fs/ocfs2/namei.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * namei.c
   5  *
   6  * Create and rename file, directory, symlinks
   7  *
   8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   9  *
  10  *  Portions of this code from linux/fs/ext3/dir.c
  11  *
  12  *  Copyright (C) 1992, 1993, 1994, 1995
  13  *  Remy Card (card@masi.ibp.fr)
  14  *  Laboratoire MASI - Institut Blaise pascal
  15  *  Universite Pierre et Marie Curie (Paris VI)
  16  *
  17  *   from
  18  *
  19  *   linux/fs/minix/dir.c
  20  *
  21  *   Copyright (C) 1991, 1992 Linux Torvalds
  22  *
  23  * This program is free software; you can redistribute it and/or
  24  * modify it under the terms of the GNU General Public
  25  * License as published by the Free Software Foundation; either
  26  * version 2 of the License, or (at your option) any later version.
  27  *
  28  * This program is distributed in the hope that it will be useful,
  29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  31  * General Public License for more details.
  32  *
  33  * You should have received a copy of the GNU General Public
  34  * License along with this program; if not, write to the
  35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  36  * Boston, MA 021110-1307, USA.
  37  */
  38
  39 #include <linux/fs.h>
  40 #include <linux/types.h>
  41 #include <linux/slab.h>
  42 #include <linux/highmem.h>
  43 #include <linux/quotaops.h>
  44
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48
  49 #include "alloc.h"
  50 #include "dcache.h"
  51 #include "dir.h"
  52 #include "dlmglue.h"
  53 #include "extent_map.h"
  54 #include "file.h"
  55 #include "inode.h"
  56 #include "journal.h"
  57 #include "namei.h"
  58 #include "suballoc.h"
  59 #include "super.h"
  60 #include "symlink.h"
  61 #include "sysfile.h"
  62 #include "uptodate.h"
  63 #include "xattr.h"
  64 #include "acl.h"
  65 #include "ocfs2_trace.h"
  66
  67 #include "buffer_head_io.h"
  68
  69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  70                               struct inode *dir,
  71                               struct inode *inode,
  72                               dev_t dev,
  73                               struct buffer_head **new_fe_bh,
  74                               struct buffer_head *parent_fe_bh,
  75                               handle_t *handle,
  76                               struct ocfs2_alloc_context *inode_ac);
  77
  78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  79                                     struct inode **ret_orphan_dir,
  80                                     u64 blkno,
  81                                     char *name,
  82                                     struct ocfs2_dir_lookup_result *lookup);
  83
  84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  85                             handle_t *handle,
  86                             struct inode *inode,
  87                             struct buffer_head *fe_bh,
  88                             char *name,
  89                             struct ocfs2_dir_lookup_result *lookup,
  90                             struct inode *orphan_dir_inode);
  91
  92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  93                                      handle_t *handle,
  94                                      struct inode *inode,
  95                                      const char *symname);
  96
  97 /* An orphan dir name is an 8 byte value, printed as a hex string */
  98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  99
 100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 101                                    unsigned int flags)
 102 {
 103         int status;
 104         u64 blkno;
 105         struct inode *inode = NULL;
 106         struct dentry *ret;
 107         struct ocfs2_inode_info *oi;
 108
 109         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 110                            dentry->d_name.name,
 111                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 112
 113         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 114                 ret = ERR_PTR(-ENAMETOOLONG);
 115                 goto bail;
 116         }
 117
 118         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 119         if (status < 0) {
 120                 if (status != -ENOENT)
 121                         mlog_errno(status);
 122                 ret = ERR_PTR(status);
 123                 goto bail;
 124         }
 125
 126         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 127                                             dentry->d_name.len, &blkno);
 128         if (status < 0)
 129                 goto bail_add;
 130
 131         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 132         if (IS_ERR(inode)) {
 133                 ret = ERR_PTR(-EACCES);
 134                 goto bail_unlock;
 135         }
 136
 137         oi = OCFS2_I(inode);
 138         /* Clear any orphaned state... If we were able to look up the
 139          * inode from a directory, it certainly can't be orphaned. We
 140          * might have the bad state from a node which intended to
 141          * orphan this inode but crashed before it could commit the
 142          * unlink. */
 143         spin_lock(&oi->ip_lock);
 144         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 145         spin_unlock(&oi->ip_lock);
 146
 147 bail_add:
 148         ret = d_splice_alias(inode, dentry);
 149
 150         if (inode) {
 151                 /*
 152                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 153                  * dentry, it will d_move() it on top of ourse. The
 154                  * return value will indicate this however, so in
 155                  * those cases, we switch them around for the locking
 156                  * code.
 157                  *
 158                  * NOTE: This dentry already has ->d_op set from
 159                  * ocfs2_get_parent() and ocfs2_get_dentry()
 160                  */
 161                 if (ret)
 162                         dentry = ret;
 163
 164                 status = ocfs2_dentry_attach_lock(dentry, inode,
 165                                                   OCFS2_I(dir)->ip_blkno);
 166                 if (status) {
 167                         mlog_errno(status);
 168                         ret = ERR_PTR(status);
 169                         goto bail_unlock;
 170                 }
 171         } else
 172                 ocfs2_dentry_attach_gen(dentry);
 173
 174 bail_unlock:
 175         /* Don't drop the cluster lock until *after* the d_add --
 176          * unlink on another node will message us to remove that
 177          * dentry under this lock so otherwise we can race this with
 178          * the downconvert thread and have a stale dentry. */
 179         ocfs2_inode_unlock(dir, 0);
 180
 181 bail:
 182
 183         trace_ocfs2_lookup_ret(ret);
 184
 185         return ret;
 186 }
 187
 188 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 189 {
 190         struct inode *inode;
 191
 192         inode = new_inode(dir->i_sb);
 193         if (!inode) {
 194                 mlog(ML_ERROR, "new_inode failed!\n");
 195                 return NULL;
 196         }
 197
 198         /* populate as many fields early on as possible - many of
 199          * these are used by the support functions here and in
 200          * callers. */
 201         if (S_ISDIR(mode))
 202                 set_nlink(inode, 2);
 203         inode_init_owner(inode, dir, mode);
 204         dquot_initialize(inode);
 205         return inode;
 206 }
 207
 208 static int ocfs2_mknod(struct inode *dir,
 209                        struct dentry *dentry,
 210                        umode_t mode,
 211                        dev_t dev)
 212 {
 213         int status = 0;
 214         struct buffer_head *parent_fe_bh = NULL;
 215         handle_t *handle = NULL;
 216         struct ocfs2_super *osb;
 217         struct ocfs2_dinode *dirfe;
 218         struct buffer_head *new_fe_bh = NULL;
 219         struct inode *inode = NULL;
 220         struct ocfs2_alloc_context *inode_ac = NULL;
 221         struct ocfs2_alloc_context *data_ac = NULL;
 222         struct ocfs2_alloc_context *meta_ac = NULL;
 223         int want_clusters = 0;
 224         int want_meta = 0;
 225         int xattr_credits = 0;
 226         struct ocfs2_security_xattr_info si = {
 227                 .enable = 1,
 228         };
 229         int did_quota_inode = 0;
 230         struct ocfs2_dir_lookup_result lookup = { NULL, };
 231         sigset_t oldset;
 232         int did_block_signals = 0;
 233         struct posix_acl *default_acl = NULL, *acl = NULL;
 234
 235         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 236                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 237                           (unsigned long)dev, mode);
 238
 239         dquot_initialize(dir);
 240
 241         /* get our super block */
 242         osb = OCFS2_SB(dir->i_sb);
 243
 244         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 245         if (status < 0) {
 246                 if (status != -ENOENT)
 247                         mlog_errno(status);
 248                 return status;
 249         }
 250
 251         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 252                 status = -EMLINK;
 253                 goto leave;
 254         }
 255
 256         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 257         if (!ocfs2_read_links_count(dirfe)) {
 258                 /* can't make a file in a deleted directory. */
 259                 status = -ENOENT;
 260                 goto leave;
 261         }
 262
 263         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 264                                            dentry->d_name.len);
 265         if (status)
 266                 goto leave;
 267
 268         /* get a spot inside the dir. */
 269         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 270                                               dentry->d_name.name,
 271                                               dentry->d_name.len, &lookup);
 272         if (status < 0) {
 273                 mlog_errno(status);
 274                 goto leave;
 275         }
 276
 277         /* reserve an inode spot */
 278         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 279         if (status < 0) {
 280                 if (status != -ENOSPC)
 281                         mlog_errno(status);
 282                 goto leave;
 283         }
 284
 285         inode = ocfs2_get_init_inode(dir, mode);
 286         if (!inode) {
 287                 status = -ENOMEM;
 288                 mlog_errno(status);
 289                 goto leave;
 290         }
 291
 292         /* get security xattr */
 293         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 294         if (status) {
 295                 if (status == -EOPNOTSUPP)
 296                         si.enable = 0;
 297                 else {
 298                         mlog_errno(status);
 299                         goto leave;
 300                 }
 301         }
 302
 303         /* calculate meta data/clusters for setting security and acl xattr */
 304         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 305                                        &si, &want_clusters,
 306                                        &xattr_credits, &want_meta);
 307         if (status < 0) {
 308                 mlog_errno(status);
 309                 goto leave;
 310         }
 311
 312         /* Reserve a cluster if creating an extent based directory. */
 313         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 314                 want_clusters += 1;
 315
 316                 /* Dir indexing requires extra space as well */
 317                 if (ocfs2_supports_indexed_dirs(osb))
 318                         want_meta++;
 319         }
 320
 321         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 322         if (status < 0) {
 323                 if (status != -ENOSPC)
 324                         mlog_errno(status);
 325                 goto leave;
 326         }
 327
 328         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 329         if (status < 0) {
 330                 if (status != -ENOSPC)
 331                         mlog_errno(status);
 332                 goto leave;
 333         }
 334
 335         status = posix_acl_create(dir, &mode, &default_acl, &acl);
 336         if (status) {
 337                 mlog_errno(status);
 338                 goto leave;
 339         }
 340
 341         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 342                                                             S_ISDIR(mode),
 343                                                             xattr_credits));
 344         if (IS_ERR(handle)) {
 345                 status = PTR_ERR(handle);
 346                 handle = NULL;
 347                 mlog_errno(status);
 348                 goto leave;
 349         }
 350
 351         /* Starting to change things, restart is no longer possible. */
 352         ocfs2_block_signals(&oldset);
 353         did_block_signals = 1;
 354
 355         status = dquot_alloc_inode(inode);
 356         if (status)
 357                 goto leave;
 358         did_quota_inode = 1;
 359
 360         /* do the real work now. */
 361         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 362                                     &new_fe_bh, parent_fe_bh, handle,
 363                                     inode_ac);
 364         if (status < 0) {
 365                 mlog_errno(status);
 366                 goto leave;
 367         }
 368
 369         if (S_ISDIR(mode)) {
 370                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 371                                             new_fe_bh, data_ac, meta_ac);
 372                 if (status < 0) {
 373                         mlog_errno(status);
 374                         goto leave;
 375                 }
 376
 377                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 378                                                  parent_fe_bh,
 379                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 380                 if (status < 0) {
 381                         mlog_errno(status);
 382                         goto leave;
 383                 }
 384                 ocfs2_add_links_count(dirfe, 1);
 385                 ocfs2_journal_dirty(handle, parent_fe_bh);
 386                 inc_nlink(dir);
 387         }
 388
 389         if (default_acl) {
 390                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
 391                                        ACL_TYPE_DEFAULT, default_acl,
 392                                        meta_ac, data_ac);
 393         }
 394         if (!status && acl) {
 395                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
 396                                        ACL_TYPE_ACCESS, acl,
 397                                        meta_ac, data_ac);
 398         }
 399
 400         if (status < 0) {
 401                 mlog_errno(status);
 402                 goto leave;
 403         }
 404
 405         if (si.enable) {
 406                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 407                                                  meta_ac, data_ac);
 408                 if (status < 0) {
 409                         mlog_errno(status);
 410                         goto leave;
 411                 }
 412         }
 413
 414         /*
 415          * Do this before adding the entry to the directory. We add
 416          * also set d_op after success so that ->d_iput() will cleanup
 417          * the dentry lock even if ocfs2_add_entry() fails below.
 418          */
 419         status = ocfs2_dentry_attach_lock(dentry, inode,
 420                                           OCFS2_I(dir)->ip_blkno);
 421         if (status) {
 422                 mlog_errno(status);
 423                 goto leave;
 424         }
 425
 426         status = ocfs2_add_entry(handle, dentry, inode,
 427                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 428                                  &lookup);
 429         if (status < 0) {
 430                 mlog_errno(status);
 431                 goto leave;
 432         }
 433
 434         insert_inode_hash(inode);
 435         d_instantiate(dentry, inode);
 436         status = 0;
 437 leave:
 438         if (default_acl)
 439                 posix_acl_release(default_acl);
 440         if (acl)
 441                 posix_acl_release(acl);
 442         if (status < 0 && did_quota_inode)
 443                 dquot_free_inode(inode);
 444         if (handle)
 445                 ocfs2_commit_trans(osb, handle);
 446
 447         ocfs2_inode_unlock(dir, 1);
 448         if (did_block_signals)
 449                 ocfs2_unblock_signals(&oldset);
 450
 451         brelse(new_fe_bh);
 452         brelse(parent_fe_bh);
 453         kfree(si.name);
 454         kfree(si.value);
 455
 456         ocfs2_free_dir_lookup_result(&lookup);
 457
 458         if (inode_ac)
 459                 ocfs2_free_alloc_context(inode_ac);
 460
 461         if (data_ac)
 462                 ocfs2_free_alloc_context(data_ac);
 463
 464         if (meta_ac)
 465                 ocfs2_free_alloc_context(meta_ac);
 466
 467         /*
 468          * We should call iput after the i_mutex of the bitmap been
 469          * unlocked in ocfs2_free_alloc_context, or the
 470          * ocfs2_delete_inode will mutex_lock again.
 471          */
 472         if ((status < 0) && inode) {
 473                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 474                 clear_nlink(inode);
 475                 iput(inode);
 476         }
 477
 478         if (status)
 479                 mlog_errno(status);
 480
 481         return status;
 482 }
 483
 484 static int __ocfs2_mknod_locked(struct inode *dir,
 485                                 struct inode *inode,
 486                                 dev_t dev,
 487                                 struct buffer_head **new_fe_bh,
 488                                 struct buffer_head *parent_fe_bh,
 489                                 handle_t *handle,
 490                                 struct ocfs2_alloc_context *inode_ac,
 491                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 492 {
 493         int status = 0;
 494         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 495         struct ocfs2_dinode *fe = NULL;
 496         struct ocfs2_extent_list *fel;
 497         u16 feat;
 498
 499         *new_fe_bh = NULL;
 500
 501         /* populate as many fields early on as possible - many of
 502          * these are used by the support functions here and in
 503          * callers. */
 504         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 505         OCFS2_I(inode)->ip_blkno = fe_blkno;
 506         spin_lock(&osb->osb_lock);
 507         inode->i_generation = osb->s_next_generation++;
 508         spin_unlock(&osb->osb_lock);
 509
 510         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 511         if (!*new_fe_bh) {
 512                 status = -ENOMEM;
 513                 mlog_errno(status);
 514                 goto leave;
 515         }
 516         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 517
 518         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 519                                          *new_fe_bh,
 520                                          OCFS2_JOURNAL_ACCESS_CREATE);
 521         if (status < 0) {
 522                 mlog_errno(status);
 523                 goto leave;
 524         }
 525
 526         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 527         memset(fe, 0, osb->sb->s_blocksize);
 528
 529         fe->i_generation = cpu_to_le32(inode->i_generation);
 530         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 531         fe->i_blkno = cpu_to_le64(fe_blkno);
 532         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 533         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 534         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 535         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 536         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 537         fe->i_mode = cpu_to_le16(inode->i_mode);
 538         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 539                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 540
 541         ocfs2_set_links_count(fe, inode->i_nlink);
 542
 543         fe->i_last_eb_blk = 0;
 544         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 545         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 546         fe->i_atime = fe->i_ctime = fe->i_mtime =
 547                 cpu_to_le64(CURRENT_TIME.tv_sec);
 548         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 549                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 550         fe->i_dtime = 0;
 551
 552         /*
 553          * If supported, directories start with inline data. If inline
 554          * isn't supported, but indexing is, we start them as indexed.
 555          */
 556         feat = le16_to_cpu(fe->i_dyn_features);
 557         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 558                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 559
 560                 fe->id2.i_data.id_count = cpu_to_le16(
 561                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 562         } else {
 563                 fel = &fe->id2.i_list;
 564                 fel->l_tree_depth = 0;
 565                 fel->l_next_free_rec = 0;
 566                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 567         }
 568
 569         ocfs2_journal_dirty(handle, *new_fe_bh);
 570
 571         ocfs2_populate_inode(inode, fe, 1);
 572         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 573         if (!ocfs2_mount_local(osb)) {
 574                 status = ocfs2_create_new_inode_locks(inode);
 575                 if (status < 0)
 576                         mlog_errno(status);
 577         }
 578
 579         status = 0; /* error in ocfs2_create_new_inode_locks is not
 580                      * critical */
 581
 582 leave:
 583         if (status < 0) {
 584                 if (*new_fe_bh) {
 585                         brelse(*new_fe_bh);
 586                         *new_fe_bh = NULL;
 587                 }
 588         }
 589
 590         if (status)
 591                 mlog_errno(status);
 592         return status;
 593 }
 594
 595 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 596                               struct inode *dir,
 597                               struct inode *inode,
 598                               dev_t dev,
 599                               struct buffer_head **new_fe_bh,
 600                               struct buffer_head *parent_fe_bh,
 601                               handle_t *handle,
 602                               struct ocfs2_alloc_context *inode_ac)
 603 {
 604         int status = 0;
 605         u64 suballoc_loc, fe_blkno = 0;
 606         u16 suballoc_bit;
 607
 608         *new_fe_bh = NULL;
 609
 610         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 611                                        inode_ac, &suballoc_loc,
 612                                        &suballoc_bit, &fe_blkno);
 613         if (status < 0) {
 614                 mlog_errno(status);
 615                 return status;
 616         }
 617
 618         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 619                                     parent_fe_bh, handle, inode_ac,
 620                                     fe_blkno, suballoc_loc, suballoc_bit);
 621 }
 622
 623 static int ocfs2_mkdir(struct inode *dir,
 624                        struct dentry *dentry,
 625                        umode_t mode)
 626 {
 627         int ret;
 628
 629         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 630                           OCFS2_I(dir)->ip_blkno, mode);
 631         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 632         if (ret)
 633                 mlog_errno(ret);
 634
 635         return ret;
 636 }
 637
 638 static int ocfs2_create(struct inode *dir,
 639                         struct dentry *dentry,
 640                         umode_t mode,
 641                         bool excl)
 642 {
 643         int ret;
 644
 645         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 646                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 647         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 648         if (ret)
 649                 mlog_errno(ret);
 650
 651         return ret;
 652 }
 653
 654 static int ocfs2_link(struct dentry *old_dentry,
 655                       struct inode *dir,
 656                       struct dentry *dentry)
 657 {
 658         handle_t *handle;
 659         struct inode *inode = old_dentry->d_inode;
 660         int err;
 661         struct buffer_head *fe_bh = NULL;
 662         struct buffer_head *parent_fe_bh = NULL;
 663         struct ocfs2_dinode *fe = NULL;
 664         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 665         struct ocfs2_dir_lookup_result lookup = { NULL, };
 666         sigset_t oldset;
 667         u64 old_de_ino;
 668
 669         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 670                          old_dentry->d_name.len, old_dentry->d_name.name,
 671                          dentry->d_name.len, dentry->d_name.name);
 672
 673         if (S_ISDIR(inode->i_mode))
 674                 return -EPERM;
 675
 676         dquot_initialize(dir);
 677
 678         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 679         if (err < 0) {
 680                 if (err != -ENOENT)
 681                         mlog_errno(err);
 682                 return err;
 683         }
 684
 685         if (!dir->i_nlink) {
 686                 err = -ENOENT;
 687                 goto out;
 688         }
 689
 690         err = ocfs2_lookup_ino_from_name(dir, old_dentry->d_name.name,
 691                         old_dentry->d_name.len, &old_de_ino);
 692         if (err) {
 693                 err = -ENOENT;
 694                 goto out;
 695         }
 696
 697         /*
 698          * Check whether another node removed the source inode while we
 699          * were in the vfs.
 700          */
 701         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 702                 err = -ENOENT;
 703                 goto out;
 704         }
 705
 706         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 707                                         dentry->d_name.len);
 708         if (err)
 709                 goto out;
 710
 711         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 712                                            dentry->d_name.name,
 713                                            dentry->d_name.len, &lookup);
 714         if (err < 0) {
 715                 mlog_errno(err);
 716                 goto out;
 717         }
 718
 719         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 720         if (err < 0) {
 721                 if (err != -ENOENT)
 722                         mlog_errno(err);
 723                 goto out;
 724         }
 725
 726         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 727         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 728                 err = -EMLINK;
 729                 goto out_unlock_inode;
 730         }
 731
 732         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 733         if (IS_ERR(handle)) {
 734                 err = PTR_ERR(handle);
 735                 handle = NULL;
 736                 mlog_errno(err);
 737                 goto out_unlock_inode;
 738         }
 739
 740         /* Starting to change things, restart is no longer possible. */
 741         ocfs2_block_signals(&oldset);
 742
 743         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 744                                       OCFS2_JOURNAL_ACCESS_WRITE);
 745         if (err < 0) {
 746                 mlog_errno(err);
 747                 goto out_commit;
 748         }
 749
 750         inc_nlink(inode);
 751         inode->i_ctime = CURRENT_TIME;
 752         ocfs2_set_links_count(fe, inode->i_nlink);
 753         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 754         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 755         ocfs2_journal_dirty(handle, fe_bh);
 756
 757         err = ocfs2_add_entry(handle, dentry, inode,
 758                               OCFS2_I(inode)->ip_blkno,
 759                               parent_fe_bh, &lookup);
 760         if (err) {
 761                 ocfs2_add_links_count(fe, -1);
 762                 drop_nlink(inode);
 763                 mlog_errno(err);
 764                 goto out_commit;
 765         }
 766
 767         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 768         if (err) {
 769                 mlog_errno(err);
 770                 goto out_commit;
 771         }
 772
 773         ihold(inode);
 774         d_instantiate(dentry, inode);
 775
 776 out_commit:
 777         ocfs2_commit_trans(osb, handle);
 778         ocfs2_unblock_signals(&oldset);
 779 out_unlock_inode:
 780         ocfs2_inode_unlock(inode, 1);
 781
 782 out:
 783         ocfs2_inode_unlock(dir, 1);
 784
 785         brelse(fe_bh);
 786         brelse(parent_fe_bh);
 787
 788         ocfs2_free_dir_lookup_result(&lookup);
 789
 790         if (err)
 791                 mlog_errno(err);
 792
 793         return err;
 794 }
 795
 796 /*
 797  * Takes and drops an exclusive lock on the given dentry. This will
 798  * force other nodes to drop it.
 799  */
 800 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 801 {
 802         int ret;
 803
 804         ret = ocfs2_dentry_lock(dentry, 1);
 805         if (ret)
 806                 mlog_errno(ret);
 807         else
 808                 ocfs2_dentry_unlock(dentry, 1);
 809
 810         return ret;
 811 }
 812
 813 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 814 {
 815         if (S_ISDIR(inode->i_mode)) {
 816                 if (inode->i_nlink == 2)
 817                         return 1;
 818                 return 0;
 819         }
 820
 821         if (inode->i_nlink == 1)
 822                 return 1;
 823         return 0;
 824 }
 825
 826 static int ocfs2_unlink(struct inode *dir,
 827                         struct dentry *dentry)
 828 {
 829         int status;
 830         int child_locked = 0;
 831         bool is_unlinkable = false;
 832         struct inode *inode = dentry->d_inode;
 833         struct inode *orphan_dir = NULL;
 834         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 835         u64 blkno;
 836         struct ocfs2_dinode *fe = NULL;
 837         struct buffer_head *fe_bh = NULL;
 838         struct buffer_head *parent_node_bh = NULL;
 839         handle_t *handle = NULL;
 840         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 841         struct ocfs2_dir_lookup_result lookup = { NULL, };
 842         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 843
 844         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 845                            dentry->d_name.name,
 846                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 847                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 848
 849         dquot_initialize(dir);
 850
 851         BUG_ON(dentry->d_parent->d_inode != dir);
 852
 853         if (inode == osb->root_inode)
 854                 return -EPERM;
 855
 856         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 857                                          OI_LS_PARENT);
 858         if (status < 0) {
 859                 if (status != -ENOENT)
 860                         mlog_errno(status);
 861                 return status;
 862         }
 863
 864         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 865                                           dentry->d_name.len, &blkno, dir,
 866                                           &lookup);
 867         if (status < 0) {
 868                 if (status != -ENOENT)
 869                         mlog_errno(status);
 870                 goto leave;
 871         }
 872
 873         if (OCFS2_I(inode)->ip_blkno != blkno) {
 874                 status = -ENOENT;
 875
 876                 trace_ocfs2_unlink_noent(
 877                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 878                                 (unsigned long long)blkno,
 879                                 OCFS2_I(inode)->ip_flags);
 880                 goto leave;
 881         }
 882
 883         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 884         if (status < 0) {
 885                 if (status != -ENOENT)
 886                         mlog_errno(status);
 887                 goto leave;
 888         }
 889         child_locked = 1;
 890
 891         if (S_ISDIR(inode->i_mode)) {
 892                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 893                         status = -ENOTEMPTY;
 894                         goto leave;
 895                 }
 896         }
 897
 898         status = ocfs2_remote_dentry_delete(dentry);
 899         if (status < 0) {
 900                 /* This remote delete should succeed under all normal
 901                  * circumstances. */
 902                 mlog_errno(status);
 903                 goto leave;
 904         }
 905
 906         if (ocfs2_inode_is_unlinkable(inode)) {
 907                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 908                                                   OCFS2_I(inode)->ip_blkno,
 909                                                   orphan_name, &orphan_insert);
 910                 if (status < 0) {
 911                         mlog_errno(status);
 912                         goto leave;
 913                 }
 914                 is_unlinkable = true;
 915         }
 916
 917         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 918         if (IS_ERR(handle)) {
 919                 status = PTR_ERR(handle);
 920                 handle = NULL;
 921                 mlog_errno(status);
 922                 goto leave;
 923         }
 924
 925         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 926                                          OCFS2_JOURNAL_ACCESS_WRITE);
 927         if (status < 0) {
 928                 mlog_errno(status);
 929                 goto leave;
 930         }
 931
 932         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 933
 934         /* delete the name from the parent dir */
 935         status = ocfs2_delete_entry(handle, dir, &lookup);
 936         if (status < 0) {
 937                 mlog_errno(status);
 938                 goto leave;
 939         }
 940
 941         if (S_ISDIR(inode->i_mode))
 942                 drop_nlink(inode);
 943         drop_nlink(inode);
 944         ocfs2_set_links_count(fe, inode->i_nlink);
 945         ocfs2_journal_dirty(handle, fe_bh);
 946
 947         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 948         if (S_ISDIR(inode->i_mode))
 949                 drop_nlink(dir);
 950
 951         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 952         if (status < 0) {
 953                 mlog_errno(status);
 954                 if (S_ISDIR(inode->i_mode))
 955                         inc_nlink(dir);
 956                 goto leave;
 957         }
 958
 959         if (is_unlinkable) {
 960                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
 961                                 orphan_name, &orphan_insert, orphan_dir);
 962                 if (status < 0)
 963                         mlog_errno(status);
 964         }
 965
 966 leave:
 967         if (handle)
 968                 ocfs2_commit_trans(osb, handle);
 969
 970         if (child_locked)
 971                 ocfs2_inode_unlock(inode, 1);
 972
 973         ocfs2_inode_unlock(dir, 1);
 974
 975         if (orphan_dir) {
 976                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 977                 ocfs2_inode_unlock(orphan_dir, 1);
 978                 mutex_unlock(&orphan_dir->i_mutex);
 979                 iput(orphan_dir);
 980         }
 981
 982         brelse(fe_bh);
 983         brelse(parent_node_bh);
 984
 985         ocfs2_free_dir_lookup_result(&orphan_insert);
 986         ocfs2_free_dir_lookup_result(&lookup);
 987
 988         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
 989                 mlog_errno(status);
 990
 991         return status;
 992 }
 993
 994 /*
 995  * The only place this should be used is rename!
 996  * if they have the same id, then the 1st one is the only one locked.
 997  */
 998 static int ocfs2_double_lock(struct ocfs2_super *osb,
 999                              struct buffer_head **bh1,
1000                              struct inode *inode1,
1001                              struct buffer_head **bh2,
1002                              struct inode *inode2)
1003 {
1004         int status;
1005         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1006         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1007         struct buffer_head **tmpbh;
1008         struct inode *tmpinode;
1009
1010         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1011                                 (unsigned long long)oi2->ip_blkno);
1012
1013         if (*bh1)
1014                 *bh1 = NULL;
1015         if (*bh2)
1016                 *bh2 = NULL;
1017
1018         /* we always want to lock the one with the lower lockid first. */
1019         if (oi1->ip_blkno != oi2->ip_blkno) {
1020                 if (oi1->ip_blkno < oi2->ip_blkno) {
1021                         /* switch id1 and id2 around */
1022                         tmpbh = bh2;
1023                         bh2 = bh1;
1024                         bh1 = tmpbh;
1025
1026                         tmpinode = inode2;
1027                         inode2 = inode1;
1028                         inode1 = tmpinode;
1029                 }
1030                 /* lock id2 */
1031                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1032                                                  OI_LS_RENAME1);
1033                 if (status < 0) {
1034                         if (status != -ENOENT)
1035                                 mlog_errno(status);
1036                         goto bail;
1037                 }
1038         }
1039
1040         /* lock id1 */
1041         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1042         if (status < 0) {
1043                 /*
1044                  * An error return must mean that no cluster locks
1045                  * were held on function exit.
1046                  */
1047                 if (oi1->ip_blkno != oi2->ip_blkno) {
1048                         ocfs2_inode_unlock(inode2, 1);
1049                         brelse(*bh2);
1050                         *bh2 = NULL;
1051                 }
1052
1053                 if (status != -ENOENT)
1054                         mlog_errno(status);
1055         }
1056
1057         trace_ocfs2_double_lock_end(
1058                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1059                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1060
1061 bail:
1062         if (status)
1063                 mlog_errno(status);
1064         return status;
1065 }
1066
1067 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1068 {
1069         ocfs2_inode_unlock(inode1, 1);
1070
1071         if (inode1 != inode2)
1072                 ocfs2_inode_unlock(inode2, 1);
1073 }
1074
1075 static int ocfs2_rename(struct inode *old_dir,
1076                         struct dentry *old_dentry,
1077                         struct inode *new_dir,
1078                         struct dentry *new_dentry)
1079 {
1080         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1081         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1082         struct inode *old_inode = old_dentry->d_inode;
1083         struct inode *new_inode = new_dentry->d_inode;
1084         struct inode *orphan_dir = NULL;
1085         struct ocfs2_dinode *newfe = NULL;
1086         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1087         struct buffer_head *newfe_bh = NULL;
1088         struct buffer_head *old_inode_bh = NULL;
1089         struct ocfs2_super *osb = NULL;
1090         u64 newfe_blkno, old_de_ino;
1091         handle_t *handle = NULL;
1092         struct buffer_head *old_dir_bh = NULL;
1093         struct buffer_head *new_dir_bh = NULL;
1094         u32 old_dir_nlink = old_dir->i_nlink;
1095         struct ocfs2_dinode *old_di;
1096         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1097         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1098         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1099         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1100         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1101
1102         /* At some point it might be nice to break this function up a
1103          * bit. */
1104
1105         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1106                            old_dentry->d_name.len, old_dentry->d_name.name,
1107                            new_dentry->d_name.len, new_dentry->d_name.name);
1108
1109         dquot_initialize(old_dir);
1110         dquot_initialize(new_dir);
1111
1112         osb = OCFS2_SB(old_dir->i_sb);
1113
1114         if (new_inode) {
1115                 if (!igrab(new_inode))
1116                         BUG();
1117         }
1118
1119         /* Assume a directory hierarchy thusly:
1120          * a/b/c
1121          * a/d
1122          * a,b,c, and d are all directories.
1123          *
1124          * from cwd of 'a' on both nodes:
1125          * node1: mv b/c d
1126          * node2: mv d   b/c
1127          *
1128          * And that's why, just like the VFS, we need a file system
1129          * rename lock. */
1130         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1131                 status = ocfs2_rename_lock(osb);
1132                 if (status < 0) {
1133                         mlog_errno(status);
1134                         goto bail;
1135                 }
1136                 rename_lock = 1;
1137         }
1138
1139         /* if old and new are the same, this'll just do one lock. */
1140         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1141                                    &new_dir_bh, new_dir);
1142         if (status < 0) {
1143                 mlog_errno(status);
1144                 goto bail;
1145         }
1146         parents_locked = 1;
1147
1148         /* make sure both dirs have bhs
1149          * get an extra ref on old_dir_bh if old==new */
1150         if (!new_dir_bh) {
1151                 if (old_dir_bh) {
1152                         new_dir_bh = old_dir_bh;
1153                         get_bh(new_dir_bh);
1154                 } else {
1155                         mlog(ML_ERROR, "no old_dir_bh!\n");
1156                         status = -EIO;
1157                         goto bail;
1158                 }
1159         }
1160
1161         /*
1162          * Aside from allowing a meta data update, the locking here
1163          * also ensures that the downconvert thread on other nodes
1164          * won't have to concurrently downconvert the inode and the
1165          * dentry locks.
1166          */
1167         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1168                                          OI_LS_PARENT);
1169         if (status < 0) {
1170                 if (status != -ENOENT)
1171                         mlog_errno(status);
1172                 goto bail;
1173         }
1174         old_child_locked = 1;
1175
1176         status = ocfs2_remote_dentry_delete(old_dentry);
1177         if (status < 0) {
1178                 mlog_errno(status);
1179                 goto bail;
1180         }
1181
1182         if (S_ISDIR(old_inode->i_mode)) {
1183                 u64 old_inode_parent;
1184
1185                 update_dot_dot = 1;
1186                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1187                                                   old_inode,
1188                                                   &old_inode_dot_dot_res);
1189                 if (status) {
1190                         status = -EIO;
1191                         goto bail;
1192                 }
1193
1194                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1195                         status = -EIO;
1196                         goto bail;
1197                 }
1198
1199                 if (!new_inode && new_dir != old_dir &&
1200                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1201                         status = -EMLINK;
1202                         goto bail;
1203                 }
1204         }
1205
1206         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1207                                             old_dentry->d_name.len,
1208                                             &old_de_ino);
1209         if (status) {
1210                 status = -ENOENT;
1211                 goto bail;
1212         }
1213
1214         /*
1215          *  Check for inode number is _not_ due to possible IO errors.
1216          *  We might rmdir the source, keep it as pwd of some process
1217          *  and merrily kill the link to whatever was created under the
1218          *  same name. Goodbye sticky bit ;-<
1219          */
1220         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1221                 status = -ENOENT;
1222                 goto bail;
1223         }
1224
1225         /* check if the target already exists (in which case we need
1226          * to delete it */
1227         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1228                                           new_dentry->d_name.len,
1229                                           &newfe_blkno, new_dir,
1230                                           &target_lookup_res);
1231         /* The only error we allow here is -ENOENT because the new
1232          * file not existing is perfectly valid. */
1233         if ((status < 0) && (status != -ENOENT)) {
1234                 /* If we cannot find the file specified we should just */
1235                 /* return the error... */
1236                 mlog_errno(status);
1237                 goto bail;
1238         }
1239         if (status == 0)
1240                 target_exists = 1;
1241
1242         if (!target_exists && new_inode) {
1243                 /*
1244                  * Target was unlinked by another node while we were
1245                  * waiting to get to ocfs2_rename(). There isn't
1246                  * anything we can do here to help the situation, so
1247                  * bubble up the appropriate error.
1248                  */
1249                 status = -ENOENT;
1250                 goto bail;
1251         }
1252
1253         /* In case we need to overwrite an existing file, we blow it
1254          * away first */
1255         if (target_exists) {
1256                 /* VFS didn't think there existed an inode here, but
1257                  * someone else in the cluster must have raced our
1258                  * rename to create one. Today we error cleanly, in
1259                  * the future we should consider calling iget to build
1260                  * a new struct inode for this entry. */
1261                 if (!new_inode) {
1262                         status = -EACCES;
1263
1264                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1265                                                 new_dentry->d_name.name);
1266                         goto bail;
1267                 }
1268
1269                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1270                         status = -EACCES;
1271
1272                         trace_ocfs2_rename_disagree(
1273                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1274                              (unsigned long long)newfe_blkno,
1275                              OCFS2_I(new_inode)->ip_flags);
1276                         goto bail;
1277                 }
1278
1279                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1280                 if (status < 0) {
1281                         if (status != -ENOENT)
1282                                 mlog_errno(status);
1283                         goto bail;
1284                 }
1285                 new_child_locked = 1;
1286
1287                 status = ocfs2_remote_dentry_delete(new_dentry);
1288                 if (status < 0) {
1289                         mlog_errno(status);
1290                         goto bail;
1291                 }
1292
1293                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1294
1295                 trace_ocfs2_rename_over_existing(
1296                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1297                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1298
1299                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1300                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1301                                                 OCFS2_I(new_inode)->ip_blkno,
1302                                                 orphan_name, &orphan_insert);
1303                         if (status < 0) {
1304                                 mlog_errno(status);
1305                                 goto bail;
1306                         }
1307                 }
1308         } else {
1309                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1310
1311                 status = ocfs2_check_dir_for_entry(new_dir,
1312                                                    new_dentry->d_name.name,
1313                                                    new_dentry->d_name.len);
1314                 if (status)
1315                         goto bail;
1316
1317                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1318                                                       new_dentry->d_name.name,
1319                                                       new_dentry->d_name.len,
1320                                                       &target_insert);
1321                 if (status < 0) {
1322                         mlog_errno(status);
1323                         goto bail;
1324                 }
1325         }
1326
1327         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1328         if (IS_ERR(handle)) {
1329                 status = PTR_ERR(handle);
1330                 handle = NULL;
1331                 mlog_errno(status);
1332                 goto bail;
1333         }
1334
1335         if (target_exists) {
1336                 if (S_ISDIR(new_inode->i_mode)) {
1337                         if (new_inode->i_nlink != 2 ||
1338                             !ocfs2_empty_dir(new_inode)) {
1339                                 status = -ENOTEMPTY;
1340                                 goto bail;
1341                         }
1342                 }
1343                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1344                                                  newfe_bh,
1345                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1346                 if (status < 0) {
1347                         mlog_errno(status);
1348                         goto bail;
1349                 }
1350
1351                 if (S_ISDIR(new_inode->i_mode) ||
1352                     (ocfs2_read_links_count(newfe) == 1)) {
1353                         status = ocfs2_orphan_add(osb, handle, new_inode,
1354                                                   newfe_bh, orphan_name,
1355                                                   &orphan_insert, orphan_dir);
1356                         if (status < 0) {
1357                                 mlog_errno(status);
1358                                 goto bail;
1359                         }
1360                 }
1361
1362                 /* change the dirent to point to the correct inode */
1363                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1364                                             old_inode);
1365                 if (status < 0) {
1366                         mlog_errno(status);
1367                         goto bail;
1368                 }
1369                 new_dir->i_version++;
1370
1371                 if (S_ISDIR(new_inode->i_mode))
1372                         ocfs2_set_links_count(newfe, 0);
1373                 else
1374                         ocfs2_add_links_count(newfe, -1);
1375                 ocfs2_journal_dirty(handle, newfe_bh);
1376         } else {
1377                 /* if the name was not found in new_dir, add it now */
1378                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1379                                          OCFS2_I(old_inode)->ip_blkno,
1380                                          new_dir_bh, &target_insert);
1381         }
1382
1383         old_inode->i_ctime = CURRENT_TIME;
1384         mark_inode_dirty(old_inode);
1385
1386         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1387                                          old_inode_bh,
1388                                          OCFS2_JOURNAL_ACCESS_WRITE);
1389         if (status >= 0) {
1390                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1391
1392                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1393                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1394                 ocfs2_journal_dirty(handle, old_inode_bh);
1395         } else
1396                 mlog_errno(status);
1397
1398         /*
1399          * Now that the name has been added to new_dir, remove the old name.
1400          *
1401          * We don't keep any directory entry context around until now
1402          * because the insert might have changed the type of directory
1403          * we're dealing with.
1404          */
1405         status = ocfs2_find_entry(old_dentry->d_name.name,
1406                                   old_dentry->d_name.len, old_dir,
1407                                   &old_entry_lookup);
1408         if (status)
1409                 goto bail;
1410
1411         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1412         if (status < 0) {
1413                 mlog_errno(status);
1414                 goto bail;
1415         }
1416
1417         if (new_inode) {
1418                 drop_nlink(new_inode);
1419                 new_inode->i_ctime = CURRENT_TIME;
1420         }
1421         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1422
1423         if (update_dot_dot) {
1424                 status = ocfs2_update_entry(old_inode, handle,
1425                                             &old_inode_dot_dot_res, new_dir);
1426                 drop_nlink(old_dir);
1427                 if (new_inode) {
1428                         drop_nlink(new_inode);
1429                 } else {
1430                         inc_nlink(new_dir);
1431                         mark_inode_dirty(new_dir);
1432                 }
1433         }
1434         mark_inode_dirty(old_dir);
1435         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1436         if (new_inode) {
1437                 mark_inode_dirty(new_inode);
1438                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1439         }
1440
1441         if (old_dir != new_dir) {
1442                 /* Keep the same times on both directories.*/
1443                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1444
1445                 /*
1446                  * This will also pick up the i_nlink change from the
1447                  * block above.
1448                  */
1449                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1450         }
1451
1452         if (old_dir_nlink != old_dir->i_nlink) {
1453                 if (!old_dir_bh) {
1454                         mlog(ML_ERROR, "need to change nlink for old dir "
1455                              "%llu from %d to %d but bh is NULL!\n",
1456                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1457                              (int)old_dir_nlink, old_dir->i_nlink);
1458                 } else {
1459                         struct ocfs2_dinode *fe;
1460                         status = ocfs2_journal_access_di(handle,
1461                                                          INODE_CACHE(old_dir),
1462                                                          old_dir_bh,
1463                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1464                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1465                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1466                         ocfs2_journal_dirty(handle, old_dir_bh);
1467                 }
1468         }
1469         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1470         status = 0;
1471 bail:
1472         if (rename_lock)
1473                 ocfs2_rename_unlock(osb);
1474
1475         if (handle)
1476                 ocfs2_commit_trans(osb, handle);
1477
1478         if (parents_locked)
1479                 ocfs2_double_unlock(old_dir, new_dir);
1480
1481         if (old_child_locked)
1482                 ocfs2_inode_unlock(old_inode, 1);
1483
1484         if (new_child_locked)
1485                 ocfs2_inode_unlock(new_inode, 1);
1486
1487         if (orphan_dir) {
1488                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1489                 ocfs2_inode_unlock(orphan_dir, 1);
1490                 mutex_unlock(&orphan_dir->i_mutex);
1491                 iput(orphan_dir);
1492         }
1493
1494         if (new_inode)
1495                 sync_mapping_buffers(old_inode->i_mapping);
1496
1497         if (new_inode)
1498                 iput(new_inode);
1499
1500         ocfs2_free_dir_lookup_result(&target_lookup_res);
1501         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1502         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1503         ocfs2_free_dir_lookup_result(&orphan_insert);
1504         ocfs2_free_dir_lookup_result(&target_insert);
1505
1506         brelse(newfe_bh);
1507         brelse(old_inode_bh);
1508         brelse(old_dir_bh);
1509         brelse(new_dir_bh);
1510
1511         if (status)
1512                 mlog_errno(status);
1513
1514         return status;
1515 }
1516
1517 /*
1518  * we expect i_size = strlen(symname). Copy symname into the file
1519  * data, including the null terminator.
1520  */
1521 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1522                                      handle_t *handle,
1523                                      struct inode *inode,
1524                                      const char *symname)
1525 {
1526         struct buffer_head **bhs = NULL;
1527         const char *c;
1528         struct super_block *sb = osb->sb;
1529         u64 p_blkno, p_blocks;
1530         int virtual, blocks, status, i, bytes_left;
1531
1532         bytes_left = i_size_read(inode) + 1;
1533         /* we can't trust i_blocks because we're actually going to
1534          * write i_size + 1 bytes. */
1535         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1536
1537         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1538                                         i_size_read(inode), blocks);
1539
1540         /* Sanity check -- make sure we're going to fit. */
1541         if (bytes_left >
1542             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1543                 status = -EIO;
1544                 mlog_errno(status);
1545                 goto bail;
1546         }
1547
1548         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1549         if (!bhs) {
1550                 status = -ENOMEM;
1551                 mlog_errno(status);
1552                 goto bail;
1553         }
1554
1555         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1556                                              NULL);
1557         if (status < 0) {
1558                 mlog_errno(status);
1559                 goto bail;
1560         }
1561
1562         /* links can never be larger than one cluster so we know this
1563          * is all going to be contiguous, but do a sanity check
1564          * anyway. */
1565         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1566                 status = -EIO;
1567                 mlog_errno(status);
1568                 goto bail;
1569         }
1570
1571         virtual = 0;
1572         while(bytes_left > 0) {
1573                 c = &symname[virtual * sb->s_blocksize];
1574
1575                 bhs[virtual] = sb_getblk(sb, p_blkno);
1576                 if (!bhs[virtual]) {
1577                         status = -ENOMEM;
1578                         mlog_errno(status);
1579                         goto bail;
1580                 }
1581                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1582                                               bhs[virtual]);
1583
1584                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1585                                               bhs[virtual],
1586                                               OCFS2_JOURNAL_ACCESS_CREATE);
1587                 if (status < 0) {
1588                         mlog_errno(status);
1589                         goto bail;
1590                 }
1591
1592                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1593
1594                 memcpy(bhs[virtual]->b_data, c,
1595                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1596                        bytes_left);
1597
1598                 ocfs2_journal_dirty(handle, bhs[virtual]);
1599
1600                 virtual++;
1601                 p_blkno++;
1602                 bytes_left -= sb->s_blocksize;
1603         }
1604
1605         status = 0;
1606 bail:
1607
1608         if (bhs) {
1609                 for(i = 0; i < blocks; i++)
1610                         brelse(bhs[i]);
1611                 kfree(bhs);
1612         }
1613
1614         if (status)
1615                 mlog_errno(status);
1616         return status;
1617 }
1618
1619 static int ocfs2_symlink(struct inode *dir,
1620                          struct dentry *dentry,
1621                          const char *symname)
1622 {
1623         int status, l, credits;
1624         u64 newsize;
1625         struct ocfs2_super *osb = NULL;
1626         struct inode *inode = NULL;
1627         struct super_block *sb;
1628         struct buffer_head *new_fe_bh = NULL;
1629         struct buffer_head *parent_fe_bh = NULL;
1630         struct ocfs2_dinode *fe = NULL;
1631         struct ocfs2_dinode *dirfe;
1632         handle_t *handle = NULL;
1633         struct ocfs2_alloc_context *inode_ac = NULL;
1634         struct ocfs2_alloc_context *data_ac = NULL;
1635         struct ocfs2_alloc_context *xattr_ac = NULL;
1636         int want_clusters = 0;
1637         int xattr_credits = 0;
1638         struct ocfs2_security_xattr_info si = {
1639                 .enable = 1,
1640         };
1641         int did_quota = 0, did_quota_inode = 0;
1642         struct ocfs2_dir_lookup_result lookup = { NULL, };
1643         sigset_t oldset;
1644         int did_block_signals = 0;
1645
1646         trace_ocfs2_symlink_begin(dir, dentry, symname,
1647                                   dentry->d_name.len, dentry->d_name.name);
1648
1649         dquot_initialize(dir);
1650
1651         sb = dir->i_sb;
1652         osb = OCFS2_SB(sb);
1653
1654         l = strlen(symname) + 1;
1655
1656         credits = ocfs2_calc_symlink_credits(sb);
1657
1658         /* lock the parent directory */
1659         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1660         if (status < 0) {
1661                 if (status != -ENOENT)
1662                         mlog_errno(status);
1663                 return status;
1664         }
1665
1666         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1667         if (!ocfs2_read_links_count(dirfe)) {
1668                 /* can't make a file in a deleted directory. */
1669                 status = -ENOENT;
1670                 goto bail;
1671         }
1672
1673         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1674                                            dentry->d_name.len);
1675         if (status)
1676                 goto bail;
1677
1678         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1679                                               dentry->d_name.name,
1680                                               dentry->d_name.len, &lookup);
1681         if (status < 0) {
1682                 mlog_errno(status);
1683                 goto bail;
1684         }
1685
1686         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1687         if (status < 0) {
1688                 if (status != -ENOSPC)
1689                         mlog_errno(status);
1690                 goto bail;
1691         }
1692
1693         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1694         if (!inode) {
1695                 status = -ENOMEM;
1696                 mlog_errno(status);
1697                 goto bail;
1698         }
1699
1700         /* get security xattr */
1701         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1702         if (status) {
1703                 if (status == -EOPNOTSUPP)
1704                         si.enable = 0;
1705                 else {
1706                         mlog_errno(status);
1707                         goto bail;
1708                 }
1709         }
1710
1711         /* calculate meta data/clusters for setting security xattr */
1712         if (si.enable) {
1713                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1714                                                   &xattr_credits, &xattr_ac);
1715                 if (status < 0) {
1716                         mlog_errno(status);
1717                         goto bail;
1718                 }
1719         }
1720
1721         /* don't reserve bitmap space for fast symlinks. */
1722         if (l > ocfs2_fast_symlink_chars(sb))
1723                 want_clusters += 1;
1724
1725         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1726         if (status < 0) {
1727                 if (status != -ENOSPC)
1728                         mlog_errno(status);
1729                 goto bail;
1730         }
1731
1732         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1733         if (IS_ERR(handle)) {
1734                 status = PTR_ERR(handle);
1735                 handle = NULL;
1736                 mlog_errno(status);
1737                 goto bail;
1738         }
1739
1740         /* Starting to change things, restart is no longer possible. */
1741         ocfs2_block_signals(&oldset);
1742         did_block_signals = 1;
1743
1744         status = dquot_alloc_inode(inode);
1745         if (status)
1746                 goto bail;
1747         did_quota_inode = 1;
1748
1749         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1750                                    dentry->d_name.name,
1751                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1752                                    inode->i_mode);
1753
1754         status = ocfs2_mknod_locked(osb, dir, inode,
1755                                     0, &new_fe_bh, parent_fe_bh, handle,
1756                                     inode_ac);
1757         if (status < 0) {
1758                 mlog_errno(status);
1759                 goto bail;
1760         }
1761
1762         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1763         inode->i_rdev = 0;
1764         newsize = l - 1;
1765         inode->i_op = &ocfs2_symlink_inode_operations;
1766         if (l > ocfs2_fast_symlink_chars(sb)) {
1767                 u32 offset = 0;
1768
1769                 status = dquot_alloc_space_nodirty(inode,
1770                     ocfs2_clusters_to_bytes(osb->sb, 1));
1771                 if (status)
1772                         goto bail;
1773                 did_quota = 1;
1774                 inode->i_mapping->a_ops = &ocfs2_aops;
1775                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1776                                               new_fe_bh,
1777                                               handle, data_ac, NULL,
1778                                               NULL);
1779                 if (status < 0) {
1780                         if (status != -ENOSPC && status != -EINTR) {
1781                                 mlog(ML_ERROR,
1782                                      "Failed to extend file to %llu\n",
1783                                      (unsigned long long)newsize);
1784                                 mlog_errno(status);
1785                                 status = -ENOSPC;
1786                         }
1787                         goto bail;
1788                 }
1789                 i_size_write(inode, newsize);
1790                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1791         } else {
1792                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1793                 memcpy((char *) fe->id2.i_symlink, symname, l);
1794                 i_size_write(inode, newsize);
1795                 inode->i_blocks = 0;
1796         }
1797
1798         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1799         if (status < 0) {
1800                 mlog_errno(status);
1801                 goto bail;
1802         }
1803
1804         if (!ocfs2_inode_is_fast_symlink(inode)) {
1805                 status = ocfs2_create_symlink_data(osb, handle, inode,
1806                                                    symname);
1807                 if (status < 0) {
1808                         mlog_errno(status);
1809                         goto bail;
1810                 }
1811         }
1812
1813         if (si.enable) {
1814                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1815                                                  xattr_ac, data_ac);
1816                 if (status < 0) {
1817                         mlog_errno(status);
1818                         goto bail;
1819                 }
1820         }
1821
1822         /*
1823          * Do this before adding the entry to the directory. We add
1824          * also set d_op after success so that ->d_iput() will cleanup
1825          * the dentry lock even if ocfs2_add_entry() fails below.
1826          */
1827         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1828         if (status) {
1829                 mlog_errno(status);
1830                 goto bail;
1831         }
1832
1833         status = ocfs2_add_entry(handle, dentry, inode,
1834                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1835                                  &lookup);
1836         if (status < 0) {
1837                 mlog_errno(status);
1838                 goto bail;
1839         }
1840
1841         insert_inode_hash(inode);
1842         d_instantiate(dentry, inode);
1843 bail:
1844         if (status < 0 && did_quota)
1845                 dquot_free_space_nodirty(inode,
1846                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1847         if (status < 0 && did_quota_inode)
1848                 dquot_free_inode(inode);
1849         if (handle)
1850                 ocfs2_commit_trans(osb, handle);
1851
1852         ocfs2_inode_unlock(dir, 1);
1853         if (did_block_signals)
1854                 ocfs2_unblock_signals(&oldset);
1855
1856         brelse(new_fe_bh);
1857         brelse(parent_fe_bh);
1858         kfree(si.name);
1859         kfree(si.value);
1860         ocfs2_free_dir_lookup_result(&lookup);
1861         if (inode_ac)
1862                 ocfs2_free_alloc_context(inode_ac);
1863         if (data_ac)
1864                 ocfs2_free_alloc_context(data_ac);
1865         if (xattr_ac)
1866                 ocfs2_free_alloc_context(xattr_ac);
1867         if ((status < 0) && inode) {
1868                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1869                 clear_nlink(inode);
1870                 iput(inode);
1871         }
1872
1873         if (status)
1874                 mlog_errno(status);
1875
1876         return status;
1877 }
1878
1879 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1880 {
1881         int status, namelen;
1882
1883         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1884                            (long long)blkno);
1885         if (namelen <= 0) {
1886                 if (namelen)
1887                         status = namelen;
1888                 else
1889                         status = -EINVAL;
1890                 mlog_errno(status);
1891                 goto bail;
1892         }
1893         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1894                 status = -EINVAL;
1895                 mlog_errno(status);
1896                 goto bail;
1897         }
1898
1899         trace_ocfs2_blkno_stringify(blkno, name, namelen);
1900
1901         status = 0;
1902 bail:
1903         if (status < 0)
1904                 mlog_errno(status);
1905         return status;
1906 }
1907
1908 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1909                                         struct inode **ret_orphan_dir,
1910                                         struct buffer_head **ret_orphan_dir_bh)
1911 {
1912         struct inode *orphan_dir_inode;
1913         struct buffer_head *orphan_dir_bh = NULL;
1914         int ret = 0;
1915
1916         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1917                                                        ORPHAN_DIR_SYSTEM_INODE,
1918                                                        osb->slot_num);
1919         if (!orphan_dir_inode) {
1920                 ret = -ENOENT;
1921                 mlog_errno(ret);
1922                 return ret;
1923         }
1924
1925         mutex_lock(&orphan_dir_inode->i_mutex);
1926
1927         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1928         if (ret < 0) {
1929                 mutex_unlock(&orphan_dir_inode->i_mutex);
1930                 iput(orphan_dir_inode);
1931
1932                 mlog_errno(ret);
1933                 return ret;
1934         }
1935
1936         *ret_orphan_dir = orphan_dir_inode;
1937         *ret_orphan_dir_bh = orphan_dir_bh;
1938
1939         return 0;
1940 }
1941
1942 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1943                                       struct buffer_head *orphan_dir_bh,
1944                                       u64 blkno,
1945                                       char *name,
1946                                       struct ocfs2_dir_lookup_result *lookup)
1947 {
1948         int ret;
1949         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1950
1951         ret = ocfs2_blkno_stringify(blkno, name);
1952         if (ret < 0) {
1953                 mlog_errno(ret);
1954                 return ret;
1955         }
1956
1957         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1958                                            orphan_dir_bh, name,
1959                                            OCFS2_ORPHAN_NAMELEN, lookup);
1960         if (ret < 0) {
1961                 mlog_errno(ret);
1962                 return ret;
1963         }
1964
1965         return 0;
1966 }
1967
1968 /**
1969  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1970  * insertion of an orphan.
1971  * @osb: ocfs2 file system
1972  * @ret_orphan_dir: Orphan dir inode - returned locked!
1973  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1974  * @lookup: dir lookup result, to be passed back into functions like
1975  *          ocfs2_orphan_add
1976  *
1977  * Returns zero on success and the ret_orphan_dir, name and lookup
1978  * fields will be populated.
1979  *
1980  * Returns non-zero on failure.
1981  */
1982 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1983                                     struct inode **ret_orphan_dir,
1984                                     u64 blkno,
1985                                     char *name,
1986                                     struct ocfs2_dir_lookup_result *lookup)
1987 {
1988         struct inode *orphan_dir_inode = NULL;
1989         struct buffer_head *orphan_dir_bh = NULL;
1990         int ret = 0;
1991
1992         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1993                                            &orphan_dir_bh);
1994         if (ret < 0) {
1995                 mlog_errno(ret);
1996                 return ret;
1997         }
1998
1999         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2000                                          blkno, name, lookup);
2001         if (ret < 0) {
2002                 mlog_errno(ret);
2003                 goto out;
2004         }
2005
2006         *ret_orphan_dir = orphan_dir_inode;
2007
2008 out:
2009         brelse(orphan_dir_bh);
2010
2011         if (ret) {
2012                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2013                 mutex_unlock(&orphan_dir_inode->i_mutex);
2014                 iput(orphan_dir_inode);
2015         }
2016
2017         if (ret)
2018                 mlog_errno(ret);
2019         return ret;
2020 }
2021
2022 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2023                             handle_t *handle,
2024                             struct inode *inode,
2025                             struct buffer_head *fe_bh,
2026                             char *name,
2027                             struct ocfs2_dir_lookup_result *lookup,
2028                             struct inode *orphan_dir_inode)
2029 {
2030         struct buffer_head *orphan_dir_bh = NULL;
2031         int status = 0;
2032         struct ocfs2_dinode *orphan_fe;
2033         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2034
2035         trace_ocfs2_orphan_add_begin(
2036                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2037
2038         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2039         if (status < 0) {
2040                 mlog_errno(status);
2041                 goto leave;
2042         }
2043
2044         status = ocfs2_journal_access_di(handle,
2045                                          INODE_CACHE(orphan_dir_inode),
2046                                          orphan_dir_bh,
2047                                          OCFS2_JOURNAL_ACCESS_WRITE);
2048         if (status < 0) {
2049                 mlog_errno(status);
2050                 goto leave;
2051         }
2052
2053         /*
2054          * We're going to journal the change of i_flags and i_orphaned_slot.
2055          * It's safe anyway, though some callers may duplicate the journaling.
2056          * Journaling within the func just make the logic look more
2057          * straightforward.
2058          */
2059         status = ocfs2_journal_access_di(handle,
2060                                          INODE_CACHE(inode),
2061                                          fe_bh,
2062                                          OCFS2_JOURNAL_ACCESS_WRITE);
2063         if (status < 0) {
2064                 mlog_errno(status);
2065                 goto leave;
2066         }
2067
2068         /* we're a cluster, and nlink can change on disk from
2069          * underneath us... */
2070         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2071         if (S_ISDIR(inode->i_mode))
2072                 ocfs2_add_links_count(orphan_fe, 1);
2073         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2074         ocfs2_journal_dirty(handle, orphan_dir_bh);
2075
2076         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2077                                    OCFS2_ORPHAN_NAMELEN, inode,
2078                                    OCFS2_I(inode)->ip_blkno,
2079                                    orphan_dir_bh, lookup);
2080         if (status < 0) {
2081                 mlog_errno(status);
2082                 goto rollback;
2083         }
2084
2085         fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2086         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2087
2088         /* Record which orphan dir our inode now resides
2089          * in. delete_inode will use this to determine which orphan
2090          * dir to lock. */
2091         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2092
2093         ocfs2_journal_dirty(handle, fe_bh);
2094
2095         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2096                                    osb->slot_num);
2097
2098 rollback:
2099         if (status < 0) {
2100                 if (S_ISDIR(inode->i_mode))
2101                         ocfs2_add_links_count(orphan_fe, -1);
2102                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2103         }
2104
2105 leave:
2106         brelse(orphan_dir_bh);
2107
2108         return status;
2109 }
2110
2111 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2112 int ocfs2_orphan_del(struct ocfs2_super *osb,
2113                      handle_t *handle,
2114                      struct inode *orphan_dir_inode,
2115                      struct inode *inode,
2116                      struct buffer_head *orphan_dir_bh)
2117 {
2118         char name[OCFS2_ORPHAN_NAMELEN + 1];
2119         struct ocfs2_dinode *orphan_fe;
2120         int status = 0;
2121         struct ocfs2_dir_lookup_result lookup = { NULL, };
2122
2123         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2124         if (status < 0) {
2125                 mlog_errno(status);
2126                 goto leave;
2127         }
2128
2129         trace_ocfs2_orphan_del(
2130              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2131              name, OCFS2_ORPHAN_NAMELEN);
2132
2133         /* find it's spot in the orphan directory */
2134         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2135                                   &lookup);
2136         if (status) {
2137                 mlog_errno(status);
2138                 goto leave;
2139         }
2140
2141         /* remove it from the orphan directory */
2142         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2143         if (status < 0) {
2144                 mlog_errno(status);
2145                 goto leave;
2146         }
2147
2148         status = ocfs2_journal_access_di(handle,
2149                                          INODE_CACHE(orphan_dir_inode),
2150                                          orphan_dir_bh,
2151                                          OCFS2_JOURNAL_ACCESS_WRITE);
2152         if (status < 0) {
2153                 mlog_errno(status);
2154                 goto leave;
2155         }
2156
2157         /* do the i_nlink dance! :) */
2158         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2159         if (S_ISDIR(inode->i_mode))
2160                 ocfs2_add_links_count(orphan_fe, -1);
2161         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2162         ocfs2_journal_dirty(handle, orphan_dir_bh);
2163
2164 leave:
2165         ocfs2_free_dir_lookup_result(&lookup);
2166
2167         if (status)
2168                 mlog_errno(status);
2169         return status;
2170 }
2171
2172 /**
2173  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2174  * allocated file. This is different from the typical 'add to orphan dir'
2175  * operation in that the inode does not yet exist. This is a problem because
2176  * the orphan dir stringifies the inode block number to come up with it's
2177  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2178  * problem. This function works around it by calling deeper into the orphan
2179  * and suballoc code than other callers. Use this only by necessity.
2180  * @dir: The directory which this inode will ultimately wind up under - not the
2181  * orphan dir!
2182  * @dir_bh: buffer_head the @dir inode block
2183  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2184  * with the string to be used for orphan dirent. Pass back to the orphan dir
2185  * code.
2186  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2187  * dir code.
2188  * @ret_di_blkno: block number where the new inode will be allocated.
2189  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2190  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2191  *
2192  * Returns zero on success and the ret_orphan_dir, name and lookup
2193  * fields will be populated.
2194  *
2195  * Returns non-zero on failure.
2196  */
2197 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2198                                         struct buffer_head *dir_bh,
2199                                         char *orphan_name,
2200                                         struct inode **ret_orphan_dir,
2201                                         u64 *ret_di_blkno,
2202                                         struct ocfs2_dir_lookup_result *orphan_insert,
2203                                         struct ocfs2_alloc_context **ret_inode_ac)
2204 {
2205         int ret;
2206         u64 di_blkno;
2207         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2208         struct inode *orphan_dir = NULL;
2209         struct buffer_head *orphan_dir_bh = NULL;
2210         struct ocfs2_alloc_context *inode_ac = NULL;
2211
2212         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2213         if (ret < 0) {
2214                 mlog_errno(ret);
2215                 return ret;
2216         }
2217
2218         /* reserve an inode spot */
2219         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2220         if (ret < 0) {
2221                 if (ret != -ENOSPC)
2222                         mlog_errno(ret);
2223                 goto out;
2224         }
2225
2226         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2227                                        &di_blkno);
2228         if (ret) {
2229                 mlog_errno(ret);
2230                 goto out;
2231         }
2232
2233         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2234                                          di_blkno, orphan_name, orphan_insert);
2235         if (ret < 0) {
2236                 mlog_errno(ret);
2237                 goto out;
2238         }
2239
2240 out:
2241         if (ret == 0) {
2242                 *ret_orphan_dir = orphan_dir;
2243                 *ret_di_blkno = di_blkno;
2244                 *ret_inode_ac = inode_ac;
2245                 /*
2246                  * orphan_name and orphan_insert are already up to
2247                  * date via prepare_orphan_dir
2248                  */
2249         } else {
2250                 /* Unroll reserve_new_inode* */
2251                 if (inode_ac)
2252                         ocfs2_free_alloc_context(inode_ac);
2253
2254                 /* Unroll orphan dir locking */
2255                 mutex_unlock(&orphan_dir->i_mutex);
2256                 ocfs2_inode_unlock(orphan_dir, 1);
2257                 iput(orphan_dir);
2258         }
2259
2260         brelse(orphan_dir_bh);
2261
2262         return ret;
2263 }
2264
2265 int ocfs2_create_inode_in_orphan(struct inode *dir,
2266                                  int mode,
2267                                  struct inode **new_inode)
2268 {
2269         int status, did_quota_inode = 0;
2270         struct inode *inode = NULL;
2271         struct inode *orphan_dir = NULL;
2272         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2273         struct ocfs2_dinode *di = NULL;
2274         handle_t *handle = NULL;
2275         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2276         struct buffer_head *parent_di_bh = NULL;
2277         struct buffer_head *new_di_bh = NULL;
2278         struct ocfs2_alloc_context *inode_ac = NULL;
2279         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2280         u64 uninitialized_var(di_blkno), suballoc_loc;
2281         u16 suballoc_bit;
2282
2283         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2284         if (status < 0) {
2285                 if (status != -ENOENT)
2286                         mlog_errno(status);
2287                 return status;
2288         }
2289
2290         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2291                                               orphan_name, &orphan_dir,
2292                                               &di_blkno, &orphan_insert, &inode_ac);
2293         if (status < 0) {
2294                 if (status != -ENOSPC)
2295                         mlog_errno(status);
2296                 goto leave;
2297         }
2298
2299         inode = ocfs2_get_init_inode(dir, mode);
2300         if (!inode) {
2301                 status = -ENOMEM;
2302                 mlog_errno(status);
2303                 goto leave;
2304         }
2305
2306         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2307         if (IS_ERR(handle)) {
2308                 status = PTR_ERR(handle);
2309                 handle = NULL;
2310                 mlog_errno(status);
2311                 goto leave;
2312         }
2313
2314         status = dquot_alloc_inode(inode);
2315         if (status)
2316                 goto leave;
2317         did_quota_inode = 1;
2318
2319         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2320                                               &suballoc_loc,
2321                                               &suballoc_bit, di_blkno);
2322         if (status < 0) {
2323                 mlog_errno(status);
2324                 goto leave;
2325         }
2326
2327         clear_nlink(inode);
2328         /* do the real work now. */
2329         status = __ocfs2_mknod_locked(dir, inode,
2330                                       0, &new_di_bh, parent_di_bh, handle,
2331                                       inode_ac, di_blkno, suballoc_loc,
2332                                       suballoc_bit);
2333         if (status < 0) {
2334                 mlog_errno(status);
2335                 goto leave;
2336         }
2337
2338         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2339         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2340                                   &orphan_insert, orphan_dir);
2341         if (status < 0) {
2342                 mlog_errno(status);
2343                 goto leave;
2344         }
2345
2346         /* get open lock so that only nodes can't remove it from orphan dir. */
2347         status = ocfs2_open_lock(inode);
2348         if (status < 0)
2349                 mlog_errno(status);
2350
2351         insert_inode_hash(inode);
2352 leave:
2353         if (status < 0 && did_quota_inode)
2354                 dquot_free_inode(inode);
2355         if (handle)
2356                 ocfs2_commit_trans(osb, handle);
2357
2358         if (orphan_dir) {
2359                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2360                 ocfs2_inode_unlock(orphan_dir, 1);
2361                 mutex_unlock(&orphan_dir->i_mutex);
2362                 iput(orphan_dir);
2363         }
2364
2365         if ((status < 0) && inode) {
2366                 clear_nlink(inode);
2367                 iput(inode);
2368         }
2369
2370         if (inode_ac)
2371                 ocfs2_free_alloc_context(inode_ac);
2372
2373         brelse(new_di_bh);
2374
2375         if (!status)
2376                 *new_inode = inode;
2377
2378         ocfs2_free_dir_lookup_result(&orphan_insert);
2379
2380         ocfs2_inode_unlock(dir, 1);
2381         brelse(parent_di_bh);
2382         return status;
2383 }
2384
2385 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2386                                    struct inode *inode,
2387                                    struct dentry *dentry)
2388 {
2389         int status = 0;
2390         struct buffer_head *parent_di_bh = NULL;
2391         handle_t *handle = NULL;
2392         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2393         struct ocfs2_dinode *dir_di, *di;
2394         struct inode *orphan_dir_inode = NULL;
2395         struct buffer_head *orphan_dir_bh = NULL;
2396         struct buffer_head *di_bh = NULL;
2397         struct ocfs2_dir_lookup_result lookup = { NULL, };
2398
2399         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2400                                 dentry->d_name.len, dentry->d_name.name,
2401                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2402                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2403
2404         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2405         if (status < 0) {
2406                 if (status != -ENOENT)
2407                         mlog_errno(status);
2408                 return status;
2409         }
2410
2411         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2412         if (!dir_di->i_links_count) {
2413                 /* can't make a file in a deleted directory. */
2414                 status = -ENOENT;
2415                 goto leave;
2416         }
2417
2418         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2419                                            dentry->d_name.len);
2420         if (status)
2421                 goto leave;
2422
2423         /* get a spot inside the dir. */
2424         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2425                                               dentry->d_name.name,
2426                                               dentry->d_name.len, &lookup);
2427         if (status < 0) {
2428                 mlog_errno(status);
2429                 goto leave;
2430         }
2431
2432         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2433                                                        ORPHAN_DIR_SYSTEM_INODE,
2434                                                        osb->slot_num);
2435         if (!orphan_dir_inode) {
2436                 status = -EEXIST;
2437                 mlog_errno(status);
2438                 goto leave;
2439         }
2440
2441         mutex_lock(&orphan_dir_inode->i_mutex);
2442
2443         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2444         if (status < 0) {
2445                 mlog_errno(status);
2446                 mutex_unlock(&orphan_dir_inode->i_mutex);
2447                 iput(orphan_dir_inode);
2448                 goto leave;
2449         }
2450
2451         status = ocfs2_read_inode_block(inode, &di_bh);
2452         if (status < 0) {
2453                 mlog_errno(status);
2454                 goto orphan_unlock;
2455         }
2456
2457         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2458         if (IS_ERR(handle)) {
2459                 status = PTR_ERR(handle);
2460                 handle = NULL;
2461                 mlog_errno(status);
2462                 goto orphan_unlock;
2463         }
2464
2465         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2466                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2467         if (status < 0) {
2468                 mlog_errno(status);
2469                 goto out_commit;
2470         }
2471
2472         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2473                                   orphan_dir_bh);
2474         if (status < 0) {
2475                 mlog_errno(status);
2476                 goto out_commit;
2477         }
2478
2479         di = (struct ocfs2_dinode *)di_bh->b_data;
2480         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2481         di->i_orphaned_slot = 0;
2482         set_nlink(inode, 1);
2483         ocfs2_set_links_count(di, inode->i_nlink);
2484         ocfs2_journal_dirty(handle, di_bh);
2485
2486         status = ocfs2_add_entry(handle, dentry, inode,
2487                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2488                                  &lookup);
2489         if (status < 0) {
2490                 mlog_errno(status);
2491                 goto out_commit;
2492         }
2493
2494         status = ocfs2_dentry_attach_lock(dentry, inode,
2495                                           OCFS2_I(dir)->ip_blkno);
2496         if (status) {
2497                 mlog_errno(status);
2498                 goto out_commit;
2499         }
2500
2501         d_instantiate(dentry, inode);
2502         status = 0;
2503 out_commit:
2504         ocfs2_commit_trans(osb, handle);
2505 orphan_unlock:
2506         ocfs2_inode_unlock(orphan_dir_inode, 1);
2507         mutex_unlock(&orphan_dir_inode->i_mutex);
2508         iput(orphan_dir_inode);
2509 leave:
2510
2511         ocfs2_inode_unlock(dir, 1);
2512
2513         brelse(di_bh);
2514         brelse(parent_di_bh);
2515         brelse(orphan_dir_bh);
2516
2517         ocfs2_free_dir_lookup_result(&lookup);
2518
2519         if (status)
2520                 mlog_errno(status);
2521
2522         return status;
2523 }
2524
2525 const struct inode_operations ocfs2_dir_iops = {
2526         .create         = ocfs2_create,
2527         .lookup         = ocfs2_lookup,
2528         .link           = ocfs2_link,
2529         .unlink         = ocfs2_unlink,
2530         .rmdir          = ocfs2_unlink,
2531         .symlink        = ocfs2_symlink,
2532         .mkdir          = ocfs2_mkdir,
2533         .mknod          = ocfs2_mknod,
2534         .rename         = ocfs2_rename,
2535         .setattr        = ocfs2_setattr,
2536         .getattr        = ocfs2_getattr,
2537         .permission     = ocfs2_permission,
2538         .setxattr       = generic_setxattr,
2539         .getxattr       = generic_getxattr,
2540         .listxattr      = ocfs2_listxattr,
2541         .removexattr    = generic_removexattr,
2542         .fiemap         = ocfs2_fiemap,
2543         .get_acl        = ocfs2_iop_get_acl,
2544         .set_acl        = ocfs2_iop_set_acl,
2545 };