
2416  * @ret_di_blkno: block number where the new inode will be allocated.
2417  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2418  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2419  *
2420  * Returns zero on success and the ret_orphan_dir, name and lookup
2421  * fields will be populated.
2422  *
2423  * Returns non-zero on failure.
2424  */
2425 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2426                                         struct buffer_head *dir_bh,
2427                                         char *orphan_name,
2428                                         struct inode **ret_orphan_dir,
2429                                         u64 *ret_di_blkno,
2430                                         struct ocfs2_dir_lookup_result *orphan_insert,
2431                                         struct ocfs2_alloc_context **ret_inode_ac)
2432 {
2433         int ret;
2434         u64 di_blkno;
2435         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2436         struct inode *orphan_dir = NULL;
2437         struct buffer_head *orphan_dir_bh = NULL;
2438         struct ocfs2_alloc_context *inode_ac = NULL;
2439
2440         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2441         if (ret < 0) {
2442                 mlog_errno(ret);
2443                 return ret;
2444         }
2445
2446         /* reserve an inode spot */
2447         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2448         if (ret < 0) {
2449                 if (ret != -ENOSPC)
2450                         mlog_errno(ret);
2451                 goto out;
2452         }
2453
2454         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2455                                        &di_blkno);
2456         if (ret) {
2457                 mlog_errno(ret);
2458                 goto out;
2459         }
2460
2461         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2462                                          di_blkno, orphan_name, orphan_insert,
2463                                          false);
2464         if (ret < 0) {
2465                 mlog_errno(ret);
2466                 goto out;
2467         }
2468
2469 out:
2470         if (ret == 0) {
2471                 *ret_orphan_dir = orphan_dir;
2472                 *ret_di_blkno = di_blkno;
2473                 *ret_inode_ac = inode_ac;
2474                 /*
2475                  * orphan_name and orphan_insert are already up to
2476                  * date via prepare_orphan_dir
2477                  */
2478         } else {
2479                 /* Unroll reserve_new_inode* */
2480                 if (inode_ac)
2481                         ocfs2_free_alloc_context(inode_ac);
2482
2483                 /* Unroll orphan dir locking */
2484                 inode_unlock(orphan_dir);
2485                 ocfs2_inode_unlock(orphan_dir, 1);
2486                 iput(orphan_dir);
2487         }
2488
2489         brelse(orphan_dir_bh);
2490
2491         return ret;
2492 }
2493
2494 int ocfs2_create_inode_in_orphan(struct inode *dir,
2495                                  int mode,
2496                                  struct inode **new_inode)
2497 {
2498         int status, did_quota_inode = 0;
2499         struct inode *inode = NULL;
2500         struct inode *orphan_dir = NULL;
2501         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2502         struct ocfs2_dinode *di = NULL;
2503         handle_t *handle = NULL;
2504         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2505         struct buffer_head *parent_di_bh = NULL;
2506         struct buffer_head *new_di_bh = NULL;
2507         struct ocfs2_alloc_context *inode_ac = NULL;
2508         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2509         u64 uninitialized_var(di_blkno), suballoc_loc;
2510         u16 suballoc_bit;
2511
2512         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2513         if (status < 0) {
2514                 if (status != -ENOENT)
2515                         mlog_errno(status);
2516                 return status;
2517         }
2518
2519         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2520                                               orphan_name, &orphan_dir,
2521                                               &di_blkno, &orphan_insert, &inode_ac);
2522         if (status < 0) {
2523                 if (status != -ENOSPC)
2524                         mlog_errno(status);
2525                 goto leave;
2526         }
2527
2528         inode = ocfs2_get_init_inode(dir, mode);
2529         if (IS_ERR(inode)) {
2530                 status = PTR_ERR(inode);
2531                 inode = NULL;
2532                 mlog_errno(status);
2533                 goto leave;
2534         }
2535
2536         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2537         if (IS_ERR(handle)) {
2538                 status = PTR_ERR(handle);
2539                 handle = NULL;
2540                 mlog_errno(status);
2541                 goto leave;
2542         }
2543
2544         status = dquot_alloc_inode(inode);
2545         if (status)
2546                 goto leave;
2547         did_quota_inode = 1;
2548
2549         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2550                                               &suballoc_loc,
2551                                               &suballoc_bit, di_blkno);
2552         if (status < 0) {
2553                 mlog_errno(status);
2554                 goto leave;
2555         }
2556
2557         clear_nlink(inode);
2558         /* do the real work now. */
2559         status = __ocfs2_mknod_locked(dir, inode,
2560                                       0, &new_di_bh, parent_di_bh, handle,
2561                                       inode_ac, di_blkno, suballoc_loc,
2562                                       suballoc_bit);
2563         if (status < 0) {
2564                 mlog_errno(status);
2565                 goto leave;
2566         }
2567
2568         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2569         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2570                                   &orphan_insert, orphan_dir, false);
2571         if (status < 0) {
2572                 mlog_errno(status);
2573                 goto leave;
2574         }
2575
2576         /* get open lock so that only nodes can't remove it from orphan dir. */
2577         status = ocfs2_open_lock(inode);
2578         if (status < 0)
2579                 mlog_errno(status);
2580
2581         insert_inode_hash(inode);
2582 leave:
2583         if (status < 0 && did_quota_inode)
2584                 dquot_free_inode(inode);
2585         if (handle)
2586                 ocfs2_commit_trans(osb, handle);
2587
2588         if (orphan_dir) {
2589                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2590                 ocfs2_inode_unlock(orphan_dir, 1);
2591                 inode_unlock(orphan_dir);
2592                 iput(orphan_dir);
2593         }
2594
2595         if ((status < 0) && inode) {
2596                 clear_nlink(inode);
2597                 iput(inode);
2598         }
2599
2600         if (inode_ac)
2601                 ocfs2_free_alloc_context(inode_ac);
2602
2603         brelse(new_di_bh);
2604
2605         if (!status)
2606                 *new_inode = inode;
2607
2608         ocfs2_free_dir_lookup_result(&orphan_insert);
2609
2610         ocfs2_inode_unlock(dir, 1);
2611         brelse(parent_di_bh);
2612         return status;
2613 }
2614
2615 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2616         struct inode *inode)
2617 {
2618         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2619         struct inode *orphan_dir_inode = NULL;
2620         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2621         struct buffer_head *di_bh = NULL;
2622         int status = 0;
2623         handle_t *handle = NULL;
2624         struct ocfs2_dinode *di = NULL;
2625
2626         status = ocfs2_inode_lock(inode, &di_bh, 1);
2627         if (status < 0) {
2628                 mlog_errno(status);
2629                 goto bail;
2630         }
2631
2632         di = (struct ocfs2_dinode *) di_bh->b_data;
2633         /*
2634          * Another append dio crashed?
2635          * If so, manually recover it first.
2636          */
2637         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2638                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2639                 if (status < 0) {
2640                         if (status != -ENOSPC)
2641                                 mlog_errno(status);
2642                         goto bail_unlock_inode;
2643                 }
2644
2645                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2646                 if (status < 0) {
2647                         mlog_errno(status);
2648                         goto bail_unlock_inode;
2649                 }
2650         }
2651
2652         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2653                         OCFS2_I(inode)->ip_blkno,
2654                         orphan_name,
2655                         &orphan_insert,
2656                         true);
2657         if (status < 0) {
2658                 mlog_errno(status);
2659                 goto bail_unlock_inode;
2660         }
2661
2662         handle = ocfs2_start_trans(osb,
2663                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2664         if (IS_ERR(handle)) {
2665                 status = PTR_ERR(handle);
2666                 goto bail_unlock_orphan;
2667         }
2668
2669         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2670                         &orphan_insert, orphan_dir_inode, true);
2671         if (status)
2672                 mlog_errno(status);
2673
2674         ocfs2_commit_trans(osb, handle);
2675
2676 bail_unlock_orphan:
2677         ocfs2_inode_unlock(orphan_dir_inode, 1);
2678         inode_unlock(orphan_dir_inode);
2679         iput(orphan_dir_inode);
2680
2681         ocfs2_free_dir_lookup_result(&orphan_insert);
2682
2683 bail_unlock_inode:
2684         ocfs2_inode_unlock(inode, 1);
2685         brelse(di_bh);
2686
2687 bail:
2688         return status;
2689 }
2690
2691 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2692                 struct inode *inode, struct buffer_head *di_bh,
2693                 int update_isize, loff_t end)
2694 {
2695         struct inode *orphan_dir_inode = NULL;
2696         struct buffer_head *orphan_dir_bh = NULL;
2697         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2698         handle_t *handle = NULL;
2699         int status = 0;
2700
2701         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2702                         ORPHAN_DIR_SYSTEM_INODE,
2703                         le16_to_cpu(di->i_dio_orphaned_slot));
2704         if (!orphan_dir_inode) {
2705                 status = -ENOENT;
2706                 mlog_errno(status);
2707                 goto bail;
2708         }
2709
2710         inode_lock(orphan_dir_inode);
2711         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2712         if (status < 0) {
2713                 inode_unlock(orphan_dir_inode);
2714                 iput(orphan_dir_inode);
2715                 mlog_errno(status);
2716                 goto bail;
2717         }
2718
2719         handle = ocfs2_start_trans(osb,
2720                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2721         if (IS_ERR(handle)) {
2722                 status = PTR_ERR(handle);
2723                 goto bail_unlock_orphan;
2724         }
2725
2726         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2727
2728         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2729                                 inode, orphan_dir_bh, true);
2730         if (status < 0) {
2731                 mlog_errno(status);
2732                 goto bail_commit;
2733         }
2734
2735         status = ocfs2_journal_access_di(handle,
2736                         INODE_CACHE(inode),
2737                         di_bh,
2738                         OCFS2_JOURNAL_ACCESS_WRITE);
2739         if (status < 0) {
2740                 mlog_errno(status);
2741                 goto bail_commit;
2742         }
2743
2744         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2745         di->i_dio_orphaned_slot = 0;
2746
2747         if (update_isize) {
2748                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2749                 if (status)
2750                         mlog_errno(status);
2751         } else
2752                 ocfs2_journal_dirty(handle, di_bh);
2753
2754 bail_commit:
2755         ocfs2_commit_trans(osb, handle);
2756
2757 bail_unlock_orphan:
2758         ocfs2_inode_unlock(orphan_dir_inode, 1);
2759         inode_unlock(orphan_dir_inode);
2760         brelse(orphan_dir_bh);
2761         iput(orphan_dir_inode);
2762
2763 bail:
2764         return status;
2765 }
2766
2767 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2768                                    struct inode *inode,
2769                                    struct dentry *dentry)
2770 {
2771         int status = 0;
2772         struct buffer_head *parent_di_bh = NULL;
2773         handle_t *handle = NULL;
2774         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2775         struct ocfs2_dinode *dir_di, *di;
2776         struct inode *orphan_dir_inode = NULL;
2777         struct buffer_head *orphan_dir_bh = NULL;
2778         struct buffer_head *di_bh = NULL;
2779         struct ocfs2_dir_lookup_result lookup = { NULL, };
2780
2781         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2782                                 dentry->d_name.len, dentry->d_name.name,
2783                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2784                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2785
2786         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2787         if (status < 0) {
2788                 if (status != -ENOENT)
2789                         mlog_errno(status);
2790                 return status;
2791         }
2792
2793         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2794         if (!dir_di->i_links_count) {
2795                 /* can't make a file in a deleted directory. */
2796                 status = -ENOENT;
2797                 goto leave;
2798         }
2799
2800         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2801                                            dentry->d_name.len);
2802         if (status)
2803                 goto leave;
2804
2805         /* get a spot inside the dir. */
2806         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2807                                               dentry->d_name.name,
2808                                               dentry->d_name.len, &lookup);
2809         if (status < 0) {
2810                 mlog_errno(status);
2811                 goto leave;
2812         }
2813
2814         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2815                                                        ORPHAN_DIR_SYSTEM_INODE,
2816                                                        osb->slot_num);
2817         if (!orphan_dir_inode) {
2818                 status = -ENOENT;
2819                 mlog_errno(status);
2820                 goto leave;
2821         }
2822
2823         inode_lock(orphan_dir_inode);
2824
2825         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2826         if (status < 0) {
2827                 mlog_errno(status);
2828                 inode_unlock(orphan_dir_inode);
2829                 iput(orphan_dir_inode);
2830                 goto leave;
2831         }
2832
2833         status = ocfs2_read_inode_block(inode, &di_bh);
2834         if (status < 0) {
2835                 mlog_errno(status);
2836                 goto orphan_unlock;
2837         }
2838
2839         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2840         if (IS_ERR(handle)) {
2841                 status = PTR_ERR(handle);
2842                 handle = NULL;
2843                 mlog_errno(status);
2844                 goto orphan_unlock;
2845         }
2846
2847         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2848                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2849         if (status < 0) {
2850                 mlog_errno(status);
2851                 goto out_commit;
2852         }
2853
2854         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2855                                   orphan_dir_bh, false);
2856         if (status < 0) {
2857                 mlog_errno(status);
2858                 goto out_commit;
2859         }
2860
2861         di = (struct ocfs2_dinode *)di_bh->b_data;
2862         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2863         di->i_orphaned_slot = 0;
2864         set_nlink(inode, 1);
2865         ocfs2_set_links_count(di, inode->i_nlink);
2866         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2867         ocfs2_journal_dirty(handle, di_bh);
2868
2869         status = ocfs2_add_entry(handle, dentry, inode,
2870                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2871                                  &lookup);
2872         if (status < 0) {
2873                 mlog_errno(status);
2874                 goto out_commit;
2875         }
2876
2877         status = ocfs2_dentry_attach_lock(dentry, inode,
2878                                           OCFS2_I(dir)->ip_blkno);
2879         if (status) {
2880                 mlog_errno(status);
2881                 goto out_commit;
2882         }
2883
2884         d_instantiate(dentry, inode);
2885         status = 0;
2886 out_commit:
2887         ocfs2_commit_trans(osb, handle);
2888 orphan_unlock:
2889         ocfs2_inode_unlock(orphan_dir_inode, 1);
2890         inode_unlock(orphan_dir_inode);
2891         iput(orphan_dir_inode);
2892 leave:
2893
2894         ocfs2_inode_unlock(dir, 1);
2895
2896         brelse(di_bh);
2897         brelse(parent_di_bh);
2898         brelse(orphan_dir_bh);
2899
2900         ocfs2_free_dir_lookup_result(&lookup);
2901
2902         if (status)
2903                 mlog_errno(status);
2904
2905         return status;
2906 }
2907
2908 const struct inode_operations ocfs2_dir_iops = {
2909         .create         = ocfs2_create,
2910         .lookup         = ocfs2_lookup,
2911         .link           = ocfs2_link,
2912         .unlink         = ocfs2_unlink,
2913         .rmdir          = ocfs2_unlink,
2914         .symlink        = ocfs2_symlink,
2915         .mkdir          = ocfs2_mkdir,
2916         .mknod          = ocfs2_mknod,
2917         .rename         = ocfs2_rename,
2918         .setattr        = ocfs2_setattr,
2919         .getattr        = ocfs2_getattr,
2920         .permission     = ocfs2_permission,
2921         .listxattr      = ocfs2_listxattr,
2922         .fiemap         = ocfs2_fiemap,
2923         .get_acl        = ocfs2_iop_get_acl,
2924         .set_acl        = ocfs2_iop_set_acl,
2925 };