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                 inode->i_nlink = 2;
 203         else
 204                 inode->i_nlink = 1;
 205         inode_init_owner(inode, dir, mode);
 206         dquot_initialize(inode);
 207         return inode;
 208 }
 209
 210 static int ocfs2_mknod(struct inode *dir,
 211                        struct dentry *dentry,
 212                        int mode,
 213                        dev_t dev)
 214 {
 215         int status = 0;
 216         struct buffer_head *parent_fe_bh = NULL;
 217         handle_t *handle = NULL;
 218         struct ocfs2_super *osb;
 219         struct ocfs2_dinode *dirfe;
 220         struct buffer_head *new_fe_bh = NULL;
 221         struct inode *inode = NULL;
 222         struct ocfs2_alloc_context *inode_ac = NULL;
 223         struct ocfs2_alloc_context *data_ac = NULL;
 224         struct ocfs2_alloc_context *meta_ac = NULL;
 225         int want_clusters = 0;
 226         int want_meta = 0;
 227         int xattr_credits = 0;
 228         struct ocfs2_security_xattr_info si = {
 229                 .enable = 1,
 230         };
 231         int did_quota_inode = 0;
 232         struct ocfs2_dir_lookup_result lookup = { NULL, };
 233         sigset_t oldset;
 234         int did_block_signals = 0;
 235
 236         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 237                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 238                           (unsigned long)dev, mode);
 239
 240         dquot_initialize(dir);
 241
 242         /* get our super block */
 243         osb = OCFS2_SB(dir->i_sb);
 244
 245         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 246         if (status < 0) {
 247                 if (status != -ENOENT)
 248                         mlog_errno(status);
 249                 return status;
 250         }
 251
 252         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 253                 status = -EMLINK;
 254                 goto leave;
 255         }
 256
 257         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 258         if (!ocfs2_read_links_count(dirfe)) {
 259                 /* can't make a file in a deleted directory. */
 260                 status = -ENOENT;
 261                 goto leave;
 262         }
 263
 264         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 265                                            dentry->d_name.len);
 266         if (status)
 267                 goto leave;
 268
 269         /* get a spot inside the dir. */
 270         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 271                                               dentry->d_name.name,
 272                                               dentry->d_name.len, &lookup);
 273         if (status < 0) {
 274                 mlog_errno(status);
 275                 goto leave;
 276         }
 277
 278         /* reserve an inode spot */
 279         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 280         if (status < 0) {
 281                 if (status != -ENOSPC)
 282                         mlog_errno(status);
 283                 goto leave;
 284         }
 285
 286         inode = ocfs2_get_init_inode(dir, mode);
 287         if (!inode) {
 288                 status = -ENOMEM;
 289                 mlog_errno(status);
 290                 goto leave;
 291         }
 292
 293         /* get security xattr */
 294         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 295         if (status) {
 296                 if (status == -EOPNOTSUPP)
 297                         si.enable = 0;
 298                 else {
 299                         mlog_errno(status);
 300                         goto leave;
 301                 }
 302         }
 303
 304         /* calculate meta data/clusters for setting security and acl xattr */
 305         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 306                                        &si, &want_clusters,
 307                                        &xattr_credits, &want_meta);
 308         if (status < 0) {
 309                 mlog_errno(status);
 310                 goto leave;
 311         }
 312
 313         /* Reserve a cluster if creating an extent based directory. */
 314         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 315                 want_clusters += 1;
 316
 317                 /* Dir indexing requires extra space as well */
 318                 if (ocfs2_supports_indexed_dirs(osb))
 319                         want_meta++;
 320         }
 321
 322         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 323         if (status < 0) {
 324                 if (status != -ENOSPC)
 325                         mlog_errno(status);
 326                 goto leave;
 327         }
 328
 329         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 330         if (status < 0) {
 331                 if (status != -ENOSPC)
 332                         mlog_errno(status);
 333                 goto leave;
 334         }
 335
 336         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 337                                                             S_ISDIR(mode),
 338                                                             xattr_credits));
 339         if (IS_ERR(handle)) {
 340                 status = PTR_ERR(handle);
 341                 handle = NULL;
 342                 mlog_errno(status);
 343                 goto leave;
 344         }
 345
 346         /* Starting to change things, restart is no longer possible. */
 347         ocfs2_block_signals(&oldset);
 348         did_block_signals = 1;
 349
 350         status = dquot_alloc_inode(inode);
 351         if (status)
 352                 goto leave;
 353         did_quota_inode = 1;
 354
 355         /* do the real work now. */
 356         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 357                                     &new_fe_bh, parent_fe_bh, handle,
 358                                     inode_ac);
 359         if (status < 0) {
 360                 mlog_errno(status);
 361                 goto leave;
 362         }
 363
 364         if (S_ISDIR(mode)) {
 365                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 366                                             new_fe_bh, data_ac, meta_ac);
 367                 if (status < 0) {
 368                         mlog_errno(status);
 369                         goto leave;
 370                 }
 371
 372                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 373                                                  parent_fe_bh,
 374                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 375                 if (status < 0) {
 376                         mlog_errno(status);
 377                         goto leave;
 378                 }
 379                 ocfs2_add_links_count(dirfe, 1);
 380                 ocfs2_journal_dirty(handle, parent_fe_bh);
 381                 inc_nlink(dir);
 382         }
 383
 384         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 385                                 meta_ac, data_ac);
 386         if (status < 0) {
 387                 mlog_errno(status);
 388                 goto leave;
 389         }
 390
 391         if (si.enable) {
 392                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 393                                                  meta_ac, data_ac);
 394                 if (status < 0) {
 395                         mlog_errno(status);
 396                         goto leave;
 397                 }
 398         }
 399
 400         /*
 401          * Do this before adding the entry to the directory. We add
 402          * also set d_op after success so that ->d_iput() will cleanup
 403          * the dentry lock even if ocfs2_add_entry() fails below.
 404          */
 405         status = ocfs2_dentry_attach_lock(dentry, inode,
 406                                           OCFS2_I(dir)->ip_blkno);
 407         if (status) {
 408                 mlog_errno(status);
 409                 goto leave;
 410         }
 411
 412         status = ocfs2_add_entry(handle, dentry, inode,
 413                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 414                                  &lookup);
 415         if (status < 0) {
 416                 mlog_errno(status);
 417                 goto leave;
 418         }
 419
 420         insert_inode_hash(inode);
 421         d_instantiate(dentry, inode);
 422         status = 0;
 423 leave:
 424         if (status < 0 && did_quota_inode)
 425                 dquot_free_inode(inode);
 426         if (handle)
 427                 ocfs2_commit_trans(osb, handle);
 428
 429         ocfs2_inode_unlock(dir, 1);
 430         if (did_block_signals)
 431                 ocfs2_unblock_signals(&oldset);
 432
 433         brelse(new_fe_bh);
 434         brelse(parent_fe_bh);
 435         kfree(si.name);
 436         kfree(si.value);
 437
 438         ocfs2_free_dir_lookup_result(&lookup);
 439
 440         if (inode_ac)
 441                 ocfs2_free_alloc_context(inode_ac);
 442
 443         if (data_ac)
 444                 ocfs2_free_alloc_context(data_ac);
 445
 446         if (meta_ac)
 447                 ocfs2_free_alloc_context(meta_ac);
 448
 449         /*
 450          * We should call iput after the i_mutex of the bitmap been
 451          * unlocked in ocfs2_free_alloc_context, or the
 452          * ocfs2_delete_inode will mutex_lock again.
 453          */
 454         if ((status < 0) && inode) {
 455                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 456                 clear_nlink(inode);
 457                 iput(inode);
 458         }
 459
 460         if (status)
 461                 mlog_errno(status);
 462
 463         return status;
 464 }
 465
 466 static int __ocfs2_mknod_locked(struct inode *dir,
 467                                 struct inode *inode,
 468                                 dev_t dev,
 469                                 struct buffer_head **new_fe_bh,
 470                                 struct buffer_head *parent_fe_bh,
 471                                 handle_t *handle,
 472                                 struct ocfs2_alloc_context *inode_ac,
 473                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 474 {
 475         int status = 0;
 476         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 477         struct ocfs2_dinode *fe = NULL;
 478         struct ocfs2_extent_list *fel;
 479         u16 feat;
 480
 481         *new_fe_bh = NULL;
 482
 483         /* populate as many fields early on as possible - many of
 484          * these are used by the support functions here and in
 485          * callers. */
 486         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 487         OCFS2_I(inode)->ip_blkno = fe_blkno;
 488         spin_lock(&osb->osb_lock);
 489         inode->i_generation = osb->s_next_generation++;
 490         spin_unlock(&osb->osb_lock);
 491
 492         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 493         if (!*new_fe_bh) {
 494                 status = -EIO;
 495                 mlog_errno(status);
 496                 goto leave;
 497         }
 498         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 499
 500         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 501                                          *new_fe_bh,
 502                                          OCFS2_JOURNAL_ACCESS_CREATE);
 503         if (status < 0) {
 504                 mlog_errno(status);
 505                 goto leave;
 506         }
 507
 508         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 509         memset(fe, 0, osb->sb->s_blocksize);
 510
 511         fe->i_generation = cpu_to_le32(inode->i_generation);
 512         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 513         fe->i_blkno = cpu_to_le64(fe_blkno);
 514         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 515         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 516         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 517         fe->i_uid = cpu_to_le32(inode->i_uid);
 518         fe->i_gid = cpu_to_le32(inode->i_gid);
 519         fe->i_mode = cpu_to_le16(inode->i_mode);
 520         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 521                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 522
 523         ocfs2_set_links_count(fe, inode->i_nlink);
 524
 525         fe->i_last_eb_blk = 0;
 526         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 527         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
 528         fe->i_atime = fe->i_ctime = fe->i_mtime =
 529                 cpu_to_le64(CURRENT_TIME.tv_sec);
 530         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 531                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 532         fe->i_dtime = 0;
 533
 534         /*
 535          * If supported, directories start with inline data. If inline
 536          * isn't supported, but indexing is, we start them as indexed.
 537          */
 538         feat = le16_to_cpu(fe->i_dyn_features);
 539         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 540                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 541
 542                 fe->id2.i_data.id_count = cpu_to_le16(
 543                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 544         } else {
 545                 fel = &fe->id2.i_list;
 546                 fel->l_tree_depth = 0;
 547                 fel->l_next_free_rec = 0;
 548                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 549         }
 550
 551         ocfs2_journal_dirty(handle, *new_fe_bh);
 552
 553         ocfs2_populate_inode(inode, fe, 1);
 554         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 555         if (!ocfs2_mount_local(osb)) {
 556                 status = ocfs2_create_new_inode_locks(inode);
 557                 if (status < 0)
 558                         mlog_errno(status);
 559         }
 560
 561         status = 0; /* error in ocfs2_create_new_inode_locks is not
 562                      * critical */
 563
 564 leave:
 565         if (status < 0) {
 566                 if (*new_fe_bh) {
 567                         brelse(*new_fe_bh);
 568                         *new_fe_bh = NULL;
 569                 }
 570         }
 571
 572         if (status)
 573                 mlog_errno(status);
 574         return status;
 575 }
 576
 577 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 578                               struct inode *dir,
 579                               struct inode *inode,
 580                               dev_t dev,
 581                               struct buffer_head **new_fe_bh,
 582                               struct buffer_head *parent_fe_bh,
 583                               handle_t *handle,
 584                               struct ocfs2_alloc_context *inode_ac)
 585 {
 586         int status = 0;
 587         u64 suballoc_loc, fe_blkno = 0;
 588         u16 suballoc_bit;
 589
 590         *new_fe_bh = NULL;
 591
 592         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 593                                        inode_ac, &suballoc_loc,
 594                                        &suballoc_bit, &fe_blkno);
 595         if (status < 0) {
 596                 mlog_errno(status);
 597                 return status;
 598         }
 599
 600         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 601                                     parent_fe_bh, handle, inode_ac,
 602                                     fe_blkno, suballoc_loc, suballoc_bit);
 603 }
 604
 605 static int ocfs2_mkdir(struct inode *dir,
 606                        struct dentry *dentry,
 607                        int mode)
 608 {
 609         int ret;
 610
 611         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 612                           OCFS2_I(dir)->ip_blkno, mode);
 613         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 614         if (ret)
 615                 mlog_errno(ret);
 616
 617         return ret;
 618 }
 619
 620 static int ocfs2_create(struct inode *dir,
 621                         struct dentry *dentry,
 622                         int mode,
 623                         struct nameidata *nd)
 624 {
 625         int ret;
 626
 627         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 628                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 629         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 630         if (ret)
 631                 mlog_errno(ret);
 632
 633         return ret;
 634 }
 635
 636 static int ocfs2_link(struct dentry *old_dentry,
 637                       struct inode *dir,
 638                       struct dentry *dentry)
 639 {
 640         handle_t *handle;
 641         struct inode *inode = old_dentry->d_inode;
 642         int err;
 643         struct buffer_head *fe_bh = NULL;
 644         struct buffer_head *parent_fe_bh = NULL;
 645         struct ocfs2_dinode *fe = NULL;
 646         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 647         struct ocfs2_dir_lookup_result lookup = { NULL, };
 648         sigset_t oldset;
 649
 650         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 651                          old_dentry->d_name.len, old_dentry->d_name.name,
 652                          dentry->d_name.len, dentry->d_name.name);
 653
 654         if (S_ISDIR(inode->i_mode))
 655                 return -EPERM;
 656
 657         dquot_initialize(dir);
 658
 659         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
 660         if (err < 0) {
 661                 if (err != -ENOENT)
 662                         mlog_errno(err);
 663                 return err;
 664         }
 665
 666         if (!dir->i_nlink) {
 667                 err = -ENOENT;
 668                 goto out;
 669         }
 670
 671         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 672                                         dentry->d_name.len);
 673         if (err)
 674                 goto out;
 675
 676         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 677                                            dentry->d_name.name,
 678                                            dentry->d_name.len, &lookup);
 679         if (err < 0) {
 680                 mlog_errno(err);
 681                 goto out;
 682         }
 683
 684         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 685         if (err < 0) {
 686                 if (err != -ENOENT)
 687                         mlog_errno(err);
 688                 goto out;
 689         }
 690
 691         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 692         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 693                 err = -EMLINK;
 694                 goto out_unlock_inode;
 695         }
 696
 697         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 698         if (IS_ERR(handle)) {
 699                 err = PTR_ERR(handle);
 700                 handle = NULL;
 701                 mlog_errno(err);
 702                 goto out_unlock_inode;
 703         }
 704
 705         /* Starting to change things, restart is no longer possible. */
 706         ocfs2_block_signals(&oldset);
 707
 708         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 709                                       OCFS2_JOURNAL_ACCESS_WRITE);
 710         if (err < 0) {
 711                 mlog_errno(err);
 712                 goto out_commit;
 713         }
 714
 715         inc_nlink(inode);
 716         inode->i_ctime = CURRENT_TIME;
 717         ocfs2_set_links_count(fe, inode->i_nlink);
 718         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 719         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 720         ocfs2_journal_dirty(handle, fe_bh);
 721
 722         err = ocfs2_add_entry(handle, dentry, inode,
 723                               OCFS2_I(inode)->ip_blkno,
 724                               parent_fe_bh, &lookup);
 725         if (err) {
 726                 ocfs2_add_links_count(fe, -1);
 727                 drop_nlink(inode);
 728                 mlog_errno(err);
 729                 goto out_commit;
 730         }
 731
 732         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 733         if (err) {
 734                 mlog_errno(err);
 735                 goto out_commit;
 736         }
 737
 738         ihold(inode);
 739         d_instantiate(dentry, inode);
 740
 741 out_commit:
 742         ocfs2_commit_trans(osb, handle);
 743         ocfs2_unblock_signals(&oldset);
 744 out_unlock_inode:
 745         ocfs2_inode_unlock(inode, 1);
 746
 747 out:
 748         ocfs2_inode_unlock(dir, 1);
 749
 750         brelse(fe_bh);
 751         brelse(parent_fe_bh);
 752
 753         ocfs2_free_dir_lookup_result(&lookup);
 754
 755         if (err)
 756                 mlog_errno(err);
 757
 758         return err;
 759 }
 760
 761 /*
 762  * Takes and drops an exclusive lock on the given dentry. This will
 763  * force other nodes to drop it.
 764  */
 765 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 766 {
 767         int ret;
 768
 769         ret = ocfs2_dentry_lock(dentry, 1);
 770         if (ret)
 771                 mlog_errno(ret);
 772         else
 773                 ocfs2_dentry_unlock(dentry, 1);
 774
 775         return ret;
 776 }
 777
 778 static inline int inode_is_unlinkable(struct inode *inode)
 779 {
 780         if (S_ISDIR(inode->i_mode)) {
 781                 if (inode->i_nlink == 2)
 782                         return 1;
 783                 return 0;
 784         }
 785
 786         if (inode->i_nlink == 1)
 787                 return 1;
 788         return 0;
 789 }
 790
 791 static int ocfs2_unlink(struct inode *dir,
 792                         struct dentry *dentry)
 793 {
 794         int status;
 795         int child_locked = 0;
 796         struct inode *inode = dentry->d_inode;
 797         struct inode *orphan_dir = NULL;
 798         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 799         u64 blkno;
 800         struct ocfs2_dinode *fe = NULL;
 801         struct buffer_head *fe_bh = NULL;
 802         struct buffer_head *parent_node_bh = NULL;
 803         handle_t *handle = NULL;
 804         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 805         struct ocfs2_dir_lookup_result lookup = { NULL, };
 806         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 807
 808         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 809                            dentry->d_name.name,
 810                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 811                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 812
 813         dquot_initialize(dir);
 814
 815         BUG_ON(dentry->d_parent->d_inode != dir);
 816
 817         if (inode == osb->root_inode)
 818                 return -EPERM;
 819
 820         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 821                                          OI_LS_PARENT);
 822         if (status < 0) {
 823                 if (status != -ENOENT)
 824                         mlog_errno(status);
 825                 return status;
 826         }
 827
 828         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 829                                           dentry->d_name.len, &blkno, dir,
 830                                           &lookup);
 831         if (status < 0) {
 832                 if (status != -ENOENT)
 833                         mlog_errno(status);
 834                 goto leave;
 835         }
 836
 837         if (OCFS2_I(inode)->ip_blkno != blkno) {
 838                 status = -ENOENT;
 839
 840                 trace_ocfs2_unlink_noent(
 841                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 842                                 (unsigned long long)blkno,
 843                                 OCFS2_I(inode)->ip_flags);
 844                 goto leave;
 845         }
 846
 847         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 848         if (status < 0) {
 849                 if (status != -ENOENT)
 850                         mlog_errno(status);
 851                 goto leave;
 852         }
 853         child_locked = 1;
 854
 855         if (S_ISDIR(inode->i_mode)) {
 856                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 857                         status = -ENOTEMPTY;
 858                         goto leave;
 859                 }
 860         }
 861
 862         status = ocfs2_remote_dentry_delete(dentry);
 863         if (status < 0) {
 864                 /* This remote delete should succeed under all normal
 865                  * circumstances. */
 866                 mlog_errno(status);
 867                 goto leave;
 868         }
 869
 870         if (inode_is_unlinkable(inode)) {
 871                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 872                                                   OCFS2_I(inode)->ip_blkno,
 873                                                   orphan_name, &orphan_insert);
 874                 if (status < 0) {
 875                         mlog_errno(status);
 876                         goto leave;
 877                 }
 878         }
 879
 880         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 881         if (IS_ERR(handle)) {
 882                 status = PTR_ERR(handle);
 883                 handle = NULL;
 884                 mlog_errno(status);
 885                 goto leave;
 886         }
 887
 888         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 889                                          OCFS2_JOURNAL_ACCESS_WRITE);
 890         if (status < 0) {
 891                 mlog_errno(status);
 892                 goto leave;
 893         }
 894
 895         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 896
 897         if (inode_is_unlinkable(inode)) {
 898                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
 899                                           &orphan_insert, orphan_dir);
 900                 if (status < 0) {
 901                         mlog_errno(status);
 902                         goto leave;
 903                 }
 904         }
 905
 906         /* delete the name from the parent dir */
 907         status = ocfs2_delete_entry(handle, dir, &lookup);
 908         if (status < 0) {
 909                 mlog_errno(status);
 910                 goto leave;
 911         }
 912
 913         if (S_ISDIR(inode->i_mode))
 914                 drop_nlink(inode);
 915         drop_nlink(inode);
 916         ocfs2_set_links_count(fe, inode->i_nlink);
 917         ocfs2_journal_dirty(handle, fe_bh);
 918
 919         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
 920         if (S_ISDIR(inode->i_mode))
 921                 drop_nlink(dir);
 922
 923         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
 924         if (status < 0) {
 925                 mlog_errno(status);
 926                 if (S_ISDIR(inode->i_mode))
 927                         inc_nlink(dir);
 928         }
 929
 930 leave:
 931         if (handle)
 932                 ocfs2_commit_trans(osb, handle);
 933
 934         if (child_locked)
 935                 ocfs2_inode_unlock(inode, 1);
 936
 937         ocfs2_inode_unlock(dir, 1);
 938
 939         if (orphan_dir) {
 940                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
 941                 ocfs2_inode_unlock(orphan_dir, 1);
 942                 mutex_unlock(&orphan_dir->i_mutex);
 943                 iput(orphan_dir);
 944         }
 945
 946         brelse(fe_bh);
 947         brelse(parent_node_bh);
 948
 949         ocfs2_free_dir_lookup_result(&orphan_insert);
 950         ocfs2_free_dir_lookup_result(&lookup);
 951
 952         if (status)
 953                 mlog_errno(status);
 954
 955         return status;
 956 }
 957
 958 /*
 959  * The only place this should be used is rename!
 960  * if they have the same id, then the 1st one is the only one locked.
 961  */
 962 static int ocfs2_double_lock(struct ocfs2_super *osb,
 963                              struct buffer_head **bh1,
 964                              struct inode *inode1,
 965                              struct buffer_head **bh2,
 966                              struct inode *inode2)
 967 {
 968         int status;
 969         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
 970         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
 971         struct buffer_head **tmpbh;
 972         struct inode *tmpinode;
 973
 974         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
 975                                 (unsigned long long)oi2->ip_blkno);
 976
 977         if (*bh1)
 978                 *bh1 = NULL;
 979         if (*bh2)
 980                 *bh2 = NULL;
 981
 982         /* we always want to lock the one with the lower lockid first. */
 983         if (oi1->ip_blkno != oi2->ip_blkno) {
 984                 if (oi1->ip_blkno < oi2->ip_blkno) {
 985                         /* switch id1 and id2 around */
 986                         tmpbh = bh2;
 987                         bh2 = bh1;
 988                         bh1 = tmpbh;
 989
 990                         tmpinode = inode2;
 991                         inode2 = inode1;
 992                         inode1 = tmpinode;
 993                 }
 994                 /* lock id2 */
 995                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
 996                                                  OI_LS_RENAME1);
 997                 if (status < 0) {
 998                         if (status != -ENOENT)
 999                                 mlog_errno(status);
1000                         goto bail;
1001                 }
1002         }
1003
1004         /* lock id1 */
1005         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1006         if (status < 0) {
1007                 /*
1008                  * An error return must mean that no cluster locks
1009                  * were held on function exit.
1010                  */
1011                 if (oi1->ip_blkno != oi2->ip_blkno) {
1012                         ocfs2_inode_unlock(inode2, 1);
1013                         brelse(*bh2);
1014                         *bh2 = NULL;
1015                 }
1016
1017                 if (status != -ENOENT)
1018                         mlog_errno(status);
1019         }
1020
1021         trace_ocfs2_double_lock_end(
1022                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1023                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1024
1025 bail:
1026         if (status)
1027                 mlog_errno(status);
1028         return status;
1029 }
1030
1031 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1032 {
1033         ocfs2_inode_unlock(inode1, 1);
1034
1035         if (inode1 != inode2)
1036                 ocfs2_inode_unlock(inode2, 1);
1037 }
1038
1039 static int ocfs2_rename(struct inode *old_dir,
1040                         struct dentry *old_dentry,
1041                         struct inode *new_dir,
1042                         struct dentry *new_dentry)
1043 {
1044         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1045         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1046         struct inode *old_inode = old_dentry->d_inode;
1047         struct inode *new_inode = new_dentry->d_inode;
1048         struct inode *orphan_dir = NULL;
1049         struct ocfs2_dinode *newfe = NULL;
1050         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1051         struct buffer_head *newfe_bh = NULL;
1052         struct buffer_head *old_inode_bh = NULL;
1053         struct ocfs2_super *osb = NULL;
1054         u64 newfe_blkno, old_de_ino;
1055         handle_t *handle = NULL;
1056         struct buffer_head *old_dir_bh = NULL;
1057         struct buffer_head *new_dir_bh = NULL;
1058         nlink_t old_dir_nlink = old_dir->i_nlink;
1059         struct ocfs2_dinode *old_di;
1060         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1061         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1062         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1063         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1064         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1065
1066         /* At some point it might be nice to break this function up a
1067          * bit. */
1068
1069         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1070                            old_dentry->d_name.len, old_dentry->d_name.name,
1071                            new_dentry->d_name.len, new_dentry->d_name.name);
1072
1073         dquot_initialize(old_dir);
1074         dquot_initialize(new_dir);
1075
1076         osb = OCFS2_SB(old_dir->i_sb);
1077
1078         if (new_inode) {
1079                 if (!igrab(new_inode))
1080                         BUG();
1081         }
1082
1083         /* Assume a directory hierarchy thusly:
1084          * a/b/c
1085          * a/d
1086          * a,b,c, and d are all directories.
1087          *
1088          * from cwd of 'a' on both nodes:
1089          * node1: mv b/c d
1090          * node2: mv d   b/c
1091          *
1092          * And that's why, just like the VFS, we need a file system
1093          * rename lock. */
1094         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1095                 status = ocfs2_rename_lock(osb);
1096                 if (status < 0) {
1097                         mlog_errno(status);
1098                         goto bail;
1099                 }
1100                 rename_lock = 1;
1101         }
1102
1103         /* if old and new are the same, this'll just do one lock. */
1104         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1105                                    &new_dir_bh, new_dir);
1106         if (status < 0) {
1107                 mlog_errno(status);
1108                 goto bail;
1109         }
1110         parents_locked = 1;
1111
1112         /* make sure both dirs have bhs
1113          * get an extra ref on old_dir_bh if old==new */
1114         if (!new_dir_bh) {
1115                 if (old_dir_bh) {
1116                         new_dir_bh = old_dir_bh;
1117                         get_bh(new_dir_bh);
1118                 } else {
1119                         mlog(ML_ERROR, "no old_dir_bh!\n");
1120                         status = -EIO;
1121                         goto bail;
1122                 }
1123         }
1124
1125         /*
1126          * Aside from allowing a meta data update, the locking here
1127          * also ensures that the downconvert thread on other nodes
1128          * won't have to concurrently downconvert the inode and the
1129          * dentry locks.
1130          */
1131         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1132                                          OI_LS_PARENT);
1133         if (status < 0) {
1134                 if (status != -ENOENT)
1135                         mlog_errno(status);
1136                 goto bail;
1137         }
1138         old_child_locked = 1;
1139
1140         status = ocfs2_remote_dentry_delete(old_dentry);
1141         if (status < 0) {
1142                 mlog_errno(status);
1143                 goto bail;
1144         }
1145
1146         if (S_ISDIR(old_inode->i_mode)) {
1147                 u64 old_inode_parent;
1148
1149                 update_dot_dot = 1;
1150                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1151                                                   old_inode,
1152                                                   &old_inode_dot_dot_res);
1153                 if (status) {
1154                         status = -EIO;
1155                         goto bail;
1156                 }
1157
1158                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1159                         status = -EIO;
1160                         goto bail;
1161                 }
1162
1163                 if (!new_inode && new_dir != old_dir &&
1164                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1165                         status = -EMLINK;
1166                         goto bail;
1167                 }
1168         }
1169
1170         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1171                                             old_dentry->d_name.len,
1172                                             &old_de_ino);
1173         if (status) {
1174                 status = -ENOENT;
1175                 goto bail;
1176         }
1177
1178         /*
1179          *  Check for inode number is _not_ due to possible IO errors.
1180          *  We might rmdir the source, keep it as pwd of some process
1181          *  and merrily kill the link to whatever was created under the
1182          *  same name. Goodbye sticky bit ;-<
1183          */
1184         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1185                 status = -ENOENT;
1186                 goto bail;
1187         }
1188
1189         /* check if the target already exists (in which case we need
1190          * to delete it */
1191         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1192                                           new_dentry->d_name.len,
1193                                           &newfe_blkno, new_dir,
1194                                           &target_lookup_res);
1195         /* The only error we allow here is -ENOENT because the new
1196          * file not existing is perfectly valid. */
1197         if ((status < 0) && (status != -ENOENT)) {
1198                 /* If we cannot find the file specified we should just */
1199                 /* return the error... */
1200                 mlog_errno(status);
1201                 goto bail;
1202         }
1203         if (status == 0)
1204                 target_exists = 1;
1205
1206         if (!target_exists && new_inode) {
1207                 /*
1208                  * Target was unlinked by another node while we were
1209                  * waiting to get to ocfs2_rename(). There isn't
1210                  * anything we can do here to help the situation, so
1211                  * bubble up the appropriate error.
1212                  */
1213                 status = -ENOENT;
1214                 goto bail;
1215         }
1216
1217         /* In case we need to overwrite an existing file, we blow it
1218          * away first */
1219         if (target_exists) {
1220                 /* VFS didn't think there existed an inode here, but
1221                  * someone else in the cluster must have raced our
1222                  * rename to create one. Today we error cleanly, in
1223                  * the future we should consider calling iget to build
1224                  * a new struct inode for this entry. */
1225                 if (!new_inode) {
1226                         status = -EACCES;
1227
1228                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1229                                                 new_dentry->d_name.name);
1230                         goto bail;
1231                 }
1232
1233                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1234                         status = -EACCES;
1235
1236                         trace_ocfs2_rename_disagree(
1237                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1238                              (unsigned long long)newfe_blkno,
1239                              OCFS2_I(new_inode)->ip_flags);
1240                         goto bail;
1241                 }
1242
1243                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1244                 if (status < 0) {
1245                         if (status != -ENOENT)
1246                                 mlog_errno(status);
1247                         goto bail;
1248                 }
1249                 new_child_locked = 1;
1250
1251                 status = ocfs2_remote_dentry_delete(new_dentry);
1252                 if (status < 0) {
1253                         mlog_errno(status);
1254                         goto bail;
1255                 }
1256
1257                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1258
1259                 trace_ocfs2_rename_over_existing(
1260                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1261                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1262
1263                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1264                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1265                                                 OCFS2_I(new_inode)->ip_blkno,
1266                                                 orphan_name, &orphan_insert);
1267                         if (status < 0) {
1268                                 mlog_errno(status);
1269                                 goto bail;
1270                         }
1271                 }
1272         } else {
1273                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1274
1275                 status = ocfs2_check_dir_for_entry(new_dir,
1276                                                    new_dentry->d_name.name,
1277                                                    new_dentry->d_name.len);
1278                 if (status)
1279                         goto bail;
1280
1281                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1282                                                       new_dentry->d_name.name,
1283                                                       new_dentry->d_name.len,
1284                                                       &target_insert);
1285                 if (status < 0) {
1286                         mlog_errno(status);
1287                         goto bail;
1288                 }
1289         }
1290
1291         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1292         if (IS_ERR(handle)) {
1293                 status = PTR_ERR(handle);
1294                 handle = NULL;
1295                 mlog_errno(status);
1296                 goto bail;
1297         }
1298
1299         if (target_exists) {
1300                 if (S_ISDIR(new_inode->i_mode)) {
1301                         if (new_inode->i_nlink != 2 ||
1302                             !ocfs2_empty_dir(new_inode)) {
1303                                 status = -ENOTEMPTY;
1304                                 goto bail;
1305                         }
1306                 }
1307                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1308                                                  newfe_bh,
1309                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1310                 if (status < 0) {
1311                         mlog_errno(status);
1312                         goto bail;
1313                 }
1314
1315                 if (S_ISDIR(new_inode->i_mode) ||
1316                     (ocfs2_read_links_count(newfe) == 1)) {
1317                         status = ocfs2_orphan_add(osb, handle, new_inode,
1318                                                   newfe_bh, orphan_name,
1319                                                   &orphan_insert, orphan_dir);
1320                         if (status < 0) {
1321                                 mlog_errno(status);
1322                                 goto bail;
1323                         }
1324                 }
1325
1326                 /* change the dirent to point to the correct inode */
1327                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1328                                             old_inode);
1329                 if (status < 0) {
1330                         mlog_errno(status);
1331                         goto bail;
1332                 }
1333                 new_dir->i_version++;
1334
1335                 if (S_ISDIR(new_inode->i_mode))
1336                         ocfs2_set_links_count(newfe, 0);
1337                 else
1338                         ocfs2_add_links_count(newfe, -1);
1339                 ocfs2_journal_dirty(handle, newfe_bh);
1340         } else {
1341                 /* if the name was not found in new_dir, add it now */
1342                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1343                                          OCFS2_I(old_inode)->ip_blkno,
1344                                          new_dir_bh, &target_insert);
1345         }
1346
1347         old_inode->i_ctime = CURRENT_TIME;
1348         mark_inode_dirty(old_inode);
1349
1350         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1351                                          old_inode_bh,
1352                                          OCFS2_JOURNAL_ACCESS_WRITE);
1353         if (status >= 0) {
1354                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1355
1356                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1357                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1358                 ocfs2_journal_dirty(handle, old_inode_bh);
1359         } else
1360                 mlog_errno(status);
1361
1362         /*
1363          * Now that the name has been added to new_dir, remove the old name.
1364          *
1365          * We don't keep any directory entry context around until now
1366          * because the insert might have changed the type of directory
1367          * we're dealing with.
1368          */
1369         status = ocfs2_find_entry(old_dentry->d_name.name,
1370                                   old_dentry->d_name.len, old_dir,
1371                                   &old_entry_lookup);
1372         if (status)
1373                 goto bail;
1374
1375         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1376         if (status < 0) {
1377                 mlog_errno(status);
1378                 goto bail;
1379         }
1380
1381         if (new_inode) {
1382                 new_inode->i_nlink--;
1383                 new_inode->i_ctime = CURRENT_TIME;
1384         }
1385         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1386
1387         if (update_dot_dot) {
1388                 status = ocfs2_update_entry(old_inode, handle,
1389                                             &old_inode_dot_dot_res, new_dir);
1390                 old_dir->i_nlink--;
1391                 if (new_inode) {
1392                         new_inode->i_nlink--;
1393                 } else {
1394                         inc_nlink(new_dir);
1395                         mark_inode_dirty(new_dir);
1396                 }
1397         }
1398         mark_inode_dirty(old_dir);
1399         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1400         if (new_inode) {
1401                 mark_inode_dirty(new_inode);
1402                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1403         }
1404
1405         if (old_dir != new_dir) {
1406                 /* Keep the same times on both directories.*/
1407                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1408
1409                 /*
1410                  * This will also pick up the i_nlink change from the
1411                  * block above.
1412                  */
1413                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1414         }
1415
1416         if (old_dir_nlink != old_dir->i_nlink) {
1417                 if (!old_dir_bh) {
1418                         mlog(ML_ERROR, "need to change nlink for old dir "
1419                              "%llu from %d to %d but bh is NULL!\n",
1420                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1421                              (int)old_dir_nlink, old_dir->i_nlink);
1422                 } else {
1423                         struct ocfs2_dinode *fe;
1424                         status = ocfs2_journal_access_di(handle,
1425                                                          INODE_CACHE(old_dir),
1426                                                          old_dir_bh,
1427                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1428                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1429                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1430                         ocfs2_journal_dirty(handle, old_dir_bh);
1431                 }
1432         }
1433         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1434         status = 0;
1435 bail:
1436         if (rename_lock)
1437                 ocfs2_rename_unlock(osb);
1438
1439         if (handle)
1440                 ocfs2_commit_trans(osb, handle);
1441
1442         if (parents_locked)
1443                 ocfs2_double_unlock(old_dir, new_dir);
1444
1445         if (old_child_locked)
1446                 ocfs2_inode_unlock(old_inode, 1);
1447
1448         if (new_child_locked)
1449                 ocfs2_inode_unlock(new_inode, 1);
1450
1451         if (orphan_dir) {
1452                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1453                 ocfs2_inode_unlock(orphan_dir, 1);
1454                 mutex_unlock(&orphan_dir->i_mutex);
1455                 iput(orphan_dir);
1456         }
1457
1458         if (new_inode)
1459                 sync_mapping_buffers(old_inode->i_mapping);
1460
1461         if (new_inode)
1462                 iput(new_inode);
1463
1464         ocfs2_free_dir_lookup_result(&target_lookup_res);
1465         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1466         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1467         ocfs2_free_dir_lookup_result(&orphan_insert);
1468         ocfs2_free_dir_lookup_result(&target_insert);
1469
1470         brelse(newfe_bh);
1471         brelse(old_inode_bh);
1472         brelse(old_dir_bh);
1473         brelse(new_dir_bh);
1474
1475         if (status)
1476                 mlog_errno(status);
1477
1478         return status;
1479 }
1480
1481 /*
1482  * we expect i_size = strlen(symname). Copy symname into the file
1483  * data, including the null terminator.
1484  */
1485 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1486                                      handle_t *handle,
1487                                      struct inode *inode,
1488                                      const char *symname)
1489 {
1490         struct buffer_head **bhs = NULL;
1491         const char *c;
1492         struct super_block *sb = osb->sb;
1493         u64 p_blkno, p_blocks;
1494         int virtual, blocks, status, i, bytes_left;
1495
1496         bytes_left = i_size_read(inode) + 1;
1497         /* we can't trust i_blocks because we're actually going to
1498          * write i_size + 1 bytes. */
1499         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1500
1501         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1502                                         i_size_read(inode), blocks);
1503
1504         /* Sanity check -- make sure we're going to fit. */
1505         if (bytes_left >
1506             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1507                 status = -EIO;
1508                 mlog_errno(status);
1509                 goto bail;
1510         }
1511
1512         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1513         if (!bhs) {
1514                 status = -ENOMEM;
1515                 mlog_errno(status);
1516                 goto bail;
1517         }
1518
1519         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1520                                              NULL);
1521         if (status < 0) {
1522                 mlog_errno(status);
1523                 goto bail;
1524         }
1525
1526         /* links can never be larger than one cluster so we know this
1527          * is all going to be contiguous, but do a sanity check
1528          * anyway. */
1529         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1530                 status = -EIO;
1531                 mlog_errno(status);
1532                 goto bail;
1533         }
1534
1535         virtual = 0;
1536         while(bytes_left > 0) {
1537                 c = &symname[virtual * sb->s_blocksize];
1538
1539                 bhs[virtual] = sb_getblk(sb, p_blkno);
1540                 if (!bhs[virtual]) {
1541                         status = -ENOMEM;
1542                         mlog_errno(status);
1543                         goto bail;
1544                 }
1545                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1546                                               bhs[virtual]);
1547
1548                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1549                                               bhs[virtual],
1550                                               OCFS2_JOURNAL_ACCESS_CREATE);
1551                 if (status < 0) {
1552                         mlog_errno(status);
1553                         goto bail;
1554                 }
1555
1556                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1557
1558                 memcpy(bhs[virtual]->b_data, c,
1559                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1560                        bytes_left);
1561
1562                 ocfs2_journal_dirty(handle, bhs[virtual]);
1563
1564                 virtual++;
1565                 p_blkno++;
1566                 bytes_left -= sb->s_blocksize;
1567         }
1568
1569         status = 0;
1570 bail:
1571
1572         if (bhs) {
1573                 for(i = 0; i < blocks; i++)
1574                         brelse(bhs[i]);
1575                 kfree(bhs);
1576         }
1577
1578         if (status)
1579                 mlog_errno(status);
1580         return status;
1581 }
1582
1583 static int ocfs2_symlink(struct inode *dir,
1584                          struct dentry *dentry,
1585                          const char *symname)
1586 {
1587         int status, l, credits;
1588         u64 newsize;
1589         struct ocfs2_super *osb = NULL;
1590         struct inode *inode = NULL;
1591         struct super_block *sb;
1592         struct buffer_head *new_fe_bh = NULL;
1593         struct buffer_head *parent_fe_bh = NULL;
1594         struct ocfs2_dinode *fe = NULL;
1595         struct ocfs2_dinode *dirfe;
1596         handle_t *handle = NULL;
1597         struct ocfs2_alloc_context *inode_ac = NULL;
1598         struct ocfs2_alloc_context *data_ac = NULL;
1599         struct ocfs2_alloc_context *xattr_ac = NULL;
1600         int want_clusters = 0;
1601         int xattr_credits = 0;
1602         struct ocfs2_security_xattr_info si = {
1603                 .enable = 1,
1604         };
1605         int did_quota = 0, did_quota_inode = 0;
1606         struct ocfs2_dir_lookup_result lookup = { NULL, };
1607         sigset_t oldset;
1608         int did_block_signals = 0;
1609
1610         trace_ocfs2_symlink_begin(dir, dentry, symname,
1611                                   dentry->d_name.len, dentry->d_name.name);
1612
1613         dquot_initialize(dir);
1614
1615         sb = dir->i_sb;
1616         osb = OCFS2_SB(sb);
1617
1618         l = strlen(symname) + 1;
1619
1620         credits = ocfs2_calc_symlink_credits(sb);
1621
1622         /* lock the parent directory */
1623         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1624         if (status < 0) {
1625                 if (status != -ENOENT)
1626                         mlog_errno(status);
1627                 return status;
1628         }
1629
1630         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1631         if (!ocfs2_read_links_count(dirfe)) {
1632                 /* can't make a file in a deleted directory. */
1633                 status = -ENOENT;
1634                 goto bail;
1635         }
1636
1637         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1638                                            dentry->d_name.len);
1639         if (status)
1640                 goto bail;
1641
1642         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1643                                               dentry->d_name.name,
1644                                               dentry->d_name.len, &lookup);
1645         if (status < 0) {
1646                 mlog_errno(status);
1647                 goto bail;
1648         }
1649
1650         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1651         if (status < 0) {
1652                 if (status != -ENOSPC)
1653                         mlog_errno(status);
1654                 goto bail;
1655         }
1656
1657         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1658         if (!inode) {
1659                 status = -ENOMEM;
1660                 mlog_errno(status);
1661                 goto bail;
1662         }
1663
1664         /* get security xattr */
1665         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1666         if (status) {
1667                 if (status == -EOPNOTSUPP)
1668                         si.enable = 0;
1669                 else {
1670                         mlog_errno(status);
1671                         goto bail;
1672                 }
1673         }
1674
1675         /* calculate meta data/clusters for setting security xattr */
1676         if (si.enable) {
1677                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1678                                                   &xattr_credits, &xattr_ac);
1679                 if (status < 0) {
1680                         mlog_errno(status);
1681                         goto bail;
1682                 }
1683         }
1684
1685         /* don't reserve bitmap space for fast symlinks. */
1686         if (l > ocfs2_fast_symlink_chars(sb))
1687                 want_clusters += 1;
1688
1689         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1690         if (status < 0) {
1691                 if (status != -ENOSPC)
1692                         mlog_errno(status);
1693                 goto bail;
1694         }
1695
1696         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1697         if (IS_ERR(handle)) {
1698                 status = PTR_ERR(handle);
1699                 handle = NULL;
1700                 mlog_errno(status);
1701                 goto bail;
1702         }
1703
1704         /* Starting to change things, restart is no longer possible. */
1705         ocfs2_block_signals(&oldset);
1706         did_block_signals = 1;
1707
1708         status = dquot_alloc_inode(inode);
1709         if (status)
1710                 goto bail;
1711         did_quota_inode = 1;
1712
1713         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1714                                    dentry->d_name.name,
1715                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1716                                    inode->i_mode);
1717
1718         status = ocfs2_mknod_locked(osb, dir, inode,
1719                                     0, &new_fe_bh, parent_fe_bh, handle,
1720                                     inode_ac);
1721         if (status < 0) {
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1727         inode->i_rdev = 0;
1728         newsize = l - 1;
1729         if (l > ocfs2_fast_symlink_chars(sb)) {
1730                 u32 offset = 0;
1731
1732                 inode->i_op = &ocfs2_symlink_inode_operations;
1733                 status = dquot_alloc_space_nodirty(inode,
1734                     ocfs2_clusters_to_bytes(osb->sb, 1));
1735                 if (status)
1736                         goto bail;
1737                 did_quota = 1;
1738                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1739                                               new_fe_bh,
1740                                               handle, data_ac, NULL,
1741                                               NULL);
1742                 if (status < 0) {
1743                         if (status != -ENOSPC && status != -EINTR) {
1744                                 mlog(ML_ERROR,
1745                                      "Failed to extend file to %llu\n",
1746                                      (unsigned long long)newsize);
1747                                 mlog_errno(status);
1748                                 status = -ENOSPC;
1749                         }
1750                         goto bail;
1751                 }
1752                 i_size_write(inode, newsize);
1753                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1754         } else {
1755                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1756                 memcpy((char *) fe->id2.i_symlink, symname, l);
1757                 i_size_write(inode, newsize);
1758                 inode->i_blocks = 0;
1759         }
1760
1761         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1762         if (status < 0) {
1763                 mlog_errno(status);
1764                 goto bail;
1765         }
1766
1767         if (!ocfs2_inode_is_fast_symlink(inode)) {
1768                 status = ocfs2_create_symlink_data(osb, handle, inode,
1769                                                    symname);
1770                 if (status < 0) {
1771                         mlog_errno(status);
1772                         goto bail;
1773                 }
1774         }
1775
1776         if (si.enable) {
1777                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1778                                                  xattr_ac, data_ac);
1779                 if (status < 0) {
1780                         mlog_errno(status);
1781                         goto bail;
1782                 }
1783         }
1784
1785         /*
1786          * Do this before adding the entry to the directory. We add
1787          * also set d_op after success so that ->d_iput() will cleanup
1788          * the dentry lock even if ocfs2_add_entry() fails below.
1789          */
1790         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1791         if (status) {
1792                 mlog_errno(status);
1793                 goto bail;
1794         }
1795
1796         status = ocfs2_add_entry(handle, dentry, inode,
1797                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1798                                  &lookup);
1799         if (status < 0) {
1800                 mlog_errno(status);
1801                 goto bail;
1802         }
1803
1804         insert_inode_hash(inode);
1805         d_instantiate(dentry, inode);
1806 bail:
1807         if (status < 0 && did_quota)
1808                 dquot_free_space_nodirty(inode,
1809                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1810         if (status < 0 && did_quota_inode)
1811                 dquot_free_inode(inode);
1812         if (handle)
1813                 ocfs2_commit_trans(osb, handle);
1814
1815         ocfs2_inode_unlock(dir, 1);
1816         if (did_block_signals)
1817                 ocfs2_unblock_signals(&oldset);
1818
1819         brelse(new_fe_bh);
1820         brelse(parent_fe_bh);
1821         kfree(si.name);
1822         kfree(si.value);
1823         ocfs2_free_dir_lookup_result(&lookup);
1824         if (inode_ac)
1825                 ocfs2_free_alloc_context(inode_ac);
1826         if (data_ac)
1827                 ocfs2_free_alloc_context(data_ac);
1828         if (xattr_ac)
1829                 ocfs2_free_alloc_context(xattr_ac);
1830         if ((status < 0) && inode) {
1831                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1832                 clear_nlink(inode);
1833                 iput(inode);
1834         }
1835
1836         if (status)
1837                 mlog_errno(status);
1838
1839         return status;
1840 }
1841
1842 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1843 {
1844         int status, namelen;
1845
1846         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1847                            (long long)blkno);
1848         if (namelen <= 0) {
1849                 if (namelen)
1850                         status = namelen;
1851                 else
1852                         status = -EINVAL;
1853                 mlog_errno(status);
1854                 goto bail;
1855         }
1856         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1857                 status = -EINVAL;
1858                 mlog_errno(status);
1859                 goto bail;
1860         }
1861
1862         trace_ocfs2_blkno_stringify(blkno, name, namelen);
1863
1864         status = 0;
1865 bail:
1866         if (status < 0)
1867                 mlog_errno(status);
1868         return status;
1869 }
1870
1871 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1872                                         struct inode **ret_orphan_dir,
1873                                         struct buffer_head **ret_orphan_dir_bh)
1874 {
1875         struct inode *orphan_dir_inode;
1876         struct buffer_head *orphan_dir_bh = NULL;
1877         int ret = 0;
1878
1879         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1880                                                        ORPHAN_DIR_SYSTEM_INODE,
1881                                                        osb->slot_num);
1882         if (!orphan_dir_inode) {
1883                 ret = -ENOENT;
1884                 mlog_errno(ret);
1885                 return ret;
1886         }
1887
1888         mutex_lock(&orphan_dir_inode->i_mutex);
1889
1890         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1891         if (ret < 0) {
1892                 mutex_unlock(&orphan_dir_inode->i_mutex);
1893                 iput(orphan_dir_inode);
1894
1895                 mlog_errno(ret);
1896                 return ret;
1897         }
1898
1899         *ret_orphan_dir = orphan_dir_inode;
1900         *ret_orphan_dir_bh = orphan_dir_bh;
1901
1902         return 0;
1903 }
1904
1905 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1906                                       struct buffer_head *orphan_dir_bh,
1907                                       u64 blkno,
1908                                       char *name,
1909                                       struct ocfs2_dir_lookup_result *lookup)
1910 {
1911         int ret;
1912         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1913
1914         ret = ocfs2_blkno_stringify(blkno, name);
1915         if (ret < 0) {
1916                 mlog_errno(ret);
1917                 return ret;
1918         }
1919
1920         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1921                                            orphan_dir_bh, name,
1922                                            OCFS2_ORPHAN_NAMELEN, lookup);
1923         if (ret < 0) {
1924                 mlog_errno(ret);
1925                 return ret;
1926         }
1927
1928         return 0;
1929 }
1930
1931 /**
1932  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1933  * insertion of an orphan.
1934  * @osb: ocfs2 file system
1935  * @ret_orphan_dir: Orphan dir inode - returned locked!
1936  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1937  * @lookup: dir lookup result, to be passed back into functions like
1938  *          ocfs2_orphan_add
1939  *
1940  * Returns zero on success and the ret_orphan_dir, name and lookup
1941  * fields will be populated.
1942  *
1943  * Returns non-zero on failure.
1944  */
1945 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1946                                     struct inode **ret_orphan_dir,
1947                                     u64 blkno,
1948                                     char *name,
1949                                     struct ocfs2_dir_lookup_result *lookup)
1950 {
1951         struct inode *orphan_dir_inode = NULL;
1952         struct buffer_head *orphan_dir_bh = NULL;
1953         int ret = 0;
1954
1955         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1956                                            &orphan_dir_bh);
1957         if (ret < 0) {
1958                 mlog_errno(ret);
1959                 return ret;
1960         }
1961
1962         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1963                                          blkno, name, lookup);
1964         if (ret < 0) {
1965                 mlog_errno(ret);
1966                 goto out;
1967         }
1968
1969         *ret_orphan_dir = orphan_dir_inode;
1970
1971 out:
1972         brelse(orphan_dir_bh);
1973
1974         if (ret) {
1975                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1976                 mutex_unlock(&orphan_dir_inode->i_mutex);
1977                 iput(orphan_dir_inode);
1978         }
1979
1980         if (ret)
1981                 mlog_errno(ret);
1982         return ret;
1983 }
1984
1985 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1986                             handle_t *handle,
1987                             struct inode *inode,
1988                             struct buffer_head *fe_bh,
1989                             char *name,
1990                             struct ocfs2_dir_lookup_result *lookup,
1991                             struct inode *orphan_dir_inode)
1992 {
1993         struct buffer_head *orphan_dir_bh = NULL;
1994         int status = 0;
1995         struct ocfs2_dinode *orphan_fe;
1996         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1997
1998         trace_ocfs2_orphan_add_begin(
1999                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2000
2001         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2002         if (status < 0) {
2003                 mlog_errno(status);
2004                 goto leave;
2005         }
2006
2007         status = ocfs2_journal_access_di(handle,
2008                                          INODE_CACHE(orphan_dir_inode),
2009                                          orphan_dir_bh,
2010                                          OCFS2_JOURNAL_ACCESS_WRITE);
2011         if (status < 0) {
2012                 mlog_errno(status);
2013                 goto leave;
2014         }
2015
2016         /* we're a cluster, and nlink can change on disk from
2017          * underneath us... */
2018         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2019         if (S_ISDIR(inode->i_mode))
2020                 ocfs2_add_links_count(orphan_fe, 1);
2021         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2022         ocfs2_journal_dirty(handle, orphan_dir_bh);
2023
2024         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2025                                    OCFS2_ORPHAN_NAMELEN, inode,
2026                                    OCFS2_I(inode)->ip_blkno,
2027                                    orphan_dir_bh, lookup);
2028         if (status < 0) {
2029                 mlog_errno(status);
2030                 goto leave;
2031         }
2032
2033         /*
2034          * We're going to journal the change of i_flags and i_orphaned_slot.
2035          * It's safe anyway, though some callers may duplicate the journaling.
2036          * Journaling within the func just make the logic look more
2037          * straightforward.
2038          */
2039         status = ocfs2_journal_access_di(handle,
2040                                          INODE_CACHE(inode),
2041                                          fe_bh,
2042                                          OCFS2_JOURNAL_ACCESS_WRITE);
2043         if (status < 0) {
2044                 mlog_errno(status);
2045                 goto leave;
2046         }
2047
2048         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2049         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2050
2051         /* Record which orphan dir our inode now resides
2052          * in. delete_inode will use this to determine which orphan
2053          * dir to lock. */
2054         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2055
2056         ocfs2_journal_dirty(handle, fe_bh);
2057
2058         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2059                                    osb->slot_num);
2060
2061 leave:
2062         brelse(orphan_dir_bh);
2063
2064         if (status)
2065                 mlog_errno(status);
2066         return status;
2067 }
2068
2069 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2070 int ocfs2_orphan_del(struct ocfs2_super *osb,
2071                      handle_t *handle,
2072                      struct inode *orphan_dir_inode,
2073                      struct inode *inode,
2074                      struct buffer_head *orphan_dir_bh)
2075 {
2076         char name[OCFS2_ORPHAN_NAMELEN + 1];
2077         struct ocfs2_dinode *orphan_fe;
2078         int status = 0;
2079         struct ocfs2_dir_lookup_result lookup = { NULL, };
2080
2081         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2082         if (status < 0) {
2083                 mlog_errno(status);
2084                 goto leave;
2085         }
2086
2087         trace_ocfs2_orphan_del(
2088              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2089              name, OCFS2_ORPHAN_NAMELEN);
2090
2091         /* find it's spot in the orphan directory */
2092         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2093                                   &lookup);
2094         if (status) {
2095                 mlog_errno(status);
2096                 goto leave;
2097         }
2098
2099         /* remove it from the orphan directory */
2100         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2101         if (status < 0) {
2102                 mlog_errno(status);
2103                 goto leave;
2104         }
2105
2106         status = ocfs2_journal_access_di(handle,
2107                                          INODE_CACHE(orphan_dir_inode),
2108                                          orphan_dir_bh,
2109                                          OCFS2_JOURNAL_ACCESS_WRITE);
2110         if (status < 0) {
2111                 mlog_errno(status);
2112                 goto leave;
2113         }
2114
2115         /* do the i_nlink dance! :) */
2116         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2117         if (S_ISDIR(inode->i_mode))
2118                 ocfs2_add_links_count(orphan_fe, -1);
2119         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2120         ocfs2_journal_dirty(handle, orphan_dir_bh);
2121
2122 leave:
2123         ocfs2_free_dir_lookup_result(&lookup);
2124
2125         if (status)
2126                 mlog_errno(status);
2127         return status;
2128 }
2129
2130 /**
2131  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to recieve a newly
2132  * allocated file. This is different from the typical 'add to orphan dir'
2133  * operation in that the inode does not yet exist. This is a problem because
2134  * the orphan dir stringifies the inode block number to come up with it's
2135  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2136  * problem. This function works around it by calling deeper into the orphan
2137  * and suballoc code than other callers. Use this only by necessity.
2138  * @dir: The directory which this inode will ultimately wind up under - not the
2139  * orphan dir!
2140  * @dir_bh: buffer_head the @dir inode block
2141  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2142  * with the string to be used for orphan dirent. Pass back to the orphan dir
2143  * code.
2144  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2145  * dir code.
2146  * @ret_di_blkno: block number where the new inode will be allocated.
2147  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2148  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2149  *
2150  * Returns zero on success and the ret_orphan_dir, name and lookup
2151  * fields will be populated.
2152  *
2153  * Returns non-zero on failure.
2154  */
2155 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2156                                         struct buffer_head *dir_bh,
2157                                         char *orphan_name,
2158                                         struct inode **ret_orphan_dir,
2159                                         u64 *ret_di_blkno,
2160                                         struct ocfs2_dir_lookup_result *orphan_insert,
2161                                         struct ocfs2_alloc_context **ret_inode_ac)
2162 {
2163         int ret;
2164         u64 di_blkno;
2165         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2166         struct inode *orphan_dir = NULL;
2167         struct buffer_head *orphan_dir_bh = NULL;
2168         struct ocfs2_alloc_context *inode_ac = NULL;
2169
2170         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2171         if (ret < 0) {
2172                 mlog_errno(ret);
2173                 return ret;
2174         }
2175
2176         /* reserve an inode spot */
2177         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2178         if (ret < 0) {
2179                 if (ret != -ENOSPC)
2180                         mlog_errno(ret);
2181                 goto out;
2182         }
2183
2184         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2185                                        &di_blkno);
2186         if (ret) {
2187                 mlog_errno(ret);
2188                 goto out;
2189         }
2190
2191         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2192                                          di_blkno, orphan_name, orphan_insert);
2193         if (ret < 0) {
2194                 mlog_errno(ret);
2195                 goto out;
2196         }
2197
2198 out:
2199         if (ret == 0) {
2200                 *ret_orphan_dir = orphan_dir;
2201                 *ret_di_blkno = di_blkno;
2202                 *ret_inode_ac = inode_ac;
2203                 /*
2204                  * orphan_name and orphan_insert are already up to
2205                  * date via prepare_orphan_dir
2206                  */
2207         } else {
2208                 /* Unroll reserve_new_inode* */
2209                 if (inode_ac)
2210                         ocfs2_free_alloc_context(inode_ac);
2211
2212                 /* Unroll orphan dir locking */
2213                 mutex_unlock(&orphan_dir->i_mutex);
2214                 ocfs2_inode_unlock(orphan_dir, 1);
2215                 iput(orphan_dir);
2216         }
2217
2218         brelse(orphan_dir_bh);
2219
2220         return 0;
2221 }
2222
2223 int ocfs2_create_inode_in_orphan(struct inode *dir,
2224                                  int mode,
2225                                  struct inode **new_inode)
2226 {
2227         int status, did_quota_inode = 0;
2228         struct inode *inode = NULL;
2229         struct inode *orphan_dir = NULL;
2230         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2231         struct ocfs2_dinode *di = NULL;
2232         handle_t *handle = NULL;
2233         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2234         struct buffer_head *parent_di_bh = NULL;
2235         struct buffer_head *new_di_bh = NULL;
2236         struct ocfs2_alloc_context *inode_ac = NULL;
2237         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2238         u64 uninitialized_var(di_blkno), suballoc_loc;
2239         u16 suballoc_bit;
2240
2241         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2242         if (status < 0) {
2243                 if (status != -ENOENT)
2244                         mlog_errno(status);
2245                 return status;
2246         }
2247
2248         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2249                                               orphan_name, &orphan_dir,
2250                                               &di_blkno, &orphan_insert, &inode_ac);
2251         if (status < 0) {
2252                 if (status != -ENOSPC)
2253                         mlog_errno(status);
2254                 goto leave;
2255         }
2256
2257         inode = ocfs2_get_init_inode(dir, mode);
2258         if (!inode) {
2259                 status = -ENOMEM;
2260                 mlog_errno(status);
2261                 goto leave;
2262         }
2263
2264         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2265         if (IS_ERR(handle)) {
2266                 status = PTR_ERR(handle);
2267                 handle = NULL;
2268                 mlog_errno(status);
2269                 goto leave;
2270         }
2271
2272         status = dquot_alloc_inode(inode);
2273         if (status)
2274                 goto leave;
2275         did_quota_inode = 1;
2276
2277         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2278                                               &suballoc_loc,
2279                                               &suballoc_bit, di_blkno);
2280         if (status < 0) {
2281                 mlog_errno(status);
2282                 goto leave;
2283         }
2284
2285         inode->i_nlink = 0;
2286         /* do the real work now. */
2287         status = __ocfs2_mknod_locked(dir, inode,
2288                                       0, &new_di_bh, parent_di_bh, handle,
2289                                       inode_ac, di_blkno, suballoc_loc,
2290                                       suballoc_bit);
2291         if (status < 0) {
2292                 mlog_errno(status);
2293                 goto leave;
2294         }
2295
2296         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2297         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2298                                   &orphan_insert, orphan_dir);
2299         if (status < 0) {
2300                 mlog_errno(status);
2301                 goto leave;
2302         }
2303
2304         /* get open lock so that only nodes can't remove it from orphan dir. */
2305         status = ocfs2_open_lock(inode);
2306         if (status < 0)
2307                 mlog_errno(status);
2308
2309         insert_inode_hash(inode);
2310 leave:
2311         if (status < 0 && did_quota_inode)
2312                 dquot_free_inode(inode);
2313         if (handle)
2314                 ocfs2_commit_trans(osb, handle);
2315
2316         if (orphan_dir) {
2317                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2318                 ocfs2_inode_unlock(orphan_dir, 1);
2319                 mutex_unlock(&orphan_dir->i_mutex);
2320                 iput(orphan_dir);
2321         }
2322
2323         if ((status < 0) && inode) {
2324                 clear_nlink(inode);
2325                 iput(inode);
2326         }
2327
2328         if (inode_ac)
2329                 ocfs2_free_alloc_context(inode_ac);
2330
2331         brelse(new_di_bh);
2332
2333         if (!status)
2334                 *new_inode = inode;
2335
2336         ocfs2_free_dir_lookup_result(&orphan_insert);
2337
2338         ocfs2_inode_unlock(dir, 1);
2339         brelse(parent_di_bh);
2340         return status;
2341 }
2342
2343 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2344                                    struct inode *inode,
2345                                    struct dentry *dentry)
2346 {
2347         int status = 0;
2348         struct buffer_head *parent_di_bh = NULL;
2349         handle_t *handle = NULL;
2350         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2351         struct ocfs2_dinode *dir_di, *di;
2352         struct inode *orphan_dir_inode = NULL;
2353         struct buffer_head *orphan_dir_bh = NULL;
2354         struct buffer_head *di_bh = NULL;
2355         struct ocfs2_dir_lookup_result lookup = { NULL, };
2356
2357         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2358                                 dentry->d_name.len, dentry->d_name.name,
2359                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2360                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2361
2362         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2363         if (status < 0) {
2364                 if (status != -ENOENT)
2365                         mlog_errno(status);
2366                 return status;
2367         }
2368
2369         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2370         if (!dir_di->i_links_count) {
2371                 /* can't make a file in a deleted directory. */
2372                 status = -ENOENT;
2373                 goto leave;
2374         }
2375
2376         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2377                                            dentry->d_name.len);
2378         if (status)
2379                 goto leave;
2380
2381         /* get a spot inside the dir. */
2382         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2383                                               dentry->d_name.name,
2384                                               dentry->d_name.len, &lookup);
2385         if (status < 0) {
2386                 mlog_errno(status);
2387                 goto leave;
2388         }
2389
2390         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2391                                                        ORPHAN_DIR_SYSTEM_INODE,
2392                                                        osb->slot_num);
2393         if (!orphan_dir_inode) {
2394                 status = -EEXIST;
2395                 mlog_errno(status);
2396                 goto leave;
2397         }
2398
2399         mutex_lock(&orphan_dir_inode->i_mutex);
2400
2401         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2402         if (status < 0) {
2403                 mlog_errno(status);
2404                 mutex_unlock(&orphan_dir_inode->i_mutex);
2405                 iput(orphan_dir_inode);
2406                 goto leave;
2407         }
2408
2409         status = ocfs2_read_inode_block(inode, &di_bh);
2410         if (status < 0) {
2411                 mlog_errno(status);
2412                 goto orphan_unlock;
2413         }
2414
2415         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2416         if (IS_ERR(handle)) {
2417                 status = PTR_ERR(handle);
2418                 handle = NULL;
2419                 mlog_errno(status);
2420                 goto orphan_unlock;
2421         }
2422
2423         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2424                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2425         if (status < 0) {
2426                 mlog_errno(status);
2427                 goto out_commit;
2428         }
2429
2430         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2431                                   orphan_dir_bh);
2432         if (status < 0) {
2433                 mlog_errno(status);
2434                 goto out_commit;
2435         }
2436
2437         di = (struct ocfs2_dinode *)di_bh->b_data;
2438         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2439         di->i_orphaned_slot = 0;
2440         inode->i_nlink = 1;
2441         ocfs2_set_links_count(di, inode->i_nlink);
2442         ocfs2_journal_dirty(handle, di_bh);
2443
2444         status = ocfs2_add_entry(handle, dentry, inode,
2445                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2446                                  &lookup);
2447         if (status < 0) {
2448                 mlog_errno(status);
2449                 goto out_commit;
2450         }
2451
2452         status = ocfs2_dentry_attach_lock(dentry, inode,
2453                                           OCFS2_I(dir)->ip_blkno);
2454         if (status) {
2455                 mlog_errno(status);
2456                 goto out_commit;
2457         }
2458
2459         d_instantiate(dentry, inode);
2460         status = 0;
2461 out_commit:
2462         ocfs2_commit_trans(osb, handle);
2463 orphan_unlock:
2464         ocfs2_inode_unlock(orphan_dir_inode, 1);
2465         mutex_unlock(&orphan_dir_inode->i_mutex);
2466         iput(orphan_dir_inode);
2467 leave:
2468
2469         ocfs2_inode_unlock(dir, 1);
2470
2471         brelse(di_bh);
2472         brelse(parent_di_bh);
2473         brelse(orphan_dir_bh);
2474
2475         ocfs2_free_dir_lookup_result(&lookup);
2476
2477         if (status)
2478                 mlog_errno(status);
2479
2480         return status;
2481 }
2482
2483 const struct inode_operations ocfs2_dir_iops = {
2484         .create         = ocfs2_create,
2485         .lookup         = ocfs2_lookup,
2486         .link           = ocfs2_link,
2487         .unlink         = ocfs2_unlink,
2488         .rmdir          = ocfs2_unlink,
2489         .symlink        = ocfs2_symlink,
2490         .mkdir          = ocfs2_mkdir,
2491         .mknod          = ocfs2_mknod,
2492         .rename         = ocfs2_rename,
2493         .setattr        = ocfs2_setattr,
2494         .getattr        = ocfs2_getattr,
2495         .permission     = ocfs2_permission,
2496         .setxattr       = generic_setxattr,
2497         .getxattr       = generic_getxattr,
2498         .listxattr      = ocfs2_listxattr,
2499         .removexattr    = generic_removexattr,
2500         .fiemap         = ocfs2_fiemap,
2501 };