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
 110 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
 111                                    unsigned int flags)
 112 {
 113         int status;
 114         u64 blkno;
 115         struct inode *inode = NULL;
 116         struct dentry *ret;
 117         struct ocfs2_inode_info *oi;
 118
 119         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 120                            dentry->d_name.name,
 121                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 122
 123         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 124                 ret = ERR_PTR(-ENAMETOOLONG);
 125                 goto bail;
 126         }
 127
 128         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 129         if (status < 0) {
 130                 if (status != -ENOENT)
 131                         mlog_errno(status);
 132                 ret = ERR_PTR(status);
 133                 goto bail;
 134         }
 135
 136         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 137                                             dentry->d_name.len, &blkno);
 138         if (status < 0)
 139                 goto bail_add;
 140
 141         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 142         if (IS_ERR(inode)) {
 143                 ret = ERR_PTR(-EACCES);
 144                 goto bail_unlock;
 145         }
 146
 147         oi = OCFS2_I(inode);
 148         /* Clear any orphaned state... If we were able to look up the
 149          * inode from a directory, it certainly can't be orphaned. We
 150          * might have the bad state from a node which intended to
 151          * orphan this inode but crashed before it could commit the
 152          * unlink. */
 153         spin_lock(&oi->ip_lock);
 154         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 155         spin_unlock(&oi->ip_lock);
 156
 157 bail_add:
 158         ret = d_splice_alias(inode, dentry);
 159
 160         if (inode) {
 161                 /*
 162                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 163                  * dentry, it will d_move() it on top of ourse. The
 164                  * return value will indicate this however, so in
 165                  * those cases, we switch them around for the locking
 166                  * code.
 167                  *
 168                  * NOTE: This dentry already has ->d_op set from
 169                  * ocfs2_get_parent() and ocfs2_get_dentry()
 170                  */
 171                 if (!IS_ERR_OR_NULL(ret))
 172                         dentry = ret;
 173
 174                 status = ocfs2_dentry_attach_lock(dentry, inode,
 175                                                   OCFS2_I(dir)->ip_blkno);
 176                 if (status) {
 177                         mlog_errno(status);
 178                         ret = ERR_PTR(status);
 179                         goto bail_unlock;
 180                 }
 181         } else
 182                 ocfs2_dentry_attach_gen(dentry);
 183
 184 bail_unlock:
 185         /* Don't drop the cluster lock until *after* the d_add --
 186          * unlink on another node will message us to remove that
 187          * dentry under this lock so otherwise we can race this with
 188          * the downconvert thread and have a stale dentry. */
 189         ocfs2_inode_unlock(dir, 0);
 190
 191 bail:
 192
 193         trace_ocfs2_lookup_ret(ret);
 194
 195         return ret;
 196 }
 197
 198 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 199 {
 200         struct inode *inode;
 201         int status;
 202
 203         inode = new_inode(dir->i_sb);
 204         if (!inode) {
 205                 mlog(ML_ERROR, "new_inode failed!\n");
 206                 return ERR_PTR(-ENOMEM);
 207         }
 208
 209         /* populate as many fields early on as possible - many of
 210          * these are used by the support functions here and in
 211          * callers. */
 212         if (S_ISDIR(mode))
 213                 set_nlink(inode, 2);
 214         inode_init_owner(inode, dir, mode);
 215         status = dquot_initialize(inode);
 216         if (status)
 217                 return ERR_PTR(status);
 218
 219         return inode;
 220 }
 221
 222 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 223                 struct dentry *dentry, struct inode *inode)
 224 {
 225         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 226
 227         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 228         ocfs2_lock_res_free(&dl->dl_lockres);
 229         BUG_ON(dl->dl_count != 1);
 230         spin_lock(&dentry_attach_lock);
 231         dentry->d_fsdata = NULL;
 232         spin_unlock(&dentry_attach_lock);
 233         kfree(dl);
 234         iput(inode);
 235 }
 236
 237 static int ocfs2_mknod(struct inode *dir,
 238                        struct dentry *dentry,
 239                        umode_t mode,
 240                        dev_t dev)
 241 {
 242         int status = 0;
 243         struct buffer_head *parent_fe_bh = NULL;
 244         handle_t *handle = NULL;
 245         struct ocfs2_super *osb;
 246         struct ocfs2_dinode *dirfe;
 247         struct buffer_head *new_fe_bh = NULL;
 248         struct inode *inode = NULL;
 249         struct ocfs2_alloc_context *inode_ac = NULL;
 250         struct ocfs2_alloc_context *data_ac = NULL;
 251         struct ocfs2_alloc_context *meta_ac = NULL;
 252         int want_clusters = 0;
 253         int want_meta = 0;
 254         int xattr_credits = 0;
 255         struct ocfs2_security_xattr_info si = {
 256                 .enable = 1,
 257         };
 258         int did_quota_inode = 0;
 259         struct ocfs2_dir_lookup_result lookup = { NULL, };
 260         sigset_t oldset;
 261         int did_block_signals = 0;
 262         struct ocfs2_dentry_lock *dl = NULL;
 263
 264         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 265                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 266                           (unsigned long)dev, mode);
 267
 268         status = dquot_initialize(dir);
 269         if (status) {
 270                 mlog_errno(status);
 271                 return status;
 272         }
 273
 274         /* get our super block */
 275         osb = OCFS2_SB(dir->i_sb);
 276
 277         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 278         if (status < 0) {
 279                 if (status != -ENOENT)
 280                         mlog_errno(status);
 281                 return status;
 282         }
 283
 284         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 285                 status = -EMLINK;
 286                 goto leave;
 287         }
 288
 289         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 290         if (!ocfs2_read_links_count(dirfe)) {
 291                 /* can't make a file in a deleted directory. */
 292                 status = -ENOENT;
 293                 goto leave;
 294         }
 295
 296         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 297                                            dentry->d_name.len);
 298         if (status)
 299                 goto leave;
 300
 301         /* get a spot inside the dir. */
 302         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 303                                               dentry->d_name.name,
 304                                               dentry->d_name.len, &lookup);
 305         if (status < 0) {
 306                 mlog_errno(status);
 307                 goto leave;
 308         }
 309
 310         /* reserve an inode spot */
 311         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 312         if (status < 0) {
 313                 if (status != -ENOSPC)
 314                         mlog_errno(status);
 315                 goto leave;
 316         }
 317
 318         inode = ocfs2_get_init_inode(dir, mode);
 319         if (IS_ERR(inode)) {
 320                 status = PTR_ERR(inode);
 321                 inode = NULL;
 322                 mlog_errno(status);
 323                 goto leave;
 324         }
 325
 326         /* get security xattr */
 327         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 328         if (status) {
 329                 if (status == -EOPNOTSUPP)
 330                         si.enable = 0;
 331                 else {
 332                         mlog_errno(status);
 333                         goto leave;
 334                 }
 335         }
 336
 337         /* calculate meta data/clusters for setting security and acl xattr */
 338         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 339                                        &si, &want_clusters,
 340                                        &xattr_credits, &want_meta);
 341         if (status < 0) {
 342                 mlog_errno(status);
 343                 goto leave;
 344         }
 345
 346         /* Reserve a cluster if creating an extent based directory. */
 347         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 348                 want_clusters += 1;
 349
 350                 /* Dir indexing requires extra space as well */
 351                 if (ocfs2_supports_indexed_dirs(osb))
 352                         want_meta++;
 353         }
 354
 355         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 356         if (status < 0) {
 357                 if (status != -ENOSPC)
 358                         mlog_errno(status);
 359                 goto leave;
 360         }
 361
 362         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 363         if (status < 0) {
 364                 if (status != -ENOSPC)
 365                         mlog_errno(status);
 366                 goto leave;
 367         }
 368
 369         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 370                                                             S_ISDIR(mode),
 371                                                             xattr_credits));
 372         if (IS_ERR(handle)) {
 373                 status = PTR_ERR(handle);
 374                 handle = NULL;
 375                 mlog_errno(status);
 376                 goto leave;
 377         }
 378
 379         /* Starting to change things, restart is no longer possible. */
 380         ocfs2_block_signals(&oldset);
 381         did_block_signals = 1;
 382
 383         status = dquot_alloc_inode(inode);
 384         if (status)
 385                 goto leave;
 386         did_quota_inode = 1;
 387
 388         /* do the real work now. */
 389         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 390                                     &new_fe_bh, parent_fe_bh, handle,
 391                                     inode_ac);
 392         if (status < 0) {
 393                 mlog_errno(status);
 394                 goto leave;
 395         }
 396
 397         if (S_ISDIR(mode)) {
 398                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 399                                             new_fe_bh, data_ac, meta_ac);
 400                 if (status < 0) {
 401                         mlog_errno(status);
 402                         goto leave;
 403                 }
 404
 405                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 406                                                  parent_fe_bh,
 407                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 408                 if (status < 0) {
 409                         mlog_errno(status);
 410                         goto leave;
 411                 }
 412                 ocfs2_add_links_count(dirfe, 1);
 413                 ocfs2_journal_dirty(handle, parent_fe_bh);
 414                 inc_nlink(dir);
 415         }
 416
 417         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 418                          meta_ac, data_ac);
 419
 420         if (status < 0) {
 421                 mlog_errno(status);
 422                 goto leave;
 423         }
 424
 425         if (si.enable) {
 426                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 427                                                  meta_ac, data_ac);
 428                 if (status < 0) {
 429                         mlog_errno(status);
 430                         goto leave;
 431                 }
 432         }
 433
 434         /*
 435          * Do this before adding the entry to the directory. We add
 436          * also set d_op after success so that ->d_iput() will cleanup
 437          * the dentry lock even if ocfs2_add_entry() fails below.
 438          */
 439         status = ocfs2_dentry_attach_lock(dentry, inode,
 440                                           OCFS2_I(dir)->ip_blkno);
 441         if (status) {
 442                 mlog_errno(status);
 443                 goto leave;
 444         }
 445
 446         dl = dentry->d_fsdata;
 447
 448         status = ocfs2_add_entry(handle, dentry, inode,
 449                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 450                                  &lookup);
 451         if (status < 0) {
 452                 mlog_errno(status);
 453                 goto leave;
 454         }
 455
 456         insert_inode_hash(inode);
 457         d_instantiate(dentry, inode);
 458         status = 0;
 459 leave:
 460         if (status < 0 && did_quota_inode)
 461                 dquot_free_inode(inode);
 462         if (handle)
 463                 ocfs2_commit_trans(osb, handle);
 464
 465         ocfs2_inode_unlock(dir, 1);
 466         if (did_block_signals)
 467                 ocfs2_unblock_signals(&oldset);
 468
 469         brelse(new_fe_bh);
 470         brelse(parent_fe_bh);
 471         kfree(si.value);
 472
 473         ocfs2_free_dir_lookup_result(&lookup);
 474
 475         if (inode_ac)
 476                 ocfs2_free_alloc_context(inode_ac);
 477
 478         if (data_ac)
 479                 ocfs2_free_alloc_context(data_ac);
 480
 481         if (meta_ac)
 482                 ocfs2_free_alloc_context(meta_ac);
 483
 484         /*
 485          * We should call iput after the i_mutex of the bitmap been
 486          * unlocked in ocfs2_free_alloc_context, or the
 487          * ocfs2_delete_inode will mutex_lock again.
 488          */
 489         if ((status < 0) && inode) {
 490                 if (dl)
 491                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 492
 493                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 494                 clear_nlink(inode);
 495                 iput(inode);
 496         }
 497
 498         if (status)
 499                 mlog_errno(status);
 500
 501         return status;
 502 }
 503
 504 static int __ocfs2_mknod_locked(struct inode *dir,
 505                                 struct inode *inode,
 506                                 dev_t dev,
 507                                 struct buffer_head **new_fe_bh,
 508                                 struct buffer_head *parent_fe_bh,
 509                                 handle_t *handle,
 510                                 struct ocfs2_alloc_context *inode_ac,
 511                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 512 {
 513         int status = 0;
 514         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 515         struct ocfs2_dinode *fe = NULL;
 516         struct ocfs2_extent_list *fel;
 517         u16 feat;
 518         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 519
 520         *new_fe_bh = NULL;
 521
 522         /* populate as many fields early on as possible - many of
 523          * these are used by the support functions here and in
 524          * callers. */
 525         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 526         OCFS2_I(inode)->ip_blkno = fe_blkno;
 527         spin_lock(&osb->osb_lock);
 528         inode->i_generation = osb->s_next_generation++;
 529         spin_unlock(&osb->osb_lock);
 530
 531         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 532         if (!*new_fe_bh) {
 533                 status = -ENOMEM;
 534                 mlog_errno(status);
 535                 goto leave;
 536         }
 537         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 538
 539         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 540                                          *new_fe_bh,
 541                                          OCFS2_JOURNAL_ACCESS_CREATE);
 542         if (status < 0) {
 543                 mlog_errno(status);
 544                 goto leave;
 545         }
 546
 547         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 548         memset(fe, 0, osb->sb->s_blocksize);
 549
 550         fe->i_generation = cpu_to_le32(inode->i_generation);
 551         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 552         fe->i_blkno = cpu_to_le64(fe_blkno);
 553         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 554         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 555         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 556         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 557         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 558         fe->i_mode = cpu_to_le16(inode->i_mode);
 559         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 560                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 561
 562         ocfs2_set_links_count(fe, inode->i_nlink);
 563
 564         fe->i_last_eb_blk = 0;
 565         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 566         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 567         fe->i_atime = fe->i_ctime = fe->i_mtime =
 568                 cpu_to_le64(CURRENT_TIME.tv_sec);
 569         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 570                 cpu_to_le32(CURRENT_TIME.tv_nsec);
 571         fe->i_dtime = 0;
 572
 573         /*
 574          * If supported, directories start with inline data. If inline
 575          * isn't supported, but indexing is, we start them as indexed.
 576          */
 577         feat = le16_to_cpu(fe->i_dyn_features);
 578         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 579                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 580
 581                 fe->id2.i_data.id_count = cpu_to_le16(
 582                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 583         } else {
 584                 fel = &fe->id2.i_list;
 585                 fel->l_tree_depth = 0;
 586                 fel->l_next_free_rec = 0;
 587                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 588         }
 589
 590         ocfs2_journal_dirty(handle, *new_fe_bh);
 591
 592         ocfs2_populate_inode(inode, fe, 1);
 593         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 594         if (!ocfs2_mount_local(osb)) {
 595                 status = ocfs2_create_new_inode_locks(inode);
 596                 if (status < 0)
 597                         mlog_errno(status);
 598         }
 599
 600         oi->i_sync_tid = handle->h_transaction->t_tid;
 601         oi->i_datasync_tid = handle->h_transaction->t_tid;
 602
 603 leave:
 604         if (status < 0) {
 605                 if (*new_fe_bh) {
 606                         brelse(*new_fe_bh);
 607                         *new_fe_bh = NULL;
 608                 }
 609         }
 610
 611         if (status)
 612                 mlog_errno(status);
 613         return status;
 614 }
 615
 616 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 617                               struct inode *dir,
 618                               struct inode *inode,
 619                               dev_t dev,
 620                               struct buffer_head **new_fe_bh,
 621                               struct buffer_head *parent_fe_bh,
 622                               handle_t *handle,
 623                               struct ocfs2_alloc_context *inode_ac)
 624 {
 625         int status = 0;
 626         u64 suballoc_loc, fe_blkno = 0;
 627         u16 suballoc_bit;
 628
 629         *new_fe_bh = NULL;
 630
 631         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 632                                        inode_ac, &suballoc_loc,
 633                                        &suballoc_bit, &fe_blkno);
 634         if (status < 0) {
 635                 mlog_errno(status);
 636                 return status;
 637         }
 638
 639         status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 640                                     parent_fe_bh, handle, inode_ac,
 641                                     fe_blkno, suballoc_loc, suballoc_bit);
 642         if (status < 0) {
 643                 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
 644                 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
 645                                 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
 646                 if (tmp)
 647                         mlog_errno(tmp);
 648         }
 649
 650         return status;
 651 }
 652
 653 static int ocfs2_mkdir(struct inode *dir,
 654                        struct dentry *dentry,
 655                        umode_t mode)
 656 {
 657         int ret;
 658
 659         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 660                           OCFS2_I(dir)->ip_blkno, mode);
 661         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
 662         if (ret)
 663                 mlog_errno(ret);
 664
 665         return ret;
 666 }
 667
 668 static int ocfs2_create(struct inode *dir,
 669                         struct dentry *dentry,
 670                         umode_t mode,
 671                         bool excl)
 672 {
 673         int ret;
 674
 675         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 676                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 677         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
 678         if (ret)
 679                 mlog_errno(ret);
 680
 681         return ret;
 682 }
 683
 684 static int ocfs2_link(struct dentry *old_dentry,
 685                       struct inode *dir,
 686                       struct dentry *dentry)
 687 {
 688         handle_t *handle;
 689         struct inode *inode = d_inode(old_dentry);
 690         struct inode *old_dir = d_inode(old_dentry->d_parent);
 691         int err;
 692         struct buffer_head *fe_bh = NULL;
 693         struct buffer_head *old_dir_bh = NULL;
 694         struct buffer_head *parent_fe_bh = NULL;
 695         struct ocfs2_dinode *fe = NULL;
 696         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 697         struct ocfs2_dir_lookup_result lookup = { NULL, };
 698         sigset_t oldset;
 699         u64 old_de_ino;
 700
 701         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 702                          old_dentry->d_name.len, old_dentry->d_name.name,
 703                          dentry->d_name.len, dentry->d_name.name);
 704
 705         if (S_ISDIR(inode->i_mode))
 706                 return -EPERM;
 707
 708         err = dquot_initialize(dir);
 709         if (err) {
 710                 mlog_errno(err);
 711                 return err;
 712         }
 713
 714         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 715                         &parent_fe_bh, dir, 0);
 716         if (err < 0) {
 717                 if (err != -ENOENT)
 718                         mlog_errno(err);
 719                 return err;
 720         }
 721
 722         /* make sure both dirs have bhs
 723          * get an extra ref on old_dir_bh if old==new */
 724         if (!parent_fe_bh) {
 725                 if (old_dir_bh) {
 726                         parent_fe_bh = old_dir_bh;
 727                         get_bh(parent_fe_bh);
 728                 } else {
 729                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 730                         err = -EIO;
 731                         goto out;
 732                 }
 733         }
 734
 735         if (!dir->i_nlink) {
 736                 err = -ENOENT;
 737                 goto out;
 738         }
 739
 740         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 741                         old_dentry->d_name.len, &old_de_ino);
 742         if (err) {
 743                 err = -ENOENT;
 744                 goto out;
 745         }
 746
 747         /*
 748          * Check whether another node removed the source inode while we
 749          * were in the vfs.
 750          */
 751         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 752                 err = -ENOENT;
 753                 goto out;
 754         }
 755
 756         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 757                                         dentry->d_name.len);
 758         if (err)
 759                 goto out;
 760
 761         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 762                                            dentry->d_name.name,
 763                                            dentry->d_name.len, &lookup);
 764         if (err < 0) {
 765                 mlog_errno(err);
 766                 goto out;
 767         }
 768
 769         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 770         if (err < 0) {
 771                 if (err != -ENOENT)
 772                         mlog_errno(err);
 773                 goto out;
 774         }
 775
 776         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 777         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 778                 err = -EMLINK;
 779                 goto out_unlock_inode;
 780         }
 781
 782         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 783         if (IS_ERR(handle)) {
 784                 err = PTR_ERR(handle);
 785                 handle = NULL;
 786                 mlog_errno(err);
 787                 goto out_unlock_inode;
 788         }
 789
 790         /* Starting to change things, restart is no longer possible. */
 791         ocfs2_block_signals(&oldset);
 792
 793         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 794                                       OCFS2_JOURNAL_ACCESS_WRITE);
 795         if (err < 0) {
 796                 mlog_errno(err);
 797                 goto out_commit;
 798         }
 799
 800         inc_nlink(inode);
 801         inode->i_ctime = current_time(inode);
 802         ocfs2_set_links_count(fe, inode->i_nlink);
 803         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 804         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 805         ocfs2_journal_dirty(handle, fe_bh);
 806
 807         err = ocfs2_add_entry(handle, dentry, inode,
 808                               OCFS2_I(inode)->ip_blkno,
 809                               parent_fe_bh, &lookup);
 810         if (err) {
 811                 ocfs2_add_links_count(fe, -1);
 812                 drop_nlink(inode);
 813                 mlog_errno(err);
 814                 goto out_commit;
 815         }
 816
 817         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 818         if (err) {
 819                 mlog_errno(err);
 820                 goto out_commit;
 821         }
 822
 823         ihold(inode);
 824         d_instantiate(dentry, inode);
 825
 826 out_commit:
 827         ocfs2_commit_trans(osb, handle);
 828         ocfs2_unblock_signals(&oldset);
 829 out_unlock_inode:
 830         ocfs2_inode_unlock(inode, 1);
 831
 832 out:
 833         ocfs2_double_unlock(old_dir, dir);
 834
 835         brelse(fe_bh);
 836         brelse(parent_fe_bh);
 837         brelse(old_dir_bh);
 838
 839         ocfs2_free_dir_lookup_result(&lookup);
 840
 841         if (err)
 842                 mlog_errno(err);
 843
 844         return err;
 845 }
 846
 847 /*
 848  * Takes and drops an exclusive lock on the given dentry. This will
 849  * force other nodes to drop it.
 850  */
 851 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 852 {
 853         int ret;
 854
 855         ret = ocfs2_dentry_lock(dentry, 1);
 856         if (ret)
 857                 mlog_errno(ret);
 858         else
 859                 ocfs2_dentry_unlock(dentry, 1);
 860
 861         return ret;
 862 }
 863
 864 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 865 {
 866         if (S_ISDIR(inode->i_mode)) {
 867                 if (inode->i_nlink == 2)
 868                         return 1;
 869                 return 0;
 870         }
 871
 872         if (inode->i_nlink == 1)
 873                 return 1;
 874         return 0;
 875 }
 876
 877 static int ocfs2_unlink(struct inode *dir,
 878                         struct dentry *dentry)
 879 {
 880         int status;
 881         int child_locked = 0;
 882         bool is_unlinkable = false;
 883         struct inode *inode = d_inode(dentry);
 884         struct inode *orphan_dir = NULL;
 885         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 886         u64 blkno;
 887         struct ocfs2_dinode *fe = NULL;
 888         struct buffer_head *fe_bh = NULL;
 889         struct buffer_head *parent_node_bh = NULL;
 890         handle_t *handle = NULL;
 891         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 892         struct ocfs2_dir_lookup_result lookup = { NULL, };
 893         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 894
 895         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 896                            dentry->d_name.name,
 897                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 898                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 899
 900         status = dquot_initialize(dir);
 901         if (status) {
 902                 mlog_errno(status);
 903                 return status;
 904         }
 905
 906         BUG_ON(d_inode(dentry->d_parent) != dir);
 907
 908         if (inode == osb->root_inode)
 909                 return -EPERM;
 910
 911         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 912                                          OI_LS_PARENT);
 913         if (status < 0) {
 914                 if (status != -ENOENT)
 915                         mlog_errno(status);
 916                 return status;
 917         }
 918
 919         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 920                                           dentry->d_name.len, &blkno, dir,
 921                                           &lookup);
 922         if (status < 0) {
 923                 if (status != -ENOENT)
 924                         mlog_errno(status);
 925                 goto leave;
 926         }
 927
 928         if (OCFS2_I(inode)->ip_blkno != blkno) {
 929                 status = -ENOENT;
 930
 931                 trace_ocfs2_unlink_noent(
 932                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 933                                 (unsigned long long)blkno,
 934                                 OCFS2_I(inode)->ip_flags);
 935                 goto leave;
 936         }
 937
 938         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 939         if (status < 0) {
 940                 if (status != -ENOENT)
 941                         mlog_errno(status);
 942                 goto leave;
 943         }
 944         child_locked = 1;
 945
 946         if (S_ISDIR(inode->i_mode)) {
 947                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 948                         status = -ENOTEMPTY;
 949                         goto leave;
 950                 }
 951         }
 952
 953         status = ocfs2_remote_dentry_delete(dentry);
 954         if (status < 0) {
 955                 /* This remote delete should succeed under all normal
 956                  * circumstances. */
 957                 mlog_errno(status);
 958                 goto leave;
 959         }
 960
 961         if (ocfs2_inode_is_unlinkable(inode)) {
 962                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 963                                                   OCFS2_I(inode)->ip_blkno,
 964                                                   orphan_name, &orphan_insert,
 965                                                   false);
 966                 if (status < 0) {
 967                         mlog_errno(status);
 968                         goto leave;
 969                 }
 970                 is_unlinkable = true;
 971         }
 972
 973         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 974         if (IS_ERR(handle)) {
 975                 status = PTR_ERR(handle);
 976                 handle = NULL;
 977                 mlog_errno(status);
 978                 goto leave;
 979         }
 980
 981         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 982                                          OCFS2_JOURNAL_ACCESS_WRITE);
 983         if (status < 0) {
 984                 mlog_errno(status);
 985                 goto leave;
 986         }
 987
 988         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 989
 990         /* delete the name from the parent dir */
 991         status = ocfs2_delete_entry(handle, dir, &lookup);
 992         if (status < 0) {
 993                 mlog_errno(status);
 994                 goto leave;
 995         }
 996
 997         if (S_ISDIR(inode->i_mode))
 998                 drop_nlink(inode);
 999         drop_nlink(inode);
1000         ocfs2_set_links_count(fe, inode->i_nlink);
1001         ocfs2_journal_dirty(handle, fe_bh);
1002
1003         dir->i_ctime = dir->i_mtime = current_time(dir);
1004         if (S_ISDIR(inode->i_mode))
1005                 drop_nlink(dir);
1006
1007         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1008         if (status < 0) {
1009                 mlog_errno(status);
1010                 if (S_ISDIR(inode->i_mode))
1011                         inc_nlink(dir);
1012                 goto leave;
1013         }
1014
1015         if (is_unlinkable) {
1016                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1017                                 orphan_name, &orphan_insert, orphan_dir, false);
1018                 if (status < 0)
1019                         mlog_errno(status);
1020         }
1021
1022 leave:
1023         if (handle)
1024                 ocfs2_commit_trans(osb, handle);
1025
1026         if (orphan_dir) {
1027                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1028                 ocfs2_inode_unlock(orphan_dir, 1);
1029                 inode_unlock(orphan_dir);
1030                 iput(orphan_dir);
1031         }
1032
1033         if (child_locked)
1034                 ocfs2_inode_unlock(inode, 1);
1035
1036         ocfs2_inode_unlock(dir, 1);
1037
1038         brelse(fe_bh);
1039         brelse(parent_node_bh);
1040
1041         ocfs2_free_dir_lookup_result(&orphan_insert);
1042         ocfs2_free_dir_lookup_result(&lookup);
1043
1044         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1045                 mlog_errno(status);
1046
1047         return status;
1048 }
1049
1050 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1051                 u64 src_inode_no, u64 dest_inode_no)
1052 {
1053         int ret = 0, i = 0;
1054         u64 parent_inode_no = 0;
1055         u64 child_inode_no = src_inode_no;
1056         struct inode *child_inode;
1057
1058 #define MAX_LOOKUP_TIMES 32
1059         while (1) {
1060                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1061                 if (IS_ERR(child_inode)) {
1062                         ret = PTR_ERR(child_inode);
1063                         break;
1064                 }
1065
1066                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1067                 if (ret < 0) {
1068                         iput(child_inode);
1069                         if (ret != -ENOENT)
1070                                 mlog_errno(ret);
1071                         break;
1072                 }
1073
1074                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1075                                 &parent_inode_no);
1076                 ocfs2_inode_unlock(child_inode, 0);
1077                 iput(child_inode);
1078                 if (ret < 0) {
1079                         ret = -ENOENT;
1080                         break;
1081                 }
1082
1083                 if (parent_inode_no == dest_inode_no) {
1084                         ret = 1;
1085                         break;
1086                 }
1087
1088                 if (parent_inode_no == osb->root_inode->i_ino) {
1089                         ret = 0;
1090                         break;
1091                 }
1092
1093                 child_inode_no = parent_inode_no;
1094
1095                 if (++i >= MAX_LOOKUP_TIMES) {
1096                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1097                                         "may have nested directories, "
1098                                         "src inode: %llu, dest inode: %llu.\n",
1099                                         (unsigned long long)src_inode_no,
1100                                         (unsigned long long)dest_inode_no);
1101                         ret = 0;
1102                         break;
1103                 }
1104         }
1105
1106         return ret;
1107 }
1108
1109 /*
1110  * The only place this should be used is rename and link!
1111  * if they have the same id, then the 1st one is the only one locked.
1112  */
1113 static int ocfs2_double_lock(struct ocfs2_super *osb,
1114                              struct buffer_head **bh1,
1115                              struct inode *inode1,
1116                              struct buffer_head **bh2,
1117                              struct inode *inode2,
1118                              int rename)
1119 {
1120         int status;
1121         int inode1_is_ancestor, inode2_is_ancestor;
1122         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1123         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1124
1125         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1126                                 (unsigned long long)oi2->ip_blkno);
1127
1128         if (*bh1)
1129                 *bh1 = NULL;
1130         if (*bh2)
1131                 *bh2 = NULL;
1132
1133         /* we always want to lock the one with the lower lockid first.
1134          * and if they are nested, we lock ancestor first */
1135         if (oi1->ip_blkno != oi2->ip_blkno) {
1136                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1137                                 oi1->ip_blkno);
1138                 if (inode1_is_ancestor < 0) {
1139                         status = inode1_is_ancestor;
1140                         goto bail;
1141                 }
1142
1143                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1144                                 oi2->ip_blkno);
1145                 if (inode2_is_ancestor < 0) {
1146                         status = inode2_is_ancestor;
1147                         goto bail;
1148                 }
1149
1150                 if ((inode1_is_ancestor == 1) ||
1151                                 (oi1->ip_blkno < oi2->ip_blkno &&
1152                                 inode2_is_ancestor == 0)) {
1153                         /* switch id1 and id2 around */
1154                         swap(bh2, bh1);
1155                         swap(inode2, inode1);
1156                 }
1157                 /* lock id2 */
1158                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1159                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1160                 if (status < 0) {
1161                         if (status != -ENOENT)
1162                                 mlog_errno(status);
1163                         goto bail;
1164                 }
1165         }
1166
1167         /* lock id1 */
1168         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1169                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1170         if (status < 0) {
1171                 /*
1172                  * An error return must mean that no cluster locks
1173                  * were held on function exit.
1174                  */
1175                 if (oi1->ip_blkno != oi2->ip_blkno) {
1176                         ocfs2_inode_unlock(inode2, 1);
1177                         brelse(*bh2);
1178                         *bh2 = NULL;
1179                 }
1180
1181                 if (status != -ENOENT)
1182                         mlog_errno(status);
1183         }
1184
1185         trace_ocfs2_double_lock_end(
1186                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1187                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1188
1189 bail:
1190         if (status)
1191                 mlog_errno(status);
1192         return status;
1193 }
1194
1195 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1196 {
1197         ocfs2_inode_unlock(inode1, 1);
1198
1199         if (inode1 != inode2)
1200                 ocfs2_inode_unlock(inode2, 1);
1201 }
1202
1203 static int ocfs2_rename(struct inode *old_dir,
1204                         struct dentry *old_dentry,
1205                         struct inode *new_dir,
1206                         struct dentry *new_dentry,
1207                         unsigned int flags)
1208 {
1209         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1210         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1211         struct inode *old_inode = d_inode(old_dentry);
1212         struct inode *new_inode = d_inode(new_dentry);
1213         struct inode *orphan_dir = NULL;
1214         struct ocfs2_dinode *newfe = NULL;
1215         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1216         struct buffer_head *newfe_bh = NULL;
1217         struct buffer_head *old_inode_bh = NULL;
1218         struct ocfs2_super *osb = NULL;
1219         u64 newfe_blkno, old_de_ino;
1220         handle_t *handle = NULL;
1221         struct buffer_head *old_dir_bh = NULL;
1222         struct buffer_head *new_dir_bh = NULL;
1223         u32 old_dir_nlink = old_dir->i_nlink;
1224         struct ocfs2_dinode *old_di;
1225         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1226         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1227         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1228         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1229         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1230         bool should_add_orphan = false;
1231
1232         if (flags)
1233                 return -EINVAL;
1234
1235         /* At some point it might be nice to break this function up a
1236          * bit. */
1237
1238         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1239                            old_dentry->d_name.len, old_dentry->d_name.name,
1240                            new_dentry->d_name.len, new_dentry->d_name.name);
1241
1242         status = dquot_initialize(old_dir);
1243         if (status) {
1244                 mlog_errno(status);
1245                 goto bail;
1246         }
1247         status = dquot_initialize(new_dir);
1248         if (status) {
1249                 mlog_errno(status);
1250                 goto bail;
1251         }
1252
1253         osb = OCFS2_SB(old_dir->i_sb);
1254
1255         if (new_inode) {
1256                 if (!igrab(new_inode))
1257                         BUG();
1258         }
1259
1260         /* Assume a directory hierarchy thusly:
1261          * a/b/c
1262          * a/d
1263          * a,b,c, and d are all directories.
1264          *
1265          * from cwd of 'a' on both nodes:
1266          * node1: mv b/c d
1267          * node2: mv d   b/c
1268          *
1269          * And that's why, just like the VFS, we need a file system
1270          * rename lock. */
1271         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1272                 status = ocfs2_rename_lock(osb);
1273                 if (status < 0) {
1274                         mlog_errno(status);
1275                         goto bail;
1276                 }
1277                 rename_lock = 1;
1278
1279                 /* here we cannot guarantee the inodes haven't just been
1280                  * changed, so check if they are nested again */
1281                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1282                                 old_inode->i_ino);
1283                 if (status < 0) {
1284                         mlog_errno(status);
1285                         goto bail;
1286                 } else if (status == 1) {
1287                         status = -EPERM;
1288                         trace_ocfs2_rename_not_permitted(
1289                                         (unsigned long long)old_inode->i_ino,
1290                                         (unsigned long long)new_dir->i_ino);
1291                         goto bail;
1292                 }
1293         }
1294
1295         /* if old and new are the same, this'll just do one lock. */
1296         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1297                                    &new_dir_bh, new_dir, 1);
1298         if (status < 0) {
1299                 mlog_errno(status);
1300                 goto bail;
1301         }
1302         parents_locked = 1;
1303
1304         if (!new_dir->i_nlink) {
1305                 status = -EACCES;
1306                 goto bail;
1307         }
1308
1309         /* make sure both dirs have bhs
1310          * get an extra ref on old_dir_bh if old==new */
1311         if (!new_dir_bh) {
1312                 if (old_dir_bh) {
1313                         new_dir_bh = old_dir_bh;
1314                         get_bh(new_dir_bh);
1315                 } else {
1316                         mlog(ML_ERROR, "no old_dir_bh!\n");
1317                         status = -EIO;
1318                         goto bail;
1319                 }
1320         }
1321
1322         /*
1323          * Aside from allowing a meta data update, the locking here
1324          * also ensures that the downconvert thread on other nodes
1325          * won't have to concurrently downconvert the inode and the
1326          * dentry locks.
1327          */
1328         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1329                                          OI_LS_PARENT);
1330         if (status < 0) {
1331                 if (status != -ENOENT)
1332                         mlog_errno(status);
1333                 goto bail;
1334         }
1335         old_child_locked = 1;
1336
1337         status = ocfs2_remote_dentry_delete(old_dentry);
1338         if (status < 0) {
1339                 mlog_errno(status);
1340                 goto bail;
1341         }
1342
1343         if (S_ISDIR(old_inode->i_mode)) {
1344                 u64 old_inode_parent;
1345
1346                 update_dot_dot = 1;
1347                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1348                                                   old_inode,
1349                                                   &old_inode_dot_dot_res);
1350                 if (status) {
1351                         status = -EIO;
1352                         goto bail;
1353                 }
1354
1355                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1356                         status = -EIO;
1357                         goto bail;
1358                 }
1359
1360                 if (!new_inode && new_dir != old_dir &&
1361                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1362                         status = -EMLINK;
1363                         goto bail;
1364                 }
1365         }
1366
1367         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1368                                             old_dentry->d_name.len,
1369                                             &old_de_ino);
1370         if (status) {
1371                 status = -ENOENT;
1372                 goto bail;
1373         }
1374
1375         /*
1376          *  Check for inode number is _not_ due to possible IO errors.
1377          *  We might rmdir the source, keep it as pwd of some process
1378          *  and merrily kill the link to whatever was created under the
1379          *  same name. Goodbye sticky bit ;-<
1380          */
1381         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1382                 status = -ENOENT;
1383                 goto bail;
1384         }
1385
1386         /* check if the target already exists (in which case we need
1387          * to delete it */
1388         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1389                                           new_dentry->d_name.len,
1390                                           &newfe_blkno, new_dir,
1391                                           &target_lookup_res);
1392         /* The only error we allow here is -ENOENT because the new
1393          * file not existing is perfectly valid. */
1394         if ((status < 0) && (status != -ENOENT)) {
1395                 /* If we cannot find the file specified we should just */
1396                 /* return the error... */
1397                 mlog_errno(status);
1398                 goto bail;
1399         }
1400         if (status == 0)
1401                 target_exists = 1;
1402
1403         if (!target_exists && new_inode) {
1404                 /*
1405                  * Target was unlinked by another node while we were
1406                  * waiting to get to ocfs2_rename(). There isn't
1407                  * anything we can do here to help the situation, so
1408                  * bubble up the appropriate error.
1409                  */
1410                 status = -ENOENT;
1411                 goto bail;
1412         }
1413
1414         /* In case we need to overwrite an existing file, we blow it
1415          * away first */
1416         if (target_exists) {
1417                 /* VFS didn't think there existed an inode here, but
1418                  * someone else in the cluster must have raced our
1419                  * rename to create one. Today we error cleanly, in
1420                  * the future we should consider calling iget to build
1421                  * a new struct inode for this entry. */
1422                 if (!new_inode) {
1423                         status = -EACCES;
1424
1425                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1426                                                 new_dentry->d_name.name);
1427                         goto bail;
1428                 }
1429
1430                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1431                         status = -EACCES;
1432
1433                         trace_ocfs2_rename_disagree(
1434                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1435                              (unsigned long long)newfe_blkno,
1436                              OCFS2_I(new_inode)->ip_flags);
1437                         goto bail;
1438                 }
1439
1440                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1441                 if (status < 0) {
1442                         if (status != -ENOENT)
1443                                 mlog_errno(status);
1444                         goto bail;
1445                 }
1446                 new_child_locked = 1;
1447
1448                 status = ocfs2_remote_dentry_delete(new_dentry);
1449                 if (status < 0) {
1450                         mlog_errno(status);
1451                         goto bail;
1452                 }
1453
1454                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1455
1456                 trace_ocfs2_rename_over_existing(
1457                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1458                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1459
1460                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1461                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1462                                                 OCFS2_I(new_inode)->ip_blkno,
1463                                                 orphan_name, &orphan_insert,
1464                                                 false);
1465                         if (status < 0) {
1466                                 mlog_errno(status);
1467                                 goto bail;
1468                         }
1469                         should_add_orphan = true;
1470                 }
1471         } else {
1472                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1473
1474                 status = ocfs2_check_dir_for_entry(new_dir,
1475                                                    new_dentry->d_name.name,
1476                                                    new_dentry->d_name.len);
1477                 if (status)
1478                         goto bail;
1479
1480                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1481                                                       new_dentry->d_name.name,
1482                                                       new_dentry->d_name.len,
1483                                                       &target_insert);
1484                 if (status < 0) {
1485                         mlog_errno(status);
1486                         goto bail;
1487                 }
1488         }
1489
1490         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1491         if (IS_ERR(handle)) {
1492                 status = PTR_ERR(handle);
1493                 handle = NULL;
1494                 mlog_errno(status);
1495                 goto bail;
1496         }
1497
1498         if (target_exists) {
1499                 if (S_ISDIR(new_inode->i_mode)) {
1500                         if (new_inode->i_nlink != 2 ||
1501                             !ocfs2_empty_dir(new_inode)) {
1502                                 status = -ENOTEMPTY;
1503                                 goto bail;
1504                         }
1505                 }
1506                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1507                                                  newfe_bh,
1508                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1509                 if (status < 0) {
1510                         mlog_errno(status);
1511                         goto bail;
1512                 }
1513
1514                 /* change the dirent to point to the correct inode */
1515                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1516                                             old_inode);
1517                 if (status < 0) {
1518                         mlog_errno(status);
1519                         goto bail;
1520                 }
1521                 new_dir->i_version++;
1522
1523                 if (S_ISDIR(new_inode->i_mode))
1524                         ocfs2_set_links_count(newfe, 0);
1525                 else
1526                         ocfs2_add_links_count(newfe, -1);
1527                 ocfs2_journal_dirty(handle, newfe_bh);
1528                 if (should_add_orphan) {
1529                         status = ocfs2_orphan_add(osb, handle, new_inode,
1530                                         newfe_bh, orphan_name,
1531                                         &orphan_insert, orphan_dir, false);
1532                         if (status < 0) {
1533                                 mlog_errno(status);
1534                                 goto bail;
1535                         }
1536                 }
1537         } else {
1538                 /* if the name was not found in new_dir, add it now */
1539                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1540                                          OCFS2_I(old_inode)->ip_blkno,
1541                                          new_dir_bh, &target_insert);
1542         }
1543
1544         old_inode->i_ctime = current_time(old_inode);
1545         mark_inode_dirty(old_inode);
1546
1547         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1548                                          old_inode_bh,
1549                                          OCFS2_JOURNAL_ACCESS_WRITE);
1550         if (status >= 0) {
1551                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1552
1553                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1554                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1555                 ocfs2_journal_dirty(handle, old_inode_bh);
1556         } else
1557                 mlog_errno(status);
1558
1559         /*
1560          * Now that the name has been added to new_dir, remove the old name.
1561          *
1562          * We don't keep any directory entry context around until now
1563          * because the insert might have changed the type of directory
1564          * we're dealing with.
1565          */
1566         status = ocfs2_find_entry(old_dentry->d_name.name,
1567                                   old_dentry->d_name.len, old_dir,
1568                                   &old_entry_lookup);
1569         if (status) {
1570                 if (!is_journal_aborted(osb->journal->j_journal)) {
1571                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1572                                         "is not deleted.",
1573                                         new_dentry->d_name.len, new_dentry->d_name.name,
1574                                         old_dentry->d_name.len, old_dentry->d_name.name);
1575                 }
1576                 goto bail;
1577         }
1578
1579         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1580         if (status < 0) {
1581                 mlog_errno(status);
1582                 if (!is_journal_aborted(osb->journal->j_journal)) {
1583                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1584                                         "is not deleted.",
1585                                         new_dentry->d_name.len, new_dentry->d_name.name,
1586                                         old_dentry->d_name.len, old_dentry->d_name.name);
1587                 }
1588                 goto bail;
1589         }
1590
1591         if (new_inode) {
1592                 drop_nlink(new_inode);
1593                 new_inode->i_ctime = current_time(new_inode);
1594         }
1595         old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1596
1597         if (update_dot_dot) {
1598                 status = ocfs2_update_entry(old_inode, handle,
1599                                             &old_inode_dot_dot_res, new_dir);
1600                 drop_nlink(old_dir);
1601                 if (new_inode) {
1602                         drop_nlink(new_inode);
1603                 } else {
1604                         inc_nlink(new_dir);
1605                         mark_inode_dirty(new_dir);
1606                 }
1607         }
1608         mark_inode_dirty(old_dir);
1609         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1610         if (new_inode) {
1611                 mark_inode_dirty(new_inode);
1612                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1613         }
1614
1615         if (old_dir != new_dir) {
1616                 /* Keep the same times on both directories.*/
1617                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1618
1619                 /*
1620                  * This will also pick up the i_nlink change from the
1621                  * block above.
1622                  */
1623                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1624         }
1625
1626         if (old_dir_nlink != old_dir->i_nlink) {
1627                 if (!old_dir_bh) {
1628                         mlog(ML_ERROR, "need to change nlink for old dir "
1629                              "%llu from %d to %d but bh is NULL!\n",
1630                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1631                              (int)old_dir_nlink, old_dir->i_nlink);
1632                 } else {
1633                         struct ocfs2_dinode *fe;
1634                         status = ocfs2_journal_access_di(handle,
1635                                                          INODE_CACHE(old_dir),
1636                                                          old_dir_bh,
1637                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1638                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1639                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1640                         ocfs2_journal_dirty(handle, old_dir_bh);
1641                 }
1642         }
1643         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1644         status = 0;
1645 bail:
1646         if (handle)
1647                 ocfs2_commit_trans(osb, handle);
1648
1649         if (orphan_dir) {
1650                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1651                 ocfs2_inode_unlock(orphan_dir, 1);
1652                 inode_unlock(orphan_dir);
1653                 iput(orphan_dir);
1654         }
1655
1656         if (new_child_locked)
1657                 ocfs2_inode_unlock(new_inode, 1);
1658
1659         if (old_child_locked)
1660                 ocfs2_inode_unlock(old_inode, 1);
1661
1662         if (parents_locked)
1663                 ocfs2_double_unlock(old_dir, new_dir);
1664
1665         if (rename_lock)
1666                 ocfs2_rename_unlock(osb);
1667
1668         if (new_inode)
1669                 sync_mapping_buffers(old_inode->i_mapping);
1670
1671         iput(new_inode);
1672
1673         ocfs2_free_dir_lookup_result(&target_lookup_res);
1674         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1675         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1676         ocfs2_free_dir_lookup_result(&orphan_insert);
1677         ocfs2_free_dir_lookup_result(&target_insert);
1678
1679         brelse(newfe_bh);
1680         brelse(old_inode_bh);
1681         brelse(old_dir_bh);
1682         brelse(new_dir_bh);
1683
1684         if (status)
1685                 mlog_errno(status);
1686
1687         return status;
1688 }
1689
1690 /*
1691  * we expect i_size = strlen(symname). Copy symname into the file
1692  * data, including the null terminator.
1693  */
1694 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1695                                      handle_t *handle,
1696                                      struct inode *inode,
1697                                      const char *symname)
1698 {
1699         struct buffer_head **bhs = NULL;
1700         const char *c;
1701         struct super_block *sb = osb->sb;
1702         u64 p_blkno, p_blocks;
1703         int virtual, blocks, status, i, bytes_left;
1704
1705         bytes_left = i_size_read(inode) + 1;
1706         /* we can't trust i_blocks because we're actually going to
1707          * write i_size + 1 bytes. */
1708         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1709
1710         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1711                                         i_size_read(inode), blocks);
1712
1713         /* Sanity check -- make sure we're going to fit. */
1714         if (bytes_left >
1715             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1716                 status = -EIO;
1717                 mlog_errno(status);
1718                 goto bail;
1719         }
1720
1721         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1722         if (!bhs) {
1723                 status = -ENOMEM;
1724                 mlog_errno(status);
1725                 goto bail;
1726         }
1727
1728         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1729                                              NULL);
1730         if (status < 0) {
1731                 mlog_errno(status);
1732                 goto bail;
1733         }
1734
1735         /* links can never be larger than one cluster so we know this
1736          * is all going to be contiguous, but do a sanity check
1737          * anyway. */
1738         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1739                 status = -EIO;
1740                 mlog_errno(status);
1741                 goto bail;
1742         }
1743
1744         virtual = 0;
1745         while(bytes_left > 0) {
1746                 c = &symname[virtual * sb->s_blocksize];
1747
1748                 bhs[virtual] = sb_getblk(sb, p_blkno);
1749                 if (!bhs[virtual]) {
1750                         status = -ENOMEM;
1751                         mlog_errno(status);
1752                         goto bail;
1753                 }
1754                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1755                                               bhs[virtual]);
1756
1757                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1758                                               bhs[virtual],
1759                                               OCFS2_JOURNAL_ACCESS_CREATE);
1760                 if (status < 0) {
1761                         mlog_errno(status);
1762                         goto bail;
1763                 }
1764
1765                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1766
1767                 memcpy(bhs[virtual]->b_data, c,
1768                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1769                        bytes_left);
1770
1771                 ocfs2_journal_dirty(handle, bhs[virtual]);
1772
1773                 virtual++;
1774                 p_blkno++;
1775                 bytes_left -= sb->s_blocksize;
1776         }
1777
1778         status = 0;
1779 bail:
1780
1781         if (bhs) {
1782                 for(i = 0; i < blocks; i++)
1783                         brelse(bhs[i]);
1784                 kfree(bhs);
1785         }
1786
1787         if (status)
1788                 mlog_errno(status);
1789         return status;
1790 }
1791
1792 static int ocfs2_symlink(struct inode *dir,
1793                          struct dentry *dentry,
1794                          const char *symname)
1795 {
1796         int status, l, credits;
1797         u64 newsize;
1798         struct ocfs2_super *osb = NULL;
1799         struct inode *inode = NULL;
1800         struct super_block *sb;
1801         struct buffer_head *new_fe_bh = NULL;
1802         struct buffer_head *parent_fe_bh = NULL;
1803         struct ocfs2_dinode *fe = NULL;
1804         struct ocfs2_dinode *dirfe;
1805         handle_t *handle = NULL;
1806         struct ocfs2_alloc_context *inode_ac = NULL;
1807         struct ocfs2_alloc_context *data_ac = NULL;
1808         struct ocfs2_alloc_context *xattr_ac = NULL;
1809         int want_clusters = 0;
1810         int xattr_credits = 0;
1811         struct ocfs2_security_xattr_info si = {
1812                 .enable = 1,
1813         };
1814         int did_quota = 0, did_quota_inode = 0;
1815         struct ocfs2_dir_lookup_result lookup = { NULL, };
1816         sigset_t oldset;
1817         int did_block_signals = 0;
1818         struct ocfs2_dentry_lock *dl = NULL;
1819
1820         trace_ocfs2_symlink_begin(dir, dentry, symname,
1821                                   dentry->d_name.len, dentry->d_name.name);
1822
1823         status = dquot_initialize(dir);
1824         if (status) {
1825                 mlog_errno(status);
1826                 goto bail;
1827         }
1828
1829         sb = dir->i_sb;
1830         osb = OCFS2_SB(sb);
1831
1832         l = strlen(symname) + 1;
1833
1834         credits = ocfs2_calc_symlink_credits(sb);
1835
1836         /* lock the parent directory */
1837         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1838         if (status < 0) {
1839                 if (status != -ENOENT)
1840                         mlog_errno(status);
1841                 return status;
1842         }
1843
1844         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1845         if (!ocfs2_read_links_count(dirfe)) {
1846                 /* can't make a file in a deleted directory. */
1847                 status = -ENOENT;
1848                 goto bail;
1849         }
1850
1851         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1852                                            dentry->d_name.len);
1853         if (status)
1854                 goto bail;
1855
1856         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1857                                               dentry->d_name.name,
1858                                               dentry->d_name.len, &lookup);
1859         if (status < 0) {
1860                 mlog_errno(status);
1861                 goto bail;
1862         }
1863
1864         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1865         if (status < 0) {
1866                 if (status != -ENOSPC)
1867                         mlog_errno(status);
1868                 goto bail;
1869         }
1870
1871         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1872         if (IS_ERR(inode)) {
1873                 status = PTR_ERR(inode);
1874                 inode = NULL;
1875                 mlog_errno(status);
1876                 goto bail;
1877         }
1878
1879         /* get security xattr */
1880         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1881         if (status) {
1882                 if (status == -EOPNOTSUPP)
1883                         si.enable = 0;
1884                 else {
1885                         mlog_errno(status);
1886                         goto bail;
1887                 }
1888         }
1889
1890         /* calculate meta data/clusters for setting security xattr */
1891         if (si.enable) {
1892                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1893                                                   &xattr_credits, &xattr_ac);
1894                 if (status < 0) {
1895                         mlog_errno(status);
1896                         goto bail;
1897                 }
1898         }
1899
1900         /* don't reserve bitmap space for fast symlinks. */
1901         if (l > ocfs2_fast_symlink_chars(sb))
1902                 want_clusters += 1;
1903
1904         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1905         if (status < 0) {
1906                 if (status != -ENOSPC)
1907                         mlog_errno(status);
1908                 goto bail;
1909         }
1910
1911         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1912         if (IS_ERR(handle)) {
1913                 status = PTR_ERR(handle);
1914                 handle = NULL;
1915                 mlog_errno(status);
1916                 goto bail;
1917         }
1918
1919         /* Starting to change things, restart is no longer possible. */
1920         ocfs2_block_signals(&oldset);
1921         did_block_signals = 1;
1922
1923         status = dquot_alloc_inode(inode);
1924         if (status)
1925                 goto bail;
1926         did_quota_inode = 1;
1927
1928         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1929                                    dentry->d_name.name,
1930                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1931                                    inode->i_mode);
1932
1933         status = ocfs2_mknod_locked(osb, dir, inode,
1934                                     0, &new_fe_bh, parent_fe_bh, handle,
1935                                     inode_ac);
1936         if (status < 0) {
1937                 mlog_errno(status);
1938                 goto bail;
1939         }
1940
1941         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1942         inode->i_rdev = 0;
1943         newsize = l - 1;
1944         inode->i_op = &ocfs2_symlink_inode_operations;
1945         inode_nohighmem(inode);
1946         if (l > ocfs2_fast_symlink_chars(sb)) {
1947                 u32 offset = 0;
1948
1949                 status = dquot_alloc_space_nodirty(inode,
1950                     ocfs2_clusters_to_bytes(osb->sb, 1));
1951                 if (status)
1952                         goto bail;
1953                 did_quota = 1;
1954                 inode->i_mapping->a_ops = &ocfs2_aops;
1955                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1956                                               new_fe_bh,
1957                                               handle, data_ac, NULL,
1958                                               NULL);
1959                 if (status < 0) {
1960                         if (status != -ENOSPC && status != -EINTR) {
1961                                 mlog(ML_ERROR,
1962                                      "Failed to extend file to %llu\n",
1963                                      (unsigned long long)newsize);
1964                                 mlog_errno(status);
1965                                 status = -ENOSPC;
1966                         }
1967                         goto bail;
1968                 }
1969                 i_size_write(inode, newsize);
1970                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1971         } else {
1972                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1973                 memcpy((char *) fe->id2.i_symlink, symname, l);
1974                 i_size_write(inode, newsize);
1975                 inode->i_blocks = 0;
1976         }
1977
1978         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1979         if (status < 0) {
1980                 mlog_errno(status);
1981                 goto bail;
1982         }
1983
1984         if (!ocfs2_inode_is_fast_symlink(inode)) {
1985                 status = ocfs2_create_symlink_data(osb, handle, inode,
1986                                                    symname);
1987                 if (status < 0) {
1988                         mlog_errno(status);
1989                         goto bail;
1990                 }
1991         }
1992
1993         if (si.enable) {
1994                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1995                                                  xattr_ac, data_ac);
1996                 if (status < 0) {
1997                         mlog_errno(status);
1998                         goto bail;
1999                 }
2000         }
2001
2002         /*
2003          * Do this before adding the entry to the directory. We add
2004          * also set d_op after success so that ->d_iput() will cleanup
2005          * the dentry lock even if ocfs2_add_entry() fails below.
2006          */
2007         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2008         if (status) {
2009                 mlog_errno(status);
2010                 goto bail;
2011         }
2012
2013         dl = dentry->d_fsdata;
2014
2015         status = ocfs2_add_entry(handle, dentry, inode,
2016                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2017                                  &lookup);
2018         if (status < 0) {
2019                 mlog_errno(status);
2020                 goto bail;
2021         }
2022
2023         insert_inode_hash(inode);
2024         d_instantiate(dentry, inode);
2025 bail:
2026         if (status < 0 && did_quota)
2027                 dquot_free_space_nodirty(inode,
2028                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2029         if (status < 0 && did_quota_inode)
2030                 dquot_free_inode(inode);
2031         if (handle)
2032                 ocfs2_commit_trans(osb, handle);
2033
2034         ocfs2_inode_unlock(dir, 1);
2035         if (did_block_signals)
2036                 ocfs2_unblock_signals(&oldset);
2037
2038         brelse(new_fe_bh);
2039         brelse(parent_fe_bh);
2040         kfree(si.value);
2041         ocfs2_free_dir_lookup_result(&lookup);
2042         if (inode_ac)
2043                 ocfs2_free_alloc_context(inode_ac);
2044         if (data_ac)
2045                 ocfs2_free_alloc_context(data_ac);
2046         if (xattr_ac)
2047                 ocfs2_free_alloc_context(xattr_ac);
2048         if ((status < 0) && inode) {
2049                 if (dl)
2050                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2051
2052                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2053                 clear_nlink(inode);
2054                 iput(inode);
2055         }
2056
2057         if (status)
2058                 mlog_errno(status);
2059
2060         return status;
2061 }
2062
2063 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2064 {
2065         int status, namelen;
2066
2067         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2068                            (long long)blkno);
2069         if (namelen <= 0) {
2070                 if (namelen)
2071                         status = namelen;
2072                 else
2073                         status = -EINVAL;
2074                 mlog_errno(status);
2075                 goto bail;
2076         }
2077         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2078                 status = -EINVAL;
2079                 mlog_errno(status);
2080                 goto bail;
2081         }
2082
2083         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2084
2085         status = 0;
2086 bail:
2087         if (status < 0)
2088                 mlog_errno(status);
2089         return status;
2090 }
2091
2092 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2093                                         struct inode **ret_orphan_dir,
2094                                         struct buffer_head **ret_orphan_dir_bh)
2095 {
2096         struct inode *orphan_dir_inode;
2097         struct buffer_head *orphan_dir_bh = NULL;
2098         int ret = 0;
2099
2100         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2101                                                        ORPHAN_DIR_SYSTEM_INODE,
2102                                                        osb->slot_num);
2103         if (!orphan_dir_inode) {
2104                 ret = -ENOENT;
2105                 mlog_errno(ret);
2106                 return ret;
2107         }
2108
2109         inode_lock(orphan_dir_inode);
2110
2111         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2112         if (ret < 0) {
2113                 inode_unlock(orphan_dir_inode);
2114                 iput(orphan_dir_inode);
2115
2116                 mlog_errno(ret);
2117                 return ret;
2118         }
2119
2120         *ret_orphan_dir = orphan_dir_inode;
2121         *ret_orphan_dir_bh = orphan_dir_bh;
2122
2123         return 0;
2124 }
2125
2126 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2127                                       struct buffer_head *orphan_dir_bh,
2128                                       u64 blkno,
2129                                       char *name,
2130                                       struct ocfs2_dir_lookup_result *lookup,
2131                                       bool dio)
2132 {
2133         int ret;
2134         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2135         int namelen = dio ?
2136                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2137                         OCFS2_ORPHAN_NAMELEN;
2138
2139         if (dio) {
2140                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2141                                 OCFS2_DIO_ORPHAN_PREFIX);
2142                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2143                         ret = -EINVAL;
2144                         mlog_errno(ret);
2145                         return ret;
2146                 }
2147
2148                 ret = ocfs2_blkno_stringify(blkno,
2149                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2150         } else
2151                 ret = ocfs2_blkno_stringify(blkno, name);
2152         if (ret < 0) {
2153                 mlog_errno(ret);
2154                 return ret;
2155         }
2156
2157         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2158                                            orphan_dir_bh, name,
2159                                            namelen, lookup);
2160         if (ret < 0) {
2161                 mlog_errno(ret);
2162                 return ret;
2163         }
2164
2165         return 0;
2166 }
2167
2168 /**
2169  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2170  * insertion of an orphan.
2171  * @osb: ocfs2 file system
2172  * @ret_orphan_dir: Orphan dir inode - returned locked!
2173  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2174  * @lookup: dir lookup result, to be passed back into functions like
2175  *          ocfs2_orphan_add
2176  *
2177  * Returns zero on success and the ret_orphan_dir, name and lookup
2178  * fields will be populated.
2179  *
2180  * Returns non-zero on failure.
2181  */
2182 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2183                                     struct inode **ret_orphan_dir,
2184                                     u64 blkno,
2185                                     char *name,
2186                                     struct ocfs2_dir_lookup_result *lookup,
2187                                     bool dio)
2188 {
2189         struct inode *orphan_dir_inode = NULL;
2190         struct buffer_head *orphan_dir_bh = NULL;
2191         int ret = 0;
2192
2193         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2194                                            &orphan_dir_bh);
2195         if (ret < 0) {
2196                 mlog_errno(ret);
2197                 return ret;
2198         }
2199
2200         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2201                                          blkno, name, lookup, dio);
2202         if (ret < 0) {
2203                 mlog_errno(ret);
2204                 goto out;
2205         }
2206
2207         *ret_orphan_dir = orphan_dir_inode;
2208
2209 out:
2210         brelse(orphan_dir_bh);
2211
2212         if (ret) {
2213                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2214                 inode_unlock(orphan_dir_inode);
2215                 iput(orphan_dir_inode);
2216         }
2217
2218         if (ret)
2219                 mlog_errno(ret);
2220         return ret;
2221 }
2222
2223 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2224                             handle_t *handle,
2225                             struct inode *inode,
2226                             struct buffer_head *fe_bh,
2227                             char *name,
2228                             struct ocfs2_dir_lookup_result *lookup,
2229                             struct inode *orphan_dir_inode,
2230                             bool dio)
2231 {
2232         struct buffer_head *orphan_dir_bh = NULL;
2233         int status = 0;
2234         struct ocfs2_dinode *orphan_fe;
2235         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2236         int namelen = dio ?
2237                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2238                         OCFS2_ORPHAN_NAMELEN;
2239
2240         trace_ocfs2_orphan_add_begin(
2241                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2242
2243         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2244         if (status < 0) {
2245                 mlog_errno(status);
2246                 goto leave;
2247         }
2248
2249         status = ocfs2_journal_access_di(handle,
2250                                          INODE_CACHE(orphan_dir_inode),
2251                                          orphan_dir_bh,
2252                                          OCFS2_JOURNAL_ACCESS_WRITE);
2253         if (status < 0) {
2254                 mlog_errno(status);
2255                 goto leave;
2256         }
2257
2258         /*
2259          * We're going to journal the change of i_flags and i_orphaned_slot.
2260          * It's safe anyway, though some callers may duplicate the journaling.
2261          * Journaling within the func just make the logic look more
2262          * straightforward.
2263          */
2264         status = ocfs2_journal_access_di(handle,
2265                                          INODE_CACHE(inode),
2266                                          fe_bh,
2267                                          OCFS2_JOURNAL_ACCESS_WRITE);
2268         if (status < 0) {
2269                 mlog_errno(status);
2270                 goto leave;
2271         }
2272
2273         /* we're a cluster, and nlink can change on disk from
2274          * underneath us... */
2275         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2276         if (S_ISDIR(inode->i_mode))
2277                 ocfs2_add_links_count(orphan_fe, 1);
2278         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2279         ocfs2_journal_dirty(handle, orphan_dir_bh);
2280
2281         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2282                                    namelen, inode,
2283                                    OCFS2_I(inode)->ip_blkno,
2284                                    orphan_dir_bh, lookup);
2285         if (status < 0) {
2286                 mlog_errno(status);
2287                 goto rollback;
2288         }
2289
2290         if (dio) {
2291                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2292                  * slot.
2293                  */
2294                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2295                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2296         } else {
2297                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2298                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2299
2300                 /* Record which orphan dir our inode now resides
2301                  * in. delete_inode will use this to determine which orphan
2302                  * dir to lock. */
2303                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2304         }
2305
2306         ocfs2_journal_dirty(handle, fe_bh);
2307
2308         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2309                                    osb->slot_num);
2310
2311 rollback:
2312         if (status < 0) {
2313                 if (S_ISDIR(inode->i_mode))
2314                         ocfs2_add_links_count(orphan_fe, -1);
2315                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2316         }
2317
2318 leave:
2319         brelse(orphan_dir_bh);
2320
2321         return status;
2322 }
2323
2324 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2325 int ocfs2_orphan_del(struct ocfs2_super *osb,
2326                      handle_t *handle,
2327                      struct inode *orphan_dir_inode,
2328                      struct inode *inode,
2329                      struct buffer_head *orphan_dir_bh,
2330                      bool dio)
2331 {
2332         const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
2333         char name[namelen + 1];
2334         struct ocfs2_dinode *orphan_fe;
2335         int status = 0;
2336         struct ocfs2_dir_lookup_result lookup = { NULL, };
2337
2338         if (dio) {
2339                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2340                                 OCFS2_DIO_ORPHAN_PREFIX);
2341                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2342                         status = -EINVAL;
2343                         mlog_errno(status);
2344                         return status;
2345                 }
2346
2347                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2348                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2349         } else
2350                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2351         if (status < 0) {
2352                 mlog_errno(status);
2353                 goto leave;
2354         }
2355
2356         trace_ocfs2_orphan_del(
2357              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2358              name, strlen(name));
2359
2360         status = ocfs2_journal_access_di(handle,
2361                                          INODE_CACHE(orphan_dir_inode),
2362                                          orphan_dir_bh,
2363                                          OCFS2_JOURNAL_ACCESS_WRITE);
2364         if (status < 0) {
2365                 mlog_errno(status);
2366                 goto leave;
2367         }
2368
2369         /* find it's spot in the orphan directory */
2370         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2371                                   &lookup);
2372         if (status) {
2373                 mlog_errno(status);
2374                 goto leave;
2375         }
2376
2377         /* remove it from the orphan directory */
2378         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2379         if (status < 0) {
2380                 mlog_errno(status);
2381                 goto leave;
2382         }
2383
2384         /* do the i_nlink dance! :) */
2385         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2386         if (S_ISDIR(inode->i_mode))
2387                 ocfs2_add_links_count(orphan_fe, -1);
2388         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2389         ocfs2_journal_dirty(handle, orphan_dir_bh);
2390
2391 leave:
2392         ocfs2_free_dir_lookup_result(&lookup);
2393
2394         if (status)
2395                 mlog_errno(status);
2396         return status;
2397 }
2398
2399 /**
2400  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2401  * allocated file. This is different from the typical 'add to orphan dir'
2402  * operation in that the inode does not yet exist. This is a problem because
2403  * the orphan dir stringifies the inode block number to come up with it's
2404  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2405  * problem. This function works around it by calling deeper into the orphan
2406  * and suballoc code than other callers. Use this only by necessity.
2407  * @dir: The directory which this inode will ultimately wind up under - not the
2408  * orphan dir!
2409  * @dir_bh: buffer_head the @dir inode block
2410  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2411  * with the string to be used for orphan dirent. Pass back to the orphan dir
2412  * code.
2413  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2414  * dir code.
2415  * @ret_di_blkno: block number where the new inode will be allocated.
2416  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2417  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2418  *
2419  * Returns zero on success and the ret_orphan_dir, name and lookup
2420  * fields will be populated.
2421  *
2422  * Returns non-zero on failure.
2423  */
2424 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2425                                         struct buffer_head *dir_bh,
2426                                         char *orphan_name,
2427                                         struct inode **ret_orphan_dir,
2428                                         u64 *ret_di_blkno,
2429                                         struct ocfs2_dir_lookup_result *orphan_insert,
2430                                         struct ocfs2_alloc_context **ret_inode_ac)
2431 {
2432         int ret;
2433         u64 di_blkno;
2434         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2435         struct inode *orphan_dir = NULL;
2436         struct buffer_head *orphan_dir_bh = NULL;
2437         struct ocfs2_alloc_context *inode_ac = NULL;
2438
2439         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2440         if (ret < 0) {
2441                 mlog_errno(ret);
2442                 return ret;
2443         }
2444
2445         /* reserve an inode spot */
2446         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2447         if (ret < 0) {
2448                 if (ret != -ENOSPC)
2449                         mlog_errno(ret);
2450                 goto out;
2451         }
2452
2453         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2454                                        &di_blkno);
2455         if (ret) {
2456                 mlog_errno(ret);
2457                 goto out;
2458         }
2459
2460         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2461                                          di_blkno, orphan_name, orphan_insert,
2462                                          false);
2463         if (ret < 0) {
2464                 mlog_errno(ret);
2465                 goto out;
2466         }
2467
2468 out:
2469         if (ret == 0) {
2470                 *ret_orphan_dir = orphan_dir;
2471                 *ret_di_blkno = di_blkno;
2472                 *ret_inode_ac = inode_ac;
2473                 /*
2474                  * orphan_name and orphan_insert are already up to
2475                  * date via prepare_orphan_dir
2476                  */
2477         } else {
2478                 /* Unroll reserve_new_inode* */
2479                 if (inode_ac)
2480                         ocfs2_free_alloc_context(inode_ac);
2481
2482                 /* Unroll orphan dir locking */
2483                 inode_unlock(orphan_dir);
2484                 ocfs2_inode_unlock(orphan_dir, 1);
2485                 iput(orphan_dir);
2486         }
2487
2488         brelse(orphan_dir_bh);
2489
2490         return ret;
2491 }
2492
2493 int ocfs2_create_inode_in_orphan(struct inode *dir,
2494                                  int mode,
2495                                  struct inode **new_inode)
2496 {
2497         int status, did_quota_inode = 0;
2498         struct inode *inode = NULL;
2499         struct inode *orphan_dir = NULL;
2500         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2501         struct ocfs2_dinode *di = NULL;
2502         handle_t *handle = NULL;
2503         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2504         struct buffer_head *parent_di_bh = NULL;
2505         struct buffer_head *new_di_bh = NULL;
2506         struct ocfs2_alloc_context *inode_ac = NULL;
2507         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2508         u64 uninitialized_var(di_blkno), suballoc_loc;
2509         u16 suballoc_bit;
2510
2511         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2512         if (status < 0) {
2513                 if (status != -ENOENT)
2514                         mlog_errno(status);
2515                 return status;
2516         }
2517
2518         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2519                                               orphan_name, &orphan_dir,
2520                                               &di_blkno, &orphan_insert, &inode_ac);
2521         if (status < 0) {
2522                 if (status != -ENOSPC)
2523                         mlog_errno(status);
2524                 goto leave;
2525         }
2526
2527         inode = ocfs2_get_init_inode(dir, mode);
2528         if (IS_ERR(inode)) {
2529                 status = PTR_ERR(inode);
2530                 inode = NULL;
2531                 mlog_errno(status);
2532                 goto leave;
2533         }
2534
2535         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2536         if (IS_ERR(handle)) {
2537                 status = PTR_ERR(handle);
2538                 handle = NULL;
2539                 mlog_errno(status);
2540                 goto leave;
2541         }
2542
2543         status = dquot_alloc_inode(inode);
2544         if (status)
2545                 goto leave;
2546         did_quota_inode = 1;
2547
2548         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2549                                               &suballoc_loc,
2550                                               &suballoc_bit, di_blkno);
2551         if (status < 0) {
2552                 mlog_errno(status);
2553                 goto leave;
2554         }
2555
2556         clear_nlink(inode);
2557         /* do the real work now. */
2558         status = __ocfs2_mknod_locked(dir, inode,
2559                                       0, &new_di_bh, parent_di_bh, handle,
2560                                       inode_ac, di_blkno, suballoc_loc,
2561                                       suballoc_bit);
2562         if (status < 0) {
2563                 mlog_errno(status);
2564                 goto leave;
2565         }
2566
2567         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2568         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2569                                   &orphan_insert, orphan_dir, false);
2570         if (status < 0) {
2571                 mlog_errno(status);
2572                 goto leave;
2573         }
2574
2575         /* get open lock so that only nodes can't remove it from orphan dir. */
2576         status = ocfs2_open_lock(inode);
2577         if (status < 0)
2578                 mlog_errno(status);
2579
2580         insert_inode_hash(inode);
2581 leave:
2582         if (status < 0 && did_quota_inode)
2583                 dquot_free_inode(inode);
2584         if (handle)
2585                 ocfs2_commit_trans(osb, handle);
2586
2587         if (orphan_dir) {
2588                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2589                 ocfs2_inode_unlock(orphan_dir, 1);
2590                 inode_unlock(orphan_dir);
2591                 iput(orphan_dir);
2592         }
2593
2594         if ((status < 0) && inode) {
2595                 clear_nlink(inode);
2596                 iput(inode);
2597         }
2598
2599         if (inode_ac)
2600                 ocfs2_free_alloc_context(inode_ac);
2601
2602         brelse(new_di_bh);
2603
2604         if (!status)
2605                 *new_inode = inode;
2606
2607         ocfs2_free_dir_lookup_result(&orphan_insert);
2608
2609         ocfs2_inode_unlock(dir, 1);
2610         brelse(parent_di_bh);
2611         return status;
2612 }
2613
2614 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2615         struct inode *inode)
2616 {
2617         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2618         struct inode *orphan_dir_inode = NULL;
2619         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2620         struct buffer_head *di_bh = NULL;
2621         int status = 0;
2622         handle_t *handle = NULL;
2623         struct ocfs2_dinode *di = NULL;
2624
2625         status = ocfs2_inode_lock(inode, &di_bh, 1);
2626         if (status < 0) {
2627                 mlog_errno(status);
2628                 goto bail;
2629         }
2630
2631         di = (struct ocfs2_dinode *) di_bh->b_data;
2632         /*
2633          * Another append dio crashed?
2634          * If so, manually recover it first.
2635          */
2636         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2637                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2638                 if (status < 0) {
2639                         if (status != -ENOSPC)
2640                                 mlog_errno(status);
2641                         goto bail_unlock_inode;
2642                 }
2643
2644                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2645                 if (status < 0) {
2646                         mlog_errno(status);
2647                         goto bail_unlock_inode;
2648                 }
2649         }
2650
2651         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2652                         OCFS2_I(inode)->ip_blkno,
2653                         orphan_name,
2654                         &orphan_insert,
2655                         true);
2656         if (status < 0) {
2657                 mlog_errno(status);
2658                 goto bail_unlock_inode;
2659         }
2660
2661         handle = ocfs2_start_trans(osb,
2662                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2663         if (IS_ERR(handle)) {
2664                 status = PTR_ERR(handle);
2665                 goto bail_unlock_orphan;
2666         }
2667
2668         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2669                         &orphan_insert, orphan_dir_inode, true);
2670         if (status)
2671                 mlog_errno(status);
2672
2673         ocfs2_commit_trans(osb, handle);
2674
2675 bail_unlock_orphan:
2676         ocfs2_inode_unlock(orphan_dir_inode, 1);
2677         inode_unlock(orphan_dir_inode);
2678         iput(orphan_dir_inode);
2679
2680         ocfs2_free_dir_lookup_result(&orphan_insert);
2681
2682 bail_unlock_inode:
2683         ocfs2_inode_unlock(inode, 1);
2684         brelse(di_bh);
2685
2686 bail:
2687         return status;
2688 }
2689
2690 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2691                 struct inode *inode, struct buffer_head *di_bh,
2692                 int update_isize, loff_t end)
2693 {
2694         struct inode *orphan_dir_inode = NULL;
2695         struct buffer_head *orphan_dir_bh = NULL;
2696         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2697         handle_t *handle = NULL;
2698         int status = 0;
2699
2700         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2701                         ORPHAN_DIR_SYSTEM_INODE,
2702                         le16_to_cpu(di->i_dio_orphaned_slot));
2703         if (!orphan_dir_inode) {
2704                 status = -ENOENT;
2705                 mlog_errno(status);
2706                 goto bail;
2707         }
2708
2709         inode_lock(orphan_dir_inode);
2710         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2711         if (status < 0) {
2712                 inode_unlock(orphan_dir_inode);
2713                 iput(orphan_dir_inode);
2714                 mlog_errno(status);
2715                 goto bail;
2716         }
2717
2718         handle = ocfs2_start_trans(osb,
2719                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2720         if (IS_ERR(handle)) {
2721                 status = PTR_ERR(handle);
2722                 goto bail_unlock_orphan;
2723         }
2724
2725         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2726
2727         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2728                                 inode, orphan_dir_bh, true);
2729         if (status < 0) {
2730                 mlog_errno(status);
2731                 goto bail_commit;
2732         }
2733
2734         status = ocfs2_journal_access_di(handle,
2735                         INODE_CACHE(inode),
2736                         di_bh,
2737                         OCFS2_JOURNAL_ACCESS_WRITE);
2738         if (status < 0) {
2739                 mlog_errno(status);
2740                 goto bail_commit;
2741         }
2742
2743         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2744         di->i_dio_orphaned_slot = 0;
2745
2746         if (update_isize) {
2747                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2748                 if (status)
2749                         mlog_errno(status);
2750         } else
2751                 ocfs2_journal_dirty(handle, di_bh);
2752
2753 bail_commit:
2754         ocfs2_commit_trans(osb, handle);
2755
2756 bail_unlock_orphan:
2757         ocfs2_inode_unlock(orphan_dir_inode, 1);
2758         inode_unlock(orphan_dir_inode);
2759         brelse(orphan_dir_bh);
2760         iput(orphan_dir_inode);
2761
2762 bail:
2763         return status;
2764 }
2765
2766 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2767                                    struct inode *inode,
2768                                    struct dentry *dentry)
2769 {
2770         int status = 0;
2771         struct buffer_head *parent_di_bh = NULL;
2772         handle_t *handle = NULL;
2773         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2774         struct ocfs2_dinode *dir_di, *di;
2775         struct inode *orphan_dir_inode = NULL;
2776         struct buffer_head *orphan_dir_bh = NULL;
2777         struct buffer_head *di_bh = NULL;
2778         struct ocfs2_dir_lookup_result lookup = { NULL, };
2779
2780         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2781                                 dentry->d_name.len, dentry->d_name.name,
2782                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2783                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2784
2785         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2786         if (status < 0) {
2787                 if (status != -ENOENT)
2788                         mlog_errno(status);
2789                 return status;
2790         }
2791
2792         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2793         if (!dir_di->i_links_count) {
2794                 /* can't make a file in a deleted directory. */
2795                 status = -ENOENT;
2796                 goto leave;
2797         }
2798
2799         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2800                                            dentry->d_name.len);
2801         if (status)
2802                 goto leave;
2803
2804         /* get a spot inside the dir. */
2805         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2806                                               dentry->d_name.name,
2807                                               dentry->d_name.len, &lookup);
2808         if (status < 0) {
2809                 mlog_errno(status);
2810                 goto leave;
2811         }
2812
2813         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2814                                                        ORPHAN_DIR_SYSTEM_INODE,
2815                                                        osb->slot_num);
2816         if (!orphan_dir_inode) {
2817                 status = -ENOENT;
2818                 mlog_errno(status);
2819                 goto leave;
2820         }
2821
2822         inode_lock(orphan_dir_inode);
2823
2824         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2825         if (status < 0) {
2826                 mlog_errno(status);
2827                 inode_unlock(orphan_dir_inode);
2828                 iput(orphan_dir_inode);
2829                 goto leave;
2830         }
2831
2832         status = ocfs2_read_inode_block(inode, &di_bh);
2833         if (status < 0) {
2834                 mlog_errno(status);
2835                 goto orphan_unlock;
2836         }
2837
2838         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2839         if (IS_ERR(handle)) {
2840                 status = PTR_ERR(handle);
2841                 handle = NULL;
2842                 mlog_errno(status);
2843                 goto orphan_unlock;
2844         }
2845
2846         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2847                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2848         if (status < 0) {
2849                 mlog_errno(status);
2850                 goto out_commit;
2851         }
2852
2853         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2854                                   orphan_dir_bh, false);
2855         if (status < 0) {
2856                 mlog_errno(status);
2857                 goto out_commit;
2858         }
2859
2860         di = (struct ocfs2_dinode *)di_bh->b_data;
2861         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2862         di->i_orphaned_slot = 0;
2863         set_nlink(inode, 1);
2864         ocfs2_set_links_count(di, inode->i_nlink);
2865         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2866         ocfs2_journal_dirty(handle, di_bh);
2867
2868         status = ocfs2_add_entry(handle, dentry, inode,
2869                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2870                                  &lookup);
2871         if (status < 0) {
2872                 mlog_errno(status);
2873                 goto out_commit;
2874         }
2875
2876         status = ocfs2_dentry_attach_lock(dentry, inode,
2877                                           OCFS2_I(dir)->ip_blkno);
2878         if (status) {
2879                 mlog_errno(status);
2880                 goto out_commit;
2881         }
2882
2883         d_instantiate(dentry, inode);
2884         status = 0;
2885 out_commit:
2886         ocfs2_commit_trans(osb, handle);
2887 orphan_unlock:
2888         ocfs2_inode_unlock(orphan_dir_inode, 1);
2889         inode_unlock(orphan_dir_inode);
2890         iput(orphan_dir_inode);
2891 leave:
2892
2893         ocfs2_inode_unlock(dir, 1);
2894
2895         brelse(di_bh);
2896         brelse(parent_di_bh);
2897         brelse(orphan_dir_bh);
2898
2899         ocfs2_free_dir_lookup_result(&lookup);
2900
2901         if (status)
2902                 mlog_errno(status);
2903
2904         return status;
2905 }
2906
2907 const struct inode_operations ocfs2_dir_iops = {
2908         .create         = ocfs2_create,
2909         .lookup         = ocfs2_lookup,
2910         .link           = ocfs2_link,
2911         .unlink         = ocfs2_unlink,
2912         .rmdir          = ocfs2_unlink,
2913         .symlink        = ocfs2_symlink,
2914         .mkdir          = ocfs2_mkdir,
2915         .mknod          = ocfs2_mknod,
2916         .rename         = ocfs2_rename,
2917         .setattr        = ocfs2_setattr,
2918         .getattr        = ocfs2_getattr,
2919         .permission     = ocfs2_permission,
2920         .listxattr      = ocfs2_listxattr,
2921         .fiemap         = ocfs2_fiemap,
2922         .get_acl        = ocfs2_iop_get_acl,
2923         .set_acl        = ocfs2_iop_set_acl,
2924 };