kernel/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 buffer_head *new_fe_bh = NULL;
 248         struct inode *inode = NULL;
 249         struct ocfs2_alloc_context *inode_ac = NULL;
 250         struct ocfs2_alloc_context *data_ac = NULL;
 251         struct ocfs2_alloc_context *meta_ac = NULL;
 252         int want_clusters = 0;
 253         int want_meta = 0;
 254         int xattr_credits = 0;
 255         struct ocfs2_security_xattr_info si = {
 256                 .enable = 1,
 257         };
 258         int did_quota_inode = 0;
 259         struct ocfs2_dir_lookup_result lookup = { NULL, };
 260         sigset_t oldset;
 261         int did_block_signals = 0;
 262         struct posix_acl *default_acl = NULL, *acl = NULL;
 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         status = posix_acl_create(dir, &inode->i_mode, &default_acl, &acl);
 371         if (status) {
 372                 mlog_errno(status);
 373                 goto leave;
 374         }
 375
 376         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 377                                                             S_ISDIR(mode),
 378                                                             xattr_credits));
 379         if (IS_ERR(handle)) {
 380                 status = PTR_ERR(handle);
 381                 handle = NULL;
 382                 mlog_errno(status);
 383                 goto leave;
 384         }
 385
 386         /* Starting to change things, restart is no longer possible. */
 387         ocfs2_block_signals(&oldset);
 388         did_block_signals = 1;
 389
 390         status = dquot_alloc_inode(inode);
 391         if (status)
 392                 goto leave;
 393         did_quota_inode = 1;
 394
 395         /* do the real work now. */
 396         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 397                                     &new_fe_bh, parent_fe_bh, handle,
 398                                     inode_ac);
 399         if (status < 0) {
 400                 mlog_errno(status);
 401                 goto leave;
 402         }
 403
 404         if (S_ISDIR(mode)) {
 405                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 406                                             new_fe_bh, data_ac, meta_ac);
 407                 if (status < 0) {
 408                         mlog_errno(status);
 409                         goto leave;
 410                 }
 411
 412                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 413                                                  parent_fe_bh,
 414                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 415                 if (status < 0) {
 416                         mlog_errno(status);
 417                         goto leave;
 418                 }
 419                 ocfs2_add_links_count(dirfe, 1);
 420                 ocfs2_journal_dirty(handle, parent_fe_bh);
 421                 inc_nlink(dir);
 422         }
 423
 424         if (default_acl) {
 425                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
 426                                        ACL_TYPE_DEFAULT, default_acl,
 427                                        meta_ac, data_ac);
 428         }
 429         if (!status && acl) {
 430                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
 431                                        ACL_TYPE_ACCESS, acl,
 432                                        meta_ac, data_ac);
 433         }
 434
 435         if (status < 0) {
 436                 mlog_errno(status);
 437                 goto leave;
 438         }
 439
 440         if (si.enable) {
 441                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 442                                                  meta_ac, data_ac);
 443                 if (status < 0) {
 444                         mlog_errno(status);
 445                         goto leave;
 446                 }
 447         }
 448
 449         /*
 450          * Do this before adding the entry to the directory. We add
 451          * also set d_op after success so that ->d_iput() will cleanup
 452          * the dentry lock even if ocfs2_add_entry() fails below.
 453          */
 454         status = ocfs2_dentry_attach_lock(dentry, inode,
 455                                           OCFS2_I(dir)->ip_blkno);
 456         if (status) {
 457                 mlog_errno(status);
 458                 goto leave;
 459         }
 460
 461         dl = dentry->d_fsdata;
 462
 463         status = ocfs2_add_entry(handle, dentry, inode,
 464                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 465                                  &lookup);
 466         if (status < 0) {
 467                 mlog_errno(status);
 468                 goto leave;
 469         }
 470
 471         insert_inode_hash(inode);
 472         d_instantiate(dentry, inode);
 473         status = 0;
 474 leave:
 475         if (default_acl)
 476                 posix_acl_release(default_acl);
 477         if (acl)
 478                 posix_acl_release(acl);
 479         if (status < 0 && did_quota_inode)
 480                 dquot_free_inode(inode);
 481         if (handle)
 482                 ocfs2_commit_trans(osb, handle);
 483
 484         ocfs2_inode_unlock(dir, 1);
 485         if (did_block_signals)
 486                 ocfs2_unblock_signals(&oldset);
 487
 488         brelse(new_fe_bh);
 489         brelse(parent_fe_bh);
 490         kfree(si.value);
 491
 492         ocfs2_free_dir_lookup_result(&lookup);
 493
 494         if (inode_ac)
 495                 ocfs2_free_alloc_context(inode_ac);
 496
 497         if (data_ac)
 498                 ocfs2_free_alloc_context(data_ac);
 499
 500         if (meta_ac)
 501                 ocfs2_free_alloc_context(meta_ac);
 502
 503         /*
 504          * We should call iput after the i_mutex of the bitmap been
 505          * unlocked in ocfs2_free_alloc_context, or the
 506          * ocfs2_delete_inode will mutex_lock again.
 507          */
 508         if ((status < 0) && inode) {
 509                 if (dl)
 510                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 511
 512                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 513                 clear_nlink(inode);
 514                 iput(inode);
 515         }
 516
 517         if (status)
 518                 mlog_errno(status);
 519
 520         return status;
 521 }
 522
 523 static int __ocfs2_mknod_locked(struct inode *dir,
 524                                 struct inode *inode,
 525                                 dev_t dev,
 526                                 struct buffer_head **new_fe_bh,
 527                                 struct buffer_head *parent_fe_bh,
 528                                 handle_t *handle,
 529                                 struct ocfs2_alloc_context *inode_ac,
 530                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 531 {
 532         int status = 0;
 533         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 534         struct ocfs2_dinode *fe = NULL;
 535         struct ocfs2_extent_list *fel;
 536         u16 feat;
 537         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 538
 539         *new_fe_bh = NULL;
 540
 541         /* populate as many fields early on as possible - many of
 542          * these are used by the support functions here and in
 543          * callers. */
 544         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 545         OCFS2_I(inode)->ip_blkno = fe_blkno;
 546         spin_lock(&osb->osb_lock);
 547         inode->i_generation = osb->s_next_generation++;
 548         spin_unlock(&osb->osb_lock);
 549
 550         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 551         if (!*new_fe_bh) {
 552                 status = -ENOMEM;
 553                 mlog_errno(status);
 554                 goto leave;
 555         }
 556         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 557
 558         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 559                                          *new_fe_bh,
 560                                          OCFS2_JOURNAL_ACCESS_CREATE);
 561         if (status < 0) {
 562                 mlog_errno(status);
 563                 goto leave;
 564         }
 565
 566         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 567         memset(fe, 0, osb->sb->s_blocksize);
 568
 569         fe->i_generation = cpu_to_le32(inode->i_generation);
 570         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 571         fe->i_blkno = cpu_to_le64(fe_blkno);
 572         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 573         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 574         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 575         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 576         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 577         fe->i_mode = cpu_to_le16(inode->i_mode);
 578         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 579                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 580
 581         ocfs2_set_links_count(fe, inode->i_nlink);
 582
 583         fe->i_last_eb_blk = 0;
 584         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 585         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 586         fe->i_atime = fe->i_ctime = fe->i_mtime =
 587                 cpu_to_le64(CURRENT_TIME.tv_sec);
 588         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 589                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 590         fe->i_dtime = 0;
 591
 592         /*
 593          * If supported, directories start with inline data. If inline
 594          * isn't supported, but indexing is, we start them as indexed.
 595          */
 596         feat = le16_to_cpu(fe->i_dyn_features);
 597         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 598                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 599
 600                 fe->id2.i_data.id_count = cpu_to_le16(
 601                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 602         } else {
 603                 fel = &fe->id2.i_list;
 604                 fel->l_tree_depth = 0;
 605                 fel->l_next_free_rec = 0;
 606                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 607         }
 608
 609         ocfs2_journal_dirty(handle, *new_fe_bh);
 610
 611         ocfs2_populate_inode(inode, fe, 1);
 612         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 613         if (!ocfs2_mount_local(osb)) {
 614                 status = ocfs2_create_new_inode_locks(inode);
 615                 if (status < 0)
 616                         mlog_errno(status);
 617         }
 618
 619         oi->i_sync_tid = handle->h_transaction->t_tid;
 620         oi->i_datasync_tid = handle->h_transaction->t_tid;
 621
 622 leave:
 623         if (status < 0) {
 624                 if (*new_fe_bh) {
 625                         brelse(*new_fe_bh);
 626                         *new_fe_bh = NULL;
 627                 }
 628         }
 629
 630         if (status)
 631                 mlog_errno(status);
 632         return status;
 633 }
 634
 635 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 636                               struct inode *dir,
 637                               struct inode *inode,
 638                               dev_t dev,
 639                               struct buffer_head **new_fe_bh,
 640                               struct buffer_head *parent_fe_bh,
 641                               handle_t *handle,
 642                               struct ocfs2_alloc_context *inode_ac)
 643 {
 644         int status = 0;
 645         u64 suballoc_loc, fe_blkno = 0;
 646         u16 suballoc_bit;
 647
 648         *new_fe_bh = NULL;
 649
 650         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 651                                        inode_ac, &suballoc_loc,
 652                                        &suballoc_bit, &fe_blkno);
 653         if (status < 0) {
 654                 mlog_errno(status);
 655                 return status;
 656         }
 657
 658         status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 659                                     parent_fe_bh, handle, inode_ac,
 660                                     fe_blkno, suballoc_loc, suballoc_bit);
 661         if (status < 0) {
 662                 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
 663                 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
 664                                 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
 665                 if (tmp)
 666                         mlog_errno(tmp);
 667         }
 668
 669         return status;
 670 }
 671
 672 static int ocfs2_mkdir(struct inode *dir,
 673                        struct dentry *dentry,
 674                        umode_t mode)
 675 {
 676         int ret;
 677
 678         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 679                           OCFS2_I(dir)->ip_blkno, mode);
 680         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 681         if (ret)
 682                 mlog_errno(ret);
 683
 684         return ret;
 685 }
 686
 687 static int ocfs2_create(struct inode *dir,
 688                         struct dentry *dentry,
 689                         umode_t mode,
 690                         bool excl)
 691 {
 692         int ret;
 693
 694         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 695                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 696         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 697         if (ret)
 698                 mlog_errno(ret);
 699
 700         return ret;
 701 }
 702
 703 static int ocfs2_link(struct dentry *old_dentry,
 704                       struct inode *dir,
 705                       struct dentry *dentry)
 706 {
 707         handle_t *handle;
 708         struct inode *inode = d_inode(old_dentry);
 709         struct inode *old_dir = d_inode(old_dentry->d_parent);
 710         int err;
 711         struct buffer_head *fe_bh = NULL;
 712         struct buffer_head *old_dir_bh = NULL;
 713         struct buffer_head *parent_fe_bh = NULL;
 714         struct ocfs2_dinode *fe = NULL;
 715         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 716         struct ocfs2_dir_lookup_result lookup = { NULL, };
 717         sigset_t oldset;
 718         u64 old_de_ino;
 719
 720         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 721                          old_dentry->d_name.len, old_dentry->d_name.name,
 722                          dentry->d_name.len, dentry->d_name.name);
 723
 724         if (S_ISDIR(inode->i_mode))
 725                 return -EPERM;
 726
 727         err = dquot_initialize(dir);
 728         if (err) {
 729                 mlog_errno(err);
 730                 return err;
 731         }
 732
 733         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 734                         &parent_fe_bh, dir, 0);
 735         if (err < 0) {
 736                 if (err != -ENOENT)
 737                         mlog_errno(err);
 738                 return err;
 739         }
 740
 741         /* make sure both dirs have bhs
 742          * get an extra ref on old_dir_bh if old==new */
 743         if (!parent_fe_bh) {
 744                 if (old_dir_bh) {
 745                         parent_fe_bh = old_dir_bh;
 746                         get_bh(parent_fe_bh);
 747                 } else {
 748                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 749                         err = -EIO;
 750                         goto out;
 751                 }
 752         }
 753
 754         if (!dir->i_nlink) {
 755                 err = -ENOENT;
 756                 goto out;
 757         }
 758
 759         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 760                         old_dentry->d_name.len, &old_de_ino);
 761         if (err) {
 762                 err = -ENOENT;
 763                 goto out;
 764         }
 765
 766         /*
 767          * Check whether another node removed the source inode while we
 768          * were in the vfs.
 769          */
 770         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 771                 err = -ENOENT;
 772                 goto out;
 773         }
 774
 775         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 776                                         dentry->d_name.len);
 777         if (err)
 778                 goto out;
 779
 780         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 781                                            dentry->d_name.name,
 782                                            dentry->d_name.len, &lookup);
 783         if (err < 0) {
 784                 mlog_errno(err);
 785                 goto out;
 786         }
 787
 788         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 789         if (err < 0) {
 790                 if (err != -ENOENT)
 791                         mlog_errno(err);
 792                 goto out;
 793         }
 794
 795         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 796         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 797                 err = -EMLINK;
 798                 goto out_unlock_inode;
 799         }
 800
 801         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 802         if (IS_ERR(handle)) {
 803                 err = PTR_ERR(handle);
 804                 handle = NULL;
 805                 mlog_errno(err);
 806                 goto out_unlock_inode;
 807         }
 808
 809         /* Starting to change things, restart is no longer possible. */
 810         ocfs2_block_signals(&oldset);
 811
 812         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 813                                       OCFS2_JOURNAL_ACCESS_WRITE);
 814         if (err < 0) {
 815                 mlog_errno(err);
 816                 goto out_commit;
 817         }
 818
 819         inc_nlink(inode);
 820         inode->i_ctime = CURRENT_TIME;
 821         ocfs2_set_links_count(fe, inode->i_nlink);
 822         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 823         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 824         ocfs2_journal_dirty(handle, fe_bh);
 825
 826         err = ocfs2_add_entry(handle, dentry, inode,
 827                               OCFS2_I(inode)->ip_blkno,
 828                               parent_fe_bh, &lookup);
 829         if (err) {
 830                 ocfs2_add_links_count(fe, -1);
 831                 drop_nlink(inode);
 832                 mlog_errno(err);
 833                 goto out_commit;
 834         }
 835
 836         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 837         if (err) {
 838                 mlog_errno(err);
 839                 goto out_commit;
 840         }
 841
 842         ihold(inode);
 843         d_instantiate(dentry, inode);
 844
 845 out_commit:
 846         ocfs2_commit_trans(osb, handle);
 847         ocfs2_unblock_signals(&oldset);
 848 out_unlock_inode:
 849         ocfs2_inode_unlock(inode, 1);
 850
 851 out:
 852         ocfs2_double_unlock(old_dir, dir);
 853
 854         brelse(fe_bh);
 855         brelse(parent_fe_bh);
 856         brelse(old_dir_bh);
 857
 858         ocfs2_free_dir_lookup_result(&lookup);
 859
 860         if (err)
 861                 mlog_errno(err);
 862
 863         return err;
 864 }
 865
 866 /*
 867  * Takes and drops an exclusive lock on the given dentry. This will
 868  * force other nodes to drop it.
 869  */
 870 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 871 {
 872         int ret;
 873
 874         ret = ocfs2_dentry_lock(dentry, 1);
 875         if (ret)
 876                 mlog_errno(ret);
 877         else
 878                 ocfs2_dentry_unlock(dentry, 1);
 879
 880         return ret;
 881 }
 882
 883 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 884 {
 885         if (S_ISDIR(inode->i_mode)) {
 886                 if (inode->i_nlink == 2)
 887                         return 1;
 888                 return 0;
 889         }
 890
 891         if (inode->i_nlink == 1)
 892                 return 1;
 893         return 0;
 894 }
 895
 896 static int ocfs2_unlink(struct inode *dir,
 897                         struct dentry *dentry)
 898 {
 899         int status;
 900         int child_locked = 0;
 901         bool is_unlinkable = false;
 902         struct inode *inode = d_inode(dentry);
 903         struct inode *orphan_dir = NULL;
 904         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 905         u64 blkno;
 906         struct ocfs2_dinode *fe = NULL;
 907         struct buffer_head *fe_bh = NULL;
 908         struct buffer_head *parent_node_bh = NULL;
 909         handle_t *handle = NULL;
 910         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 911         struct ocfs2_dir_lookup_result lookup = { NULL, };
 912         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 913
 914         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 915                            dentry->d_name.name,
 916                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 917                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 918
 919         status = dquot_initialize(dir);
 920         if (status) {
 921                 mlog_errno(status);
 922                 return status;
 923         }
 924
 925         BUG_ON(d_inode(dentry->d_parent) != dir);
 926
 927         if (inode == osb->root_inode)
 928                 return -EPERM;
 929
 930         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 931                                          OI_LS_PARENT);
 932         if (status < 0) {
 933                 if (status != -ENOENT)
 934                         mlog_errno(status);
 935                 return status;
 936         }
 937
 938         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 939                                           dentry->d_name.len, &blkno, dir,
 940                                           &lookup);
 941         if (status < 0) {
 942                 if (status != -ENOENT)
 943                         mlog_errno(status);
 944                 goto leave;
 945         }
 946
 947         if (OCFS2_I(inode)->ip_blkno != blkno) {
 948                 status = -ENOENT;
 949
 950                 trace_ocfs2_unlink_noent(
 951                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 952                                 (unsigned long long)blkno,
 953                                 OCFS2_I(inode)->ip_flags);
 954                 goto leave;
 955         }
 956
 957         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 958         if (status < 0) {
 959                 if (status != -ENOENT)
 960                         mlog_errno(status);
 961                 goto leave;
 962         }
 963         child_locked = 1;
 964
 965         if (S_ISDIR(inode->i_mode)) {
 966                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 967                         status = -ENOTEMPTY;
 968                         goto leave;
 969                 }
 970         }
 971
 972         status = ocfs2_remote_dentry_delete(dentry);
 973         if (status < 0) {
 974                 /* This remote delete should succeed under all normal
 975                  * circumstances. */
 976                 mlog_errno(status);
 977                 goto leave;
 978         }
 979
 980         if (ocfs2_inode_is_unlinkable(inode)) {
 981                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 982                                                   OCFS2_I(inode)->ip_blkno,
 983                                                   orphan_name, &orphan_insert,
 984                                                   false);
 985                 if (status < 0) {
 986                         mlog_errno(status);
 987                         goto leave;
 988                 }
 989                 is_unlinkable = true;
 990         }
 991
 992         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 993         if (IS_ERR(handle)) {
 994                 status = PTR_ERR(handle);
 995                 handle = NULL;
 996                 mlog_errno(status);
 997                 goto leave;
 998         }
 999
1000         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
1001                                          OCFS2_JOURNAL_ACCESS_WRITE);
1002         if (status < 0) {
1003                 mlog_errno(status);
1004                 goto leave;
1005         }
1006
1007         fe = (struct ocfs2_dinode *) fe_bh->b_data;
1008
1009         /* delete the name from the parent dir */
1010         status = ocfs2_delete_entry(handle, dir, &lookup);
1011         if (status < 0) {
1012                 mlog_errno(status);
1013                 goto leave;
1014         }
1015
1016         if (S_ISDIR(inode->i_mode))
1017                 drop_nlink(inode);
1018         drop_nlink(inode);
1019         ocfs2_set_links_count(fe, inode->i_nlink);
1020         ocfs2_journal_dirty(handle, fe_bh);
1021
1022         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1023         if (S_ISDIR(inode->i_mode))
1024                 drop_nlink(dir);
1025
1026         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1027         if (status < 0) {
1028                 mlog_errno(status);
1029                 if (S_ISDIR(inode->i_mode))
1030                         inc_nlink(dir);
1031                 goto leave;
1032         }
1033
1034         if (is_unlinkable) {
1035                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1036                                 orphan_name, &orphan_insert, orphan_dir, false);
1037                 if (status < 0)
1038                         mlog_errno(status);
1039         }
1040
1041 leave:
1042         if (handle)
1043                 ocfs2_commit_trans(osb, handle);
1044
1045         if (orphan_dir) {
1046                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1047                 ocfs2_inode_unlock(orphan_dir, 1);
1048                 mutex_unlock(&orphan_dir->i_mutex);
1049                 iput(orphan_dir);
1050         }
1051
1052         if (child_locked)
1053                 ocfs2_inode_unlock(inode, 1);
1054
1055         ocfs2_inode_unlock(dir, 1);
1056
1057         brelse(fe_bh);
1058         brelse(parent_node_bh);
1059
1060         ocfs2_free_dir_lookup_result(&orphan_insert);
1061         ocfs2_free_dir_lookup_result(&lookup);
1062
1063         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1064                 mlog_errno(status);
1065
1066         return status;
1067 }
1068
1069 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1070                 u64 src_inode_no, u64 dest_inode_no)
1071 {
1072         int ret = 0, i = 0;
1073         u64 parent_inode_no = 0;
1074         u64 child_inode_no = src_inode_no;
1075         struct inode *child_inode;
1076
1077 #define MAX_LOOKUP_TIMES 32
1078         while (1) {
1079                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1080                 if (IS_ERR(child_inode)) {
1081                         ret = PTR_ERR(child_inode);
1082                         break;
1083                 }
1084
1085                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1086                 if (ret < 0) {
1087                         iput(child_inode);
1088                         if (ret != -ENOENT)
1089                                 mlog_errno(ret);
1090                         break;
1091                 }
1092
1093                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1094                                 &parent_inode_no);
1095                 ocfs2_inode_unlock(child_inode, 0);
1096                 iput(child_inode);
1097                 if (ret < 0) {
1098                         ret = -ENOENT;
1099                         break;
1100                 }
1101
1102                 if (parent_inode_no == dest_inode_no) {
1103                         ret = 1;
1104                         break;
1105                 }
1106
1107                 if (parent_inode_no == osb->root_inode->i_ino) {
1108                         ret = 0;
1109                         break;
1110                 }
1111
1112                 child_inode_no = parent_inode_no;
1113
1114                 if (++i >= MAX_LOOKUP_TIMES) {
1115                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1116                                         "may have nested directories, "
1117                                         "src inode: %llu, dest inode: %llu.\n",
1118                                         (unsigned long long)src_inode_no,
1119                                         (unsigned long long)dest_inode_no);
1120                         ret = 0;
1121                         break;
1122                 }
1123         }
1124
1125         return ret;
1126 }
1127
1128 /*
1129  * The only place this should be used is rename and link!
1130  * if they have the same id, then the 1st one is the only one locked.
1131  */
1132 static int ocfs2_double_lock(struct ocfs2_super *osb,
1133                              struct buffer_head **bh1,
1134                              struct inode *inode1,
1135                              struct buffer_head **bh2,
1136                              struct inode *inode2,
1137                              int rename)
1138 {
1139         int status;
1140         int inode1_is_ancestor, inode2_is_ancestor;
1141         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1142         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1143
1144         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1145                                 (unsigned long long)oi2->ip_blkno);
1146
1147         if (*bh1)
1148                 *bh1 = NULL;
1149         if (*bh2)
1150                 *bh2 = NULL;
1151
1152         /* we always want to lock the one with the lower lockid first.
1153          * and if they are nested, we lock ancestor first */
1154         if (oi1->ip_blkno != oi2->ip_blkno) {
1155                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1156                                 oi1->ip_blkno);
1157                 if (inode1_is_ancestor < 0) {
1158                         status = inode1_is_ancestor;
1159                         goto bail;
1160                 }
1161
1162                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1163                                 oi2->ip_blkno);
1164                 if (inode2_is_ancestor < 0) {
1165                         status = inode2_is_ancestor;
1166                         goto bail;
1167                 }
1168
1169                 if ((inode1_is_ancestor == 1) ||
1170                                 (oi1->ip_blkno < oi2->ip_blkno &&
1171                                 inode2_is_ancestor == 0)) {
1172                         /* switch id1 and id2 around */
1173                         swap(bh2, bh1);
1174                         swap(inode2, inode1);
1175                 }
1176                 /* lock id2 */
1177                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1178                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1179                 if (status < 0) {
1180                         if (status != -ENOENT)
1181                                 mlog_errno(status);
1182                         goto bail;
1183                 }
1184         }
1185
1186         /* lock id1 */
1187         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1188                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1189         if (status < 0) {
1190                 /*
1191                  * An error return must mean that no cluster locks
1192                  * were held on function exit.
1193                  */
1194                 if (oi1->ip_blkno != oi2->ip_blkno) {
1195                         ocfs2_inode_unlock(inode2, 1);
1196                         brelse(*bh2);
1197                         *bh2 = NULL;
1198                 }
1199
1200                 if (status != -ENOENT)
1201                         mlog_errno(status);
1202         }
1203
1204         trace_ocfs2_double_lock_end(
1205                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1206                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1207
1208 bail:
1209         if (status)
1210                 mlog_errno(status);
1211         return status;
1212 }
1213
1214 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1215 {
1216         ocfs2_inode_unlock(inode1, 1);
1217
1218         if (inode1 != inode2)
1219                 ocfs2_inode_unlock(inode2, 1);
1220 }
1221
1222 static int ocfs2_rename(struct inode *old_dir,
1223                         struct dentry *old_dentry,
1224                         struct inode *new_dir,
1225                         struct dentry *new_dentry)
1226 {
1227         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1228         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1229         struct inode *old_inode = d_inode(old_dentry);
1230         struct inode *new_inode = d_inode(new_dentry);
1231         struct inode *orphan_dir = NULL;
1232         struct ocfs2_dinode *newfe = NULL;
1233         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1234         struct buffer_head *newfe_bh = NULL;
1235         struct buffer_head *old_inode_bh = NULL;
1236         struct ocfs2_super *osb = NULL;
1237         u64 newfe_blkno, old_de_ino;
1238         handle_t *handle = NULL;
1239         struct buffer_head *old_dir_bh = NULL;
1240         struct buffer_head *new_dir_bh = NULL;
1241         u32 old_dir_nlink = old_dir->i_nlink;
1242         struct ocfs2_dinode *old_di;
1243         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1244         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1245         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1246         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1247         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1248         bool should_add_orphan = false;
1249
1250         /* At some point it might be nice to break this function up a
1251          * bit. */
1252
1253         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1254                            old_dentry->d_name.len, old_dentry->d_name.name,
1255                            new_dentry->d_name.len, new_dentry->d_name.name);
1256
1257         status = dquot_initialize(old_dir);
1258         if (status) {
1259                 mlog_errno(status);
1260                 goto bail;
1261         }
1262         status = dquot_initialize(new_dir);
1263         if (status) {
1264                 mlog_errno(status);
1265                 goto bail;
1266         }
1267
1268         osb = OCFS2_SB(old_dir->i_sb);
1269
1270         if (new_inode) {
1271                 if (!igrab(new_inode))
1272                         BUG();
1273         }
1274
1275         /* Assume a directory hierarchy thusly:
1276          * a/b/c
1277          * a/d
1278          * a,b,c, and d are all directories.
1279          *
1280          * from cwd of 'a' on both nodes:
1281          * node1: mv b/c d
1282          * node2: mv d   b/c
1283          *
1284          * And that's why, just like the VFS, we need a file system
1285          * rename lock. */
1286         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1287                 status = ocfs2_rename_lock(osb);
1288                 if (status < 0) {
1289                         mlog_errno(status);
1290                         goto bail;
1291                 }
1292                 rename_lock = 1;
1293
1294                 /* here we cannot guarantee the inodes haven't just been
1295                  * changed, so check if they are nested again */
1296                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1297                                 old_inode->i_ino);
1298                 if (status < 0) {
1299                         mlog_errno(status);
1300                         goto bail;
1301                 } else if (status == 1) {
1302                         status = -EPERM;
1303                         trace_ocfs2_rename_not_permitted(
1304                                         (unsigned long long)old_inode->i_ino,
1305                                         (unsigned long long)new_dir->i_ino);
1306                         goto bail;
1307                 }
1308         }
1309
1310         /* if old and new are the same, this'll just do one lock. */
1311         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1312                                    &new_dir_bh, new_dir, 1);
1313         if (status < 0) {
1314                 mlog_errno(status);
1315                 goto bail;
1316         }
1317         parents_locked = 1;
1318
1319         if (!new_dir->i_nlink) {
1320                 status = -EACCES;
1321                 goto bail;
1322         }
1323
1324         /* make sure both dirs have bhs
1325          * get an extra ref on old_dir_bh if old==new */
1326         if (!new_dir_bh) {
1327                 if (old_dir_bh) {
1328                         new_dir_bh = old_dir_bh;
1329                         get_bh(new_dir_bh);
1330                 } else {
1331                         mlog(ML_ERROR, "no old_dir_bh!\n");
1332                         status = -EIO;
1333                         goto bail;
1334                 }
1335         }
1336
1337         /*
1338          * Aside from allowing a meta data update, the locking here
1339          * also ensures that the downconvert thread on other nodes
1340          * won't have to concurrently downconvert the inode and the
1341          * dentry locks.
1342          */
1343         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1344                                          OI_LS_PARENT);
1345         if (status < 0) {
1346                 if (status != -ENOENT)
1347                         mlog_errno(status);
1348                 goto bail;
1349         }
1350         old_child_locked = 1;
1351
1352         status = ocfs2_remote_dentry_delete(old_dentry);
1353         if (status < 0) {
1354                 mlog_errno(status);
1355                 goto bail;
1356         }
1357
1358         if (S_ISDIR(old_inode->i_mode)) {
1359                 u64 old_inode_parent;
1360
1361                 update_dot_dot = 1;
1362                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1363                                                   old_inode,
1364                                                   &old_inode_dot_dot_res);
1365                 if (status) {
1366                         status = -EIO;
1367                         goto bail;
1368                 }
1369
1370                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1371                         status = -EIO;
1372                         goto bail;
1373                 }
1374
1375                 if (!new_inode && new_dir != old_dir &&
1376                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1377                         status = -EMLINK;
1378                         goto bail;
1379                 }
1380         }
1381
1382         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1383                                             old_dentry->d_name.len,
1384                                             &old_de_ino);
1385         if (status) {
1386                 status = -ENOENT;
1387                 goto bail;
1388         }
1389
1390         /*
1391          *  Check for inode number is _not_ due to possible IO errors.
1392          *  We might rmdir the source, keep it as pwd of some process
1393          *  and merrily kill the link to whatever was created under the
1394          *  same name. Goodbye sticky bit ;-<
1395          */
1396         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1397                 status = -ENOENT;
1398                 goto bail;
1399         }
1400
1401         /* check if the target already exists (in which case we need
1402          * to delete it */
1403         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1404                                           new_dentry->d_name.len,
1405                                           &newfe_blkno, new_dir,
1406                                           &target_lookup_res);
1407         /* The only error we allow here is -ENOENT because the new
1408          * file not existing is perfectly valid. */
1409         if ((status < 0) && (status != -ENOENT)) {
1410                 /* If we cannot find the file specified we should just */
1411                 /* return the error... */
1412                 mlog_errno(status);
1413                 goto bail;
1414         }
1415         if (status == 0)
1416                 target_exists = 1;
1417
1418         if (!target_exists && new_inode) {
1419                 /*
1420                  * Target was unlinked by another node while we were
1421                  * waiting to get to ocfs2_rename(). There isn't
1422                  * anything we can do here to help the situation, so
1423                  * bubble up the appropriate error.
1424                  */
1425                 status = -ENOENT;
1426                 goto bail;
1427         }
1428
1429         /* In case we need to overwrite an existing file, we blow it
1430          * away first */
1431         if (target_exists) {
1432                 /* VFS didn't think there existed an inode here, but
1433                  * someone else in the cluster must have raced our
1434                  * rename to create one. Today we error cleanly, in
1435                  * the future we should consider calling iget to build
1436                  * a new struct inode for this entry. */
1437                 if (!new_inode) {
1438                         status = -EACCES;
1439
1440                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1441                                                 new_dentry->d_name.name);
1442                         goto bail;
1443                 }
1444
1445                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1446                         status = -EACCES;
1447
1448                         trace_ocfs2_rename_disagree(
1449                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1450                              (unsigned long long)newfe_blkno,
1451                              OCFS2_I(new_inode)->ip_flags);
1452                         goto bail;
1453                 }
1454
1455                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1456                 if (status < 0) {
1457                         if (status != -ENOENT)
1458                                 mlog_errno(status);
1459                         goto bail;
1460                 }
1461                 new_child_locked = 1;
1462
1463                 status = ocfs2_remote_dentry_delete(new_dentry);
1464                 if (status < 0) {
1465                         mlog_errno(status);
1466                         goto bail;
1467                 }
1468
1469                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1470
1471                 trace_ocfs2_rename_over_existing(
1472                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1473                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1474
1475                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1476                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1477                                                 OCFS2_I(new_inode)->ip_blkno,
1478                                                 orphan_name, &orphan_insert,
1479                                                 false);
1480                         if (status < 0) {
1481                                 mlog_errno(status);
1482                                 goto bail;
1483                         }
1484                         should_add_orphan = true;
1485                 }
1486         } else {
1487                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1488
1489                 status = ocfs2_check_dir_for_entry(new_dir,
1490                                                    new_dentry->d_name.name,
1491                                                    new_dentry->d_name.len);
1492                 if (status)
1493                         goto bail;
1494
1495                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1496                                                       new_dentry->d_name.name,
1497                                                       new_dentry->d_name.len,
1498                                                       &target_insert);
1499                 if (status < 0) {
1500                         mlog_errno(status);
1501                         goto bail;
1502                 }
1503         }
1504
1505         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1506         if (IS_ERR(handle)) {
1507                 status = PTR_ERR(handle);
1508                 handle = NULL;
1509                 mlog_errno(status);
1510                 goto bail;
1511         }
1512
1513         if (target_exists) {
1514                 if (S_ISDIR(new_inode->i_mode)) {
1515                         if (new_inode->i_nlink != 2 ||
1516                             !ocfs2_empty_dir(new_inode)) {
1517                                 status = -ENOTEMPTY;
1518                                 goto bail;
1519                         }
1520                 }
1521                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1522                                                  newfe_bh,
1523                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1524                 if (status < 0) {
1525                         mlog_errno(status);
1526                         goto bail;
1527                 }
1528
1529                 /* change the dirent to point to the correct inode */
1530                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1531                                             old_inode);
1532                 if (status < 0) {
1533                         mlog_errno(status);
1534                         goto bail;
1535                 }
1536                 new_dir->i_version++;
1537
1538                 if (S_ISDIR(new_inode->i_mode))
1539                         ocfs2_set_links_count(newfe, 0);
1540                 else
1541                         ocfs2_add_links_count(newfe, -1);
1542                 ocfs2_journal_dirty(handle, newfe_bh);
1543                 if (should_add_orphan) {
1544                         status = ocfs2_orphan_add(osb, handle, new_inode,
1545                                         newfe_bh, orphan_name,
1546                                         &orphan_insert, orphan_dir, false);
1547                         if (status < 0) {
1548                                 mlog_errno(status);
1549                                 goto bail;
1550                         }
1551                 }
1552         } else {
1553                 /* if the name was not found in new_dir, add it now */
1554                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1555                                          OCFS2_I(old_inode)->ip_blkno,
1556                                          new_dir_bh, &target_insert);
1557         }
1558
1559         old_inode->i_ctime = CURRENT_TIME;
1560         mark_inode_dirty(old_inode);
1561
1562         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1563                                          old_inode_bh,
1564                                          OCFS2_JOURNAL_ACCESS_WRITE);
1565         if (status >= 0) {
1566                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1567
1568                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1569                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1570                 ocfs2_journal_dirty(handle, old_inode_bh);
1571         } else
1572                 mlog_errno(status);
1573
1574         /*
1575          * Now that the name has been added to new_dir, remove the old name.
1576          *
1577          * We don't keep any directory entry context around until now
1578          * because the insert might have changed the type of directory
1579          * we're dealing with.
1580          */
1581         status = ocfs2_find_entry(old_dentry->d_name.name,
1582                                   old_dentry->d_name.len, old_dir,
1583                                   &old_entry_lookup);
1584         if (status) {
1585                 if (!is_journal_aborted(osb->journal->j_journal)) {
1586                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1587                                         "is not deleted.",
1588                                         new_dentry->d_name.len, new_dentry->d_name.name,
1589                                         old_dentry->d_name.len, old_dentry->d_name.name);
1590                 }
1591                 goto bail;
1592         }
1593
1594         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1595         if (status < 0) {
1596                 mlog_errno(status);
1597                 if (!is_journal_aborted(osb->journal->j_journal)) {
1598                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1599                                         "is not deleted.",
1600                                         new_dentry->d_name.len, new_dentry->d_name.name,
1601                                         old_dentry->d_name.len, old_dentry->d_name.name);
1602                 }
1603                 goto bail;
1604         }
1605
1606         if (new_inode) {
1607                 drop_nlink(new_inode);
1608                 new_inode->i_ctime = CURRENT_TIME;
1609         }
1610         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1611
1612         if (update_dot_dot) {
1613                 status = ocfs2_update_entry(old_inode, handle,
1614                                             &old_inode_dot_dot_res, new_dir);
1615                 drop_nlink(old_dir);
1616                 if (new_inode) {
1617                         drop_nlink(new_inode);
1618                 } else {
1619                         inc_nlink(new_dir);
1620                         mark_inode_dirty(new_dir);
1621                 }
1622         }
1623         mark_inode_dirty(old_dir);
1624         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1625         if (new_inode) {
1626                 mark_inode_dirty(new_inode);
1627                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1628         }
1629
1630         if (old_dir != new_dir) {
1631                 /* Keep the same times on both directories.*/
1632                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1633
1634                 /*
1635                  * This will also pick up the i_nlink change from the
1636                  * block above.
1637                  */
1638                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1639         }
1640
1641         if (old_dir_nlink != old_dir->i_nlink) {
1642                 if (!old_dir_bh) {
1643                         mlog(ML_ERROR, "need to change nlink for old dir "
1644                              "%llu from %d to %d but bh is NULL!\n",
1645                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1646                              (int)old_dir_nlink, old_dir->i_nlink);
1647                 } else {
1648                         struct ocfs2_dinode *fe;
1649                         status = ocfs2_journal_access_di(handle,
1650                                                          INODE_CACHE(old_dir),
1651                                                          old_dir_bh,
1652                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1653                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1654                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1655                         ocfs2_journal_dirty(handle, old_dir_bh);
1656                 }
1657         }
1658         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1659         status = 0;
1660 bail:
1661         if (handle)
1662                 ocfs2_commit_trans(osb, handle);
1663
1664         if (orphan_dir) {
1665                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1666                 ocfs2_inode_unlock(orphan_dir, 1);
1667                 mutex_unlock(&orphan_dir->i_mutex);
1668                 iput(orphan_dir);
1669         }
1670
1671         if (new_child_locked)
1672                 ocfs2_inode_unlock(new_inode, 1);
1673
1674         if (old_child_locked)
1675                 ocfs2_inode_unlock(old_inode, 1);
1676
1677         if (parents_locked)
1678                 ocfs2_double_unlock(old_dir, new_dir);
1679
1680         if (rename_lock)
1681                 ocfs2_rename_unlock(osb);
1682
1683         if (new_inode)
1684                 sync_mapping_buffers(old_inode->i_mapping);
1685
1686         if (new_inode)
1687                 iput(new_inode);
1688
1689         ocfs2_free_dir_lookup_result(&target_lookup_res);
1690         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1691         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1692         ocfs2_free_dir_lookup_result(&orphan_insert);
1693         ocfs2_free_dir_lookup_result(&target_insert);
1694
1695         brelse(newfe_bh);
1696         brelse(old_inode_bh);
1697         brelse(old_dir_bh);
1698         brelse(new_dir_bh);
1699
1700         if (status)
1701                 mlog_errno(status);
1702
1703         return status;
1704 }
1705
1706 /*
1707  * we expect i_size = strlen(symname). Copy symname into the file
1708  * data, including the null terminator.
1709  */
1710 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1711                                      handle_t *handle,
1712                                      struct inode *inode,
1713                                      const char *symname)
1714 {
1715         struct buffer_head **bhs = NULL;
1716         const char *c;
1717         struct super_block *sb = osb->sb;
1718         u64 p_blkno, p_blocks;
1719         int virtual, blocks, status, i, bytes_left;
1720
1721         bytes_left = i_size_read(inode) + 1;
1722         /* we can't trust i_blocks because we're actually going to
1723          * write i_size + 1 bytes. */
1724         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1725
1726         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1727                                         i_size_read(inode), blocks);
1728
1729         /* Sanity check -- make sure we're going to fit. */
1730         if (bytes_left >
1731             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1732                 status = -EIO;
1733                 mlog_errno(status);
1734                 goto bail;
1735         }
1736
1737         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1738         if (!bhs) {
1739                 status = -ENOMEM;
1740                 mlog_errno(status);
1741                 goto bail;
1742         }
1743
1744         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1745                                              NULL);
1746         if (status < 0) {
1747                 mlog_errno(status);
1748                 goto bail;
1749         }
1750
1751         /* links can never be larger than one cluster so we know this
1752          * is all going to be contiguous, but do a sanity check
1753          * anyway. */
1754         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1755                 status = -EIO;
1756                 mlog_errno(status);
1757                 goto bail;
1758         }
1759
1760         virtual = 0;
1761         while(bytes_left > 0) {
1762                 c = &symname[virtual * sb->s_blocksize];
1763
1764                 bhs[virtual] = sb_getblk(sb, p_blkno);
1765                 if (!bhs[virtual]) {
1766                         status = -ENOMEM;
1767                         mlog_errno(status);
1768                         goto bail;
1769                 }
1770                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1771                                               bhs[virtual]);
1772
1773                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1774                                               bhs[virtual],
1775                                               OCFS2_JOURNAL_ACCESS_CREATE);
1776                 if (status < 0) {
1777                         mlog_errno(status);
1778                         goto bail;
1779                 }
1780
1781                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1782
1783                 memcpy(bhs[virtual]->b_data, c,
1784                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1785                        bytes_left);
1786
1787                 ocfs2_journal_dirty(handle, bhs[virtual]);
1788
1789                 virtual++;
1790                 p_blkno++;
1791                 bytes_left -= sb->s_blocksize;
1792         }
1793
1794         status = 0;
1795 bail:
1796
1797         if (bhs) {
1798                 for(i = 0; i < blocks; i++)
1799                         brelse(bhs[i]);
1800                 kfree(bhs);
1801         }
1802
1803         if (status)
1804                 mlog_errno(status);
1805         return status;
1806 }
1807
1808 static int ocfs2_symlink(struct inode *dir,
1809                          struct dentry *dentry,
1810                          const char *symname)
1811 {
1812         int status, l, credits;
1813         u64 newsize;
1814         struct ocfs2_super *osb = NULL;
1815         struct inode *inode = NULL;
1816         struct super_block *sb;
1817         struct buffer_head *new_fe_bh = NULL;
1818         struct buffer_head *parent_fe_bh = NULL;
1819         struct ocfs2_dinode *fe = NULL;
1820         struct ocfs2_dinode *dirfe;
1821         handle_t *handle = NULL;
1822         struct ocfs2_alloc_context *inode_ac = NULL;
1823         struct ocfs2_alloc_context *data_ac = NULL;
1824         struct ocfs2_alloc_context *xattr_ac = NULL;
1825         int want_clusters = 0;
1826         int xattr_credits = 0;
1827         struct ocfs2_security_xattr_info si = {
1828                 .enable = 1,
1829         };
1830         int did_quota = 0, did_quota_inode = 0;
1831         struct ocfs2_dir_lookup_result lookup = { NULL, };
1832         sigset_t oldset;
1833         int did_block_signals = 0;
1834         struct ocfs2_dentry_lock *dl = NULL;
1835
1836         trace_ocfs2_symlink_begin(dir, dentry, symname,
1837                                   dentry->d_name.len, dentry->d_name.name);
1838
1839         status = dquot_initialize(dir);
1840         if (status) {
1841                 mlog_errno(status);
1842                 goto bail;
1843         }
1844
1845         sb = dir->i_sb;
1846         osb = OCFS2_SB(sb);
1847
1848         l = strlen(symname) + 1;
1849
1850         credits = ocfs2_calc_symlink_credits(sb);
1851
1852         /* lock the parent directory */
1853         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1854         if (status < 0) {
1855                 if (status != -ENOENT)
1856                         mlog_errno(status);
1857                 return status;
1858         }
1859
1860         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1861         if (!ocfs2_read_links_count(dirfe)) {
1862                 /* can't make a file in a deleted directory. */
1863                 status = -ENOENT;
1864                 goto bail;
1865         }
1866
1867         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1868                                            dentry->d_name.len);
1869         if (status)
1870                 goto bail;
1871
1872         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1873                                               dentry->d_name.name,
1874                                               dentry->d_name.len, &lookup);
1875         if (status < 0) {
1876                 mlog_errno(status);
1877                 goto bail;
1878         }
1879
1880         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1881         if (status < 0) {
1882                 if (status != -ENOSPC)
1883                         mlog_errno(status);
1884                 goto bail;
1885         }
1886
1887         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1888         if (IS_ERR(inode)) {
1889                 status = PTR_ERR(inode);
1890                 inode = NULL;
1891                 mlog_errno(status);
1892                 goto bail;
1893         }
1894
1895         /* get security xattr */
1896         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1897         if (status) {
1898                 if (status == -EOPNOTSUPP)
1899                         si.enable = 0;
1900                 else {
1901                         mlog_errno(status);
1902                         goto bail;
1903                 }
1904         }
1905
1906         /* calculate meta data/clusters for setting security xattr */
1907         if (si.enable) {
1908                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1909                                                   &xattr_credits, &xattr_ac);
1910                 if (status < 0) {
1911                         mlog_errno(status);
1912                         goto bail;
1913                 }
1914         }
1915
1916         /* don't reserve bitmap space for fast symlinks. */
1917         if (l > ocfs2_fast_symlink_chars(sb))
1918                 want_clusters += 1;
1919
1920         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1921         if (status < 0) {
1922                 if (status != -ENOSPC)
1923                         mlog_errno(status);
1924                 goto bail;
1925         }
1926
1927         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1928         if (IS_ERR(handle)) {
1929                 status = PTR_ERR(handle);
1930                 handle = NULL;
1931                 mlog_errno(status);
1932                 goto bail;
1933         }
1934
1935         /* Starting to change things, restart is no longer possible. */
1936         ocfs2_block_signals(&oldset);
1937         did_block_signals = 1;
1938
1939         status = dquot_alloc_inode(inode);
1940         if (status)
1941                 goto bail;
1942         did_quota_inode = 1;
1943
1944         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1945                                    dentry->d_name.name,
1946                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1947                                    inode->i_mode);
1948
1949         status = ocfs2_mknod_locked(osb, dir, inode,
1950                                     0, &new_fe_bh, parent_fe_bh, handle,
1951                                     inode_ac);
1952         if (status < 0) {
1953                 mlog_errno(status);
1954                 goto bail;
1955         }
1956
1957         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1958         inode->i_rdev = 0;
1959         newsize = l - 1;
1960         inode->i_op = &ocfs2_symlink_inode_operations;
1961         if (l > ocfs2_fast_symlink_chars(sb)) {
1962                 u32 offset = 0;
1963
1964                 status = dquot_alloc_space_nodirty(inode,
1965                     ocfs2_clusters_to_bytes(osb->sb, 1));
1966                 if (status)
1967                         goto bail;
1968                 did_quota = 1;
1969                 inode->i_mapping->a_ops = &ocfs2_aops;
1970                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1971                                               new_fe_bh,
1972                                               handle, data_ac, NULL,
1973                                               NULL);
1974                 if (status < 0) {
1975                         if (status != -ENOSPC && status != -EINTR) {
1976                                 mlog(ML_ERROR,
1977                                      "Failed to extend file to %llu\n",
1978                                      (unsigned long long)newsize);
1979                                 mlog_errno(status);
1980                                 status = -ENOSPC;
1981                         }
1982                         goto bail;
1983                 }
1984                 i_size_write(inode, newsize);
1985                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1986         } else {
1987                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1988                 memcpy((char *) fe->id2.i_symlink, symname, l);
1989                 i_size_write(inode, newsize);
1990                 inode->i_blocks = 0;
1991         }
1992
1993         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1994         if (status < 0) {
1995                 mlog_errno(status);
1996                 goto bail;
1997         }
1998
1999         if (!ocfs2_inode_is_fast_symlink(inode)) {
2000                 status = ocfs2_create_symlink_data(osb, handle, inode,
2001                                                    symname);
2002                 if (status < 0) {
2003                         mlog_errno(status);
2004                         goto bail;
2005                 }
2006         }
2007
2008         if (si.enable) {
2009                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
2010                                                  xattr_ac, data_ac);
2011                 if (status < 0) {
2012                         mlog_errno(status);
2013                         goto bail;
2014                 }
2015         }
2016
2017         /*
2018          * Do this before adding the entry to the directory. We add
2019          * also set d_op after success so that ->d_iput() will cleanup
2020          * the dentry lock even if ocfs2_add_entry() fails below.
2021          */
2022         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2023         if (status) {
2024                 mlog_errno(status);
2025                 goto bail;
2026         }
2027
2028         dl = dentry->d_fsdata;
2029
2030         status = ocfs2_add_entry(handle, dentry, inode,
2031                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2032                                  &lookup);
2033         if (status < 0) {
2034                 mlog_errno(status);
2035                 goto bail;
2036         }
2037
2038         insert_inode_hash(inode);
2039         d_instantiate(dentry, inode);
2040 bail:
2041         if (status < 0 && did_quota)
2042                 dquot_free_space_nodirty(inode,
2043                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2044         if (status < 0 && did_quota_inode)
2045                 dquot_free_inode(inode);
2046         if (handle)
2047                 ocfs2_commit_trans(osb, handle);
2048
2049         ocfs2_inode_unlock(dir, 1);
2050         if (did_block_signals)
2051                 ocfs2_unblock_signals(&oldset);
2052
2053         brelse(new_fe_bh);
2054         brelse(parent_fe_bh);
2055         kfree(si.value);
2056         ocfs2_free_dir_lookup_result(&lookup);
2057         if (inode_ac)
2058                 ocfs2_free_alloc_context(inode_ac);
2059         if (data_ac)
2060                 ocfs2_free_alloc_context(data_ac);
2061         if (xattr_ac)
2062                 ocfs2_free_alloc_context(xattr_ac);
2063         if ((status < 0) && inode) {
2064                 if (dl)
2065                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2066
2067                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2068                 clear_nlink(inode);
2069                 iput(inode);
2070         }
2071
2072         if (status)
2073                 mlog_errno(status);
2074
2075         return status;
2076 }
2077
2078 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2079 {
2080         int status, namelen;
2081
2082         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2083                            (long long)blkno);
2084         if (namelen <= 0) {
2085                 if (namelen)
2086                         status = namelen;
2087                 else
2088                         status = -EINVAL;
2089                 mlog_errno(status);
2090                 goto bail;
2091         }
2092         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2093                 status = -EINVAL;
2094                 mlog_errno(status);
2095                 goto bail;
2096         }
2097
2098         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2099
2100         status = 0;
2101 bail:
2102         if (status < 0)
2103                 mlog_errno(status);
2104         return status;
2105 }
2106
2107 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2108                                         struct inode **ret_orphan_dir,
2109                                         struct buffer_head **ret_orphan_dir_bh)
2110 {
2111         struct inode *orphan_dir_inode;
2112         struct buffer_head *orphan_dir_bh = NULL;
2113         int ret = 0;
2114
2115         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2116                                                        ORPHAN_DIR_SYSTEM_INODE,
2117                                                        osb->slot_num);
2118         if (!orphan_dir_inode) {
2119                 ret = -ENOENT;
2120                 mlog_errno(ret);
2121                 return ret;
2122         }
2123
2124         mutex_lock(&orphan_dir_inode->i_mutex);
2125
2126         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2127         if (ret < 0) {
2128                 mutex_unlock(&orphan_dir_inode->i_mutex);
2129                 iput(orphan_dir_inode);
2130
2131                 mlog_errno(ret);
2132                 return ret;
2133         }
2134
2135         *ret_orphan_dir = orphan_dir_inode;
2136         *ret_orphan_dir_bh = orphan_dir_bh;
2137
2138         return 0;
2139 }
2140
2141 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2142                                       struct buffer_head *orphan_dir_bh,
2143                                       u64 blkno,
2144                                       char *name,
2145                                       struct ocfs2_dir_lookup_result *lookup,
2146                                       bool dio)
2147 {
2148         int ret;
2149         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2150         int namelen = dio ?
2151                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2152                         OCFS2_ORPHAN_NAMELEN;
2153
2154         if (dio) {
2155                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2156                                 OCFS2_DIO_ORPHAN_PREFIX);
2157                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2158                         ret = -EINVAL;
2159                         mlog_errno(ret);
2160                         return ret;
2161                 }
2162
2163                 ret = ocfs2_blkno_stringify(blkno,
2164                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2165         } else
2166                 ret = ocfs2_blkno_stringify(blkno, name);
2167         if (ret < 0) {
2168                 mlog_errno(ret);
2169                 return ret;
2170         }
2171
2172         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2173                                            orphan_dir_bh, name,
2174                                            namelen, lookup);
2175         if (ret < 0) {
2176                 mlog_errno(ret);
2177                 return ret;
2178         }
2179
2180         return 0;
2181 }
2182
2183 /**
2184  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2185  * insertion of an orphan.
2186  * @osb: ocfs2 file system
2187  * @ret_orphan_dir: Orphan dir inode - returned locked!
2188  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2189  * @lookup: dir lookup result, to be passed back into functions like
2190  *          ocfs2_orphan_add
2191  *
2192  * Returns zero on success and the ret_orphan_dir, name and lookup
2193  * fields will be populated.
2194  *
2195  * Returns non-zero on failure.
2196  */
2197 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2198                                     struct inode **ret_orphan_dir,
2199                                     u64 blkno,
2200                                     char *name,
2201                                     struct ocfs2_dir_lookup_result *lookup,
2202                                     bool dio)
2203 {
2204         struct inode *orphan_dir_inode = NULL;
2205         struct buffer_head *orphan_dir_bh = NULL;
2206         int ret = 0;
2207
2208         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2209                                            &orphan_dir_bh);
2210         if (ret < 0) {
2211                 mlog_errno(ret);
2212                 return ret;
2213         }
2214
2215         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2216                                          blkno, name, lookup, dio);
2217         if (ret < 0) {
2218                 mlog_errno(ret);
2219                 goto out;
2220         }
2221
2222         *ret_orphan_dir = orphan_dir_inode;
2223
2224 out:
2225         brelse(orphan_dir_bh);
2226
2227         if (ret) {
2228                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2229                 mutex_unlock(&orphan_dir_inode->i_mutex);
2230                 iput(orphan_dir_inode);
2231         }
2232
2233         if (ret)
2234                 mlog_errno(ret);
2235         return ret;
2236 }
2237
2238 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2239                             handle_t *handle,
2240                             struct inode *inode,
2241                             struct buffer_head *fe_bh,
2242                             char *name,
2243                             struct ocfs2_dir_lookup_result *lookup,
2244                             struct inode *orphan_dir_inode,
2245                             bool dio)
2246 {
2247         struct buffer_head *orphan_dir_bh = NULL;
2248         int status = 0;
2249         struct ocfs2_dinode *orphan_fe;
2250         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2251         int namelen = dio ?
2252                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2253                         OCFS2_ORPHAN_NAMELEN;
2254
2255         trace_ocfs2_orphan_add_begin(
2256                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2257
2258         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2259         if (status < 0) {
2260                 mlog_errno(status);
2261                 goto leave;
2262         }
2263
2264         status = ocfs2_journal_access_di(handle,
2265                                          INODE_CACHE(orphan_dir_inode),
2266                                          orphan_dir_bh,
2267                                          OCFS2_JOURNAL_ACCESS_WRITE);
2268         if (status < 0) {
2269                 mlog_errno(status);
2270                 goto leave;
2271         }
2272
2273         /*
2274          * We're going to journal the change of i_flags and i_orphaned_slot.
2275          * It's safe anyway, though some callers may duplicate the journaling.
2276          * Journaling within the func just make the logic look more
2277          * straightforward.
2278          */
2279         status = ocfs2_journal_access_di(handle,
2280                                          INODE_CACHE(inode),
2281                                          fe_bh,
2282                                          OCFS2_JOURNAL_ACCESS_WRITE);
2283         if (status < 0) {
2284                 mlog_errno(status);
2285                 goto leave;
2286         }
2287
2288         /* we're a cluster, and nlink can change on disk from
2289          * underneath us... */
2290         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2291         if (S_ISDIR(inode->i_mode))
2292                 ocfs2_add_links_count(orphan_fe, 1);
2293         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2294         ocfs2_journal_dirty(handle, orphan_dir_bh);
2295
2296         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2297                                    namelen, inode,
2298                                    OCFS2_I(inode)->ip_blkno,
2299                                    orphan_dir_bh, lookup);
2300         if (status < 0) {
2301                 mlog_errno(status);
2302                 goto rollback;
2303         }
2304
2305         if (dio) {
2306                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2307                  * slot.
2308                  */
2309                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2310                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2311         } else {
2312                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2313                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2314
2315                 /* Record which orphan dir our inode now resides
2316                  * in. delete_inode will use this to determine which orphan
2317                  * dir to lock. */
2318                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2319         }
2320
2321         ocfs2_journal_dirty(handle, fe_bh);
2322
2323         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2324                                    osb->slot_num);
2325
2326 rollback:
2327         if (status < 0) {
2328                 if (S_ISDIR(inode->i_mode))
2329                         ocfs2_add_links_count(orphan_fe, -1);
2330                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2331         }
2332
2333 leave:
2334         brelse(orphan_dir_bh);
2335
2336         return status;
2337 }
2338
2339 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2340 int ocfs2_orphan_del(struct ocfs2_super *osb,
2341                      handle_t *handle,
2342                      struct inode *orphan_dir_inode,
2343                      struct inode *inode,
2344                      struct buffer_head *orphan_dir_bh,
2345                      bool dio)
2346 {
2347         const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
2348         char name[namelen + 1];
2349         struct ocfs2_dinode *orphan_fe;
2350         int status = 0;
2351         struct ocfs2_dir_lookup_result lookup = { NULL, };
2352
2353         if (dio) {
2354                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2355                                 OCFS2_DIO_ORPHAN_PREFIX);
2356                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2357                         status = -EINVAL;
2358                         mlog_errno(status);
2359                         return status;
2360                 }
2361
2362                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2363                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2364         } else
2365                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2366         if (status < 0) {
2367                 mlog_errno(status);
2368                 goto leave;
2369         }
2370
2371         trace_ocfs2_orphan_del(
2372              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2373              name, strlen(name));
2374
2375         /* find it's spot in the orphan directory */
2376         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2377                                   &lookup);
2378         if (status) {
2379                 mlog_errno(status);
2380                 goto leave;
2381         }
2382
2383         /* remove it from the orphan directory */
2384         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2385         if (status < 0) {
2386                 mlog_errno(status);
2387                 goto leave;
2388         }
2389
2390         status = ocfs2_journal_access_di(handle,
2391                                          INODE_CACHE(orphan_dir_inode),
2392                                          orphan_dir_bh,
2393                                          OCFS2_JOURNAL_ACCESS_WRITE);
2394         if (status < 0) {
2395                 mlog_errno(status);
2396                 goto leave;
2397         }
2398
2399         /* do the i_nlink dance! :) */
2400         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2401         if (S_ISDIR(inode->i_mode))
2402                 ocfs2_add_links_count(orphan_fe, -1);
2403         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2404         ocfs2_journal_dirty(handle, orphan_dir_bh);
2405
2406 leave:
2407         ocfs2_free_dir_lookup_result(&lookup);
2408
2409         if (status)
2410                 mlog_errno(status);
2411         return status;
2412 }
2413
2414 /**
2415  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2416  * allocated file. This is different from the typical 'add to orphan dir'
2417  * operation in that the inode does not yet exist. This is a problem because
2418  * the orphan dir stringifies the inode block number to come up with it's
2419  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2420  * problem. This function works around it by calling deeper into the orphan
2421  * and suballoc code than other callers. Use this only by necessity.
2422  * @dir: The directory which this inode will ultimately wind up under - not the
2423  * orphan dir!
2424  * @dir_bh: buffer_head the @dir inode block
2425  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2426  * with the string to be used for orphan dirent. Pass back to the orphan dir
2427  * code.
2428  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2429  * dir code.
2430  * @ret_di_blkno: block number where the new inode will be allocated.
2431  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2432  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2433  *
2434  * Returns zero on success and the ret_orphan_dir, name and lookup
2435  * fields will be populated.
2436  *
2437  * Returns non-zero on failure.
2438  */
2439 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2440                                         struct buffer_head *dir_bh,
2441                                         char *orphan_name,
2442                                         struct inode **ret_orphan_dir,
2443                                         u64 *ret_di_blkno,
2444                                         struct ocfs2_dir_lookup_result *orphan_insert,
2445                                         struct ocfs2_alloc_context **ret_inode_ac)
2446 {
2447         int ret;
2448         u64 di_blkno;
2449         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2450         struct inode *orphan_dir = NULL;
2451         struct buffer_head *orphan_dir_bh = NULL;
2452         struct ocfs2_alloc_context *inode_ac = NULL;
2453
2454         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2455         if (ret < 0) {
2456                 mlog_errno(ret);
2457                 return ret;
2458         }
2459
2460         /* reserve an inode spot */
2461         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2462         if (ret < 0) {
2463                 if (ret != -ENOSPC)
2464                         mlog_errno(ret);
2465                 goto out;
2466         }
2467
2468         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2469                                        &di_blkno);
2470         if (ret) {
2471                 mlog_errno(ret);
2472                 goto out;
2473         }
2474
2475         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2476                                          di_blkno, orphan_name, orphan_insert,
2477                                          false);
2478         if (ret < 0) {
2479                 mlog_errno(ret);
2480                 goto out;
2481         }
2482
2483 out:
2484         if (ret == 0) {
2485                 *ret_orphan_dir = orphan_dir;
2486                 *ret_di_blkno = di_blkno;
2487                 *ret_inode_ac = inode_ac;
2488                 /*
2489                  * orphan_name and orphan_insert are already up to
2490                  * date via prepare_orphan_dir
2491                  */
2492         } else {
2493                 /* Unroll reserve_new_inode* */
2494                 if (inode_ac)
2495                         ocfs2_free_alloc_context(inode_ac);
2496
2497                 /* Unroll orphan dir locking */
2498                 mutex_unlock(&orphan_dir->i_mutex);
2499                 ocfs2_inode_unlock(orphan_dir, 1);
2500                 iput(orphan_dir);
2501         }
2502
2503         brelse(orphan_dir_bh);
2504
2505         return ret;
2506 }
2507
2508 int ocfs2_create_inode_in_orphan(struct inode *dir,
2509                                  int mode,
2510                                  struct inode **new_inode)
2511 {
2512         int status, did_quota_inode = 0;
2513         struct inode *inode = NULL;
2514         struct inode *orphan_dir = NULL;
2515         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2516         struct ocfs2_dinode *di = NULL;
2517         handle_t *handle = NULL;
2518         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2519         struct buffer_head *parent_di_bh = NULL;
2520         struct buffer_head *new_di_bh = NULL;
2521         struct ocfs2_alloc_context *inode_ac = NULL;
2522         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2523         u64 uninitialized_var(di_blkno), suballoc_loc;
2524         u16 suballoc_bit;
2525
2526         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2527         if (status < 0) {
2528                 if (status != -ENOENT)
2529                         mlog_errno(status);
2530                 return status;
2531         }
2532
2533         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2534                                               orphan_name, &orphan_dir,
2535                                               &di_blkno, &orphan_insert, &inode_ac);
2536         if (status < 0) {
2537                 if (status != -ENOSPC)
2538                         mlog_errno(status);
2539                 goto leave;
2540         }
2541
2542         inode = ocfs2_get_init_inode(dir, mode);
2543         if (IS_ERR(inode)) {
2544                 status = PTR_ERR(inode);
2545                 inode = NULL;
2546                 mlog_errno(status);
2547                 goto leave;
2548         }
2549
2550         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2551         if (IS_ERR(handle)) {
2552                 status = PTR_ERR(handle);
2553                 handle = NULL;
2554                 mlog_errno(status);
2555                 goto leave;
2556         }
2557
2558         status = dquot_alloc_inode(inode);
2559         if (status)
2560                 goto leave;
2561         did_quota_inode = 1;
2562
2563         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2564                                               &suballoc_loc,
2565                                               &suballoc_bit, di_blkno);
2566         if (status < 0) {
2567                 mlog_errno(status);
2568                 goto leave;
2569         }
2570
2571         clear_nlink(inode);
2572         /* do the real work now. */
2573         status = __ocfs2_mknod_locked(dir, inode,
2574                                       0, &new_di_bh, parent_di_bh, handle,
2575                                       inode_ac, di_blkno, suballoc_loc,
2576                                       suballoc_bit);
2577         if (status < 0) {
2578                 mlog_errno(status);
2579                 goto leave;
2580         }
2581
2582         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2583         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2584                                   &orphan_insert, orphan_dir, false);
2585         if (status < 0) {
2586                 mlog_errno(status);
2587                 goto leave;
2588         }
2589
2590         /* get open lock so that only nodes can't remove it from orphan dir. */
2591         status = ocfs2_open_lock(inode);
2592         if (status < 0)
2593                 mlog_errno(status);
2594
2595         insert_inode_hash(inode);
2596 leave:
2597         if (status < 0 && did_quota_inode)
2598                 dquot_free_inode(inode);
2599         if (handle)
2600                 ocfs2_commit_trans(osb, handle);
2601
2602         if (orphan_dir) {
2603                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2604                 ocfs2_inode_unlock(orphan_dir, 1);
2605                 mutex_unlock(&orphan_dir->i_mutex);
2606                 iput(orphan_dir);
2607         }
2608
2609         if ((status < 0) && inode) {
2610                 clear_nlink(inode);
2611                 iput(inode);
2612         }
2613
2614         if (inode_ac)
2615                 ocfs2_free_alloc_context(inode_ac);
2616
2617         brelse(new_di_bh);
2618
2619         if (!status)
2620                 *new_inode = inode;
2621
2622         ocfs2_free_dir_lookup_result(&orphan_insert);
2623
2624         ocfs2_inode_unlock(dir, 1);
2625         brelse(parent_di_bh);
2626         return status;
2627 }
2628
2629 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2630         struct inode *inode)
2631 {
2632         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2633         struct inode *orphan_dir_inode = NULL;
2634         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2635         struct buffer_head *di_bh = NULL;
2636         int status = 0;
2637         handle_t *handle = NULL;
2638         struct ocfs2_dinode *di = NULL;
2639
2640         status = ocfs2_inode_lock(inode, &di_bh, 1);
2641         if (status < 0) {
2642                 mlog_errno(status);
2643                 goto bail;
2644         }
2645
2646         di = (struct ocfs2_dinode *) di_bh->b_data;
2647         /*
2648          * Another append dio crashed?
2649          * If so, manually recover it first.
2650          */
2651         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2652                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2653                 if (status < 0) {
2654                         if (status != -ENOSPC)
2655                                 mlog_errno(status);
2656                         goto bail_unlock_inode;
2657                 }
2658
2659                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2660                 if (status < 0) {
2661                         mlog_errno(status);
2662                         goto bail_unlock_inode;
2663                 }
2664         }
2665
2666         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2667                         OCFS2_I(inode)->ip_blkno,
2668                         orphan_name,
2669                         &orphan_insert,
2670                         true);
2671         if (status < 0) {
2672                 mlog_errno(status);
2673                 goto bail_unlock_inode;
2674         }
2675
2676         handle = ocfs2_start_trans(osb,
2677                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2678         if (IS_ERR(handle)) {
2679                 status = PTR_ERR(handle);
2680                 goto bail_unlock_orphan;
2681         }
2682
2683         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2684                         &orphan_insert, orphan_dir_inode, true);
2685         if (status)
2686                 mlog_errno(status);
2687
2688         ocfs2_commit_trans(osb, handle);
2689
2690 bail_unlock_orphan:
2691         ocfs2_inode_unlock(orphan_dir_inode, 1);
2692         mutex_unlock(&orphan_dir_inode->i_mutex);
2693         iput(orphan_dir_inode);
2694
2695         ocfs2_free_dir_lookup_result(&orphan_insert);
2696
2697 bail_unlock_inode:
2698         ocfs2_inode_unlock(inode, 1);
2699         brelse(di_bh);
2700
2701 bail:
2702         return status;
2703 }
2704
2705 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2706                 struct inode *inode, struct buffer_head *di_bh,
2707                 int update_isize, loff_t end)
2708 {
2709         struct inode *orphan_dir_inode = NULL;
2710         struct buffer_head *orphan_dir_bh = NULL;
2711         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2712         handle_t *handle = NULL;
2713         int status = 0;
2714
2715         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2716                         ORPHAN_DIR_SYSTEM_INODE,
2717                         le16_to_cpu(di->i_dio_orphaned_slot));
2718         if (!orphan_dir_inode) {
2719                 status = -ENOENT;
2720                 mlog_errno(status);
2721                 goto bail;
2722         }
2723
2724         mutex_lock(&orphan_dir_inode->i_mutex);
2725         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2726         if (status < 0) {
2727                 mutex_unlock(&orphan_dir_inode->i_mutex);
2728                 iput(orphan_dir_inode);
2729                 mlog_errno(status);
2730                 goto bail;
2731         }
2732
2733         handle = ocfs2_start_trans(osb,
2734                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2735         if (IS_ERR(handle)) {
2736                 status = PTR_ERR(handle);
2737                 goto bail_unlock_orphan;
2738         }
2739
2740         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2741
2742         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2743                                 inode, orphan_dir_bh, true);
2744         if (status < 0) {
2745                 mlog_errno(status);
2746                 goto bail_commit;
2747         }
2748
2749         status = ocfs2_journal_access_di(handle,
2750                         INODE_CACHE(inode),
2751                         di_bh,
2752                         OCFS2_JOURNAL_ACCESS_WRITE);
2753         if (status < 0) {
2754                 mlog_errno(status);
2755                 goto bail_commit;
2756         }
2757
2758         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2759         di->i_dio_orphaned_slot = 0;
2760
2761         if (update_isize) {
2762                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2763                 if (status)
2764                         mlog_errno(status);
2765         } else
2766                 ocfs2_journal_dirty(handle, di_bh);
2767
2768 bail_commit:
2769         ocfs2_commit_trans(osb, handle);
2770
2771 bail_unlock_orphan:
2772         ocfs2_inode_unlock(orphan_dir_inode, 1);
2773         mutex_unlock(&orphan_dir_inode->i_mutex);
2774         brelse(orphan_dir_bh);
2775         iput(orphan_dir_inode);
2776
2777 bail:
2778         return status;
2779 }
2780
2781 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2782                                    struct inode *inode,
2783                                    struct dentry *dentry)
2784 {
2785         int status = 0;
2786         struct buffer_head *parent_di_bh = NULL;
2787         handle_t *handle = NULL;
2788         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2789         struct ocfs2_dinode *dir_di, *di;
2790         struct inode *orphan_dir_inode = NULL;
2791         struct buffer_head *orphan_dir_bh = NULL;
2792         struct buffer_head *di_bh = NULL;
2793         struct ocfs2_dir_lookup_result lookup = { NULL, };
2794
2795         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2796                                 dentry->d_name.len, dentry->d_name.name,
2797                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2798                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2799
2800         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2801         if (status < 0) {
2802                 if (status != -ENOENT)
2803                         mlog_errno(status);
2804                 return status;
2805         }
2806
2807         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2808         if (!dir_di->i_links_count) {
2809                 /* can't make a file in a deleted directory. */
2810                 status = -ENOENT;
2811                 goto leave;
2812         }
2813
2814         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2815                                            dentry->d_name.len);
2816         if (status)
2817                 goto leave;
2818
2819         /* get a spot inside the dir. */
2820         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2821                                               dentry->d_name.name,
2822                                               dentry->d_name.len, &lookup);
2823         if (status < 0) {
2824                 mlog_errno(status);
2825                 goto leave;
2826         }
2827
2828         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2829                                                        ORPHAN_DIR_SYSTEM_INODE,
2830                                                        osb->slot_num);
2831         if (!orphan_dir_inode) {
2832                 status = -ENOENT;
2833                 mlog_errno(status);
2834                 goto leave;
2835         }
2836
2837         mutex_lock(&orphan_dir_inode->i_mutex);
2838
2839         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2840         if (status < 0) {
2841                 mlog_errno(status);
2842                 mutex_unlock(&orphan_dir_inode->i_mutex);
2843                 iput(orphan_dir_inode);
2844                 goto leave;
2845         }
2846
2847         status = ocfs2_read_inode_block(inode, &di_bh);
2848         if (status < 0) {
2849                 mlog_errno(status);
2850                 goto orphan_unlock;
2851         }
2852
2853         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2854         if (IS_ERR(handle)) {
2855                 status = PTR_ERR(handle);
2856                 handle = NULL;
2857                 mlog_errno(status);
2858                 goto orphan_unlock;
2859         }
2860
2861         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2862                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2863         if (status < 0) {
2864                 mlog_errno(status);
2865                 goto out_commit;
2866         }
2867
2868         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2869                                   orphan_dir_bh, false);
2870         if (status < 0) {
2871                 mlog_errno(status);
2872                 goto out_commit;
2873         }
2874
2875         di = (struct ocfs2_dinode *)di_bh->b_data;
2876         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2877         di->i_orphaned_slot = 0;
2878         set_nlink(inode, 1);
2879         ocfs2_set_links_count(di, inode->i_nlink);
2880         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2881         ocfs2_journal_dirty(handle, di_bh);
2882
2883         status = ocfs2_add_entry(handle, dentry, inode,
2884                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2885                                  &lookup);
2886         if (status < 0) {
2887                 mlog_errno(status);
2888                 goto out_commit;
2889         }
2890
2891         status = ocfs2_dentry_attach_lock(dentry, inode,
2892                                           OCFS2_I(dir)->ip_blkno);
2893         if (status) {
2894                 mlog_errno(status);
2895                 goto out_commit;
2896         }
2897
2898         d_instantiate(dentry, inode);
2899         status = 0;
2900 out_commit:
2901         ocfs2_commit_trans(osb, handle);
2902 orphan_unlock:
2903         ocfs2_inode_unlock(orphan_dir_inode, 1);
2904         mutex_unlock(&orphan_dir_inode->i_mutex);
2905         iput(orphan_dir_inode);
2906 leave:
2907
2908         ocfs2_inode_unlock(dir, 1);
2909
2910         brelse(di_bh);
2911         brelse(parent_di_bh);
2912         brelse(orphan_dir_bh);
2913
2914         ocfs2_free_dir_lookup_result(&lookup);
2915
2916         if (status)
2917                 mlog_errno(status);
2918
2919         return status;
2920 }
2921
2922 const struct inode_operations ocfs2_dir_iops = {
2923         .create         = ocfs2_create,
2924         .lookup         = ocfs2_lookup,
2925         .link           = ocfs2_link,
2926         .unlink         = ocfs2_unlink,
2927         .rmdir          = ocfs2_unlink,
2928         .symlink        = ocfs2_symlink,
2929         .mkdir          = ocfs2_mkdir,
2930         .mknod          = ocfs2_mknod,
2931         .rename         = ocfs2_rename,
2932         .setattr        = ocfs2_setattr,
2933         .getattr        = ocfs2_getattr,
2934         .permission     = ocfs2_permission,
2935         .setxattr       = generic_setxattr,
2936         .getxattr       = generic_getxattr,
2937         .listxattr      = ocfs2_listxattr,
2938         .removexattr    = generic_removexattr,
2939         .fiemap         = ocfs2_fiemap,
2940         .get_acl        = ocfs2_iop_get_acl,
2941         .set_acl        = ocfs2_iop_set_acl,
2942 };