fs/ocfs2/namei.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * namei.c
   6  *
   7  * Create and rename file, directory, symlinks
   8  *
   9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
  10  *
  11  *  Portions of this code from linux/fs/ext3/dir.c
  12  *
  13  *  Copyright (C) 1992, 1993, 1994, 1995
  14  *  Remy Card (card@masi.ibp.fr)
  15  *  Laboratoire MASI - Institut Blaise pascal
  16  *  Universite Pierre et Marie Curie (Paris VI)
  17  *
  18  *   from
  19  *
  20  *   linux/fs/minix/dir.c
  21  *
  22  *   Copyright (C) 1991, 1992 Linux Torvalds
  23  */
  24
  25 #include <linux/fs.h>
  26 #include <linux/types.h>
  27 #include <linux/slab.h>
  28 #include <linux/highmem.h>
  29 #include <linux/quotaops.h>
  30 #include <linux/iversion.h>
  31
  32 #include <cluster/masklog.h>
  33
  34 #include "ocfs2.h"
  35
  36 #include "alloc.h"
  37 #include "dcache.h"
  38 #include "dir.h"
  39 #include "dlmglue.h"
  40 #include "extent_map.h"
  41 #include "file.h"
  42 #include "inode.h"
  43 #include "journal.h"
  44 #include "namei.h"
  45 #include "suballoc.h"
  46 #include "super.h"
  47 #include "symlink.h"
  48 #include "sysfile.h"
  49 #include "uptodate.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "ocfs2_trace.h"
  53
  54 #include "buffer_head_io.h"
  55
  56 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  57                               struct inode *dir,
  58                               struct inode *inode,
  59                               dev_t dev,
  60                               struct buffer_head **new_fe_bh,
  61                               struct buffer_head *parent_fe_bh,
  62                               handle_t *handle,
  63                               struct ocfs2_alloc_context *inode_ac);
  64
  65 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  66                                     struct inode **ret_orphan_dir,
  67                                     u64 blkno,
  68                                     char *name,
  69                                     struct ocfs2_dir_lookup_result *lookup,
  70                                     bool dio);
  71
  72 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  73                             handle_t *handle,
  74                             struct inode *inode,
  75                             struct buffer_head *fe_bh,
  76                             char *name,
  77                             struct ocfs2_dir_lookup_result *lookup,
  78                             struct inode *orphan_dir_inode,
  79                             bool dio);
  80
  81 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  82                                      handle_t *handle,
  83                                      struct inode *inode,
  84                                      const char *symname);
  85
  86 static int ocfs2_double_lock(struct ocfs2_super *osb,
  87                              struct buffer_head **bh1,
  88                              struct inode *inode1,
  89                              struct buffer_head **bh2,
  90                              struct inode *inode2,
  91                              int rename);
  92
  93 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
  94 /* An orphan dir name is an 8 byte value, printed as a hex string */
  95 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  96
  97 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
  98                                    unsigned int flags)
  99 {
 100         int status;
 101         u64 blkno;
 102         struct inode *inode = NULL;
 103         struct dentry *ret;
 104         struct ocfs2_inode_info *oi;
 105
 106         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 107                            dentry->d_name.name,
 108                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 109
 110         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 111                 ret = ERR_PTR(-ENAMETOOLONG);
 112                 goto bail;
 113         }
 114
 115         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 116         if (status < 0) {
 117                 if (status != -ENOENT)
 118                         mlog_errno(status);
 119                 ret = ERR_PTR(status);
 120                 goto bail;
 121         }
 122
 123         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 124                                             dentry->d_name.len, &blkno);
 125         if (status < 0)
 126                 goto bail_add;
 127
 128         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 129         if (IS_ERR(inode)) {
 130                 ret = ERR_PTR(-EACCES);
 131                 goto bail_unlock;
 132         }
 133
 134         oi = OCFS2_I(inode);
 135         /* Clear any orphaned state... If we were able to look up the
 136          * inode from a directory, it certainly can't be orphaned. We
 137          * might have the bad state from a node which intended to
 138          * orphan this inode but crashed before it could commit the
 139          * unlink. */
 140         spin_lock(&oi->ip_lock);
 141         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 142         spin_unlock(&oi->ip_lock);
 143
 144 bail_add:
 145         ret = d_splice_alias(inode, dentry);
 146
 147         if (inode) {
 148                 /*
 149                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 150                  * dentry, it will d_move() it on top of ourse. The
 151                  * return value will indicate this however, so in
 152                  * those cases, we switch them around for the locking
 153                  * code.
 154                  *
 155                  * NOTE: This dentry already has ->d_op set from
 156                  * ocfs2_get_parent() and ocfs2_get_dentry()
 157                  */
 158                 if (!IS_ERR_OR_NULL(ret))
 159                         dentry = ret;
 160
 161                 status = ocfs2_dentry_attach_lock(dentry, inode,
 162                                                   OCFS2_I(dir)->ip_blkno);
 163                 if (status) {
 164                         mlog_errno(status);
 165                         ret = ERR_PTR(status);
 166                         goto bail_unlock;
 167                 }
 168         } else
 169                 ocfs2_dentry_attach_gen(dentry);
 170
 171 bail_unlock:
 172         /* Don't drop the cluster lock until *after* the d_add --
 173          * unlink on another node will message us to remove that
 174          * dentry under this lock so otherwise we can race this with
 175          * the downconvert thread and have a stale dentry. */
 176         ocfs2_inode_unlock(dir, 0);
 177
 178 bail:
 179
 180         trace_ocfs2_lookup_ret(ret);
 181
 182         return ret;
 183 }
 184
 185 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 186 {
 187         struct inode *inode;
 188         int status;
 189
 190         inode = new_inode(dir->i_sb);
 191         if (!inode) {
 192                 mlog(ML_ERROR, "new_inode failed!\n");
 193                 return ERR_PTR(-ENOMEM);
 194         }
 195
 196         /* populate as many fields early on as possible - many of
 197          * these are used by the support functions here and in
 198          * callers. */
 199         if (S_ISDIR(mode))
 200                 set_nlink(inode, 2);
 201         inode_init_owner(inode, dir, mode);
 202         status = dquot_initialize(inode);
 203         if (status)
 204                 return ERR_PTR(status);
 205
 206         return inode;
 207 }
 208
 209 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 210                 struct dentry *dentry, struct inode *inode)
 211 {
 212         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 213
 214         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 215         ocfs2_lock_res_free(&dl->dl_lockres);
 216         BUG_ON(dl->dl_count != 1);
 217         spin_lock(&dentry_attach_lock);
 218         dentry->d_fsdata = NULL;
 219         spin_unlock(&dentry_attach_lock);
 220         kfree(dl);
 221         iput(inode);
 222 }
 223
 224 static int ocfs2_mknod(struct inode *dir,
 225                        struct dentry *dentry,
 226                        umode_t mode,
 227                        dev_t dev)
 228 {
 229         int status = 0;
 230         struct buffer_head *parent_fe_bh = NULL;
 231         handle_t *handle = NULL;
 232         struct ocfs2_super *osb;
 233         struct ocfs2_dinode *dirfe;
 234         struct ocfs2_dinode *fe = NULL;
 235         struct buffer_head *new_fe_bh = NULL;
 236         struct inode *inode = NULL;
 237         struct ocfs2_alloc_context *inode_ac = NULL;
 238         struct ocfs2_alloc_context *data_ac = NULL;
 239         struct ocfs2_alloc_context *meta_ac = NULL;
 240         int want_clusters = 0;
 241         int want_meta = 0;
 242         int xattr_credits = 0;
 243         struct ocfs2_security_xattr_info si = {
 244                 .enable = 1,
 245         };
 246         int did_quota_inode = 0;
 247         struct ocfs2_dir_lookup_result lookup = { NULL, };
 248         sigset_t oldset;
 249         int did_block_signals = 0;
 250         struct ocfs2_dentry_lock *dl = NULL;
 251
 252         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 253                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 254                           (unsigned long)dev, mode);
 255
 256         status = dquot_initialize(dir);
 257         if (status) {
 258                 mlog_errno(status);
 259                 return status;
 260         }
 261
 262         /* get our super block */
 263         osb = OCFS2_SB(dir->i_sb);
 264
 265         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 266         if (status < 0) {
 267                 if (status != -ENOENT)
 268                         mlog_errno(status);
 269                 return status;
 270         }
 271
 272         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 273                 status = -EMLINK;
 274                 goto leave;
 275         }
 276
 277         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 278         if (!ocfs2_read_links_count(dirfe)) {
 279                 /* can't make a file in a deleted directory. */
 280                 status = -ENOENT;
 281                 goto leave;
 282         }
 283
 284         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 285                                            dentry->d_name.len);
 286         if (status)
 287                 goto leave;
 288
 289         /* get a spot inside the dir. */
 290         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 291                                               dentry->d_name.name,
 292                                               dentry->d_name.len, &lookup);
 293         if (status < 0) {
 294                 mlog_errno(status);
 295                 goto leave;
 296         }
 297
 298         /* reserve an inode spot */
 299         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 300         if (status < 0) {
 301                 if (status != -ENOSPC)
 302                         mlog_errno(status);
 303                 goto leave;
 304         }
 305
 306         inode = ocfs2_get_init_inode(dir, mode);
 307         if (IS_ERR(inode)) {
 308                 status = PTR_ERR(inode);
 309                 inode = NULL;
 310                 mlog_errno(status);
 311                 goto leave;
 312         }
 313
 314         /* get security xattr */
 315         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 316         if (status) {
 317                 if (status == -EOPNOTSUPP)
 318                         si.enable = 0;
 319                 else {
 320                         mlog_errno(status);
 321                         goto leave;
 322                 }
 323         }
 324
 325         /* calculate meta data/clusters for setting security and acl xattr */
 326         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 327                                        &si, &want_clusters,
 328                                        &xattr_credits, &want_meta);
 329         if (status < 0) {
 330                 mlog_errno(status);
 331                 goto leave;
 332         }
 333
 334         /* Reserve a cluster if creating an extent based directory. */
 335         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 336                 want_clusters += 1;
 337
 338                 /* Dir indexing requires extra space as well */
 339                 if (ocfs2_supports_indexed_dirs(osb))
 340                         want_meta++;
 341         }
 342
 343         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 344         if (status < 0) {
 345                 if (status != -ENOSPC)
 346                         mlog_errno(status);
 347                 goto leave;
 348         }
 349
 350         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 351         if (status < 0) {
 352                 if (status != -ENOSPC)
 353                         mlog_errno(status);
 354                 goto leave;
 355         }
 356
 357         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 358                                                             S_ISDIR(mode),
 359                                                             xattr_credits));
 360         if (IS_ERR(handle)) {
 361                 status = PTR_ERR(handle);
 362                 handle = NULL;
 363                 mlog_errno(status);
 364                 goto leave;
 365         }
 366
 367         /* Starting to change things, restart is no longer possible. */
 368         ocfs2_block_signals(&oldset);
 369         did_block_signals = 1;
 370
 371         status = dquot_alloc_inode(inode);
 372         if (status)
 373                 goto leave;
 374         did_quota_inode = 1;
 375
 376         /* do the real work now. */
 377         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 378                                     &new_fe_bh, parent_fe_bh, handle,
 379                                     inode_ac);
 380         if (status < 0) {
 381                 mlog_errno(status);
 382                 goto leave;
 383         }
 384
 385         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
 386         if (S_ISDIR(mode)) {
 387                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 388                                             new_fe_bh, data_ac, meta_ac);
 389                 if (status < 0) {
 390                         mlog_errno(status);
 391                         goto leave;
 392                 }
 393
 394                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 395                                                  parent_fe_bh,
 396                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 397                 if (status < 0) {
 398                         mlog_errno(status);
 399                         goto leave;
 400                 }
 401                 ocfs2_add_links_count(dirfe, 1);
 402                 ocfs2_journal_dirty(handle, parent_fe_bh);
 403                 inc_nlink(dir);
 404         }
 405
 406         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 407                          meta_ac, data_ac);
 408
 409         if (status < 0) {
 410                 mlog_errno(status);
 411                 goto leave;
 412         }
 413
 414         if (si.enable) {
 415                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 416                                                  meta_ac, data_ac);
 417                 if (status < 0) {
 418                         mlog_errno(status);
 419                         goto leave;
 420                 }
 421         }
 422
 423         /*
 424          * Do this before adding the entry to the directory. We add
 425          * also set d_op after success so that ->d_iput() will cleanup
 426          * the dentry lock even if ocfs2_add_entry() fails below.
 427          */
 428         status = ocfs2_dentry_attach_lock(dentry, inode,
 429                                           OCFS2_I(dir)->ip_blkno);
 430         if (status) {
 431                 mlog_errno(status);
 432                 goto leave;
 433         }
 434
 435         dl = dentry->d_fsdata;
 436
 437         status = ocfs2_add_entry(handle, dentry, inode,
 438                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 439                                  &lookup);
 440         if (status < 0) {
 441                 mlog_errno(status);
 442                 goto leave;
 443         }
 444
 445         insert_inode_hash(inode);
 446         d_instantiate(dentry, inode);
 447         status = 0;
 448 leave:
 449         if (status < 0 && did_quota_inode)
 450                 dquot_free_inode(inode);
 451         if (handle) {
 452                 if (status < 0 && fe)
 453                         ocfs2_set_links_count(fe, 0);
 454                 ocfs2_commit_trans(osb, handle);
 455         }
 456
 457         ocfs2_inode_unlock(dir, 1);
 458         if (did_block_signals)
 459                 ocfs2_unblock_signals(&oldset);
 460
 461         brelse(new_fe_bh);
 462         brelse(parent_fe_bh);
 463         kfree(si.value);
 464
 465         ocfs2_free_dir_lookup_result(&lookup);
 466
 467         if (inode_ac)
 468                 ocfs2_free_alloc_context(inode_ac);
 469
 470         if (data_ac)
 471                 ocfs2_free_alloc_context(data_ac);
 472
 473         if (meta_ac)
 474                 ocfs2_free_alloc_context(meta_ac);
 475
 476         /*
 477          * We should call iput after the i_mutex of the bitmap been
 478          * unlocked in ocfs2_free_alloc_context, or the
 479          * ocfs2_delete_inode will mutex_lock again.
 480          */
 481         if ((status < 0) && inode) {
 482                 if (dl)
 483                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 484
 485                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 486                 clear_nlink(inode);
 487                 iput(inode);
 488         }
 489
 490         if (status)
 491                 mlog_errno(status);
 492
 493         return status;
 494 }
 495
 496 static int __ocfs2_mknod_locked(struct inode *dir,
 497                                 struct inode *inode,
 498                                 dev_t dev,
 499                                 struct buffer_head **new_fe_bh,
 500                                 struct buffer_head *parent_fe_bh,
 501                                 handle_t *handle,
 502                                 struct ocfs2_alloc_context *inode_ac,
 503                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 504 {
 505         int status = 0;
 506         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 507         struct ocfs2_dinode *fe = NULL;
 508         struct ocfs2_extent_list *fel;
 509         u16 feat;
 510         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 511         struct timespec64 ts;
 512
 513         *new_fe_bh = NULL;
 514
 515         /* populate as many fields early on as possible - many of
 516          * these are used by the support functions here and in
 517          * callers. */
 518         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 519         oi->ip_blkno = fe_blkno;
 520         spin_lock(&osb->osb_lock);
 521         inode->i_generation = osb->s_next_generation++;
 522         spin_unlock(&osb->osb_lock);
 523
 524         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 525         if (!*new_fe_bh) {
 526                 status = -ENOMEM;
 527                 mlog_errno(status);
 528                 goto leave;
 529         }
 530         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 531
 532         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 533                                          *new_fe_bh,
 534                                          OCFS2_JOURNAL_ACCESS_CREATE);
 535         if (status < 0) {
 536                 mlog_errno(status);
 537                 goto leave;
 538         }
 539
 540         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 541         memset(fe, 0, osb->sb->s_blocksize);
 542
 543         fe->i_generation = cpu_to_le32(inode->i_generation);
 544         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 545         fe->i_blkno = cpu_to_le64(fe_blkno);
 546         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 547         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 548         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 549         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 550         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 551         fe->i_mode = cpu_to_le16(inode->i_mode);
 552         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 553                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 554
 555         ocfs2_set_links_count(fe, inode->i_nlink);
 556
 557         fe->i_last_eb_blk = 0;
 558         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 559         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 560         ktime_get_real_ts64(&ts);
 561         fe->i_atime = fe->i_ctime = fe->i_mtime =
 562                 cpu_to_le64(ts.tv_sec);
 563         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 564                 cpu_to_le32(ts.tv_nsec);
 565         fe->i_dtime = 0;
 566
 567         /*
 568          * If supported, directories start with inline data. If inline
 569          * isn't supported, but indexing is, we start them as indexed.
 570          */
 571         feat = le16_to_cpu(fe->i_dyn_features);
 572         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 573                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 574
 575                 fe->id2.i_data.id_count = cpu_to_le16(
 576                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 577         } else {
 578                 fel = &fe->id2.i_list;
 579                 fel->l_tree_depth = 0;
 580                 fel->l_next_free_rec = 0;
 581                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 582         }
 583
 584         ocfs2_journal_dirty(handle, *new_fe_bh);
 585
 586         ocfs2_populate_inode(inode, fe, 1);
 587         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 588         if (!ocfs2_mount_local(osb)) {
 589                 status = ocfs2_create_new_inode_locks(inode);
 590                 if (status < 0)
 591                         mlog_errno(status);
 592         }
 593
 594         oi->i_sync_tid = handle->h_transaction->t_tid;
 595         oi->i_datasync_tid = handle->h_transaction->t_tid;
 596
 597 leave:
 598         if (status < 0) {
 599                 if (*new_fe_bh) {
 600                         brelse(*new_fe_bh);
 601                         *new_fe_bh = NULL;
 602                 }
 603         }
 604
 605         if (status)
 606                 mlog_errno(status);
 607         return status;
 608 }
 609
 610 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 611                               struct inode *dir,
 612                               struct inode *inode,
 613                               dev_t dev,
 614                               struct buffer_head **new_fe_bh,
 615                               struct buffer_head *parent_fe_bh,
 616                               handle_t *handle,
 617                               struct ocfs2_alloc_context *inode_ac)
 618 {
 619         int status = 0;
 620         u64 suballoc_loc, fe_blkno = 0;
 621         u16 suballoc_bit;
 622
 623         *new_fe_bh = NULL;
 624
 625         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 626                                        inode_ac, &suballoc_loc,
 627                                        &suballoc_bit, &fe_blkno);
 628         if (status < 0) {
 629                 mlog_errno(status);
 630                 return status;
 631         }
 632
 633         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 634                                     parent_fe_bh, handle, inode_ac,
 635                                     fe_blkno, suballoc_loc, suballoc_bit);
 636 }
 637
 638 static int ocfs2_mkdir(struct inode *dir,
 639                        struct dentry *dentry,
 640                        umode_t mode)
 641 {
 642         int ret;
 643
 644         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 645                           OCFS2_I(dir)->ip_blkno, mode);
 646         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 647         if (ret)
 648                 mlog_errno(ret);
 649
 650         return ret;
 651 }
 652
 653 static int ocfs2_create(struct inode *dir,
 654                         struct dentry *dentry,
 655                         umode_t mode,
 656                         bool excl)
 657 {
 658         int ret;
 659
 660         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 661                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 662         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 663         if (ret)
 664                 mlog_errno(ret);
 665
 666         return ret;
 667 }
 668
 669 static int ocfs2_link(struct dentry *old_dentry,
 670                       struct inode *dir,
 671                       struct dentry *dentry)
 672 {
 673         handle_t *handle;
 674         struct inode *inode = d_inode(old_dentry);
 675         struct inode *old_dir = d_inode(old_dentry->d_parent);
 676         int err;
 677         struct buffer_head *fe_bh = NULL;
 678         struct buffer_head *old_dir_bh = NULL;
 679         struct buffer_head *parent_fe_bh = NULL;
 680         struct ocfs2_dinode *fe = NULL;
 681         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 682         struct ocfs2_dir_lookup_result lookup = { NULL, };
 683         sigset_t oldset;
 684         u64 old_de_ino;
 685
 686         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 687                          old_dentry->d_name.len, old_dentry->d_name.name,
 688                          dentry->d_name.len, dentry->d_name.name);
 689
 690         if (S_ISDIR(inode->i_mode))
 691                 return -EPERM;
 692
 693         err = dquot_initialize(dir);
 694         if (err) {
 695                 mlog_errno(err);
 696                 return err;
 697         }
 698
 699         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 700                         &parent_fe_bh, dir, 0);
 701         if (err < 0) {
 702                 if (err != -ENOENT)
 703                         mlog_errno(err);
 704                 return err;
 705         }
 706
 707         /* make sure both dirs have bhs
 708          * get an extra ref on old_dir_bh if old==new */
 709         if (!parent_fe_bh) {
 710                 if (old_dir_bh) {
 711                         parent_fe_bh = old_dir_bh;
 712                         get_bh(parent_fe_bh);
 713                 } else {
 714                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 715                         err = -EIO;
 716                         goto out;
 717                 }
 718         }
 719
 720         if (!dir->i_nlink) {
 721                 err = -ENOENT;
 722                 goto out;
 723         }
 724
 725         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 726                         old_dentry->d_name.len, &old_de_ino);
 727         if (err) {
 728                 err = -ENOENT;
 729                 goto out;
 730         }
 731
 732         /*
 733          * Check whether another node removed the source inode while we
 734          * were in the vfs.
 735          */
 736         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 737                 err = -ENOENT;
 738                 goto out;
 739         }
 740
 741         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 742                                         dentry->d_name.len);
 743         if (err)
 744                 goto out;
 745
 746         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 747                                            dentry->d_name.name,
 748                                            dentry->d_name.len, &lookup);
 749         if (err < 0) {
 750                 mlog_errno(err);
 751                 goto out;
 752         }
 753
 754         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 755         if (err < 0) {
 756                 if (err != -ENOENT)
 757                         mlog_errno(err);
 758                 goto out;
 759         }
 760
 761         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 762         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 763                 err = -EMLINK;
 764                 goto out_unlock_inode;
 765         }
 766
 767         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 768         if (IS_ERR(handle)) {
 769                 err = PTR_ERR(handle);
 770                 handle = NULL;
 771                 mlog_errno(err);
 772                 goto out_unlock_inode;
 773         }
 774
 775         /* Starting to change things, restart is no longer possible. */
 776         ocfs2_block_signals(&oldset);
 777
 778         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 779                                       OCFS2_JOURNAL_ACCESS_WRITE);
 780         if (err < 0) {
 781                 mlog_errno(err);
 782                 goto out_commit;
 783         }
 784
 785         inc_nlink(inode);
 786         inode->i_ctime = current_time(inode);
 787         ocfs2_set_links_count(fe, inode->i_nlink);
 788         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 789         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 790         ocfs2_journal_dirty(handle, fe_bh);
 791
 792         err = ocfs2_add_entry(handle, dentry, inode,
 793                               OCFS2_I(inode)->ip_blkno,
 794                               parent_fe_bh, &lookup);
 795         if (err) {
 796                 ocfs2_add_links_count(fe, -1);
 797                 drop_nlink(inode);
 798                 mlog_errno(err);
 799                 goto out_commit;
 800         }
 801
 802         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 803         if (err) {
 804                 mlog_errno(err);
 805                 goto out_commit;
 806         }
 807
 808         ihold(inode);
 809         d_instantiate(dentry, inode);
 810
 811 out_commit:
 812         ocfs2_commit_trans(osb, handle);
 813         ocfs2_unblock_signals(&oldset);
 814 out_unlock_inode:
 815         ocfs2_inode_unlock(inode, 1);
 816
 817 out:
 818         ocfs2_double_unlock(old_dir, dir);
 819
 820         brelse(fe_bh);
 821         brelse(parent_fe_bh);
 822         brelse(old_dir_bh);
 823
 824         ocfs2_free_dir_lookup_result(&lookup);
 825
 826         if (err)
 827                 mlog_errno(err);
 828
 829         return err;
 830 }
 831
 832 /*
 833  * Takes and drops an exclusive lock on the given dentry. This will
 834  * force other nodes to drop it.
 835  */
 836 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 837 {
 838         int ret;
 839
 840         ret = ocfs2_dentry_lock(dentry, 1);
 841         if (ret)
 842                 mlog_errno(ret);
 843         else
 844                 ocfs2_dentry_unlock(dentry, 1);
 845
 846         return ret;
 847 }
 848
 849 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 850 {
 851         if (S_ISDIR(inode->i_mode)) {
 852                 if (inode->i_nlink == 2)
 853                         return 1;
 854                 return 0;
 855         }
 856
 857         if (inode->i_nlink == 1)
 858                 return 1;
 859         return 0;
 860 }
 861
 862 static int ocfs2_unlink(struct inode *dir,
 863                         struct dentry *dentry)
 864 {
 865         int status;
 866         int child_locked = 0;
 867         bool is_unlinkable = false;
 868         struct inode *inode = d_inode(dentry);
 869         struct inode *orphan_dir = NULL;
 870         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 871         u64 blkno;
 872         struct ocfs2_dinode *fe = NULL;
 873         struct buffer_head *fe_bh = NULL;
 874         struct buffer_head *parent_node_bh = NULL;
 875         handle_t *handle = NULL;
 876         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 877         struct ocfs2_dir_lookup_result lookup = { NULL, };
 878         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 879
 880         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 881                            dentry->d_name.name,
 882                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 883                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 884
 885         status = dquot_initialize(dir);
 886         if (status) {
 887                 mlog_errno(status);
 888                 return status;
 889         }
 890
 891         BUG_ON(d_inode(dentry->d_parent) != dir);
 892
 893         if (inode == osb->root_inode)
 894                 return -EPERM;
 895
 896         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 897                                          OI_LS_PARENT);
 898         if (status < 0) {
 899                 if (status != -ENOENT)
 900                         mlog_errno(status);
 901                 return status;
 902         }
 903
 904         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 905                                           dentry->d_name.len, &blkno, dir,
 906                                           &lookup);
 907         if (status < 0) {
 908                 if (status != -ENOENT)
 909                         mlog_errno(status);
 910                 goto leave;
 911         }
 912
 913         if (OCFS2_I(inode)->ip_blkno != blkno) {
 914                 status = -ENOENT;
 915
 916                 trace_ocfs2_unlink_noent(
 917                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 918                                 (unsigned long long)blkno,
 919                                 OCFS2_I(inode)->ip_flags);
 920                 goto leave;
 921         }
 922
 923         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 924         if (status < 0) {
 925                 if (status != -ENOENT)
 926                         mlog_errno(status);
 927                 goto leave;
 928         }
 929         child_locked = 1;
 930
 931         if (S_ISDIR(inode->i_mode)) {
 932                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 933                         status = -ENOTEMPTY;
 934                         goto leave;
 935                 }
 936         }
 937
 938         status = ocfs2_remote_dentry_delete(dentry);
 939         if (status < 0) {
 940                 /* This remote delete should succeed under all normal
 941                  * circumstances. */
 942                 mlog_errno(status);
 943                 goto leave;
 944         }
 945
 946         if (ocfs2_inode_is_unlinkable(inode)) {
 947                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 948                                                   OCFS2_I(inode)->ip_blkno,
 949                                                   orphan_name, &orphan_insert,
 950                                                   false);
 951                 if (status < 0) {
 952                         mlog_errno(status);
 953                         goto leave;
 954                 }
 955                 is_unlinkable = true;
 956         }
 957
 958         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 959         if (IS_ERR(handle)) {
 960                 status = PTR_ERR(handle);
 961                 handle = NULL;
 962                 mlog_errno(status);
 963                 goto leave;
 964         }
 965
 966         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 967                                          OCFS2_JOURNAL_ACCESS_WRITE);
 968         if (status < 0) {
 969                 mlog_errno(status);
 970                 goto leave;
 971         }
 972
 973         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 974
 975         /* delete the name from the parent dir */
 976         status = ocfs2_delete_entry(handle, dir, &lookup);
 977         if (status < 0) {
 978                 mlog_errno(status);
 979                 goto leave;
 980         }
 981
 982         if (S_ISDIR(inode->i_mode))
 983                 drop_nlink(inode);
 984         drop_nlink(inode);
 985         ocfs2_set_links_count(fe, inode->i_nlink);
 986         ocfs2_journal_dirty(handle, fe_bh);
 987
 988         dir->i_ctime = dir->i_mtime = current_time(dir);
 989         if (S_ISDIR(inode->i_mode))
 990                 drop_nlink(dir);
 991
 992         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 993         if (status < 0) {
 994                 mlog_errno(status);
 995                 if (S_ISDIR(inode->i_mode))
 996                         inc_nlink(dir);
 997                 goto leave;
 998         }
 999
1000         if (is_unlinkable) {
1001                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1002                                 orphan_name, &orphan_insert, orphan_dir, false);
1003                 if (status < 0)
1004                         mlog_errno(status);
1005         }
1006
1007 leave:
1008         if (handle)
1009                 ocfs2_commit_trans(osb, handle);
1010
1011         if (orphan_dir) {
1012                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1013                 ocfs2_inode_unlock(orphan_dir, 1);
1014                 inode_unlock(orphan_dir);
1015                 iput(orphan_dir);
1016         }
1017
1018         if (child_locked)
1019                 ocfs2_inode_unlock(inode, 1);
1020
1021         ocfs2_inode_unlock(dir, 1);
1022
1023         brelse(fe_bh);
1024         brelse(parent_node_bh);
1025
1026         ocfs2_free_dir_lookup_result(&orphan_insert);
1027         ocfs2_free_dir_lookup_result(&lookup);
1028
1029         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1030                 mlog_errno(status);
1031
1032         return status;
1033 }
1034
1035 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1036                 u64 src_inode_no, u64 dest_inode_no)
1037 {
1038         int ret = 0, i = 0;
1039         u64 parent_inode_no = 0;
1040         u64 child_inode_no = src_inode_no;
1041         struct inode *child_inode;
1042
1043 #define MAX_LOOKUP_TIMES 32
1044         while (1) {
1045                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1046                 if (IS_ERR(child_inode)) {
1047                         ret = PTR_ERR(child_inode);
1048                         break;
1049                 }
1050
1051                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1052                 if (ret < 0) {
1053                         iput(child_inode);
1054                         if (ret != -ENOENT)
1055                                 mlog_errno(ret);
1056                         break;
1057                 }
1058
1059                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1060                                 &parent_inode_no);
1061                 ocfs2_inode_unlock(child_inode, 0);
1062                 iput(child_inode);
1063                 if (ret < 0) {
1064                         ret = -ENOENT;
1065                         break;
1066                 }
1067
1068                 if (parent_inode_no == dest_inode_no) {
1069                         ret = 1;
1070                         break;
1071                 }
1072
1073                 if (parent_inode_no == osb->root_inode->i_ino) {
1074                         ret = 0;
1075                         break;
1076                 }
1077
1078                 child_inode_no = parent_inode_no;
1079
1080                 if (++i >= MAX_LOOKUP_TIMES) {
1081                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1082                                         "may have nested directories, "
1083                                         "src inode: %llu, dest inode: %llu.\n",
1084                                         (unsigned long long)src_inode_no,
1085                                         (unsigned long long)dest_inode_no);
1086                         ret = 0;
1087                         break;
1088                 }
1089         }
1090
1091         return ret;
1092 }
1093
1094 /*
1095  * The only place this should be used is rename and link!
1096  * if they have the same id, then the 1st one is the only one locked.
1097  */
1098 static int ocfs2_double_lock(struct ocfs2_super *osb,
1099                              struct buffer_head **bh1,
1100                              struct inode *inode1,
1101                              struct buffer_head **bh2,
1102                              struct inode *inode2,
1103                              int rename)
1104 {
1105         int status;
1106         int inode1_is_ancestor, inode2_is_ancestor;
1107         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1108         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1109
1110         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1111                                 (unsigned long long)oi2->ip_blkno);
1112
1113         if (*bh1)
1114                 *bh1 = NULL;
1115         if (*bh2)
1116                 *bh2 = NULL;
1117
1118         /* we always want to lock the one with the lower lockid first.
1119          * and if they are nested, we lock ancestor first */
1120         if (oi1->ip_blkno != oi2->ip_blkno) {
1121                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1122                                 oi1->ip_blkno);
1123                 if (inode1_is_ancestor < 0) {
1124                         status = inode1_is_ancestor;
1125                         goto bail;
1126                 }
1127
1128                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1129                                 oi2->ip_blkno);
1130                 if (inode2_is_ancestor < 0) {
1131                         status = inode2_is_ancestor;
1132                         goto bail;
1133                 }
1134
1135                 if ((inode1_is_ancestor == 1) ||
1136                                 (oi1->ip_blkno < oi2->ip_blkno &&
1137                                 inode2_is_ancestor == 0)) {
1138                         /* switch id1 and id2 around */
1139                         swap(bh2, bh1);
1140                         swap(inode2, inode1);
1141                 }
1142                 /* lock id2 */
1143                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1144                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1145                 if (status < 0) {
1146                         if (status != -ENOENT)
1147                                 mlog_errno(status);
1148                         goto bail;
1149                 }
1150         }
1151
1152         /* lock id1 */
1153         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1154                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1155         if (status < 0) {
1156                 /*
1157                  * An error return must mean that no cluster locks
1158                  * were held on function exit.
1159                  */
1160                 if (oi1->ip_blkno != oi2->ip_blkno) {
1161                         ocfs2_inode_unlock(inode2, 1);
1162                         brelse(*bh2);
1163                         *bh2 = NULL;
1164                 }
1165
1166                 if (status != -ENOENT)
1167                         mlog_errno(status);
1168         }
1169
1170         trace_ocfs2_double_lock_end(
1171                         (unsigned long long)oi1->ip_blkno,
1172                         (unsigned long long)oi2->ip_blkno);
1173
1174 bail:
1175         if (status)
1176                 mlog_errno(status);
1177         return status;
1178 }
1179
1180 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1181 {
1182         ocfs2_inode_unlock(inode1, 1);
1183
1184         if (inode1 != inode2)
1185                 ocfs2_inode_unlock(inode2, 1);
1186 }
1187
1188 static int ocfs2_rename(struct inode *old_dir,
1189                         struct dentry *old_dentry,
1190                         struct inode *new_dir,
1191                         struct dentry *new_dentry,
1192                         unsigned int flags)
1193 {
1194         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1195         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1196         struct inode *old_inode = d_inode(old_dentry);
1197         struct inode *new_inode = d_inode(new_dentry);
1198         struct inode *orphan_dir = NULL;
1199         struct ocfs2_dinode *newfe = NULL;
1200         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1201         struct buffer_head *newfe_bh = NULL;
1202         struct buffer_head *old_inode_bh = NULL;
1203         struct ocfs2_super *osb = NULL;
1204         u64 newfe_blkno, old_de_ino;
1205         handle_t *handle = NULL;
1206         struct buffer_head *old_dir_bh = NULL;
1207         struct buffer_head *new_dir_bh = NULL;
1208         u32 old_dir_nlink = old_dir->i_nlink;
1209         struct ocfs2_dinode *old_di;
1210         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1211         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1212         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1213         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1214         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1215         bool should_add_orphan = false;
1216
1217         if (flags)
1218                 return -EINVAL;
1219
1220         /* At some point it might be nice to break this function up a
1221          * bit. */
1222
1223         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1224                            old_dentry->d_name.len, old_dentry->d_name.name,
1225                            new_dentry->d_name.len, new_dentry->d_name.name);
1226
1227         status = dquot_initialize(old_dir);
1228         if (status) {
1229                 mlog_errno(status);
1230                 goto bail;
1231         }
1232         status = dquot_initialize(new_dir);
1233         if (status) {
1234                 mlog_errno(status);
1235                 goto bail;
1236         }
1237
1238         osb = OCFS2_SB(old_dir->i_sb);
1239
1240         if (new_inode) {
1241                 if (!igrab(new_inode))
1242                         BUG();
1243         }
1244
1245         /* Assume a directory hierarchy thusly:
1246          * a/b/c
1247          * a/d
1248          * a,b,c, and d are all directories.
1249          *
1250          * from cwd of 'a' on both nodes:
1251          * node1: mv b/c d
1252          * node2: mv d   b/c
1253          *
1254          * And that's why, just like the VFS, we need a file system
1255          * rename lock. */
1256         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1257                 status = ocfs2_rename_lock(osb);
1258                 if (status < 0) {
1259                         mlog_errno(status);
1260                         goto bail;
1261                 }
1262                 rename_lock = 1;
1263
1264                 /* here we cannot guarantee the inodes haven't just been
1265                  * changed, so check if they are nested again */
1266                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1267                                 old_inode->i_ino);
1268                 if (status < 0) {
1269                         mlog_errno(status);
1270                         goto bail;
1271                 } else if (status == 1) {
1272                         status = -EPERM;
1273                         trace_ocfs2_rename_not_permitted(
1274                                         (unsigned long long)old_inode->i_ino,
1275                                         (unsigned long long)new_dir->i_ino);
1276                         goto bail;
1277                 }
1278         }
1279
1280         /* if old and new are the same, this'll just do one lock. */
1281         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1282                                    &new_dir_bh, new_dir, 1);
1283         if (status < 0) {
1284                 mlog_errno(status);
1285                 goto bail;
1286         }
1287         parents_locked = 1;
1288
1289         if (!new_dir->i_nlink) {
1290                 status = -EACCES;
1291                 goto bail;
1292         }
1293
1294         /* make sure both dirs have bhs
1295          * get an extra ref on old_dir_bh if old==new */
1296         if (!new_dir_bh) {
1297                 if (old_dir_bh) {
1298                         new_dir_bh = old_dir_bh;
1299                         get_bh(new_dir_bh);
1300                 } else {
1301                         mlog(ML_ERROR, "no old_dir_bh!\n");
1302                         status = -EIO;
1303                         goto bail;
1304                 }
1305         }
1306
1307         /*
1308          * Aside from allowing a meta data update, the locking here
1309          * also ensures that the downconvert thread on other nodes
1310          * won't have to concurrently downconvert the inode and the
1311          * dentry locks.
1312          */
1313         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1314                                          OI_LS_PARENT);
1315         if (status < 0) {
1316                 if (status != -ENOENT)
1317                         mlog_errno(status);
1318                 goto bail;
1319         }
1320         old_child_locked = 1;
1321
1322         status = ocfs2_remote_dentry_delete(old_dentry);
1323         if (status < 0) {
1324                 mlog_errno(status);
1325                 goto bail;
1326         }
1327
1328         if (S_ISDIR(old_inode->i_mode)) {
1329                 u64 old_inode_parent;
1330
1331                 update_dot_dot = 1;
1332                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1333                                                   old_inode,
1334                                                   &old_inode_dot_dot_res);
1335                 if (status) {
1336                         status = -EIO;
1337                         goto bail;
1338                 }
1339
1340                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1341                         status = -EIO;
1342                         goto bail;
1343                 }
1344
1345                 if (!new_inode && new_dir != old_dir &&
1346                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1347                         status = -EMLINK;
1348                         goto bail;
1349                 }
1350         }
1351
1352         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1353                                             old_dentry->d_name.len,
1354                                             &old_de_ino);
1355         if (status) {
1356                 status = -ENOENT;
1357                 goto bail;
1358         }
1359
1360         /*
1361          *  Check for inode number is _not_ due to possible IO errors.
1362          *  We might rmdir the source, keep it as pwd of some process
1363          *  and merrily kill the link to whatever was created under the
1364          *  same name. Goodbye sticky bit ;-<
1365          */
1366         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1367                 status = -ENOENT;
1368                 goto bail;
1369         }
1370
1371         /* check if the target already exists (in which case we need
1372          * to delete it */
1373         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1374                                           new_dentry->d_name.len,
1375                                           &newfe_blkno, new_dir,
1376                                           &target_lookup_res);
1377         /* The only error we allow here is -ENOENT because the new
1378          * file not existing is perfectly valid. */
1379         if ((status < 0) && (status != -ENOENT)) {
1380                 /* If we cannot find the file specified we should just */
1381                 /* return the error... */
1382                 mlog_errno(status);
1383                 goto bail;
1384         }
1385         if (status == 0)
1386                 target_exists = 1;
1387
1388         if (!target_exists && new_inode) {
1389                 /*
1390                  * Target was unlinked by another node while we were
1391                  * waiting to get to ocfs2_rename(). There isn't
1392                  * anything we can do here to help the situation, so
1393                  * bubble up the appropriate error.
1394                  */
1395                 status = -ENOENT;
1396                 goto bail;
1397         }
1398
1399         /* In case we need to overwrite an existing file, we blow it
1400          * away first */
1401         if (target_exists) {
1402                 /* VFS didn't think there existed an inode here, but
1403                  * someone else in the cluster must have raced our
1404                  * rename to create one. Today we error cleanly, in
1405                  * the future we should consider calling iget to build
1406                  * a new struct inode for this entry. */
1407                 if (!new_inode) {
1408                         status = -EACCES;
1409
1410                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1411                                                 new_dentry->d_name.name);
1412                         goto bail;
1413                 }
1414
1415                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1416                         status = -EACCES;
1417
1418                         trace_ocfs2_rename_disagree(
1419                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1420                              (unsigned long long)newfe_blkno,
1421                              OCFS2_I(new_inode)->ip_flags);
1422                         goto bail;
1423                 }
1424
1425                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1426                 if (status < 0) {
1427                         if (status != -ENOENT)
1428                                 mlog_errno(status);
1429                         goto bail;
1430                 }
1431                 new_child_locked = 1;
1432
1433                 status = ocfs2_remote_dentry_delete(new_dentry);
1434                 if (status < 0) {
1435                         mlog_errno(status);
1436                         goto bail;
1437                 }
1438
1439                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1440
1441                 trace_ocfs2_rename_over_existing(
1442                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1443                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1444
1445                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1446                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1447                                                 OCFS2_I(new_inode)->ip_blkno,
1448                                                 orphan_name, &orphan_insert,
1449                                                 false);
1450                         if (status < 0) {
1451                                 mlog_errno(status);
1452                                 goto bail;
1453                         }
1454                         should_add_orphan = true;
1455                 }
1456         } else {
1457                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1458
1459                 status = ocfs2_check_dir_for_entry(new_dir,
1460                                                    new_dentry->d_name.name,
1461                                                    new_dentry->d_name.len);
1462                 if (status)
1463                         goto bail;
1464
1465                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1466                                                       new_dentry->d_name.name,
1467                                                       new_dentry->d_name.len,
1468                                                       &target_insert);
1469                 if (status < 0) {
1470                         mlog_errno(status);
1471                         goto bail;
1472                 }
1473         }
1474
1475         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1476         if (IS_ERR(handle)) {
1477                 status = PTR_ERR(handle);
1478                 handle = NULL;
1479                 mlog_errno(status);
1480                 goto bail;
1481         }
1482
1483         if (target_exists) {
1484                 if (S_ISDIR(new_inode->i_mode)) {
1485                         if (new_inode->i_nlink != 2 ||
1486                             !ocfs2_empty_dir(new_inode)) {
1487                                 status = -ENOTEMPTY;
1488                                 goto bail;
1489                         }
1490                 }
1491                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1492                                                  newfe_bh,
1493                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1494                 if (status < 0) {
1495                         mlog_errno(status);
1496                         goto bail;
1497                 }
1498
1499                 /* change the dirent to point to the correct inode */
1500                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1501                                             old_inode);
1502                 if (status < 0) {
1503                         mlog_errno(status);
1504                         goto bail;
1505                 }
1506                 inode_inc_iversion(new_dir);
1507
1508                 if (S_ISDIR(new_inode->i_mode))
1509                         ocfs2_set_links_count(newfe, 0);
1510                 else
1511                         ocfs2_add_links_count(newfe, -1);
1512                 ocfs2_journal_dirty(handle, newfe_bh);
1513                 if (should_add_orphan) {
1514                         status = ocfs2_orphan_add(osb, handle, new_inode,
1515                                         newfe_bh, orphan_name,
1516                                         &orphan_insert, orphan_dir, false);
1517                         if (status < 0) {
1518                                 mlog_errno(status);
1519                                 goto bail;
1520                         }
1521                 }
1522         } else {
1523                 /* if the name was not found in new_dir, add it now */
1524                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1525                                          OCFS2_I(old_inode)->ip_blkno,
1526                                          new_dir_bh, &target_insert);
1527                 if (status < 0) {
1528                         mlog_errno(status);
1529                         goto bail;
1530                 }
1531         }
1532
1533         old_inode->i_ctime = current_time(old_inode);
1534         mark_inode_dirty(old_inode);
1535
1536         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1537                                          old_inode_bh,
1538                                          OCFS2_JOURNAL_ACCESS_WRITE);
1539         if (status >= 0) {
1540                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1541
1542                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1543                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1544                 ocfs2_journal_dirty(handle, old_inode_bh);
1545         } else
1546                 mlog_errno(status);
1547
1548         /*
1549          * Now that the name has been added to new_dir, remove the old name.
1550          *
1551          * We don't keep any directory entry context around until now
1552          * because the insert might have changed the type of directory
1553          * we're dealing with.
1554          */
1555         status = ocfs2_find_entry(old_dentry->d_name.name,
1556                                   old_dentry->d_name.len, old_dir,
1557                                   &old_entry_lookup);
1558         if (status) {
1559                 if (!is_journal_aborted(osb->journal->j_journal)) {
1560                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1561                                         "is not deleted.",
1562                                         new_dentry->d_name.len, new_dentry->d_name.name,
1563                                         old_dentry->d_name.len, old_dentry->d_name.name);
1564                 }
1565                 goto bail;
1566         }
1567
1568         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1569         if (status < 0) {
1570                 mlog_errno(status);
1571                 if (!is_journal_aborted(osb->journal->j_journal)) {
1572                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1573                                         "is not deleted.",
1574                                         new_dentry->d_name.len, new_dentry->d_name.name,
1575                                         old_dentry->d_name.len, old_dentry->d_name.name);
1576                 }
1577                 goto bail;
1578         }
1579
1580         if (new_inode) {
1581                 drop_nlink(new_inode);
1582                 new_inode->i_ctime = current_time(new_inode);
1583         }
1584         old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1585
1586         if (update_dot_dot) {
1587                 status = ocfs2_update_entry(old_inode, handle,
1588                                             &old_inode_dot_dot_res, new_dir);
1589                 drop_nlink(old_dir);
1590                 if (new_inode) {
1591                         drop_nlink(new_inode);
1592                 } else {
1593                         inc_nlink(new_dir);
1594                         mark_inode_dirty(new_dir);
1595                 }
1596         }
1597         mark_inode_dirty(old_dir);
1598         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1599         if (new_inode) {
1600                 mark_inode_dirty(new_inode);
1601                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1602         }
1603
1604         if (old_dir != new_dir) {
1605                 /* Keep the same times on both directories.*/
1606                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1607
1608                 /*
1609                  * This will also pick up the i_nlink change from the
1610                  * block above.
1611                  */
1612                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1613         }
1614
1615         if (old_dir_nlink != old_dir->i_nlink) {
1616                 if (!old_dir_bh) {
1617                         mlog(ML_ERROR, "need to change nlink for old dir "
1618                              "%llu from %d to %d but bh is NULL!\n",
1619                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1620                              (int)old_dir_nlink, old_dir->i_nlink);
1621                 } else {
1622                         struct ocfs2_dinode *fe;
1623                         status = ocfs2_journal_access_di(handle,
1624                                                          INODE_CACHE(old_dir),
1625                                                          old_dir_bh,
1626                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1627                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1628                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1629                         ocfs2_journal_dirty(handle, old_dir_bh);
1630                 }
1631         }
1632         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1633         status = 0;
1634 bail:
1635         if (handle)
1636                 ocfs2_commit_trans(osb, handle);
1637
1638         if (orphan_dir) {
1639                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1640                 ocfs2_inode_unlock(orphan_dir, 1);
1641                 inode_unlock(orphan_dir);
1642                 iput(orphan_dir);
1643         }
1644
1645         if (new_child_locked)
1646                 ocfs2_inode_unlock(new_inode, 1);
1647
1648         if (old_child_locked)
1649                 ocfs2_inode_unlock(old_inode, 1);
1650
1651         if (parents_locked)
1652                 ocfs2_double_unlock(old_dir, new_dir);
1653
1654         if (rename_lock)
1655                 ocfs2_rename_unlock(osb);
1656
1657         if (new_inode)
1658                 sync_mapping_buffers(old_inode->i_mapping);
1659
1660         iput(new_inode);
1661
1662         ocfs2_free_dir_lookup_result(&target_lookup_res);
1663         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1664         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1665         ocfs2_free_dir_lookup_result(&orphan_insert);
1666         ocfs2_free_dir_lookup_result(&target_insert);
1667
1668         brelse(newfe_bh);
1669         brelse(old_inode_bh);
1670         brelse(old_dir_bh);
1671         brelse(new_dir_bh);
1672
1673         if (status)
1674                 mlog_errno(status);
1675
1676         return status;
1677 }
1678
1679 /*
1680  * we expect i_size = strlen(symname). Copy symname into the file
1681  * data, including the null terminator.
1682  */
1683 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1684                                      handle_t *handle,
1685                                      struct inode *inode,
1686                                      const char *symname)
1687 {
1688         struct buffer_head **bhs = NULL;
1689         const char *c;
1690         struct super_block *sb = osb->sb;
1691         u64 p_blkno, p_blocks;
1692         int virtual, blocks, status, i, bytes_left;
1693
1694         bytes_left = i_size_read(inode) + 1;
1695         /* we can't trust i_blocks because we're actually going to
1696          * write i_size + 1 bytes. */
1697         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1698
1699         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1700                                         i_size_read(inode), blocks);
1701
1702         /* Sanity check -- make sure we're going to fit. */
1703         if (bytes_left >
1704             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1705                 status = -EIO;
1706                 mlog_errno(status);
1707                 goto bail;
1708         }
1709
1710         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1711         if (!bhs) {
1712                 status = -ENOMEM;
1713                 mlog_errno(status);
1714                 goto bail;
1715         }
1716
1717         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1718                                              NULL);
1719         if (status < 0) {
1720                 mlog_errno(status);
1721                 goto bail;
1722         }
1723
1724         /* links can never be larger than one cluster so we know this
1725          * is all going to be contiguous, but do a sanity check
1726          * anyway. */
1727         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1728                 status = -EIO;
1729                 mlog_errno(status);
1730                 goto bail;
1731         }
1732
1733         virtual = 0;
1734         while(bytes_left > 0) {
1735                 c = &symname[virtual * sb->s_blocksize];
1736
1737                 bhs[virtual] = sb_getblk(sb, p_blkno);
1738                 if (!bhs[virtual]) {
1739                         status = -ENOMEM;
1740                         mlog_errno(status);
1741                         goto bail;
1742                 }
1743                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1744                                               bhs[virtual]);
1745
1746                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1747                                               bhs[virtual],
1748                                               OCFS2_JOURNAL_ACCESS_CREATE);
1749                 if (status < 0) {
1750                         mlog_errno(status);
1751                         goto bail;
1752                 }
1753
1754                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1755
1756                 memcpy(bhs[virtual]->b_data, c,
1757                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1758                        bytes_left);
1759
1760                 ocfs2_journal_dirty(handle, bhs[virtual]);
1761
1762                 virtual++;
1763                 p_blkno++;
1764                 bytes_left -= sb->s_blocksize;
1765         }
1766
1767         status = 0;
1768 bail:
1769
1770         if (bhs) {
1771                 for(i = 0; i < blocks; i++)
1772                         brelse(bhs[i]);
1773                 kfree(bhs);
1774         }
1775
1776         if (status)
1777                 mlog_errno(status);
1778         return status;
1779 }
1780
1781 static int ocfs2_symlink(struct inode *dir,
1782                          struct dentry *dentry,
1783                          const char *symname)
1784 {
1785         int status, l, credits;
1786         u64 newsize;
1787         struct ocfs2_super *osb = NULL;
1788         struct inode *inode = NULL;
1789         struct super_block *sb;
1790         struct buffer_head *new_fe_bh = NULL;
1791         struct buffer_head *parent_fe_bh = NULL;
1792         struct ocfs2_dinode *fe = NULL;
1793         struct ocfs2_dinode *dirfe;
1794         handle_t *handle = NULL;
1795         struct ocfs2_alloc_context *inode_ac = NULL;
1796         struct ocfs2_alloc_context *data_ac = NULL;
1797         struct ocfs2_alloc_context *xattr_ac = NULL;
1798         int want_clusters = 0;
1799         int xattr_credits = 0;
1800         struct ocfs2_security_xattr_info si = {
1801                 .enable = 1,
1802         };
1803         int did_quota = 0, did_quota_inode = 0;
1804         struct ocfs2_dir_lookup_result lookup = { NULL, };
1805         sigset_t oldset;
1806         int did_block_signals = 0;
1807         struct ocfs2_dentry_lock *dl = NULL;
1808
1809         trace_ocfs2_symlink_begin(dir, dentry, symname,
1810                                   dentry->d_name.len, dentry->d_name.name);
1811
1812         status = dquot_initialize(dir);
1813         if (status) {
1814                 mlog_errno(status);
1815                 goto bail;
1816         }
1817
1818         sb = dir->i_sb;
1819         osb = OCFS2_SB(sb);
1820
1821         l = strlen(symname) + 1;
1822
1823         credits = ocfs2_calc_symlink_credits(sb);
1824
1825         /* lock the parent directory */
1826         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1827         if (status < 0) {
1828                 if (status != -ENOENT)
1829                         mlog_errno(status);
1830                 return status;
1831         }
1832
1833         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1834         if (!ocfs2_read_links_count(dirfe)) {
1835                 /* can't make a file in a deleted directory. */
1836                 status = -ENOENT;
1837                 goto bail;
1838         }
1839
1840         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1841                                            dentry->d_name.len);
1842         if (status)
1843                 goto bail;
1844
1845         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1846                                               dentry->d_name.name,
1847                                               dentry->d_name.len, &lookup);
1848         if (status < 0) {
1849                 mlog_errno(status);
1850                 goto bail;
1851         }
1852
1853         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1854         if (status < 0) {
1855                 if (status != -ENOSPC)
1856                         mlog_errno(status);
1857                 goto bail;
1858         }
1859
1860         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1861         if (IS_ERR(inode)) {
1862                 status = PTR_ERR(inode);
1863                 inode = NULL;
1864                 mlog_errno(status);
1865                 goto bail;
1866         }
1867
1868         /* get security xattr */
1869         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1870         if (status) {
1871                 if (status == -EOPNOTSUPP)
1872                         si.enable = 0;
1873                 else {
1874                         mlog_errno(status);
1875                         goto bail;
1876                 }
1877         }
1878
1879         /* calculate meta data/clusters for setting security xattr */
1880         if (si.enable) {
1881                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1882                                                   &xattr_credits, &xattr_ac);
1883                 if (status < 0) {
1884                         mlog_errno(status);
1885                         goto bail;
1886                 }
1887         }
1888
1889         /* don't reserve bitmap space for fast symlinks. */
1890         if (l > ocfs2_fast_symlink_chars(sb))
1891                 want_clusters += 1;
1892
1893         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1894         if (status < 0) {
1895                 if (status != -ENOSPC)
1896                         mlog_errno(status);
1897                 goto bail;
1898         }
1899
1900         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1901         if (IS_ERR(handle)) {
1902                 status = PTR_ERR(handle);
1903                 handle = NULL;
1904                 mlog_errno(status);
1905                 goto bail;
1906         }
1907
1908         /* Starting to change things, restart is no longer possible. */
1909         ocfs2_block_signals(&oldset);
1910         did_block_signals = 1;
1911
1912         status = dquot_alloc_inode(inode);
1913         if (status)
1914                 goto bail;
1915         did_quota_inode = 1;
1916
1917         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1918                                    dentry->d_name.name,
1919                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1920                                    inode->i_mode);
1921
1922         status = ocfs2_mknod_locked(osb, dir, inode,
1923                                     0, &new_fe_bh, parent_fe_bh, handle,
1924                                     inode_ac);
1925         if (status < 0) {
1926                 mlog_errno(status);
1927                 goto bail;
1928         }
1929
1930         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1931         inode->i_rdev = 0;
1932         newsize = l - 1;
1933         inode->i_op = &ocfs2_symlink_inode_operations;
1934         inode_nohighmem(inode);
1935         if (l > ocfs2_fast_symlink_chars(sb)) {
1936                 u32 offset = 0;
1937
1938                 status = dquot_alloc_space_nodirty(inode,
1939                     ocfs2_clusters_to_bytes(osb->sb, 1));
1940                 if (status)
1941                         goto bail;
1942                 did_quota = 1;
1943                 inode->i_mapping->a_ops = &ocfs2_aops;
1944                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1945                                               new_fe_bh,
1946                                               handle, data_ac, NULL,
1947                                               NULL);
1948                 if (status < 0) {
1949                         if (status != -ENOSPC && status != -EINTR) {
1950                                 mlog(ML_ERROR,
1951                                      "Failed to extend file to %llu\n",
1952                                      (unsigned long long)newsize);
1953                                 mlog_errno(status);
1954                                 status = -ENOSPC;
1955                         }
1956                         goto bail;
1957                 }
1958                 i_size_write(inode, newsize);
1959                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1960         } else {
1961                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1962                 memcpy((char *) fe->id2.i_symlink, symname, l);
1963                 i_size_write(inode, newsize);
1964                 inode->i_blocks = 0;
1965         }
1966
1967         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1968         if (status < 0) {
1969                 mlog_errno(status);
1970                 goto bail;
1971         }
1972
1973         if (!ocfs2_inode_is_fast_symlink(inode)) {
1974                 status = ocfs2_create_symlink_data(osb, handle, inode,
1975                                                    symname);
1976                 if (status < 0) {
1977                         mlog_errno(status);
1978                         goto bail;
1979                 }
1980         }
1981
1982         if (si.enable) {
1983                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1984                                                  xattr_ac, data_ac);
1985                 if (status < 0) {
1986                         mlog_errno(status);
1987                         goto bail;
1988                 }
1989         }
1990
1991         /*
1992          * Do this before adding the entry to the directory. We add
1993          * also set d_op after success so that ->d_iput() will cleanup
1994          * the dentry lock even if ocfs2_add_entry() fails below.
1995          */
1996         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1997         if (status) {
1998                 mlog_errno(status);
1999                 goto bail;
2000         }
2001
2002         dl = dentry->d_fsdata;
2003
2004         status = ocfs2_add_entry(handle, dentry, inode,
2005                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2006                                  &lookup);
2007         if (status < 0) {
2008                 mlog_errno(status);
2009                 goto bail;
2010         }
2011
2012         insert_inode_hash(inode);
2013         d_instantiate(dentry, inode);
2014 bail:
2015         if (status < 0 && did_quota)
2016                 dquot_free_space_nodirty(inode,
2017                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2018         if (status < 0 && did_quota_inode)
2019                 dquot_free_inode(inode);
2020         if (handle) {
2021                 if (status < 0 && fe)
2022                         ocfs2_set_links_count(fe, 0);
2023                 ocfs2_commit_trans(osb, handle);
2024         }
2025
2026         ocfs2_inode_unlock(dir, 1);
2027         if (did_block_signals)
2028                 ocfs2_unblock_signals(&oldset);
2029
2030         brelse(new_fe_bh);
2031         brelse(parent_fe_bh);
2032         kfree(si.value);
2033         ocfs2_free_dir_lookup_result(&lookup);
2034         if (inode_ac)
2035                 ocfs2_free_alloc_context(inode_ac);
2036         if (data_ac)
2037                 ocfs2_free_alloc_context(data_ac);
2038         if (xattr_ac)
2039                 ocfs2_free_alloc_context(xattr_ac);
2040         if ((status < 0) && inode) {
2041                 if (dl)
2042                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2043
2044                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2045                 clear_nlink(inode);
2046                 iput(inode);
2047         }
2048
2049         if (status)
2050                 mlog_errno(status);
2051
2052         return status;
2053 }
2054
2055 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2056 {
2057         int status, namelen;
2058
2059         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2060                            (long long)blkno);
2061         if (namelen <= 0) {
2062                 if (namelen)
2063                         status = namelen;
2064                 else
2065                         status = -EINVAL;
2066                 mlog_errno(status);
2067                 goto bail;
2068         }
2069         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2070                 status = -EINVAL;
2071                 mlog_errno(status);
2072                 goto bail;
2073         }
2074
2075         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2076
2077         status = 0;
2078 bail:
2079         if (status < 0)
2080                 mlog_errno(status);
2081         return status;
2082 }
2083
2084 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2085                                         struct inode **ret_orphan_dir,
2086                                         struct buffer_head **ret_orphan_dir_bh)
2087 {
2088         struct inode *orphan_dir_inode;
2089         struct buffer_head *orphan_dir_bh = NULL;
2090         int ret = 0;
2091
2092         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2093                                                        ORPHAN_DIR_SYSTEM_INODE,
2094                                                        osb->slot_num);
2095         if (!orphan_dir_inode) {
2096                 ret = -ENOENT;
2097                 mlog_errno(ret);
2098                 return ret;
2099         }
2100
2101         inode_lock(orphan_dir_inode);
2102
2103         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2104         if (ret < 0) {
2105                 inode_unlock(orphan_dir_inode);
2106                 iput(orphan_dir_inode);
2107
2108                 mlog_errno(ret);
2109                 return ret;
2110         }
2111
2112         *ret_orphan_dir = orphan_dir_inode;
2113         *ret_orphan_dir_bh = orphan_dir_bh;
2114
2115         return 0;
2116 }
2117
2118 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2119                                       struct buffer_head *orphan_dir_bh,
2120                                       u64 blkno,
2121                                       char *name,
2122                                       struct ocfs2_dir_lookup_result *lookup,
2123                                       bool dio)
2124 {
2125         int ret;
2126         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2127         int namelen = dio ?
2128                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2129                         OCFS2_ORPHAN_NAMELEN;
2130
2131         if (dio) {
2132                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2133                                 OCFS2_DIO_ORPHAN_PREFIX);
2134                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2135                         ret = -EINVAL;
2136                         mlog_errno(ret);
2137                         return ret;
2138                 }
2139
2140                 ret = ocfs2_blkno_stringify(blkno,
2141                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2142         } else
2143                 ret = ocfs2_blkno_stringify(blkno, name);
2144         if (ret < 0) {
2145                 mlog_errno(ret);
2146                 return ret;
2147         }
2148
2149         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2150                                            orphan_dir_bh, name,
2151                                            namelen, lookup);
2152         if (ret < 0) {
2153                 mlog_errno(ret);
2154                 return ret;
2155         }
2156
2157         return 0;
2158 }
2159
2160 /**
2161  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2162  * insertion of an orphan.
2163  * @osb: ocfs2 file system
2164  * @ret_orphan_dir: Orphan dir inode - returned locked!
2165  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2166  * @lookup: dir lookup result, to be passed back into functions like
2167  *          ocfs2_orphan_add
2168  *
2169  * Returns zero on success and the ret_orphan_dir, name and lookup
2170  * fields will be populated.
2171  *
2172  * Returns non-zero on failure.
2173  */
2174 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2175                                     struct inode **ret_orphan_dir,
2176                                     u64 blkno,
2177                                     char *name,
2178                                     struct ocfs2_dir_lookup_result *lookup,
2179                                     bool dio)
2180 {
2181         struct inode *orphan_dir_inode = NULL;
2182         struct buffer_head *orphan_dir_bh = NULL;
2183         int ret = 0;
2184
2185         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2186                                            &orphan_dir_bh);
2187         if (ret < 0) {
2188                 mlog_errno(ret);
2189                 return ret;
2190         }
2191
2192         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2193                                          blkno, name, lookup, dio);
2194         if (ret < 0) {
2195                 mlog_errno(ret);
2196                 goto out;
2197         }
2198
2199         *ret_orphan_dir = orphan_dir_inode;
2200
2201 out:
2202         brelse(orphan_dir_bh);
2203
2204         if (ret) {
2205                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2206                 inode_unlock(orphan_dir_inode);
2207                 iput(orphan_dir_inode);
2208         }
2209
2210         if (ret)
2211                 mlog_errno(ret);
2212         return ret;
2213 }
2214
2215 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2216                             handle_t *handle,
2217                             struct inode *inode,
2218                             struct buffer_head *fe_bh,
2219                             char *name,
2220                             struct ocfs2_dir_lookup_result *lookup,
2221                             struct inode *orphan_dir_inode,
2222                             bool dio)
2223 {
2224         struct buffer_head *orphan_dir_bh = NULL;
2225         int status = 0;
2226         struct ocfs2_dinode *orphan_fe;
2227         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2228         int namelen = dio ?
2229                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2230                         OCFS2_ORPHAN_NAMELEN;
2231
2232         trace_ocfs2_orphan_add_begin(
2233                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2234
2235         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2236         if (status < 0) {
2237                 mlog_errno(status);
2238                 goto leave;
2239         }
2240
2241         status = ocfs2_journal_access_di(handle,
2242                                          INODE_CACHE(orphan_dir_inode),
2243                                          orphan_dir_bh,
2244                                          OCFS2_JOURNAL_ACCESS_WRITE);
2245         if (status < 0) {
2246                 mlog_errno(status);
2247                 goto leave;
2248         }
2249
2250         /*
2251          * We're going to journal the change of i_flags and i_orphaned_slot.
2252          * It's safe anyway, though some callers may duplicate the journaling.
2253          * Journaling within the func just make the logic look more
2254          * straightforward.
2255          */
2256         status = ocfs2_journal_access_di(handle,
2257                                          INODE_CACHE(inode),
2258                                          fe_bh,
2259                                          OCFS2_JOURNAL_ACCESS_WRITE);
2260         if (status < 0) {
2261                 mlog_errno(status);
2262                 goto leave;
2263         }
2264
2265         /* we're a cluster, and nlink can change on disk from
2266          * underneath us... */
2267         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2268         if (S_ISDIR(inode->i_mode))
2269                 ocfs2_add_links_count(orphan_fe, 1);
2270         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2271         ocfs2_journal_dirty(handle, orphan_dir_bh);
2272
2273         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2274                                    namelen, inode,
2275                                    OCFS2_I(inode)->ip_blkno,
2276                                    orphan_dir_bh, lookup);
2277         if (status < 0) {
2278                 mlog_errno(status);
2279                 goto rollback;
2280         }
2281
2282         if (dio) {
2283                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2284                  * slot.
2285                  */
2286                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2287                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2288         } else {
2289                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2290                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2291
2292                 /* Record which orphan dir our inode now resides
2293                  * in. delete_inode will use this to determine which orphan
2294                  * dir to lock. */
2295                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2296         }
2297
2298         ocfs2_journal_dirty(handle, fe_bh);
2299
2300         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2301                                    osb->slot_num);
2302
2303 rollback:
2304         if (status < 0) {
2305                 if (S_ISDIR(inode->i_mode))
2306                         ocfs2_add_links_count(orphan_fe, -1);
2307                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2308         }
2309
2310 leave:
2311         brelse(orphan_dir_bh);
2312
2313         return status;
2314 }
2315
2316 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2317 int ocfs2_orphan_del(struct ocfs2_super *osb,
2318                      handle_t *handle,
2319                      struct inode *orphan_dir_inode,
2320                      struct inode *inode,
2321                      struct buffer_head *orphan_dir_bh,
2322                      bool dio)
2323 {
2324         char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2325         struct ocfs2_dinode *orphan_fe;
2326         int status = 0;
2327         struct ocfs2_dir_lookup_result lookup = { NULL, };
2328
2329         if (dio) {
2330                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2331                                 OCFS2_DIO_ORPHAN_PREFIX);
2332                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2333                         status = -EINVAL;
2334                         mlog_errno(status);
2335                         return status;
2336                 }
2337
2338                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2339                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2340         } else
2341                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2342         if (status < 0) {
2343                 mlog_errno(status);
2344                 goto leave;
2345         }
2346
2347         trace_ocfs2_orphan_del(
2348              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2349              name, strlen(name));
2350
2351         status = ocfs2_journal_access_di(handle,
2352                                          INODE_CACHE(orphan_dir_inode),
2353                                          orphan_dir_bh,
2354                                          OCFS2_JOURNAL_ACCESS_WRITE);
2355         if (status < 0) {
2356                 mlog_errno(status);
2357                 goto leave;
2358         }
2359
2360         /* find it's spot in the orphan directory */
2361         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2362                                   &lookup);
2363         if (status) {
2364                 mlog_errno(status);
2365                 goto leave;
2366         }
2367
2368         /* remove it from the orphan directory */
2369         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2370         if (status < 0) {
2371                 mlog_errno(status);
2372                 goto leave;
2373         }
2374
2375         /* do the i_nlink dance! :) */
2376         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2377         if (S_ISDIR(inode->i_mode))
2378                 ocfs2_add_links_count(orphan_fe, -1);
2379         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2380         ocfs2_journal_dirty(handle, orphan_dir_bh);
2381
2382 leave:
2383         ocfs2_free_dir_lookup_result(&lookup);
2384
2385         if (status)
2386                 mlog_errno(status);
2387         return status;
2388 }
2389
2390 /**
2391  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2392  * allocated file. This is different from the typical 'add to orphan dir'
2393  * operation in that the inode does not yet exist. This is a problem because
2394  * the orphan dir stringifies the inode block number to come up with it's
2395  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2396  * problem. This function works around it by calling deeper into the orphan
2397  * and suballoc code than other callers. Use this only by necessity.
2398  * @dir: The directory which this inode will ultimately wind up under - not the
2399  * orphan dir!
2400  * @dir_bh: buffer_head the @dir inode block
2401  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2402  * with the string to be used for orphan dirent. Pass back to the orphan dir
2403  * code.
2404  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2405  * dir code.
2406  * @ret_di_blkno: block number where the new inode will be allocated.
2407  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2408  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2409  *
2410  * Returns zero on success and the ret_orphan_dir, name and lookup
2411  * fields will be populated.
2412  *
2413  * Returns non-zero on failure.
2414  */
2415 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2416                                         struct buffer_head *dir_bh,
2417                                         char *orphan_name,
2418                                         struct inode **ret_orphan_dir,
2419                                         u64 *ret_di_blkno,
2420                                         struct ocfs2_dir_lookup_result *orphan_insert,
2421                                         struct ocfs2_alloc_context **ret_inode_ac)
2422 {
2423         int ret;
2424         u64 di_blkno;
2425         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2426         struct inode *orphan_dir = NULL;
2427         struct buffer_head *orphan_dir_bh = NULL;
2428         struct ocfs2_alloc_context *inode_ac = NULL;
2429
2430         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2431         if (ret < 0) {
2432                 mlog_errno(ret);
2433                 return ret;
2434         }
2435
2436         /* reserve an inode spot */
2437         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2438         if (ret < 0) {
2439                 if (ret != -ENOSPC)
2440                         mlog_errno(ret);
2441                 goto out;
2442         }
2443
2444         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2445                                        &di_blkno);
2446         if (ret) {
2447                 mlog_errno(ret);
2448                 goto out;
2449         }
2450
2451         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2452                                          di_blkno, orphan_name, orphan_insert,
2453                                          false);
2454         if (ret < 0) {
2455                 mlog_errno(ret);
2456                 goto out;
2457         }
2458
2459 out:
2460         if (ret == 0) {
2461                 *ret_orphan_dir = orphan_dir;
2462                 *ret_di_blkno = di_blkno;
2463                 *ret_inode_ac = inode_ac;
2464                 /*
2465                  * orphan_name and orphan_insert are already up to
2466                  * date via prepare_orphan_dir
2467                  */
2468         } else {
2469                 /* Unroll reserve_new_inode* */
2470                 if (inode_ac)
2471                         ocfs2_free_alloc_context(inode_ac);
2472
2473                 /* Unroll orphan dir locking */
2474                 inode_unlock(orphan_dir);
2475                 ocfs2_inode_unlock(orphan_dir, 1);
2476                 iput(orphan_dir);
2477         }
2478
2479         brelse(orphan_dir_bh);
2480
2481         return ret;
2482 }
2483
2484 int ocfs2_create_inode_in_orphan(struct inode *dir,
2485                                  int mode,
2486                                  struct inode **new_inode)
2487 {
2488         int status, did_quota_inode = 0;
2489         struct inode *inode = NULL;
2490         struct inode *orphan_dir = NULL;
2491         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2492         handle_t *handle = NULL;
2493         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2494         struct buffer_head *parent_di_bh = NULL;
2495         struct buffer_head *new_di_bh = NULL;
2496         struct ocfs2_alloc_context *inode_ac = NULL;
2497         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2498         u64 di_blkno, suballoc_loc;
2499         u16 suballoc_bit;
2500
2501         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2502         if (status < 0) {
2503                 if (status != -ENOENT)
2504                         mlog_errno(status);
2505                 return status;
2506         }
2507
2508         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2509                                               orphan_name, &orphan_dir,
2510                                               &di_blkno, &orphan_insert, &inode_ac);
2511         if (status < 0) {
2512                 if (status != -ENOSPC)
2513                         mlog_errno(status);
2514                 goto leave;
2515         }
2516
2517         inode = ocfs2_get_init_inode(dir, mode);
2518         if (IS_ERR(inode)) {
2519                 status = PTR_ERR(inode);
2520                 inode = NULL;
2521                 mlog_errno(status);
2522                 goto leave;
2523         }
2524
2525         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2526         if (IS_ERR(handle)) {
2527                 status = PTR_ERR(handle);
2528                 handle = NULL;
2529                 mlog_errno(status);
2530                 goto leave;
2531         }
2532
2533         status = dquot_alloc_inode(inode);
2534         if (status)
2535                 goto leave;
2536         did_quota_inode = 1;
2537
2538         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2539                                               &suballoc_loc,
2540                                               &suballoc_bit, di_blkno);
2541         if (status < 0) {
2542                 mlog_errno(status);
2543                 goto leave;
2544         }
2545
2546         clear_nlink(inode);
2547         /* do the real work now. */
2548         status = __ocfs2_mknod_locked(dir, inode,
2549                                       0, &new_di_bh, parent_di_bh, handle,
2550                                       inode_ac, di_blkno, suballoc_loc,
2551                                       suballoc_bit);
2552         if (status < 0) {
2553                 mlog_errno(status);
2554                 goto leave;
2555         }
2556
2557         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2558                                   &orphan_insert, orphan_dir, false);
2559         if (status < 0) {
2560                 mlog_errno(status);
2561                 goto leave;
2562         }
2563
2564         /* get open lock so that only nodes can't remove it from orphan dir. */
2565         status = ocfs2_open_lock(inode);
2566         if (status < 0)
2567                 mlog_errno(status);
2568
2569         insert_inode_hash(inode);
2570 leave:
2571         if (status < 0 && did_quota_inode)
2572                 dquot_free_inode(inode);
2573         if (handle)
2574                 ocfs2_commit_trans(osb, handle);
2575
2576         if (orphan_dir) {
2577                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2578                 ocfs2_inode_unlock(orphan_dir, 1);
2579                 inode_unlock(orphan_dir);
2580                 iput(orphan_dir);
2581         }
2582
2583         if ((status < 0) && inode) {
2584                 clear_nlink(inode);
2585                 iput(inode);
2586         }
2587
2588         if (inode_ac)
2589                 ocfs2_free_alloc_context(inode_ac);
2590
2591         brelse(new_di_bh);
2592
2593         if (!status)
2594                 *new_inode = inode;
2595
2596         ocfs2_free_dir_lookup_result(&orphan_insert);
2597
2598         ocfs2_inode_unlock(dir, 1);
2599         brelse(parent_di_bh);
2600         return status;
2601 }
2602
2603 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2604         struct inode *inode)
2605 {
2606         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2607         struct inode *orphan_dir_inode = NULL;
2608         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2609         struct buffer_head *di_bh = NULL;
2610         int status = 0;
2611         handle_t *handle = NULL;
2612         struct ocfs2_dinode *di = NULL;
2613
2614         status = ocfs2_inode_lock(inode, &di_bh, 1);
2615         if (status < 0) {
2616                 mlog_errno(status);
2617                 goto bail;
2618         }
2619
2620         di = (struct ocfs2_dinode *) di_bh->b_data;
2621         /*
2622          * Another append dio crashed?
2623          * If so, manually recover it first.
2624          */
2625         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2626                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2627                 if (status < 0) {
2628                         if (status != -ENOSPC)
2629                                 mlog_errno(status);
2630                         goto bail_unlock_inode;
2631                 }
2632
2633                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2634                 if (status < 0) {
2635                         mlog_errno(status);
2636                         goto bail_unlock_inode;
2637                 }
2638         }
2639
2640         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2641                         OCFS2_I(inode)->ip_blkno,
2642                         orphan_name,
2643                         &orphan_insert,
2644                         true);
2645         if (status < 0) {
2646                 mlog_errno(status);
2647                 goto bail_unlock_inode;
2648         }
2649
2650         handle = ocfs2_start_trans(osb,
2651                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2652         if (IS_ERR(handle)) {
2653                 status = PTR_ERR(handle);
2654                 goto bail_unlock_orphan;
2655         }
2656
2657         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2658                         &orphan_insert, orphan_dir_inode, true);
2659         if (status)
2660                 mlog_errno(status);
2661
2662         ocfs2_commit_trans(osb, handle);
2663
2664 bail_unlock_orphan:
2665         ocfs2_inode_unlock(orphan_dir_inode, 1);
2666         inode_unlock(orphan_dir_inode);
2667         iput(orphan_dir_inode);
2668
2669         ocfs2_free_dir_lookup_result(&orphan_insert);
2670
2671 bail_unlock_inode:
2672         ocfs2_inode_unlock(inode, 1);
2673         brelse(di_bh);
2674
2675 bail:
2676         return status;
2677 }
2678
2679 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2680                 struct inode *inode, struct buffer_head *di_bh,
2681                 int update_isize, loff_t end)
2682 {
2683         struct inode *orphan_dir_inode = NULL;
2684         struct buffer_head *orphan_dir_bh = NULL;
2685         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2686         handle_t *handle = NULL;
2687         int status = 0;
2688
2689         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2690                         ORPHAN_DIR_SYSTEM_INODE,
2691                         le16_to_cpu(di->i_dio_orphaned_slot));
2692         if (!orphan_dir_inode) {
2693                 status = -ENOENT;
2694                 mlog_errno(status);
2695                 goto bail;
2696         }
2697
2698         inode_lock(orphan_dir_inode);
2699         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2700         if (status < 0) {
2701                 inode_unlock(orphan_dir_inode);
2702                 iput(orphan_dir_inode);
2703                 mlog_errno(status);
2704                 goto bail;
2705         }
2706
2707         handle = ocfs2_start_trans(osb,
2708                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2709         if (IS_ERR(handle)) {
2710                 status = PTR_ERR(handle);
2711                 goto bail_unlock_orphan;
2712         }
2713
2714         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2715
2716         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2717                                 inode, orphan_dir_bh, true);
2718         if (status < 0) {
2719                 mlog_errno(status);
2720                 goto bail_commit;
2721         }
2722
2723         status = ocfs2_journal_access_di(handle,
2724                         INODE_CACHE(inode),
2725                         di_bh,
2726                         OCFS2_JOURNAL_ACCESS_WRITE);
2727         if (status < 0) {
2728                 mlog_errno(status);
2729                 goto bail_commit;
2730         }
2731
2732         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2733         di->i_dio_orphaned_slot = 0;
2734
2735         if (update_isize) {
2736                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2737                 if (status)
2738                         mlog_errno(status);
2739         } else
2740                 ocfs2_journal_dirty(handle, di_bh);
2741
2742 bail_commit:
2743         ocfs2_commit_trans(osb, handle);
2744
2745 bail_unlock_orphan:
2746         ocfs2_inode_unlock(orphan_dir_inode, 1);
2747         inode_unlock(orphan_dir_inode);
2748         brelse(orphan_dir_bh);
2749         iput(orphan_dir_inode);
2750
2751 bail:
2752         return status;
2753 }
2754
2755 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2756                                    struct inode *inode,
2757                                    struct dentry *dentry)
2758 {
2759         int status = 0;
2760         struct buffer_head *parent_di_bh = NULL;
2761         handle_t *handle = NULL;
2762         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2763         struct ocfs2_dinode *dir_di, *di;
2764         struct inode *orphan_dir_inode = NULL;
2765         struct buffer_head *orphan_dir_bh = NULL;
2766         struct buffer_head *di_bh = NULL;
2767         struct ocfs2_dir_lookup_result lookup = { NULL, };
2768
2769         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2770                                 dentry->d_name.len, dentry->d_name.name,
2771                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2772                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2773
2774         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2775         if (status < 0) {
2776                 if (status != -ENOENT)
2777                         mlog_errno(status);
2778                 return status;
2779         }
2780
2781         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2782         if (!dir_di->i_links_count) {
2783                 /* can't make a file in a deleted directory. */
2784                 status = -ENOENT;
2785                 goto leave;
2786         }
2787
2788         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2789                                            dentry->d_name.len);
2790         if (status)
2791                 goto leave;
2792
2793         /* get a spot inside the dir. */
2794         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2795                                               dentry->d_name.name,
2796                                               dentry->d_name.len, &lookup);
2797         if (status < 0) {
2798                 mlog_errno(status);
2799                 goto leave;
2800         }
2801
2802         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2803                                                        ORPHAN_DIR_SYSTEM_INODE,
2804                                                        osb->slot_num);
2805         if (!orphan_dir_inode) {
2806                 status = -ENOENT;
2807                 mlog_errno(status);
2808                 goto leave;
2809         }
2810
2811         inode_lock(orphan_dir_inode);
2812
2813         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2814         if (status < 0) {
2815                 mlog_errno(status);
2816                 inode_unlock(orphan_dir_inode);
2817                 iput(orphan_dir_inode);
2818                 goto leave;
2819         }
2820
2821         status = ocfs2_read_inode_block(inode, &di_bh);
2822         if (status < 0) {
2823                 mlog_errno(status);
2824                 goto orphan_unlock;
2825         }
2826
2827         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2828         if (IS_ERR(handle)) {
2829                 status = PTR_ERR(handle);
2830                 handle = NULL;
2831                 mlog_errno(status);
2832                 goto orphan_unlock;
2833         }
2834
2835         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2836                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2837         if (status < 0) {
2838                 mlog_errno(status);
2839                 goto out_commit;
2840         }
2841
2842         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2843                                   orphan_dir_bh, false);
2844         if (status < 0) {
2845                 mlog_errno(status);
2846                 goto out_commit;
2847         }
2848
2849         di = (struct ocfs2_dinode *)di_bh->b_data;
2850         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2851         di->i_orphaned_slot = 0;
2852         set_nlink(inode, 1);
2853         ocfs2_set_links_count(di, inode->i_nlink);
2854         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2855         ocfs2_journal_dirty(handle, di_bh);
2856
2857         status = ocfs2_add_entry(handle, dentry, inode,
2858                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2859                                  &lookup);
2860         if (status < 0) {
2861                 mlog_errno(status);
2862                 goto out_commit;
2863         }
2864
2865         status = ocfs2_dentry_attach_lock(dentry, inode,
2866                                           OCFS2_I(dir)->ip_blkno);
2867         if (status) {
2868                 mlog_errno(status);
2869                 goto out_commit;
2870         }
2871
2872         d_instantiate(dentry, inode);
2873         status = 0;
2874 out_commit:
2875         ocfs2_commit_trans(osb, handle);
2876 orphan_unlock:
2877         ocfs2_inode_unlock(orphan_dir_inode, 1);
2878         inode_unlock(orphan_dir_inode);
2879         iput(orphan_dir_inode);
2880 leave:
2881
2882         ocfs2_inode_unlock(dir, 1);
2883
2884         brelse(di_bh);
2885         brelse(parent_di_bh);
2886         brelse(orphan_dir_bh);
2887
2888         ocfs2_free_dir_lookup_result(&lookup);
2889
2890         if (status)
2891                 mlog_errno(status);
2892
2893         return status;
2894 }
2895
2896 const struct inode_operations ocfs2_dir_iops = {
2897         .create         = ocfs2_create,
2898         .lookup         = ocfs2_lookup,
2899         .link           = ocfs2_link,
2900         .unlink         = ocfs2_unlink,
2901         .rmdir          = ocfs2_unlink,
2902         .symlink        = ocfs2_symlink,
2903         .mkdir          = ocfs2_mkdir,
2904         .mknod          = ocfs2_mknod,
2905         .rename         = ocfs2_rename,
2906         .setattr        = ocfs2_setattr,
2907         .getattr        = ocfs2_getattr,
2908         .permission     = ocfs2_permission,
2909         .listxattr      = ocfs2_listxattr,
2910         .fiemap         = ocfs2_fiemap,
2911         .get_acl        = ocfs2_iop_get_acl,
2912         .set_acl        = ocfs2_iop_set_acl,
2913 };