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                                     bool dio);
  84
  85 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  86                             handle_t *handle,
  87                             struct inode *inode,
  88                             struct buffer_head *fe_bh,
  89                             char *name,
  90                             struct ocfs2_dir_lookup_result *lookup,
  91                             struct inode *orphan_dir_inode,
  92                             bool dio);
  93
  94 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  95                                      handle_t *handle,
  96                                      struct inode *inode,
  97                                      const char *symname);
  98
  99 static int ocfs2_double_lock(struct ocfs2_super *osb,
 100                              struct buffer_head **bh1,
 101                              struct inode *inode1,
 102                              struct buffer_head **bh2,
 103                              struct inode *inode2,
 104                              int rename);
 105
 106 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
 107 /* An orphan dir name is an 8 byte value, printed as a hex string */
 108 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
 109 #define OCFS2_DIO_ORPHAN_PREFIX "dio-"
 110 #define OCFS2_DIO_ORPHAN_PREFIX_LEN 4
 111
 112 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 113                                    unsigned int flags)
 114 {
 115         int status;
 116         u64 blkno;
 117         struct inode *inode = NULL;
 118         struct dentry *ret;
 119         struct ocfs2_inode_info *oi;
 120
 121         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 122                            dentry->d_name.name,
 123                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 124
 125         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 126                 ret = ERR_PTR(-ENAMETOOLONG);
 127                 goto bail;
 128         }
 129
 130         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 131         if (status < 0) {
 132                 if (status != -ENOENT)
 133                         mlog_errno(status);
 134                 ret = ERR_PTR(status);
 135                 goto bail;
 136         }
 137
 138         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 139                                             dentry->d_name.len, &blkno);
 140         if (status < 0)
 141                 goto bail_add;
 142
 143         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 144         if (IS_ERR(inode)) {
 145                 ret = ERR_PTR(-EACCES);
 146                 goto bail_unlock;
 147         }
 148
 149         oi = OCFS2_I(inode);
 150         /* Clear any orphaned state... If we were able to look up the
 151          * inode from a directory, it certainly can't be orphaned. We
 152          * might have the bad state from a node which intended to
 153          * orphan this inode but crashed before it could commit the
 154          * unlink. */
 155         spin_lock(&oi->ip_lock);
 156         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 157         spin_unlock(&oi->ip_lock);
 158
 159 bail_add:
 160         ret = d_splice_alias(inode, dentry);
 161
 162         if (inode) {
 163                 /*
 164                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 165                  * dentry, it will d_move() it on top of ourse. The
 166                  * return value will indicate this however, so in
 167                  * those cases, we switch them around for the locking
 168                  * code.
 169                  *
 170                  * NOTE: This dentry already has ->d_op set from
 171                  * ocfs2_get_parent() and ocfs2_get_dentry()
 172                  */
 173                 if (!IS_ERR_OR_NULL(ret))
 174                         dentry = ret;
 175
 176                 status = ocfs2_dentry_attach_lock(dentry, inode,
 177                                                   OCFS2_I(dir)->ip_blkno);
 178                 if (status) {
 179                         mlog_errno(status);
 180                         ret = ERR_PTR(status);
 181                         goto bail_unlock;
 182                 }
 183         } else
 184                 ocfs2_dentry_attach_gen(dentry);
 185
 186 bail_unlock:
 187         /* Don't drop the cluster lock until *after* the d_add --
 188          * unlink on another node will message us to remove that
 189          * dentry under this lock so otherwise we can race this with
 190          * the downconvert thread and have a stale dentry. */
 191         ocfs2_inode_unlock(dir, 0);
 192
 193 bail:
 194
 195         trace_ocfs2_lookup_ret(ret);
 196
 197         return ret;
 198 }
 199
 200 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 201 {
 202         struct inode *inode;
 203         int status;
 204
 205         inode = new_inode(dir->i_sb);
 206         if (!inode) {
 207                 mlog(ML_ERROR, "new_inode failed!\n");
 208                 return ERR_PTR(-ENOMEM);
 209         }
 210
 211         /* populate as many fields early on as possible - many of
 212          * these are used by the support functions here and in
 213          * callers. */
 214         if (S_ISDIR(mode))
 215                 set_nlink(inode, 2);
 216         inode_init_owner(inode, dir, mode);
 217         status = dquot_initialize(inode);
 218         if (status)
 219                 return ERR_PTR(status);
 220
 221         return inode;
 222 }
 223
 224 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 225                 struct dentry *dentry, struct inode *inode)
 226 {
 227         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 228
 229         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 230         ocfs2_lock_res_free(&dl->dl_lockres);
 231         BUG_ON(dl->dl_count != 1);
 232         spin_lock(&dentry_attach_lock);
 233         dentry->d_fsdata = NULL;
 234         spin_unlock(&dentry_attach_lock);
 235         kfree(dl);
 236         iput(inode);
 237 }
 238
 239 static int ocfs2_mknod(struct inode *dir,
 240                        struct dentry *dentry,
 241                        umode_t mode,
 242                        dev_t dev)
 243 {
 244         int status = 0;
 245         struct buffer_head *parent_fe_bh = NULL;
 246         handle_t *handle = NULL;
 247         struct ocfs2_super *osb;
 248         struct ocfs2_dinode *dirfe;
 249         struct buffer_head *new_fe_bh = NULL;
 250         struct inode *inode = NULL;
 251         struct ocfs2_alloc_context *inode_ac = NULL;
 252         struct ocfs2_alloc_context *data_ac = NULL;
 253         struct ocfs2_alloc_context *meta_ac = NULL;
 254         int want_clusters = 0;
 255         int want_meta = 0;
 256         int xattr_credits = 0;
 257         struct ocfs2_security_xattr_info si = {
 258                 .enable = 1,
 259         };
 260         int did_quota_inode = 0;
 261         struct ocfs2_dir_lookup_result lookup = { NULL, };
 262         sigset_t oldset;
 263         int did_block_signals = 0;
 264         struct posix_acl *default_acl = NULL, *acl = NULL;
 265         struct ocfs2_dentry_lock *dl = NULL;
 266
 267         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 268                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 269                           (unsigned long)dev, mode);
 270
 271         status = dquot_initialize(dir);
 272         if (status) {
 273                 mlog_errno(status);
 274                 return status;
 275         }
 276
 277         /* get our super block */
 278         osb = OCFS2_SB(dir->i_sb);
 279
 280         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 281         if (status < 0) {
 282                 if (status != -ENOENT)
 283                         mlog_errno(status);
 284                 return status;
 285         }
 286
 287         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 288                 status = -EMLINK;
 289                 goto leave;
 290         }
 291
 292         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 293         if (!ocfs2_read_links_count(dirfe)) {
 294                 /* can't make a file in a deleted directory. */
 295                 status = -ENOENT;
 296                 goto leave;
 297         }
 298
 299         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 300                                            dentry->d_name.len);
 301         if (status)
 302                 goto leave;
 303
 304         /* get a spot inside the dir. */
 305         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 306                                               dentry->d_name.name,
 307                                               dentry->d_name.len, &lookup);
 308         if (status < 0) {
 309                 mlog_errno(status);
 310                 goto leave;
 311         }
 312
 313         /* reserve an inode spot */
 314         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 315         if (status < 0) {
 316                 if (status != -ENOSPC)
 317                         mlog_errno(status);
 318                 goto leave;
 319         }
 320
 321         inode = ocfs2_get_init_inode(dir, mode);
 322         if (IS_ERR(inode)) {
 323                 status = PTR_ERR(inode);
 324                 inode = NULL;
 325                 mlog_errno(status);
 326                 goto leave;
 327         }
 328
 329         /* get security xattr */
 330         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 331         if (status) {
 332                 if (status == -EOPNOTSUPP)
 333                         si.enable = 0;
 334                 else {
 335                         mlog_errno(status);
 336                         goto leave;
 337                 }
 338         }
 339
 340         /* calculate meta data/clusters for setting security and acl xattr */
 341         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 342                                        &si, &want_clusters,
 343                                        &xattr_credits, &want_meta);
 344         if (status < 0) {
 345                 mlog_errno(status);
 346                 goto leave;
 347         }
 348
 349         /* Reserve a cluster if creating an extent based directory. */
 350         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 351                 want_clusters += 1;
 352
 353                 /* Dir indexing requires extra space as well */
 354                 if (ocfs2_supports_indexed_dirs(osb))
 355                         want_meta++;
 356         }
 357
 358         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 359         if (status < 0) {
 360                 if (status != -ENOSPC)
 361                         mlog_errno(status);
 362                 goto leave;
 363         }
 364
 365         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 366         if (status < 0) {
 367                 if (status != -ENOSPC)
 368                         mlog_errno(status);
 369                 goto leave;
 370         }
 371
 372         status = posix_acl_create(dir, &mode, &default_acl, &acl);
 373         if (status) {
 374                 mlog_errno(status);
 375                 goto leave;
 376         }
 377
 378         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 379                                                             S_ISDIR(mode),
 380                                                             xattr_credits));
 381         if (IS_ERR(handle)) {
 382                 status = PTR_ERR(handle);
 383                 handle = NULL;
 384                 mlog_errno(status);
 385                 goto leave;
 386         }
 387
 388         /* Starting to change things, restart is no longer possible. */
 389         ocfs2_block_signals(&oldset);
 390         did_block_signals = 1;
 391
 392         status = dquot_alloc_inode(inode);
 393         if (status)
 394                 goto leave;
 395         did_quota_inode = 1;
 396
 397         /* do the real work now. */
 398         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 399                                     &new_fe_bh, parent_fe_bh, handle,
 400                                     inode_ac);
 401         if (status < 0) {
 402                 mlog_errno(status);
 403                 goto leave;
 404         }
 405
 406         if (S_ISDIR(mode)) {
 407                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 408                                             new_fe_bh, data_ac, meta_ac);
 409                 if (status < 0) {
 410                         mlog_errno(status);
 411                         goto leave;
 412                 }
 413
 414                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 415                                                  parent_fe_bh,
 416                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 417                 if (status < 0) {
 418                         mlog_errno(status);
 419                         goto leave;
 420                 }
 421                 ocfs2_add_links_count(dirfe, 1);
 422                 ocfs2_journal_dirty(handle, parent_fe_bh);
 423                 inc_nlink(dir);
 424         }
 425
 426         if (default_acl) {
 427                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
 428                                        ACL_TYPE_DEFAULT, default_acl,
 429                                        meta_ac, data_ac);
 430         }
 431         if (!status && acl) {
 432                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
 433                                        ACL_TYPE_ACCESS, acl,
 434                                        meta_ac, data_ac);
 435         }
 436
 437         if (status < 0) {
 438                 mlog_errno(status);
 439                 goto leave;
 440         }
 441
 442         if (si.enable) {
 443                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 444                                                  meta_ac, data_ac);
 445                 if (status < 0) {
 446                         mlog_errno(status);
 447                         goto leave;
 448                 }
 449         }
 450
 451         /*
 452          * Do this before adding the entry to the directory. We add
 453          * also set d_op after success so that ->d_iput() will cleanup
 454          * the dentry lock even if ocfs2_add_entry() fails below.
 455          */
 456         status = ocfs2_dentry_attach_lock(dentry, inode,
 457                                           OCFS2_I(dir)->ip_blkno);
 458         if (status) {
 459                 mlog_errno(status);
 460                 goto leave;
 461         }
 462
 463         dl = dentry->d_fsdata;
 464
 465         status = ocfs2_add_entry(handle, dentry, inode,
 466                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 467                                  &lookup);
 468         if (status < 0) {
 469                 mlog_errno(status);
 470                 goto leave;
 471         }
 472
 473         insert_inode_hash(inode);
 474         d_instantiate(dentry, inode);
 475         status = 0;
 476 leave:
 477         if (default_acl)
 478                 posix_acl_release(default_acl);
 479         if (acl)
 480                 posix_acl_release(acl);
 481         if (status < 0 && did_quota_inode)
 482                 dquot_free_inode(inode);
 483         if (handle)
 484                 ocfs2_commit_trans(osb, handle);
 485
 486         ocfs2_inode_unlock(dir, 1);
 487         if (did_block_signals)
 488                 ocfs2_unblock_signals(&oldset);
 489
 490         brelse(new_fe_bh);
 491         brelse(parent_fe_bh);
 492         kfree(si.value);
 493
 494         ocfs2_free_dir_lookup_result(&lookup);
 495
 496         if (inode_ac)
 497                 ocfs2_free_alloc_context(inode_ac);
 498
 499         if (data_ac)
 500                 ocfs2_free_alloc_context(data_ac);
 501
 502         if (meta_ac)
 503                 ocfs2_free_alloc_context(meta_ac);
 504
 505         /*
 506          * We should call iput after the i_mutex of the bitmap been
 507          * unlocked in ocfs2_free_alloc_context, or the
 508          * ocfs2_delete_inode will mutex_lock again.
 509          */
 510         if ((status < 0) && inode) {
 511                 if (dl)
 512                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 513
 514                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 515                 clear_nlink(inode);
 516                 iput(inode);
 517         }
 518
 519         if (status)
 520                 mlog_errno(status);
 521
 522         return status;
 523 }
 524
 525 static int __ocfs2_mknod_locked(struct inode *dir,
 526                                 struct inode *inode,
 527                                 dev_t dev,
 528                                 struct buffer_head **new_fe_bh,
 529                                 struct buffer_head *parent_fe_bh,
 530                                 handle_t *handle,
 531                                 struct ocfs2_alloc_context *inode_ac,
 532                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 533 {
 534         int status = 0;
 535         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 536         struct ocfs2_dinode *fe = NULL;
 537         struct ocfs2_extent_list *fel;
 538         u16 feat;
 539         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 540
 541         *new_fe_bh = NULL;
 542
 543         /* populate as many fields early on as possible - many of
 544          * these are used by the support functions here and in
 545          * callers. */
 546         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 547         OCFS2_I(inode)->ip_blkno = fe_blkno;
 548         spin_lock(&osb->osb_lock);
 549         inode->i_generation = osb->s_next_generation++;
 550         spin_unlock(&osb->osb_lock);
 551
 552         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 553         if (!*new_fe_bh) {
 554                 status = -ENOMEM;
 555                 mlog_errno(status);
 556                 goto leave;
 557         }
 558         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 559
 560         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 561                                          *new_fe_bh,
 562                                          OCFS2_JOURNAL_ACCESS_CREATE);
 563         if (status < 0) {
 564                 mlog_errno(status);
 565                 goto leave;
 566         }
 567
 568         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 569         memset(fe, 0, osb->sb->s_blocksize);
 570
 571         fe->i_generation = cpu_to_le32(inode->i_generation);
 572         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 573         fe->i_blkno = cpu_to_le64(fe_blkno);
 574         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 575         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 576         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 577         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 578         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 579         fe->i_mode = cpu_to_le16(inode->i_mode);
 580         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 581                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 582
 583         ocfs2_set_links_count(fe, inode->i_nlink);
 584
 585         fe->i_last_eb_blk = 0;
 586         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 587         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 588         fe->i_atime = fe->i_ctime = fe->i_mtime =
 589                 cpu_to_le64(CURRENT_TIME.tv_sec);
 590         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 591                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 592         fe->i_dtime = 0;
 593
 594         /*
 595          * If supported, directories start with inline data. If inline
 596          * isn't supported, but indexing is, we start them as indexed.
 597          */
 598         feat = le16_to_cpu(fe->i_dyn_features);
 599         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 600                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 601
 602                 fe->id2.i_data.id_count = cpu_to_le16(
 603                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 604         } else {
 605                 fel = &fe->id2.i_list;
 606                 fel->l_tree_depth = 0;
 607                 fel->l_next_free_rec = 0;
 608                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 609         }
 610
 611         ocfs2_journal_dirty(handle, *new_fe_bh);
 612
 613         ocfs2_populate_inode(inode, fe, 1);
 614         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 615         if (!ocfs2_mount_local(osb)) {
 616                 status = ocfs2_create_new_inode_locks(inode);
 617                 if (status < 0)
 618                         mlog_errno(status);
 619         }
 620
 621         oi->i_sync_tid = handle->h_transaction->t_tid;
 622         oi->i_datasync_tid = handle->h_transaction->t_tid;
 623
 624 leave:
 625         if (status < 0) {
 626                 if (*new_fe_bh) {
 627                         brelse(*new_fe_bh);
 628                         *new_fe_bh = NULL;
 629                 }
 630         }
 631
 632         if (status)
 633                 mlog_errno(status);
 634         return status;
 635 }
 636
 637 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 638                               struct inode *dir,
 639                               struct inode *inode,
 640                               dev_t dev,
 641                               struct buffer_head **new_fe_bh,
 642                               struct buffer_head *parent_fe_bh,
 643                               handle_t *handle,
 644                               struct ocfs2_alloc_context *inode_ac)
 645 {
 646         int status = 0;
 647         u64 suballoc_loc, fe_blkno = 0;
 648         u16 suballoc_bit;
 649
 650         *new_fe_bh = NULL;
 651
 652         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 653                                        inode_ac, &suballoc_loc,
 654                                        &suballoc_bit, &fe_blkno);
 655         if (status < 0) {
 656                 mlog_errno(status);
 657                 return status;
 658         }
 659
 660         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 661                                     parent_fe_bh, handle, inode_ac,
 662                                     fe_blkno, suballoc_loc, suballoc_bit);
 663 }
 664
 665 static int ocfs2_mkdir(struct inode *dir,
 666                        struct dentry *dentry,
 667                        umode_t mode)
 668 {
 669         int ret;
 670
 671         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 672                           OCFS2_I(dir)->ip_blkno, mode);
 673         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 674         if (ret)
 675                 mlog_errno(ret);
 676
 677         return ret;
 678 }
 679
 680 static int ocfs2_create(struct inode *dir,
 681                         struct dentry *dentry,
 682                         umode_t mode,
 683                         bool excl)
 684 {
 685         int ret;
 686
 687         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 688                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 689         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 690         if (ret)
 691                 mlog_errno(ret);
 692
 693         return ret;
 694 }
 695
 696 static int ocfs2_link(struct dentry *old_dentry,
 697                       struct inode *dir,
 698                       struct dentry *dentry)
 699 {
 700         handle_t *handle;
 701         struct inode *inode = d_inode(old_dentry);
 702         struct inode *old_dir = d_inode(old_dentry->d_parent);
 703         int err;
 704         struct buffer_head *fe_bh = NULL;
 705         struct buffer_head *old_dir_bh = NULL;
 706         struct buffer_head *parent_fe_bh = NULL;
 707         struct ocfs2_dinode *fe = NULL;
 708         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 709         struct ocfs2_dir_lookup_result lookup = { NULL, };
 710         sigset_t oldset;
 711         u64 old_de_ino;
 712
 713         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 714                          old_dentry->d_name.len, old_dentry->d_name.name,
 715                          dentry->d_name.len, dentry->d_name.name);
 716
 717         if (S_ISDIR(inode->i_mode))
 718                 return -EPERM;
 719
 720         err = dquot_initialize(dir);
 721         if (err) {
 722                 mlog_errno(err);
 723                 return err;
 724         }
 725
 726         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 727                         &parent_fe_bh, dir, 0);
 728         if (err < 0) {
 729                 if (err != -ENOENT)
 730                         mlog_errno(err);
 731                 return err;
 732         }
 733
 734         /* make sure both dirs have bhs
 735          * get an extra ref on old_dir_bh if old==new */
 736         if (!parent_fe_bh) {
 737                 if (old_dir_bh) {
 738                         parent_fe_bh = old_dir_bh;
 739                         get_bh(parent_fe_bh);
 740                 } else {
 741                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 742                         err = -EIO;
 743                         goto out;
 744                 }
 745         }
 746
 747         if (!dir->i_nlink) {
 748                 err = -ENOENT;
 749                 goto out;
 750         }
 751
 752         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 753                         old_dentry->d_name.len, &old_de_ino);
 754         if (err) {
 755                 err = -ENOENT;
 756                 goto out;
 757         }
 758
 759         /*
 760          * Check whether another node removed the source inode while we
 761          * were in the vfs.
 762          */
 763         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 764                 err = -ENOENT;
 765                 goto out;
 766         }
 767
 768         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 769                                         dentry->d_name.len);
 770         if (err)
 771                 goto out;
 772
 773         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 774                                            dentry->d_name.name,
 775                                            dentry->d_name.len, &lookup);
 776         if (err < 0) {
 777                 mlog_errno(err);
 778                 goto out;
 779         }
 780
 781         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 782         if (err < 0) {
 783                 if (err != -ENOENT)
 784                         mlog_errno(err);
 785                 goto out;
 786         }
 787
 788         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 789         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 790                 err = -EMLINK;
 791                 goto out_unlock_inode;
 792         }
 793
 794         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 795         if (IS_ERR(handle)) {
 796                 err = PTR_ERR(handle);
 797                 handle = NULL;
 798                 mlog_errno(err);
 799                 goto out_unlock_inode;
 800         }
 801
 802         /* Starting to change things, restart is no longer possible. */
 803         ocfs2_block_signals(&oldset);
 804
 805         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 806                                       OCFS2_JOURNAL_ACCESS_WRITE);
 807         if (err < 0) {
 808                 mlog_errno(err);
 809                 goto out_commit;
 810         }
 811
 812         inc_nlink(inode);
 813         inode->i_ctime = CURRENT_TIME;
 814         ocfs2_set_links_count(fe, inode->i_nlink);
 815         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 816         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 817         ocfs2_journal_dirty(handle, fe_bh);
 818
 819         err = ocfs2_add_entry(handle, dentry, inode,
 820                               OCFS2_I(inode)->ip_blkno,
 821                               parent_fe_bh, &lookup);
 822         if (err) {
 823                 ocfs2_add_links_count(fe, -1);
 824                 drop_nlink(inode);
 825                 mlog_errno(err);
 826                 goto out_commit;
 827         }
 828
 829         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 830         if (err) {
 831                 mlog_errno(err);
 832                 goto out_commit;
 833         }
 834
 835         ihold(inode);
 836         d_instantiate(dentry, inode);
 837
 838 out_commit:
 839         ocfs2_commit_trans(osb, handle);
 840         ocfs2_unblock_signals(&oldset);
 841 out_unlock_inode:
 842         ocfs2_inode_unlock(inode, 1);
 843
 844 out:
 845         ocfs2_double_unlock(old_dir, dir);
 846
 847         brelse(fe_bh);
 848         brelse(parent_fe_bh);
 849         brelse(old_dir_bh);
 850
 851         ocfs2_free_dir_lookup_result(&lookup);
 852
 853         if (err)
 854                 mlog_errno(err);
 855
 856         return err;
 857 }
 858
 859 /*
 860  * Takes and drops an exclusive lock on the given dentry. This will
 861  * force other nodes to drop it.
 862  */
 863 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 864 {
 865         int ret;
 866
 867         ret = ocfs2_dentry_lock(dentry, 1);
 868         if (ret)
 869                 mlog_errno(ret);
 870         else
 871                 ocfs2_dentry_unlock(dentry, 1);
 872
 873         return ret;
 874 }
 875
 876 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 877 {
 878         if (S_ISDIR(inode->i_mode)) {
 879                 if (inode->i_nlink == 2)
 880                         return 1;
 881                 return 0;
 882         }
 883
 884         if (inode->i_nlink == 1)
 885                 return 1;
 886         return 0;
 887 }
 888
 889 static int ocfs2_unlink(struct inode *dir,
 890                         struct dentry *dentry)
 891 {
 892         int status;
 893         int child_locked = 0;
 894         bool is_unlinkable = false;
 895         struct inode *inode = d_inode(dentry);
 896         struct inode *orphan_dir = NULL;
 897         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 898         u64 blkno;
 899         struct ocfs2_dinode *fe = NULL;
 900         struct buffer_head *fe_bh = NULL;
 901         struct buffer_head *parent_node_bh = NULL;
 902         handle_t *handle = NULL;
 903         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 904         struct ocfs2_dir_lookup_result lookup = { NULL, };
 905         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 906
 907         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 908                            dentry->d_name.name,
 909                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 910                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 911
 912         status = dquot_initialize(dir);
 913         if (status) {
 914                 mlog_errno(status);
 915                 return status;
 916         }
 917
 918         BUG_ON(d_inode(dentry->d_parent) != dir);
 919
 920         if (inode == osb->root_inode)
 921                 return -EPERM;
 922
 923         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 924                                          OI_LS_PARENT);
 925         if (status < 0) {
 926                 if (status != -ENOENT)
 927                         mlog_errno(status);
 928                 return status;
 929         }
 930
 931         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 932                                           dentry->d_name.len, &blkno, dir,
 933                                           &lookup);
 934         if (status < 0) {
 935                 if (status != -ENOENT)
 936                         mlog_errno(status);
 937                 goto leave;
 938         }
 939
 940         if (OCFS2_I(inode)->ip_blkno != blkno) {
 941                 status = -ENOENT;
 942
 943                 trace_ocfs2_unlink_noent(
 944                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 945                                 (unsigned long long)blkno,
 946                                 OCFS2_I(inode)->ip_flags);
 947                 goto leave;
 948         }
 949
 950         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 951         if (status < 0) {
 952                 if (status != -ENOENT)
 953                         mlog_errno(status);
 954                 goto leave;
 955         }
 956         child_locked = 1;
 957
 958         if (S_ISDIR(inode->i_mode)) {
 959                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 960                         status = -ENOTEMPTY;
 961                         goto leave;
 962                 }
 963         }
 964
 965         status = ocfs2_remote_dentry_delete(dentry);
 966         if (status < 0) {
 967                 /* This remote delete should succeed under all normal
 968                  * circumstances. */
 969                 mlog_errno(status);
 970                 goto leave;
 971         }
 972
 973         if (ocfs2_inode_is_unlinkable(inode)) {
 974                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 975                                                   OCFS2_I(inode)->ip_blkno,
 976                                                   orphan_name, &orphan_insert,
 977                                                   false);
 978                 if (status < 0) {
 979                         mlog_errno(status);
 980                         goto leave;
 981                 }
 982                 is_unlinkable = true;
 983         }
 984
 985         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 986         if (IS_ERR(handle)) {
 987                 status = PTR_ERR(handle);
 988                 handle = NULL;
 989                 mlog_errno(status);
 990                 goto leave;
 991         }
 992
 993         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 994                                          OCFS2_JOURNAL_ACCESS_WRITE);
 995         if (status < 0) {
 996                 mlog_errno(status);
 997                 goto leave;
 998         }
 999
1000         fe = (struct ocfs2_dinode *) fe_bh->b_data;
1001
1002         /* delete the name from the parent dir */
1003         status = ocfs2_delete_entry(handle, dir, &lookup);
1004         if (status < 0) {
1005                 mlog_errno(status);
1006                 goto leave;
1007         }
1008
1009         if (S_ISDIR(inode->i_mode))
1010                 drop_nlink(inode);
1011         drop_nlink(inode);
1012         ocfs2_set_links_count(fe, inode->i_nlink);
1013         ocfs2_journal_dirty(handle, fe_bh);
1014
1015         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1016         if (S_ISDIR(inode->i_mode))
1017                 drop_nlink(dir);
1018
1019         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1020         if (status < 0) {
1021                 mlog_errno(status);
1022                 if (S_ISDIR(inode->i_mode))
1023                         inc_nlink(dir);
1024                 goto leave;
1025         }
1026
1027         if (is_unlinkable) {
1028                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1029                                 orphan_name, &orphan_insert, orphan_dir, false);
1030                 if (status < 0)
1031                         mlog_errno(status);
1032         }
1033
1034 leave:
1035         if (handle)
1036                 ocfs2_commit_trans(osb, handle);
1037
1038         if (child_locked)
1039                 ocfs2_inode_unlock(inode, 1);
1040
1041         ocfs2_inode_unlock(dir, 1);
1042
1043         if (orphan_dir) {
1044                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1045                 ocfs2_inode_unlock(orphan_dir, 1);
1046                 mutex_unlock(&orphan_dir->i_mutex);
1047                 iput(orphan_dir);
1048         }
1049
1050         brelse(fe_bh);
1051         brelse(parent_node_bh);
1052
1053         ocfs2_free_dir_lookup_result(&orphan_insert);
1054         ocfs2_free_dir_lookup_result(&lookup);
1055
1056         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1057                 mlog_errno(status);
1058
1059         return status;
1060 }
1061
1062 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1063                 u64 src_inode_no, u64 dest_inode_no)
1064 {
1065         int ret = 0, i = 0;
1066         u64 parent_inode_no = 0;
1067         u64 child_inode_no = src_inode_no;
1068         struct inode *child_inode;
1069
1070 #define MAX_LOOKUP_TIMES 32
1071         while (1) {
1072                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1073                 if (IS_ERR(child_inode)) {
1074                         ret = PTR_ERR(child_inode);
1075                         break;
1076                 }
1077
1078                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1079                 if (ret < 0) {
1080                         iput(child_inode);
1081                         if (ret != -ENOENT)
1082                                 mlog_errno(ret);
1083                         break;
1084                 }
1085
1086                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1087                                 &parent_inode_no);
1088                 ocfs2_inode_unlock(child_inode, 0);
1089                 iput(child_inode);
1090                 if (ret < 0) {
1091                         ret = -ENOENT;
1092                         break;
1093                 }
1094
1095                 if (parent_inode_no == dest_inode_no) {
1096                         ret = 1;
1097                         break;
1098                 }
1099
1100                 if (parent_inode_no == osb->root_inode->i_ino) {
1101                         ret = 0;
1102                         break;
1103                 }
1104
1105                 child_inode_no = parent_inode_no;
1106
1107                 if (++i >= MAX_LOOKUP_TIMES) {
1108                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1109                                         "may have nested directories, "
1110                                         "src inode: %llu, dest inode: %llu.\n",
1111                                         (unsigned long long)src_inode_no,
1112                                         (unsigned long long)dest_inode_no);
1113                         ret = 0;
1114                         break;
1115                 }
1116         }
1117
1118         return ret;
1119 }
1120
1121 /*
1122  * The only place this should be used is rename and link!
1123  * if they have the same id, then the 1st one is the only one locked.
1124  */
1125 static int ocfs2_double_lock(struct ocfs2_super *osb,
1126                              struct buffer_head **bh1,
1127                              struct inode *inode1,
1128                              struct buffer_head **bh2,
1129                              struct inode *inode2,
1130                              int rename)
1131 {
1132         int status;
1133         int inode1_is_ancestor, inode2_is_ancestor;
1134         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1135         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1136
1137         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1138                                 (unsigned long long)oi2->ip_blkno);
1139
1140         if (*bh1)
1141                 *bh1 = NULL;
1142         if (*bh2)
1143                 *bh2 = NULL;
1144
1145         /* we always want to lock the one with the lower lockid first.
1146          * and if they are nested, we lock ancestor first */
1147         if (oi1->ip_blkno != oi2->ip_blkno) {
1148                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1149                                 oi1->ip_blkno);
1150                 if (inode1_is_ancestor < 0) {
1151                         status = inode1_is_ancestor;
1152                         goto bail;
1153                 }
1154
1155                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1156                                 oi2->ip_blkno);
1157                 if (inode2_is_ancestor < 0) {
1158                         status = inode2_is_ancestor;
1159                         goto bail;
1160                 }
1161
1162                 if ((inode1_is_ancestor == 1) ||
1163                                 (oi1->ip_blkno < oi2->ip_blkno &&
1164                                 inode2_is_ancestor == 0)) {
1165                         /* switch id1 and id2 around */
1166                         swap(bh2, bh1);
1167                         swap(inode2, inode1);
1168                 }
1169                 /* lock id2 */
1170                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1171                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1172                 if (status < 0) {
1173                         if (status != -ENOENT)
1174                                 mlog_errno(status);
1175                         goto bail;
1176                 }
1177         }
1178
1179         /* lock id1 */
1180         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1181                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1182         if (status < 0) {
1183                 /*
1184                  * An error return must mean that no cluster locks
1185                  * were held on function exit.
1186                  */
1187                 if (oi1->ip_blkno != oi2->ip_blkno) {
1188                         ocfs2_inode_unlock(inode2, 1);
1189                         brelse(*bh2);
1190                         *bh2 = NULL;
1191                 }
1192
1193                 if (status != -ENOENT)
1194                         mlog_errno(status);
1195         }
1196
1197         trace_ocfs2_double_lock_end(
1198                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1199                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1200
1201 bail:
1202         if (status)
1203                 mlog_errno(status);
1204         return status;
1205 }
1206
1207 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1208 {
1209         ocfs2_inode_unlock(inode1, 1);
1210
1211         if (inode1 != inode2)
1212                 ocfs2_inode_unlock(inode2, 1);
1213 }
1214
1215 static int ocfs2_rename(struct inode *old_dir,
1216                         struct dentry *old_dentry,
1217                         struct inode *new_dir,
1218                         struct dentry *new_dentry)
1219 {
1220         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1221         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1222         struct inode *old_inode = d_inode(old_dentry);
1223         struct inode *new_inode = d_inode(new_dentry);
1224         struct inode *orphan_dir = NULL;
1225         struct ocfs2_dinode *newfe = NULL;
1226         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1227         struct buffer_head *newfe_bh = NULL;
1228         struct buffer_head *old_inode_bh = NULL;
1229         struct ocfs2_super *osb = NULL;
1230         u64 newfe_blkno, old_de_ino;
1231         handle_t *handle = NULL;
1232         struct buffer_head *old_dir_bh = NULL;
1233         struct buffer_head *new_dir_bh = NULL;
1234         u32 old_dir_nlink = old_dir->i_nlink;
1235         struct ocfs2_dinode *old_di;
1236         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1237         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1238         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1239         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1240         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1241         bool should_add_orphan = false;
1242
1243         /* At some point it might be nice to break this function up a
1244          * bit. */
1245
1246         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1247                            old_dentry->d_name.len, old_dentry->d_name.name,
1248                            new_dentry->d_name.len, new_dentry->d_name.name);
1249
1250         status = dquot_initialize(old_dir);
1251         if (status) {
1252                 mlog_errno(status);
1253                 goto bail;
1254         }
1255         status = dquot_initialize(new_dir);
1256         if (status) {
1257                 mlog_errno(status);
1258                 goto bail;
1259         }
1260
1261         osb = OCFS2_SB(old_dir->i_sb);
1262
1263         if (new_inode) {
1264                 if (!igrab(new_inode))
1265                         BUG();
1266         }
1267
1268         /* Assume a directory hierarchy thusly:
1269          * a/b/c
1270          * a/d
1271          * a,b,c, and d are all directories.
1272          *
1273          * from cwd of 'a' on both nodes:
1274          * node1: mv b/c d
1275          * node2: mv d   b/c
1276          *
1277          * And that's why, just like the VFS, we need a file system
1278          * rename lock. */
1279         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1280                 status = ocfs2_rename_lock(osb);
1281                 if (status < 0) {
1282                         mlog_errno(status);
1283                         goto bail;
1284                 }
1285                 rename_lock = 1;
1286
1287                 /* here we cannot guarantee the inodes haven't just been
1288                  * changed, so check if they are nested again */
1289                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1290                                 old_inode->i_ino);
1291                 if (status < 0) {
1292                         mlog_errno(status);
1293                         goto bail;
1294                 } else if (status == 1) {
1295                         status = -EPERM;
1296                         trace_ocfs2_rename_not_permitted(
1297                                         (unsigned long long)old_inode->i_ino,
1298                                         (unsigned long long)new_dir->i_ino);
1299                         goto bail;
1300                 }
1301         }
1302
1303         /* if old and new are the same, this'll just do one lock. */
1304         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1305                                    &new_dir_bh, new_dir, 1);
1306         if (status < 0) {
1307                 mlog_errno(status);
1308                 goto bail;
1309         }
1310         parents_locked = 1;
1311
1312         /* make sure both dirs have bhs
1313          * get an extra ref on old_dir_bh if old==new */
1314         if (!new_dir_bh) {
1315                 if (old_dir_bh) {
1316                         new_dir_bh = old_dir_bh;
1317                         get_bh(new_dir_bh);
1318                 } else {
1319                         mlog(ML_ERROR, "no old_dir_bh!\n");
1320                         status = -EIO;
1321                         goto bail;
1322                 }
1323         }
1324
1325         /*
1326          * Aside from allowing a meta data update, the locking here
1327          * also ensures that the downconvert thread on other nodes
1328          * won't have to concurrently downconvert the inode and the
1329          * dentry locks.
1330          */
1331         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1332                                          OI_LS_PARENT);
1333         if (status < 0) {
1334                 if (status != -ENOENT)
1335                         mlog_errno(status);
1336                 goto bail;
1337         }
1338         old_child_locked = 1;
1339
1340         status = ocfs2_remote_dentry_delete(old_dentry);
1341         if (status < 0) {
1342                 mlog_errno(status);
1343                 goto bail;
1344         }
1345
1346         if (S_ISDIR(old_inode->i_mode)) {
1347                 u64 old_inode_parent;
1348
1349                 update_dot_dot = 1;
1350                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1351                                                   old_inode,
1352                                                   &old_inode_dot_dot_res);
1353                 if (status) {
1354                         status = -EIO;
1355                         goto bail;
1356                 }
1357
1358                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1359                         status = -EIO;
1360                         goto bail;
1361                 }
1362
1363                 if (!new_inode && new_dir != old_dir &&
1364                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1365                         status = -EMLINK;
1366                         goto bail;
1367                 }
1368         }
1369
1370         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1371                                             old_dentry->d_name.len,
1372                                             &old_de_ino);
1373         if (status) {
1374                 status = -ENOENT;
1375                 goto bail;
1376         }
1377
1378         /*
1379          *  Check for inode number is _not_ due to possible IO errors.
1380          *  We might rmdir the source, keep it as pwd of some process
1381          *  and merrily kill the link to whatever was created under the
1382          *  same name. Goodbye sticky bit ;-<
1383          */
1384         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1385                 status = -ENOENT;
1386                 goto bail;
1387         }
1388
1389         /* check if the target already exists (in which case we need
1390          * to delete it */
1391         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1392                                           new_dentry->d_name.len,
1393                                           &newfe_blkno, new_dir,
1394                                           &target_lookup_res);
1395         /* The only error we allow here is -ENOENT because the new
1396          * file not existing is perfectly valid. */
1397         if ((status < 0) && (status != -ENOENT)) {
1398                 /* If we cannot find the file specified we should just */
1399                 /* return the error... */
1400                 mlog_errno(status);
1401                 goto bail;
1402         }
1403         if (status == 0)
1404                 target_exists = 1;
1405
1406         if (!target_exists && new_inode) {
1407                 /*
1408                  * Target was unlinked by another node while we were
1409                  * waiting to get to ocfs2_rename(). There isn't
1410                  * anything we can do here to help the situation, so
1411                  * bubble up the appropriate error.
1412                  */
1413                 status = -ENOENT;
1414                 goto bail;
1415         }
1416
1417         /* In case we need to overwrite an existing file, we blow it
1418          * away first */
1419         if (target_exists) {
1420                 /* VFS didn't think there existed an inode here, but
1421                  * someone else in the cluster must have raced our
1422                  * rename to create one. Today we error cleanly, in
1423                  * the future we should consider calling iget to build
1424                  * a new struct inode for this entry. */
1425                 if (!new_inode) {
1426                         status = -EACCES;
1427
1428                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1429                                                 new_dentry->d_name.name);
1430                         goto bail;
1431                 }
1432
1433                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1434                         status = -EACCES;
1435
1436                         trace_ocfs2_rename_disagree(
1437                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1438                              (unsigned long long)newfe_blkno,
1439                              OCFS2_I(new_inode)->ip_flags);
1440                         goto bail;
1441                 }
1442
1443                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1444                 if (status < 0) {
1445                         if (status != -ENOENT)
1446                                 mlog_errno(status);
1447                         goto bail;
1448                 }
1449                 new_child_locked = 1;
1450
1451                 status = ocfs2_remote_dentry_delete(new_dentry);
1452                 if (status < 0) {
1453                         mlog_errno(status);
1454                         goto bail;
1455                 }
1456
1457                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1458
1459                 trace_ocfs2_rename_over_existing(
1460                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1461                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1462
1463                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1464                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1465                                                 OCFS2_I(new_inode)->ip_blkno,
1466                                                 orphan_name, &orphan_insert,
1467                                                 false);
1468                         if (status < 0) {
1469                                 mlog_errno(status);
1470                                 goto bail;
1471                         }
1472                         should_add_orphan = true;
1473                 }
1474         } else {
1475                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1476
1477                 status = ocfs2_check_dir_for_entry(new_dir,
1478                                                    new_dentry->d_name.name,
1479                                                    new_dentry->d_name.len);
1480                 if (status)
1481                         goto bail;
1482
1483                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1484                                                       new_dentry->d_name.name,
1485                                                       new_dentry->d_name.len,
1486                                                       &target_insert);
1487                 if (status < 0) {
1488                         mlog_errno(status);
1489                         goto bail;
1490                 }
1491         }
1492
1493         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1494         if (IS_ERR(handle)) {
1495                 status = PTR_ERR(handle);
1496                 handle = NULL;
1497                 mlog_errno(status);
1498                 goto bail;
1499         }
1500
1501         if (target_exists) {
1502                 if (S_ISDIR(new_inode->i_mode)) {
1503                         if (new_inode->i_nlink != 2 ||
1504                             !ocfs2_empty_dir(new_inode)) {
1505                                 status = -ENOTEMPTY;
1506                                 goto bail;
1507                         }
1508                 }
1509                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1510                                                  newfe_bh,
1511                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1512                 if (status < 0) {
1513                         mlog_errno(status);
1514                         goto bail;
1515                 }
1516
1517                 /* change the dirent to point to the correct inode */
1518                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1519                                             old_inode);
1520                 if (status < 0) {
1521                         mlog_errno(status);
1522                         goto bail;
1523                 }
1524                 new_dir->i_version++;
1525
1526                 if (S_ISDIR(new_inode->i_mode))
1527                         ocfs2_set_links_count(newfe, 0);
1528                 else
1529                         ocfs2_add_links_count(newfe, -1);
1530                 ocfs2_journal_dirty(handle, newfe_bh);
1531                 if (should_add_orphan) {
1532                         status = ocfs2_orphan_add(osb, handle, new_inode,
1533                                         newfe_bh, orphan_name,
1534                                         &orphan_insert, orphan_dir, false);
1535                         if (status < 0) {
1536                                 mlog_errno(status);
1537                                 goto bail;
1538                         }
1539                 }
1540         } else {
1541                 /* if the name was not found in new_dir, add it now */
1542                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1543                                          OCFS2_I(old_inode)->ip_blkno,
1544                                          new_dir_bh, &target_insert);
1545         }
1546
1547         old_inode->i_ctime = CURRENT_TIME;
1548         mark_inode_dirty(old_inode);
1549
1550         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1551                                          old_inode_bh,
1552                                          OCFS2_JOURNAL_ACCESS_WRITE);
1553         if (status >= 0) {
1554                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1555
1556                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1557                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1558                 ocfs2_journal_dirty(handle, old_inode_bh);
1559         } else
1560                 mlog_errno(status);
1561
1562         /*
1563          * Now that the name has been added to new_dir, remove the old name.
1564          *
1565          * We don't keep any directory entry context around until now
1566          * because the insert might have changed the type of directory
1567          * we're dealing with.
1568          */
1569         status = ocfs2_find_entry(old_dentry->d_name.name,
1570                                   old_dentry->d_name.len, old_dir,
1571                                   &old_entry_lookup);
1572         if (status)
1573                 goto bail;
1574
1575         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1576         if (status < 0) {
1577                 mlog_errno(status);
1578                 goto bail;
1579         }
1580
1581         if (new_inode) {
1582                 drop_nlink(new_inode);
1583                 new_inode->i_ctime = CURRENT_TIME;
1584         }
1585         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1586
1587         if (update_dot_dot) {
1588                 status = ocfs2_update_entry(old_inode, handle,
1589                                             &old_inode_dot_dot_res, new_dir);
1590                 drop_nlink(old_dir);
1591                 if (new_inode) {
1592                         drop_nlink(new_inode);
1593                 } else {
1594                         inc_nlink(new_dir);
1595                         mark_inode_dirty(new_dir);
1596                 }
1597         }
1598         mark_inode_dirty(old_dir);
1599         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1600         if (new_inode) {
1601                 mark_inode_dirty(new_inode);
1602                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1603         }
1604
1605         if (old_dir != new_dir) {
1606                 /* Keep the same times on both directories.*/
1607                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1608
1609                 /*
1610                  * This will also pick up the i_nlink change from the
1611                  * block above.
1612                  */
1613                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1614         }
1615
1616         if (old_dir_nlink != old_dir->i_nlink) {
1617                 if (!old_dir_bh) {
1618                         mlog(ML_ERROR, "need to change nlink for old dir "
1619                              "%llu from %d to %d but bh is NULL!\n",
1620                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1621                              (int)old_dir_nlink, old_dir->i_nlink);
1622                 } else {
1623                         struct ocfs2_dinode *fe;
1624                         status = ocfs2_journal_access_di(handle,
1625                                                          INODE_CACHE(old_dir),
1626                                                          old_dir_bh,
1627                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1628                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1629                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1630                         ocfs2_journal_dirty(handle, old_dir_bh);
1631                 }
1632         }
1633         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1634         status = 0;
1635 bail:
1636         if (rename_lock)
1637                 ocfs2_rename_unlock(osb);
1638
1639         if (handle)
1640                 ocfs2_commit_trans(osb, handle);
1641
1642         if (parents_locked)
1643                 ocfs2_double_unlock(old_dir, new_dir);
1644
1645         if (old_child_locked)
1646                 ocfs2_inode_unlock(old_inode, 1);
1647
1648         if (new_child_locked)
1649                 ocfs2_inode_unlock(new_inode, 1);
1650
1651         if (orphan_dir) {
1652                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1653                 ocfs2_inode_unlock(orphan_dir, 1);
1654                 mutex_unlock(&orphan_dir->i_mutex);
1655                 iput(orphan_dir);
1656         }
1657
1658         if (new_inode)
1659                 sync_mapping_buffers(old_inode->i_mapping);
1660
1661         if (new_inode)
1662                 iput(new_inode);
1663
1664         ocfs2_free_dir_lookup_result(&target_lookup_res);
1665         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1666         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1667         ocfs2_free_dir_lookup_result(&orphan_insert);
1668         ocfs2_free_dir_lookup_result(&target_insert);
1669
1670         brelse(newfe_bh);
1671         brelse(old_inode_bh);
1672         brelse(old_dir_bh);
1673         brelse(new_dir_bh);
1674
1675         if (status)
1676                 mlog_errno(status);
1677
1678         return status;
1679 }
1680
1681 /*
1682  * we expect i_size = strlen(symname). Copy symname into the file
1683  * data, including the null terminator.
1684  */
1685 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1686                                      handle_t *handle,
1687                                      struct inode *inode,
1688                                      const char *symname)
1689 {
1690         struct buffer_head **bhs = NULL;
1691         const char *c;
1692         struct super_block *sb = osb->sb;
1693         u64 p_blkno, p_blocks;
1694         int virtual, blocks, status, i, bytes_left;
1695
1696         bytes_left = i_size_read(inode) + 1;
1697         /* we can't trust i_blocks because we're actually going to
1698          * write i_size + 1 bytes. */
1699         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1700
1701         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1702                                         i_size_read(inode), blocks);
1703
1704         /* Sanity check -- make sure we're going to fit. */
1705         if (bytes_left >
1706             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1707                 status = -EIO;
1708                 mlog_errno(status);
1709                 goto bail;
1710         }
1711
1712         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1713         if (!bhs) {
1714                 status = -ENOMEM;
1715                 mlog_errno(status);
1716                 goto bail;
1717         }
1718
1719         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1720                                              NULL);
1721         if (status < 0) {
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         /* links can never be larger than one cluster so we know this
1727          * is all going to be contiguous, but do a sanity check
1728          * anyway. */
1729         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1730                 status = -EIO;
1731                 mlog_errno(status);
1732                 goto bail;
1733         }
1734
1735         virtual = 0;
1736         while(bytes_left > 0) {
1737                 c = &symname[virtual * sb->s_blocksize];
1738
1739                 bhs[virtual] = sb_getblk(sb, p_blkno);
1740                 if (!bhs[virtual]) {
1741                         status = -ENOMEM;
1742                         mlog_errno(status);
1743                         goto bail;
1744                 }
1745                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1746                                               bhs[virtual]);
1747
1748                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1749                                               bhs[virtual],
1750                                               OCFS2_JOURNAL_ACCESS_CREATE);
1751                 if (status < 0) {
1752                         mlog_errno(status);
1753                         goto bail;
1754                 }
1755
1756                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1757
1758                 memcpy(bhs[virtual]->b_data, c,
1759                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1760                        bytes_left);
1761
1762                 ocfs2_journal_dirty(handle, bhs[virtual]);
1763
1764                 virtual++;
1765                 p_blkno++;
1766                 bytes_left -= sb->s_blocksize;
1767         }
1768
1769         status = 0;
1770 bail:
1771
1772         if (bhs) {
1773                 for(i = 0; i < blocks; i++)
1774                         brelse(bhs[i]);
1775                 kfree(bhs);
1776         }
1777
1778         if (status)
1779                 mlog_errno(status);
1780         return status;
1781 }
1782
1783 static int ocfs2_symlink(struct inode *dir,
1784                          struct dentry *dentry,
1785                          const char *symname)
1786 {
1787         int status, l, credits;
1788         u64 newsize;
1789         struct ocfs2_super *osb = NULL;
1790         struct inode *inode = NULL;
1791         struct super_block *sb;
1792         struct buffer_head *new_fe_bh = NULL;
1793         struct buffer_head *parent_fe_bh = NULL;
1794         struct ocfs2_dinode *fe = NULL;
1795         struct ocfs2_dinode *dirfe;
1796         handle_t *handle = NULL;
1797         struct ocfs2_alloc_context *inode_ac = NULL;
1798         struct ocfs2_alloc_context *data_ac = NULL;
1799         struct ocfs2_alloc_context *xattr_ac = NULL;
1800         int want_clusters = 0;
1801         int xattr_credits = 0;
1802         struct ocfs2_security_xattr_info si = {
1803                 .enable = 1,
1804         };
1805         int did_quota = 0, did_quota_inode = 0;
1806         struct ocfs2_dir_lookup_result lookup = { NULL, };
1807         sigset_t oldset;
1808         int did_block_signals = 0;
1809         struct ocfs2_dentry_lock *dl = NULL;
1810
1811         trace_ocfs2_symlink_begin(dir, dentry, symname,
1812                                   dentry->d_name.len, dentry->d_name.name);
1813
1814         status = dquot_initialize(dir);
1815         if (status) {
1816                 mlog_errno(status);
1817                 goto bail;
1818         }
1819
1820         sb = dir->i_sb;
1821         osb = OCFS2_SB(sb);
1822
1823         l = strlen(symname) + 1;
1824
1825         credits = ocfs2_calc_symlink_credits(sb);
1826
1827         /* lock the parent directory */
1828         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1829         if (status < 0) {
1830                 if (status != -ENOENT)
1831                         mlog_errno(status);
1832                 return status;
1833         }
1834
1835         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1836         if (!ocfs2_read_links_count(dirfe)) {
1837                 /* can't make a file in a deleted directory. */
1838                 status = -ENOENT;
1839                 goto bail;
1840         }
1841
1842         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1843                                            dentry->d_name.len);
1844         if (status)
1845                 goto bail;
1846
1847         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1848                                               dentry->d_name.name,
1849                                               dentry->d_name.len, &lookup);
1850         if (status < 0) {
1851                 mlog_errno(status);
1852                 goto bail;
1853         }
1854
1855         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1856         if (status < 0) {
1857                 if (status != -ENOSPC)
1858                         mlog_errno(status);
1859                 goto bail;
1860         }
1861
1862         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1863         if (IS_ERR(inode)) {
1864                 status = PTR_ERR(inode);
1865                 inode = NULL;
1866                 mlog_errno(status);
1867                 goto bail;
1868         }
1869
1870         /* get security xattr */
1871         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1872         if (status) {
1873                 if (status == -EOPNOTSUPP)
1874                         si.enable = 0;
1875                 else {
1876                         mlog_errno(status);
1877                         goto bail;
1878                 }
1879         }
1880
1881         /* calculate meta data/clusters for setting security xattr */
1882         if (si.enable) {
1883                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1884                                                   &xattr_credits, &xattr_ac);
1885                 if (status < 0) {
1886                         mlog_errno(status);
1887                         goto bail;
1888                 }
1889         }
1890
1891         /* don't reserve bitmap space for fast symlinks. */
1892         if (l > ocfs2_fast_symlink_chars(sb))
1893                 want_clusters += 1;
1894
1895         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1896         if (status < 0) {
1897                 if (status != -ENOSPC)
1898                         mlog_errno(status);
1899                 goto bail;
1900         }
1901
1902         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1903         if (IS_ERR(handle)) {
1904                 status = PTR_ERR(handle);
1905                 handle = NULL;
1906                 mlog_errno(status);
1907                 goto bail;
1908         }
1909
1910         /* Starting to change things, restart is no longer possible. */
1911         ocfs2_block_signals(&oldset);
1912         did_block_signals = 1;
1913
1914         status = dquot_alloc_inode(inode);
1915         if (status)
1916                 goto bail;
1917         did_quota_inode = 1;
1918
1919         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1920                                    dentry->d_name.name,
1921                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1922                                    inode->i_mode);
1923
1924         status = ocfs2_mknod_locked(osb, dir, inode,
1925                                     0, &new_fe_bh, parent_fe_bh, handle,
1926                                     inode_ac);
1927         if (status < 0) {
1928                 mlog_errno(status);
1929                 goto bail;
1930         }
1931
1932         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1933         inode->i_rdev = 0;
1934         newsize = l - 1;
1935         inode->i_op = &ocfs2_symlink_inode_operations;
1936         if (l > ocfs2_fast_symlink_chars(sb)) {
1937                 u32 offset = 0;
1938
1939                 status = dquot_alloc_space_nodirty(inode,
1940                     ocfs2_clusters_to_bytes(osb->sb, 1));
1941                 if (status)
1942                         goto bail;
1943                 did_quota = 1;
1944                 inode->i_mapping->a_ops = &ocfs2_aops;
1945                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1946                                               new_fe_bh,
1947                                               handle, data_ac, NULL,
1948                                               NULL);
1949                 if (status < 0) {
1950                         if (status != -ENOSPC && status != -EINTR) {
1951                                 mlog(ML_ERROR,
1952                                      "Failed to extend file to %llu\n",
1953                                      (unsigned long long)newsize);
1954                                 mlog_errno(status);
1955                                 status = -ENOSPC;
1956                         }
1957                         goto bail;
1958                 }
1959                 i_size_write(inode, newsize);
1960                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1961         } else {
1962                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1963                 memcpy((char *) fe->id2.i_symlink, symname, l);
1964                 i_size_write(inode, newsize);
1965                 inode->i_blocks = 0;
1966         }
1967
1968         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1969         if (status < 0) {
1970                 mlog_errno(status);
1971                 goto bail;
1972         }
1973
1974         if (!ocfs2_inode_is_fast_symlink(inode)) {
1975                 status = ocfs2_create_symlink_data(osb, handle, inode,
1976                                                    symname);
1977                 if (status < 0) {
1978                         mlog_errno(status);
1979                         goto bail;
1980                 }
1981         }
1982
1983         if (si.enable) {
1984                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1985                                                  xattr_ac, data_ac);
1986                 if (status < 0) {
1987                         mlog_errno(status);
1988                         goto bail;
1989                 }
1990         }
1991
1992         /*
1993          * Do this before adding the entry to the directory. We add
1994          * also set d_op after success so that ->d_iput() will cleanup
1995          * the dentry lock even if ocfs2_add_entry() fails below.
1996          */
1997         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1998         if (status) {
1999                 mlog_errno(status);
2000                 goto bail;
2001         }
2002
2003         dl = dentry->d_fsdata;
2004
2005         status = ocfs2_add_entry(handle, dentry, inode,
2006                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2007                                  &lookup);
2008         if (status < 0) {
2009                 mlog_errno(status);
2010                 goto bail;
2011         }
2012
2013         insert_inode_hash(inode);
2014         d_instantiate(dentry, inode);
2015 bail:
2016         if (status < 0 && did_quota)
2017                 dquot_free_space_nodirty(inode,
2018                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2019         if (status < 0 && did_quota_inode)
2020                 dquot_free_inode(inode);
2021         if (handle)
2022                 ocfs2_commit_trans(osb, handle);
2023
2024         ocfs2_inode_unlock(dir, 1);
2025         if (did_block_signals)
2026                 ocfs2_unblock_signals(&oldset);
2027
2028         brelse(new_fe_bh);
2029         brelse(parent_fe_bh);
2030         kfree(si.value);
2031         ocfs2_free_dir_lookup_result(&lookup);
2032         if (inode_ac)
2033                 ocfs2_free_alloc_context(inode_ac);
2034         if (data_ac)
2035                 ocfs2_free_alloc_context(data_ac);
2036         if (xattr_ac)
2037                 ocfs2_free_alloc_context(xattr_ac);
2038         if ((status < 0) && inode) {
2039                 if (dl)
2040                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2041
2042                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2043                 clear_nlink(inode);
2044                 iput(inode);
2045         }
2046
2047         if (status)
2048                 mlog_errno(status);
2049
2050         return status;
2051 }
2052
2053 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2054 {
2055         int status, namelen;
2056
2057         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2058                            (long long)blkno);
2059         if (namelen <= 0) {
2060                 if (namelen)
2061                         status = namelen;
2062                 else
2063                         status = -EINVAL;
2064                 mlog_errno(status);
2065                 goto bail;
2066         }
2067         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2068                 status = -EINVAL;
2069                 mlog_errno(status);
2070                 goto bail;
2071         }
2072
2073         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2074
2075         status = 0;
2076 bail:
2077         if (status < 0)
2078                 mlog_errno(status);
2079         return status;
2080 }
2081
2082 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2083                                         struct inode **ret_orphan_dir,
2084                                         struct buffer_head **ret_orphan_dir_bh)
2085 {
2086         struct inode *orphan_dir_inode;
2087         struct buffer_head *orphan_dir_bh = NULL;
2088         int ret = 0;
2089
2090         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2091                                                        ORPHAN_DIR_SYSTEM_INODE,
2092                                                        osb->slot_num);
2093         if (!orphan_dir_inode) {
2094                 ret = -ENOENT;
2095                 mlog_errno(ret);
2096                 return ret;
2097         }
2098
2099         mutex_lock(&orphan_dir_inode->i_mutex);
2100
2101         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2102         if (ret < 0) {
2103                 mutex_unlock(&orphan_dir_inode->i_mutex);
2104                 iput(orphan_dir_inode);
2105
2106                 mlog_errno(ret);
2107                 return ret;
2108         }
2109
2110         *ret_orphan_dir = orphan_dir_inode;
2111         *ret_orphan_dir_bh = orphan_dir_bh;
2112
2113         return 0;
2114 }
2115
2116 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2117                                       struct buffer_head *orphan_dir_bh,
2118                                       u64 blkno,
2119                                       char *name,
2120                                       struct ocfs2_dir_lookup_result *lookup,
2121                                       bool dio)
2122 {
2123         int ret;
2124         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2125         int namelen = dio ?
2126                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2127                         OCFS2_ORPHAN_NAMELEN;
2128
2129         if (dio) {
2130                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2131                                 OCFS2_DIO_ORPHAN_PREFIX);
2132                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2133                         ret = -EINVAL;
2134                         mlog_errno(ret);
2135                         return ret;
2136                 }
2137
2138                 ret = ocfs2_blkno_stringify(blkno,
2139                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2140         } else
2141                 ret = ocfs2_blkno_stringify(blkno, name);
2142         if (ret < 0) {
2143                 mlog_errno(ret);
2144                 return ret;
2145         }
2146
2147         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2148                                            orphan_dir_bh, name,
2149                                            namelen, lookup);
2150         if (ret < 0) {
2151                 mlog_errno(ret);
2152                 return ret;
2153         }
2154
2155         return 0;
2156 }
2157
2158 /**
2159  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2160  * insertion of an orphan.
2161  * @osb: ocfs2 file system
2162  * @ret_orphan_dir: Orphan dir inode - returned locked!
2163  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2164  * @lookup: dir lookup result, to be passed back into functions like
2165  *          ocfs2_orphan_add
2166  *
2167  * Returns zero on success and the ret_orphan_dir, name and lookup
2168  * fields will be populated.
2169  *
2170  * Returns non-zero on failure.
2171  */
2172 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2173                                     struct inode **ret_orphan_dir,
2174                                     u64 blkno,
2175                                     char *name,
2176                                     struct ocfs2_dir_lookup_result *lookup,
2177                                     bool dio)
2178 {
2179         struct inode *orphan_dir_inode = NULL;
2180         struct buffer_head *orphan_dir_bh = NULL;
2181         int ret = 0;
2182
2183         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2184                                            &orphan_dir_bh);
2185         if (ret < 0) {
2186                 mlog_errno(ret);
2187                 return ret;
2188         }
2189
2190         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2191                                          blkno, name, lookup, dio);
2192         if (ret < 0) {
2193                 mlog_errno(ret);
2194                 goto out;
2195         }
2196
2197         *ret_orphan_dir = orphan_dir_inode;
2198
2199 out:
2200         brelse(orphan_dir_bh);
2201
2202         if (ret) {
2203                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2204                 mutex_unlock(&orphan_dir_inode->i_mutex);
2205                 iput(orphan_dir_inode);
2206         }
2207
2208         if (ret)
2209                 mlog_errno(ret);
2210         return ret;
2211 }
2212
2213 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2214                             handle_t *handle,
2215                             struct inode *inode,
2216                             struct buffer_head *fe_bh,
2217                             char *name,
2218                             struct ocfs2_dir_lookup_result *lookup,
2219                             struct inode *orphan_dir_inode,
2220                             bool dio)
2221 {
2222         struct buffer_head *orphan_dir_bh = NULL;
2223         int status = 0;
2224         struct ocfs2_dinode *orphan_fe;
2225         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2226         int namelen = dio ?
2227                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2228                         OCFS2_ORPHAN_NAMELEN;
2229
2230         trace_ocfs2_orphan_add_begin(
2231                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2232
2233         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2234         if (status < 0) {
2235                 mlog_errno(status);
2236                 goto leave;
2237         }
2238
2239         status = ocfs2_journal_access_di(handle,
2240                                          INODE_CACHE(orphan_dir_inode),
2241                                          orphan_dir_bh,
2242                                          OCFS2_JOURNAL_ACCESS_WRITE);
2243         if (status < 0) {
2244                 mlog_errno(status);
2245                 goto leave;
2246         }
2247
2248         /*
2249          * We're going to journal the change of i_flags and i_orphaned_slot.
2250          * It's safe anyway, though some callers may duplicate the journaling.
2251          * Journaling within the func just make the logic look more
2252          * straightforward.
2253          */
2254         status = ocfs2_journal_access_di(handle,
2255                                          INODE_CACHE(inode),
2256                                          fe_bh,
2257                                          OCFS2_JOURNAL_ACCESS_WRITE);
2258         if (status < 0) {
2259                 mlog_errno(status);
2260                 goto leave;
2261         }
2262
2263         /* we're a cluster, and nlink can change on disk from
2264          * underneath us... */
2265         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2266         if (S_ISDIR(inode->i_mode))
2267                 ocfs2_add_links_count(orphan_fe, 1);
2268         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2269         ocfs2_journal_dirty(handle, orphan_dir_bh);
2270
2271         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2272                                    namelen, inode,
2273                                    OCFS2_I(inode)->ip_blkno,
2274                                    orphan_dir_bh, lookup);
2275         if (status < 0) {
2276                 mlog_errno(status);
2277                 goto rollback;
2278         }
2279
2280         if (dio) {
2281                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2282                  * slot.
2283                  */
2284                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2285                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2286         } else {
2287                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2288                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2289
2290                 /* Record which orphan dir our inode now resides
2291                  * in. delete_inode will use this to determine which orphan
2292                  * dir to lock. */
2293                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2294         }
2295
2296         ocfs2_journal_dirty(handle, fe_bh);
2297
2298         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2299                                    osb->slot_num);
2300
2301 rollback:
2302         if (status < 0) {
2303                 if (S_ISDIR(inode->i_mode))
2304                         ocfs2_add_links_count(orphan_fe, -1);
2305                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2306         }
2307
2308 leave:
2309         brelse(orphan_dir_bh);
2310
2311         return status;
2312 }
2313
2314 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2315 int ocfs2_orphan_del(struct ocfs2_super *osb,
2316                      handle_t *handle,
2317                      struct inode *orphan_dir_inode,
2318                      struct inode *inode,
2319                      struct buffer_head *orphan_dir_bh,
2320                      bool dio)
2321 {
2322         const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
2323         char name[namelen + 1];
2324         struct ocfs2_dinode *orphan_fe;
2325         int status = 0;
2326         struct ocfs2_dir_lookup_result lookup = { NULL, };
2327
2328         if (dio) {
2329                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2330                                 OCFS2_DIO_ORPHAN_PREFIX);
2331                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2332                         status = -EINVAL;
2333                         mlog_errno(status);
2334                         return status;
2335                 }
2336
2337                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2338                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2339         } else
2340                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2341         if (status < 0) {
2342                 mlog_errno(status);
2343                 goto leave;
2344         }
2345
2346         trace_ocfs2_orphan_del(
2347              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2348              name, strlen(name));
2349
2350         /* find it's spot in the orphan directory */
2351         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2352                                   &lookup);
2353         if (status) {
2354                 mlog_errno(status);
2355                 goto leave;
2356         }
2357
2358         /* remove it from the orphan directory */
2359         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2360         if (status < 0) {
2361                 mlog_errno(status);
2362                 goto leave;
2363         }
2364
2365         status = ocfs2_journal_access_di(handle,
2366                                          INODE_CACHE(orphan_dir_inode),
2367                                          orphan_dir_bh,
2368                                          OCFS2_JOURNAL_ACCESS_WRITE);
2369         if (status < 0) {
2370                 mlog_errno(status);
2371                 goto leave;
2372         }
2373
2374         /* do the i_nlink dance! :) */
2375         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2376         if (S_ISDIR(inode->i_mode))
2377                 ocfs2_add_links_count(orphan_fe, -1);
2378         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2379         ocfs2_journal_dirty(handle, orphan_dir_bh);
2380
2381 leave:
2382         ocfs2_free_dir_lookup_result(&lookup);
2383
2384         if (status)
2385                 mlog_errno(status);
2386         return status;
2387 }
2388
2389 /**
2390  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2391  * allocated file. This is different from the typical 'add to orphan dir'
2392  * operation in that the inode does not yet exist. This is a problem because
2393  * the orphan dir stringifies the inode block number to come up with it's
2394  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2395  * problem. This function works around it by calling deeper into the orphan
2396  * and suballoc code than other callers. Use this only by necessity.
2397  * @dir: The directory which this inode will ultimately wind up under - not the
2398  * orphan dir!
2399  * @dir_bh: buffer_head the @dir inode block
2400  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2401  * with the string to be used for orphan dirent. Pass back to the orphan dir
2402  * code.
2403  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2404  * dir code.
2405  * @ret_di_blkno: block number where the new inode will be allocated.
2406  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2407  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2408  *
2409  * Returns zero on success and the ret_orphan_dir, name and lookup
2410  * fields will be populated.
2411  *
2412  * Returns non-zero on failure.
2413  */
2414 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2415                                         struct buffer_head *dir_bh,
2416                                         char *orphan_name,
2417                                         struct inode **ret_orphan_dir,
2418                                         u64 *ret_di_blkno,
2419                                         struct ocfs2_dir_lookup_result *orphan_insert,
2420                                         struct ocfs2_alloc_context **ret_inode_ac)
2421 {
2422         int ret;
2423         u64 di_blkno;
2424         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2425         struct inode *orphan_dir = NULL;
2426         struct buffer_head *orphan_dir_bh = NULL;
2427         struct ocfs2_alloc_context *inode_ac = NULL;
2428
2429         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2430         if (ret < 0) {
2431                 mlog_errno(ret);
2432                 return ret;
2433         }
2434
2435         /* reserve an inode spot */
2436         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2437         if (ret < 0) {
2438                 if (ret != -ENOSPC)
2439                         mlog_errno(ret);
2440                 goto out;
2441         }
2442
2443         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2444                                        &di_blkno);
2445         if (ret) {
2446                 mlog_errno(ret);
2447                 goto out;
2448         }
2449
2450         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2451                                          di_blkno, orphan_name, orphan_insert,
2452                                          false);
2453         if (ret < 0) {
2454                 mlog_errno(ret);
2455                 goto out;
2456         }
2457
2458 out:
2459         if (ret == 0) {
2460                 *ret_orphan_dir = orphan_dir;
2461                 *ret_di_blkno = di_blkno;
2462                 *ret_inode_ac = inode_ac;
2463                 /*
2464                  * orphan_name and orphan_insert are already up to
2465                  * date via prepare_orphan_dir
2466                  */
2467         } else {
2468                 /* Unroll reserve_new_inode* */
2469                 if (inode_ac)
2470                         ocfs2_free_alloc_context(inode_ac);
2471
2472                 /* Unroll orphan dir locking */
2473                 mutex_unlock(&orphan_dir->i_mutex);
2474                 ocfs2_inode_unlock(orphan_dir, 1);
2475                 iput(orphan_dir);
2476         }
2477
2478         brelse(orphan_dir_bh);
2479
2480         return ret;
2481 }
2482
2483 int ocfs2_create_inode_in_orphan(struct inode *dir,
2484                                  int mode,
2485                                  struct inode **new_inode)
2486 {
2487         int status, did_quota_inode = 0;
2488         struct inode *inode = NULL;
2489         struct inode *orphan_dir = NULL;
2490         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2491         struct ocfs2_dinode *di = NULL;
2492         handle_t *handle = NULL;
2493         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2494         struct buffer_head *parent_di_bh = NULL;
2495         struct buffer_head *new_di_bh = NULL;
2496         struct ocfs2_alloc_context *inode_ac = NULL;
2497         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2498         u64 uninitialized_var(di_blkno), suballoc_loc;
2499         u16 suballoc_bit;
2500
2501         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2502         if (status < 0) {
2503                 if (status != -ENOENT)
2504                         mlog_errno(status);
2505                 return status;
2506         }
2507
2508         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2509                                               orphan_name, &orphan_dir,
2510                                               &di_blkno, &orphan_insert, &inode_ac);
2511         if (status < 0) {
2512                 if (status != -ENOSPC)
2513                         mlog_errno(status);
2514                 goto leave;
2515         }
2516
2517         inode = ocfs2_get_init_inode(dir, mode);
2518         if (IS_ERR(inode)) {
2519                 status = PTR_ERR(inode);
2520                 inode = NULL;
2521                 mlog_errno(status);
2522                 goto leave;
2523         }
2524
2525         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2526         if (IS_ERR(handle)) {
2527                 status = PTR_ERR(handle);
2528                 handle = NULL;
2529                 mlog_errno(status);
2530                 goto leave;
2531         }
2532
2533         status = dquot_alloc_inode(inode);
2534         if (status)
2535                 goto leave;
2536         did_quota_inode = 1;
2537
2538         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2539                                               &suballoc_loc,
2540                                               &suballoc_bit, di_blkno);
2541         if (status < 0) {
2542                 mlog_errno(status);
2543                 goto leave;
2544         }
2545
2546         clear_nlink(inode);
2547         /* do the real work now. */
2548         status = __ocfs2_mknod_locked(dir, inode,
2549                                       0, &new_di_bh, parent_di_bh, handle,
2550                                       inode_ac, di_blkno, suballoc_loc,
2551                                       suballoc_bit);
2552         if (status < 0) {
2553                 mlog_errno(status);
2554                 goto leave;
2555         }
2556
2557         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2558         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2559                                   &orphan_insert, orphan_dir, false);
2560         if (status < 0) {
2561                 mlog_errno(status);
2562                 goto leave;
2563         }
2564
2565         /* get open lock so that only nodes can't remove it from orphan dir. */
2566         status = ocfs2_open_lock(inode);
2567         if (status < 0)
2568                 mlog_errno(status);
2569
2570         insert_inode_hash(inode);
2571 leave:
2572         if (status < 0 && did_quota_inode)
2573                 dquot_free_inode(inode);
2574         if (handle)
2575                 ocfs2_commit_trans(osb, handle);
2576
2577         if (orphan_dir) {
2578                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2579                 ocfs2_inode_unlock(orphan_dir, 1);
2580                 mutex_unlock(&orphan_dir->i_mutex);
2581                 iput(orphan_dir);
2582         }
2583
2584         if ((status < 0) && inode) {
2585                 clear_nlink(inode);
2586                 iput(inode);
2587         }
2588
2589         if (inode_ac)
2590                 ocfs2_free_alloc_context(inode_ac);
2591
2592         brelse(new_di_bh);
2593
2594         if (!status)
2595                 *new_inode = inode;
2596
2597         ocfs2_free_dir_lookup_result(&orphan_insert);
2598
2599         ocfs2_inode_unlock(dir, 1);
2600         brelse(parent_di_bh);
2601         return status;
2602 }
2603
2604 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2605         struct inode *inode)
2606 {
2607         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2608         struct inode *orphan_dir_inode = NULL;
2609         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2610         struct buffer_head *di_bh = NULL;
2611         int status = 0;
2612         handle_t *handle = NULL;
2613         struct ocfs2_dinode *di = NULL;
2614
2615         status = ocfs2_inode_lock(inode, &di_bh, 1);
2616         if (status < 0) {
2617                 mlog_errno(status);
2618                 goto bail;
2619         }
2620
2621         di = (struct ocfs2_dinode *) di_bh->b_data;
2622         /*
2623          * Another append dio crashed?
2624          * If so, manually recover it first.
2625          */
2626         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2627                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2628                 if (status < 0) {
2629                         if (status != -ENOSPC)
2630                                 mlog_errno(status);
2631                         goto bail_unlock_inode;
2632                 }
2633
2634                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2635                 if (status < 0) {
2636                         mlog_errno(status);
2637                         goto bail_unlock_inode;
2638                 }
2639         }
2640
2641         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2642                         OCFS2_I(inode)->ip_blkno,
2643                         orphan_name,
2644                         &orphan_insert,
2645                         true);
2646         if (status < 0) {
2647                 mlog_errno(status);
2648                 goto bail_unlock_inode;
2649         }
2650
2651         handle = ocfs2_start_trans(osb,
2652                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2653         if (IS_ERR(handle)) {
2654                 status = PTR_ERR(handle);
2655                 goto bail_unlock_orphan;
2656         }
2657
2658         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2659                         &orphan_insert, orphan_dir_inode, true);
2660         if (status)
2661                 mlog_errno(status);
2662
2663         ocfs2_commit_trans(osb, handle);
2664
2665 bail_unlock_orphan:
2666         ocfs2_inode_unlock(orphan_dir_inode, 1);
2667         mutex_unlock(&orphan_dir_inode->i_mutex);
2668         iput(orphan_dir_inode);
2669
2670         ocfs2_free_dir_lookup_result(&orphan_insert);
2671
2672 bail_unlock_inode:
2673         ocfs2_inode_unlock(inode, 1);
2674         brelse(di_bh);
2675
2676 bail:
2677         return status;
2678 }
2679
2680 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2681                 struct inode *inode, struct buffer_head *di_bh,
2682                 int update_isize, loff_t end)
2683 {
2684         struct inode *orphan_dir_inode = NULL;
2685         struct buffer_head *orphan_dir_bh = NULL;
2686         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2687         handle_t *handle = NULL;
2688         int status = 0;
2689
2690         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2691                         ORPHAN_DIR_SYSTEM_INODE,
2692                         le16_to_cpu(di->i_dio_orphaned_slot));
2693         if (!orphan_dir_inode) {
2694                 status = -ENOENT;
2695                 mlog_errno(status);
2696                 goto bail;
2697         }
2698
2699         mutex_lock(&orphan_dir_inode->i_mutex);
2700         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2701         if (status < 0) {
2702                 mutex_unlock(&orphan_dir_inode->i_mutex);
2703                 iput(orphan_dir_inode);
2704                 mlog_errno(status);
2705                 goto bail;
2706         }
2707
2708         handle = ocfs2_start_trans(osb,
2709                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2710         if (IS_ERR(handle)) {
2711                 status = PTR_ERR(handle);
2712                 goto bail_unlock_orphan;
2713         }
2714
2715         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2716
2717         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2718                                 inode, orphan_dir_bh, true);
2719         if (status < 0) {
2720                 mlog_errno(status);
2721                 goto bail_commit;
2722         }
2723
2724         status = ocfs2_journal_access_di(handle,
2725                         INODE_CACHE(inode),
2726                         di_bh,
2727                         OCFS2_JOURNAL_ACCESS_WRITE);
2728         if (status < 0) {
2729                 mlog_errno(status);
2730                 goto bail_commit;
2731         }
2732
2733         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2734         di->i_dio_orphaned_slot = 0;
2735
2736         if (update_isize) {
2737                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2738                 if (status)
2739                         mlog_errno(status);
2740         } else
2741                 ocfs2_journal_dirty(handle, di_bh);
2742
2743 bail_commit:
2744         ocfs2_commit_trans(osb, handle);
2745
2746 bail_unlock_orphan:
2747         ocfs2_inode_unlock(orphan_dir_inode, 1);
2748         mutex_unlock(&orphan_dir_inode->i_mutex);
2749         brelse(orphan_dir_bh);
2750         iput(orphan_dir_inode);
2751
2752 bail:
2753         return status;
2754 }
2755
2756 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2757                                    struct inode *inode,
2758                                    struct dentry *dentry)
2759 {
2760         int status = 0;
2761         struct buffer_head *parent_di_bh = NULL;
2762         handle_t *handle = NULL;
2763         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2764         struct ocfs2_dinode *dir_di, *di;
2765         struct inode *orphan_dir_inode = NULL;
2766         struct buffer_head *orphan_dir_bh = NULL;
2767         struct buffer_head *di_bh = NULL;
2768         struct ocfs2_dir_lookup_result lookup = { NULL, };
2769
2770         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2771                                 dentry->d_name.len, dentry->d_name.name,
2772                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2773                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2774
2775         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2776         if (status < 0) {
2777                 if (status != -ENOENT)
2778                         mlog_errno(status);
2779                 return status;
2780         }
2781
2782         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2783         if (!dir_di->i_links_count) {
2784                 /* can't make a file in a deleted directory. */
2785                 status = -ENOENT;
2786                 goto leave;
2787         }
2788
2789         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2790                                            dentry->d_name.len);
2791         if (status)
2792                 goto leave;
2793
2794         /* get a spot inside the dir. */
2795         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2796                                               dentry->d_name.name,
2797                                               dentry->d_name.len, &lookup);
2798         if (status < 0) {
2799                 mlog_errno(status);
2800                 goto leave;
2801         }
2802
2803         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2804                                                        ORPHAN_DIR_SYSTEM_INODE,
2805                                                        osb->slot_num);
2806         if (!orphan_dir_inode) {
2807                 status = -ENOENT;
2808                 mlog_errno(status);
2809                 goto leave;
2810         }
2811
2812         mutex_lock(&orphan_dir_inode->i_mutex);
2813
2814         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2815         if (status < 0) {
2816                 mlog_errno(status);
2817                 mutex_unlock(&orphan_dir_inode->i_mutex);
2818                 iput(orphan_dir_inode);
2819                 goto leave;
2820         }
2821
2822         status = ocfs2_read_inode_block(inode, &di_bh);
2823         if (status < 0) {
2824                 mlog_errno(status);
2825                 goto orphan_unlock;
2826         }
2827
2828         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2829         if (IS_ERR(handle)) {
2830                 status = PTR_ERR(handle);
2831                 handle = NULL;
2832                 mlog_errno(status);
2833                 goto orphan_unlock;
2834         }
2835
2836         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2837                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2838         if (status < 0) {
2839                 mlog_errno(status);
2840                 goto out_commit;
2841         }
2842
2843         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2844                                   orphan_dir_bh, false);
2845         if (status < 0) {
2846                 mlog_errno(status);
2847                 goto out_commit;
2848         }
2849
2850         di = (struct ocfs2_dinode *)di_bh->b_data;
2851         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2852         di->i_orphaned_slot = 0;
2853         set_nlink(inode, 1);
2854         ocfs2_set_links_count(di, inode->i_nlink);
2855         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2856         ocfs2_journal_dirty(handle, di_bh);
2857
2858         status = ocfs2_add_entry(handle, dentry, inode,
2859                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2860                                  &lookup);
2861         if (status < 0) {
2862                 mlog_errno(status);
2863                 goto out_commit;
2864         }
2865
2866         status = ocfs2_dentry_attach_lock(dentry, inode,
2867                                           OCFS2_I(dir)->ip_blkno);
2868         if (status) {
2869                 mlog_errno(status);
2870                 goto out_commit;
2871         }
2872
2873         d_instantiate(dentry, inode);
2874         status = 0;
2875 out_commit:
2876         ocfs2_commit_trans(osb, handle);
2877 orphan_unlock:
2878         ocfs2_inode_unlock(orphan_dir_inode, 1);
2879         mutex_unlock(&orphan_dir_inode->i_mutex);
2880         iput(orphan_dir_inode);
2881 leave:
2882
2883         ocfs2_inode_unlock(dir, 1);
2884
2885         brelse(di_bh);
2886         brelse(parent_di_bh);
2887         brelse(orphan_dir_bh);
2888
2889         ocfs2_free_dir_lookup_result(&lookup);
2890
2891         if (status)
2892                 mlog_errno(status);
2893
2894         return status;
2895 }
2896
2897 const struct inode_operations ocfs2_dir_iops = {
2898         .create         = ocfs2_create,
2899         .lookup         = ocfs2_lookup,
2900         .link           = ocfs2_link,
2901         .unlink         = ocfs2_unlink,
2902         .rmdir          = ocfs2_unlink,
2903         .symlink        = ocfs2_symlink,
2904         .mkdir          = ocfs2_mkdir,
2905         .mknod          = ocfs2_mknod,
2906         .rename         = ocfs2_rename,
2907         .setattr        = ocfs2_setattr,
2908         .getattr        = ocfs2_getattr,
2909         .permission     = ocfs2_permission,
2910         .setxattr       = generic_setxattr,
2911         .getxattr       = generic_getxattr,
2912         .listxattr      = ocfs2_listxattr,
2913         .removexattr    = generic_removexattr,
2914         .fiemap         = ocfs2_fiemap,
2915         .get_acl        = ocfs2_iop_get_acl,
2916         .set_acl        = ocfs2_iop_set_acl,
2917 };