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