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                                    struct nameidata *nd)
 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, int 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                        int 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
 234         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 235                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 236                           (unsigned long)dev, mode);
 237
 238         dquot_initialize(dir);
 239
 240         /* get our super block */
 241         osb = OCFS2_SB(dir->i_sb);
 242
 243         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 244         if (status < 0) {
 245                 if (status != -ENOENT)
 246                         mlog_errno(status);
 247                 return status;
 248         }
 249
 250         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 251                 status = -EMLINK;
 252                 goto leave;
 253         }
 254
 255         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 256         if (!ocfs2_read_links_count(dirfe)) {
 257                 /* can't make a file in a deleted directory. */
 258                 status = -ENOENT;
 259                 goto leave;
 260         }
 261
 262         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 263                                            dentry->d_name.len);
 264         if (status)
 265                 goto leave;
 266
 267         /* get a spot inside the dir. */
 268         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 269                                               dentry->d_name.name,
 270                                               dentry->d_name.len, &lookup);
 271         if (status < 0) {
 272                 mlog_errno(status);
 273                 goto leave;
 274         }
 275
 276         /* reserve an inode spot */
 277         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 278         if (status < 0) {
 279                 if (status != -ENOSPC)
 280                         mlog_errno(status);
 281                 goto leave;
 282         }
 283
 284         inode = ocfs2_get_init_inode(dir, mode);
 285         if (!inode) {
 286                 status = -ENOMEM;
 287                 mlog_errno(status);
 288                 goto leave;
 289         }
 290
 291         /* get security xattr */
 292         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 293         if (status) {
 294                 if (status == -EOPNOTSUPP)
 295                         si.enable = 0;
 296                 else {
 297                         mlog_errno(status);
 298                         goto leave;
 299                 }
 300         }
 301
 302         /* calculate meta data/clusters for setting security and acl xattr */
 303         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 304                                        &si, &want_clusters,
 305                                        &xattr_credits, &want_meta);
 306         if (status < 0) {
 307                 mlog_errno(status);
 308                 goto leave;
 309         }
 310
 311         /* Reserve a cluster if creating an extent based directory. */
 312         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 313                 want_clusters += 1;
 314
 315                 /* Dir indexing requires extra space as well */
 316                 if (ocfs2_supports_indexed_dirs(osb))
 317                         want_meta++;
 318         }
 319
 320         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 321         if (status < 0) {
 322                 if (status != -ENOSPC)
 323                         mlog_errno(status);
 324                 goto leave;
 325         }
 326
 327         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 328         if (status < 0) {
 329                 if (status != -ENOSPC)
 330                         mlog_errno(status);
 331                 goto leave;
 332         }
 333
 334         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 335                                                             S_ISDIR(mode),
 336                                                             xattr_credits));
 337         if (IS_ERR(handle)) {
 338                 status = PTR_ERR(handle);
 339                 handle = NULL;
 340                 mlog_errno(status);
 341                 goto leave;
 342         }
 343
 344         /* Starting to change things, restart is no longer possible. */
 345         ocfs2_block_signals(&oldset);
 346         did_block_signals = 1;
 347
 348         status = dquot_alloc_inode(inode);
 349         if (status)
 350                 goto leave;
 351         did_quota_inode = 1;
 352
 353         /* do the real work now. */
 354         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 355                                     &new_fe_bh, parent_fe_bh, handle,
 356                                     inode_ac);
 357         if (status < 0) {
 358                 mlog_errno(status);
 359                 goto leave;
 360         }
 361
 362         if (S_ISDIR(mode)) {
 363                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 364                                             new_fe_bh, data_ac, meta_ac);
 365                 if (status < 0) {
 366                         mlog_errno(status);
 367                         goto leave;
 368                 }
 369
 370                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 371                                                  parent_fe_bh,
 372                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 373                 if (status < 0) {
 374                         mlog_errno(status);
 375                         goto leave;
 376                 }
 377                 ocfs2_add_links_count(dirfe, 1);
 378                 ocfs2_journal_dirty(handle, parent_fe_bh);
 379                 inc_nlink(dir);
 380         }
 381
 382         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 383                                 meta_ac, data_ac);
 384         if (status < 0) {
 385                 mlog_errno(status);
 386                 goto leave;
 387         }
 388
 389         if (si.enable) {
 390                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 391                                                  meta_ac, data_ac);
 392                 if (status < 0) {
 393                         mlog_errno(status);
 394                         goto leave;
 395                 }
 396         }
 397
 398         /*
 399          * Do this before adding the entry to the directory. We add
 400          * also set d_op after success so that ->d_iput() will cleanup
 401          * the dentry lock even if ocfs2_add_entry() fails below.
 402          */
 403         status = ocfs2_dentry_attach_lock(dentry, inode,
 404                                           OCFS2_I(dir)->ip_blkno);
 405         if (status) {
 406                 mlog_errno(status);
 407                 goto leave;
 408         }
 409
 410         status = ocfs2_add_entry(handle, dentry, inode,
 411                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 412                                  &lookup);
 413         if (status < 0) {
 414                 mlog_errno(status);
 415                 goto leave;
 416         }
 417
 418         insert_inode_hash(inode);
 419         d_instantiate(dentry, inode);
 420         status = 0;
 421 leave:
 422         if (status < 0 && did_quota_inode)
 423                 dquot_free_inode(inode);
 424         if (handle)
 425                 ocfs2_commit_trans(osb, handle);
 426
 427         ocfs2_inode_unlock(dir, 1);
 428         if (did_block_signals)
 429                 ocfs2_unblock_signals(&oldset);
 430
 431         brelse(new_fe_bh);
 432         brelse(parent_fe_bh);
 433         kfree(si.name);
 434         kfree(si.value);
 435
 436         ocfs2_free_dir_lookup_result(&lookup);
 437
 438         if (inode_ac)
 439                 ocfs2_free_alloc_context(inode_ac);
 440
 441         if (data_ac)
 442                 ocfs2_free_alloc_context(data_ac);
 443
 444         if (meta_ac)
 445                 ocfs2_free_alloc_context(meta_ac);
 446
 447         /*
 448          * We should call iput after the i_mutex of the bitmap been
 449          * unlocked in ocfs2_free_alloc_context, or the
 450          * ocfs2_delete_inode will mutex_lock again.
 451          */
 452         if ((status < 0) && inode) {
 453                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 454                 clear_nlink(inode);
 455                 iput(inode);
 456         }
 457
 458         if (status)
 459                 mlog_errno(status);
 460
 461         return status;
 462 }
 463
 464 static int __ocfs2_mknod_locked(struct inode *dir,
 465                                 struct inode *inode,
 466                                 dev_t dev,
 467                                 struct buffer_head **new_fe_bh,
 468                                 struct buffer_head *parent_fe_bh,
 469                                 handle_t *handle,
 470                                 struct ocfs2_alloc_context *inode_ac,
 471                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 472 {
 473         int status = 0;
 474         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 475         struct ocfs2_dinode *fe = NULL;
 476         struct ocfs2_extent_list *fel;
 477         u16 feat;
 478
 479         *new_fe_bh = NULL;
 480
 481         /* populate as many fields early on as possible - many of
 482          * these are used by the support functions here and in
 483          * callers. */
 484         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 485         OCFS2_I(inode)->ip_blkno = fe_blkno;
 486         spin_lock(&osb->osb_lock);
 487         inode->i_generation = osb->s_next_generation++;
 488         spin_unlock(&osb->osb_lock);
 489
 490         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 491         if (!*new_fe_bh) {
 492                 status = -EIO;
 493                 mlog_errno(status);
 494                 goto leave;
 495         }
 496         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 497
 498         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 499                                          *new_fe_bh,
 500                                          OCFS2_JOURNAL_ACCESS_CREATE);
 501         if (status < 0) {
 502                 mlog_errno(status);
 503                 goto leave;
 504         }
 505
 506         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 507         memset(fe, 0, osb->sb->s_blocksize);
 508
 509         fe->i_generation = cpu_to_le32(inode->i_generation);
 510         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 511         fe->i_blkno = cpu_to_le64(fe_blkno);
 512         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 513         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 514         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 515         fe->i_uid = cpu_to_le32(inode->i_uid);
 516         fe->i_gid = cpu_to_le32(inode->i_gid);
 517         fe->i_mode = cpu_to_le16(inode->i_mode);
 518         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 519                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 520
 521         ocfs2_set_links_count(fe, inode->i_nlink);
 522
 523         fe->i_last_eb_blk = 0;
 524         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 525         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 526         fe->i_atime = fe->i_ctime = fe->i_mtime =
 527                 cpu_to_le64(CURRENT_TIME.tv_sec);
 528         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 529                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 530         fe->i_dtime = 0;
 531
 532         /*
 533          * If supported, directories start with inline data. If inline
 534          * isn't supported, but indexing is, we start them as indexed.
 535          */
 536         feat = le16_to_cpu(fe->i_dyn_features);
 537         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 538                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 539
 540                 fe->id2.i_data.id_count = cpu_to_le16(
 541                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 542         } else {
 543                 fel = &fe->id2.i_list;
 544                 fel->l_tree_depth = 0;
 545                 fel->l_next_free_rec = 0;
 546                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 547         }
 548
 549         ocfs2_journal_dirty(handle, *new_fe_bh);
 550
 551         ocfs2_populate_inode(inode, fe, 1);
 552         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 553         if (!ocfs2_mount_local(osb)) {
 554                 status = ocfs2_create_new_inode_locks(inode);
 555                 if (status < 0)
 556                         mlog_errno(status);
 557         }
 558
 559         status = 0; /* error in ocfs2_create_new_inode_locks is not
 560                      * critical */
 561
 562 leave:
 563         if (status < 0) {
 564                 if (*new_fe_bh) {
 565                         brelse(*new_fe_bh);
 566                         *new_fe_bh = NULL;
 567                 }
 568         }
 569
 570         if (status)
 571                 mlog_errno(status);
 572         return status;
 573 }
 574
 575 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 576                               struct inode *dir,
 577                               struct inode *inode,
 578                               dev_t dev,
 579                               struct buffer_head **new_fe_bh,
 580                               struct buffer_head *parent_fe_bh,
 581                               handle_t *handle,
 582                               struct ocfs2_alloc_context *inode_ac)
 583 {
 584         int status = 0;
 585         u64 suballoc_loc, fe_blkno = 0;
 586         u16 suballoc_bit;
 587
 588         *new_fe_bh = NULL;
 589
 590         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 591                                        inode_ac, &suballoc_loc,
 592                                        &suballoc_bit, &fe_blkno);
 593         if (status < 0) {
 594                 mlog_errno(status);
 595                 return status;
 596         }
 597
 598         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 599                                     parent_fe_bh, handle, inode_ac,
 600                                     fe_blkno, suballoc_loc, suballoc_bit);
 601 }
 602
 603 static int ocfs2_mkdir(struct inode *dir,
 604                        struct dentry *dentry,
 605                        int mode)
 606 {
 607         int ret;
 608
 609         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 610                           OCFS2_I(dir)->ip_blkno, mode);
 611         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 612         if (ret)
 613                 mlog_errno(ret);
 614
 615         return ret;
 616 }
 617
 618 static int ocfs2_create(struct inode *dir,
 619                         struct dentry *dentry,
 620                         int mode,
 621                         struct nameidata *nd)
 622 {
 623         int ret;
 624
 625         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 626                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 627         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 628         if (ret)
 629                 mlog_errno(ret);
 630
 631         return ret;
 632 }
 633
 634 static int ocfs2_link(struct dentry *old_dentry,
 635                       struct inode *dir,
 636                       struct dentry *dentry)
 637 {
 638         handle_t *handle;
 639         struct inode *inode = old_dentry->d_inode;
 640         int err;
 641         struct buffer_head *fe_bh = NULL;
 642         struct buffer_head *parent_fe_bh = NULL;
 643         struct ocfs2_dinode *fe = NULL;
 644         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 645         struct ocfs2_dir_lookup_result lookup = { NULL, };
 646         sigset_t oldset;
 647
 648         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 649                          old_dentry->d_name.len, old_dentry->d_name.name,
 650                          dentry->d_name.len, dentry->d_name.name);
 651
 652         if (S_ISDIR(inode->i_mode))
 653                 return -EPERM;
 654
 655         dquot_initialize(dir);
 656
 657         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 658         if (err < 0) {
 659                 if (err != -ENOENT)
 660                         mlog_errno(err);
 661                 return err;
 662         }
 663
 664         if (!dir->i_nlink) {
 665                 err = -ENOENT;
 666                 goto out;
 667         }
 668
 669         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 670                                         dentry->d_name.len);
 671         if (err)
 672                 goto out;
 673
 674         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 675                                            dentry->d_name.name,
 676                                            dentry->d_name.len, &lookup);
 677         if (err < 0) {
 678                 mlog_errno(err);
 679                 goto out;
 680         }
 681
 682         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 683         if (err < 0) {
 684                 if (err != -ENOENT)
 685                         mlog_errno(err);
 686                 goto out;
 687         }
 688
 689         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 690         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 691                 err = -EMLINK;
 692                 goto out_unlock_inode;
 693         }
 694
 695         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 696         if (IS_ERR(handle)) {
 697                 err = PTR_ERR(handle);
 698                 handle = NULL;
 699                 mlog_errno(err);
 700                 goto out_unlock_inode;
 701         }
 702
 703         /* Starting to change things, restart is no longer possible. */
 704         ocfs2_block_signals(&oldset);
 705
 706         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 707                                       OCFS2_JOURNAL_ACCESS_WRITE);
 708         if (err < 0) {
 709                 mlog_errno(err);
 710                 goto out_commit;
 711         }
 712
 713         inc_nlink(inode);
 714         inode->i_ctime = CURRENT_TIME;
 715         ocfs2_set_links_count(fe, inode->i_nlink);
 716         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 717         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 718         ocfs2_journal_dirty(handle, fe_bh);
 719
 720         err = ocfs2_add_entry(handle, dentry, inode,
 721                               OCFS2_I(inode)->ip_blkno,
 722                               parent_fe_bh, &lookup);
 723         if (err) {
 724                 ocfs2_add_links_count(fe, -1);
 725                 drop_nlink(inode);
 726                 mlog_errno(err);
 727                 goto out_commit;
 728         }
 729
 730         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 731         if (err) {
 732                 mlog_errno(err);
 733                 goto out_commit;
 734         }
 735
 736         ihold(inode);
 737         d_instantiate(dentry, inode);
 738
 739 out_commit:
 740         ocfs2_commit_trans(osb, handle);
 741         ocfs2_unblock_signals(&oldset);
 742 out_unlock_inode:
 743         ocfs2_inode_unlock(inode, 1);
 744
 745 out:
 746         ocfs2_inode_unlock(dir, 1);
 747
 748         brelse(fe_bh);
 749         brelse(parent_fe_bh);
 750
 751         ocfs2_free_dir_lookup_result(&lookup);
 752
 753         if (err)
 754                 mlog_errno(err);
 755
 756         return err;
 757 }
 758
 759 /*
 760  * Takes and drops an exclusive lock on the given dentry. This will
 761  * force other nodes to drop it.
 762  */
 763 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 764 {
 765         int ret;
 766
 767         ret = ocfs2_dentry_lock(dentry, 1);
 768         if (ret)
 769                 mlog_errno(ret);
 770         else
 771                 ocfs2_dentry_unlock(dentry, 1);
 772
 773         return ret;
 774 }
 775
 776 static inline int inode_is_unlinkable(struct inode *inode)
 777 {
 778         if (S_ISDIR(inode->i_mode)) {
 779                 if (inode->i_nlink == 2)
 780                         return 1;
 781                 return 0;
 782         }
 783
 784         if (inode->i_nlink == 1)
 785                 return 1;
 786         return 0;
 787 }
 788
 789 static int ocfs2_unlink(struct inode *dir,
 790                         struct dentry *dentry)
 791 {
 792         int status;
 793         int child_locked = 0;
 794         struct inode *inode = dentry->d_inode;
 795         struct inode *orphan_dir = NULL;
 796         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 797         u64 blkno;
 798         struct ocfs2_dinode *fe = NULL;
 799         struct buffer_head *fe_bh = NULL;
 800         struct buffer_head *parent_node_bh = NULL;
 801         handle_t *handle = NULL;
 802         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 803         struct ocfs2_dir_lookup_result lookup = { NULL, };
 804         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 805
 806         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 807                            dentry->d_name.name,
 808                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 809                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 810
 811         dquot_initialize(dir);
 812
 813         BUG_ON(dentry->d_parent->d_inode != dir);
 814
 815         if (inode == osb->root_inode)
 816                 return -EPERM;
 817
 818         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 819                                          OI_LS_PARENT);
 820         if (status < 0) {
 821                 if (status != -ENOENT)
 822                         mlog_errno(status);
 823                 return status;
 824         }
 825
 826         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 827                                           dentry->d_name.len, &blkno, dir,
 828                                           &lookup);
 829         if (status < 0) {
 830                 if (status != -ENOENT)
 831                         mlog_errno(status);
 832                 goto leave;
 833         }
 834
 835         if (OCFS2_I(inode)->ip_blkno != blkno) {
 836                 status = -ENOENT;
 837
 838                 trace_ocfs2_unlink_noent(
 839                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 840                                 (unsigned long long)blkno,
 841                                 OCFS2_I(inode)->ip_flags);
 842                 goto leave;
 843         }
 844
 845         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 846         if (status < 0) {
 847                 if (status != -ENOENT)
 848                         mlog_errno(status);
 849                 goto leave;
 850         }
 851         child_locked = 1;
 852
 853         if (S_ISDIR(inode->i_mode)) {
 854                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 855                         status = -ENOTEMPTY;
 856                         goto leave;
 857                 }
 858         }
 859
 860         status = ocfs2_remote_dentry_delete(dentry);
 861         if (status < 0) {
 862                 /* This remote delete should succeed under all normal
 863                  * circumstances. */
 864                 mlog_errno(status);
 865                 goto leave;
 866         }
 867
 868         if (inode_is_unlinkable(inode)) {
 869                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 870                                                   OCFS2_I(inode)->ip_blkno,
 871                                                   orphan_name, &orphan_insert);
 872                 if (status < 0) {
 873                         mlog_errno(status);
 874                         goto leave;
 875                 }
 876         }
 877
 878         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 879         if (IS_ERR(handle)) {
 880                 status = PTR_ERR(handle);
 881                 handle = NULL;
 882                 mlog_errno(status);
 883                 goto leave;
 884         }
 885
 886         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 887                                          OCFS2_JOURNAL_ACCESS_WRITE);
 888         if (status < 0) {
 889                 mlog_errno(status);
 890                 goto leave;
 891         }
 892
 893         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 894
 895         if (inode_is_unlinkable(inode)) {
 896                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 897                                           &orphan_insert, orphan_dir);
 898                 if (status < 0) {
 899                         mlog_errno(status);
 900                         goto leave;
 901                 }
 902         }
 903
 904         /* delete the name from the parent dir */
 905         status = ocfs2_delete_entry(handle, dir, &lookup);
 906         if (status < 0) {
 907                 mlog_errno(status);
 908                 goto leave;
 909         }
 910
 911         if (S_ISDIR(inode->i_mode))
 912                 drop_nlink(inode);
 913         drop_nlink(inode);
 914         ocfs2_set_links_count(fe, inode->i_nlink);
 915         ocfs2_journal_dirty(handle, fe_bh);
 916
 917         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 918         if (S_ISDIR(inode->i_mode))
 919                 drop_nlink(dir);
 920
 921         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 922         if (status < 0) {
 923                 mlog_errno(status);
 924                 if (S_ISDIR(inode->i_mode))
 925                         inc_nlink(dir);
 926         }
 927
 928 leave:
 929         if (handle)
 930                 ocfs2_commit_trans(osb, handle);
 931
 932         if (child_locked)
 933                 ocfs2_inode_unlock(inode, 1);
 934
 935         ocfs2_inode_unlock(dir, 1);
 936
 937         if (orphan_dir) {
 938                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 939                 ocfs2_inode_unlock(orphan_dir, 1);
 940                 mutex_unlock(&orphan_dir->i_mutex);
 941                 iput(orphan_dir);
 942         }
 943
 944         brelse(fe_bh);
 945         brelse(parent_node_bh);
 946
 947         ocfs2_free_dir_lookup_result(&orphan_insert);
 948         ocfs2_free_dir_lookup_result(&lookup);
 949
 950         if (status)
 951                 mlog_errno(status);
 952
 953         return status;
 954 }
 955
 956 /*
 957  * The only place this should be used is rename!
 958  * if they have the same id, then the 1st one is the only one locked.
 959  */
 960 static int ocfs2_double_lock(struct ocfs2_super *osb,
 961                              struct buffer_head **bh1,
 962                              struct inode *inode1,
 963                              struct buffer_head **bh2,
 964                              struct inode *inode2)
 965 {
 966         int status;
 967         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 968         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 969         struct buffer_head **tmpbh;
 970         struct inode *tmpinode;
 971
 972         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
 973                                 (unsigned long long)oi2->ip_blkno);
 974
 975         if (*bh1)
 976                 *bh1 = NULL;
 977         if (*bh2)
 978                 *bh2 = NULL;
 979
 980         /* we always want to lock the one with the lower lockid first. */
 981         if (oi1->ip_blkno != oi2->ip_blkno) {
 982                 if (oi1->ip_blkno < oi2->ip_blkno) {
 983                         /* switch id1 and id2 around */
 984                         tmpbh = bh2;
 985                         bh2 = bh1;
 986                         bh1 = tmpbh;
 987
 988                         tmpinode = inode2;
 989                         inode2 = inode1;
 990                         inode1 = tmpinode;
 991                 }
 992                 /* lock id2 */
 993                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 994                                                  OI_LS_RENAME1);
 995                 if (status < 0) {
 996                         if (status != -ENOENT)
 997                                 mlog_errno(status);
 998                         goto bail;
 999                 }
1000         }
1001
1002         /* lock id1 */
1003         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1004         if (status < 0) {
1005                 /*
1006                  * An error return must mean that no cluster locks
1007                  * were held on function exit.
1008                  */
1009                 if (oi1->ip_blkno != oi2->ip_blkno) {
1010                         ocfs2_inode_unlock(inode2, 1);
1011                         brelse(*bh2);
1012                         *bh2 = NULL;
1013                 }
1014
1015                 if (status != -ENOENT)
1016                         mlog_errno(status);
1017         }
1018
1019         trace_ocfs2_double_lock_end(
1020                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1021                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1022
1023 bail:
1024         if (status)
1025                 mlog_errno(status);
1026         return status;
1027 }
1028
1029 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1030 {
1031         ocfs2_inode_unlock(inode1, 1);
1032
1033         if (inode1 != inode2)
1034                 ocfs2_inode_unlock(inode2, 1);
1035 }
1036
1037 static int ocfs2_rename(struct inode *old_dir,
1038                         struct dentry *old_dentry,
1039                         struct inode *new_dir,
1040                         struct dentry *new_dentry)
1041 {
1042         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1043         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1044         struct inode *old_inode = old_dentry->d_inode;
1045         struct inode *new_inode = new_dentry->d_inode;
1046         struct inode *orphan_dir = NULL;
1047         struct ocfs2_dinode *newfe = NULL;
1048         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1049         struct buffer_head *newfe_bh = NULL;
1050         struct buffer_head *old_inode_bh = NULL;
1051         struct ocfs2_super *osb = NULL;
1052         u64 newfe_blkno, old_de_ino;
1053         handle_t *handle = NULL;
1054         struct buffer_head *old_dir_bh = NULL;
1055         struct buffer_head *new_dir_bh = NULL;
1056         nlink_t old_dir_nlink = old_dir->i_nlink;
1057         struct ocfs2_dinode *old_di;
1058         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1059         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1060         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1061         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1062         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1063
1064         /* At some point it might be nice to break this function up a
1065          * bit. */
1066
1067         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1068                            old_dentry->d_name.len, old_dentry->d_name.name,
1069                            new_dentry->d_name.len, new_dentry->d_name.name);
1070
1071         dquot_initialize(old_dir);
1072         dquot_initialize(new_dir);
1073
1074         osb = OCFS2_SB(old_dir->i_sb);
1075
1076         if (new_inode) {
1077                 if (!igrab(new_inode))
1078                         BUG();
1079         }
1080
1081         /* Assume a directory hierarchy thusly:
1082          * a/b/c
1083          * a/d
1084          * a,b,c, and d are all directories.
1085          *
1086          * from cwd of 'a' on both nodes:
1087          * node1: mv b/c d
1088          * node2: mv d   b/c
1089          *
1090          * And that's why, just like the VFS, we need a file system
1091          * rename lock. */
1092         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1093                 status = ocfs2_rename_lock(osb);
1094                 if (status < 0) {
1095                         mlog_errno(status);
1096                         goto bail;
1097                 }
1098                 rename_lock = 1;
1099         }
1100
1101         /* if old and new are the same, this'll just do one lock. */
1102         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1103                                    &new_dir_bh, new_dir);
1104         if (status < 0) {
1105                 mlog_errno(status);
1106                 goto bail;
1107         }
1108         parents_locked = 1;
1109
1110         /* make sure both dirs have bhs
1111          * get an extra ref on old_dir_bh if old==new */
1112         if (!new_dir_bh) {
1113                 if (old_dir_bh) {
1114                         new_dir_bh = old_dir_bh;
1115                         get_bh(new_dir_bh);
1116                 } else {
1117                         mlog(ML_ERROR, "no old_dir_bh!\n");
1118                         status = -EIO;
1119                         goto bail;
1120                 }
1121         }
1122
1123         /*
1124          * Aside from allowing a meta data update, the locking here
1125          * also ensures that the downconvert thread on other nodes
1126          * won't have to concurrently downconvert the inode and the
1127          * dentry locks.
1128          */
1129         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1130                                          OI_LS_PARENT);
1131         if (status < 0) {
1132                 if (status != -ENOENT)
1133                         mlog_errno(status);
1134                 goto bail;
1135         }
1136         old_child_locked = 1;
1137
1138         status = ocfs2_remote_dentry_delete(old_dentry);
1139         if (status < 0) {
1140                 mlog_errno(status);
1141                 goto bail;
1142         }
1143
1144         if (S_ISDIR(old_inode->i_mode)) {
1145                 u64 old_inode_parent;
1146
1147                 update_dot_dot = 1;
1148                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1149                                                   old_inode,
1150                                                   &old_inode_dot_dot_res);
1151                 if (status) {
1152                         status = -EIO;
1153                         goto bail;
1154                 }
1155
1156                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1157                         status = -EIO;
1158                         goto bail;
1159                 }
1160
1161                 if (!new_inode && new_dir != old_dir &&
1162                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1163                         status = -EMLINK;
1164                         goto bail;
1165                 }
1166         }
1167
1168         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1169                                             old_dentry->d_name.len,
1170                                             &old_de_ino);
1171         if (status) {
1172                 status = -ENOENT;
1173                 goto bail;
1174         }
1175
1176         /*
1177          *  Check for inode number is _not_ due to possible IO errors.
1178          *  We might rmdir the source, keep it as pwd of some process
1179          *  and merrily kill the link to whatever was created under the
1180          *  same name. Goodbye sticky bit ;-<
1181          */
1182         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1183                 status = -ENOENT;
1184                 goto bail;
1185         }
1186
1187         /* check if the target already exists (in which case we need
1188          * to delete it */
1189         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1190                                           new_dentry->d_name.len,
1191                                           &newfe_blkno, new_dir,
1192                                           &target_lookup_res);
1193         /* The only error we allow here is -ENOENT because the new
1194          * file not existing is perfectly valid. */
1195         if ((status < 0) && (status != -ENOENT)) {
1196                 /* If we cannot find the file specified we should just */
1197                 /* return the error... */
1198                 mlog_errno(status);
1199                 goto bail;
1200         }
1201         if (status == 0)
1202                 target_exists = 1;
1203
1204         if (!target_exists && new_inode) {
1205                 /*
1206                  * Target was unlinked by another node while we were
1207                  * waiting to get to ocfs2_rename(). There isn't
1208                  * anything we can do here to help the situation, so
1209                  * bubble up the appropriate error.
1210                  */
1211                 status = -ENOENT;
1212                 goto bail;
1213         }
1214
1215         /* In case we need to overwrite an existing file, we blow it
1216          * away first */
1217         if (target_exists) {
1218                 /* VFS didn't think there existed an inode here, but
1219                  * someone else in the cluster must have raced our
1220                  * rename to create one. Today we error cleanly, in
1221                  * the future we should consider calling iget to build
1222                  * a new struct inode for this entry. */
1223                 if (!new_inode) {
1224                         status = -EACCES;
1225
1226                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1227                                                 new_dentry->d_name.name);
1228                         goto bail;
1229                 }
1230
1231                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1232                         status = -EACCES;
1233
1234                         trace_ocfs2_rename_disagree(
1235                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1236                              (unsigned long long)newfe_blkno,
1237                              OCFS2_I(new_inode)->ip_flags);
1238                         goto bail;
1239                 }
1240
1241                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1242                 if (status < 0) {
1243                         if (status != -ENOENT)
1244                                 mlog_errno(status);
1245                         goto bail;
1246                 }
1247                 new_child_locked = 1;
1248
1249                 status = ocfs2_remote_dentry_delete(new_dentry);
1250                 if (status < 0) {
1251                         mlog_errno(status);
1252                         goto bail;
1253                 }
1254
1255                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1256
1257                 trace_ocfs2_rename_over_existing(
1258                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1259                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1260
1261                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1262                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1263                                                 OCFS2_I(new_inode)->ip_blkno,
1264                                                 orphan_name, &orphan_insert);
1265                         if (status < 0) {
1266                                 mlog_errno(status);
1267                                 goto bail;
1268                         }
1269                 }
1270         } else {
1271                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1272
1273                 status = ocfs2_check_dir_for_entry(new_dir,
1274                                                    new_dentry->d_name.name,
1275                                                    new_dentry->d_name.len);
1276                 if (status)
1277                         goto bail;
1278
1279                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1280                                                       new_dentry->d_name.name,
1281                                                       new_dentry->d_name.len,
1282                                                       &target_insert);
1283                 if (status < 0) {
1284                         mlog_errno(status);
1285                         goto bail;
1286                 }
1287         }
1288
1289         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1290         if (IS_ERR(handle)) {
1291                 status = PTR_ERR(handle);
1292                 handle = NULL;
1293                 mlog_errno(status);
1294                 goto bail;
1295         }
1296
1297         if (target_exists) {
1298                 if (S_ISDIR(new_inode->i_mode)) {
1299                         if (new_inode->i_nlink != 2 ||
1300                             !ocfs2_empty_dir(new_inode)) {
1301                                 status = -ENOTEMPTY;
1302                                 goto bail;
1303                         }
1304                 }
1305                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1306                                                  newfe_bh,
1307                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1308                 if (status < 0) {
1309                         mlog_errno(status);
1310                         goto bail;
1311                 }
1312
1313                 if (S_ISDIR(new_inode->i_mode) ||
1314                     (ocfs2_read_links_count(newfe) == 1)) {
1315                         status = ocfs2_orphan_add(osb, handle, new_inode,
1316                                                   newfe_bh, orphan_name,
1317                                                   &orphan_insert, orphan_dir);
1318                         if (status < 0) {
1319                                 mlog_errno(status);
1320                                 goto bail;
1321                         }
1322                 }
1323
1324                 /* change the dirent to point to the correct inode */
1325                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1326                                             old_inode);
1327                 if (status < 0) {
1328                         mlog_errno(status);
1329                         goto bail;
1330                 }
1331                 new_dir->i_version++;
1332
1333                 if (S_ISDIR(new_inode->i_mode))
1334                         ocfs2_set_links_count(newfe, 0);
1335                 else
1336                         ocfs2_add_links_count(newfe, -1);
1337                 ocfs2_journal_dirty(handle, newfe_bh);
1338         } else {
1339                 /* if the name was not found in new_dir, add it now */
1340                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1341                                          OCFS2_I(old_inode)->ip_blkno,
1342                                          new_dir_bh, &target_insert);
1343         }
1344
1345         old_inode->i_ctime = CURRENT_TIME;
1346         mark_inode_dirty(old_inode);
1347
1348         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1349                                          old_inode_bh,
1350                                          OCFS2_JOURNAL_ACCESS_WRITE);
1351         if (status >= 0) {
1352                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1353
1354                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1355                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1356                 ocfs2_journal_dirty(handle, old_inode_bh);
1357         } else
1358                 mlog_errno(status);
1359
1360         /*
1361          * Now that the name has been added to new_dir, remove the old name.
1362          *
1363          * We don't keep any directory entry context around until now
1364          * because the insert might have changed the type of directory
1365          * we're dealing with.
1366          */
1367         status = ocfs2_find_entry(old_dentry->d_name.name,
1368                                   old_dentry->d_name.len, old_dir,
1369                                   &old_entry_lookup);
1370         if (status)
1371                 goto bail;
1372
1373         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1374         if (status < 0) {
1375                 mlog_errno(status);
1376                 goto bail;
1377         }
1378
1379         if (new_inode) {
1380                 drop_nlink(new_inode);
1381                 new_inode->i_ctime = CURRENT_TIME;
1382         }
1383         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1384
1385         if (update_dot_dot) {
1386                 status = ocfs2_update_entry(old_inode, handle,
1387                                             &old_inode_dot_dot_res, new_dir);
1388                 drop_nlink(old_dir);
1389                 if (new_inode) {
1390                         drop_nlink(new_inode);
1391                 } else {
1392                         inc_nlink(new_dir);
1393                         mark_inode_dirty(new_dir);
1394                 }
1395         }
1396         mark_inode_dirty(old_dir);
1397         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1398         if (new_inode) {
1399                 mark_inode_dirty(new_inode);
1400                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1401         }
1402
1403         if (old_dir != new_dir) {
1404                 /* Keep the same times on both directories.*/
1405                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1406
1407                 /*
1408                  * This will also pick up the i_nlink change from the
1409                  * block above.
1410                  */
1411                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1412         }
1413
1414         if (old_dir_nlink != old_dir->i_nlink) {
1415                 if (!old_dir_bh) {
1416                         mlog(ML_ERROR, "need to change nlink for old dir "
1417                              "%llu from %d to %d but bh is NULL!\n",
1418                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1419                              (int)old_dir_nlink, old_dir->i_nlink);
1420                 } else {
1421                         struct ocfs2_dinode *fe;
1422                         status = ocfs2_journal_access_di(handle,
1423                                                          INODE_CACHE(old_dir),
1424                                                          old_dir_bh,
1425                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1426                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1427                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1428                         ocfs2_journal_dirty(handle, old_dir_bh);
1429                 }
1430         }
1431         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1432         status = 0;
1433 bail:
1434         if (rename_lock)
1435                 ocfs2_rename_unlock(osb);
1436
1437         if (handle)
1438                 ocfs2_commit_trans(osb, handle);
1439
1440         if (parents_locked)
1441                 ocfs2_double_unlock(old_dir, new_dir);
1442
1443         if (old_child_locked)
1444                 ocfs2_inode_unlock(old_inode, 1);
1445
1446         if (new_child_locked)
1447                 ocfs2_inode_unlock(new_inode, 1);
1448
1449         if (orphan_dir) {
1450                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1451                 ocfs2_inode_unlock(orphan_dir, 1);
1452                 mutex_unlock(&orphan_dir->i_mutex);
1453                 iput(orphan_dir);
1454         }
1455
1456         if (new_inode)
1457                 sync_mapping_buffers(old_inode->i_mapping);
1458
1459         if (new_inode)
1460                 iput(new_inode);
1461
1462         ocfs2_free_dir_lookup_result(&target_lookup_res);
1463         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1464         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1465         ocfs2_free_dir_lookup_result(&orphan_insert);
1466         ocfs2_free_dir_lookup_result(&target_insert);
1467
1468         brelse(newfe_bh);
1469         brelse(old_inode_bh);
1470         brelse(old_dir_bh);
1471         brelse(new_dir_bh);
1472
1473         if (status)
1474                 mlog_errno(status);
1475
1476         return status;
1477 }
1478
1479 /*
1480  * we expect i_size = strlen(symname). Copy symname into the file
1481  * data, including the null terminator.
1482  */
1483 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1484                                      handle_t *handle,
1485                                      struct inode *inode,
1486                                      const char *symname)
1487 {
1488         struct buffer_head **bhs = NULL;
1489         const char *c;
1490         struct super_block *sb = osb->sb;
1491         u64 p_blkno, p_blocks;
1492         int virtual, blocks, status, i, bytes_left;
1493
1494         bytes_left = i_size_read(inode) + 1;
1495         /* we can't trust i_blocks because we're actually going to
1496          * write i_size + 1 bytes. */
1497         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1498
1499         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1500                                         i_size_read(inode), blocks);
1501
1502         /* Sanity check -- make sure we're going to fit. */
1503         if (bytes_left >
1504             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1505                 status = -EIO;
1506                 mlog_errno(status);
1507                 goto bail;
1508         }
1509
1510         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1511         if (!bhs) {
1512                 status = -ENOMEM;
1513                 mlog_errno(status);
1514                 goto bail;
1515         }
1516
1517         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1518                                              NULL);
1519         if (status < 0) {
1520                 mlog_errno(status);
1521                 goto bail;
1522         }
1523
1524         /* links can never be larger than one cluster so we know this
1525          * is all going to be contiguous, but do a sanity check
1526          * anyway. */
1527         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1528                 status = -EIO;
1529                 mlog_errno(status);
1530                 goto bail;
1531         }
1532
1533         virtual = 0;
1534         while(bytes_left > 0) {
1535                 c = &symname[virtual * sb->s_blocksize];
1536
1537                 bhs[virtual] = sb_getblk(sb, p_blkno);
1538                 if (!bhs[virtual]) {
1539                         status = -ENOMEM;
1540                         mlog_errno(status);
1541                         goto bail;
1542                 }
1543                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1544                                               bhs[virtual]);
1545
1546                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1547                                               bhs[virtual],
1548                                               OCFS2_JOURNAL_ACCESS_CREATE);
1549                 if (status < 0) {
1550                         mlog_errno(status);
1551                         goto bail;
1552                 }
1553
1554                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1555
1556                 memcpy(bhs[virtual]->b_data, c,
1557                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1558                        bytes_left);
1559
1560                 ocfs2_journal_dirty(handle, bhs[virtual]);
1561
1562                 virtual++;
1563                 p_blkno++;
1564                 bytes_left -= sb->s_blocksize;
1565         }
1566
1567         status = 0;
1568 bail:
1569
1570         if (bhs) {
1571                 for(i = 0; i < blocks; i++)
1572                         brelse(bhs[i]);
1573                 kfree(bhs);
1574         }
1575
1576         if (status)
1577                 mlog_errno(status);
1578         return status;
1579 }
1580
1581 static int ocfs2_symlink(struct inode *dir,
1582                          struct dentry *dentry,
1583                          const char *symname)
1584 {
1585         int status, l, credits;
1586         u64 newsize;
1587         struct ocfs2_super *osb = NULL;
1588         struct inode *inode = NULL;
1589         struct super_block *sb;
1590         struct buffer_head *new_fe_bh = NULL;
1591         struct buffer_head *parent_fe_bh = NULL;
1592         struct ocfs2_dinode *fe = NULL;
1593         struct ocfs2_dinode *dirfe;
1594         handle_t *handle = NULL;
1595         struct ocfs2_alloc_context *inode_ac = NULL;
1596         struct ocfs2_alloc_context *data_ac = NULL;
1597         struct ocfs2_alloc_context *xattr_ac = NULL;
1598         int want_clusters = 0;
1599         int xattr_credits = 0;
1600         struct ocfs2_security_xattr_info si = {
1601                 .enable = 1,
1602         };
1603         int did_quota = 0, did_quota_inode = 0;
1604         struct ocfs2_dir_lookup_result lookup = { NULL, };
1605         sigset_t oldset;
1606         int did_block_signals = 0;
1607
1608         trace_ocfs2_symlink_begin(dir, dentry, symname,
1609                                   dentry->d_name.len, dentry->d_name.name);
1610
1611         dquot_initialize(dir);
1612
1613         sb = dir->i_sb;
1614         osb = OCFS2_SB(sb);
1615
1616         l = strlen(symname) + 1;
1617
1618         credits = ocfs2_calc_symlink_credits(sb);
1619
1620         /* lock the parent directory */
1621         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1622         if (status < 0) {
1623                 if (status != -ENOENT)
1624                         mlog_errno(status);
1625                 return status;
1626         }
1627
1628         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1629         if (!ocfs2_read_links_count(dirfe)) {
1630                 /* can't make a file in a deleted directory. */
1631                 status = -ENOENT;
1632                 goto bail;
1633         }
1634
1635         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1636                                            dentry->d_name.len);
1637         if (status)
1638                 goto bail;
1639
1640         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1641                                               dentry->d_name.name,
1642                                               dentry->d_name.len, &lookup);
1643         if (status < 0) {
1644                 mlog_errno(status);
1645                 goto bail;
1646         }
1647
1648         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1649         if (status < 0) {
1650                 if (status != -ENOSPC)
1651                         mlog_errno(status);
1652                 goto bail;
1653         }
1654
1655         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1656         if (!inode) {
1657                 status = -ENOMEM;
1658                 mlog_errno(status);
1659                 goto bail;
1660         }
1661
1662         /* get security xattr */
1663         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1664         if (status) {
1665                 if (status == -EOPNOTSUPP)
1666                         si.enable = 0;
1667                 else {
1668                         mlog_errno(status);
1669                         goto bail;
1670                 }
1671         }
1672
1673         /* calculate meta data/clusters for setting security xattr */
1674         if (si.enable) {
1675                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1676                                                   &xattr_credits, &xattr_ac);
1677                 if (status < 0) {
1678                         mlog_errno(status);
1679                         goto bail;
1680                 }
1681         }
1682
1683         /* don't reserve bitmap space for fast symlinks. */
1684         if (l > ocfs2_fast_symlink_chars(sb))
1685                 want_clusters += 1;
1686
1687         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1688         if (status < 0) {
1689                 if (status != -ENOSPC)
1690                         mlog_errno(status);
1691                 goto bail;
1692         }
1693
1694         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1695         if (IS_ERR(handle)) {
1696                 status = PTR_ERR(handle);
1697                 handle = NULL;
1698                 mlog_errno(status);
1699                 goto bail;
1700         }
1701
1702         /* Starting to change things, restart is no longer possible. */
1703         ocfs2_block_signals(&oldset);
1704         did_block_signals = 1;
1705
1706         status = dquot_alloc_inode(inode);
1707         if (status)
1708                 goto bail;
1709         did_quota_inode = 1;
1710
1711         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1712                                    dentry->d_name.name,
1713                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1714                                    inode->i_mode);
1715
1716         status = ocfs2_mknod_locked(osb, dir, inode,
1717                                     0, &new_fe_bh, parent_fe_bh, handle,
1718                                     inode_ac);
1719         if (status < 0) {
1720                 mlog_errno(status);
1721                 goto bail;
1722         }
1723
1724         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1725         inode->i_rdev = 0;
1726         newsize = l - 1;
1727         if (l > ocfs2_fast_symlink_chars(sb)) {
1728                 u32 offset = 0;
1729
1730                 inode->i_op = &ocfs2_symlink_inode_operations;
1731                 status = dquot_alloc_space_nodirty(inode,
1732                     ocfs2_clusters_to_bytes(osb->sb, 1));
1733                 if (status)
1734                         goto bail;
1735                 did_quota = 1;
1736                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1737                                               new_fe_bh,
1738                                               handle, data_ac, NULL,
1739                                               NULL);
1740                 if (status < 0) {
1741                         if (status != -ENOSPC && status != -EINTR) {
1742                                 mlog(ML_ERROR,
1743                                      "Failed to extend file to %llu\n",
1744                                      (unsigned long long)newsize);
1745                                 mlog_errno(status);
1746                                 status = -ENOSPC;
1747                         }
1748                         goto bail;
1749                 }
1750                 i_size_write(inode, newsize);
1751                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1752         } else {
1753                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1754                 memcpy((char *) fe->id2.i_symlink, symname, l);
1755                 i_size_write(inode, newsize);
1756                 inode->i_blocks = 0;
1757         }
1758
1759         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1760         if (status < 0) {
1761                 mlog_errno(status);
1762                 goto bail;
1763         }
1764
1765         if (!ocfs2_inode_is_fast_symlink(inode)) {
1766                 status = ocfs2_create_symlink_data(osb, handle, inode,
1767                                                    symname);
1768                 if (status < 0) {
1769                         mlog_errno(status);
1770                         goto bail;
1771                 }
1772         }
1773
1774         if (si.enable) {
1775                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1776                                                  xattr_ac, data_ac);
1777                 if (status < 0) {
1778                         mlog_errno(status);
1779                         goto bail;
1780                 }
1781         }
1782
1783         /*
1784          * Do this before adding the entry to the directory. We add
1785          * also set d_op after success so that ->d_iput() will cleanup
1786          * the dentry lock even if ocfs2_add_entry() fails below.
1787          */
1788         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1789         if (status) {
1790                 mlog_errno(status);
1791                 goto bail;
1792         }
1793
1794         status = ocfs2_add_entry(handle, dentry, inode,
1795                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1796                                  &lookup);
1797         if (status < 0) {
1798                 mlog_errno(status);
1799                 goto bail;
1800         }
1801
1802         insert_inode_hash(inode);
1803         d_instantiate(dentry, inode);
1804 bail:
1805         if (status < 0 && did_quota)
1806                 dquot_free_space_nodirty(inode,
1807                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1808         if (status < 0 && did_quota_inode)
1809                 dquot_free_inode(inode);
1810         if (handle)
1811                 ocfs2_commit_trans(osb, handle);
1812
1813         ocfs2_inode_unlock(dir, 1);
1814         if (did_block_signals)
1815                 ocfs2_unblock_signals(&oldset);
1816
1817         brelse(new_fe_bh);
1818         brelse(parent_fe_bh);
1819         kfree(si.name);
1820         kfree(si.value);
1821         ocfs2_free_dir_lookup_result(&lookup);
1822         if (inode_ac)
1823                 ocfs2_free_alloc_context(inode_ac);
1824         if (data_ac)
1825                 ocfs2_free_alloc_context(data_ac);
1826         if (xattr_ac)
1827                 ocfs2_free_alloc_context(xattr_ac);
1828         if ((status < 0) && inode) {
1829                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1830                 clear_nlink(inode);
1831                 iput(inode);
1832         }
1833
1834         if (status)
1835                 mlog_errno(status);
1836
1837         return status;
1838 }
1839
1840 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1841 {
1842         int status, namelen;
1843
1844         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1845                            (long long)blkno);
1846         if (namelen <= 0) {
1847                 if (namelen)
1848                         status = namelen;
1849                 else
1850                         status = -EINVAL;
1851                 mlog_errno(status);
1852                 goto bail;
1853         }
1854         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1855                 status = -EINVAL;
1856                 mlog_errno(status);
1857                 goto bail;
1858         }
1859
1860         trace_ocfs2_blkno_stringify(blkno, name, namelen);
1861
1862         status = 0;
1863 bail:
1864         if (status < 0)
1865                 mlog_errno(status);
1866         return status;
1867 }
1868
1869 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1870                                         struct inode **ret_orphan_dir,
1871                                         struct buffer_head **ret_orphan_dir_bh)
1872 {
1873         struct inode *orphan_dir_inode;
1874         struct buffer_head *orphan_dir_bh = NULL;
1875         int ret = 0;
1876
1877         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1878                                                        ORPHAN_DIR_SYSTEM_INODE,
1879                                                        osb->slot_num);
1880         if (!orphan_dir_inode) {
1881                 ret = -ENOENT;
1882                 mlog_errno(ret);
1883                 return ret;
1884         }
1885
1886         mutex_lock(&orphan_dir_inode->i_mutex);
1887
1888         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1889         if (ret < 0) {
1890                 mutex_unlock(&orphan_dir_inode->i_mutex);
1891                 iput(orphan_dir_inode);
1892
1893                 mlog_errno(ret);
1894                 return ret;
1895         }
1896
1897         *ret_orphan_dir = orphan_dir_inode;
1898         *ret_orphan_dir_bh = orphan_dir_bh;
1899
1900         return 0;
1901 }
1902
1903 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1904                                       struct buffer_head *orphan_dir_bh,
1905                                       u64 blkno,
1906                                       char *name,
1907                                       struct ocfs2_dir_lookup_result *lookup)
1908 {
1909         int ret;
1910         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1911
1912         ret = ocfs2_blkno_stringify(blkno, name);
1913         if (ret < 0) {
1914                 mlog_errno(ret);
1915                 return ret;
1916         }
1917
1918         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1919                                            orphan_dir_bh, name,
1920                                            OCFS2_ORPHAN_NAMELEN, lookup);
1921         if (ret < 0) {
1922                 mlog_errno(ret);
1923                 return ret;
1924         }
1925
1926         return 0;
1927 }
1928
1929 /**
1930  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1931  * insertion of an orphan.
1932  * @osb: ocfs2 file system
1933  * @ret_orphan_dir: Orphan dir inode - returned locked!
1934  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1935  * @lookup: dir lookup result, to be passed back into functions like
1936  *          ocfs2_orphan_add
1937  *
1938  * Returns zero on success and the ret_orphan_dir, name and lookup
1939  * fields will be populated.
1940  *
1941  * Returns non-zero on failure.
1942  */
1943 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1944                                     struct inode **ret_orphan_dir,
1945                                     u64 blkno,
1946                                     char *name,
1947                                     struct ocfs2_dir_lookup_result *lookup)
1948 {
1949         struct inode *orphan_dir_inode = NULL;
1950         struct buffer_head *orphan_dir_bh = NULL;
1951         int ret = 0;
1952
1953         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1954                                            &orphan_dir_bh);
1955         if (ret < 0) {
1956                 mlog_errno(ret);
1957                 return ret;
1958         }
1959
1960         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1961                                          blkno, name, lookup);
1962         if (ret < 0) {
1963                 mlog_errno(ret);
1964                 goto out;
1965         }
1966
1967         *ret_orphan_dir = orphan_dir_inode;
1968
1969 out:
1970         brelse(orphan_dir_bh);
1971
1972         if (ret) {
1973                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1974                 mutex_unlock(&orphan_dir_inode->i_mutex);
1975                 iput(orphan_dir_inode);
1976         }
1977
1978         if (ret)
1979                 mlog_errno(ret);
1980         return ret;
1981 }
1982
1983 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1984                             handle_t *handle,
1985                             struct inode *inode,
1986                             struct buffer_head *fe_bh,
1987                             char *name,
1988                             struct ocfs2_dir_lookup_result *lookup,
1989                             struct inode *orphan_dir_inode)
1990 {
1991         struct buffer_head *orphan_dir_bh = NULL;
1992         int status = 0;
1993         struct ocfs2_dinode *orphan_fe;
1994         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1995
1996         trace_ocfs2_orphan_add_begin(
1997                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
1998
1999         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2000         if (status < 0) {
2001                 mlog_errno(status);
2002                 goto leave;
2003         }
2004
2005         status = ocfs2_journal_access_di(handle,
2006                                          INODE_CACHE(orphan_dir_inode),
2007                                          orphan_dir_bh,
2008                                          OCFS2_JOURNAL_ACCESS_WRITE);
2009         if (status < 0) {
2010                 mlog_errno(status);
2011                 goto leave;
2012         }
2013
2014         /* we're a cluster, and nlink can change on disk from
2015          * underneath us... */
2016         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2017         if (S_ISDIR(inode->i_mode))
2018                 ocfs2_add_links_count(orphan_fe, 1);
2019         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2020         ocfs2_journal_dirty(handle, orphan_dir_bh);
2021
2022         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2023                                    OCFS2_ORPHAN_NAMELEN, inode,
2024                                    OCFS2_I(inode)->ip_blkno,
2025                                    orphan_dir_bh, lookup);
2026         if (status < 0) {
2027                 mlog_errno(status);
2028                 goto leave;
2029         }
2030
2031         /*
2032          * We're going to journal the change of i_flags and i_orphaned_slot.
2033          * It's safe anyway, though some callers may duplicate the journaling.
2034          * Journaling within the func just make the logic look more
2035          * straightforward.
2036          */
2037         status = ocfs2_journal_access_di(handle,
2038                                          INODE_CACHE(inode),
2039                                          fe_bh,
2040                                          OCFS2_JOURNAL_ACCESS_WRITE);
2041         if (status < 0) {
2042                 mlog_errno(status);
2043                 goto leave;
2044         }
2045
2046         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2047         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2048
2049         /* Record which orphan dir our inode now resides
2050          * in. delete_inode will use this to determine which orphan
2051          * dir to lock. */
2052         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2053
2054         ocfs2_journal_dirty(handle, fe_bh);
2055
2056         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2057                                    osb->slot_num);
2058
2059 leave:
2060         brelse(orphan_dir_bh);
2061
2062         if (status)
2063                 mlog_errno(status);
2064         return status;
2065 }
2066
2067 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2068 int ocfs2_orphan_del(struct ocfs2_super *osb,
2069                      handle_t *handle,
2070                      struct inode *orphan_dir_inode,
2071                      struct inode *inode,
2072                      struct buffer_head *orphan_dir_bh)
2073 {
2074         char name[OCFS2_ORPHAN_NAMELEN + 1];
2075         struct ocfs2_dinode *orphan_fe;
2076         int status = 0;
2077         struct ocfs2_dir_lookup_result lookup = { NULL, };
2078
2079         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2080         if (status < 0) {
2081                 mlog_errno(status);
2082                 goto leave;
2083         }
2084
2085         trace_ocfs2_orphan_del(
2086              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2087              name, OCFS2_ORPHAN_NAMELEN);
2088
2089         /* find it's spot in the orphan directory */
2090         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2091                                   &lookup);
2092         if (status) {
2093                 mlog_errno(status);
2094                 goto leave;
2095         }
2096
2097         /* remove it from the orphan directory */
2098         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2099         if (status < 0) {
2100                 mlog_errno(status);
2101                 goto leave;
2102         }
2103
2104         status = ocfs2_journal_access_di(handle,
2105                                          INODE_CACHE(orphan_dir_inode),
2106                                          orphan_dir_bh,
2107                                          OCFS2_JOURNAL_ACCESS_WRITE);
2108         if (status < 0) {
2109                 mlog_errno(status);
2110                 goto leave;
2111         }
2112
2113         /* do the i_nlink dance! :) */
2114         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2115         if (S_ISDIR(inode->i_mode))
2116                 ocfs2_add_links_count(orphan_fe, -1);
2117         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2118         ocfs2_journal_dirty(handle, orphan_dir_bh);
2119
2120 leave:
2121         ocfs2_free_dir_lookup_result(&lookup);
2122
2123         if (status)
2124                 mlog_errno(status);
2125         return status;
2126 }
2127
2128 /**
2129  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2130  * allocated file. This is different from the typical 'add to orphan dir'
2131  * operation in that the inode does not yet exist. This is a problem because
2132  * the orphan dir stringifies the inode block number to come up with it's
2133  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2134  * problem. This function works around it by calling deeper into the orphan
2135  * and suballoc code than other callers. Use this only by necessity.
2136  * @dir: The directory which this inode will ultimately wind up under - not the
2137  * orphan dir!
2138  * @dir_bh: buffer_head the @dir inode block
2139  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2140  * with the string to be used for orphan dirent. Pass back to the orphan dir
2141  * code.
2142  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2143  * dir code.
2144  * @ret_di_blkno: block number where the new inode will be allocated.
2145  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2146  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2147  *
2148  * Returns zero on success and the ret_orphan_dir, name and lookup
2149  * fields will be populated.
2150  *
2151  * Returns non-zero on failure.
2152  */
2153 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2154                                         struct buffer_head *dir_bh,
2155                                         char *orphan_name,
2156                                         struct inode **ret_orphan_dir,
2157                                         u64 *ret_di_blkno,
2158                                         struct ocfs2_dir_lookup_result *orphan_insert,
2159                                         struct ocfs2_alloc_context **ret_inode_ac)
2160 {
2161         int ret;
2162         u64 di_blkno;
2163         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2164         struct inode *orphan_dir = NULL;
2165         struct buffer_head *orphan_dir_bh = NULL;
2166         struct ocfs2_alloc_context *inode_ac = NULL;
2167
2168         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2169         if (ret < 0) {
2170                 mlog_errno(ret);
2171                 return ret;
2172         }
2173
2174         /* reserve an inode spot */
2175         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2176         if (ret < 0) {
2177                 if (ret != -ENOSPC)
2178                         mlog_errno(ret);
2179                 goto out;
2180         }
2181
2182         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2183                                        &di_blkno);
2184         if (ret) {
2185                 mlog_errno(ret);
2186                 goto out;
2187         }
2188
2189         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2190                                          di_blkno, orphan_name, orphan_insert);
2191         if (ret < 0) {
2192                 mlog_errno(ret);
2193                 goto out;
2194         }
2195
2196 out:
2197         if (ret == 0) {
2198                 *ret_orphan_dir = orphan_dir;
2199                 *ret_di_blkno = di_blkno;
2200                 *ret_inode_ac = inode_ac;
2201                 /*
2202                  * orphan_name and orphan_insert are already up to
2203                  * date via prepare_orphan_dir
2204                  */
2205         } else {
2206                 /* Unroll reserve_new_inode* */
2207                 if (inode_ac)
2208                         ocfs2_free_alloc_context(inode_ac);
2209
2210                 /* Unroll orphan dir locking */
2211                 mutex_unlock(&orphan_dir->i_mutex);
2212                 ocfs2_inode_unlock(orphan_dir, 1);
2213                 iput(orphan_dir);
2214         }
2215
2216         brelse(orphan_dir_bh);
2217
2218         return 0;
2219 }
2220
2221 int ocfs2_create_inode_in_orphan(struct inode *dir,
2222                                  int mode,
2223                                  struct inode **new_inode)
2224 {
2225         int status, did_quota_inode = 0;
2226         struct inode *inode = NULL;
2227         struct inode *orphan_dir = NULL;
2228         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2229         struct ocfs2_dinode *di = NULL;
2230         handle_t *handle = NULL;
2231         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2232         struct buffer_head *parent_di_bh = NULL;
2233         struct buffer_head *new_di_bh = NULL;
2234         struct ocfs2_alloc_context *inode_ac = NULL;
2235         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2236         u64 uninitialized_var(di_blkno), suballoc_loc;
2237         u16 suballoc_bit;
2238
2239         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2240         if (status < 0) {
2241                 if (status != -ENOENT)
2242                         mlog_errno(status);
2243                 return status;
2244         }
2245
2246         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2247                                               orphan_name, &orphan_dir,
2248                                               &di_blkno, &orphan_insert, &inode_ac);
2249         if (status < 0) {
2250                 if (status != -ENOSPC)
2251                         mlog_errno(status);
2252                 goto leave;
2253         }
2254
2255         inode = ocfs2_get_init_inode(dir, mode);
2256         if (!inode) {
2257                 status = -ENOMEM;
2258                 mlog_errno(status);
2259                 goto leave;
2260         }
2261
2262         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2263         if (IS_ERR(handle)) {
2264                 status = PTR_ERR(handle);
2265                 handle = NULL;
2266                 mlog_errno(status);
2267                 goto leave;
2268         }
2269
2270         status = dquot_alloc_inode(inode);
2271         if (status)
2272                 goto leave;
2273         did_quota_inode = 1;
2274
2275         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2276                                               &suballoc_loc,
2277                                               &suballoc_bit, di_blkno);
2278         if (status < 0) {
2279                 mlog_errno(status);
2280                 goto leave;
2281         }
2282
2283         clear_nlink(inode);
2284         /* do the real work now. */
2285         status = __ocfs2_mknod_locked(dir, inode,
2286                                       0, &new_di_bh, parent_di_bh, handle,
2287                                       inode_ac, di_blkno, suballoc_loc,
2288                                       suballoc_bit);
2289         if (status < 0) {
2290                 mlog_errno(status);
2291                 goto leave;
2292         }
2293
2294         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2295         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2296                                   &orphan_insert, orphan_dir);
2297         if (status < 0) {
2298                 mlog_errno(status);
2299                 goto leave;
2300         }
2301
2302         /* get open lock so that only nodes can't remove it from orphan dir. */
2303         status = ocfs2_open_lock(inode);
2304         if (status < 0)
2305                 mlog_errno(status);
2306
2307         insert_inode_hash(inode);
2308 leave:
2309         if (status < 0 && did_quota_inode)
2310                 dquot_free_inode(inode);
2311         if (handle)
2312                 ocfs2_commit_trans(osb, handle);
2313
2314         if (orphan_dir) {
2315                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2316                 ocfs2_inode_unlock(orphan_dir, 1);
2317                 mutex_unlock(&orphan_dir->i_mutex);
2318                 iput(orphan_dir);
2319         }
2320
2321         if ((status < 0) && inode) {
2322                 clear_nlink(inode);
2323                 iput(inode);
2324         }
2325
2326         if (inode_ac)
2327                 ocfs2_free_alloc_context(inode_ac);
2328
2329         brelse(new_di_bh);
2330
2331         if (!status)
2332                 *new_inode = inode;
2333
2334         ocfs2_free_dir_lookup_result(&orphan_insert);
2335
2336         ocfs2_inode_unlock(dir, 1);
2337         brelse(parent_di_bh);
2338         return status;
2339 }
2340
2341 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2342                                    struct inode *inode,
2343                                    struct dentry *dentry)
2344 {
2345         int status = 0;
2346         struct buffer_head *parent_di_bh = NULL;
2347         handle_t *handle = NULL;
2348         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2349         struct ocfs2_dinode *dir_di, *di;
2350         struct inode *orphan_dir_inode = NULL;
2351         struct buffer_head *orphan_dir_bh = NULL;
2352         struct buffer_head *di_bh = NULL;
2353         struct ocfs2_dir_lookup_result lookup = { NULL, };
2354
2355         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2356                                 dentry->d_name.len, dentry->d_name.name,
2357                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2358                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2359
2360         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2361         if (status < 0) {
2362                 if (status != -ENOENT)
2363                         mlog_errno(status);
2364                 return status;
2365         }
2366
2367         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2368         if (!dir_di->i_links_count) {
2369                 /* can't make a file in a deleted directory. */
2370                 status = -ENOENT;
2371                 goto leave;
2372         }
2373
2374         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2375                                            dentry->d_name.len);
2376         if (status)
2377                 goto leave;
2378
2379         /* get a spot inside the dir. */
2380         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2381                                               dentry->d_name.name,
2382                                               dentry->d_name.len, &lookup);
2383         if (status < 0) {
2384                 mlog_errno(status);
2385                 goto leave;
2386         }
2387
2388         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2389                                                        ORPHAN_DIR_SYSTEM_INODE,
2390                                                        osb->slot_num);
2391         if (!orphan_dir_inode) {
2392                 status = -EEXIST;
2393                 mlog_errno(status);
2394                 goto leave;
2395         }
2396
2397         mutex_lock(&orphan_dir_inode->i_mutex);
2398
2399         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2400         if (status < 0) {
2401                 mlog_errno(status);
2402                 mutex_unlock(&orphan_dir_inode->i_mutex);
2403                 iput(orphan_dir_inode);
2404                 goto leave;
2405         }
2406
2407         status = ocfs2_read_inode_block(inode, &di_bh);
2408         if (status < 0) {
2409                 mlog_errno(status);
2410                 goto orphan_unlock;
2411         }
2412
2413         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2414         if (IS_ERR(handle)) {
2415                 status = PTR_ERR(handle);
2416                 handle = NULL;
2417                 mlog_errno(status);
2418                 goto orphan_unlock;
2419         }
2420
2421         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2422                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2423         if (status < 0) {
2424                 mlog_errno(status);
2425                 goto out_commit;
2426         }
2427
2428         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2429                                   orphan_dir_bh);
2430         if (status < 0) {
2431                 mlog_errno(status);
2432                 goto out_commit;
2433         }
2434
2435         di = (struct ocfs2_dinode *)di_bh->b_data;
2436         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2437         di->i_orphaned_slot = 0;
2438         set_nlink(inode, 1);
2439         ocfs2_set_links_count(di, inode->i_nlink);
2440         ocfs2_journal_dirty(handle, di_bh);
2441
2442         status = ocfs2_add_entry(handle, dentry, inode,
2443                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2444                                  &lookup);
2445         if (status < 0) {
2446                 mlog_errno(status);
2447                 goto out_commit;
2448         }
2449
2450         status = ocfs2_dentry_attach_lock(dentry, inode,
2451                                           OCFS2_I(dir)->ip_blkno);
2452         if (status) {
2453                 mlog_errno(status);
2454                 goto out_commit;
2455         }
2456
2457         d_instantiate(dentry, inode);
2458         status = 0;
2459 out_commit:
2460         ocfs2_commit_trans(osb, handle);
2461 orphan_unlock:
2462         ocfs2_inode_unlock(orphan_dir_inode, 1);
2463         mutex_unlock(&orphan_dir_inode->i_mutex);
2464         iput(orphan_dir_inode);
2465 leave:
2466
2467         ocfs2_inode_unlock(dir, 1);
2468
2469         brelse(di_bh);
2470         brelse(parent_di_bh);
2471         brelse(orphan_dir_bh);
2472
2473         ocfs2_free_dir_lookup_result(&lookup);
2474
2475         if (status)
2476                 mlog_errno(status);
2477
2478         return status;
2479 }
2480
2481 const struct inode_operations ocfs2_dir_iops = {
2482         .create         = ocfs2_create,
2483         .lookup         = ocfs2_lookup,
2484         .link           = ocfs2_link,
2485         .unlink         = ocfs2_unlink,
2486         .rmdir          = ocfs2_unlink,
2487         .symlink        = ocfs2_symlink,
2488         .mkdir          = ocfs2_mkdir,
2489         .mknod          = ocfs2_mknod,
2490         .rename         = ocfs2_rename,
2491         .setattr        = ocfs2_setattr,
2492         .getattr        = ocfs2_getattr,
2493         .permission     = ocfs2_permission,
2494         .setxattr       = generic_setxattr,
2495         .getxattr       = generic_getxattr,
2496         .listxattr      = ocfs2_listxattr,
2497         .removexattr    = generic_removexattr,
2498         .fiemap         = ocfs2_fiemap,
2499         .get_acl        = ocfs2_iop_get_acl,
2500 };