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