fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/quotaops.h>
  44 #include <linux/iversion.h>
  45
  46 #include <cluster/masklog.h>
  47
  48 #include "ocfs2.h"
  49
  50 #include "alloc.h"
  51 #include "dcache.h"
  52 #include "dir.h"
  53 #include "dlmglue.h"
  54 #include "extent_map.h"
  55 #include "file.h"
  56 #include "inode.h"
  57 #include "journal.h"
  58 #include "namei.h"
  59 #include "suballoc.h"
  60 #include "super.h"
  61 #include "symlink.h"
  62 #include "sysfile.h"
  63 #include "uptodate.h"
  64 #include "xattr.h"
  65 #include "acl.h"
  66 #include "ocfs2_trace.h"
  67
  68 #include "buffer_head_io.h"
  69
  70 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  71                               struct inode *dir,
  72                               struct inode *inode,
  73                               dev_t dev,
  74                               struct buffer_head **new_fe_bh,
  75                               struct buffer_head *parent_fe_bh,
  76                               handle_t *handle,
  77                               struct ocfs2_alloc_context *inode_ac);
  78
  79 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  80                                     struct inode **ret_orphan_dir,
  81                                     u64 blkno,
  82                                     char *name,
  83                                     struct ocfs2_dir_lookup_result *lookup,
  84                                     bool dio);
  85
  86 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  87                             handle_t *handle,
  88                             struct inode *inode,
  89                             struct buffer_head *fe_bh,
  90                             char *name,
  91                             struct ocfs2_dir_lookup_result *lookup,
  92                             struct inode *orphan_dir_inode,
  93                             bool dio);
  94
  95 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  96                                      handle_t *handle,
  97                                      struct inode *inode,
  98                                      const char *symname);
  99
 100 static int ocfs2_double_lock(struct ocfs2_super *osb,
 101                              struct buffer_head **bh1,
 102                              struct inode *inode1,
 103                              struct buffer_head **bh2,
 104                              struct inode *inode2,
 105                              int rename);
 106
 107 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
 108 /* An orphan dir name is an 8 byte value, printed as a hex string */
 109 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 110
 111 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 112                                    unsigned int flags)
 113 {
 114         int status;
 115         u64 blkno;
 116         struct inode *inode = NULL;
 117         struct dentry *ret;
 118         struct ocfs2_inode_info *oi;
 119
 120         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 121                            dentry->d_name.name,
 122                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 123
 124         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 125                 ret = ERR_PTR(-ENAMETOOLONG);
 126                 goto bail;
 127         }
 128
 129         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 130         if (status < 0) {
 131                 if (status != -ENOENT)
 132                         mlog_errno(status);
 133                 ret = ERR_PTR(status);
 134                 goto bail;
 135         }
 136
 137         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 138                                             dentry->d_name.len, &blkno);
 139         if (status < 0)
 140                 goto bail_add;
 141
 142         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 143         if (IS_ERR(inode)) {
 144                 ret = ERR_PTR(-EACCES);
 145                 goto bail_unlock;
 146         }
 147
 148         oi = OCFS2_I(inode);
 149         /* Clear any orphaned state... If we were able to look up the
 150          * inode from a directory, it certainly can't be orphaned. We
 151          * might have the bad state from a node which intended to
 152          * orphan this inode but crashed before it could commit the
 153          * unlink. */
 154         spin_lock(&oi->ip_lock);
 155         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 156         spin_unlock(&oi->ip_lock);
 157
 158 bail_add:
 159         ret = d_splice_alias(inode, dentry);
 160
 161         if (inode) {
 162                 /*
 163                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 164                  * dentry, it will d_move() it on top of ourse. The
 165                  * return value will indicate this however, so in
 166                  * those cases, we switch them around for the locking
 167                  * code.
 168                  *
 169                  * NOTE: This dentry already has ->d_op set from
 170                  * ocfs2_get_parent() and ocfs2_get_dentry()
 171                  */
 172                 if (!IS_ERR_OR_NULL(ret))
 173                         dentry = ret;
 174
 175                 status = ocfs2_dentry_attach_lock(dentry, inode,
 176                                                   OCFS2_I(dir)->ip_blkno);
 177                 if (status) {
 178                         mlog_errno(status);
 179                         ret = ERR_PTR(status);
 180                         goto bail_unlock;
 181                 }
 182         } else
 183                 ocfs2_dentry_attach_gen(dentry);
 184
 185 bail_unlock:
 186         /* Don't drop the cluster lock until *after* the d_add --
 187          * unlink on another node will message us to remove that
 188          * dentry under this lock so otherwise we can race this with
 189          * the downconvert thread and have a stale dentry. */
 190         ocfs2_inode_unlock(dir, 0);
 191
 192 bail:
 193
 194         trace_ocfs2_lookup_ret(ret);
 195
 196         return ret;
 197 }
 198
 199 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 200 {
 201         struct inode *inode;
 202         int status;
 203
 204         inode = new_inode(dir->i_sb);
 205         if (!inode) {
 206                 mlog(ML_ERROR, "new_inode failed!\n");
 207                 return ERR_PTR(-ENOMEM);
 208         }
 209
 210         /* populate as many fields early on as possible - many of
 211          * these are used by the support functions here and in
 212          * callers. */
 213         if (S_ISDIR(mode))
 214                 set_nlink(inode, 2);
 215         inode_init_owner(inode, dir, mode);
 216         status = dquot_initialize(inode);
 217         if (status)
 218                 return ERR_PTR(status);
 219
 220         return inode;
 221 }
 222
 223 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 224                 struct dentry *dentry, struct inode *inode)
 225 {
 226         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 227
 228         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 229         ocfs2_lock_res_free(&dl->dl_lockres);
 230         BUG_ON(dl->dl_count != 1);
 231         spin_lock(&dentry_attach_lock);
 232         dentry->d_fsdata = NULL;
 233         spin_unlock(&dentry_attach_lock);
 234         kfree(dl);
 235         iput(inode);
 236 }
 237
 238 static int ocfs2_mknod(struct inode *dir,
 239                        struct dentry *dentry,
 240                        umode_t mode,
 241                        dev_t dev)
 242 {
 243         int status = 0;
 244         struct buffer_head *parent_fe_bh = NULL;
 245         handle_t *handle = NULL;
 246         struct ocfs2_super *osb;
 247         struct ocfs2_dinode *dirfe;
 248         struct ocfs2_dinode *fe = NULL;
 249         struct buffer_head *new_fe_bh = NULL;
 250         struct inode *inode = NULL;
 251         struct ocfs2_alloc_context *inode_ac = NULL;
 252         struct ocfs2_alloc_context *data_ac = NULL;
 253         struct ocfs2_alloc_context *meta_ac = NULL;
 254         int want_clusters = 0;
 255         int want_meta = 0;
 256         int xattr_credits = 0;
 257         struct ocfs2_security_xattr_info si = {
 258                 .enable = 1,
 259         };
 260         int did_quota_inode = 0;
 261         struct ocfs2_dir_lookup_result lookup = { NULL, };
 262         sigset_t oldset;
 263         int did_block_signals = 0;
 264         struct ocfs2_dentry_lock *dl = NULL;
 265
 266         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 267                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 268                           (unsigned long)dev, mode);
 269
 270         status = dquot_initialize(dir);
 271         if (status) {
 272                 mlog_errno(status);
 273                 return status;
 274         }
 275
 276         /* get our super block */
 277         osb = OCFS2_SB(dir->i_sb);
 278
 279         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 280         if (status < 0) {
 281                 if (status != -ENOENT)
 282                         mlog_errno(status);
 283                 return status;
 284         }
 285
 286         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 287                 status = -EMLINK;
 288                 goto leave;
 289         }
 290
 291         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 292         if (!ocfs2_read_links_count(dirfe)) {
 293                 /* can't make a file in a deleted directory. */
 294                 status = -ENOENT;
 295                 goto leave;
 296         }
 297
 298         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 299                                            dentry->d_name.len);
 300         if (status)
 301                 goto leave;
 302
 303         /* get a spot inside the dir. */
 304         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 305                                               dentry->d_name.name,
 306                                               dentry->d_name.len, &lookup);
 307         if (status < 0) {
 308                 mlog_errno(status);
 309                 goto leave;
 310         }
 311
 312         /* reserve an inode spot */
 313         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 314         if (status < 0) {
 315                 if (status != -ENOSPC)
 316                         mlog_errno(status);
 317                 goto leave;
 318         }
 319
 320         inode = ocfs2_get_init_inode(dir, mode);
 321         if (IS_ERR(inode)) {
 322                 status = PTR_ERR(inode);
 323                 inode = NULL;
 324                 mlog_errno(status);
 325                 goto leave;
 326         }
 327
 328         /* get security xattr */
 329         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 330         if (status) {
 331                 if (status == -EOPNOTSUPP)
 332                         si.enable = 0;
 333                 else {
 334                         mlog_errno(status);
 335                         goto leave;
 336                 }
 337         }
 338
 339         /* calculate meta data/clusters for setting security and acl xattr */
 340         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 341                                        &si, &want_clusters,
 342                                        &xattr_credits, &want_meta);
 343         if (status < 0) {
 344                 mlog_errno(status);
 345                 goto leave;
 346         }
 347
 348         /* Reserve a cluster if creating an extent based directory. */
 349         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 350                 want_clusters += 1;
 351
 352                 /* Dir indexing requires extra space as well */
 353                 if (ocfs2_supports_indexed_dirs(osb))
 354                         want_meta++;
 355         }
 356
 357         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 358         if (status < 0) {
 359                 if (status != -ENOSPC)
 360                         mlog_errno(status);
 361                 goto leave;
 362         }
 363
 364         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 365         if (status < 0) {
 366                 if (status != -ENOSPC)
 367                         mlog_errno(status);
 368                 goto leave;
 369         }
 370
 371         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 372                                                             S_ISDIR(mode),
 373                                                             xattr_credits));
 374         if (IS_ERR(handle)) {
 375                 status = PTR_ERR(handle);
 376                 handle = NULL;
 377                 mlog_errno(status);
 378                 goto leave;
 379         }
 380
 381         /* Starting to change things, restart is no longer possible. */
 382         ocfs2_block_signals(&oldset);
 383         did_block_signals = 1;
 384
 385         status = dquot_alloc_inode(inode);
 386         if (status)
 387                 goto leave;
 388         did_quota_inode = 1;
 389
 390         /* do the real work now. */
 391         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 392                                     &new_fe_bh, parent_fe_bh, handle,
 393                                     inode_ac);
 394         if (status < 0) {
 395                 mlog_errno(status);
 396                 goto leave;
 397         }
 398
 399         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
 400         if (S_ISDIR(mode)) {
 401                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 402                                             new_fe_bh, data_ac, meta_ac);
 403                 if (status < 0) {
 404                         mlog_errno(status);
 405                         goto leave;
 406                 }
 407
 408                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 409                                                  parent_fe_bh,
 410                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 411                 if (status < 0) {
 412                         mlog_errno(status);
 413                         goto leave;
 414                 }
 415                 ocfs2_add_links_count(dirfe, 1);
 416                 ocfs2_journal_dirty(handle, parent_fe_bh);
 417                 inc_nlink(dir);
 418         }
 419
 420         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 421                          meta_ac, data_ac);
 422
 423         if (status < 0) {
 424                 mlog_errno(status);
 425                 goto leave;
 426         }
 427
 428         if (si.enable) {
 429                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 430                                                  meta_ac, data_ac);
 431                 if (status < 0) {
 432                         mlog_errno(status);
 433                         goto leave;
 434                 }
 435         }
 436
 437         /*
 438          * Do this before adding the entry to the directory. We add
 439          * also set d_op after success so that ->d_iput() will cleanup
 440          * the dentry lock even if ocfs2_add_entry() fails below.
 441          */
 442         status = ocfs2_dentry_attach_lock(dentry, inode,
 443                                           OCFS2_I(dir)->ip_blkno);
 444         if (status) {
 445                 mlog_errno(status);
 446                 goto leave;
 447         }
 448
 449         dl = dentry->d_fsdata;
 450
 451         status = ocfs2_add_entry(handle, dentry, inode,
 452                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 453                                  &lookup);
 454         if (status < 0) {
 455                 mlog_errno(status);
 456                 goto leave;
 457         }
 458
 459         insert_inode_hash(inode);
 460         d_instantiate(dentry, inode);
 461         status = 0;
 462 leave:
 463         if (status < 0 && did_quota_inode)
 464                 dquot_free_inode(inode);
 465         if (handle) {
 466                 if (status < 0 && fe)
 467                         ocfs2_set_links_count(fe, 0);
 468                 ocfs2_commit_trans(osb, handle);
 469         }
 470
 471         ocfs2_inode_unlock(dir, 1);
 472         if (did_block_signals)
 473                 ocfs2_unblock_signals(&oldset);
 474
 475         brelse(new_fe_bh);
 476         brelse(parent_fe_bh);
 477         kfree(si.value);
 478
 479         ocfs2_free_dir_lookup_result(&lookup);
 480
 481         if (inode_ac)
 482                 ocfs2_free_alloc_context(inode_ac);
 483
 484         if (data_ac)
 485                 ocfs2_free_alloc_context(data_ac);
 486
 487         if (meta_ac)
 488                 ocfs2_free_alloc_context(meta_ac);
 489
 490         /*
 491          * We should call iput after the i_mutex of the bitmap been
 492          * unlocked in ocfs2_free_alloc_context, or the
 493          * ocfs2_delete_inode will mutex_lock again.
 494          */
 495         if ((status < 0) && inode) {
 496                 if (dl)
 497                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 498
 499                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 500                 clear_nlink(inode);
 501                 iput(inode);
 502         }
 503
 504         if (status)
 505                 mlog_errno(status);
 506
 507         return status;
 508 }
 509
 510 static int __ocfs2_mknod_locked(struct inode *dir,
 511                                 struct inode *inode,
 512                                 dev_t dev,
 513                                 struct buffer_head **new_fe_bh,
 514                                 struct buffer_head *parent_fe_bh,
 515                                 handle_t *handle,
 516                                 struct ocfs2_alloc_context *inode_ac,
 517                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 518 {
 519         int status = 0;
 520         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 521         struct ocfs2_dinode *fe = NULL;
 522         struct ocfs2_extent_list *fel;
 523         u16 feat;
 524         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 525         struct timespec64 ts;
 526
 527         *new_fe_bh = NULL;
 528
 529         /* populate as many fields early on as possible - many of
 530          * these are used by the support functions here and in
 531          * callers. */
 532         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 533         oi->ip_blkno = fe_blkno;
 534         spin_lock(&osb->osb_lock);
 535         inode->i_generation = osb->s_next_generation++;
 536         spin_unlock(&osb->osb_lock);
 537
 538         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 539         if (!*new_fe_bh) {
 540                 status = -ENOMEM;
 541                 mlog_errno(status);
 542                 goto leave;
 543         }
 544         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 545
 546         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 547                                          *new_fe_bh,
 548                                          OCFS2_JOURNAL_ACCESS_CREATE);
 549         if (status < 0) {
 550                 mlog_errno(status);
 551                 goto leave;
 552         }
 553
 554         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 555         memset(fe, 0, osb->sb->s_blocksize);
 556
 557         fe->i_generation = cpu_to_le32(inode->i_generation);
 558         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 559         fe->i_blkno = cpu_to_le64(fe_blkno);
 560         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 561         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 562         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 563         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 564         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 565         fe->i_mode = cpu_to_le16(inode->i_mode);
 566         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 567                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 568
 569         ocfs2_set_links_count(fe, inode->i_nlink);
 570
 571         fe->i_last_eb_blk = 0;
 572         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 573         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 574         ktime_get_real_ts64(&ts);
 575         fe->i_atime = fe->i_ctime = fe->i_mtime =
 576                 cpu_to_le64(ts.tv_sec);
 577         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 578                 cpu_to_le32(ts.tv_nsec);
 579         fe->i_dtime = 0;
 580
 581         /*
 582          * If supported, directories start with inline data. If inline
 583          * isn't supported, but indexing is, we start them as indexed.
 584          */
 585         feat = le16_to_cpu(fe->i_dyn_features);
 586         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 587                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 588
 589                 fe->id2.i_data.id_count = cpu_to_le16(
 590                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 591         } else {
 592                 fel = &fe->id2.i_list;
 593                 fel->l_tree_depth = 0;
 594                 fel->l_next_free_rec = 0;
 595                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 596         }
 597
 598         ocfs2_journal_dirty(handle, *new_fe_bh);
 599
 600         ocfs2_populate_inode(inode, fe, 1);
 601         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 602         if (!ocfs2_mount_local(osb)) {
 603                 status = ocfs2_create_new_inode_locks(inode);
 604                 if (status < 0)
 605                         mlog_errno(status);
 606         }
 607
 608         oi->i_sync_tid = handle->h_transaction->t_tid;
 609         oi->i_datasync_tid = handle->h_transaction->t_tid;
 610
 611 leave:
 612         if (status < 0) {
 613                 if (*new_fe_bh) {
 614                         brelse(*new_fe_bh);
 615                         *new_fe_bh = NULL;
 616                 }
 617         }
 618
 619         if (status)
 620                 mlog_errno(status);
 621         return status;
 622 }
 623
 624 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 625                               struct inode *dir,
 626                               struct inode *inode,
 627                               dev_t dev,
 628                               struct buffer_head **new_fe_bh,
 629                               struct buffer_head *parent_fe_bh,
 630                               handle_t *handle,
 631                               struct ocfs2_alloc_context *inode_ac)
 632 {
 633         int status = 0;
 634         u64 suballoc_loc, fe_blkno = 0;
 635         u16 suballoc_bit;
 636
 637         *new_fe_bh = NULL;
 638
 639         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 640                                        inode_ac, &suballoc_loc,
 641                                        &suballoc_bit, &fe_blkno);
 642         if (status < 0) {
 643                 mlog_errno(status);
 644                 return status;
 645         }
 646
 647         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 648                                     parent_fe_bh, handle, inode_ac,
 649                                     fe_blkno, suballoc_loc, suballoc_bit);
 650 }
 651
 652 static int ocfs2_mkdir(struct inode *dir,
 653                        struct dentry *dentry,
 654                        umode_t mode)
 655 {
 656         int ret;
 657
 658         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 659                           OCFS2_I(dir)->ip_blkno, mode);
 660         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 661         if (ret)
 662                 mlog_errno(ret);
 663
 664         return ret;
 665 }
 666
 667 static int ocfs2_create(struct inode *dir,
 668                         struct dentry *dentry,
 669                         umode_t mode,
 670                         bool excl)
 671 {
 672         int ret;
 673
 674         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 675                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 676         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 677         if (ret)
 678                 mlog_errno(ret);
 679
 680         return ret;
 681 }
 682
 683 static int ocfs2_link(struct dentry *old_dentry,
 684                       struct inode *dir,
 685                       struct dentry *dentry)
 686 {
 687         handle_t *handle;
 688         struct inode *inode = d_inode(old_dentry);
 689         struct inode *old_dir = d_inode(old_dentry->d_parent);
 690         int err;
 691         struct buffer_head *fe_bh = NULL;
 692         struct buffer_head *old_dir_bh = NULL;
 693         struct buffer_head *parent_fe_bh = NULL;
 694         struct ocfs2_dinode *fe = NULL;
 695         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 696         struct ocfs2_dir_lookup_result lookup = { NULL, };
 697         sigset_t oldset;
 698         u64 old_de_ino;
 699
 700         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 701                          old_dentry->d_name.len, old_dentry->d_name.name,
 702                          dentry->d_name.len, dentry->d_name.name);
 703
 704         if (S_ISDIR(inode->i_mode))
 705                 return -EPERM;
 706
 707         err = dquot_initialize(dir);
 708         if (err) {
 709                 mlog_errno(err);
 710                 return err;
 711         }
 712
 713         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 714                         &parent_fe_bh, dir, 0);
 715         if (err < 0) {
 716                 if (err != -ENOENT)
 717                         mlog_errno(err);
 718                 return err;
 719         }
 720
 721         /* make sure both dirs have bhs
 722          * get an extra ref on old_dir_bh if old==new */
 723         if (!parent_fe_bh) {
 724                 if (old_dir_bh) {
 725                         parent_fe_bh = old_dir_bh;
 726                         get_bh(parent_fe_bh);
 727                 } else {
 728                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 729                         err = -EIO;
 730                         goto out;
 731                 }
 732         }
 733
 734         if (!dir->i_nlink) {
 735                 err = -ENOENT;
 736                 goto out;
 737         }
 738
 739         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 740                         old_dentry->d_name.len, &old_de_ino);
 741         if (err) {
 742                 err = -ENOENT;
 743                 goto out;
 744         }
 745
 746         /*
 747          * Check whether another node removed the source inode while we
 748          * were in the vfs.
 749          */
 750         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 751                 err = -ENOENT;
 752                 goto out;
 753         }
 754
 755         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 756                                         dentry->d_name.len);
 757         if (err)
 758                 goto out;
 759
 760         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 761                                            dentry->d_name.name,
 762                                            dentry->d_name.len, &lookup);
 763         if (err < 0) {
 764                 mlog_errno(err);
 765                 goto out;
 766         }
 767
 768         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 769         if (err < 0) {
 770                 if (err != -ENOENT)
 771                         mlog_errno(err);
 772                 goto out;
 773         }
 774
 775         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 776         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 777                 err = -EMLINK;
 778                 goto out_unlock_inode;
 779         }
 780
 781         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 782         if (IS_ERR(handle)) {
 783                 err = PTR_ERR(handle);
 784                 handle = NULL;
 785                 mlog_errno(err);
 786                 goto out_unlock_inode;
 787         }
 788
 789         /* Starting to change things, restart is no longer possible. */
 790         ocfs2_block_signals(&oldset);
 791
 792         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 793                                       OCFS2_JOURNAL_ACCESS_WRITE);
 794         if (err < 0) {
 795                 mlog_errno(err);
 796                 goto out_commit;
 797         }
 798
 799         inc_nlink(inode);
 800         inode->i_ctime = current_time(inode);
 801         ocfs2_set_links_count(fe, inode->i_nlink);
 802         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 803         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 804         ocfs2_journal_dirty(handle, fe_bh);
 805
 806         err = ocfs2_add_entry(handle, dentry, inode,
 807                               OCFS2_I(inode)->ip_blkno,
 808                               parent_fe_bh, &lookup);
 809         if (err) {
 810                 ocfs2_add_links_count(fe, -1);
 811                 drop_nlink(inode);
 812                 mlog_errno(err);
 813                 goto out_commit;
 814         }
 815
 816         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 817         if (err) {
 818                 mlog_errno(err);
 819                 goto out_commit;
 820         }
 821
 822         ihold(inode);
 823         d_instantiate(dentry, inode);
 824
 825 out_commit:
 826         ocfs2_commit_trans(osb, handle);
 827         ocfs2_unblock_signals(&oldset);
 828 out_unlock_inode:
 829         ocfs2_inode_unlock(inode, 1);
 830
 831 out:
 832         ocfs2_double_unlock(old_dir, dir);
 833
 834         brelse(fe_bh);
 835         brelse(parent_fe_bh);
 836         brelse(old_dir_bh);
 837
 838         ocfs2_free_dir_lookup_result(&lookup);
 839
 840         if (err)
 841                 mlog_errno(err);
 842
 843         return err;
 844 }
 845
 846 /*
 847  * Takes and drops an exclusive lock on the given dentry. This will
 848  * force other nodes to drop it.
 849  */
 850 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 851 {
 852         int ret;
 853
 854         ret = ocfs2_dentry_lock(dentry, 1);
 855         if (ret)
 856                 mlog_errno(ret);
 857         else
 858                 ocfs2_dentry_unlock(dentry, 1);
 859
 860         return ret;
 861 }
 862
 863 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 864 {
 865         if (S_ISDIR(inode->i_mode)) {
 866                 if (inode->i_nlink == 2)
 867                         return 1;
 868                 return 0;
 869         }
 870
 871         if (inode->i_nlink == 1)
 872                 return 1;
 873         return 0;
 874 }
 875
 876 static int ocfs2_unlink(struct inode *dir,
 877                         struct dentry *dentry)
 878 {
 879         int status;
 880         int child_locked = 0;
 881         bool is_unlinkable = false;
 882         struct inode *inode = d_inode(dentry);
 883         struct inode *orphan_dir = NULL;
 884         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 885         u64 blkno;
 886         struct ocfs2_dinode *fe = NULL;
 887         struct buffer_head *fe_bh = NULL;
 888         struct buffer_head *parent_node_bh = NULL;
 889         handle_t *handle = NULL;
 890         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 891         struct ocfs2_dir_lookup_result lookup = { NULL, };
 892         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 893
 894         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 895                            dentry->d_name.name,
 896                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 897                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 898
 899         status = dquot_initialize(dir);
 900         if (status) {
 901                 mlog_errno(status);
 902                 return status;
 903         }
 904
 905         BUG_ON(d_inode(dentry->d_parent) != dir);
 906
 907         if (inode == osb->root_inode)
 908                 return -EPERM;
 909
 910         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 911                                          OI_LS_PARENT);
 912         if (status < 0) {
 913                 if (status != -ENOENT)
 914                         mlog_errno(status);
 915                 return status;
 916         }
 917
 918         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 919                                           dentry->d_name.len, &blkno, dir,
 920                                           &lookup);
 921         if (status < 0) {
 922                 if (status != -ENOENT)
 923                         mlog_errno(status);
 924                 goto leave;
 925         }
 926
 927         if (OCFS2_I(inode)->ip_blkno != blkno) {
 928                 status = -ENOENT;
 929
 930                 trace_ocfs2_unlink_noent(
 931                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 932                                 (unsigned long long)blkno,
 933                                 OCFS2_I(inode)->ip_flags);
 934                 goto leave;
 935         }
 936
 937         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 938         if (status < 0) {
 939                 if (status != -ENOENT)
 940                         mlog_errno(status);
 941                 goto leave;
 942         }
 943         child_locked = 1;
 944
 945         if (S_ISDIR(inode->i_mode)) {
 946                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 947                         status = -ENOTEMPTY;
 948                         goto leave;
 949                 }
 950         }
 951
 952         status = ocfs2_remote_dentry_delete(dentry);
 953         if (status < 0) {
 954                 /* This remote delete should succeed under all normal
 955                  * circumstances. */
 956                 mlog_errno(status);
 957                 goto leave;
 958         }
 959
 960         if (ocfs2_inode_is_unlinkable(inode)) {
 961                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 962                                                   OCFS2_I(inode)->ip_blkno,
 963                                                   orphan_name, &orphan_insert,
 964                                                   false);
 965                 if (status < 0) {
 966                         mlog_errno(status);
 967                         goto leave;
 968                 }
 969                 is_unlinkable = true;
 970         }
 971
 972         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 973         if (IS_ERR(handle)) {
 974                 status = PTR_ERR(handle);
 975                 handle = NULL;
 976                 mlog_errno(status);
 977                 goto leave;
 978         }
 979
 980         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 981                                          OCFS2_JOURNAL_ACCESS_WRITE);
 982         if (status < 0) {
 983                 mlog_errno(status);
 984                 goto leave;
 985         }
 986
 987         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 988
 989         /* delete the name from the parent dir */
 990         status = ocfs2_delete_entry(handle, dir, &lookup);
 991         if (status < 0) {
 992                 mlog_errno(status);
 993                 goto leave;
 994         }
 995
 996         if (S_ISDIR(inode->i_mode))
 997                 drop_nlink(inode);
 998         drop_nlink(inode);
 999         ocfs2_set_links_count(fe, inode->i_nlink);
1000         ocfs2_journal_dirty(handle, fe_bh);
1001
1002         dir->i_ctime = dir->i_mtime = current_time(dir);
1003         if (S_ISDIR(inode->i_mode))
1004                 drop_nlink(dir);
1005
1006         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1007         if (status < 0) {
1008                 mlog_errno(status);
1009                 if (S_ISDIR(inode->i_mode))
1010                         inc_nlink(dir);
1011                 goto leave;
1012         }
1013
1014         if (is_unlinkable) {
1015                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1016                                 orphan_name, &orphan_insert, orphan_dir, false);
1017                 if (status < 0)
1018                         mlog_errno(status);
1019         }
1020
1021 leave:
1022         if (handle)
1023                 ocfs2_commit_trans(osb, handle);
1024
1025         if (orphan_dir) {
1026                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1027                 ocfs2_inode_unlock(orphan_dir, 1);
1028                 inode_unlock(orphan_dir);
1029                 iput(orphan_dir);
1030         }
1031
1032         if (child_locked)
1033                 ocfs2_inode_unlock(inode, 1);
1034
1035         ocfs2_inode_unlock(dir, 1);
1036
1037         brelse(fe_bh);
1038         brelse(parent_node_bh);
1039
1040         ocfs2_free_dir_lookup_result(&orphan_insert);
1041         ocfs2_free_dir_lookup_result(&lookup);
1042
1043         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1044                 mlog_errno(status);
1045
1046         return status;
1047 }
1048
1049 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1050                 u64 src_inode_no, u64 dest_inode_no)
1051 {
1052         int ret = 0, i = 0;
1053         u64 parent_inode_no = 0;
1054         u64 child_inode_no = src_inode_no;
1055         struct inode *child_inode;
1056
1057 #define MAX_LOOKUP_TIMES 32
1058         while (1) {
1059                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1060                 if (IS_ERR(child_inode)) {
1061                         ret = PTR_ERR(child_inode);
1062                         break;
1063                 }
1064
1065                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1066                 if (ret < 0) {
1067                         iput(child_inode);
1068                         if (ret != -ENOENT)
1069                                 mlog_errno(ret);
1070                         break;
1071                 }
1072
1073                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1074                                 &parent_inode_no);
1075                 ocfs2_inode_unlock(child_inode, 0);
1076                 iput(child_inode);
1077                 if (ret < 0) {
1078                         ret = -ENOENT;
1079                         break;
1080                 }
1081
1082                 if (parent_inode_no == dest_inode_no) {
1083                         ret = 1;
1084                         break;
1085                 }
1086
1087                 if (parent_inode_no == osb->root_inode->i_ino) {
1088                         ret = 0;
1089                         break;
1090                 }
1091
1092                 child_inode_no = parent_inode_no;
1093
1094                 if (++i >= MAX_LOOKUP_TIMES) {
1095                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1096                                         "may have nested directories, "
1097                                         "src inode: %llu, dest inode: %llu.\n",
1098                                         (unsigned long long)src_inode_no,
1099                                         (unsigned long long)dest_inode_no);
1100                         ret = 0;
1101                         break;
1102                 }
1103         }
1104
1105         return ret;
1106 }
1107
1108 /*
1109  * The only place this should be used is rename and link!
1110  * if they have the same id, then the 1st one is the only one locked.
1111  */
1112 static int ocfs2_double_lock(struct ocfs2_super *osb,
1113                              struct buffer_head **bh1,
1114                              struct inode *inode1,
1115                              struct buffer_head **bh2,
1116                              struct inode *inode2,
1117                              int rename)
1118 {
1119         int status;
1120         int inode1_is_ancestor, inode2_is_ancestor;
1121         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1122         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1123
1124         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1125                                 (unsigned long long)oi2->ip_blkno);
1126
1127         if (*bh1)
1128                 *bh1 = NULL;
1129         if (*bh2)
1130                 *bh2 = NULL;
1131
1132         /* we always want to lock the one with the lower lockid first.
1133          * and if they are nested, we lock ancestor first */
1134         if (oi1->ip_blkno != oi2->ip_blkno) {
1135                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1136                                 oi1->ip_blkno);
1137                 if (inode1_is_ancestor < 0) {
1138                         status = inode1_is_ancestor;
1139                         goto bail;
1140                 }
1141
1142                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1143                                 oi2->ip_blkno);
1144                 if (inode2_is_ancestor < 0) {
1145                         status = inode2_is_ancestor;
1146                         goto bail;
1147                 }
1148
1149                 if ((inode1_is_ancestor == 1) ||
1150                                 (oi1->ip_blkno < oi2->ip_blkno &&
1151                                 inode2_is_ancestor == 0)) {
1152                         /* switch id1 and id2 around */
1153                         swap(bh2, bh1);
1154                         swap(inode2, inode1);
1155                 }
1156                 /* lock id2 */
1157                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1158                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1159                 if (status < 0) {
1160                         if (status != -ENOENT)
1161                                 mlog_errno(status);
1162                         goto bail;
1163                 }
1164         }
1165
1166         /* lock id1 */
1167         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1168                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1169         if (status < 0) {
1170                 /*
1171                  * An error return must mean that no cluster locks
1172                  * were held on function exit.
1173                  */
1174                 if (oi1->ip_blkno != oi2->ip_blkno) {
1175                         ocfs2_inode_unlock(inode2, 1);
1176                         brelse(*bh2);
1177                         *bh2 = NULL;
1178                 }
1179
1180                 if (status != -ENOENT)
1181                         mlog_errno(status);
1182         }
1183
1184         trace_ocfs2_double_lock_end(
1185                         (unsigned long long)oi1->ip_blkno,
1186                         (unsigned long long)oi2->ip_blkno);
1187
1188 bail:
1189         if (status)
1190                 mlog_errno(status);
1191         return status;
1192 }
1193
1194 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1195 {
1196         ocfs2_inode_unlock(inode1, 1);
1197
1198         if (inode1 != inode2)
1199                 ocfs2_inode_unlock(inode2, 1);
1200 }
1201
1202 static int ocfs2_rename(struct inode *old_dir,
1203                         struct dentry *old_dentry,
1204                         struct inode *new_dir,
1205                         struct dentry *new_dentry,
1206                         unsigned int flags)
1207 {
1208         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1209         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1210         struct inode *old_inode = d_inode(old_dentry);
1211         struct inode *new_inode = d_inode(new_dentry);
1212         struct inode *orphan_dir = NULL;
1213         struct ocfs2_dinode *newfe = NULL;
1214         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1215         struct buffer_head *newfe_bh = NULL;
1216         struct buffer_head *old_inode_bh = NULL;
1217         struct ocfs2_super *osb = NULL;
1218         u64 newfe_blkno, old_de_ino;
1219         handle_t *handle = NULL;
1220         struct buffer_head *old_dir_bh = NULL;
1221         struct buffer_head *new_dir_bh = NULL;
1222         u32 old_dir_nlink = old_dir->i_nlink;
1223         struct ocfs2_dinode *old_di;
1224         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1225         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1226         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1227         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1228         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1229         bool should_add_orphan = false;
1230
1231         if (flags)
1232                 return -EINVAL;
1233
1234         /* At some point it might be nice to break this function up a
1235          * bit. */
1236
1237         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1238                            old_dentry->d_name.len, old_dentry->d_name.name,
1239                            new_dentry->d_name.len, new_dentry->d_name.name);
1240
1241         status = dquot_initialize(old_dir);
1242         if (status) {
1243                 mlog_errno(status);
1244                 goto bail;
1245         }
1246         status = dquot_initialize(new_dir);
1247         if (status) {
1248                 mlog_errno(status);
1249                 goto bail;
1250         }
1251
1252         osb = OCFS2_SB(old_dir->i_sb);
1253
1254         if (new_inode) {
1255                 if (!igrab(new_inode))
1256                         BUG();
1257         }
1258
1259         /* Assume a directory hierarchy thusly:
1260          * a/b/c
1261          * a/d
1262          * a,b,c, and d are all directories.
1263          *
1264          * from cwd of 'a' on both nodes:
1265          * node1: mv b/c d
1266          * node2: mv d   b/c
1267          *
1268          * And that's why, just like the VFS, we need a file system
1269          * rename lock. */
1270         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1271                 status = ocfs2_rename_lock(osb);
1272                 if (status < 0) {
1273                         mlog_errno(status);
1274                         goto bail;
1275                 }
1276                 rename_lock = 1;
1277
1278                 /* here we cannot guarantee the inodes haven't just been
1279                  * changed, so check if they are nested again */
1280                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1281                                 old_inode->i_ino);
1282                 if (status < 0) {
1283                         mlog_errno(status);
1284                         goto bail;
1285                 } else if (status == 1) {
1286                         status = -EPERM;
1287                         trace_ocfs2_rename_not_permitted(
1288                                         (unsigned long long)old_inode->i_ino,
1289                                         (unsigned long long)new_dir->i_ino);
1290                         goto bail;
1291                 }
1292         }
1293
1294         /* if old and new are the same, this'll just do one lock. */
1295         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1296                                    &new_dir_bh, new_dir, 1);
1297         if (status < 0) {
1298                 mlog_errno(status);
1299                 goto bail;
1300         }
1301         parents_locked = 1;
1302
1303         if (!new_dir->i_nlink) {
1304                 status = -EACCES;
1305                 goto bail;
1306         }
1307
1308         /* make sure both dirs have bhs
1309          * get an extra ref on old_dir_bh if old==new */
1310         if (!new_dir_bh) {
1311                 if (old_dir_bh) {
1312                         new_dir_bh = old_dir_bh;
1313                         get_bh(new_dir_bh);
1314                 } else {
1315                         mlog(ML_ERROR, "no old_dir_bh!\n");
1316                         status = -EIO;
1317                         goto bail;
1318                 }
1319         }
1320
1321         /*
1322          * Aside from allowing a meta data update, the locking here
1323          * also ensures that the downconvert thread on other nodes
1324          * won't have to concurrently downconvert the inode and the
1325          * dentry locks.
1326          */
1327         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1328                                          OI_LS_PARENT);
1329         if (status < 0) {
1330                 if (status != -ENOENT)
1331                         mlog_errno(status);
1332                 goto bail;
1333         }
1334         old_child_locked = 1;
1335
1336         status = ocfs2_remote_dentry_delete(old_dentry);
1337         if (status < 0) {
1338                 mlog_errno(status);
1339                 goto bail;
1340         }
1341
1342         if (S_ISDIR(old_inode->i_mode)) {
1343                 u64 old_inode_parent;
1344
1345                 update_dot_dot = 1;
1346                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1347                                                   old_inode,
1348                                                   &old_inode_dot_dot_res);
1349                 if (status) {
1350                         status = -EIO;
1351                         goto bail;
1352                 }
1353
1354                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1355                         status = -EIO;
1356                         goto bail;
1357                 }
1358
1359                 if (!new_inode && new_dir != old_dir &&
1360                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1361                         status = -EMLINK;
1362                         goto bail;
1363                 }
1364         }
1365
1366         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1367                                             old_dentry->d_name.len,
1368                                             &old_de_ino);
1369         if (status) {
1370                 status = -ENOENT;
1371                 goto bail;
1372         }
1373
1374         /*
1375          *  Check for inode number is _not_ due to possible IO errors.
1376          *  We might rmdir the source, keep it as pwd of some process
1377          *  and merrily kill the link to whatever was created under the
1378          *  same name. Goodbye sticky bit ;-<
1379          */
1380         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1381                 status = -ENOENT;
1382                 goto bail;
1383         }
1384
1385         /* check if the target already exists (in which case we need
1386          * to delete it */
1387         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1388                                           new_dentry->d_name.len,
1389                                           &newfe_blkno, new_dir,
1390                                           &target_lookup_res);
1391         /* The only error we allow here is -ENOENT because the new
1392          * file not existing is perfectly valid. */
1393         if ((status < 0) && (status != -ENOENT)) {
1394                 /* If we cannot find the file specified we should just */
1395                 /* return the error... */
1396                 mlog_errno(status);
1397                 goto bail;
1398         }
1399         if (status == 0)
1400                 target_exists = 1;
1401
1402         if (!target_exists && new_inode) {
1403                 /*
1404                  * Target was unlinked by another node while we were
1405                  * waiting to get to ocfs2_rename(). There isn't
1406                  * anything we can do here to help the situation, so
1407                  * bubble up the appropriate error.
1408                  */
1409                 status = -ENOENT;
1410                 goto bail;
1411         }
1412
1413         /* In case we need to overwrite an existing file, we blow it
1414          * away first */
1415         if (target_exists) {
1416                 /* VFS didn't think there existed an inode here, but
1417                  * someone else in the cluster must have raced our
1418                  * rename to create one. Today we error cleanly, in
1419                  * the future we should consider calling iget to build
1420                  * a new struct inode for this entry. */
1421                 if (!new_inode) {
1422                         status = -EACCES;
1423
1424                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1425                                                 new_dentry->d_name.name);
1426                         goto bail;
1427                 }
1428
1429                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1430                         status = -EACCES;
1431
1432                         trace_ocfs2_rename_disagree(
1433                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1434                              (unsigned long long)newfe_blkno,
1435                              OCFS2_I(new_inode)->ip_flags);
1436                         goto bail;
1437                 }
1438
1439                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1440                 if (status < 0) {
1441                         if (status != -ENOENT)
1442                                 mlog_errno(status);
1443                         goto bail;
1444                 }
1445                 new_child_locked = 1;
1446
1447                 status = ocfs2_remote_dentry_delete(new_dentry);
1448                 if (status < 0) {
1449                         mlog_errno(status);
1450                         goto bail;
1451                 }
1452
1453                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1454
1455                 trace_ocfs2_rename_over_existing(
1456                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1457                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1458
1459                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1460                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1461                                                 OCFS2_I(new_inode)->ip_blkno,
1462                                                 orphan_name, &orphan_insert,
1463                                                 false);
1464                         if (status < 0) {
1465                                 mlog_errno(status);
1466                                 goto bail;
1467                         }
1468                         should_add_orphan = true;
1469                 }
1470         } else {
1471                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1472
1473                 status = ocfs2_check_dir_for_entry(new_dir,
1474                                                    new_dentry->d_name.name,
1475                                                    new_dentry->d_name.len);
1476                 if (status)
1477                         goto bail;
1478
1479                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1480                                                       new_dentry->d_name.name,
1481                                                       new_dentry->d_name.len,
1482                                                       &target_insert);
1483                 if (status < 0) {
1484                         mlog_errno(status);
1485                         goto bail;
1486                 }
1487         }
1488
1489         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1490         if (IS_ERR(handle)) {
1491                 status = PTR_ERR(handle);
1492                 handle = NULL;
1493                 mlog_errno(status);
1494                 goto bail;
1495         }
1496
1497         if (target_exists) {
1498                 if (S_ISDIR(new_inode->i_mode)) {
1499                         if (new_inode->i_nlink != 2 ||
1500                             !ocfs2_empty_dir(new_inode)) {
1501                                 status = -ENOTEMPTY;
1502                                 goto bail;
1503                         }
1504                 }
1505                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1506                                                  newfe_bh,
1507                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1508                 if (status < 0) {
1509                         mlog_errno(status);
1510                         goto bail;
1511                 }
1512
1513                 /* change the dirent to point to the correct inode */
1514                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1515                                             old_inode);
1516                 if (status < 0) {
1517                         mlog_errno(status);
1518                         goto bail;
1519                 }
1520                 inode_inc_iversion(new_dir);
1521
1522                 if (S_ISDIR(new_inode->i_mode))
1523                         ocfs2_set_links_count(newfe, 0);
1524                 else
1525                         ocfs2_add_links_count(newfe, -1);
1526                 ocfs2_journal_dirty(handle, newfe_bh);
1527                 if (should_add_orphan) {
1528                         status = ocfs2_orphan_add(osb, handle, new_inode,
1529                                         newfe_bh, orphan_name,
1530                                         &orphan_insert, orphan_dir, false);
1531                         if (status < 0) {
1532                                 mlog_errno(status);
1533                                 goto bail;
1534                         }
1535                 }
1536         } else {
1537                 /* if the name was not found in new_dir, add it now */
1538                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1539                                          OCFS2_I(old_inode)->ip_blkno,
1540                                          new_dir_bh, &target_insert);
1541         }
1542
1543         old_inode->i_ctime = current_time(old_inode);
1544         mark_inode_dirty(old_inode);
1545
1546         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1547                                          old_inode_bh,
1548                                          OCFS2_JOURNAL_ACCESS_WRITE);
1549         if (status >= 0) {
1550                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1551
1552                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1553                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1554                 ocfs2_journal_dirty(handle, old_inode_bh);
1555         } else
1556                 mlog_errno(status);
1557
1558         /*
1559          * Now that the name has been added to new_dir, remove the old name.
1560          *
1561          * We don't keep any directory entry context around until now
1562          * because the insert might have changed the type of directory
1563          * we're dealing with.
1564          */
1565         status = ocfs2_find_entry(old_dentry->d_name.name,
1566                                   old_dentry->d_name.len, old_dir,
1567                                   &old_entry_lookup);
1568         if (status) {
1569                 if (!is_journal_aborted(osb->journal->j_journal)) {
1570                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1571                                         "is not deleted.",
1572                                         new_dentry->d_name.len, new_dentry->d_name.name,
1573                                         old_dentry->d_name.len, old_dentry->d_name.name);
1574                 }
1575                 goto bail;
1576         }
1577
1578         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1579         if (status < 0) {
1580                 mlog_errno(status);
1581                 if (!is_journal_aborted(osb->journal->j_journal)) {
1582                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1583                                         "is not deleted.",
1584                                         new_dentry->d_name.len, new_dentry->d_name.name,
1585                                         old_dentry->d_name.len, old_dentry->d_name.name);
1586                 }
1587                 goto bail;
1588         }
1589
1590         if (new_inode) {
1591                 drop_nlink(new_inode);
1592                 new_inode->i_ctime = current_time(new_inode);
1593         }
1594         old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1595
1596         if (update_dot_dot) {
1597                 status = ocfs2_update_entry(old_inode, handle,
1598                                             &old_inode_dot_dot_res, new_dir);
1599                 drop_nlink(old_dir);
1600                 if (new_inode) {
1601                         drop_nlink(new_inode);
1602                 } else {
1603                         inc_nlink(new_dir);
1604                         mark_inode_dirty(new_dir);
1605                 }
1606         }
1607         mark_inode_dirty(old_dir);
1608         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1609         if (new_inode) {
1610                 mark_inode_dirty(new_inode);
1611                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1612         }
1613
1614         if (old_dir != new_dir) {
1615                 /* Keep the same times on both directories.*/
1616                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1617
1618                 /*
1619                  * This will also pick up the i_nlink change from the
1620                  * block above.
1621                  */
1622                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1623         }
1624
1625         if (old_dir_nlink != old_dir->i_nlink) {
1626                 if (!old_dir_bh) {
1627                         mlog(ML_ERROR, "need to change nlink for old dir "
1628                              "%llu from %d to %d but bh is NULL!\n",
1629                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1630                              (int)old_dir_nlink, old_dir->i_nlink);
1631                 } else {
1632                         struct ocfs2_dinode *fe;
1633                         status = ocfs2_journal_access_di(handle,
1634                                                          INODE_CACHE(old_dir),
1635                                                          old_dir_bh,
1636                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1637                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1638                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1639                         ocfs2_journal_dirty(handle, old_dir_bh);
1640                 }
1641         }
1642         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1643         status = 0;
1644 bail:
1645         if (handle)
1646                 ocfs2_commit_trans(osb, handle);
1647
1648         if (orphan_dir) {
1649                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1650                 ocfs2_inode_unlock(orphan_dir, 1);
1651                 inode_unlock(orphan_dir);
1652                 iput(orphan_dir);
1653         }
1654
1655         if (new_child_locked)
1656                 ocfs2_inode_unlock(new_inode, 1);
1657
1658         if (old_child_locked)
1659                 ocfs2_inode_unlock(old_inode, 1);
1660
1661         if (parents_locked)
1662                 ocfs2_double_unlock(old_dir, new_dir);
1663
1664         if (rename_lock)
1665                 ocfs2_rename_unlock(osb);
1666
1667         if (new_inode)
1668                 sync_mapping_buffers(old_inode->i_mapping);
1669
1670         iput(new_inode);
1671
1672         ocfs2_free_dir_lookup_result(&target_lookup_res);
1673         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1674         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1675         ocfs2_free_dir_lookup_result(&orphan_insert);
1676         ocfs2_free_dir_lookup_result(&target_insert);
1677
1678         brelse(newfe_bh);
1679         brelse(old_inode_bh);
1680         brelse(old_dir_bh);
1681         brelse(new_dir_bh);
1682
1683         if (status)
1684                 mlog_errno(status);
1685
1686         return status;
1687 }
1688
1689 /*
1690  * we expect i_size = strlen(symname). Copy symname into the file
1691  * data, including the null terminator.
1692  */
1693 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1694                                      handle_t *handle,
1695                                      struct inode *inode,
1696                                      const char *symname)
1697 {
1698         struct buffer_head **bhs = NULL;
1699         const char *c;
1700         struct super_block *sb = osb->sb;
1701         u64 p_blkno, p_blocks;
1702         int virtual, blocks, status, i, bytes_left;
1703
1704         bytes_left = i_size_read(inode) + 1;
1705         /* we can't trust i_blocks because we're actually going to
1706          * write i_size + 1 bytes. */
1707         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1708
1709         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1710                                         i_size_read(inode), blocks);
1711
1712         /* Sanity check -- make sure we're going to fit. */
1713         if (bytes_left >
1714             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1715                 status = -EIO;
1716                 mlog_errno(status);
1717                 goto bail;
1718         }
1719
1720         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1721         if (!bhs) {
1722                 status = -ENOMEM;
1723                 mlog_errno(status);
1724                 goto bail;
1725         }
1726
1727         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1728                                              NULL);
1729         if (status < 0) {
1730                 mlog_errno(status);
1731                 goto bail;
1732         }
1733
1734         /* links can never be larger than one cluster so we know this
1735          * is all going to be contiguous, but do a sanity check
1736          * anyway. */
1737         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1738                 status = -EIO;
1739                 mlog_errno(status);
1740                 goto bail;
1741         }
1742
1743         virtual = 0;
1744         while(bytes_left > 0) {
1745                 c = &symname[virtual * sb->s_blocksize];
1746
1747                 bhs[virtual] = sb_getblk(sb, p_blkno);
1748                 if (!bhs[virtual]) {
1749                         status = -ENOMEM;
1750                         mlog_errno(status);
1751                         goto bail;
1752                 }
1753                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1754                                               bhs[virtual]);
1755
1756                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1757                                               bhs[virtual],
1758                                               OCFS2_JOURNAL_ACCESS_CREATE);
1759                 if (status < 0) {
1760                         mlog_errno(status);
1761                         goto bail;
1762                 }
1763
1764                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1765
1766                 memcpy(bhs[virtual]->b_data, c,
1767                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1768                        bytes_left);
1769
1770                 ocfs2_journal_dirty(handle, bhs[virtual]);
1771
1772                 virtual++;
1773                 p_blkno++;
1774                 bytes_left -= sb->s_blocksize;
1775         }
1776
1777         status = 0;
1778 bail:
1779
1780         if (bhs) {
1781                 for(i = 0; i < blocks; i++)
1782                         brelse(bhs[i]);
1783                 kfree(bhs);
1784         }
1785
1786         if (status)
1787                 mlog_errno(status);
1788         return status;
1789 }
1790
1791 static int ocfs2_symlink(struct inode *dir,
1792                          struct dentry *dentry,
1793                          const char *symname)
1794 {
1795         int status, l, credits;
1796         u64 newsize;
1797         struct ocfs2_super *osb = NULL;
1798         struct inode *inode = NULL;
1799         struct super_block *sb;
1800         struct buffer_head *new_fe_bh = NULL;
1801         struct buffer_head *parent_fe_bh = NULL;
1802         struct ocfs2_dinode *fe = NULL;
1803         struct ocfs2_dinode *dirfe;
1804         handle_t *handle = NULL;
1805         struct ocfs2_alloc_context *inode_ac = NULL;
1806         struct ocfs2_alloc_context *data_ac = NULL;
1807         struct ocfs2_alloc_context *xattr_ac = NULL;
1808         int want_clusters = 0;
1809         int xattr_credits = 0;
1810         struct ocfs2_security_xattr_info si = {
1811                 .enable = 1,
1812         };
1813         int did_quota = 0, did_quota_inode = 0;
1814         struct ocfs2_dir_lookup_result lookup = { NULL, };
1815         sigset_t oldset;
1816         int did_block_signals = 0;
1817         struct ocfs2_dentry_lock *dl = NULL;
1818
1819         trace_ocfs2_symlink_begin(dir, dentry, symname,
1820                                   dentry->d_name.len, dentry->d_name.name);
1821
1822         status = dquot_initialize(dir);
1823         if (status) {
1824                 mlog_errno(status);
1825                 goto bail;
1826         }
1827
1828         sb = dir->i_sb;
1829         osb = OCFS2_SB(sb);
1830
1831         l = strlen(symname) + 1;
1832
1833         credits = ocfs2_calc_symlink_credits(sb);
1834
1835         /* lock the parent directory */
1836         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1837         if (status < 0) {
1838                 if (status != -ENOENT)
1839                         mlog_errno(status);
1840                 return status;
1841         }
1842
1843         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1844         if (!ocfs2_read_links_count(dirfe)) {
1845                 /* can't make a file in a deleted directory. */
1846                 status = -ENOENT;
1847                 goto bail;
1848         }
1849
1850         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1851                                            dentry->d_name.len);
1852         if (status)
1853                 goto bail;
1854
1855         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1856                                               dentry->d_name.name,
1857                                               dentry->d_name.len, &lookup);
1858         if (status < 0) {
1859                 mlog_errno(status);
1860                 goto bail;
1861         }
1862
1863         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1864         if (status < 0) {
1865                 if (status != -ENOSPC)
1866                         mlog_errno(status);
1867                 goto bail;
1868         }
1869
1870         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1871         if (IS_ERR(inode)) {
1872                 status = PTR_ERR(inode);
1873                 inode = NULL;
1874                 mlog_errno(status);
1875                 goto bail;
1876         }
1877
1878         /* get security xattr */
1879         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1880         if (status) {
1881                 if (status == -EOPNOTSUPP)
1882                         si.enable = 0;
1883                 else {
1884                         mlog_errno(status);
1885                         goto bail;
1886                 }
1887         }
1888
1889         /* calculate meta data/clusters for setting security xattr */
1890         if (si.enable) {
1891                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1892                                                   &xattr_credits, &xattr_ac);
1893                 if (status < 0) {
1894                         mlog_errno(status);
1895                         goto bail;
1896                 }
1897         }
1898
1899         /* don't reserve bitmap space for fast symlinks. */
1900         if (l > ocfs2_fast_symlink_chars(sb))
1901                 want_clusters += 1;
1902
1903         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1904         if (status < 0) {
1905                 if (status != -ENOSPC)
1906                         mlog_errno(status);
1907                 goto bail;
1908         }
1909
1910         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1911         if (IS_ERR(handle)) {
1912                 status = PTR_ERR(handle);
1913                 handle = NULL;
1914                 mlog_errno(status);
1915                 goto bail;
1916         }
1917
1918         /* Starting to change things, restart is no longer possible. */
1919         ocfs2_block_signals(&oldset);
1920         did_block_signals = 1;
1921
1922         status = dquot_alloc_inode(inode);
1923         if (status)
1924                 goto bail;
1925         did_quota_inode = 1;
1926
1927         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1928                                    dentry->d_name.name,
1929                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1930                                    inode->i_mode);
1931
1932         status = ocfs2_mknod_locked(osb, dir, inode,
1933                                     0, &new_fe_bh, parent_fe_bh, handle,
1934                                     inode_ac);
1935         if (status < 0) {
1936                 mlog_errno(status);
1937                 goto bail;
1938         }
1939
1940         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1941         inode->i_rdev = 0;
1942         newsize = l - 1;
1943         inode->i_op = &ocfs2_symlink_inode_operations;
1944         inode_nohighmem(inode);
1945         if (l > ocfs2_fast_symlink_chars(sb)) {
1946                 u32 offset = 0;
1947
1948                 status = dquot_alloc_space_nodirty(inode,
1949                     ocfs2_clusters_to_bytes(osb->sb, 1));
1950                 if (status)
1951                         goto bail;
1952                 did_quota = 1;
1953                 inode->i_mapping->a_ops = &ocfs2_aops;
1954                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1955                                               new_fe_bh,
1956                                               handle, data_ac, NULL,
1957                                               NULL);
1958                 if (status < 0) {
1959                         if (status != -ENOSPC && status != -EINTR) {
1960                                 mlog(ML_ERROR,
1961                                      "Failed to extend file to %llu\n",
1962                                      (unsigned long long)newsize);
1963                                 mlog_errno(status);
1964                                 status = -ENOSPC;
1965                         }
1966                         goto bail;
1967                 }
1968                 i_size_write(inode, newsize);
1969                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1970         } else {
1971                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1972                 memcpy((char *) fe->id2.i_symlink, symname, l);
1973                 i_size_write(inode, newsize);
1974                 inode->i_blocks = 0;
1975         }
1976
1977         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1978         if (status < 0) {
1979                 mlog_errno(status);
1980                 goto bail;
1981         }
1982
1983         if (!ocfs2_inode_is_fast_symlink(inode)) {
1984                 status = ocfs2_create_symlink_data(osb, handle, inode,
1985                                                    symname);
1986                 if (status < 0) {
1987                         mlog_errno(status);
1988                         goto bail;
1989                 }
1990         }
1991
1992         if (si.enable) {
1993                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1994                                                  xattr_ac, data_ac);
1995                 if (status < 0) {
1996                         mlog_errno(status);
1997                         goto bail;
1998                 }
1999         }
2000
2001         /*
2002          * Do this before adding the entry to the directory. We add
2003          * also set d_op after success so that ->d_iput() will cleanup
2004          * the dentry lock even if ocfs2_add_entry() fails below.
2005          */
2006         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2007         if (status) {
2008                 mlog_errno(status);
2009                 goto bail;
2010         }
2011
2012         dl = dentry->d_fsdata;
2013
2014         status = ocfs2_add_entry(handle, dentry, inode,
2015                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2016                                  &lookup);
2017         if (status < 0) {
2018                 mlog_errno(status);
2019                 goto bail;
2020         }
2021
2022         insert_inode_hash(inode);
2023         d_instantiate(dentry, inode);
2024 bail:
2025         if (status < 0 && did_quota)
2026                 dquot_free_space_nodirty(inode,
2027                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2028         if (status < 0 && did_quota_inode)
2029                 dquot_free_inode(inode);
2030         if (handle) {
2031                 if (status < 0 && fe)
2032                         ocfs2_set_links_count(fe, 0);
2033                 ocfs2_commit_trans(osb, handle);
2034         }
2035
2036         ocfs2_inode_unlock(dir, 1);
2037         if (did_block_signals)
2038                 ocfs2_unblock_signals(&oldset);
2039
2040         brelse(new_fe_bh);
2041         brelse(parent_fe_bh);
2042         kfree(si.value);
2043         ocfs2_free_dir_lookup_result(&lookup);
2044         if (inode_ac)
2045                 ocfs2_free_alloc_context(inode_ac);
2046         if (data_ac)
2047                 ocfs2_free_alloc_context(data_ac);
2048         if (xattr_ac)
2049                 ocfs2_free_alloc_context(xattr_ac);
2050         if ((status < 0) && inode) {
2051                 if (dl)
2052                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2053
2054                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2055                 clear_nlink(inode);
2056                 iput(inode);
2057         }
2058
2059         if (status)
2060                 mlog_errno(status);
2061
2062         return status;
2063 }
2064
2065 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2066 {
2067         int status, namelen;
2068
2069         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2070                            (long long)blkno);
2071         if (namelen <= 0) {
2072                 if (namelen)
2073                         status = namelen;
2074                 else
2075                         status = -EINVAL;
2076                 mlog_errno(status);
2077                 goto bail;
2078         }
2079         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2080                 status = -EINVAL;
2081                 mlog_errno(status);
2082                 goto bail;
2083         }
2084
2085         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2086
2087         status = 0;
2088 bail:
2089         if (status < 0)
2090                 mlog_errno(status);
2091         return status;
2092 }
2093
2094 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2095                                         struct inode **ret_orphan_dir,
2096                                         struct buffer_head **ret_orphan_dir_bh)
2097 {
2098         struct inode *orphan_dir_inode;
2099         struct buffer_head *orphan_dir_bh = NULL;
2100         int ret = 0;
2101
2102         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2103                                                        ORPHAN_DIR_SYSTEM_INODE,
2104                                                        osb->slot_num);
2105         if (!orphan_dir_inode) {
2106                 ret = -ENOENT;
2107                 mlog_errno(ret);
2108                 return ret;
2109         }
2110
2111         inode_lock(orphan_dir_inode);
2112
2113         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2114         if (ret < 0) {
2115                 inode_unlock(orphan_dir_inode);
2116                 iput(orphan_dir_inode);
2117
2118                 mlog_errno(ret);
2119                 return ret;
2120         }
2121
2122         *ret_orphan_dir = orphan_dir_inode;
2123         *ret_orphan_dir_bh = orphan_dir_bh;
2124
2125         return 0;
2126 }
2127
2128 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2129                                       struct buffer_head *orphan_dir_bh,
2130                                       u64 blkno,
2131                                       char *name,
2132                                       struct ocfs2_dir_lookup_result *lookup,
2133                                       bool dio)
2134 {
2135         int ret;
2136         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2137         int namelen = dio ?
2138                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2139                         OCFS2_ORPHAN_NAMELEN;
2140
2141         if (dio) {
2142                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2143                                 OCFS2_DIO_ORPHAN_PREFIX);
2144                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2145                         ret = -EINVAL;
2146                         mlog_errno(ret);
2147                         return ret;
2148                 }
2149
2150                 ret = ocfs2_blkno_stringify(blkno,
2151                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2152         } else
2153                 ret = ocfs2_blkno_stringify(blkno, name);
2154         if (ret < 0) {
2155                 mlog_errno(ret);
2156                 return ret;
2157         }
2158
2159         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2160                                            orphan_dir_bh, name,
2161                                            namelen, lookup);
2162         if (ret < 0) {
2163                 mlog_errno(ret);
2164                 return ret;
2165         }
2166
2167         return 0;
2168 }
2169
2170 /**
2171  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2172  * insertion of an orphan.
2173  * @osb: ocfs2 file system
2174  * @ret_orphan_dir: Orphan dir inode - returned locked!
2175  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2176  * @lookup: dir lookup result, to be passed back into functions like
2177  *          ocfs2_orphan_add
2178  *
2179  * Returns zero on success and the ret_orphan_dir, name and lookup
2180  * fields will be populated.
2181  *
2182  * Returns non-zero on failure.
2183  */
2184 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2185                                     struct inode **ret_orphan_dir,
2186                                     u64 blkno,
2187                                     char *name,
2188                                     struct ocfs2_dir_lookup_result *lookup,
2189                                     bool dio)
2190 {
2191         struct inode *orphan_dir_inode = NULL;
2192         struct buffer_head *orphan_dir_bh = NULL;
2193         int ret = 0;
2194
2195         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2196                                            &orphan_dir_bh);
2197         if (ret < 0) {
2198                 mlog_errno(ret);
2199                 return ret;
2200         }
2201
2202         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2203                                          blkno, name, lookup, dio);
2204         if (ret < 0) {
2205                 mlog_errno(ret);
2206                 goto out;
2207         }
2208
2209         *ret_orphan_dir = orphan_dir_inode;
2210
2211 out:
2212         brelse(orphan_dir_bh);
2213
2214         if (ret) {
2215                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2216                 inode_unlock(orphan_dir_inode);
2217                 iput(orphan_dir_inode);
2218         }
2219
2220         if (ret)
2221                 mlog_errno(ret);
2222         return ret;
2223 }
2224
2225 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2226                             handle_t *handle,
2227                             struct inode *inode,
2228                             struct buffer_head *fe_bh,
2229                             char *name,
2230                             struct ocfs2_dir_lookup_result *lookup,
2231                             struct inode *orphan_dir_inode,
2232                             bool dio)
2233 {
2234         struct buffer_head *orphan_dir_bh = NULL;
2235         int status = 0;
2236         struct ocfs2_dinode *orphan_fe;
2237         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2238         int namelen = dio ?
2239                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2240                         OCFS2_ORPHAN_NAMELEN;
2241
2242         trace_ocfs2_orphan_add_begin(
2243                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2244
2245         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2246         if (status < 0) {
2247                 mlog_errno(status);
2248                 goto leave;
2249         }
2250
2251         status = ocfs2_journal_access_di(handle,
2252                                          INODE_CACHE(orphan_dir_inode),
2253                                          orphan_dir_bh,
2254                                          OCFS2_JOURNAL_ACCESS_WRITE);
2255         if (status < 0) {
2256                 mlog_errno(status);
2257                 goto leave;
2258         }
2259
2260         /*
2261          * We're going to journal the change of i_flags and i_orphaned_slot.
2262          * It's safe anyway, though some callers may duplicate the journaling.
2263          * Journaling within the func just make the logic look more
2264          * straightforward.
2265          */
2266         status = ocfs2_journal_access_di(handle,
2267                                          INODE_CACHE(inode),
2268                                          fe_bh,
2269                                          OCFS2_JOURNAL_ACCESS_WRITE);
2270         if (status < 0) {
2271                 mlog_errno(status);
2272                 goto leave;
2273         }
2274
2275         /* we're a cluster, and nlink can change on disk from
2276          * underneath us... */
2277         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2278         if (S_ISDIR(inode->i_mode))
2279                 ocfs2_add_links_count(orphan_fe, 1);
2280         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2281         ocfs2_journal_dirty(handle, orphan_dir_bh);
2282
2283         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2284                                    namelen, inode,
2285                                    OCFS2_I(inode)->ip_blkno,
2286                                    orphan_dir_bh, lookup);
2287         if (status < 0) {
2288                 mlog_errno(status);
2289                 goto rollback;
2290         }
2291
2292         if (dio) {
2293                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2294                  * slot.
2295                  */
2296                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2297                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2298         } else {
2299                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2300                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2301
2302                 /* Record which orphan dir our inode now resides
2303                  * in. delete_inode will use this to determine which orphan
2304                  * dir to lock. */
2305                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2306         }
2307
2308         ocfs2_journal_dirty(handle, fe_bh);
2309
2310         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2311                                    osb->slot_num);
2312
2313 rollback:
2314         if (status < 0) {
2315                 if (S_ISDIR(inode->i_mode))
2316                         ocfs2_add_links_count(orphan_fe, -1);
2317                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2318         }
2319
2320 leave:
2321         brelse(orphan_dir_bh);
2322
2323         return status;
2324 }
2325
2326 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2327 int ocfs2_orphan_del(struct ocfs2_super *osb,
2328                      handle_t *handle,
2329                      struct inode *orphan_dir_inode,
2330                      struct inode *inode,
2331                      struct buffer_head *orphan_dir_bh,
2332                      bool dio)
2333 {
2334         char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2335         struct ocfs2_dinode *orphan_fe;
2336         int status = 0;
2337         struct ocfs2_dir_lookup_result lookup = { NULL, };
2338
2339         if (dio) {
2340                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2341                                 OCFS2_DIO_ORPHAN_PREFIX);
2342                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2343                         status = -EINVAL;
2344                         mlog_errno(status);
2345                         return status;
2346                 }
2347
2348                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2349                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2350         } else
2351                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2352         if (status < 0) {
2353                 mlog_errno(status);
2354                 goto leave;
2355         }
2356
2357         trace_ocfs2_orphan_del(
2358              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2359              name, strlen(name));
2360
2361         status = ocfs2_journal_access_di(handle,
2362                                          INODE_CACHE(orphan_dir_inode),
2363                                          orphan_dir_bh,
2364                                          OCFS2_JOURNAL_ACCESS_WRITE);
2365         if (status < 0) {
2366                 mlog_errno(status);
2367                 goto leave;
2368         }
2369
2370         /* find it's spot in the orphan directory */
2371         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2372                                   &lookup);
2373         if (status) {
2374                 mlog_errno(status);
2375                 goto leave;
2376         }
2377
2378         /* remove it from the orphan directory */
2379         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2380         if (status < 0) {
2381                 mlog_errno(status);
2382                 goto leave;
2383         }
2384
2385         /* do the i_nlink dance! :) */
2386         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2387         if (S_ISDIR(inode->i_mode))
2388                 ocfs2_add_links_count(orphan_fe, -1);
2389         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2390         ocfs2_journal_dirty(handle, orphan_dir_bh);
2391
2392 leave:
2393         ocfs2_free_dir_lookup_result(&lookup);
2394
2395         if (status)
2396                 mlog_errno(status);
2397         return status;
2398 }
2399
2400 /**
2401  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2402  * allocated file. This is different from the typical 'add to orphan dir'
2403  * operation in that the inode does not yet exist. This is a problem because
2404  * the orphan dir stringifies the inode block number to come up with it's
2405  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2406  * problem. This function works around it by calling deeper into the orphan
2407  * and suballoc code than other callers. Use this only by necessity.
2408  * @dir: The directory which this inode will ultimately wind up under - not the
2409  * orphan dir!
2410  * @dir_bh: buffer_head the @dir inode block
2411  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2412  * with the string to be used for orphan dirent. Pass back to the orphan dir
2413  * code.
2414  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2415  * dir code.
2416  * @ret_di_blkno: block number where the new inode will be allocated.
2417  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2418  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2419  *
2420  * Returns zero on success and the ret_orphan_dir, name and lookup
2421  * fields will be populated.
2422  *
2423  * Returns non-zero on failure.
2424  */
2425 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2426                                         struct buffer_head *dir_bh,
2427                                         char *orphan_name,
2428                                         struct inode **ret_orphan_dir,
2429                                         u64 *ret_di_blkno,
2430                                         struct ocfs2_dir_lookup_result *orphan_insert,
2431                                         struct ocfs2_alloc_context **ret_inode_ac)
2432 {
2433         int ret;
2434         u64 di_blkno;
2435         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2436         struct inode *orphan_dir = NULL;
2437         struct buffer_head *orphan_dir_bh = NULL;
2438         struct ocfs2_alloc_context *inode_ac = NULL;
2439
2440         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2441         if (ret < 0) {
2442                 mlog_errno(ret);
2443                 return ret;
2444         }
2445
2446         /* reserve an inode spot */
2447         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2448         if (ret < 0) {
2449                 if (ret != -ENOSPC)
2450                         mlog_errno(ret);
2451                 goto out;
2452         }
2453
2454         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2455                                        &di_blkno);
2456         if (ret) {
2457                 mlog_errno(ret);
2458                 goto out;
2459         }
2460
2461         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2462                                          di_blkno, orphan_name, orphan_insert,
2463                                          false);
2464         if (ret < 0) {
2465                 mlog_errno(ret);
2466                 goto out;
2467         }
2468
2469 out:
2470         if (ret == 0) {
2471                 *ret_orphan_dir = orphan_dir;
2472                 *ret_di_blkno = di_blkno;
2473                 *ret_inode_ac = inode_ac;
2474                 /*
2475                  * orphan_name and orphan_insert are already up to
2476                  * date via prepare_orphan_dir
2477                  */
2478         } else {
2479                 /* Unroll reserve_new_inode* */
2480                 if (inode_ac)
2481                         ocfs2_free_alloc_context(inode_ac);
2482
2483                 /* Unroll orphan dir locking */
2484                 inode_unlock(orphan_dir);
2485                 ocfs2_inode_unlock(orphan_dir, 1);
2486                 iput(orphan_dir);
2487         }
2488
2489         brelse(orphan_dir_bh);
2490
2491         return ret;
2492 }
2493
2494 int ocfs2_create_inode_in_orphan(struct inode *dir,
2495                                  int mode,
2496                                  struct inode **new_inode)
2497 {
2498         int status, did_quota_inode = 0;
2499         struct inode *inode = NULL;
2500         struct inode *orphan_dir = NULL;
2501         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2502         struct ocfs2_dinode *di = NULL;
2503         handle_t *handle = NULL;
2504         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2505         struct buffer_head *parent_di_bh = NULL;
2506         struct buffer_head *new_di_bh = NULL;
2507         struct ocfs2_alloc_context *inode_ac = NULL;
2508         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2509         u64 uninitialized_var(di_blkno), suballoc_loc;
2510         u16 suballoc_bit;
2511
2512         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2513         if (status < 0) {
2514                 if (status != -ENOENT)
2515                         mlog_errno(status);
2516                 return status;
2517         }
2518
2519         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2520                                               orphan_name, &orphan_dir,
2521                                               &di_blkno, &orphan_insert, &inode_ac);
2522         if (status < 0) {
2523                 if (status != -ENOSPC)
2524                         mlog_errno(status);
2525                 goto leave;
2526         }
2527
2528         inode = ocfs2_get_init_inode(dir, mode);
2529         if (IS_ERR(inode)) {
2530                 status = PTR_ERR(inode);
2531                 inode = NULL;
2532                 mlog_errno(status);
2533                 goto leave;
2534         }
2535
2536         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2537         if (IS_ERR(handle)) {
2538                 status = PTR_ERR(handle);
2539                 handle = NULL;
2540                 mlog_errno(status);
2541                 goto leave;
2542         }
2543
2544         status = dquot_alloc_inode(inode);
2545         if (status)
2546                 goto leave;
2547         did_quota_inode = 1;
2548
2549         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2550                                               &suballoc_loc,
2551                                               &suballoc_bit, di_blkno);
2552         if (status < 0) {
2553                 mlog_errno(status);
2554                 goto leave;
2555         }
2556
2557         clear_nlink(inode);
2558         /* do the real work now. */
2559         status = __ocfs2_mknod_locked(dir, inode,
2560                                       0, &new_di_bh, parent_di_bh, handle,
2561                                       inode_ac, di_blkno, suballoc_loc,
2562                                       suballoc_bit);
2563         if (status < 0) {
2564                 mlog_errno(status);
2565                 goto leave;
2566         }
2567
2568         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2569         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2570                                   &orphan_insert, orphan_dir, false);
2571         if (status < 0) {
2572                 mlog_errno(status);
2573                 goto leave;
2574         }
2575
2576         /* get open lock so that only nodes can't remove it from orphan dir. */
2577         status = ocfs2_open_lock(inode);
2578         if (status < 0)
2579                 mlog_errno(status);
2580
2581         insert_inode_hash(inode);
2582 leave:
2583         if (status < 0 && did_quota_inode)
2584                 dquot_free_inode(inode);
2585         if (handle)
2586                 ocfs2_commit_trans(osb, handle);
2587
2588         if (orphan_dir) {
2589                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2590                 ocfs2_inode_unlock(orphan_dir, 1);
2591                 inode_unlock(orphan_dir);
2592                 iput(orphan_dir);
2593         }
2594
2595         if ((status < 0) && inode) {
2596                 clear_nlink(inode);
2597                 iput(inode);
2598         }
2599
2600         if (inode_ac)
2601                 ocfs2_free_alloc_context(inode_ac);
2602
2603         brelse(new_di_bh);
2604
2605         if (!status)
2606                 *new_inode = inode;
2607
2608         ocfs2_free_dir_lookup_result(&orphan_insert);
2609
2610         ocfs2_inode_unlock(dir, 1);
2611         brelse(parent_di_bh);
2612         return status;
2613 }
2614
2615 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2616         struct inode *inode)
2617 {
2618         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2619         struct inode *orphan_dir_inode = NULL;
2620         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2621         struct buffer_head *di_bh = NULL;
2622         int status = 0;
2623         handle_t *handle = NULL;
2624         struct ocfs2_dinode *di = NULL;
2625
2626         status = ocfs2_inode_lock(inode, &di_bh, 1);
2627         if (status < 0) {
2628                 mlog_errno(status);
2629                 goto bail;
2630         }
2631
2632         di = (struct ocfs2_dinode *) di_bh->b_data;
2633         /*
2634          * Another append dio crashed?
2635          * If so, manually recover it first.
2636          */
2637         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2638                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2639                 if (status < 0) {
2640                         if (status != -ENOSPC)
2641                                 mlog_errno(status);
2642                         goto bail_unlock_inode;
2643                 }
2644
2645                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2646                 if (status < 0) {
2647                         mlog_errno(status);
2648                         goto bail_unlock_inode;
2649                 }
2650         }
2651
2652         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2653                         OCFS2_I(inode)->ip_blkno,
2654                         orphan_name,
2655                         &orphan_insert,
2656                         true);
2657         if (status < 0) {
2658                 mlog_errno(status);
2659                 goto bail_unlock_inode;
2660         }
2661
2662         handle = ocfs2_start_trans(osb,
2663                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2664         if (IS_ERR(handle)) {
2665                 status = PTR_ERR(handle);
2666                 goto bail_unlock_orphan;
2667         }
2668
2669         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2670                         &orphan_insert, orphan_dir_inode, true);
2671         if (status)
2672                 mlog_errno(status);
2673
2674         ocfs2_commit_trans(osb, handle);
2675
2676 bail_unlock_orphan:
2677         ocfs2_inode_unlock(orphan_dir_inode, 1);
2678         inode_unlock(orphan_dir_inode);
2679         iput(orphan_dir_inode);
2680
2681         ocfs2_free_dir_lookup_result(&orphan_insert);
2682
2683 bail_unlock_inode:
2684         ocfs2_inode_unlock(inode, 1);
2685         brelse(di_bh);
2686
2687 bail:
2688         return status;
2689 }
2690
2691 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2692                 struct inode *inode, struct buffer_head *di_bh,
2693                 int update_isize, loff_t end)
2694 {
2695         struct inode *orphan_dir_inode = NULL;
2696         struct buffer_head *orphan_dir_bh = NULL;
2697         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2698         handle_t *handle = NULL;
2699         int status = 0;
2700
2701         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2702                         ORPHAN_DIR_SYSTEM_INODE,
2703                         le16_to_cpu(di->i_dio_orphaned_slot));
2704         if (!orphan_dir_inode) {
2705                 status = -ENOENT;
2706                 mlog_errno(status);
2707                 goto bail;
2708         }
2709
2710         inode_lock(orphan_dir_inode);
2711         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2712         if (status < 0) {
2713                 inode_unlock(orphan_dir_inode);
2714                 iput(orphan_dir_inode);
2715                 mlog_errno(status);
2716                 goto bail;
2717         }
2718
2719         handle = ocfs2_start_trans(osb,
2720                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2721         if (IS_ERR(handle)) {
2722                 status = PTR_ERR(handle);
2723                 goto bail_unlock_orphan;
2724         }
2725
2726         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2727
2728         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2729                                 inode, orphan_dir_bh, true);
2730         if (status < 0) {
2731                 mlog_errno(status);
2732                 goto bail_commit;
2733         }
2734
2735         status = ocfs2_journal_access_di(handle,
2736                         INODE_CACHE(inode),
2737                         di_bh,
2738                         OCFS2_JOURNAL_ACCESS_WRITE);
2739         if (status < 0) {
2740                 mlog_errno(status);
2741                 goto bail_commit;
2742         }
2743
2744         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2745         di->i_dio_orphaned_slot = 0;
2746
2747         if (update_isize) {
2748                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2749                 if (status)
2750                         mlog_errno(status);
2751         } else
2752                 ocfs2_journal_dirty(handle, di_bh);
2753
2754 bail_commit:
2755         ocfs2_commit_trans(osb, handle);
2756
2757 bail_unlock_orphan:
2758         ocfs2_inode_unlock(orphan_dir_inode, 1);
2759         inode_unlock(orphan_dir_inode);
2760         brelse(orphan_dir_bh);
2761         iput(orphan_dir_inode);
2762
2763 bail:
2764         return status;
2765 }
2766
2767 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2768                                    struct inode *inode,
2769                                    struct dentry *dentry)
2770 {
2771         int status = 0;
2772         struct buffer_head *parent_di_bh = NULL;
2773         handle_t *handle = NULL;
2774         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2775         struct ocfs2_dinode *dir_di, *di;
2776         struct inode *orphan_dir_inode = NULL;
2777         struct buffer_head *orphan_dir_bh = NULL;
2778         struct buffer_head *di_bh = NULL;
2779         struct ocfs2_dir_lookup_result lookup = { NULL, };
2780
2781         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2782                                 dentry->d_name.len, dentry->d_name.name,
2783                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2784                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2785
2786         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2787         if (status < 0) {
2788                 if (status != -ENOENT)
2789                         mlog_errno(status);
2790                 return status;
2791         }
2792
2793         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2794         if (!dir_di->i_links_count) {
2795                 /* can't make a file in a deleted directory. */
2796                 status = -ENOENT;
2797                 goto leave;
2798         }
2799
2800         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2801                                            dentry->d_name.len);
2802         if (status)
2803                 goto leave;
2804
2805         /* get a spot inside the dir. */
2806         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2807                                               dentry->d_name.name,
2808                                               dentry->d_name.len, &lookup);
2809         if (status < 0) {
2810                 mlog_errno(status);
2811                 goto leave;
2812         }
2813
2814         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2815                                                        ORPHAN_DIR_SYSTEM_INODE,
2816                                                        osb->slot_num);
2817         if (!orphan_dir_inode) {
2818                 status = -ENOENT;
2819                 mlog_errno(status);
2820                 goto leave;
2821         }
2822
2823         inode_lock(orphan_dir_inode);
2824
2825         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2826         if (status < 0) {
2827                 mlog_errno(status);
2828                 inode_unlock(orphan_dir_inode);
2829                 iput(orphan_dir_inode);
2830                 goto leave;
2831         }
2832
2833         status = ocfs2_read_inode_block(inode, &di_bh);
2834         if (status < 0) {
2835                 mlog_errno(status);
2836                 goto orphan_unlock;
2837         }
2838
2839         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2840         if (IS_ERR(handle)) {
2841                 status = PTR_ERR(handle);
2842                 handle = NULL;
2843                 mlog_errno(status);
2844                 goto orphan_unlock;
2845         }
2846
2847         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2848                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2849         if (status < 0) {
2850                 mlog_errno(status);
2851                 goto out_commit;
2852         }
2853
2854         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2855                                   orphan_dir_bh, false);
2856         if (status < 0) {
2857                 mlog_errno(status);
2858                 goto out_commit;
2859         }
2860
2861         di = (struct ocfs2_dinode *)di_bh->b_data;
2862         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2863         di->i_orphaned_slot = 0;
2864         set_nlink(inode, 1);
2865         ocfs2_set_links_count(di, inode->i_nlink);
2866         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2867         ocfs2_journal_dirty(handle, di_bh);
2868
2869         status = ocfs2_add_entry(handle, dentry, inode,
2870                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2871                                  &lookup);
2872         if (status < 0) {
2873                 mlog_errno(status);
2874                 goto out_commit;
2875         }
2876
2877         status = ocfs2_dentry_attach_lock(dentry, inode,
2878                                           OCFS2_I(dir)->ip_blkno);
2879         if (status) {
2880                 mlog_errno(status);
2881                 goto out_commit;
2882         }
2883
2884         d_instantiate(dentry, inode);
2885         status = 0;
2886 out_commit:
2887         ocfs2_commit_trans(osb, handle);
2888 orphan_unlock:
2889         ocfs2_inode_unlock(orphan_dir_inode, 1);
2890         inode_unlock(orphan_dir_inode);
2891         iput(orphan_dir_inode);
2892 leave:
2893
2894         ocfs2_inode_unlock(dir, 1);
2895
2896         brelse(di_bh);
2897         brelse(parent_di_bh);
2898         brelse(orphan_dir_bh);
2899
2900         ocfs2_free_dir_lookup_result(&lookup);
2901
2902         if (status)
2903                 mlog_errno(status);
2904
2905         return status;
2906 }
2907
2908 const struct inode_operations ocfs2_dir_iops = {
2909         .create         = ocfs2_create,
2910         .lookup         = ocfs2_lookup,
2911         .link           = ocfs2_link,
2912         .unlink         = ocfs2_unlink,
2913         .rmdir          = ocfs2_unlink,
2914         .symlink        = ocfs2_symlink,
2915         .mkdir          = ocfs2_mkdir,
2916         .mknod          = ocfs2_mknod,
2917         .rename         = ocfs2_rename,
2918         .setattr        = ocfs2_setattr,
2919         .getattr        = ocfs2_getattr,
2920         .permission     = ocfs2_permission,
2921         .listxattr      = ocfs2_listxattr,
2922         .fiemap         = ocfs2_fiemap,
2923         .get_acl        = ocfs2_iop_get_acl,
2924         .set_acl        = ocfs2_iop_set_acl,
2925 };