Pileus Git - ~andy/linux/blob - 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
 668         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 669                          old_dentry->d_name.len, old_dentry->d_name.name,
 670                          dentry->d_name.len, dentry->d_name.name);
 671
 672         if (S_ISDIR(inode->i_mode))
 673                 return -EPERM;
 674
 675         dquot_initialize(dir);
 676
 677         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 678         if (err < 0) {
 679                 if (err != -ENOENT)
 680                         mlog_errno(err);
 681                 return err;
 682         }
 683
 684         if (!dir->i_nlink) {
 685                 err = -ENOENT;
 686                 goto out;
 687         }
 688
 689         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 690                                         dentry->d_name.len);
 691         if (err)
 692                 goto out;
 693
 694         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 695                                            dentry->d_name.name,
 696                                            dentry->d_name.len, &lookup);
 697         if (err < 0) {
 698                 mlog_errno(err);
 699                 goto out;
 700         }
 701
 702         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 703         if (err < 0) {
 704                 if (err != -ENOENT)
 705                         mlog_errno(err);
 706                 goto out;
 707         }
 708
 709         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 710         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 711                 err = -EMLINK;
 712                 goto out_unlock_inode;
 713         }
 714
 715         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 716         if (IS_ERR(handle)) {
 717                 err = PTR_ERR(handle);
 718                 handle = NULL;
 719                 mlog_errno(err);
 720                 goto out_unlock_inode;
 721         }
 722
 723         /* Starting to change things, restart is no longer possible. */
 724         ocfs2_block_signals(&oldset);
 725
 726         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 727                                       OCFS2_JOURNAL_ACCESS_WRITE);
 728         if (err < 0) {
 729                 mlog_errno(err);
 730                 goto out_commit;
 731         }
 732
 733         inc_nlink(inode);
 734         inode->i_ctime = CURRENT_TIME;
 735         ocfs2_set_links_count(fe, inode->i_nlink);
 736         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 737         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 738         ocfs2_journal_dirty(handle, fe_bh);
 739
 740         err = ocfs2_add_entry(handle, dentry, inode,
 741                               OCFS2_I(inode)->ip_blkno,
 742                               parent_fe_bh, &lookup);
 743         if (err) {
 744                 ocfs2_add_links_count(fe, -1);
 745                 drop_nlink(inode);
 746                 mlog_errno(err);
 747                 goto out_commit;
 748         }
 749
 750         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 751         if (err) {
 752                 mlog_errno(err);
 753                 goto out_commit;
 754         }
 755
 756         ihold(inode);
 757         d_instantiate(dentry, inode);
 758
 759 out_commit:
 760         ocfs2_commit_trans(osb, handle);
 761         ocfs2_unblock_signals(&oldset);
 762 out_unlock_inode:
 763         ocfs2_inode_unlock(inode, 1);
 764
 765 out:
 766         ocfs2_inode_unlock(dir, 1);
 767
 768         brelse(fe_bh);
 769         brelse(parent_fe_bh);
 770
 771         ocfs2_free_dir_lookup_result(&lookup);
 772
 773         if (err)
 774                 mlog_errno(err);
 775
 776         return err;
 777 }
 778
 779 /*
 780  * Takes and drops an exclusive lock on the given dentry. This will
 781  * force other nodes to drop it.
 782  */
 783 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 784 {
 785         int ret;
 786
 787         ret = ocfs2_dentry_lock(dentry, 1);
 788         if (ret)
 789                 mlog_errno(ret);
 790         else
 791                 ocfs2_dentry_unlock(dentry, 1);
 792
 793         return ret;
 794 }
 795
 796 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 797 {
 798         if (S_ISDIR(inode->i_mode)) {
 799                 if (inode->i_nlink == 2)
 800                         return 1;
 801                 return 0;
 802         }
 803
 804         if (inode->i_nlink == 1)
 805                 return 1;
 806         return 0;
 807 }
 808
 809 static int ocfs2_unlink(struct inode *dir,
 810                         struct dentry *dentry)
 811 {
 812         int status;
 813         int child_locked = 0;
 814         bool is_unlinkable = false;
 815         struct inode *inode = dentry->d_inode;
 816         struct inode *orphan_dir = NULL;
 817         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 818         u64 blkno;
 819         struct ocfs2_dinode *fe = NULL;
 820         struct buffer_head *fe_bh = NULL;
 821         struct buffer_head *parent_node_bh = NULL;
 822         handle_t *handle = NULL;
 823         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 824         struct ocfs2_dir_lookup_result lookup = { NULL, };
 825         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 826
 827         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 828                            dentry->d_name.name,
 829                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 830                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 831
 832         dquot_initialize(dir);
 833
 834         BUG_ON(dentry->d_parent->d_inode != dir);
 835
 836         if (inode == osb->root_inode)
 837                 return -EPERM;
 838
 839         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 840                                          OI_LS_PARENT);
 841         if (status < 0) {
 842                 if (status != -ENOENT)
 843                         mlog_errno(status);
 844                 return status;
 845         }
 846
 847         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 848                                           dentry->d_name.len, &blkno, dir,
 849                                           &lookup);
 850         if (status < 0) {
 851                 if (status != -ENOENT)
 852                         mlog_errno(status);
 853                 goto leave;
 854         }
 855
 856         if (OCFS2_I(inode)->ip_blkno != blkno) {
 857                 status = -ENOENT;
 858
 859                 trace_ocfs2_unlink_noent(
 860                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 861                                 (unsigned long long)blkno,
 862                                 OCFS2_I(inode)->ip_flags);
 863                 goto leave;
 864         }
 865
 866         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 867         if (status < 0) {
 868                 if (status != -ENOENT)
 869                         mlog_errno(status);
 870                 goto leave;
 871         }
 872         child_locked = 1;
 873
 874         if (S_ISDIR(inode->i_mode)) {
 875                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 876                         status = -ENOTEMPTY;
 877                         goto leave;
 878                 }
 879         }
 880
 881         status = ocfs2_remote_dentry_delete(dentry);
 882         if (status < 0) {
 883                 /* This remote delete should succeed under all normal
 884                  * circumstances. */
 885                 mlog_errno(status);
 886                 goto leave;
 887         }
 888
 889         if (ocfs2_inode_is_unlinkable(inode)) {
 890                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 891                                                   OCFS2_I(inode)->ip_blkno,
 892                                                   orphan_name, &orphan_insert);
 893                 if (status < 0) {
 894                         mlog_errno(status);
 895                         goto leave;
 896                 }
 897                 is_unlinkable = true;
 898         }
 899
 900         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 901         if (IS_ERR(handle)) {
 902                 status = PTR_ERR(handle);
 903                 handle = NULL;
 904                 mlog_errno(status);
 905                 goto leave;
 906         }
 907
 908         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 909                                          OCFS2_JOURNAL_ACCESS_WRITE);
 910         if (status < 0) {
 911                 mlog_errno(status);
 912                 goto leave;
 913         }
 914
 915         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 916
 917         /* delete the name from the parent dir */
 918         status = ocfs2_delete_entry(handle, dir, &lookup);
 919         if (status < 0) {
 920                 mlog_errno(status);
 921                 goto leave;
 922         }
 923
 924         if (S_ISDIR(inode->i_mode))
 925                 drop_nlink(inode);
 926         drop_nlink(inode);
 927         ocfs2_set_links_count(fe, inode->i_nlink);
 928         ocfs2_journal_dirty(handle, fe_bh);
 929
 930         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 931         if (S_ISDIR(inode->i_mode))
 932                 drop_nlink(dir);
 933
 934         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 935         if (status < 0) {
 936                 mlog_errno(status);
 937                 if (S_ISDIR(inode->i_mode))
 938                         inc_nlink(dir);
 939                 goto leave;
 940         }
 941
 942         if (is_unlinkable) {
 943                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
 944                                 orphan_name, &orphan_insert, orphan_dir);
 945                 if (status < 0)
 946                         mlog_errno(status);
 947         }
 948
 949 leave:
 950         if (handle)
 951                 ocfs2_commit_trans(osb, handle);
 952
 953         if (child_locked)
 954                 ocfs2_inode_unlock(inode, 1);
 955
 956         ocfs2_inode_unlock(dir, 1);
 957
 958         if (orphan_dir) {
 959                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 960                 ocfs2_inode_unlock(orphan_dir, 1);
 961                 mutex_unlock(&orphan_dir->i_mutex);
 962                 iput(orphan_dir);
 963         }
 964
 965         brelse(fe_bh);
 966         brelse(parent_node_bh);
 967
 968         ocfs2_free_dir_lookup_result(&orphan_insert);
 969         ocfs2_free_dir_lookup_result(&lookup);
 970
 971         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
 972                 mlog_errno(status);
 973
 974         return status;
 975 }
 976
 977 /*
 978  * The only place this should be used is rename!
 979  * if they have the same id, then the 1st one is the only one locked.
 980  */
 981 static int ocfs2_double_lock(struct ocfs2_super *osb,
 982                              struct buffer_head **bh1,
 983                              struct inode *inode1,
 984                              struct buffer_head **bh2,
 985                              struct inode *inode2)
 986 {
 987         int status;
 988         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 989         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 990         struct buffer_head **tmpbh;
 991         struct inode *tmpinode;
 992
 993         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
 994                                 (unsigned long long)oi2->ip_blkno);
 995
 996         if (*bh1)
 997                 *bh1 = NULL;
 998         if (*bh2)
 999                 *bh2 = NULL;
1000
1001         /* we always want to lock the one with the lower lockid first. */
1002         if (oi1->ip_blkno != oi2->ip_blkno) {
1003                 if (oi1->ip_blkno < oi2->ip_blkno) {
1004                         /* switch id1 and id2 around */
1005                         tmpbh = bh2;
1006                         bh2 = bh1;
1007                         bh1 = tmpbh;
1008
1009                         tmpinode = inode2;
1010                         inode2 = inode1;
1011                         inode1 = tmpinode;
1012                 }
1013                 /* lock id2 */
1014                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1015                                                  OI_LS_RENAME1);
1016                 if (status < 0) {
1017                         if (status != -ENOENT)
1018                                 mlog_errno(status);
1019                         goto bail;
1020                 }
1021         }
1022
1023         /* lock id1 */
1024         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1025         if (status < 0) {
1026                 /*
1027                  * An error return must mean that no cluster locks
1028                  * were held on function exit.
1029                  */
1030                 if (oi1->ip_blkno != oi2->ip_blkno) {
1031                         ocfs2_inode_unlock(inode2, 1);
1032                         brelse(*bh2);
1033                         *bh2 = NULL;
1034                 }
1035
1036                 if (status != -ENOENT)
1037                         mlog_errno(status);
1038         }
1039
1040         trace_ocfs2_double_lock_end(
1041                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1042                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1043
1044 bail:
1045         if (status)
1046                 mlog_errno(status);
1047         return status;
1048 }
1049
1050 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1051 {
1052         ocfs2_inode_unlock(inode1, 1);
1053
1054         if (inode1 != inode2)
1055                 ocfs2_inode_unlock(inode2, 1);
1056 }
1057
1058 static int ocfs2_rename(struct inode *old_dir,
1059                         struct dentry *old_dentry,
1060                         struct inode *new_dir,
1061                         struct dentry *new_dentry)
1062 {
1063         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1064         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1065         struct inode *old_inode = old_dentry->d_inode;
1066         struct inode *new_inode = new_dentry->d_inode;
1067         struct inode *orphan_dir = NULL;
1068         struct ocfs2_dinode *newfe = NULL;
1069         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1070         struct buffer_head *newfe_bh = NULL;
1071         struct buffer_head *old_inode_bh = NULL;
1072         struct ocfs2_super *osb = NULL;
1073         u64 newfe_blkno, old_de_ino;
1074         handle_t *handle = NULL;
1075         struct buffer_head *old_dir_bh = NULL;
1076         struct buffer_head *new_dir_bh = NULL;
1077         u32 old_dir_nlink = old_dir->i_nlink;
1078         struct ocfs2_dinode *old_di;
1079         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1080         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1081         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1082         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1083         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1084
1085         /* At some point it might be nice to break this function up a
1086          * bit. */
1087
1088         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1089                            old_dentry->d_name.len, old_dentry->d_name.name,
1090                            new_dentry->d_name.len, new_dentry->d_name.name);
1091
1092         dquot_initialize(old_dir);
1093         dquot_initialize(new_dir);
1094
1095         osb = OCFS2_SB(old_dir->i_sb);
1096
1097         if (new_inode) {
1098                 if (!igrab(new_inode))
1099                         BUG();
1100         }
1101
1102         /* Assume a directory hierarchy thusly:
1103          * a/b/c
1104          * a/d
1105          * a,b,c, and d are all directories.
1106          *
1107          * from cwd of 'a' on both nodes:
1108          * node1: mv b/c d
1109          * node2: mv d   b/c
1110          *
1111          * And that's why, just like the VFS, we need a file system
1112          * rename lock. */
1113         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1114                 status = ocfs2_rename_lock(osb);
1115                 if (status < 0) {
1116                         mlog_errno(status);
1117                         goto bail;
1118                 }
1119                 rename_lock = 1;
1120         }
1121
1122         /* if old and new are the same, this'll just do one lock. */
1123         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1124                                    &new_dir_bh, new_dir);
1125         if (status < 0) {
1126                 mlog_errno(status);
1127                 goto bail;
1128         }
1129         parents_locked = 1;
1130
1131         /* make sure both dirs have bhs
1132          * get an extra ref on old_dir_bh if old==new */
1133         if (!new_dir_bh) {
1134                 if (old_dir_bh) {
1135                         new_dir_bh = old_dir_bh;
1136                         get_bh(new_dir_bh);
1137                 } else {
1138                         mlog(ML_ERROR, "no old_dir_bh!\n");
1139                         status = -EIO;
1140                         goto bail;
1141                 }
1142         }
1143
1144         /*
1145          * Aside from allowing a meta data update, the locking here
1146          * also ensures that the downconvert thread on other nodes
1147          * won't have to concurrently downconvert the inode and the
1148          * dentry locks.
1149          */
1150         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1151                                          OI_LS_PARENT);
1152         if (status < 0) {
1153                 if (status != -ENOENT)
1154                         mlog_errno(status);
1155                 goto bail;
1156         }
1157         old_child_locked = 1;
1158
1159         status = ocfs2_remote_dentry_delete(old_dentry);
1160         if (status < 0) {
1161                 mlog_errno(status);
1162                 goto bail;
1163         }
1164
1165         if (S_ISDIR(old_inode->i_mode)) {
1166                 u64 old_inode_parent;
1167
1168                 update_dot_dot = 1;
1169                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1170                                                   old_inode,
1171                                                   &old_inode_dot_dot_res);
1172                 if (status) {
1173                         status = -EIO;
1174                         goto bail;
1175                 }
1176
1177                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1178                         status = -EIO;
1179                         goto bail;
1180                 }
1181
1182                 if (!new_inode && new_dir != old_dir &&
1183                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1184                         status = -EMLINK;
1185                         goto bail;
1186                 }
1187         }
1188
1189         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1190                                             old_dentry->d_name.len,
1191                                             &old_de_ino);
1192         if (status) {
1193                 status = -ENOENT;
1194                 goto bail;
1195         }
1196
1197         /*
1198          *  Check for inode number is _not_ due to possible IO errors.
1199          *  We might rmdir the source, keep it as pwd of some process
1200          *  and merrily kill the link to whatever was created under the
1201          *  same name. Goodbye sticky bit ;-<
1202          */
1203         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1204                 status = -ENOENT;
1205                 goto bail;
1206         }
1207
1208         /* check if the target already exists (in which case we need
1209          * to delete it */
1210         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1211                                           new_dentry->d_name.len,
1212                                           &newfe_blkno, new_dir,
1213                                           &target_lookup_res);
1214         /* The only error we allow here is -ENOENT because the new
1215          * file not existing is perfectly valid. */
1216         if ((status < 0) && (status != -ENOENT)) {
1217                 /* If we cannot find the file specified we should just */
1218                 /* return the error... */
1219                 mlog_errno(status);
1220                 goto bail;
1221         }
1222         if (status == 0)
1223                 target_exists = 1;
1224
1225         if (!target_exists && new_inode) {
1226                 /*
1227                  * Target was unlinked by another node while we were
1228                  * waiting to get to ocfs2_rename(). There isn't
1229                  * anything we can do here to help the situation, so
1230                  * bubble up the appropriate error.
1231                  */
1232                 status = -ENOENT;
1233                 goto bail;
1234         }
1235
1236         /* In case we need to overwrite an existing file, we blow it
1237          * away first */
1238         if (target_exists) {
1239                 /* VFS didn't think there existed an inode here, but
1240                  * someone else in the cluster must have raced our
1241                  * rename to create one. Today we error cleanly, in
1242                  * the future we should consider calling iget to build
1243                  * a new struct inode for this entry. */
1244                 if (!new_inode) {
1245                         status = -EACCES;
1246
1247                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1248                                                 new_dentry->d_name.name);
1249                         goto bail;
1250                 }
1251
1252                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1253                         status = -EACCES;
1254
1255                         trace_ocfs2_rename_disagree(
1256                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1257                              (unsigned long long)newfe_blkno,
1258                              OCFS2_I(new_inode)->ip_flags);
1259                         goto bail;
1260                 }
1261
1262                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1263                 if (status < 0) {
1264                         if (status != -ENOENT)
1265                                 mlog_errno(status);
1266                         goto bail;
1267                 }
1268                 new_child_locked = 1;
1269
1270                 status = ocfs2_remote_dentry_delete(new_dentry);
1271                 if (status < 0) {
1272                         mlog_errno(status);
1273                         goto bail;
1274                 }
1275
1276                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1277
1278                 trace_ocfs2_rename_over_existing(
1279                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1280                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1281
1282                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1283                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1284                                                 OCFS2_I(new_inode)->ip_blkno,
1285                                                 orphan_name, &orphan_insert);
1286                         if (status < 0) {
1287                                 mlog_errno(status);
1288                                 goto bail;
1289                         }
1290                 }
1291         } else {
1292                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1293
1294                 status = ocfs2_check_dir_for_entry(new_dir,
1295                                                    new_dentry->d_name.name,
1296                                                    new_dentry->d_name.len);
1297                 if (status)
1298                         goto bail;
1299
1300                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1301                                                       new_dentry->d_name.name,
1302                                                       new_dentry->d_name.len,
1303                                                       &target_insert);
1304                 if (status < 0) {
1305                         mlog_errno(status);
1306                         goto bail;
1307                 }
1308         }
1309
1310         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1311         if (IS_ERR(handle)) {
1312                 status = PTR_ERR(handle);
1313                 handle = NULL;
1314                 mlog_errno(status);
1315                 goto bail;
1316         }
1317
1318         if (target_exists) {
1319                 if (S_ISDIR(new_inode->i_mode)) {
1320                         if (new_inode->i_nlink != 2 ||
1321                             !ocfs2_empty_dir(new_inode)) {
1322                                 status = -ENOTEMPTY;
1323                                 goto bail;
1324                         }
1325                 }
1326                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1327                                                  newfe_bh,
1328                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1329                 if (status < 0) {
1330                         mlog_errno(status);
1331                         goto bail;
1332                 }
1333
1334                 if (S_ISDIR(new_inode->i_mode) ||
1335                     (ocfs2_read_links_count(newfe) == 1)) {
1336                         status = ocfs2_orphan_add(osb, handle, new_inode,
1337                                                   newfe_bh, orphan_name,
1338                                                   &orphan_insert, orphan_dir);
1339                         if (status < 0) {
1340                                 mlog_errno(status);
1341                                 goto bail;
1342                         }
1343                 }
1344
1345                 /* change the dirent to point to the correct inode */
1346                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1347                                             old_inode);
1348                 if (status < 0) {
1349                         mlog_errno(status);
1350                         goto bail;
1351                 }
1352                 new_dir->i_version++;
1353
1354                 if (S_ISDIR(new_inode->i_mode))
1355                         ocfs2_set_links_count(newfe, 0);
1356                 else
1357                         ocfs2_add_links_count(newfe, -1);
1358                 ocfs2_journal_dirty(handle, newfe_bh);
1359         } else {
1360                 /* if the name was not found in new_dir, add it now */
1361                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1362                                          OCFS2_I(old_inode)->ip_blkno,
1363                                          new_dir_bh, &target_insert);
1364         }
1365
1366         old_inode->i_ctime = CURRENT_TIME;
1367         mark_inode_dirty(old_inode);
1368
1369         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1370                                          old_inode_bh,
1371                                          OCFS2_JOURNAL_ACCESS_WRITE);
1372         if (status >= 0) {
1373                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1374
1375                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1376                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1377                 ocfs2_journal_dirty(handle, old_inode_bh);
1378         } else
1379                 mlog_errno(status);
1380
1381         /*
1382          * Now that the name has been added to new_dir, remove the old name.
1383          *
1384          * We don't keep any directory entry context around until now
1385          * because the insert might have changed the type of directory
1386          * we're dealing with.
1387          */
1388         status = ocfs2_find_entry(old_dentry->d_name.name,
1389                                   old_dentry->d_name.len, old_dir,
1390                                   &old_entry_lookup);
1391         if (status)
1392                 goto bail;
1393
1394         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1395         if (status < 0) {
1396                 mlog_errno(status);
1397                 goto bail;
1398         }
1399
1400         if (new_inode) {
1401                 drop_nlink(new_inode);
1402                 new_inode->i_ctime = CURRENT_TIME;
1403         }
1404         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1405
1406         if (update_dot_dot) {
1407                 status = ocfs2_update_entry(old_inode, handle,
1408                                             &old_inode_dot_dot_res, new_dir);
1409                 drop_nlink(old_dir);
1410                 if (new_inode) {
1411                         drop_nlink(new_inode);
1412                 } else {
1413                         inc_nlink(new_dir);
1414                         mark_inode_dirty(new_dir);
1415                 }
1416         }
1417         mark_inode_dirty(old_dir);
1418         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1419         if (new_inode) {
1420                 mark_inode_dirty(new_inode);
1421                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1422         }
1423
1424         if (old_dir != new_dir) {
1425                 /* Keep the same times on both directories.*/
1426                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1427
1428                 /*
1429                  * This will also pick up the i_nlink change from the
1430                  * block above.
1431                  */
1432                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1433         }
1434
1435         if (old_dir_nlink != old_dir->i_nlink) {
1436                 if (!old_dir_bh) {
1437                         mlog(ML_ERROR, "need to change nlink for old dir "
1438                              "%llu from %d to %d but bh is NULL!\n",
1439                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1440                              (int)old_dir_nlink, old_dir->i_nlink);
1441                 } else {
1442                         struct ocfs2_dinode *fe;
1443                         status = ocfs2_journal_access_di(handle,
1444                                                          INODE_CACHE(old_dir),
1445                                                          old_dir_bh,
1446                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1447                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1448                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1449                         ocfs2_journal_dirty(handle, old_dir_bh);
1450                 }
1451         }
1452         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1453         status = 0;
1454 bail:
1455         if (rename_lock)
1456                 ocfs2_rename_unlock(osb);
1457
1458         if (handle)
1459                 ocfs2_commit_trans(osb, handle);
1460
1461         if (parents_locked)
1462                 ocfs2_double_unlock(old_dir, new_dir);
1463
1464         if (old_child_locked)
1465                 ocfs2_inode_unlock(old_inode, 1);
1466
1467         if (new_child_locked)
1468                 ocfs2_inode_unlock(new_inode, 1);
1469
1470         if (orphan_dir) {
1471                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1472                 ocfs2_inode_unlock(orphan_dir, 1);
1473                 mutex_unlock(&orphan_dir->i_mutex);
1474                 iput(orphan_dir);
1475         }
1476
1477         if (new_inode)
1478                 sync_mapping_buffers(old_inode->i_mapping);
1479
1480         if (new_inode)
1481                 iput(new_inode);
1482
1483         ocfs2_free_dir_lookup_result(&target_lookup_res);
1484         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1485         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1486         ocfs2_free_dir_lookup_result(&orphan_insert);
1487         ocfs2_free_dir_lookup_result(&target_insert);
1488
1489         brelse(newfe_bh);
1490         brelse(old_inode_bh);
1491         brelse(old_dir_bh);
1492         brelse(new_dir_bh);
1493
1494         if (status)
1495                 mlog_errno(status);
1496
1497         return status;
1498 }
1499
1500 /*
1501  * we expect i_size = strlen(symname). Copy symname into the file
1502  * data, including the null terminator.
1503  */
1504 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1505                                      handle_t *handle,
1506                                      struct inode *inode,
1507                                      const char *symname)
1508 {
1509         struct buffer_head **bhs = NULL;
1510         const char *c;
1511         struct super_block *sb = osb->sb;
1512         u64 p_blkno, p_blocks;
1513         int virtual, blocks, status, i, bytes_left;
1514
1515         bytes_left = i_size_read(inode) + 1;
1516         /* we can't trust i_blocks because we're actually going to
1517          * write i_size + 1 bytes. */
1518         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1519
1520         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1521                                         i_size_read(inode), blocks);
1522
1523         /* Sanity check -- make sure we're going to fit. */
1524         if (bytes_left >
1525             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1526                 status = -EIO;
1527                 mlog_errno(status);
1528                 goto bail;
1529         }
1530
1531         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1532         if (!bhs) {
1533                 status = -ENOMEM;
1534                 mlog_errno(status);
1535                 goto bail;
1536         }
1537
1538         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1539                                              NULL);
1540         if (status < 0) {
1541                 mlog_errno(status);
1542                 goto bail;
1543         }
1544
1545         /* links can never be larger than one cluster so we know this
1546          * is all going to be contiguous, but do a sanity check
1547          * anyway. */
1548         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1549                 status = -EIO;
1550                 mlog_errno(status);
1551                 goto bail;
1552         }
1553
1554         virtual = 0;
1555         while(bytes_left > 0) {
1556                 c = &symname[virtual * sb->s_blocksize];
1557
1558                 bhs[virtual] = sb_getblk(sb, p_blkno);
1559                 if (!bhs[virtual]) {
1560                         status = -ENOMEM;
1561                         mlog_errno(status);
1562                         goto bail;
1563                 }
1564                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1565                                               bhs[virtual]);
1566
1567                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1568                                               bhs[virtual],
1569                                               OCFS2_JOURNAL_ACCESS_CREATE);
1570                 if (status < 0) {
1571                         mlog_errno(status);
1572                         goto bail;
1573                 }
1574
1575                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1576
1577                 memcpy(bhs[virtual]->b_data, c,
1578                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1579                        bytes_left);
1580
1581                 ocfs2_journal_dirty(handle, bhs[virtual]);
1582
1583                 virtual++;
1584                 p_blkno++;
1585                 bytes_left -= sb->s_blocksize;
1586         }
1587
1588         status = 0;
1589 bail:
1590
1591         if (bhs) {
1592                 for(i = 0; i < blocks; i++)
1593                         brelse(bhs[i]);
1594                 kfree(bhs);
1595         }
1596
1597         if (status)
1598                 mlog_errno(status);
1599         return status;
1600 }
1601
1602 static int ocfs2_symlink(struct inode *dir,
1603                          struct dentry *dentry,
1604                          const char *symname)
1605 {
1606         int status, l, credits;
1607         u64 newsize;
1608         struct ocfs2_super *osb = NULL;
1609         struct inode *inode = NULL;
1610         struct super_block *sb;
1611         struct buffer_head *new_fe_bh = NULL;
1612         struct buffer_head *parent_fe_bh = NULL;
1613         struct ocfs2_dinode *fe = NULL;
1614         struct ocfs2_dinode *dirfe;
1615         handle_t *handle = NULL;
1616         struct ocfs2_alloc_context *inode_ac = NULL;
1617         struct ocfs2_alloc_context *data_ac = NULL;
1618         struct ocfs2_alloc_context *xattr_ac = NULL;
1619         int want_clusters = 0;
1620         int xattr_credits = 0;
1621         struct ocfs2_security_xattr_info si = {
1622                 .enable = 1,
1623         };
1624         int did_quota = 0, did_quota_inode = 0;
1625         struct ocfs2_dir_lookup_result lookup = { NULL, };
1626         sigset_t oldset;
1627         int did_block_signals = 0;
1628
1629         trace_ocfs2_symlink_begin(dir, dentry, symname,
1630                                   dentry->d_name.len, dentry->d_name.name);
1631
1632         dquot_initialize(dir);
1633
1634         sb = dir->i_sb;
1635         osb = OCFS2_SB(sb);
1636
1637         l = strlen(symname) + 1;
1638
1639         credits = ocfs2_calc_symlink_credits(sb);
1640
1641         /* lock the parent directory */
1642         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1643         if (status < 0) {
1644                 if (status != -ENOENT)
1645                         mlog_errno(status);
1646                 return status;
1647         }
1648
1649         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1650         if (!ocfs2_read_links_count(dirfe)) {
1651                 /* can't make a file in a deleted directory. */
1652                 status = -ENOENT;
1653                 goto bail;
1654         }
1655
1656         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1657                                            dentry->d_name.len);
1658         if (status)
1659                 goto bail;
1660
1661         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1662                                               dentry->d_name.name,
1663                                               dentry->d_name.len, &lookup);
1664         if (status < 0) {
1665                 mlog_errno(status);
1666                 goto bail;
1667         }
1668
1669         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1670         if (status < 0) {
1671                 if (status != -ENOSPC)
1672                         mlog_errno(status);
1673                 goto bail;
1674         }
1675
1676         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1677         if (!inode) {
1678                 status = -ENOMEM;
1679                 mlog_errno(status);
1680                 goto bail;
1681         }
1682
1683         /* get security xattr */
1684         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1685         if (status) {
1686                 if (status == -EOPNOTSUPP)
1687                         si.enable = 0;
1688                 else {
1689                         mlog_errno(status);
1690                         goto bail;
1691                 }
1692         }
1693
1694         /* calculate meta data/clusters for setting security xattr */
1695         if (si.enable) {
1696                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1697                                                   &xattr_credits, &xattr_ac);
1698                 if (status < 0) {
1699                         mlog_errno(status);
1700                         goto bail;
1701                 }
1702         }
1703
1704         /* don't reserve bitmap space for fast symlinks. */
1705         if (l > ocfs2_fast_symlink_chars(sb))
1706                 want_clusters += 1;
1707
1708         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1709         if (status < 0) {
1710                 if (status != -ENOSPC)
1711                         mlog_errno(status);
1712                 goto bail;
1713         }
1714
1715         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1716         if (IS_ERR(handle)) {
1717                 status = PTR_ERR(handle);
1718                 handle = NULL;
1719                 mlog_errno(status);
1720                 goto bail;
1721         }
1722
1723         /* Starting to change things, restart is no longer possible. */
1724         ocfs2_block_signals(&oldset);
1725         did_block_signals = 1;
1726
1727         status = dquot_alloc_inode(inode);
1728         if (status)
1729                 goto bail;
1730         did_quota_inode = 1;
1731
1732         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1733                                    dentry->d_name.name,
1734                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1735                                    inode->i_mode);
1736
1737         status = ocfs2_mknod_locked(osb, dir, inode,
1738                                     0, &new_fe_bh, parent_fe_bh, handle,
1739                                     inode_ac);
1740         if (status < 0) {
1741                 mlog_errno(status);
1742                 goto bail;
1743         }
1744
1745         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1746         inode->i_rdev = 0;
1747         newsize = l - 1;
1748         inode->i_op = &ocfs2_symlink_inode_operations;
1749         if (l > ocfs2_fast_symlink_chars(sb)) {
1750                 u32 offset = 0;
1751
1752                 status = dquot_alloc_space_nodirty(inode,
1753                     ocfs2_clusters_to_bytes(osb->sb, 1));
1754                 if (status)
1755                         goto bail;
1756                 did_quota = 1;
1757                 inode->i_mapping->a_ops = &ocfs2_aops;
1758                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1759                                               new_fe_bh,
1760                                               handle, data_ac, NULL,
1761                                               NULL);
1762                 if (status < 0) {
1763                         if (status != -ENOSPC && status != -EINTR) {
1764                                 mlog(ML_ERROR,
1765                                      "Failed to extend file to %llu\n",
1766                                      (unsigned long long)newsize);
1767                                 mlog_errno(status);
1768                                 status = -ENOSPC;
1769                         }
1770                         goto bail;
1771                 }
1772                 i_size_write(inode, newsize);
1773                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1774         } else {
1775                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1776                 memcpy((char *) fe->id2.i_symlink, symname, l);
1777                 i_size_write(inode, newsize);
1778                 inode->i_blocks = 0;
1779         }
1780
1781         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1782         if (status < 0) {
1783                 mlog_errno(status);
1784                 goto bail;
1785         }
1786
1787         if (!ocfs2_inode_is_fast_symlink(inode)) {
1788                 status = ocfs2_create_symlink_data(osb, handle, inode,
1789                                                    symname);
1790                 if (status < 0) {
1791                         mlog_errno(status);
1792                         goto bail;
1793                 }
1794         }
1795
1796         if (si.enable) {
1797                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1798                                                  xattr_ac, data_ac);
1799                 if (status < 0) {
1800                         mlog_errno(status);
1801                         goto bail;
1802                 }
1803         }
1804
1805         /*
1806          * Do this before adding the entry to the directory. We add
1807          * also set d_op after success so that ->d_iput() will cleanup
1808          * the dentry lock even if ocfs2_add_entry() fails below.
1809          */
1810         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1811         if (status) {
1812                 mlog_errno(status);
1813                 goto bail;
1814         }
1815
1816         status = ocfs2_add_entry(handle, dentry, inode,
1817                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1818                                  &lookup);
1819         if (status < 0) {
1820                 mlog_errno(status);
1821                 goto bail;
1822         }
1823
1824         insert_inode_hash(inode);
1825         d_instantiate(dentry, inode);
1826 bail:
1827         if (status < 0 && did_quota)
1828                 dquot_free_space_nodirty(inode,
1829                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1830         if (status < 0 && did_quota_inode)
1831                 dquot_free_inode(inode);
1832         if (handle)
1833                 ocfs2_commit_trans(osb, handle);
1834
1835         ocfs2_inode_unlock(dir, 1);
1836         if (did_block_signals)
1837                 ocfs2_unblock_signals(&oldset);
1838
1839         brelse(new_fe_bh);
1840         brelse(parent_fe_bh);
1841         kfree(si.name);
1842         kfree(si.value);
1843         ocfs2_free_dir_lookup_result(&lookup);
1844         if (inode_ac)
1845                 ocfs2_free_alloc_context(inode_ac);
1846         if (data_ac)
1847                 ocfs2_free_alloc_context(data_ac);
1848         if (xattr_ac)
1849                 ocfs2_free_alloc_context(xattr_ac);
1850         if ((status < 0) && inode) {
1851                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1852                 clear_nlink(inode);
1853                 iput(inode);
1854         }
1855
1856         if (status)
1857                 mlog_errno(status);
1858
1859         return status;
1860 }
1861
1862 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1863 {
1864         int status, namelen;
1865
1866         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1867                            (long long)blkno);
1868         if (namelen <= 0) {
1869                 if (namelen)
1870                         status = namelen;
1871                 else
1872                         status = -EINVAL;
1873                 mlog_errno(status);
1874                 goto bail;
1875         }
1876         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1877                 status = -EINVAL;
1878                 mlog_errno(status);
1879                 goto bail;
1880         }
1881
1882         trace_ocfs2_blkno_stringify(blkno, name, namelen);
1883
1884         status = 0;
1885 bail:
1886         if (status < 0)
1887                 mlog_errno(status);
1888         return status;
1889 }
1890
1891 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1892                                         struct inode **ret_orphan_dir,
1893                                         struct buffer_head **ret_orphan_dir_bh)
1894 {
1895         struct inode *orphan_dir_inode;
1896         struct buffer_head *orphan_dir_bh = NULL;
1897         int ret = 0;
1898
1899         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1900                                                        ORPHAN_DIR_SYSTEM_INODE,
1901                                                        osb->slot_num);
1902         if (!orphan_dir_inode) {
1903                 ret = -ENOENT;
1904                 mlog_errno(ret);
1905                 return ret;
1906         }
1907
1908         mutex_lock(&orphan_dir_inode->i_mutex);
1909
1910         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1911         if (ret < 0) {
1912                 mutex_unlock(&orphan_dir_inode->i_mutex);
1913                 iput(orphan_dir_inode);
1914
1915                 mlog_errno(ret);
1916                 return ret;
1917         }
1918
1919         *ret_orphan_dir = orphan_dir_inode;
1920         *ret_orphan_dir_bh = orphan_dir_bh;
1921
1922         return 0;
1923 }
1924
1925 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1926                                       struct buffer_head *orphan_dir_bh,
1927                                       u64 blkno,
1928                                       char *name,
1929                                       struct ocfs2_dir_lookup_result *lookup)
1930 {
1931         int ret;
1932         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1933
1934         ret = ocfs2_blkno_stringify(blkno, name);
1935         if (ret < 0) {
1936                 mlog_errno(ret);
1937                 return ret;
1938         }
1939
1940         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1941                                            orphan_dir_bh, name,
1942                                            OCFS2_ORPHAN_NAMELEN, lookup);
1943         if (ret < 0) {
1944                 mlog_errno(ret);
1945                 return ret;
1946         }
1947
1948         return 0;
1949 }
1950
1951 /**
1952  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1953  * insertion of an orphan.
1954  * @osb: ocfs2 file system
1955  * @ret_orphan_dir: Orphan dir inode - returned locked!
1956  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1957  * @lookup: dir lookup result, to be passed back into functions like
1958  *          ocfs2_orphan_add
1959  *
1960  * Returns zero on success and the ret_orphan_dir, name and lookup
1961  * fields will be populated.
1962  *
1963  * Returns non-zero on failure.
1964  */
1965 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1966                                     struct inode **ret_orphan_dir,
1967                                     u64 blkno,
1968                                     char *name,
1969                                     struct ocfs2_dir_lookup_result *lookup)
1970 {
1971         struct inode *orphan_dir_inode = NULL;
1972         struct buffer_head *orphan_dir_bh = NULL;
1973         int ret = 0;
1974
1975         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1976                                            &orphan_dir_bh);
1977         if (ret < 0) {
1978                 mlog_errno(ret);
1979                 return ret;
1980         }
1981
1982         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1983                                          blkno, name, lookup);
1984         if (ret < 0) {
1985                 mlog_errno(ret);
1986                 goto out;
1987         }
1988
1989         *ret_orphan_dir = orphan_dir_inode;
1990
1991 out:
1992         brelse(orphan_dir_bh);
1993
1994         if (ret) {
1995                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1996                 mutex_unlock(&orphan_dir_inode->i_mutex);
1997                 iput(orphan_dir_inode);
1998         }
1999
2000         if (ret)
2001                 mlog_errno(ret);
2002         return ret;
2003 }
2004
2005 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2006                             handle_t *handle,
2007                             struct inode *inode,
2008                             struct buffer_head *fe_bh,
2009                             char *name,
2010                             struct ocfs2_dir_lookup_result *lookup,
2011                             struct inode *orphan_dir_inode)
2012 {
2013         struct buffer_head *orphan_dir_bh = NULL;
2014         int status = 0;
2015         struct ocfs2_dinode *orphan_fe;
2016         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2017
2018         trace_ocfs2_orphan_add_begin(
2019                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2020
2021         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2022         if (status < 0) {
2023                 mlog_errno(status);
2024                 goto leave;
2025         }
2026
2027         status = ocfs2_journal_access_di(handle,
2028                                          INODE_CACHE(orphan_dir_inode),
2029                                          orphan_dir_bh,
2030                                          OCFS2_JOURNAL_ACCESS_WRITE);
2031         if (status < 0) {
2032                 mlog_errno(status);
2033                 goto leave;
2034         }
2035
2036         /*
2037          * We're going to journal the change of i_flags and i_orphaned_slot.
2038          * It's safe anyway, though some callers may duplicate the journaling.
2039          * Journaling within the func just make the logic look more
2040          * straightforward.
2041          */
2042         status = ocfs2_journal_access_di(handle,
2043                                          INODE_CACHE(inode),
2044                                          fe_bh,
2045                                          OCFS2_JOURNAL_ACCESS_WRITE);
2046         if (status < 0) {
2047                 mlog_errno(status);
2048                 goto leave;
2049         }
2050
2051         /* we're a cluster, and nlink can change on disk from
2052          * underneath us... */
2053         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2054         if (S_ISDIR(inode->i_mode))
2055                 ocfs2_add_links_count(orphan_fe, 1);
2056         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2057         ocfs2_journal_dirty(handle, orphan_dir_bh);
2058
2059         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2060                                    OCFS2_ORPHAN_NAMELEN, inode,
2061                                    OCFS2_I(inode)->ip_blkno,
2062                                    orphan_dir_bh, lookup);
2063         if (status < 0) {
2064                 mlog_errno(status);
2065                 goto rollback;
2066         }
2067
2068         fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2069         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2070
2071         /* Record which orphan dir our inode now resides
2072          * in. delete_inode will use this to determine which orphan
2073          * dir to lock. */
2074         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2075
2076         ocfs2_journal_dirty(handle, fe_bh);
2077
2078         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2079                                    osb->slot_num);
2080
2081 rollback:
2082         if (status < 0) {
2083                 if (S_ISDIR(inode->i_mode))
2084                         ocfs2_add_links_count(orphan_fe, -1);
2085                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2086         }
2087
2088 leave:
2089         brelse(orphan_dir_bh);
2090
2091         return status;
2092 }
2093
2094 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2095 int ocfs2_orphan_del(struct ocfs2_super *osb,
2096                      handle_t *handle,
2097                      struct inode *orphan_dir_inode,
2098                      struct inode *inode,
2099                      struct buffer_head *orphan_dir_bh)
2100 {
2101         char name[OCFS2_ORPHAN_NAMELEN + 1];
2102         struct ocfs2_dinode *orphan_fe;
2103         int status = 0;
2104         struct ocfs2_dir_lookup_result lookup = { NULL, };
2105
2106         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2107         if (status < 0) {
2108                 mlog_errno(status);
2109                 goto leave;
2110         }
2111
2112         trace_ocfs2_orphan_del(
2113              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2114              name, OCFS2_ORPHAN_NAMELEN);
2115
2116         /* find it's spot in the orphan directory */
2117         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2118                                   &lookup);
2119         if (status) {
2120                 mlog_errno(status);
2121                 goto leave;
2122         }
2123
2124         /* remove it from the orphan directory */
2125         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2126         if (status < 0) {
2127                 mlog_errno(status);
2128                 goto leave;
2129         }
2130
2131         status = ocfs2_journal_access_di(handle,
2132                                          INODE_CACHE(orphan_dir_inode),
2133                                          orphan_dir_bh,
2134                                          OCFS2_JOURNAL_ACCESS_WRITE);
2135         if (status < 0) {
2136                 mlog_errno(status);
2137                 goto leave;
2138         }
2139
2140         /* do the i_nlink dance! :) */
2141         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2142         if (S_ISDIR(inode->i_mode))
2143                 ocfs2_add_links_count(orphan_fe, -1);
2144         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2145         ocfs2_journal_dirty(handle, orphan_dir_bh);
2146
2147 leave:
2148         ocfs2_free_dir_lookup_result(&lookup);
2149
2150         if (status)
2151                 mlog_errno(status);
2152         return status;
2153 }
2154
2155 /**
2156  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2157  * allocated file. This is different from the typical 'add to orphan dir'
2158  * operation in that the inode does not yet exist. This is a problem because
2159  * the orphan dir stringifies the inode block number to come up with it's
2160  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2161  * problem. This function works around it by calling deeper into the orphan
2162  * and suballoc code than other callers. Use this only by necessity.
2163  * @dir: The directory which this inode will ultimately wind up under - not the
2164  * orphan dir!
2165  * @dir_bh: buffer_head the @dir inode block
2166  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2167  * with the string to be used for orphan dirent. Pass back to the orphan dir
2168  * code.
2169  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2170  * dir code.
2171  * @ret_di_blkno: block number where the new inode will be allocated.
2172  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2173  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2174  *
2175  * Returns zero on success and the ret_orphan_dir, name and lookup
2176  * fields will be populated.
2177  *
2178  * Returns non-zero on failure.
2179  */
2180 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2181                                         struct buffer_head *dir_bh,
2182                                         char *orphan_name,
2183                                         struct inode **ret_orphan_dir,
2184                                         u64 *ret_di_blkno,
2185                                         struct ocfs2_dir_lookup_result *orphan_insert,
2186                                         struct ocfs2_alloc_context **ret_inode_ac)
2187 {
2188         int ret;
2189         u64 di_blkno;
2190         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2191         struct inode *orphan_dir = NULL;
2192         struct buffer_head *orphan_dir_bh = NULL;
2193         struct ocfs2_alloc_context *inode_ac = NULL;
2194
2195         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2196         if (ret < 0) {
2197                 mlog_errno(ret);
2198                 return ret;
2199         }
2200
2201         /* reserve an inode spot */
2202         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2203         if (ret < 0) {
2204                 if (ret != -ENOSPC)
2205                         mlog_errno(ret);
2206                 goto out;
2207         }
2208
2209         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2210                                        &di_blkno);
2211         if (ret) {
2212                 mlog_errno(ret);
2213                 goto out;
2214         }
2215
2216         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2217                                          di_blkno, orphan_name, orphan_insert);
2218         if (ret < 0) {
2219                 mlog_errno(ret);
2220                 goto out;
2221         }
2222
2223 out:
2224         if (ret == 0) {
2225                 *ret_orphan_dir = orphan_dir;
2226                 *ret_di_blkno = di_blkno;
2227                 *ret_inode_ac = inode_ac;
2228                 /*
2229                  * orphan_name and orphan_insert are already up to
2230                  * date via prepare_orphan_dir
2231                  */
2232         } else {
2233                 /* Unroll reserve_new_inode* */
2234                 if (inode_ac)
2235                         ocfs2_free_alloc_context(inode_ac);
2236
2237                 /* Unroll orphan dir locking */
2238                 mutex_unlock(&orphan_dir->i_mutex);
2239                 ocfs2_inode_unlock(orphan_dir, 1);
2240                 iput(orphan_dir);
2241         }
2242
2243         brelse(orphan_dir_bh);
2244
2245         return ret;
2246 }
2247
2248 int ocfs2_create_inode_in_orphan(struct inode *dir,
2249                                  int mode,
2250                                  struct inode **new_inode)
2251 {
2252         int status, did_quota_inode = 0;
2253         struct inode *inode = NULL;
2254         struct inode *orphan_dir = NULL;
2255         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2256         struct ocfs2_dinode *di = NULL;
2257         handle_t *handle = NULL;
2258         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2259         struct buffer_head *parent_di_bh = NULL;
2260         struct buffer_head *new_di_bh = NULL;
2261         struct ocfs2_alloc_context *inode_ac = NULL;
2262         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2263         u64 uninitialized_var(di_blkno), suballoc_loc;
2264         u16 suballoc_bit;
2265
2266         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2267         if (status < 0) {
2268                 if (status != -ENOENT)
2269                         mlog_errno(status);
2270                 return status;
2271         }
2272
2273         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2274                                               orphan_name, &orphan_dir,
2275                                               &di_blkno, &orphan_insert, &inode_ac);
2276         if (status < 0) {
2277                 if (status != -ENOSPC)
2278                         mlog_errno(status);
2279                 goto leave;
2280         }
2281
2282         inode = ocfs2_get_init_inode(dir, mode);
2283         if (!inode) {
2284                 status = -ENOMEM;
2285                 mlog_errno(status);
2286                 goto leave;
2287         }
2288
2289         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2290         if (IS_ERR(handle)) {
2291                 status = PTR_ERR(handle);
2292                 handle = NULL;
2293                 mlog_errno(status);
2294                 goto leave;
2295         }
2296
2297         status = dquot_alloc_inode(inode);
2298         if (status)
2299                 goto leave;
2300         did_quota_inode = 1;
2301
2302         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2303                                               &suballoc_loc,
2304                                               &suballoc_bit, di_blkno);
2305         if (status < 0) {
2306                 mlog_errno(status);
2307                 goto leave;
2308         }
2309
2310         clear_nlink(inode);
2311         /* do the real work now. */
2312         status = __ocfs2_mknod_locked(dir, inode,
2313                                       0, &new_di_bh, parent_di_bh, handle,
2314                                       inode_ac, di_blkno, suballoc_loc,
2315                                       suballoc_bit);
2316         if (status < 0) {
2317                 mlog_errno(status);
2318                 goto leave;
2319         }
2320
2321         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2322         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2323                                   &orphan_insert, orphan_dir);
2324         if (status < 0) {
2325                 mlog_errno(status);
2326                 goto leave;
2327         }
2328
2329         /* get open lock so that only nodes can't remove it from orphan dir. */
2330         status = ocfs2_open_lock(inode);
2331         if (status < 0)
2332                 mlog_errno(status);
2333
2334         insert_inode_hash(inode);
2335 leave:
2336         if (status < 0 && did_quota_inode)
2337                 dquot_free_inode(inode);
2338         if (handle)
2339                 ocfs2_commit_trans(osb, handle);
2340
2341         if (orphan_dir) {
2342                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2343                 ocfs2_inode_unlock(orphan_dir, 1);
2344                 mutex_unlock(&orphan_dir->i_mutex);
2345                 iput(orphan_dir);
2346         }
2347
2348         if ((status < 0) && inode) {
2349                 clear_nlink(inode);
2350                 iput(inode);
2351         }
2352
2353         if (inode_ac)
2354                 ocfs2_free_alloc_context(inode_ac);
2355
2356         brelse(new_di_bh);
2357
2358         if (!status)
2359                 *new_inode = inode;
2360
2361         ocfs2_free_dir_lookup_result(&orphan_insert);
2362
2363         ocfs2_inode_unlock(dir, 1);
2364         brelse(parent_di_bh);
2365         return status;
2366 }
2367
2368 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2369                                    struct inode *inode,
2370                                    struct dentry *dentry)
2371 {
2372         int status = 0;
2373         struct buffer_head *parent_di_bh = NULL;
2374         handle_t *handle = NULL;
2375         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2376         struct ocfs2_dinode *dir_di, *di;
2377         struct inode *orphan_dir_inode = NULL;
2378         struct buffer_head *orphan_dir_bh = NULL;
2379         struct buffer_head *di_bh = NULL;
2380         struct ocfs2_dir_lookup_result lookup = { NULL, };
2381
2382         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2383                                 dentry->d_name.len, dentry->d_name.name,
2384                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2385                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2386
2387         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2388         if (status < 0) {
2389                 if (status != -ENOENT)
2390                         mlog_errno(status);
2391                 return status;
2392         }
2393
2394         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2395         if (!dir_di->i_links_count) {
2396                 /* can't make a file in a deleted directory. */
2397                 status = -ENOENT;
2398                 goto leave;
2399         }
2400
2401         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2402                                            dentry->d_name.len);
2403         if (status)
2404                 goto leave;
2405
2406         /* get a spot inside the dir. */
2407         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2408                                               dentry->d_name.name,
2409                                               dentry->d_name.len, &lookup);
2410         if (status < 0) {
2411                 mlog_errno(status);
2412                 goto leave;
2413         }
2414
2415         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2416                                                        ORPHAN_DIR_SYSTEM_INODE,
2417                                                        osb->slot_num);
2418         if (!orphan_dir_inode) {
2419                 status = -EEXIST;
2420                 mlog_errno(status);
2421                 goto leave;
2422         }
2423
2424         mutex_lock(&orphan_dir_inode->i_mutex);
2425
2426         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2427         if (status < 0) {
2428                 mlog_errno(status);
2429                 mutex_unlock(&orphan_dir_inode->i_mutex);
2430                 iput(orphan_dir_inode);
2431                 goto leave;
2432         }
2433
2434         status = ocfs2_read_inode_block(inode, &di_bh);
2435         if (status < 0) {
2436                 mlog_errno(status);
2437                 goto orphan_unlock;
2438         }
2439
2440         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2441         if (IS_ERR(handle)) {
2442                 status = PTR_ERR(handle);
2443                 handle = NULL;
2444                 mlog_errno(status);
2445                 goto orphan_unlock;
2446         }
2447
2448         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2449                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2450         if (status < 0) {
2451                 mlog_errno(status);
2452                 goto out_commit;
2453         }
2454
2455         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2456                                   orphan_dir_bh);
2457         if (status < 0) {
2458                 mlog_errno(status);
2459                 goto out_commit;
2460         }
2461
2462         di = (struct ocfs2_dinode *)di_bh->b_data;
2463         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2464         di->i_orphaned_slot = 0;
2465         set_nlink(inode, 1);
2466         ocfs2_set_links_count(di, inode->i_nlink);
2467         ocfs2_journal_dirty(handle, di_bh);
2468
2469         status = ocfs2_add_entry(handle, dentry, inode,
2470                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2471                                  &lookup);
2472         if (status < 0) {
2473                 mlog_errno(status);
2474                 goto out_commit;
2475         }
2476
2477         status = ocfs2_dentry_attach_lock(dentry, inode,
2478                                           OCFS2_I(dir)->ip_blkno);
2479         if (status) {
2480                 mlog_errno(status);
2481                 goto out_commit;
2482         }
2483
2484         d_instantiate(dentry, inode);
2485         status = 0;
2486 out_commit:
2487         ocfs2_commit_trans(osb, handle);
2488 orphan_unlock:
2489         ocfs2_inode_unlock(orphan_dir_inode, 1);
2490         mutex_unlock(&orphan_dir_inode->i_mutex);
2491         iput(orphan_dir_inode);
2492 leave:
2493
2494         ocfs2_inode_unlock(dir, 1);
2495
2496         brelse(di_bh);
2497         brelse(parent_di_bh);
2498         brelse(orphan_dir_bh);
2499
2500         ocfs2_free_dir_lookup_result(&lookup);
2501
2502         if (status)
2503                 mlog_errno(status);
2504
2505         return status;
2506 }
2507
2508 const struct inode_operations ocfs2_dir_iops = {
2509         .create         = ocfs2_create,
2510         .lookup         = ocfs2_lookup,
2511         .link           = ocfs2_link,
2512         .unlink         = ocfs2_unlink,
2513         .rmdir          = ocfs2_unlink,
2514         .symlink        = ocfs2_symlink,
2515         .mkdir          = ocfs2_mkdir,
2516         .mknod          = ocfs2_mknod,
2517         .rename         = ocfs2_rename,
2518         .setattr        = ocfs2_setattr,
2519         .getattr        = ocfs2_getattr,
2520         .permission     = ocfs2_permission,
2521         .setxattr       = generic_setxattr,
2522         .getxattr       = generic_getxattr,
2523         .listxattr      = ocfs2_listxattr,
2524         .removexattr    = generic_removexattr,
2525         .fiemap         = ocfs2_fiemap,
2526         .get_acl        = ocfs2_iop_get_acl,
2527         .set_acl        = ocfs2_iop_set_acl,
2528 };