fs/ocfs2/namei.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * namei.c
   4  *
   5  * Create and rename file, directory, symlinks
   6  *
   7  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   8  *
   9  *  Portions of this code from linux/fs/ext3/dir.c
  10  *
  11  *  Copyright (C) 1992, 1993, 1994, 1995
  12  *  Remy Card (card@masi.ibp.fr)
  13  *  Laboratoire MASI - Institut Blaise pascal
  14  *  Universite Pierre et Marie Curie (Paris VI)
  15  *
  16  *   from
  17  *
  18  *   linux/fs/minix/dir.c
  19  *
  20  *   Copyright (C) 1991, 1992 Linux Torvalds
  21  */
  22
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/quotaops.h>
  28 #include <linux/iversion.h>
  29
  30 #include <cluster/masklog.h>
  31
  32 #include "ocfs2.h"
  33
  34 #include "alloc.h"
  35 #include "dcache.h"
  36 #include "dir.h"
  37 #include "dlmglue.h"
  38 #include "extent_map.h"
  39 #include "file.h"
  40 #include "inode.h"
  41 #include "journal.h"
  42 #include "namei.h"
  43 #include "suballoc.h"
  44 #include "super.h"
  45 #include "symlink.h"
  46 #include "sysfile.h"
  47 #include "uptodate.h"
  48 #include "xattr.h"
  49 #include "acl.h"
  50 #include "ocfs2_trace.h"
  51 #include "ioctl.h"
  52
  53 #include "buffer_head_io.h"
  54
  55 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  56                               struct inode *dir,
  57                               struct inode *inode,
  58                               dev_t dev,
  59                               struct buffer_head **new_fe_bh,
  60                               struct buffer_head *parent_fe_bh,
  61                               handle_t *handle,
  62                               struct ocfs2_alloc_context *inode_ac);
  63
  64 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  65                                     struct inode **ret_orphan_dir,
  66                                     u64 blkno,
  67                                     char *name,
  68                                     struct ocfs2_dir_lookup_result *lookup,
  69                                     bool dio);
  70
  71 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  72                             handle_t *handle,
  73                             struct inode *inode,
  74                             struct buffer_head *fe_bh,
  75                             char *name,
  76                             struct ocfs2_dir_lookup_result *lookup,
  77                             struct inode *orphan_dir_inode,
  78                             bool dio);
  79
  80 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  81                                      handle_t *handle,
  82                                      struct inode *inode,
  83                                      const char *symname);
  84
  85 static int ocfs2_double_lock(struct ocfs2_super *osb,
  86                              struct buffer_head **bh1,
  87                              struct inode *inode1,
  88                              struct buffer_head **bh2,
  89                              struct inode *inode2,
  90                              int rename);
  91
  92 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
  93 /* An orphan dir name is an 8 byte value, printed as a hex string */
  94 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  95
  96 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
  97                                    unsigned int flags)
  98 {
  99         int status;
 100         u64 blkno;
 101         struct inode *inode = NULL;
 102         struct dentry *ret;
 103         struct ocfs2_inode_info *oi;
 104
 105         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 106                            dentry->d_name.name,
 107                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 108
 109         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 110                 ret = ERR_PTR(-ENAMETOOLONG);
 111                 goto bail;
 112         }
 113
 114         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 115         if (status < 0) {
 116                 if (status != -ENOENT)
 117                         mlog_errno(status);
 118                 ret = ERR_PTR(status);
 119                 goto bail;
 120         }
 121
 122         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 123                                             dentry->d_name.len, &blkno);
 124         if (status < 0)
 125                 goto bail_add;
 126
 127         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 128         if (IS_ERR(inode)) {
 129                 ret = ERR_PTR(-EACCES);
 130                 goto bail_unlock;
 131         }
 132
 133         oi = OCFS2_I(inode);
 134         /* Clear any orphaned state... If we were able to look up the
 135          * inode from a directory, it certainly can't be orphaned. We
 136          * might have the bad state from a node which intended to
 137          * orphan this inode but crashed before it could commit the
 138          * unlink. */
 139         spin_lock(&oi->ip_lock);
 140         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 141         spin_unlock(&oi->ip_lock);
 142
 143 bail_add:
 144         ret = d_splice_alias(inode, dentry);
 145
 146         if (inode) {
 147                 /*
 148                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 149                  * dentry, it will d_move() it on top of ourse. The
 150                  * return value will indicate this however, so in
 151                  * those cases, we switch them around for the locking
 152                  * code.
 153                  *
 154                  * NOTE: This dentry already has ->d_op set from
 155                  * ocfs2_get_parent() and ocfs2_get_dentry()
 156                  */
 157                 if (!IS_ERR_OR_NULL(ret))
 158                         dentry = ret;
 159
 160                 status = ocfs2_dentry_attach_lock(dentry, inode,
 161                                                   OCFS2_I(dir)->ip_blkno);
 162                 if (status) {
 163                         mlog_errno(status);
 164                         ret = ERR_PTR(status);
 165                         goto bail_unlock;
 166                 }
 167         } else
 168                 ocfs2_dentry_attach_gen(dentry);
 169
 170 bail_unlock:
 171         /* Don't drop the cluster lock until *after* the d_add --
 172          * unlink on another node will message us to remove that
 173          * dentry under this lock so otherwise we can race this with
 174          * the downconvert thread and have a stale dentry. */
 175         ocfs2_inode_unlock(dir, 0);
 176
 177 bail:
 178
 179         trace_ocfs2_lookup_ret(ret);
 180
 181         return ret;
 182 }
 183
 184 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 185 {
 186         struct inode *inode;
 187         int status;
 188
 189         inode = new_inode(dir->i_sb);
 190         if (!inode) {
 191                 mlog(ML_ERROR, "new_inode failed!\n");
 192                 return ERR_PTR(-ENOMEM);
 193         }
 194
 195         /* populate as many fields early on as possible - many of
 196          * these are used by the support functions here and in
 197          * callers. */
 198         if (S_ISDIR(mode))
 199                 set_nlink(inode, 2);
 200         mode = mode_strip_sgid(&nop_mnt_idmap, dir, mode);
 201         inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
 202         status = dquot_initialize(inode);
 203         if (status)
 204                 return ERR_PTR(status);
 205
 206         return inode;
 207 }
 208
 209 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 210                 struct dentry *dentry, struct inode *inode)
 211 {
 212         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 213
 214         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 215         ocfs2_lock_res_free(&dl->dl_lockres);
 216         BUG_ON(dl->dl_count != 1);
 217         spin_lock(&dentry_attach_lock);
 218         dentry->d_fsdata = NULL;
 219         spin_unlock(&dentry_attach_lock);
 220         kfree(dl);
 221         iput(inode);
 222 }
 223
 224 static int ocfs2_mknod(struct mnt_idmap *idmap,
 225                        struct inode *dir,
 226                        struct dentry *dentry,
 227                        umode_t mode,
 228                        dev_t dev)
 229 {
 230         int status = 0;
 231         struct buffer_head *parent_fe_bh = NULL;
 232         handle_t *handle = NULL;
 233         struct ocfs2_super *osb;
 234         struct ocfs2_dinode *dirfe;
 235         struct ocfs2_dinode *fe = NULL;
 236         struct buffer_head *new_fe_bh = NULL;
 237         struct inode *inode = NULL;
 238         struct ocfs2_alloc_context *inode_ac = NULL;
 239         struct ocfs2_alloc_context *data_ac = NULL;
 240         struct ocfs2_alloc_context *meta_ac = NULL;
 241         int want_clusters = 0;
 242         int want_meta = 0;
 243         int xattr_credits = 0;
 244         struct ocfs2_security_xattr_info si = {
 245                 .name = NULL,
 246                 .enable = 1,
 247         };
 248         int did_quota_inode = 0;
 249         struct ocfs2_dir_lookup_result lookup = { NULL, };
 250         sigset_t oldset;
 251         int did_block_signals = 0;
 252         struct ocfs2_dentry_lock *dl = NULL;
 253
 254         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 255                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 256                           (unsigned long)dev, mode);
 257
 258         status = dquot_initialize(dir);
 259         if (status) {
 260                 mlog_errno(status);
 261                 return status;
 262         }
 263
 264         /* get our super block */
 265         osb = OCFS2_SB(dir->i_sb);
 266
 267         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 268         if (status < 0) {
 269                 if (status != -ENOENT)
 270                         mlog_errno(status);
 271                 return status;
 272         }
 273
 274         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 275                 status = -EMLINK;
 276                 goto leave;
 277         }
 278
 279         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 280         if (!ocfs2_read_links_count(dirfe)) {
 281                 /* can't make a file in a deleted directory. */
 282                 status = -ENOENT;
 283                 goto leave;
 284         }
 285
 286         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 287                                            dentry->d_name.len);
 288         if (status)
 289                 goto leave;
 290
 291         /* get a spot inside the dir. */
 292         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 293                                               dentry->d_name.name,
 294                                               dentry->d_name.len, &lookup);
 295         if (status < 0) {
 296                 mlog_errno(status);
 297                 goto leave;
 298         }
 299
 300         /* reserve an inode spot */
 301         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 302         if (status < 0) {
 303                 if (status != -ENOSPC)
 304                         mlog_errno(status);
 305                 goto leave;
 306         }
 307
 308         inode = ocfs2_get_init_inode(dir, mode);
 309         if (IS_ERR(inode)) {
 310                 status = PTR_ERR(inode);
 311                 inode = NULL;
 312                 mlog_errno(status);
 313                 goto leave;
 314         }
 315
 316         /* get security xattr */
 317         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 318         if (status) {
 319                 if (status == -EOPNOTSUPP)
 320                         si.enable = 0;
 321                 else {
 322                         mlog_errno(status);
 323                         goto leave;
 324                 }
 325         }
 326
 327         /* calculate meta data/clusters for setting security and acl xattr */
 328         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 329                                        &si, &want_clusters,
 330                                        &xattr_credits, &want_meta);
 331         if (status < 0) {
 332                 mlog_errno(status);
 333                 goto leave;
 334         }
 335
 336         /* Reserve a cluster if creating an extent based directory. */
 337         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 338                 want_clusters += 1;
 339
 340                 /* Dir indexing requires extra space as well */
 341                 if (ocfs2_supports_indexed_dirs(osb))
 342                         want_meta++;
 343         }
 344
 345         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 346         if (status < 0) {
 347                 if (status != -ENOSPC)
 348                         mlog_errno(status);
 349                 goto leave;
 350         }
 351
 352         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 353         if (status < 0) {
 354                 if (status != -ENOSPC)
 355                         mlog_errno(status);
 356                 goto leave;
 357         }
 358
 359         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 360                                                             S_ISDIR(mode),
 361                                                             xattr_credits));
 362         if (IS_ERR(handle)) {
 363                 status = PTR_ERR(handle);
 364                 handle = NULL;
 365                 mlog_errno(status);
 366                 goto leave;
 367         }
 368
 369         /* Starting to change things, restart is no longer possible. */
 370         ocfs2_block_signals(&oldset);
 371         did_block_signals = 1;
 372
 373         status = dquot_alloc_inode(inode);
 374         if (status)
 375                 goto leave;
 376         did_quota_inode = 1;
 377
 378         /* do the real work now. */
 379         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 380                                     &new_fe_bh, parent_fe_bh, handle,
 381                                     inode_ac);
 382         if (status < 0) {
 383                 mlog_errno(status);
 384                 goto leave;
 385         }
 386
 387         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
 388         if (S_ISDIR(mode)) {
 389                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 390                                             new_fe_bh, data_ac, meta_ac);
 391                 if (status < 0) {
 392                         mlog_errno(status);
 393                         goto leave;
 394                 }
 395
 396                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 397                                                  parent_fe_bh,
 398                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 399                 if (status < 0) {
 400                         mlog_errno(status);
 401                         goto leave;
 402                 }
 403                 ocfs2_add_links_count(dirfe, 1);
 404                 ocfs2_journal_dirty(handle, parent_fe_bh);
 405                 inc_nlink(dir);
 406         }
 407
 408         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 409                          meta_ac, data_ac);
 410
 411         if (status < 0) {
 412                 mlog_errno(status);
 413                 goto roll_back;
 414         }
 415
 416         if (si.enable) {
 417                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 418                                                  meta_ac, data_ac);
 419                 if (status < 0) {
 420                         mlog_errno(status);
 421                         goto roll_back;
 422                 }
 423         }
 424
 425         /*
 426          * Do this before adding the entry to the directory. We add
 427          * also set d_op after success so that ->d_iput() will cleanup
 428          * the dentry lock even if ocfs2_add_entry() fails below.
 429          */
 430         status = ocfs2_dentry_attach_lock(dentry, inode,
 431                                           OCFS2_I(dir)->ip_blkno);
 432         if (status) {
 433                 mlog_errno(status);
 434                 goto roll_back;
 435         }
 436
 437         dl = dentry->d_fsdata;
 438
 439         status = ocfs2_add_entry(handle, dentry, inode,
 440                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 441                                  &lookup);
 442         if (status < 0) {
 443                 mlog_errno(status);
 444                 goto roll_back;
 445         }
 446
 447         insert_inode_hash(inode);
 448         d_instantiate(dentry, inode);
 449         status = 0;
 450
 451 roll_back:
 452         if (status < 0 && S_ISDIR(mode)) {
 453                 ocfs2_add_links_count(dirfe, -1);
 454                 drop_nlink(dir);
 455         }
 456
 457 leave:
 458         if (status < 0 && did_quota_inode)
 459                 dquot_free_inode(inode);
 460         if (handle) {
 461                 if (status < 0 && fe)
 462                         ocfs2_set_links_count(fe, 0);
 463                 ocfs2_commit_trans(osb, handle);
 464         }
 465
 466         ocfs2_inode_unlock(dir, 1);
 467         if (did_block_signals)
 468                 ocfs2_unblock_signals(&oldset);
 469
 470         brelse(new_fe_bh);
 471         brelse(parent_fe_bh);
 472         kfree(si.value);
 473
 474         ocfs2_free_dir_lookup_result(&lookup);
 475
 476         if (inode_ac)
 477                 ocfs2_free_alloc_context(inode_ac);
 478
 479         if (data_ac)
 480                 ocfs2_free_alloc_context(data_ac);
 481
 482         if (meta_ac)
 483                 ocfs2_free_alloc_context(meta_ac);
 484
 485         /*
 486          * We should call iput after the i_rwsem of the bitmap been
 487          * unlocked in ocfs2_free_alloc_context, or the
 488          * ocfs2_delete_inode will mutex_lock again.
 489          */
 490         if ((status < 0) && inode) {
 491                 if (dl)
 492                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 493
 494                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 495                 clear_nlink(inode);
 496                 iput(inode);
 497         }
 498
 499         if (status)
 500                 mlog_errno(status);
 501
 502         return status;
 503 }
 504
 505 static int __ocfs2_mknod_locked(struct inode *dir,
 506                                 struct inode *inode,
 507                                 dev_t dev,
 508                                 struct buffer_head **new_fe_bh,
 509                                 struct buffer_head *parent_fe_bh,
 510                                 handle_t *handle,
 511                                 struct ocfs2_alloc_context *inode_ac,
 512                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 513 {
 514         int status = 0;
 515         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 516         struct ocfs2_dinode *fe = NULL;
 517         struct ocfs2_extent_list *fel;
 518         u16 feat;
 519         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 520         struct timespec64 ts;
 521
 522         *new_fe_bh = NULL;
 523
 524         /* populate as many fields early on as possible - many of
 525          * these are used by the support functions here and in
 526          * callers. */
 527         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 528         oi->ip_blkno = fe_blkno;
 529         spin_lock(&osb->osb_lock);
 530         inode->i_generation = osb->s_next_generation++;
 531         spin_unlock(&osb->osb_lock);
 532
 533         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 534         if (!*new_fe_bh) {
 535                 status = -ENOMEM;
 536                 mlog_errno(status);
 537                 goto leave;
 538         }
 539         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 540
 541         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 542                                          *new_fe_bh,
 543                                          OCFS2_JOURNAL_ACCESS_CREATE);
 544         if (status < 0) {
 545                 mlog_errno(status);
 546                 goto leave;
 547         }
 548
 549         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 550         memset(fe, 0, osb->sb->s_blocksize);
 551
 552         fe->i_generation = cpu_to_le32(inode->i_generation);
 553         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 554         fe->i_blkno = cpu_to_le64(fe_blkno);
 555         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 556         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 557         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 558         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 559         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 560         fe->i_mode = cpu_to_le16(inode->i_mode);
 561         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 562                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 563
 564         ocfs2_set_links_count(fe, inode->i_nlink);
 565
 566         fe->i_last_eb_blk = 0;
 567         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 568         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 569         ktime_get_real_ts64(&ts);
 570         fe->i_atime = fe->i_ctime = fe->i_mtime =
 571                 cpu_to_le64(ts.tv_sec);
 572         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 573                 cpu_to_le32(ts.tv_nsec);
 574         fe->i_dtime = 0;
 575
 576         /*
 577          * If supported, directories start with inline data. If inline
 578          * isn't supported, but indexing is, we start them as indexed.
 579          */
 580         feat = le16_to_cpu(fe->i_dyn_features);
 581         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 582                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 583
 584                 fe->id2.i_data.id_count = cpu_to_le16(
 585                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 586         } else {
 587                 fel = &fe->id2.i_list;
 588                 fel->l_tree_depth = 0;
 589                 fel->l_next_free_rec = 0;
 590                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 591         }
 592
 593         ocfs2_journal_dirty(handle, *new_fe_bh);
 594
 595         ocfs2_populate_inode(inode, fe, 1);
 596         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 597         if (!ocfs2_mount_local(osb)) {
 598                 status = ocfs2_create_new_inode_locks(inode);
 599                 if (status < 0)
 600                         mlog_errno(status);
 601         }
 602
 603         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 604
 605 leave:
 606         if (status < 0) {
 607                 if (*new_fe_bh) {
 608                         brelse(*new_fe_bh);
 609                         *new_fe_bh = NULL;
 610                 }
 611         }
 612
 613         if (status)
 614                 mlog_errno(status);
 615         return status;
 616 }
 617
 618 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 619                               struct inode *dir,
 620                               struct inode *inode,
 621                               dev_t dev,
 622                               struct buffer_head **new_fe_bh,
 623                               struct buffer_head *parent_fe_bh,
 624                               handle_t *handle,
 625                               struct ocfs2_alloc_context *inode_ac)
 626 {
 627         int status = 0;
 628         u64 suballoc_loc, fe_blkno = 0;
 629         u16 suballoc_bit;
 630
 631         *new_fe_bh = NULL;
 632
 633         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 634                                        inode_ac, &suballoc_loc,
 635                                        &suballoc_bit, &fe_blkno);
 636         if (status < 0) {
 637                 mlog_errno(status);
 638                 return status;
 639         }
 640
 641         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 642                                     parent_fe_bh, handle, inode_ac,
 643                                     fe_blkno, suballoc_loc, suballoc_bit);
 644 }
 645
 646 static int ocfs2_mkdir(struct mnt_idmap *idmap,
 647                        struct inode *dir,
 648                        struct dentry *dentry,
 649                        umode_t mode)
 650 {
 651         int ret;
 652
 653         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 654                           OCFS2_I(dir)->ip_blkno, mode);
 655         ret = ocfs2_mknod(&nop_mnt_idmap, dir, dentry, mode | S_IFDIR, 0);
 656         if (ret)
 657                 mlog_errno(ret);
 658
 659         return ret;
 660 }
 661
 662 static int ocfs2_create(struct mnt_idmap *idmap,
 663                         struct inode *dir,
 664                         struct dentry *dentry,
 665                         umode_t mode,
 666                         bool excl)
 667 {
 668         int ret;
 669
 670         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 671                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 672         ret = ocfs2_mknod(&nop_mnt_idmap, dir, dentry, mode | S_IFREG, 0);
 673         if (ret)
 674                 mlog_errno(ret);
 675
 676         return ret;
 677 }
 678
 679 static int ocfs2_link(struct dentry *old_dentry,
 680                       struct inode *dir,
 681                       struct dentry *dentry)
 682 {
 683         handle_t *handle;
 684         struct inode *inode = d_inode(old_dentry);
 685         struct inode *old_dir = d_inode(old_dentry->d_parent);
 686         int err;
 687         struct buffer_head *fe_bh = NULL;
 688         struct buffer_head *old_dir_bh = NULL;
 689         struct buffer_head *parent_fe_bh = NULL;
 690         struct ocfs2_dinode *fe = NULL;
 691         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 692         struct ocfs2_dir_lookup_result lookup = { NULL, };
 693         sigset_t oldset;
 694         u64 old_de_ino;
 695
 696         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 697                          old_dentry->d_name.len, old_dentry->d_name.name,
 698                          dentry->d_name.len, dentry->d_name.name);
 699
 700         if (S_ISDIR(inode->i_mode))
 701                 return -EPERM;
 702
 703         err = dquot_initialize(dir);
 704         if (err) {
 705                 mlog_errno(err);
 706                 return err;
 707         }
 708
 709         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 710                         &parent_fe_bh, dir, 0);
 711         if (err < 0) {
 712                 if (err != -ENOENT)
 713                         mlog_errno(err);
 714                 return err;
 715         }
 716
 717         /* make sure both dirs have bhs
 718          * get an extra ref on old_dir_bh if old==new */
 719         if (!parent_fe_bh) {
 720                 if (old_dir_bh) {
 721                         parent_fe_bh = old_dir_bh;
 722                         get_bh(parent_fe_bh);
 723                 } else {
 724                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 725                         err = -EIO;
 726                         goto out;
 727                 }
 728         }
 729
 730         if (!dir->i_nlink) {
 731                 err = -ENOENT;
 732                 goto out;
 733         }
 734
 735         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 736                         old_dentry->d_name.len, &old_de_ino);
 737         if (err) {
 738                 err = -ENOENT;
 739                 goto out;
 740         }
 741
 742         /*
 743          * Check whether another node removed the source inode while we
 744          * were in the vfs.
 745          */
 746         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 747                 err = -ENOENT;
 748                 goto out;
 749         }
 750
 751         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 752                                         dentry->d_name.len);
 753         if (err)
 754                 goto out;
 755
 756         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 757                                            dentry->d_name.name,
 758                                            dentry->d_name.len, &lookup);
 759         if (err < 0) {
 760                 mlog_errno(err);
 761                 goto out;
 762         }
 763
 764         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 765         if (err < 0) {
 766                 if (err != -ENOENT)
 767                         mlog_errno(err);
 768                 goto out;
 769         }
 770
 771         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 772         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 773                 err = -EMLINK;
 774                 goto out_unlock_inode;
 775         }
 776
 777         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 778         if (IS_ERR(handle)) {
 779                 err = PTR_ERR(handle);
 780                 handle = NULL;
 781                 mlog_errno(err);
 782                 goto out_unlock_inode;
 783         }
 784
 785         /* Starting to change things, restart is no longer possible. */
 786         ocfs2_block_signals(&oldset);
 787
 788         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 789                                       OCFS2_JOURNAL_ACCESS_WRITE);
 790         if (err < 0) {
 791                 mlog_errno(err);
 792                 goto out_commit;
 793         }
 794
 795         inc_nlink(inode);
 796         inode->i_ctime = current_time(inode);
 797         ocfs2_set_links_count(fe, inode->i_nlink);
 798         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 799         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 800         ocfs2_journal_dirty(handle, fe_bh);
 801
 802         err = ocfs2_add_entry(handle, dentry, inode,
 803                               OCFS2_I(inode)->ip_blkno,
 804                               parent_fe_bh, &lookup);
 805         if (err) {
 806                 ocfs2_add_links_count(fe, -1);
 807                 drop_nlink(inode);
 808                 mlog_errno(err);
 809                 goto out_commit;
 810         }
 811
 812         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 813         if (err) {
 814                 mlog_errno(err);
 815                 goto out_commit;
 816         }
 817
 818         ihold(inode);
 819         d_instantiate(dentry, inode);
 820
 821 out_commit:
 822         ocfs2_commit_trans(osb, handle);
 823         ocfs2_unblock_signals(&oldset);
 824 out_unlock_inode:
 825         ocfs2_inode_unlock(inode, 1);
 826
 827 out:
 828         ocfs2_double_unlock(old_dir, dir);
 829
 830         brelse(fe_bh);
 831         brelse(parent_fe_bh);
 832         brelse(old_dir_bh);
 833
 834         ocfs2_free_dir_lookup_result(&lookup);
 835
 836         if (err)
 837                 mlog_errno(err);
 838
 839         return err;
 840 }
 841
 842 /*
 843  * Takes and drops an exclusive lock on the given dentry. This will
 844  * force other nodes to drop it.
 845  */
 846 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 847 {
 848         int ret;
 849
 850         ret = ocfs2_dentry_lock(dentry, 1);
 851         if (ret)
 852                 mlog_errno(ret);
 853         else
 854                 ocfs2_dentry_unlock(dentry, 1);
 855
 856         return ret;
 857 }
 858
 859 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 860 {
 861         if (S_ISDIR(inode->i_mode)) {
 862                 if (inode->i_nlink == 2)
 863                         return 1;
 864                 return 0;
 865         }
 866
 867         if (inode->i_nlink == 1)
 868                 return 1;
 869         return 0;
 870 }
 871
 872 static int ocfs2_unlink(struct inode *dir,
 873                         struct dentry *dentry)
 874 {
 875         int status;
 876         int child_locked = 0;
 877         bool is_unlinkable = false;
 878         struct inode *inode = d_inode(dentry);
 879         struct inode *orphan_dir = NULL;
 880         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 881         u64 blkno;
 882         struct ocfs2_dinode *fe = NULL;
 883         struct buffer_head *fe_bh = NULL;
 884         struct buffer_head *parent_node_bh = NULL;
 885         handle_t *handle = NULL;
 886         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 887         struct ocfs2_dir_lookup_result lookup = { NULL, };
 888         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 889
 890         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 891                            dentry->d_name.name,
 892                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 893                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 894
 895         status = dquot_initialize(dir);
 896         if (status) {
 897                 mlog_errno(status);
 898                 return status;
 899         }
 900
 901         BUG_ON(d_inode(dentry->d_parent) != dir);
 902
 903         if (inode == osb->root_inode)
 904                 return -EPERM;
 905
 906         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 907                                          OI_LS_PARENT);
 908         if (status < 0) {
 909                 if (status != -ENOENT)
 910                         mlog_errno(status);
 911                 return status;
 912         }
 913
 914         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 915                                           dentry->d_name.len, &blkno, dir,
 916                                           &lookup);
 917         if (status < 0) {
 918                 if (status != -ENOENT)
 919                         mlog_errno(status);
 920                 goto leave;
 921         }
 922
 923         if (OCFS2_I(inode)->ip_blkno != blkno) {
 924                 status = -ENOENT;
 925
 926                 trace_ocfs2_unlink_noent(
 927                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 928                                 (unsigned long long)blkno,
 929                                 OCFS2_I(inode)->ip_flags);
 930                 goto leave;
 931         }
 932
 933         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 934         if (status < 0) {
 935                 if (status != -ENOENT)
 936                         mlog_errno(status);
 937                 goto leave;
 938         }
 939         child_locked = 1;
 940
 941         if (S_ISDIR(inode->i_mode)) {
 942                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 943                         status = -ENOTEMPTY;
 944                         goto leave;
 945                 }
 946         }
 947
 948         status = ocfs2_remote_dentry_delete(dentry);
 949         if (status < 0) {
 950                 /* This remote delete should succeed under all normal
 951                  * circumstances. */
 952                 mlog_errno(status);
 953                 goto leave;
 954         }
 955
 956         if (ocfs2_inode_is_unlinkable(inode)) {
 957                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 958                                                   OCFS2_I(inode)->ip_blkno,
 959                                                   orphan_name, &orphan_insert,
 960                                                   false);
 961                 if (status < 0) {
 962                         mlog_errno(status);
 963                         goto leave;
 964                 }
 965                 is_unlinkable = true;
 966         }
 967
 968         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 969         if (IS_ERR(handle)) {
 970                 status = PTR_ERR(handle);
 971                 handle = NULL;
 972                 mlog_errno(status);
 973                 goto leave;
 974         }
 975
 976         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 977                                          OCFS2_JOURNAL_ACCESS_WRITE);
 978         if (status < 0) {
 979                 mlog_errno(status);
 980                 goto leave;
 981         }
 982
 983         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 984
 985         /* delete the name from the parent dir */
 986         status = ocfs2_delete_entry(handle, dir, &lookup);
 987         if (status < 0) {
 988                 mlog_errno(status);
 989                 goto leave;
 990         }
 991
 992         if (S_ISDIR(inode->i_mode))
 993                 drop_nlink(inode);
 994         drop_nlink(inode);
 995         ocfs2_set_links_count(fe, inode->i_nlink);
 996         ocfs2_journal_dirty(handle, fe_bh);
 997
 998         dir->i_ctime = dir->i_mtime = current_time(dir);
 999         if (S_ISDIR(inode->i_mode))
1000                 drop_nlink(dir);
1001
1002         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1003         if (status < 0) {
1004                 mlog_errno(status);
1005                 if (S_ISDIR(inode->i_mode))
1006                         inc_nlink(dir);
1007                 goto leave;
1008         }
1009
1010         if (is_unlinkable) {
1011                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1012                                 orphan_name, &orphan_insert, orphan_dir, false);
1013                 if (status < 0)
1014                         mlog_errno(status);
1015         }
1016
1017 leave:
1018         if (handle)
1019                 ocfs2_commit_trans(osb, handle);
1020
1021         if (orphan_dir) {
1022                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1023                 ocfs2_inode_unlock(orphan_dir, 1);
1024                 inode_unlock(orphan_dir);
1025                 iput(orphan_dir);
1026         }
1027
1028         if (child_locked)
1029                 ocfs2_inode_unlock(inode, 1);
1030
1031         ocfs2_inode_unlock(dir, 1);
1032
1033         brelse(fe_bh);
1034         brelse(parent_node_bh);
1035
1036         ocfs2_free_dir_lookup_result(&orphan_insert);
1037         ocfs2_free_dir_lookup_result(&lookup);
1038
1039         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1040                 mlog_errno(status);
1041
1042         return status;
1043 }
1044
1045 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1046                 u64 src_inode_no, u64 dest_inode_no)
1047 {
1048         int ret = 0, i = 0;
1049         u64 parent_inode_no = 0;
1050         u64 child_inode_no = src_inode_no;
1051         struct inode *child_inode;
1052
1053 #define MAX_LOOKUP_TIMES 32
1054         while (1) {
1055                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1056                 if (IS_ERR(child_inode)) {
1057                         ret = PTR_ERR(child_inode);
1058                         break;
1059                 }
1060
1061                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1062                 if (ret < 0) {
1063                         iput(child_inode);
1064                         if (ret != -ENOENT)
1065                                 mlog_errno(ret);
1066                         break;
1067                 }
1068
1069                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1070                                 &parent_inode_no);
1071                 ocfs2_inode_unlock(child_inode, 0);
1072                 iput(child_inode);
1073                 if (ret < 0) {
1074                         ret = -ENOENT;
1075                         break;
1076                 }
1077
1078                 if (parent_inode_no == dest_inode_no) {
1079                         ret = 1;
1080                         break;
1081                 }
1082
1083                 if (parent_inode_no == osb->root_inode->i_ino) {
1084                         ret = 0;
1085                         break;
1086                 }
1087
1088                 child_inode_no = parent_inode_no;
1089
1090                 if (++i >= MAX_LOOKUP_TIMES) {
1091                         mlog_ratelimited(ML_NOTICE, "max lookup times reached, "
1092                                         "filesystem may have nested directories, "
1093                                         "src inode: %llu, dest inode: %llu.\n",
1094                                         (unsigned long long)src_inode_no,
1095                                         (unsigned long long)dest_inode_no);
1096                         ret = 0;
1097                         break;
1098                 }
1099         }
1100
1101         return ret;
1102 }
1103
1104 /*
1105  * The only place this should be used is rename and link!
1106  * if they have the same id, then the 1st one is the only one locked.
1107  */
1108 static int ocfs2_double_lock(struct ocfs2_super *osb,
1109                              struct buffer_head **bh1,
1110                              struct inode *inode1,
1111                              struct buffer_head **bh2,
1112                              struct inode *inode2,
1113                              int rename)
1114 {
1115         int status;
1116         int inode1_is_ancestor, inode2_is_ancestor;
1117         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1118         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1119
1120         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1121                                 (unsigned long long)oi2->ip_blkno);
1122
1123         if (*bh1)
1124                 *bh1 = NULL;
1125         if (*bh2)
1126                 *bh2 = NULL;
1127
1128         /* we always want to lock the one with the lower lockid first.
1129          * and if they are nested, we lock ancestor first */
1130         if (oi1->ip_blkno != oi2->ip_blkno) {
1131                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1132                                 oi1->ip_blkno);
1133                 if (inode1_is_ancestor < 0) {
1134                         status = inode1_is_ancestor;
1135                         goto bail;
1136                 }
1137
1138                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1139                                 oi2->ip_blkno);
1140                 if (inode2_is_ancestor < 0) {
1141                         status = inode2_is_ancestor;
1142                         goto bail;
1143                 }
1144
1145                 if ((inode1_is_ancestor == 1) ||
1146                                 (oi1->ip_blkno < oi2->ip_blkno &&
1147                                 inode2_is_ancestor == 0)) {
1148                         /* switch id1 and id2 around */
1149                         swap(bh2, bh1);
1150                         swap(inode2, inode1);
1151                 }
1152                 /* lock id2 */
1153                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1154                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1155                 if (status < 0) {
1156                         if (status != -ENOENT)
1157                                 mlog_errno(status);
1158                         goto bail;
1159                 }
1160         }
1161
1162         /* lock id1 */
1163         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1164                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1165         if (status < 0) {
1166                 /*
1167                  * An error return must mean that no cluster locks
1168                  * were held on function exit.
1169                  */
1170                 if (oi1->ip_blkno != oi2->ip_blkno) {
1171                         ocfs2_inode_unlock(inode2, 1);
1172                         brelse(*bh2);
1173                         *bh2 = NULL;
1174                 }
1175
1176                 if (status != -ENOENT)
1177                         mlog_errno(status);
1178         }
1179
1180         trace_ocfs2_double_lock_end(
1181                         (unsigned long long)oi1->ip_blkno,
1182                         (unsigned long long)oi2->ip_blkno);
1183
1184 bail:
1185         if (status)
1186                 mlog_errno(status);
1187         return status;
1188 }
1189
1190 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1191 {
1192         ocfs2_inode_unlock(inode1, 1);
1193
1194         if (inode1 != inode2)
1195                 ocfs2_inode_unlock(inode2, 1);
1196 }
1197
1198 static int ocfs2_rename(struct mnt_idmap *idmap,
1199                         struct inode *old_dir,
1200                         struct dentry *old_dentry,
1201                         struct inode *new_dir,
1202                         struct dentry *new_dentry,
1203                         unsigned int flags)
1204 {
1205         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1206         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1207         struct inode *old_inode = d_inode(old_dentry);
1208         struct inode *new_inode = d_inode(new_dentry);
1209         struct inode *orphan_dir = NULL;
1210         struct ocfs2_dinode *newfe = NULL;
1211         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1212         struct buffer_head *newfe_bh = NULL;
1213         struct buffer_head *old_inode_bh = NULL;
1214         struct ocfs2_super *osb = NULL;
1215         u64 newfe_blkno, old_de_ino;
1216         handle_t *handle = NULL;
1217         struct buffer_head *old_dir_bh = NULL;
1218         struct buffer_head *new_dir_bh = NULL;
1219         u32 old_dir_nlink = old_dir->i_nlink;
1220         struct ocfs2_dinode *old_di;
1221         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1222         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1223         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1224         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1225         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1226         bool should_add_orphan = false;
1227
1228         if (flags)
1229                 return -EINVAL;
1230
1231         /* At some point it might be nice to break this function up a
1232          * bit. */
1233
1234         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1235                            old_dentry->d_name.len, old_dentry->d_name.name,
1236                            new_dentry->d_name.len, new_dentry->d_name.name);
1237
1238         status = dquot_initialize(old_dir);
1239         if (status) {
1240                 mlog_errno(status);
1241                 goto bail;
1242         }
1243         status = dquot_initialize(new_dir);
1244         if (status) {
1245                 mlog_errno(status);
1246                 goto bail;
1247         }
1248
1249         osb = OCFS2_SB(old_dir->i_sb);
1250
1251         if (new_inode) {
1252                 if (!igrab(new_inode))
1253                         BUG();
1254         }
1255
1256         /* Assume a directory hierarchy thusly:
1257          * a/b/c
1258          * a/d
1259          * a,b,c, and d are all directories.
1260          *
1261          * from cwd of 'a' on both nodes:
1262          * node1: mv b/c d
1263          * node2: mv d   b/c
1264          *
1265          * And that's why, just like the VFS, we need a file system
1266          * rename lock. */
1267         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1268                 status = ocfs2_rename_lock(osb);
1269                 if (status < 0) {
1270                         mlog_errno(status);
1271                         goto bail;
1272                 }
1273                 rename_lock = 1;
1274
1275                 /* here we cannot guarantee the inodes haven't just been
1276                  * changed, so check if they are nested again */
1277                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1278                                 old_inode->i_ino);
1279                 if (status < 0) {
1280                         mlog_errno(status);
1281                         goto bail;
1282                 } else if (status == 1) {
1283                         status = -EPERM;
1284                         trace_ocfs2_rename_not_permitted(
1285                                         (unsigned long long)old_inode->i_ino,
1286                                         (unsigned long long)new_dir->i_ino);
1287                         goto bail;
1288                 }
1289         }
1290
1291         /* if old and new are the same, this'll just do one lock. */
1292         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1293                                    &new_dir_bh, new_dir, 1);
1294         if (status < 0) {
1295                 mlog_errno(status);
1296                 goto bail;
1297         }
1298         parents_locked = 1;
1299
1300         if (!new_dir->i_nlink) {
1301                 status = -EACCES;
1302                 goto bail;
1303         }
1304
1305         /* make sure both dirs have bhs
1306          * get an extra ref on old_dir_bh if old==new */
1307         if (!new_dir_bh) {
1308                 if (old_dir_bh) {
1309                         new_dir_bh = old_dir_bh;
1310                         get_bh(new_dir_bh);
1311                 } else {
1312                         mlog(ML_ERROR, "no old_dir_bh!\n");
1313                         status = -EIO;
1314                         goto bail;
1315                 }
1316         }
1317
1318         /*
1319          * Aside from allowing a meta data update, the locking here
1320          * also ensures that the downconvert thread on other nodes
1321          * won't have to concurrently downconvert the inode and the
1322          * dentry locks.
1323          */
1324         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1325                                          OI_LS_PARENT);
1326         if (status < 0) {
1327                 if (status != -ENOENT)
1328                         mlog_errno(status);
1329                 goto bail;
1330         }
1331         old_child_locked = 1;
1332
1333         status = ocfs2_remote_dentry_delete(old_dentry);
1334         if (status < 0) {
1335                 mlog_errno(status);
1336                 goto bail;
1337         }
1338
1339         if (S_ISDIR(old_inode->i_mode)) {
1340                 u64 old_inode_parent;
1341
1342                 update_dot_dot = 1;
1343                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1344                                                   old_inode,
1345                                                   &old_inode_dot_dot_res);
1346                 if (status) {
1347                         status = -EIO;
1348                         goto bail;
1349                 }
1350
1351                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1352                         status = -EIO;
1353                         goto bail;
1354                 }
1355
1356                 if (!new_inode && new_dir != old_dir &&
1357                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1358                         status = -EMLINK;
1359                         goto bail;
1360                 }
1361         }
1362
1363         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1364                                             old_dentry->d_name.len,
1365                                             &old_de_ino);
1366         if (status) {
1367                 status = -ENOENT;
1368                 goto bail;
1369         }
1370
1371         /*
1372          *  Check for inode number is _not_ due to possible IO errors.
1373          *  We might rmdir the source, keep it as pwd of some process
1374          *  and merrily kill the link to whatever was created under the
1375          *  same name. Goodbye sticky bit ;-<
1376          */
1377         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1378                 status = -ENOENT;
1379                 goto bail;
1380         }
1381
1382         /* check if the target already exists (in which case we need
1383          * to delete it */
1384         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1385                                           new_dentry->d_name.len,
1386                                           &newfe_blkno, new_dir,
1387                                           &target_lookup_res);
1388         /* The only error we allow here is -ENOENT because the new
1389          * file not existing is perfectly valid. */
1390         if ((status < 0) && (status != -ENOENT)) {
1391                 /* If we cannot find the file specified we should just */
1392                 /* return the error... */
1393                 mlog_errno(status);
1394                 goto bail;
1395         }
1396         if (status == 0)
1397                 target_exists = 1;
1398
1399         if (!target_exists && new_inode) {
1400                 /*
1401                  * Target was unlinked by another node while we were
1402                  * waiting to get to ocfs2_rename(). There isn't
1403                  * anything we can do here to help the situation, so
1404                  * bubble up the appropriate error.
1405                  */
1406                 status = -ENOENT;
1407                 goto bail;
1408         }
1409
1410         /* In case we need to overwrite an existing file, we blow it
1411          * away first */
1412         if (target_exists) {
1413                 /* VFS didn't think there existed an inode here, but
1414                  * someone else in the cluster must have raced our
1415                  * rename to create one. Today we error cleanly, in
1416                  * the future we should consider calling iget to build
1417                  * a new struct inode for this entry. */
1418                 if (!new_inode) {
1419                         status = -EACCES;
1420
1421                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1422                                                 new_dentry->d_name.name);
1423                         goto bail;
1424                 }
1425
1426                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1427                         status = -EACCES;
1428
1429                         trace_ocfs2_rename_disagree(
1430                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1431                              (unsigned long long)newfe_blkno,
1432                              OCFS2_I(new_inode)->ip_flags);
1433                         goto bail;
1434                 }
1435
1436                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1437                 if (status < 0) {
1438                         if (status != -ENOENT)
1439                                 mlog_errno(status);
1440                         goto bail;
1441                 }
1442                 new_child_locked = 1;
1443
1444                 status = ocfs2_remote_dentry_delete(new_dentry);
1445                 if (status < 0) {
1446                         mlog_errno(status);
1447                         goto bail;
1448                 }
1449
1450                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1451
1452                 trace_ocfs2_rename_over_existing(
1453                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1454                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1455
1456                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1457                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1458                                                 OCFS2_I(new_inode)->ip_blkno,
1459                                                 orphan_name, &orphan_insert,
1460                                                 false);
1461                         if (status < 0) {
1462                                 mlog_errno(status);
1463                                 goto bail;
1464                         }
1465                         should_add_orphan = true;
1466                 }
1467         } else {
1468                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1469
1470                 status = ocfs2_check_dir_for_entry(new_dir,
1471                                                    new_dentry->d_name.name,
1472                                                    new_dentry->d_name.len);
1473                 if (status)
1474                         goto bail;
1475
1476                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1477                                                       new_dentry->d_name.name,
1478                                                       new_dentry->d_name.len,
1479                                                       &target_insert);
1480                 if (status < 0) {
1481                         mlog_errno(status);
1482                         goto bail;
1483                 }
1484         }
1485
1486         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1487         if (IS_ERR(handle)) {
1488                 status = PTR_ERR(handle);
1489                 handle = NULL;
1490                 mlog_errno(status);
1491                 goto bail;
1492         }
1493
1494         if (target_exists) {
1495                 if (S_ISDIR(new_inode->i_mode)) {
1496                         if (new_inode->i_nlink != 2 ||
1497                             !ocfs2_empty_dir(new_inode)) {
1498                                 status = -ENOTEMPTY;
1499                                 goto bail;
1500                         }
1501                 }
1502                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1503                                                  newfe_bh,
1504                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1505                 if (status < 0) {
1506                         mlog_errno(status);
1507                         goto bail;
1508                 }
1509
1510                 /* change the dirent to point to the correct inode */
1511                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1512                                             old_inode);
1513                 if (status < 0) {
1514                         mlog_errno(status);
1515                         goto bail;
1516                 }
1517                 inode_inc_iversion(new_dir);
1518
1519                 if (S_ISDIR(new_inode->i_mode))
1520                         ocfs2_set_links_count(newfe, 0);
1521                 else
1522                         ocfs2_add_links_count(newfe, -1);
1523                 ocfs2_journal_dirty(handle, newfe_bh);
1524                 if (should_add_orphan) {
1525                         status = ocfs2_orphan_add(osb, handle, new_inode,
1526                                         newfe_bh, orphan_name,
1527                                         &orphan_insert, orphan_dir, false);
1528                         if (status < 0) {
1529                                 mlog_errno(status);
1530                                 goto bail;
1531                         }
1532                 }
1533         } else {
1534                 /* if the name was not found in new_dir, add it now */
1535                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1536                                          OCFS2_I(old_inode)->ip_blkno,
1537                                          new_dir_bh, &target_insert);
1538         }
1539
1540         old_inode->i_ctime = current_time(old_inode);
1541         mark_inode_dirty(old_inode);
1542
1543         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1544                                          old_inode_bh,
1545                                          OCFS2_JOURNAL_ACCESS_WRITE);
1546         if (status >= 0) {
1547                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1548
1549                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1550                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1551                 ocfs2_journal_dirty(handle, old_inode_bh);
1552         } else
1553                 mlog_errno(status);
1554
1555         /*
1556          * Now that the name has been added to new_dir, remove the old name.
1557          *
1558          * We don't keep any directory entry context around until now
1559          * because the insert might have changed the type of directory
1560          * we're dealing with.
1561          */
1562         status = ocfs2_find_entry(old_dentry->d_name.name,
1563                                   old_dentry->d_name.len, old_dir,
1564                                   &old_entry_lookup);
1565         if (status) {
1566                 if (!is_journal_aborted(osb->journal->j_journal)) {
1567                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1568                                         "is not deleted.",
1569                                         new_dentry->d_name.len, new_dentry->d_name.name,
1570                                         old_dentry->d_name.len, old_dentry->d_name.name);
1571                 }
1572                 goto bail;
1573         }
1574
1575         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1576         if (status < 0) {
1577                 mlog_errno(status);
1578                 if (!is_journal_aborted(osb->journal->j_journal)) {
1579                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1580                                         "is not deleted.",
1581                                         new_dentry->d_name.len, new_dentry->d_name.name,
1582                                         old_dentry->d_name.len, old_dentry->d_name.name);
1583                 }
1584                 goto bail;
1585         }
1586
1587         if (new_inode) {
1588                 drop_nlink(new_inode);
1589                 new_inode->i_ctime = current_time(new_inode);
1590         }
1591         old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1592
1593         if (update_dot_dot) {
1594                 status = ocfs2_update_entry(old_inode, handle,
1595                                             &old_inode_dot_dot_res, new_dir);
1596                 drop_nlink(old_dir);
1597                 if (new_inode) {
1598                         drop_nlink(new_inode);
1599                 } else {
1600                         inc_nlink(new_dir);
1601                         mark_inode_dirty(new_dir);
1602                 }
1603         }
1604         mark_inode_dirty(old_dir);
1605         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1606         if (new_inode) {
1607                 mark_inode_dirty(new_inode);
1608                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1609         }
1610
1611         if (old_dir != new_dir) {
1612                 /* Keep the same times on both directories.*/
1613                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1614
1615                 /*
1616                  * This will also pick up the i_nlink change from the
1617                  * block above.
1618                  */
1619                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1620         }
1621
1622         if (old_dir_nlink != old_dir->i_nlink) {
1623                 if (!old_dir_bh) {
1624                         mlog(ML_ERROR, "need to change nlink for old dir "
1625                              "%llu from %d to %d but bh is NULL!\n",
1626                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1627                              (int)old_dir_nlink, old_dir->i_nlink);
1628                 } else {
1629                         struct ocfs2_dinode *fe;
1630                         status = ocfs2_journal_access_di(handle,
1631                                                          INODE_CACHE(old_dir),
1632                                                          old_dir_bh,
1633                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1634                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1635                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1636                         ocfs2_journal_dirty(handle, old_dir_bh);
1637                 }
1638         }
1639         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1640         status = 0;
1641 bail:
1642         if (handle)
1643                 ocfs2_commit_trans(osb, handle);
1644
1645         if (orphan_dir) {
1646                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1647                 ocfs2_inode_unlock(orphan_dir, 1);
1648                 inode_unlock(orphan_dir);
1649                 iput(orphan_dir);
1650         }
1651
1652         if (new_child_locked)
1653                 ocfs2_inode_unlock(new_inode, 1);
1654
1655         if (old_child_locked)
1656                 ocfs2_inode_unlock(old_inode, 1);
1657
1658         if (parents_locked)
1659                 ocfs2_double_unlock(old_dir, new_dir);
1660
1661         if (rename_lock)
1662                 ocfs2_rename_unlock(osb);
1663
1664         if (new_inode)
1665                 sync_mapping_buffers(old_inode->i_mapping);
1666
1667         iput(new_inode);
1668
1669         ocfs2_free_dir_lookup_result(&target_lookup_res);
1670         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1671         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1672         ocfs2_free_dir_lookup_result(&orphan_insert);
1673         ocfs2_free_dir_lookup_result(&target_insert);
1674
1675         brelse(newfe_bh);
1676         brelse(old_inode_bh);
1677         brelse(old_dir_bh);
1678         brelse(new_dir_bh);
1679
1680         if (status)
1681                 mlog_errno(status);
1682
1683         return status;
1684 }
1685
1686 /*
1687  * we expect i_size = strlen(symname). Copy symname into the file
1688  * data, including the null terminator.
1689  */
1690 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1691                                      handle_t *handle,
1692                                      struct inode *inode,
1693                                      const char *symname)
1694 {
1695         struct buffer_head **bhs = NULL;
1696         const char *c;
1697         struct super_block *sb = osb->sb;
1698         u64 p_blkno, p_blocks;
1699         int virtual, blocks, status, i, bytes_left;
1700
1701         bytes_left = i_size_read(inode) + 1;
1702         /* we can't trust i_blocks because we're actually going to
1703          * write i_size + 1 bytes. */
1704         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1705
1706         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1707                                         i_size_read(inode), blocks);
1708
1709         /* Sanity check -- make sure we're going to fit. */
1710         if (bytes_left >
1711             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1712                 status = -EIO;
1713                 mlog_errno(status);
1714                 goto bail;
1715         }
1716
1717         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1718         if (!bhs) {
1719                 status = -ENOMEM;
1720                 mlog_errno(status);
1721                 goto bail;
1722         }
1723
1724         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1725                                              NULL);
1726         if (status < 0) {
1727                 mlog_errno(status);
1728                 goto bail;
1729         }
1730
1731         /* links can never be larger than one cluster so we know this
1732          * is all going to be contiguous, but do a sanity check
1733          * anyway. */
1734         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1735                 status = -EIO;
1736                 mlog_errno(status);
1737                 goto bail;
1738         }
1739
1740         virtual = 0;
1741         while(bytes_left > 0) {
1742                 c = &symname[virtual * sb->s_blocksize];
1743
1744                 bhs[virtual] = sb_getblk(sb, p_blkno);
1745                 if (!bhs[virtual]) {
1746                         status = -ENOMEM;
1747                         mlog_errno(status);
1748                         goto bail;
1749                 }
1750                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1751                                               bhs[virtual]);
1752
1753                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1754                                               bhs[virtual],
1755                                               OCFS2_JOURNAL_ACCESS_CREATE);
1756                 if (status < 0) {
1757                         mlog_errno(status);
1758                         goto bail;
1759                 }
1760
1761                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1762
1763                 memcpy(bhs[virtual]->b_data, c,
1764                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1765                        bytes_left);
1766
1767                 ocfs2_journal_dirty(handle, bhs[virtual]);
1768
1769                 virtual++;
1770                 p_blkno++;
1771                 bytes_left -= sb->s_blocksize;
1772         }
1773
1774         status = 0;
1775 bail:
1776
1777         if (bhs) {
1778                 for(i = 0; i < blocks; i++)
1779                         brelse(bhs[i]);
1780                 kfree(bhs);
1781         }
1782
1783         if (status)
1784                 mlog_errno(status);
1785         return status;
1786 }
1787
1788 static int ocfs2_symlink(struct mnt_idmap *idmap,
1789                          struct inode *dir,
1790                          struct dentry *dentry,
1791                          const char *symname)
1792 {
1793         int status, l, credits;
1794         u64 newsize;
1795         struct ocfs2_super *osb = NULL;
1796         struct inode *inode = NULL;
1797         struct super_block *sb;
1798         struct buffer_head *new_fe_bh = NULL;
1799         struct buffer_head *parent_fe_bh = NULL;
1800         struct ocfs2_dinode *fe = NULL;
1801         struct ocfs2_dinode *dirfe;
1802         handle_t *handle = NULL;
1803         struct ocfs2_alloc_context *inode_ac = NULL;
1804         struct ocfs2_alloc_context *data_ac = NULL;
1805         struct ocfs2_alloc_context *xattr_ac = NULL;
1806         int want_clusters = 0;
1807         int xattr_credits = 0;
1808         struct ocfs2_security_xattr_info si = {
1809                 .name = NULL,
1810                 .enable = 1,
1811         };
1812         int did_quota = 0, did_quota_inode = 0;
1813         struct ocfs2_dir_lookup_result lookup = { NULL, };
1814         sigset_t oldset;
1815         int did_block_signals = 0;
1816         struct ocfs2_dentry_lock *dl = NULL;
1817
1818         trace_ocfs2_symlink_begin(dir, dentry, symname,
1819                                   dentry->d_name.len, dentry->d_name.name);
1820
1821         status = dquot_initialize(dir);
1822         if (status) {
1823                 mlog_errno(status);
1824                 goto bail;
1825         }
1826
1827         sb = dir->i_sb;
1828         osb = OCFS2_SB(sb);
1829
1830         l = strlen(symname) + 1;
1831
1832         credits = ocfs2_calc_symlink_credits(sb);
1833
1834         /* lock the parent directory */
1835         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1836         if (status < 0) {
1837                 if (status != -ENOENT)
1838                         mlog_errno(status);
1839                 return status;
1840         }
1841
1842         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1843         if (!ocfs2_read_links_count(dirfe)) {
1844                 /* can't make a file in a deleted directory. */
1845                 status = -ENOENT;
1846                 goto bail;
1847         }
1848
1849         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1850                                            dentry->d_name.len);
1851         if (status)
1852                 goto bail;
1853
1854         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1855                                               dentry->d_name.name,
1856                                               dentry->d_name.len, &lookup);
1857         if (status < 0) {
1858                 mlog_errno(status);
1859                 goto bail;
1860         }
1861
1862         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1863         if (status < 0) {
1864                 if (status != -ENOSPC)
1865                         mlog_errno(status);
1866                 goto bail;
1867         }
1868
1869         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1870         if (IS_ERR(inode)) {
1871                 status = PTR_ERR(inode);
1872                 inode = NULL;
1873                 mlog_errno(status);
1874                 goto bail;
1875         }
1876
1877         /* get security xattr */
1878         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1879         if (status) {
1880                 if (status == -EOPNOTSUPP)
1881                         si.enable = 0;
1882                 else {
1883                         mlog_errno(status);
1884                         goto bail;
1885                 }
1886         }
1887
1888         /* calculate meta data/clusters for setting security xattr */
1889         if (si.enable) {
1890                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1891                                                   &xattr_credits, &xattr_ac);
1892                 if (status < 0) {
1893                         mlog_errno(status);
1894                         goto bail;
1895                 }
1896         }
1897
1898         /* don't reserve bitmap space for fast symlinks. */
1899         if (l > ocfs2_fast_symlink_chars(sb))
1900                 want_clusters += 1;
1901
1902         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1903         if (status < 0) {
1904                 if (status != -ENOSPC)
1905                         mlog_errno(status);
1906                 goto bail;
1907         }
1908
1909         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1910         if (IS_ERR(handle)) {
1911                 status = PTR_ERR(handle);
1912                 handle = NULL;
1913                 mlog_errno(status);
1914                 goto bail;
1915         }
1916
1917         /* Starting to change things, restart is no longer possible. */
1918         ocfs2_block_signals(&oldset);
1919         did_block_signals = 1;
1920
1921         status = dquot_alloc_inode(inode);
1922         if (status)
1923                 goto bail;
1924         did_quota_inode = 1;
1925
1926         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1927                                    dentry->d_name.name,
1928                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1929                                    inode->i_mode);
1930
1931         status = ocfs2_mknod_locked(osb, dir, inode,
1932                                     0, &new_fe_bh, parent_fe_bh, handle,
1933                                     inode_ac);
1934         if (status < 0) {
1935                 mlog_errno(status);
1936                 goto bail;
1937         }
1938
1939         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1940         inode->i_rdev = 0;
1941         newsize = l - 1;
1942         inode->i_op = &ocfs2_symlink_inode_operations;
1943         inode_nohighmem(inode);
1944         if (l > ocfs2_fast_symlink_chars(sb)) {
1945                 u32 offset = 0;
1946
1947                 status = dquot_alloc_space_nodirty(inode,
1948                     ocfs2_clusters_to_bytes(osb->sb, 1));
1949                 if (status)
1950                         goto bail;
1951                 did_quota = 1;
1952                 inode->i_mapping->a_ops = &ocfs2_aops;
1953                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1954                                               new_fe_bh,
1955                                               handle, data_ac, NULL,
1956                                               NULL);
1957                 if (status < 0) {
1958                         if (status != -ENOSPC && status != -EINTR) {
1959                                 mlog(ML_ERROR,
1960                                      "Failed to extend file to %llu\n",
1961                                      (unsigned long long)newsize);
1962                                 mlog_errno(status);
1963                                 status = -ENOSPC;
1964                         }
1965                         goto bail;
1966                 }
1967                 i_size_write(inode, newsize);
1968                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1969         } else {
1970                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1971                 memcpy((char *) fe->id2.i_symlink, symname, l);
1972                 i_size_write(inode, newsize);
1973                 inode->i_blocks = 0;
1974         }
1975
1976         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1977         if (status < 0) {
1978                 mlog_errno(status);
1979                 goto bail;
1980         }
1981
1982         if (!ocfs2_inode_is_fast_symlink(inode)) {
1983                 status = ocfs2_create_symlink_data(osb, handle, inode,
1984                                                    symname);
1985                 if (status < 0) {
1986                         mlog_errno(status);
1987                         goto bail;
1988                 }
1989         }
1990
1991         if (si.enable) {
1992                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1993                                                  xattr_ac, data_ac);
1994                 if (status < 0) {
1995                         mlog_errno(status);
1996                         goto bail;
1997                 }
1998         }
1999
2000         /*
2001          * Do this before adding the entry to the directory. We add
2002          * also set d_op after success so that ->d_iput() will cleanup
2003          * the dentry lock even if ocfs2_add_entry() fails below.
2004          */
2005         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2006         if (status) {
2007                 mlog_errno(status);
2008                 goto bail;
2009         }
2010
2011         dl = dentry->d_fsdata;
2012
2013         status = ocfs2_add_entry(handle, dentry, inode,
2014                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2015                                  &lookup);
2016         if (status < 0) {
2017                 mlog_errno(status);
2018                 goto bail;
2019         }
2020
2021         insert_inode_hash(inode);
2022         d_instantiate(dentry, inode);
2023 bail:
2024         if (status < 0 && did_quota)
2025                 dquot_free_space_nodirty(inode,
2026                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2027         if (status < 0 && did_quota_inode)
2028                 dquot_free_inode(inode);
2029         if (handle) {
2030                 if (status < 0 && fe)
2031                         ocfs2_set_links_count(fe, 0);
2032                 ocfs2_commit_trans(osb, handle);
2033         }
2034
2035         ocfs2_inode_unlock(dir, 1);
2036         if (did_block_signals)
2037                 ocfs2_unblock_signals(&oldset);
2038
2039         brelse(new_fe_bh);
2040         brelse(parent_fe_bh);
2041         kfree(si.value);
2042         ocfs2_free_dir_lookup_result(&lookup);
2043         if (inode_ac)
2044                 ocfs2_free_alloc_context(inode_ac);
2045         if (data_ac)
2046                 ocfs2_free_alloc_context(data_ac);
2047         if (xattr_ac)
2048                 ocfs2_free_alloc_context(xattr_ac);
2049         if ((status < 0) && inode) {
2050                 if (dl)
2051                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2052
2053                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2054                 clear_nlink(inode);
2055                 iput(inode);
2056         }
2057
2058         if (status)
2059                 mlog_errno(status);
2060
2061         return status;
2062 }
2063
2064 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2065 {
2066         int status, namelen;
2067
2068         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2069                            (long long)blkno);
2070         if (namelen <= 0) {
2071                 if (namelen)
2072                         status = namelen;
2073                 else
2074                         status = -EINVAL;
2075                 mlog_errno(status);
2076                 goto bail;
2077         }
2078         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2079                 status = -EINVAL;
2080                 mlog_errno(status);
2081                 goto bail;
2082         }
2083
2084         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2085
2086         status = 0;
2087 bail:
2088         if (status < 0)
2089                 mlog_errno(status);
2090         return status;
2091 }
2092
2093 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2094                                         struct inode **ret_orphan_dir,
2095                                         struct buffer_head **ret_orphan_dir_bh)
2096 {
2097         struct inode *orphan_dir_inode;
2098         struct buffer_head *orphan_dir_bh = NULL;
2099         int ret = 0;
2100
2101         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2102                                                        ORPHAN_DIR_SYSTEM_INODE,
2103                                                        osb->slot_num);
2104         if (!orphan_dir_inode) {
2105                 ret = -ENOENT;
2106                 mlog_errno(ret);
2107                 return ret;
2108         }
2109
2110         inode_lock(orphan_dir_inode);
2111
2112         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2113         if (ret < 0) {
2114                 inode_unlock(orphan_dir_inode);
2115                 iput(orphan_dir_inode);
2116
2117                 mlog_errno(ret);
2118                 return ret;
2119         }
2120
2121         *ret_orphan_dir = orphan_dir_inode;
2122         *ret_orphan_dir_bh = orphan_dir_bh;
2123
2124         return 0;
2125 }
2126
2127 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2128                                       struct buffer_head *orphan_dir_bh,
2129                                       u64 blkno,
2130                                       char *name,
2131                                       struct ocfs2_dir_lookup_result *lookup,
2132                                       bool dio)
2133 {
2134         int ret;
2135         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2136         int namelen = dio ?
2137                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2138                         OCFS2_ORPHAN_NAMELEN;
2139
2140         if (dio) {
2141                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2142                                 OCFS2_DIO_ORPHAN_PREFIX);
2143                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2144                         ret = -EINVAL;
2145                         mlog_errno(ret);
2146                         return ret;
2147                 }
2148
2149                 ret = ocfs2_blkno_stringify(blkno,
2150                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2151         } else
2152                 ret = ocfs2_blkno_stringify(blkno, name);
2153         if (ret < 0) {
2154                 mlog_errno(ret);
2155                 return ret;
2156         }
2157
2158         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2159                                            orphan_dir_bh, name,
2160                                            namelen, lookup);
2161         if (ret < 0) {
2162                 mlog_errno(ret);
2163                 return ret;
2164         }
2165
2166         return 0;
2167 }
2168
2169 /**
2170  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2171  * insertion of an orphan.
2172  * @osb: ocfs2 file system
2173  * @ret_orphan_dir: Orphan dir inode - returned locked!
2174  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2175  * @lookup: dir lookup result, to be passed back into functions like
2176  *          ocfs2_orphan_add
2177  *
2178  * Returns zero on success and the ret_orphan_dir, name and lookup
2179  * fields will be populated.
2180  *
2181  * Returns non-zero on failure.
2182  */
2183 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2184                                     struct inode **ret_orphan_dir,
2185                                     u64 blkno,
2186                                     char *name,
2187                                     struct ocfs2_dir_lookup_result *lookup,
2188                                     bool dio)
2189 {
2190         struct inode *orphan_dir_inode = NULL;
2191         struct buffer_head *orphan_dir_bh = NULL;
2192         int ret = 0;
2193
2194         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2195                                            &orphan_dir_bh);
2196         if (ret < 0) {
2197                 mlog_errno(ret);
2198                 return ret;
2199         }
2200
2201         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2202                                          blkno, name, lookup, dio);
2203         if (ret < 0) {
2204                 mlog_errno(ret);
2205                 goto out;
2206         }
2207
2208         *ret_orphan_dir = orphan_dir_inode;
2209
2210 out:
2211         brelse(orphan_dir_bh);
2212
2213         if (ret) {
2214                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2215                 inode_unlock(orphan_dir_inode);
2216                 iput(orphan_dir_inode);
2217         }
2218
2219         if (ret)
2220                 mlog_errno(ret);
2221         return ret;
2222 }
2223
2224 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2225                             handle_t *handle,
2226                             struct inode *inode,
2227                             struct buffer_head *fe_bh,
2228                             char *name,
2229                             struct ocfs2_dir_lookup_result *lookup,
2230                             struct inode *orphan_dir_inode,
2231                             bool dio)
2232 {
2233         struct buffer_head *orphan_dir_bh = NULL;
2234         int status = 0;
2235         struct ocfs2_dinode *orphan_fe;
2236         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2237         int namelen = dio ?
2238                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2239                         OCFS2_ORPHAN_NAMELEN;
2240
2241         trace_ocfs2_orphan_add_begin(
2242                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2243
2244         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2245         if (status < 0) {
2246                 mlog_errno(status);
2247                 goto leave;
2248         }
2249
2250         status = ocfs2_journal_access_di(handle,
2251                                          INODE_CACHE(orphan_dir_inode),
2252                                          orphan_dir_bh,
2253                                          OCFS2_JOURNAL_ACCESS_WRITE);
2254         if (status < 0) {
2255                 mlog_errno(status);
2256                 goto leave;
2257         }
2258
2259         /*
2260          * We're going to journal the change of i_flags and i_orphaned_slot.
2261          * It's safe anyway, though some callers may duplicate the journaling.
2262          * Journaling within the func just make the logic look more
2263          * straightforward.
2264          */
2265         status = ocfs2_journal_access_di(handle,
2266                                          INODE_CACHE(inode),
2267                                          fe_bh,
2268                                          OCFS2_JOURNAL_ACCESS_WRITE);
2269         if (status < 0) {
2270                 mlog_errno(status);
2271                 goto leave;
2272         }
2273
2274         /* we're a cluster, and nlink can change on disk from
2275          * underneath us... */
2276         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2277         if (S_ISDIR(inode->i_mode))
2278                 ocfs2_add_links_count(orphan_fe, 1);
2279         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2280         ocfs2_journal_dirty(handle, orphan_dir_bh);
2281
2282         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2283                                    namelen, inode,
2284                                    OCFS2_I(inode)->ip_blkno,
2285                                    orphan_dir_bh, lookup);
2286         if (status < 0) {
2287                 mlog_errno(status);
2288                 goto rollback;
2289         }
2290
2291         if (dio) {
2292                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2293                  * slot.
2294                  */
2295                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2296                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2297         } else {
2298                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2299                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2300
2301                 /* Record which orphan dir our inode now resides
2302                  * in. delete_inode will use this to determine which orphan
2303                  * dir to lock. */
2304                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2305         }
2306
2307         ocfs2_journal_dirty(handle, fe_bh);
2308
2309         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2310                                    osb->slot_num);
2311
2312 rollback:
2313         if (status < 0) {
2314                 if (S_ISDIR(inode->i_mode))
2315                         ocfs2_add_links_count(orphan_fe, -1);
2316                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2317         }
2318
2319 leave:
2320         brelse(orphan_dir_bh);
2321
2322         return status;
2323 }
2324
2325 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2326 int ocfs2_orphan_del(struct ocfs2_super *osb,
2327                      handle_t *handle,
2328                      struct inode *orphan_dir_inode,
2329                      struct inode *inode,
2330                      struct buffer_head *orphan_dir_bh,
2331                      bool dio)
2332 {
2333         char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2334         struct ocfs2_dinode *orphan_fe;
2335         int status = 0;
2336         struct ocfs2_dir_lookup_result lookup = { NULL, };
2337
2338         if (dio) {
2339                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2340                                 OCFS2_DIO_ORPHAN_PREFIX);
2341                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2342                         status = -EINVAL;
2343                         mlog_errno(status);
2344                         return status;
2345                 }
2346
2347                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2348                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2349         } else
2350                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2351         if (status < 0) {
2352                 mlog_errno(status);
2353                 goto leave;
2354         }
2355
2356         trace_ocfs2_orphan_del(
2357              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2358              name, strlen(name));
2359
2360         status = ocfs2_journal_access_di(handle,
2361                                          INODE_CACHE(orphan_dir_inode),
2362                                          orphan_dir_bh,
2363                                          OCFS2_JOURNAL_ACCESS_WRITE);
2364         if (status < 0) {
2365                 mlog_errno(status);
2366                 goto leave;
2367         }
2368
2369         /* find it's spot in the orphan directory */
2370         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2371                                   &lookup);
2372         if (status) {
2373                 mlog_errno(status);
2374                 goto leave;
2375         }
2376
2377         /* remove it from the orphan directory */
2378         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2379         if (status < 0) {
2380                 mlog_errno(status);
2381                 goto leave;
2382         }
2383
2384         /* do the i_nlink dance! :) */
2385         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2386         if (S_ISDIR(inode->i_mode))
2387                 ocfs2_add_links_count(orphan_fe, -1);
2388         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2389         ocfs2_journal_dirty(handle, orphan_dir_bh);
2390
2391 leave:
2392         ocfs2_free_dir_lookup_result(&lookup);
2393
2394         if (status)
2395                 mlog_errno(status);
2396         return status;
2397 }
2398
2399 /**
2400  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2401  * allocated file. This is different from the typical 'add to orphan dir'
2402  * operation in that the inode does not yet exist. This is a problem because
2403  * the orphan dir stringifies the inode block number to come up with it's
2404  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2405  * problem. This function works around it by calling deeper into the orphan
2406  * and suballoc code than other callers. Use this only by necessity.
2407  * @dir: The directory which this inode will ultimately wind up under - not the
2408  * orphan dir!
2409  * @dir_bh: buffer_head the @dir inode block
2410  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2411  * with the string to be used for orphan dirent. Pass back to the orphan dir
2412  * code.
2413  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2414  * dir code.
2415  * @ret_di_blkno: block number where the new inode will be allocated.
2416  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2417  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2418  *
2419  * Returns zero on success and the ret_orphan_dir, name and lookup
2420  * fields will be populated.
2421  *
2422  * Returns non-zero on failure.
2423  */
2424 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2425                                         struct buffer_head *dir_bh,
2426                                         char *orphan_name,
2427                                         struct inode **ret_orphan_dir,
2428                                         u64 *ret_di_blkno,
2429                                         struct ocfs2_dir_lookup_result *orphan_insert,
2430                                         struct ocfs2_alloc_context **ret_inode_ac)
2431 {
2432         int ret;
2433         u64 di_blkno;
2434         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2435         struct inode *orphan_dir = NULL;
2436         struct buffer_head *orphan_dir_bh = NULL;
2437         struct ocfs2_alloc_context *inode_ac = NULL;
2438
2439         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2440         if (ret < 0) {
2441                 mlog_errno(ret);
2442                 return ret;
2443         }
2444
2445         /* reserve an inode spot */
2446         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2447         if (ret < 0) {
2448                 if (ret != -ENOSPC)
2449                         mlog_errno(ret);
2450                 goto out;
2451         }
2452
2453         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2454                                        &di_blkno);
2455         if (ret) {
2456                 mlog_errno(ret);
2457                 goto out;
2458         }
2459
2460         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2461                                          di_blkno, orphan_name, orphan_insert,
2462                                          false);
2463         if (ret < 0) {
2464                 mlog_errno(ret);
2465                 goto out;
2466         }
2467
2468 out:
2469         if (ret == 0) {
2470                 *ret_orphan_dir = orphan_dir;
2471                 *ret_di_blkno = di_blkno;
2472                 *ret_inode_ac = inode_ac;
2473                 /*
2474                  * orphan_name and orphan_insert are already up to
2475                  * date via prepare_orphan_dir
2476                  */
2477         } else {
2478                 /* Unroll reserve_new_inode* */
2479                 if (inode_ac)
2480                         ocfs2_free_alloc_context(inode_ac);
2481
2482                 /* Unroll orphan dir locking */
2483                 inode_unlock(orphan_dir);
2484                 ocfs2_inode_unlock(orphan_dir, 1);
2485                 iput(orphan_dir);
2486         }
2487
2488         brelse(orphan_dir_bh);
2489
2490         return ret;
2491 }
2492
2493 int ocfs2_create_inode_in_orphan(struct inode *dir,
2494                                  int mode,
2495                                  struct inode **new_inode)
2496 {
2497         int status, did_quota_inode = 0;
2498         struct inode *inode = NULL;
2499         struct inode *orphan_dir = NULL;
2500         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2501         handle_t *handle = NULL;
2502         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2503         struct buffer_head *parent_di_bh = NULL;
2504         struct buffer_head *new_di_bh = NULL;
2505         struct ocfs2_alloc_context *inode_ac = NULL;
2506         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2507         u64 di_blkno, suballoc_loc;
2508         u16 suballoc_bit;
2509
2510         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2511         if (status < 0) {
2512                 if (status != -ENOENT)
2513                         mlog_errno(status);
2514                 return status;
2515         }
2516
2517         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2518                                               orphan_name, &orphan_dir,
2519                                               &di_blkno, &orphan_insert, &inode_ac);
2520         if (status < 0) {
2521                 if (status != -ENOSPC)
2522                         mlog_errno(status);
2523                 goto leave;
2524         }
2525
2526         inode = ocfs2_get_init_inode(dir, mode);
2527         if (IS_ERR(inode)) {
2528                 status = PTR_ERR(inode);
2529                 inode = NULL;
2530                 mlog_errno(status);
2531                 goto leave;
2532         }
2533
2534         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2535         if (IS_ERR(handle)) {
2536                 status = PTR_ERR(handle);
2537                 handle = NULL;
2538                 mlog_errno(status);
2539                 goto leave;
2540         }
2541
2542         status = dquot_alloc_inode(inode);
2543         if (status)
2544                 goto leave;
2545         did_quota_inode = 1;
2546
2547         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2548                                               &suballoc_loc,
2549                                               &suballoc_bit, di_blkno);
2550         if (status < 0) {
2551                 mlog_errno(status);
2552                 goto leave;
2553         }
2554
2555         clear_nlink(inode);
2556         /* do the real work now. */
2557         status = __ocfs2_mknod_locked(dir, inode,
2558                                       0, &new_di_bh, parent_di_bh, handle,
2559                                       inode_ac, di_blkno, suballoc_loc,
2560                                       suballoc_bit);
2561         if (status < 0) {
2562                 mlog_errno(status);
2563                 goto leave;
2564         }
2565
2566         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2567                                   &orphan_insert, orphan_dir, false);
2568         if (status < 0) {
2569                 mlog_errno(status);
2570                 goto leave;
2571         }
2572
2573         /* get open lock so that only nodes can't remove it from orphan dir. */
2574         status = ocfs2_open_lock(inode);
2575         if (status < 0)
2576                 mlog_errno(status);
2577
2578         insert_inode_hash(inode);
2579 leave:
2580         if (status < 0 && did_quota_inode)
2581                 dquot_free_inode(inode);
2582         if (handle)
2583                 ocfs2_commit_trans(osb, handle);
2584
2585         if (orphan_dir) {
2586                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2587                 ocfs2_inode_unlock(orphan_dir, 1);
2588                 inode_unlock(orphan_dir);
2589                 iput(orphan_dir);
2590         }
2591
2592         if ((status < 0) && inode) {
2593                 clear_nlink(inode);
2594                 iput(inode);
2595         }
2596
2597         if (inode_ac)
2598                 ocfs2_free_alloc_context(inode_ac);
2599
2600         brelse(new_di_bh);
2601
2602         if (!status)
2603                 *new_inode = inode;
2604
2605         ocfs2_free_dir_lookup_result(&orphan_insert);
2606
2607         ocfs2_inode_unlock(dir, 1);
2608         brelse(parent_di_bh);
2609         return status;
2610 }
2611
2612 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2613         struct inode *inode)
2614 {
2615         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2616         struct inode *orphan_dir_inode = NULL;
2617         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2618         struct buffer_head *di_bh = NULL;
2619         int status = 0;
2620         handle_t *handle = NULL;
2621         struct ocfs2_dinode *di = NULL;
2622
2623         status = ocfs2_inode_lock(inode, &di_bh, 1);
2624         if (status < 0) {
2625                 mlog_errno(status);
2626                 goto bail;
2627         }
2628
2629         di = (struct ocfs2_dinode *) di_bh->b_data;
2630         /*
2631          * Another append dio crashed?
2632          * If so, manually recover it first.
2633          */
2634         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2635                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2636                 if (status < 0) {
2637                         if (status != -ENOSPC)
2638                                 mlog_errno(status);
2639                         goto bail_unlock_inode;
2640                 }
2641
2642                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2643                 if (status < 0) {
2644                         mlog_errno(status);
2645                         goto bail_unlock_inode;
2646                 }
2647         }
2648
2649         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2650                         OCFS2_I(inode)->ip_blkno,
2651                         orphan_name,
2652                         &orphan_insert,
2653                         true);
2654         if (status < 0) {
2655                 mlog_errno(status);
2656                 goto bail_unlock_inode;
2657         }
2658
2659         handle = ocfs2_start_trans(osb,
2660                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2661         if (IS_ERR(handle)) {
2662                 status = PTR_ERR(handle);
2663                 goto bail_unlock_orphan;
2664         }
2665
2666         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2667                         &orphan_insert, orphan_dir_inode, true);
2668         if (status)
2669                 mlog_errno(status);
2670
2671         ocfs2_commit_trans(osb, handle);
2672
2673 bail_unlock_orphan:
2674         ocfs2_inode_unlock(orphan_dir_inode, 1);
2675         inode_unlock(orphan_dir_inode);
2676         iput(orphan_dir_inode);
2677
2678         ocfs2_free_dir_lookup_result(&orphan_insert);
2679
2680 bail_unlock_inode:
2681         ocfs2_inode_unlock(inode, 1);
2682         brelse(di_bh);
2683
2684 bail:
2685         return status;
2686 }
2687
2688 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2689                 struct inode *inode, struct buffer_head *di_bh,
2690                 int update_isize, loff_t end)
2691 {
2692         struct inode *orphan_dir_inode = NULL;
2693         struct buffer_head *orphan_dir_bh = NULL;
2694         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2695         handle_t *handle = NULL;
2696         int status = 0;
2697
2698         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2699                         ORPHAN_DIR_SYSTEM_INODE,
2700                         le16_to_cpu(di->i_dio_orphaned_slot));
2701         if (!orphan_dir_inode) {
2702                 status = -ENOENT;
2703                 mlog_errno(status);
2704                 goto bail;
2705         }
2706
2707         inode_lock(orphan_dir_inode);
2708         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2709         if (status < 0) {
2710                 inode_unlock(orphan_dir_inode);
2711                 iput(orphan_dir_inode);
2712                 mlog_errno(status);
2713                 goto bail;
2714         }
2715
2716         handle = ocfs2_start_trans(osb,
2717                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2718         if (IS_ERR(handle)) {
2719                 status = PTR_ERR(handle);
2720                 goto bail_unlock_orphan;
2721         }
2722
2723         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2724
2725         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2726                                 inode, orphan_dir_bh, true);
2727         if (status < 0) {
2728                 mlog_errno(status);
2729                 goto bail_commit;
2730         }
2731
2732         status = ocfs2_journal_access_di(handle,
2733                         INODE_CACHE(inode),
2734                         di_bh,
2735                         OCFS2_JOURNAL_ACCESS_WRITE);
2736         if (status < 0) {
2737                 mlog_errno(status);
2738                 goto bail_commit;
2739         }
2740
2741         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2742         di->i_dio_orphaned_slot = 0;
2743
2744         if (update_isize) {
2745                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2746                 if (status)
2747                         mlog_errno(status);
2748         } else
2749                 ocfs2_journal_dirty(handle, di_bh);
2750
2751 bail_commit:
2752         ocfs2_commit_trans(osb, handle);
2753
2754 bail_unlock_orphan:
2755         ocfs2_inode_unlock(orphan_dir_inode, 1);
2756         inode_unlock(orphan_dir_inode);
2757         brelse(orphan_dir_bh);
2758         iput(orphan_dir_inode);
2759
2760 bail:
2761         return status;
2762 }
2763
2764 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2765                                    struct inode *inode,
2766                                    struct dentry *dentry)
2767 {
2768         int status = 0;
2769         struct buffer_head *parent_di_bh = NULL;
2770         handle_t *handle = NULL;
2771         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2772         struct ocfs2_dinode *dir_di, *di;
2773         struct inode *orphan_dir_inode = NULL;
2774         struct buffer_head *orphan_dir_bh = NULL;
2775         struct buffer_head *di_bh = NULL;
2776         struct ocfs2_dir_lookup_result lookup = { NULL, };
2777
2778         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2779                                 dentry->d_name.len, dentry->d_name.name,
2780                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2781                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2782
2783         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2784         if (status < 0) {
2785                 if (status != -ENOENT)
2786                         mlog_errno(status);
2787                 return status;
2788         }
2789
2790         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2791         if (!dir_di->i_links_count) {
2792                 /* can't make a file in a deleted directory. */
2793                 status = -ENOENT;
2794                 goto leave;
2795         }
2796
2797         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2798                                            dentry->d_name.len);
2799         if (status)
2800                 goto leave;
2801
2802         /* get a spot inside the dir. */
2803         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2804                                               dentry->d_name.name,
2805                                               dentry->d_name.len, &lookup);
2806         if (status < 0) {
2807                 mlog_errno(status);
2808                 goto leave;
2809         }
2810
2811         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2812                                                        ORPHAN_DIR_SYSTEM_INODE,
2813                                                        osb->slot_num);
2814         if (!orphan_dir_inode) {
2815                 status = -ENOENT;
2816                 mlog_errno(status);
2817                 goto leave;
2818         }
2819
2820         inode_lock(orphan_dir_inode);
2821
2822         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2823         if (status < 0) {
2824                 mlog_errno(status);
2825                 inode_unlock(orphan_dir_inode);
2826                 iput(orphan_dir_inode);
2827                 goto leave;
2828         }
2829
2830         status = ocfs2_read_inode_block(inode, &di_bh);
2831         if (status < 0) {
2832                 mlog_errno(status);
2833                 goto orphan_unlock;
2834         }
2835
2836         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2837         if (IS_ERR(handle)) {
2838                 status = PTR_ERR(handle);
2839                 handle = NULL;
2840                 mlog_errno(status);
2841                 goto orphan_unlock;
2842         }
2843
2844         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2845                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2846         if (status < 0) {
2847                 mlog_errno(status);
2848                 goto out_commit;
2849         }
2850
2851         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2852                                   orphan_dir_bh, false);
2853         if (status < 0) {
2854                 mlog_errno(status);
2855                 goto out_commit;
2856         }
2857
2858         di = (struct ocfs2_dinode *)di_bh->b_data;
2859         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2860         di->i_orphaned_slot = 0;
2861         set_nlink(inode, 1);
2862         ocfs2_set_links_count(di, inode->i_nlink);
2863         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2864         ocfs2_journal_dirty(handle, di_bh);
2865
2866         status = ocfs2_add_entry(handle, dentry, inode,
2867                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2868                                  &lookup);
2869         if (status < 0) {
2870                 mlog_errno(status);
2871                 goto out_commit;
2872         }
2873
2874         status = ocfs2_dentry_attach_lock(dentry, inode,
2875                                           OCFS2_I(dir)->ip_blkno);
2876         if (status) {
2877                 mlog_errno(status);
2878                 goto out_commit;
2879         }
2880
2881         d_instantiate(dentry, inode);
2882         status = 0;
2883 out_commit:
2884         ocfs2_commit_trans(osb, handle);
2885 orphan_unlock:
2886         ocfs2_inode_unlock(orphan_dir_inode, 1);
2887         inode_unlock(orphan_dir_inode);
2888         iput(orphan_dir_inode);
2889 leave:
2890
2891         ocfs2_inode_unlock(dir, 1);
2892
2893         brelse(di_bh);
2894         brelse(parent_di_bh);
2895         brelse(orphan_dir_bh);
2896
2897         ocfs2_free_dir_lookup_result(&lookup);
2898
2899         if (status)
2900                 mlog_errno(status);
2901
2902         return status;
2903 }
2904
2905 const struct inode_operations ocfs2_dir_iops = {
2906         .create         = ocfs2_create,
2907         .lookup         = ocfs2_lookup,
2908         .link           = ocfs2_link,
2909         .unlink         = ocfs2_unlink,
2910         .rmdir          = ocfs2_unlink,
2911         .symlink        = ocfs2_symlink,
2912         .mkdir          = ocfs2_mkdir,
2913         .mknod          = ocfs2_mknod,
2914         .rename         = ocfs2_rename,
2915         .setattr        = ocfs2_setattr,
2916         .getattr        = ocfs2_getattr,
2917         .permission     = ocfs2_permission,
2918         .listxattr      = ocfs2_listxattr,
2919         .fiemap         = ocfs2_fiemap,
2920         .get_inode_acl  = ocfs2_iop_get_acl,
2921         .set_acl        = ocfs2_iop_set_acl,
2922         .fileattr_get   = ocfs2_fileattr_get,
2923         .fileattr_set   = ocfs2_fileattr_set,
2924 };