fs/ocfs2/namei.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * namei.c
   4  *
   5  * Create and rename file, directory, symlinks
   6  *
   7  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   8  *
   9  *  Portions of this code from linux/fs/ext3/dir.c
  10  *
  11  *  Copyright (C) 1992, 1993, 1994, 1995
  12  *  Remy Card (card@masi.ibp.fr)
  13  *  Laboratoire MASI - Institut Blaise pascal
  14  *  Universite Pierre et Marie Curie (Paris VI)
  15  *
  16  *   from
  17  *
  18  *   linux/fs/minix/dir.c
  19  *
  20  *   Copyright (C) 1991, 1992 Linux Torvalds
  21  */
  22
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/quotaops.h>
  28 #include <linux/iversion.h>
  29
  30 #include <cluster/masklog.h>
  31
  32 #include "ocfs2.h"
  33
  34 #include "alloc.h"
  35 #include "dcache.h"
  36 #include "dir.h"
  37 #include "dlmglue.h"
  38 #include "extent_map.h"
  39 #include "file.h"
  40 #include "inode.h"
  41 #include "journal.h"
  42 #include "namei.h"
  43 #include "suballoc.h"
  44 #include "super.h"
  45 #include "symlink.h"
  46 #include "sysfile.h"
  47 #include "uptodate.h"
  48 #include "xattr.h"
  49 #include "acl.h"
  50 #include "ocfs2_trace.h"
  51 #include "ioctl.h"
  52
  53 #include "buffer_head_io.h"
  54
  55 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  56                               struct inode *dir,
  57                               struct inode *inode,
  58                               dev_t dev,
  59                               struct buffer_head **new_fe_bh,
  60                               struct buffer_head *parent_fe_bh,
  61                               handle_t *handle,
  62                               struct ocfs2_alloc_context *inode_ac);
  63
  64 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  65                                     struct inode **ret_orphan_dir,
  66                                     u64 blkno,
  67                                     char *name,
  68                                     struct ocfs2_dir_lookup_result *lookup,
  69                                     bool dio);
  70
  71 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  72                             handle_t *handle,
  73                             struct inode *inode,
  74                             struct buffer_head *fe_bh,
  75                             char *name,
  76                             struct ocfs2_dir_lookup_result *lookup,
  77                             struct inode *orphan_dir_inode,
  78                             bool dio);
  79
  80 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  81                                      handle_t *handle,
  82                                      struct inode *inode,
  83                                      const char *symname);
  84
  85 static int ocfs2_double_lock(struct ocfs2_super *osb,
  86                              struct buffer_head **bh1,
  87                              struct inode *inode1,
  88                              struct buffer_head **bh2,
  89                              struct inode *inode2,
  90                              int rename);
  91
  92 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
  93 /* An orphan dir name is an 8 byte value, printed as a hex string */
  94 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  95
  96 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
  97                                    unsigned int flags)
  98 {
  99         int status;
 100         u64 blkno;
 101         struct inode *inode = NULL;
 102         struct dentry *ret;
 103         struct ocfs2_inode_info *oi;
 104
 105         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 106                            dentry->d_name.name,
 107                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 108
 109         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 110                 ret = ERR_PTR(-ENAMETOOLONG);
 111                 goto bail;
 112         }
 113
 114         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 115         if (status < 0) {
 116                 if (status != -ENOENT)
 117                         mlog_errno(status);
 118                 ret = ERR_PTR(status);
 119                 goto bail;
 120         }
 121
 122         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 123                                             dentry->d_name.len, &blkno);
 124         if (status < 0)
 125                 goto bail_add;
 126
 127         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 128         if (IS_ERR(inode)) {
 129                 ret = ERR_PTR(-EACCES);
 130                 goto bail_unlock;
 131         }
 132
 133         oi = OCFS2_I(inode);
 134         /* Clear any orphaned state... If we were able to look up the
 135          * inode from a directory, it certainly can't be orphaned. We
 136          * might have the bad state from a node which intended to
 137          * orphan this inode but crashed before it could commit the
 138          * unlink. */
 139         spin_lock(&oi->ip_lock);
 140         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 141         spin_unlock(&oi->ip_lock);
 142
 143 bail_add:
 144         ret = d_splice_alias(inode, dentry);
 145
 146         if (inode) {
 147                 /*
 148                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 149                  * dentry, it will d_move() it on top of ourse. The
 150                  * return value will indicate this however, so in
 151                  * those cases, we switch them around for the locking
 152                  * code.
 153                  *
 154                  * NOTE: This dentry already has ->d_op set from
 155                  * ocfs2_get_parent() and ocfs2_get_dentry()
 156                  */
 157                 if (!IS_ERR_OR_NULL(ret))
 158                         dentry = ret;
 159
 160                 status = ocfs2_dentry_attach_lock(dentry, inode,
 161                                                   OCFS2_I(dir)->ip_blkno);
 162                 if (status) {
 163                         mlog_errno(status);
 164                         ret = ERR_PTR(status);
 165                         goto bail_unlock;
 166                 }
 167         } else
 168                 ocfs2_dentry_attach_gen(dentry);
 169
 170 bail_unlock:
 171         /* Don't drop the cluster lock until *after* the d_add --
 172          * unlink on another node will message us to remove that
 173          * dentry under this lock so otherwise we can race this with
 174          * the downconvert thread and have a stale dentry. */
 175         ocfs2_inode_unlock(dir, 0);
 176
 177 bail:
 178
 179         trace_ocfs2_lookup_ret(ret);
 180
 181         return ret;
 182 }
 183
 184 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 185 {
 186         struct inode *inode;
 187         int status;
 188
 189         inode = new_inode(dir->i_sb);
 190         if (!inode) {
 191                 mlog(ML_ERROR, "new_inode failed!\n");
 192                 return ERR_PTR(-ENOMEM);
 193         }
 194
 195         /* populate as many fields early on as possible - many of
 196          * these are used by the support functions here and in
 197          * callers. */
 198         if (S_ISDIR(mode))
 199                 set_nlink(inode, 2);
 200         inode_init_owner(&init_user_ns, inode, dir, mode);
 201         status = dquot_initialize(inode);
 202         if (status)
 203                 return ERR_PTR(status);
 204
 205         return inode;
 206 }
 207
 208 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 209                 struct dentry *dentry, struct inode *inode)
 210 {
 211         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 212
 213         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 214         ocfs2_lock_res_free(&dl->dl_lockres);
 215         BUG_ON(dl->dl_count != 1);
 216         spin_lock(&dentry_attach_lock);
 217         dentry->d_fsdata = NULL;
 218         spin_unlock(&dentry_attach_lock);
 219         kfree(dl);
 220         iput(inode);
 221 }
 222
 223 static int ocfs2_mknod(struct user_namespace *mnt_userns,
 224                        struct inode *dir,
 225                        struct dentry *dentry,
 226                        umode_t mode,
 227                        dev_t dev)
 228 {
 229         int status = 0;
 230         struct buffer_head *parent_fe_bh = NULL;
 231         handle_t *handle = NULL;
 232         struct ocfs2_super *osb;
 233         struct ocfs2_dinode *dirfe;
 234         struct buffer_head *new_fe_bh = NULL;
 235         struct inode *inode = NULL;
 236         struct ocfs2_alloc_context *inode_ac = NULL;
 237         struct ocfs2_alloc_context *data_ac = NULL;
 238         struct ocfs2_alloc_context *meta_ac = NULL;
 239         int want_clusters = 0;
 240         int want_meta = 0;
 241         int xattr_credits = 0;
 242         struct ocfs2_security_xattr_info si = {
 243                 .enable = 1,
 244         };
 245         int did_quota_inode = 0;
 246         struct ocfs2_dir_lookup_result lookup = { NULL, };
 247         sigset_t oldset;
 248         int did_block_signals = 0;
 249         struct ocfs2_dentry_lock *dl = NULL;
 250
 251         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 252                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 253                           (unsigned long)dev, mode);
 254
 255         status = dquot_initialize(dir);
 256         if (status) {
 257                 mlog_errno(status);
 258                 return status;
 259         }
 260
 261         /* get our super block */
 262         osb = OCFS2_SB(dir->i_sb);
 263
 264         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 265         if (status < 0) {
 266                 if (status != -ENOENT)
 267                         mlog_errno(status);
 268                 return status;
 269         }
 270
 271         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 272                 status = -EMLINK;
 273                 goto leave;
 274         }
 275
 276         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 277         if (!ocfs2_read_links_count(dirfe)) {
 278                 /* can't make a file in a deleted directory. */
 279                 status = -ENOENT;
 280                 goto leave;
 281         }
 282
 283         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 284                                            dentry->d_name.len);
 285         if (status)
 286                 goto leave;
 287
 288         /* get a spot inside the dir. */
 289         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 290                                               dentry->d_name.name,
 291                                               dentry->d_name.len, &lookup);
 292         if (status < 0) {
 293                 mlog_errno(status);
 294                 goto leave;
 295         }
 296
 297         /* reserve an inode spot */
 298         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 299         if (status < 0) {
 300                 if (status != -ENOSPC)
 301                         mlog_errno(status);
 302                 goto leave;
 303         }
 304
 305         inode = ocfs2_get_init_inode(dir, mode);
 306         if (IS_ERR(inode)) {
 307                 status = PTR_ERR(inode);
 308                 inode = NULL;
 309                 mlog_errno(status);
 310                 goto leave;
 311         }
 312
 313         /* get security xattr */
 314         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 315         if (status) {
 316                 if (status == -EOPNOTSUPP)
 317                         si.enable = 0;
 318                 else {
 319                         mlog_errno(status);
 320                         goto leave;
 321                 }
 322         }
 323
 324         /* calculate meta data/clusters for setting security and acl xattr */
 325         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 326                                        &si, &want_clusters,
 327                                        &xattr_credits, &want_meta);
 328         if (status < 0) {
 329                 mlog_errno(status);
 330                 goto leave;
 331         }
 332
 333         /* Reserve a cluster if creating an extent based directory. */
 334         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 335                 want_clusters += 1;
 336
 337                 /* Dir indexing requires extra space as well */
 338                 if (ocfs2_supports_indexed_dirs(osb))
 339                         want_meta++;
 340         }
 341
 342         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 343         if (status < 0) {
 344                 if (status != -ENOSPC)
 345                         mlog_errno(status);
 346                 goto leave;
 347         }
 348
 349         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 350         if (status < 0) {
 351                 if (status != -ENOSPC)
 352                         mlog_errno(status);
 353                 goto leave;
 354         }
 355
 356         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 357                                                             S_ISDIR(mode),
 358                                                             xattr_credits));
 359         if (IS_ERR(handle)) {
 360                 status = PTR_ERR(handle);
 361                 handle = NULL;
 362                 mlog_errno(status);
 363                 goto leave;
 364         }
 365
 366         /* Starting to change things, restart is no longer possible. */
 367         ocfs2_block_signals(&oldset);
 368         did_block_signals = 1;
 369
 370         status = dquot_alloc_inode(inode);
 371         if (status)
 372                 goto leave;
 373         did_quota_inode = 1;
 374
 375         /* do the real work now. */
 376         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 377                                     &new_fe_bh, parent_fe_bh, handle,
 378                                     inode_ac);
 379         if (status < 0) {
 380                 mlog_errno(status);
 381                 goto leave;
 382         }
 383
 384         if (S_ISDIR(mode)) {
 385                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 386                                             new_fe_bh, data_ac, meta_ac);
 387                 if (status < 0) {
 388                         mlog_errno(status);
 389                         goto leave;
 390                 }
 391
 392                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 393                                                  parent_fe_bh,
 394                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 395                 if (status < 0) {
 396                         mlog_errno(status);
 397                         goto leave;
 398                 }
 399                 ocfs2_add_links_count(dirfe, 1);
 400                 ocfs2_journal_dirty(handle, parent_fe_bh);
 401                 inc_nlink(dir);
 402         }
 403
 404         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 405                          meta_ac, data_ac);
 406
 407         if (status < 0) {
 408                 mlog_errno(status);
 409                 goto roll_back;
 410         }
 411
 412         if (si.enable) {
 413                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 414                                                  meta_ac, data_ac);
 415                 if (status < 0) {
 416                         mlog_errno(status);
 417                         goto roll_back;
 418                 }
 419         }
 420
 421         /*
 422          * Do this before adding the entry to the directory. We add
 423          * also set d_op after success so that ->d_iput() will cleanup
 424          * the dentry lock even if ocfs2_add_entry() fails below.
 425          */
 426         status = ocfs2_dentry_attach_lock(dentry, inode,
 427                                           OCFS2_I(dir)->ip_blkno);
 428         if (status) {
 429                 mlog_errno(status);
 430                 goto roll_back;
 431         }
 432
 433         dl = dentry->d_fsdata;
 434
 435         status = ocfs2_add_entry(handle, dentry, inode,
 436                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 437                                  &lookup);
 438         if (status < 0) {
 439                 mlog_errno(status);
 440                 goto roll_back;
 441         }
 442
 443         insert_inode_hash(inode);
 444         d_instantiate(dentry, inode);
 445         status = 0;
 446
 447 roll_back:
 448         if (status < 0 && S_ISDIR(mode)) {
 449                 ocfs2_add_links_count(dirfe, -1);
 450                 drop_nlink(dir);
 451         }
 452
 453 leave:
 454         if (status < 0 && did_quota_inode)
 455                 dquot_free_inode(inode);
 456         if (handle)
 457                 ocfs2_commit_trans(osb, handle);
 458
 459         ocfs2_inode_unlock(dir, 1);
 460         if (did_block_signals)
 461                 ocfs2_unblock_signals(&oldset);
 462
 463         brelse(new_fe_bh);
 464         brelse(parent_fe_bh);
 465         kfree(si.value);
 466
 467         ocfs2_free_dir_lookup_result(&lookup);
 468
 469         if (inode_ac)
 470                 ocfs2_free_alloc_context(inode_ac);
 471
 472         if (data_ac)
 473                 ocfs2_free_alloc_context(data_ac);
 474
 475         if (meta_ac)
 476                 ocfs2_free_alloc_context(meta_ac);
 477
 478         /*
 479          * We should call iput after the i_mutex of the bitmap been
 480          * unlocked in ocfs2_free_alloc_context, or the
 481          * ocfs2_delete_inode will mutex_lock again.
 482          */
 483         if ((status < 0) && inode) {
 484                 if (dl)
 485                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 486
 487                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 488                 clear_nlink(inode);
 489                 iput(inode);
 490         }
 491
 492         if (status)
 493                 mlog_errno(status);
 494
 495         return status;
 496 }
 497
 498 static int __ocfs2_mknod_locked(struct inode *dir,
 499                                 struct inode *inode,
 500                                 dev_t dev,
 501                                 struct buffer_head **new_fe_bh,
 502                                 struct buffer_head *parent_fe_bh,
 503                                 handle_t *handle,
 504                                 struct ocfs2_alloc_context *inode_ac,
 505                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 506 {
 507         int status = 0;
 508         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 509         struct ocfs2_dinode *fe = NULL;
 510         struct ocfs2_extent_list *fel;
 511         u16 feat;
 512         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 513         struct timespec64 ts;
 514
 515         *new_fe_bh = NULL;
 516
 517         /* populate as many fields early on as possible - many of
 518          * these are used by the support functions here and in
 519          * callers. */
 520         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 521         oi->ip_blkno = fe_blkno;
 522         spin_lock(&osb->osb_lock);
 523         inode->i_generation = osb->s_next_generation++;
 524         spin_unlock(&osb->osb_lock);
 525
 526         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 527         if (!*new_fe_bh) {
 528                 status = -ENOMEM;
 529                 mlog_errno(status);
 530                 goto leave;
 531         }
 532         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 533
 534         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 535                                          *new_fe_bh,
 536                                          OCFS2_JOURNAL_ACCESS_CREATE);
 537         if (status < 0) {
 538                 mlog_errno(status);
 539                 goto leave;
 540         }
 541
 542         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 543         memset(fe, 0, osb->sb->s_blocksize);
 544
 545         fe->i_generation = cpu_to_le32(inode->i_generation);
 546         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 547         fe->i_blkno = cpu_to_le64(fe_blkno);
 548         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 549         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 550         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 551         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 552         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 553         fe->i_mode = cpu_to_le16(inode->i_mode);
 554         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 555                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 556
 557         ocfs2_set_links_count(fe, inode->i_nlink);
 558
 559         fe->i_last_eb_blk = 0;
 560         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 561         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 562         ktime_get_real_ts64(&ts);
 563         fe->i_atime = fe->i_ctime = fe->i_mtime =
 564                 cpu_to_le64(ts.tv_sec);
 565         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 566                 cpu_to_le32(ts.tv_nsec);
 567         fe->i_dtime = 0;
 568
 569         /*
 570          * If supported, directories start with inline data. If inline
 571          * isn't supported, but indexing is, we start them as indexed.
 572          */
 573         feat = le16_to_cpu(fe->i_dyn_features);
 574         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 575                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 576
 577                 fe->id2.i_data.id_count = cpu_to_le16(
 578                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 579         } else {
 580                 fel = &fe->id2.i_list;
 581                 fel->l_tree_depth = 0;
 582                 fel->l_next_free_rec = 0;
 583                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 584         }
 585
 586         ocfs2_journal_dirty(handle, *new_fe_bh);
 587
 588         ocfs2_populate_inode(inode, fe, 1);
 589         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 590         if (!ocfs2_mount_local(osb)) {
 591                 status = ocfs2_create_new_inode_locks(inode);
 592                 if (status < 0)
 593                         mlog_errno(status);
 594         }
 595
 596         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 597
 598 leave:
 599         if (status < 0) {
 600                 if (*new_fe_bh) {
 601                         brelse(*new_fe_bh);
 602                         *new_fe_bh = NULL;
 603                 }
 604         }
 605
 606         if (status)
 607                 mlog_errno(status);
 608         return status;
 609 }
 610
 611 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 612                               struct inode *dir,
 613                               struct inode *inode,
 614                               dev_t dev,
 615                               struct buffer_head **new_fe_bh,
 616                               struct buffer_head *parent_fe_bh,
 617                               handle_t *handle,
 618                               struct ocfs2_alloc_context *inode_ac)
 619 {
 620         int status = 0;
 621         u64 suballoc_loc, fe_blkno = 0;
 622         u16 suballoc_bit;
 623
 624         *new_fe_bh = NULL;
 625
 626         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 627                                        inode_ac, &suballoc_loc,
 628                                        &suballoc_bit, &fe_blkno);
 629         if (status < 0) {
 630                 mlog_errno(status);
 631                 return status;
 632         }
 633
 634         status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 635                                     parent_fe_bh, handle, inode_ac,
 636                                     fe_blkno, suballoc_loc, suballoc_bit);
 637         if (status < 0) {
 638                 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
 639                 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
 640                                 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
 641                 if (tmp)
 642                         mlog_errno(tmp);
 643         }
 644
 645         return status;
 646 }
 647
 648 static int ocfs2_mkdir(struct user_namespace *mnt_userns,
 649                        struct inode *dir,
 650                        struct dentry *dentry,
 651                        umode_t mode)
 652 {
 653         int ret;
 654
 655         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 656                           OCFS2_I(dir)->ip_blkno, mode);
 657         ret = ocfs2_mknod(&init_user_ns, dir, dentry, mode | S_IFDIR, 0);
 658         if (ret)
 659                 mlog_errno(ret);
 660
 661         return ret;
 662 }
 663
 664 static int ocfs2_create(struct user_namespace *mnt_userns,
 665                         struct inode *dir,
 666                         struct dentry *dentry,
 667                         umode_t mode,
 668                         bool excl)
 669 {
 670         int ret;
 671
 672         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 673                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 674         ret = ocfs2_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
 675         if (ret)
 676                 mlog_errno(ret);
 677
 678         return ret;
 679 }
 680
 681 static int ocfs2_link(struct dentry *old_dentry,
 682                       struct inode *dir,
 683                       struct dentry *dentry)
 684 {
 685         handle_t *handle;
 686         struct inode *inode = d_inode(old_dentry);
 687         struct inode *old_dir = d_inode(old_dentry->d_parent);
 688         int err;
 689         struct buffer_head *fe_bh = NULL;
 690         struct buffer_head *old_dir_bh = NULL;
 691         struct buffer_head *parent_fe_bh = NULL;
 692         struct ocfs2_dinode *fe = NULL;
 693         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 694         struct ocfs2_dir_lookup_result lookup = { NULL, };
 695         sigset_t oldset;
 696         u64 old_de_ino;
 697
 698         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 699                          old_dentry->d_name.len, old_dentry->d_name.name,
 700                          dentry->d_name.len, dentry->d_name.name);
 701
 702         if (S_ISDIR(inode->i_mode))
 703                 return -EPERM;
 704
 705         err = dquot_initialize(dir);
 706         if (err) {
 707                 mlog_errno(err);
 708                 return err;
 709         }
 710
 711         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 712                         &parent_fe_bh, dir, 0);
 713         if (err < 0) {
 714                 if (err != -ENOENT)
 715                         mlog_errno(err);
 716                 return err;
 717         }
 718
 719         /* make sure both dirs have bhs
 720          * get an extra ref on old_dir_bh if old==new */
 721         if (!parent_fe_bh) {
 722                 if (old_dir_bh) {
 723                         parent_fe_bh = old_dir_bh;
 724                         get_bh(parent_fe_bh);
 725                 } else {
 726                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 727                         err = -EIO;
 728                         goto out;
 729                 }
 730         }
 731
 732         if (!dir->i_nlink) {
 733                 err = -ENOENT;
 734                 goto out;
 735         }
 736
 737         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 738                         old_dentry->d_name.len, &old_de_ino);
 739         if (err) {
 740                 err = -ENOENT;
 741                 goto out;
 742         }
 743
 744         /*
 745          * Check whether another node removed the source inode while we
 746          * were in the vfs.
 747          */
 748         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 749                 err = -ENOENT;
 750                 goto out;
 751         }
 752
 753         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 754                                         dentry->d_name.len);
 755         if (err)
 756                 goto out;
 757
 758         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 759                                            dentry->d_name.name,
 760                                            dentry->d_name.len, &lookup);
 761         if (err < 0) {
 762                 mlog_errno(err);
 763                 goto out;
 764         }
 765
 766         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 767         if (err < 0) {
 768                 if (err != -ENOENT)
 769                         mlog_errno(err);
 770                 goto out;
 771         }
 772
 773         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 774         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 775                 err = -EMLINK;
 776                 goto out_unlock_inode;
 777         }
 778
 779         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 780         if (IS_ERR(handle)) {
 781                 err = PTR_ERR(handle);
 782                 handle = NULL;
 783                 mlog_errno(err);
 784                 goto out_unlock_inode;
 785         }
 786
 787         /* Starting to change things, restart is no longer possible. */
 788         ocfs2_block_signals(&oldset);
 789
 790         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 791                                       OCFS2_JOURNAL_ACCESS_WRITE);
 792         if (err < 0) {
 793                 mlog_errno(err);
 794                 goto out_commit;
 795         }
 796
 797         inc_nlink(inode);
 798         inode->i_ctime = current_time(inode);
 799         ocfs2_set_links_count(fe, inode->i_nlink);
 800         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 801         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 802         ocfs2_journal_dirty(handle, fe_bh);
 803
 804         err = ocfs2_add_entry(handle, dentry, inode,
 805                               OCFS2_I(inode)->ip_blkno,
 806                               parent_fe_bh, &lookup);
 807         if (err) {
 808                 ocfs2_add_links_count(fe, -1);
 809                 drop_nlink(inode);
 810                 mlog_errno(err);
 811                 goto out_commit;
 812         }
 813
 814         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 815         if (err) {
 816                 mlog_errno(err);
 817                 goto out_commit;
 818         }
 819
 820         ihold(inode);
 821         d_instantiate(dentry, inode);
 822
 823 out_commit:
 824         ocfs2_commit_trans(osb, handle);
 825         ocfs2_unblock_signals(&oldset);
 826 out_unlock_inode:
 827         ocfs2_inode_unlock(inode, 1);
 828
 829 out:
 830         ocfs2_double_unlock(old_dir, dir);
 831
 832         brelse(fe_bh);
 833         brelse(parent_fe_bh);
 834         brelse(old_dir_bh);
 835
 836         ocfs2_free_dir_lookup_result(&lookup);
 837
 838         if (err)
 839                 mlog_errno(err);
 840
 841         return err;
 842 }
 843
 844 /*
 845  * Takes and drops an exclusive lock on the given dentry. This will
 846  * force other nodes to drop it.
 847  */
 848 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 849 {
 850         int ret;
 851
 852         ret = ocfs2_dentry_lock(dentry, 1);
 853         if (ret)
 854                 mlog_errno(ret);
 855         else
 856                 ocfs2_dentry_unlock(dentry, 1);
 857
 858         return ret;
 859 }
 860
 861 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 862 {
 863         if (S_ISDIR(inode->i_mode)) {
 864                 if (inode->i_nlink == 2)
 865                         return 1;
 866                 return 0;
 867         }
 868
 869         if (inode->i_nlink == 1)
 870                 return 1;
 871         return 0;
 872 }
 873
 874 static int ocfs2_unlink(struct inode *dir,
 875                         struct dentry *dentry)
 876 {
 877         int status;
 878         int child_locked = 0;
 879         bool is_unlinkable = false;
 880         struct inode *inode = d_inode(dentry);
 881         struct inode *orphan_dir = NULL;
 882         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 883         u64 blkno;
 884         struct ocfs2_dinode *fe = NULL;
 885         struct buffer_head *fe_bh = NULL;
 886         struct buffer_head *parent_node_bh = NULL;
 887         handle_t *handle = NULL;
 888         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 889         struct ocfs2_dir_lookup_result lookup = { NULL, };
 890         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 891
 892         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 893                            dentry->d_name.name,
 894                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 895                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 896
 897         status = dquot_initialize(dir);
 898         if (status) {
 899                 mlog_errno(status);
 900                 return status;
 901         }
 902
 903         BUG_ON(d_inode(dentry->d_parent) != dir);
 904
 905         if (inode == osb->root_inode)
 906                 return -EPERM;
 907
 908         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 909                                          OI_LS_PARENT);
 910         if (status < 0) {
 911                 if (status != -ENOENT)
 912                         mlog_errno(status);
 913                 return status;
 914         }
 915
 916         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 917                                           dentry->d_name.len, &blkno, dir,
 918                                           &lookup);
 919         if (status < 0) {
 920                 if (status != -ENOENT)
 921                         mlog_errno(status);
 922                 goto leave;
 923         }
 924
 925         if (OCFS2_I(inode)->ip_blkno != blkno) {
 926                 status = -ENOENT;
 927
 928                 trace_ocfs2_unlink_noent(
 929                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 930                                 (unsigned long long)blkno,
 931                                 OCFS2_I(inode)->ip_flags);
 932                 goto leave;
 933         }
 934
 935         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 936         if (status < 0) {
 937                 if (status != -ENOENT)
 938                         mlog_errno(status);
 939                 goto leave;
 940         }
 941         child_locked = 1;
 942
 943         if (S_ISDIR(inode->i_mode)) {
 944                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 945                         status = -ENOTEMPTY;
 946                         goto leave;
 947                 }
 948         }
 949
 950         status = ocfs2_remote_dentry_delete(dentry);
 951         if (status < 0) {
 952                 /* This remote delete should succeed under all normal
 953                  * circumstances. */
 954                 mlog_errno(status);
 955                 goto leave;
 956         }
 957
 958         if (ocfs2_inode_is_unlinkable(inode)) {
 959                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 960                                                   OCFS2_I(inode)->ip_blkno,
 961                                                   orphan_name, &orphan_insert,
 962                                                   false);
 963                 if (status < 0) {
 964                         mlog_errno(status);
 965                         goto leave;
 966                 }
 967                 is_unlinkable = true;
 968         }
 969
 970         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 971         if (IS_ERR(handle)) {
 972                 status = PTR_ERR(handle);
 973                 handle = NULL;
 974                 mlog_errno(status);
 975                 goto leave;
 976         }
 977
 978         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 979                                          OCFS2_JOURNAL_ACCESS_WRITE);
 980         if (status < 0) {
 981                 mlog_errno(status);
 982                 goto leave;
 983         }
 984
 985         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 986
 987         /* delete the name from the parent dir */
 988         status = ocfs2_delete_entry(handle, dir, &lookup);
 989         if (status < 0) {
 990                 mlog_errno(status);
 991                 goto leave;
 992         }
 993
 994         if (S_ISDIR(inode->i_mode))
 995                 drop_nlink(inode);
 996         drop_nlink(inode);
 997         ocfs2_set_links_count(fe, inode->i_nlink);
 998         ocfs2_journal_dirty(handle, fe_bh);
 999
1000         dir->i_ctime = dir->i_mtime = current_time(dir);
1001         if (S_ISDIR(inode->i_mode))
1002                 drop_nlink(dir);
1003
1004         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1005         if (status < 0) {
1006                 mlog_errno(status);
1007                 if (S_ISDIR(inode->i_mode))
1008                         inc_nlink(dir);
1009                 goto leave;
1010         }
1011
1012         if (is_unlinkable) {
1013                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1014                                 orphan_name, &orphan_insert, orphan_dir, false);
1015                 if (status < 0)
1016                         mlog_errno(status);
1017         }
1018
1019 leave:
1020         if (handle)
1021                 ocfs2_commit_trans(osb, handle);
1022
1023         if (orphan_dir) {
1024                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1025                 ocfs2_inode_unlock(orphan_dir, 1);
1026                 inode_unlock(orphan_dir);
1027                 iput(orphan_dir);
1028         }
1029
1030         if (child_locked)
1031                 ocfs2_inode_unlock(inode, 1);
1032
1033         ocfs2_inode_unlock(dir, 1);
1034
1035         brelse(fe_bh);
1036         brelse(parent_node_bh);
1037
1038         ocfs2_free_dir_lookup_result(&orphan_insert);
1039         ocfs2_free_dir_lookup_result(&lookup);
1040
1041         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1042                 mlog_errno(status);
1043
1044         return status;
1045 }
1046
1047 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1048                 u64 src_inode_no, u64 dest_inode_no)
1049 {
1050         int ret = 0, i = 0;
1051         u64 parent_inode_no = 0;
1052         u64 child_inode_no = src_inode_no;
1053         struct inode *child_inode;
1054
1055 #define MAX_LOOKUP_TIMES 32
1056         while (1) {
1057                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1058                 if (IS_ERR(child_inode)) {
1059                         ret = PTR_ERR(child_inode);
1060                         break;
1061                 }
1062
1063                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1064                 if (ret < 0) {
1065                         iput(child_inode);
1066                         if (ret != -ENOENT)
1067                                 mlog_errno(ret);
1068                         break;
1069                 }
1070
1071                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1072                                 &parent_inode_no);
1073                 ocfs2_inode_unlock(child_inode, 0);
1074                 iput(child_inode);
1075                 if (ret < 0) {
1076                         ret = -ENOENT;
1077                         break;
1078                 }
1079
1080                 if (parent_inode_no == dest_inode_no) {
1081                         ret = 1;
1082                         break;
1083                 }
1084
1085                 if (parent_inode_no == osb->root_inode->i_ino) {
1086                         ret = 0;
1087                         break;
1088                 }
1089
1090                 child_inode_no = parent_inode_no;
1091
1092                 if (++i >= MAX_LOOKUP_TIMES) {
1093                         mlog_ratelimited(ML_NOTICE, "max lookup times reached, "
1094                                         "filesystem may have nested directories, "
1095                                         "src inode: %llu, dest inode: %llu.\n",
1096                                         (unsigned long long)src_inode_no,
1097                                         (unsigned long long)dest_inode_no);
1098                         ret = 0;
1099                         break;
1100                 }
1101         }
1102
1103         return ret;
1104 }
1105
1106 /*
1107  * The only place this should be used is rename and link!
1108  * if they have the same id, then the 1st one is the only one locked.
1109  */
1110 static int ocfs2_double_lock(struct ocfs2_super *osb,
1111                              struct buffer_head **bh1,
1112                              struct inode *inode1,
1113                              struct buffer_head **bh2,
1114                              struct inode *inode2,
1115                              int rename)
1116 {
1117         int status;
1118         int inode1_is_ancestor, inode2_is_ancestor;
1119         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1120         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1121
1122         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1123                                 (unsigned long long)oi2->ip_blkno);
1124
1125         if (*bh1)
1126                 *bh1 = NULL;
1127         if (*bh2)
1128                 *bh2 = NULL;
1129
1130         /* we always want to lock the one with the lower lockid first.
1131          * and if they are nested, we lock ancestor first */
1132         if (oi1->ip_blkno != oi2->ip_blkno) {
1133                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1134                                 oi1->ip_blkno);
1135                 if (inode1_is_ancestor < 0) {
1136                         status = inode1_is_ancestor;
1137                         goto bail;
1138                 }
1139
1140                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1141                                 oi2->ip_blkno);
1142                 if (inode2_is_ancestor < 0) {
1143                         status = inode2_is_ancestor;
1144                         goto bail;
1145                 }
1146
1147                 if ((inode1_is_ancestor == 1) ||
1148                                 (oi1->ip_blkno < oi2->ip_blkno &&
1149                                 inode2_is_ancestor == 0)) {
1150                         /* switch id1 and id2 around */
1151                         swap(bh2, bh1);
1152                         swap(inode2, inode1);
1153                 }
1154                 /* lock id2 */
1155                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1156                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1157                 if (status < 0) {
1158                         if (status != -ENOENT)
1159                                 mlog_errno(status);
1160                         goto bail;
1161                 }
1162         }
1163
1164         /* lock id1 */
1165         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1166                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1167         if (status < 0) {
1168                 /*
1169                  * An error return must mean that no cluster locks
1170                  * were held on function exit.
1171                  */
1172                 if (oi1->ip_blkno != oi2->ip_blkno) {
1173                         ocfs2_inode_unlock(inode2, 1);
1174                         brelse(*bh2);
1175                         *bh2 = NULL;
1176                 }
1177
1178                 if (status != -ENOENT)
1179                         mlog_errno(status);
1180         }
1181
1182         trace_ocfs2_double_lock_end(
1183                         (unsigned long long)oi1->ip_blkno,
1184                         (unsigned long long)oi2->ip_blkno);
1185
1186 bail:
1187         if (status)
1188                 mlog_errno(status);
1189         return status;
1190 }
1191
1192 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1193 {
1194         ocfs2_inode_unlock(inode1, 1);
1195
1196         if (inode1 != inode2)
1197                 ocfs2_inode_unlock(inode2, 1);
1198 }
1199
1200 static int ocfs2_rename(struct user_namespace *mnt_userns,
1201                         struct inode *old_dir,
1202                         struct dentry *old_dentry,
1203                         struct inode *new_dir,
1204                         struct dentry *new_dentry,
1205                         unsigned int flags)
1206 {
1207         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1208         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1209         struct inode *old_inode = d_inode(old_dentry);
1210         struct inode *new_inode = d_inode(new_dentry);
1211         struct inode *orphan_dir = NULL;
1212         struct ocfs2_dinode *newfe = NULL;
1213         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1214         struct buffer_head *newfe_bh = NULL;
1215         struct buffer_head *old_inode_bh = NULL;
1216         struct ocfs2_super *osb = NULL;
1217         u64 newfe_blkno, old_de_ino;
1218         handle_t *handle = NULL;
1219         struct buffer_head *old_dir_bh = NULL;
1220         struct buffer_head *new_dir_bh = NULL;
1221         u32 old_dir_nlink = old_dir->i_nlink;
1222         struct ocfs2_dinode *old_di;
1223         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1224         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1225         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1226         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1227         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1228         bool should_add_orphan = false;
1229
1230         if (flags)
1231                 return -EINVAL;
1232
1233         /* At some point it might be nice to break this function up a
1234          * bit. */
1235
1236         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1237                            old_dentry->d_name.len, old_dentry->d_name.name,
1238                            new_dentry->d_name.len, new_dentry->d_name.name);
1239
1240         status = dquot_initialize(old_dir);
1241         if (status) {
1242                 mlog_errno(status);
1243                 goto bail;
1244         }
1245         status = dquot_initialize(new_dir);
1246         if (status) {
1247                 mlog_errno(status);
1248                 goto bail;
1249         }
1250
1251         osb = OCFS2_SB(old_dir->i_sb);
1252
1253         if (new_inode) {
1254                 if (!igrab(new_inode))
1255                         BUG();
1256         }
1257
1258         /* Assume a directory hierarchy thusly:
1259          * a/b/c
1260          * a/d
1261          * a,b,c, and d are all directories.
1262          *
1263          * from cwd of 'a' on both nodes:
1264          * node1: mv b/c d
1265          * node2: mv d   b/c
1266          *
1267          * And that's why, just like the VFS, we need a file system
1268          * rename lock. */
1269         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1270                 status = ocfs2_rename_lock(osb);
1271                 if (status < 0) {
1272                         mlog_errno(status);
1273                         goto bail;
1274                 }
1275                 rename_lock = 1;
1276
1277                 /* here we cannot guarantee the inodes haven't just been
1278                  * changed, so check if they are nested again */
1279                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1280                                 old_inode->i_ino);
1281                 if (status < 0) {
1282                         mlog_errno(status);
1283                         goto bail;
1284                 } else if (status == 1) {
1285                         status = -EPERM;
1286                         trace_ocfs2_rename_not_permitted(
1287                                         (unsigned long long)old_inode->i_ino,
1288                                         (unsigned long long)new_dir->i_ino);
1289                         goto bail;
1290                 }
1291         }
1292
1293         /* if old and new are the same, this'll just do one lock. */
1294         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1295                                    &new_dir_bh, new_dir, 1);
1296         if (status < 0) {
1297                 mlog_errno(status);
1298                 goto bail;
1299         }
1300         parents_locked = 1;
1301
1302         if (!new_dir->i_nlink) {
1303                 status = -EACCES;
1304                 goto bail;
1305         }
1306
1307         /* make sure both dirs have bhs
1308          * get an extra ref on old_dir_bh if old==new */
1309         if (!new_dir_bh) {
1310                 if (old_dir_bh) {
1311                         new_dir_bh = old_dir_bh;
1312                         get_bh(new_dir_bh);
1313                 } else {
1314                         mlog(ML_ERROR, "no old_dir_bh!\n");
1315                         status = -EIO;
1316                         goto bail;
1317                 }
1318         }
1319
1320         /*
1321          * Aside from allowing a meta data update, the locking here
1322          * also ensures that the downconvert thread on other nodes
1323          * won't have to concurrently downconvert the inode and the
1324          * dentry locks.
1325          */
1326         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1327                                          OI_LS_PARENT);
1328         if (status < 0) {
1329                 if (status != -ENOENT)
1330                         mlog_errno(status);
1331                 goto bail;
1332         }
1333         old_child_locked = 1;
1334
1335         status = ocfs2_remote_dentry_delete(old_dentry);
1336         if (status < 0) {
1337                 mlog_errno(status);
1338                 goto bail;
1339         }
1340
1341         if (S_ISDIR(old_inode->i_mode)) {
1342                 u64 old_inode_parent;
1343
1344                 update_dot_dot = 1;
1345                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1346                                                   old_inode,
1347                                                   &old_inode_dot_dot_res);
1348                 if (status) {
1349                         status = -EIO;
1350                         goto bail;
1351                 }
1352
1353                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1354                         status = -EIO;
1355                         goto bail;
1356                 }
1357
1358                 if (!new_inode && new_dir != old_dir &&
1359                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1360                         status = -EMLINK;
1361                         goto bail;
1362                 }
1363         }
1364
1365         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1366                                             old_dentry->d_name.len,
1367                                             &old_de_ino);
1368         if (status) {
1369                 status = -ENOENT;
1370                 goto bail;
1371         }
1372
1373         /*
1374          *  Check for inode number is _not_ due to possible IO errors.
1375          *  We might rmdir the source, keep it as pwd of some process
1376          *  and merrily kill the link to whatever was created under the
1377          *  same name. Goodbye sticky bit ;-<
1378          */
1379         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1380                 status = -ENOENT;
1381                 goto bail;
1382         }
1383
1384         /* check if the target already exists (in which case we need
1385          * to delete it */
1386         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1387                                           new_dentry->d_name.len,
1388                                           &newfe_blkno, new_dir,
1389                                           &target_lookup_res);
1390         /* The only error we allow here is -ENOENT because the new
1391          * file not existing is perfectly valid. */
1392         if ((status < 0) && (status != -ENOENT)) {
1393                 /* If we cannot find the file specified we should just */
1394                 /* return the error... */
1395                 mlog_errno(status);
1396                 goto bail;
1397         }
1398         if (status == 0)
1399                 target_exists = 1;
1400
1401         if (!target_exists && new_inode) {
1402                 /*
1403                  * Target was unlinked by another node while we were
1404                  * waiting to get to ocfs2_rename(). There isn't
1405                  * anything we can do here to help the situation, so
1406                  * bubble up the appropriate error.
1407                  */
1408                 status = -ENOENT;
1409                 goto bail;
1410         }
1411
1412         /* In case we need to overwrite an existing file, we blow it
1413          * away first */
1414         if (target_exists) {
1415                 /* VFS didn't think there existed an inode here, but
1416                  * someone else in the cluster must have raced our
1417                  * rename to create one. Today we error cleanly, in
1418                  * the future we should consider calling iget to build
1419                  * a new struct inode for this entry. */
1420                 if (!new_inode) {
1421                         status = -EACCES;
1422
1423                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1424                                                 new_dentry->d_name.name);
1425                         goto bail;
1426                 }
1427
1428                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1429                         status = -EACCES;
1430
1431                         trace_ocfs2_rename_disagree(
1432                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1433                              (unsigned long long)newfe_blkno,
1434                              OCFS2_I(new_inode)->ip_flags);
1435                         goto bail;
1436                 }
1437
1438                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1439                 if (status < 0) {
1440                         if (status != -ENOENT)
1441                                 mlog_errno(status);
1442                         goto bail;
1443                 }
1444                 new_child_locked = 1;
1445
1446                 status = ocfs2_remote_dentry_delete(new_dentry);
1447                 if (status < 0) {
1448                         mlog_errno(status);
1449                         goto bail;
1450                 }
1451
1452                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1453
1454                 trace_ocfs2_rename_over_existing(
1455                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1456                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1457
1458                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1459                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1460                                                 OCFS2_I(new_inode)->ip_blkno,
1461                                                 orphan_name, &orphan_insert,
1462                                                 false);
1463                         if (status < 0) {
1464                                 mlog_errno(status);
1465                                 goto bail;
1466                         }
1467                         should_add_orphan = true;
1468                 }
1469         } else {
1470                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1471
1472                 status = ocfs2_check_dir_for_entry(new_dir,
1473                                                    new_dentry->d_name.name,
1474                                                    new_dentry->d_name.len);
1475                 if (status)
1476                         goto bail;
1477
1478                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1479                                                       new_dentry->d_name.name,
1480                                                       new_dentry->d_name.len,
1481                                                       &target_insert);
1482                 if (status < 0) {
1483                         mlog_errno(status);
1484                         goto bail;
1485                 }
1486         }
1487
1488         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1489         if (IS_ERR(handle)) {
1490                 status = PTR_ERR(handle);
1491                 handle = NULL;
1492                 mlog_errno(status);
1493                 goto bail;
1494         }
1495
1496         if (target_exists) {
1497                 if (S_ISDIR(new_inode->i_mode)) {
1498                         if (new_inode->i_nlink != 2 ||
1499                             !ocfs2_empty_dir(new_inode)) {
1500                                 status = -ENOTEMPTY;
1501                                 goto bail;
1502                         }
1503                 }
1504                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1505                                                  newfe_bh,
1506                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1507                 if (status < 0) {
1508                         mlog_errno(status);
1509                         goto bail;
1510                 }
1511
1512                 /* change the dirent to point to the correct inode */
1513                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1514                                             old_inode);
1515                 if (status < 0) {
1516                         mlog_errno(status);
1517                         goto bail;
1518                 }
1519                 inode_inc_iversion(new_dir);
1520
1521                 if (S_ISDIR(new_inode->i_mode))
1522                         ocfs2_set_links_count(newfe, 0);
1523                 else
1524                         ocfs2_add_links_count(newfe, -1);
1525                 ocfs2_journal_dirty(handle, newfe_bh);
1526                 if (should_add_orphan) {
1527                         status = ocfs2_orphan_add(osb, handle, new_inode,
1528                                         newfe_bh, orphan_name,
1529                                         &orphan_insert, orphan_dir, false);
1530                         if (status < 0) {
1531                                 mlog_errno(status);
1532                                 goto bail;
1533                         }
1534                 }
1535         } else {
1536                 /* if the name was not found in new_dir, add it now */
1537                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1538                                          OCFS2_I(old_inode)->ip_blkno,
1539                                          new_dir_bh, &target_insert);
1540         }
1541
1542         old_inode->i_ctime = current_time(old_inode);
1543         mark_inode_dirty(old_inode);
1544
1545         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1546                                          old_inode_bh,
1547                                          OCFS2_JOURNAL_ACCESS_WRITE);
1548         if (status >= 0) {
1549                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1550
1551                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1552                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1553                 ocfs2_journal_dirty(handle, old_inode_bh);
1554         } else
1555                 mlog_errno(status);
1556
1557         /*
1558          * Now that the name has been added to new_dir, remove the old name.
1559          *
1560          * We don't keep any directory entry context around until now
1561          * because the insert might have changed the type of directory
1562          * we're dealing with.
1563          */
1564         status = ocfs2_find_entry(old_dentry->d_name.name,
1565                                   old_dentry->d_name.len, old_dir,
1566                                   &old_entry_lookup);
1567         if (status) {
1568                 if (!is_journal_aborted(osb->journal->j_journal)) {
1569                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1570                                         "is not deleted.",
1571                                         new_dentry->d_name.len, new_dentry->d_name.name,
1572                                         old_dentry->d_name.len, old_dentry->d_name.name);
1573                 }
1574                 goto bail;
1575         }
1576
1577         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1578         if (status < 0) {
1579                 mlog_errno(status);
1580                 if (!is_journal_aborted(osb->journal->j_journal)) {
1581                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1582                                         "is not deleted.",
1583                                         new_dentry->d_name.len, new_dentry->d_name.name,
1584                                         old_dentry->d_name.len, old_dentry->d_name.name);
1585                 }
1586                 goto bail;
1587         }
1588
1589         if (new_inode) {
1590                 drop_nlink(new_inode);
1591                 new_inode->i_ctime = current_time(new_inode);
1592         }
1593         old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1594
1595         if (update_dot_dot) {
1596                 status = ocfs2_update_entry(old_inode, handle,
1597                                             &old_inode_dot_dot_res, new_dir);
1598                 drop_nlink(old_dir);
1599                 if (new_inode) {
1600                         drop_nlink(new_inode);
1601                 } else {
1602                         inc_nlink(new_dir);
1603                         mark_inode_dirty(new_dir);
1604                 }
1605         }
1606         mark_inode_dirty(old_dir);
1607         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1608         if (new_inode) {
1609                 mark_inode_dirty(new_inode);
1610                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1611         }
1612
1613         if (old_dir != new_dir) {
1614                 /* Keep the same times on both directories.*/
1615                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1616
1617                 /*
1618                  * This will also pick up the i_nlink change from the
1619                  * block above.
1620                  */
1621                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1622         }
1623
1624         if (old_dir_nlink != old_dir->i_nlink) {
1625                 if (!old_dir_bh) {
1626                         mlog(ML_ERROR, "need to change nlink for old dir "
1627                              "%llu from %d to %d but bh is NULL!\n",
1628                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1629                              (int)old_dir_nlink, old_dir->i_nlink);
1630                 } else {
1631                         struct ocfs2_dinode *fe;
1632                         status = ocfs2_journal_access_di(handle,
1633                                                          INODE_CACHE(old_dir),
1634                                                          old_dir_bh,
1635                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1636                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1637                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1638                         ocfs2_journal_dirty(handle, old_dir_bh);
1639                 }
1640         }
1641         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1642         status = 0;
1643 bail:
1644         if (handle)
1645                 ocfs2_commit_trans(osb, handle);
1646
1647         if (orphan_dir) {
1648                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1649                 ocfs2_inode_unlock(orphan_dir, 1);
1650                 inode_unlock(orphan_dir);
1651                 iput(orphan_dir);
1652         }
1653
1654         if (new_child_locked)
1655                 ocfs2_inode_unlock(new_inode, 1);
1656
1657         if (old_child_locked)
1658                 ocfs2_inode_unlock(old_inode, 1);
1659
1660         if (parents_locked)
1661                 ocfs2_double_unlock(old_dir, new_dir);
1662
1663         if (rename_lock)
1664                 ocfs2_rename_unlock(osb);
1665
1666         if (new_inode)
1667                 sync_mapping_buffers(old_inode->i_mapping);
1668
1669         iput(new_inode);
1670
1671         ocfs2_free_dir_lookup_result(&target_lookup_res);
1672         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1673         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1674         ocfs2_free_dir_lookup_result(&orphan_insert);
1675         ocfs2_free_dir_lookup_result(&target_insert);
1676
1677         brelse(newfe_bh);
1678         brelse(old_inode_bh);
1679         brelse(old_dir_bh);
1680         brelse(new_dir_bh);
1681
1682         if (status)
1683                 mlog_errno(status);
1684
1685         return status;
1686 }
1687
1688 /*
1689  * we expect i_size = strlen(symname). Copy symname into the file
1690  * data, including the null terminator.
1691  */
1692 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1693                                      handle_t *handle,
1694                                      struct inode *inode,
1695                                      const char *symname)
1696 {
1697         struct buffer_head **bhs = NULL;
1698         const char *c;
1699         struct super_block *sb = osb->sb;
1700         u64 p_blkno, p_blocks;
1701         int virtual, blocks, status, i, bytes_left;
1702
1703         bytes_left = i_size_read(inode) + 1;
1704         /* we can't trust i_blocks because we're actually going to
1705          * write i_size + 1 bytes. */
1706         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1707
1708         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1709                                         i_size_read(inode), blocks);
1710
1711         /* Sanity check -- make sure we're going to fit. */
1712         if (bytes_left >
1713             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1714                 status = -EIO;
1715                 mlog_errno(status);
1716                 goto bail;
1717         }
1718
1719         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1720         if (!bhs) {
1721                 status = -ENOMEM;
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1727                                              NULL);
1728         if (status < 0) {
1729                 mlog_errno(status);
1730                 goto bail;
1731         }
1732
1733         /* links can never be larger than one cluster so we know this
1734          * is all going to be contiguous, but do a sanity check
1735          * anyway. */
1736         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1737                 status = -EIO;
1738                 mlog_errno(status);
1739                 goto bail;
1740         }
1741
1742         virtual = 0;
1743         while(bytes_left > 0) {
1744                 c = &symname[virtual * sb->s_blocksize];
1745
1746                 bhs[virtual] = sb_getblk(sb, p_blkno);
1747                 if (!bhs[virtual]) {
1748                         status = -ENOMEM;
1749                         mlog_errno(status);
1750                         goto bail;
1751                 }
1752                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1753                                               bhs[virtual]);
1754
1755                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1756                                               bhs[virtual],
1757                                               OCFS2_JOURNAL_ACCESS_CREATE);
1758                 if (status < 0) {
1759                         mlog_errno(status);
1760                         goto bail;
1761                 }
1762
1763                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1764
1765                 memcpy(bhs[virtual]->b_data, c,
1766                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1767                        bytes_left);
1768
1769                 ocfs2_journal_dirty(handle, bhs[virtual]);
1770
1771                 virtual++;
1772                 p_blkno++;
1773                 bytes_left -= sb->s_blocksize;
1774         }
1775
1776         status = 0;
1777 bail:
1778
1779         if (bhs) {
1780                 for(i = 0; i < blocks; i++)
1781                         brelse(bhs[i]);
1782                 kfree(bhs);
1783         }
1784
1785         if (status)
1786                 mlog_errno(status);
1787         return status;
1788 }
1789
1790 static int ocfs2_symlink(struct user_namespace *mnt_userns,
1791                          struct inode *dir,
1792                          struct dentry *dentry,
1793                          const char *symname)
1794 {
1795         int status, l, credits;
1796         u64 newsize;
1797         struct ocfs2_super *osb = NULL;
1798         struct inode *inode = NULL;
1799         struct super_block *sb;
1800         struct buffer_head *new_fe_bh = NULL;
1801         struct buffer_head *parent_fe_bh = NULL;
1802         struct ocfs2_dinode *fe = NULL;
1803         struct ocfs2_dinode *dirfe;
1804         handle_t *handle = NULL;
1805         struct ocfs2_alloc_context *inode_ac = NULL;
1806         struct ocfs2_alloc_context *data_ac = NULL;
1807         struct ocfs2_alloc_context *xattr_ac = NULL;
1808         int want_clusters = 0;
1809         int xattr_credits = 0;
1810         struct ocfs2_security_xattr_info si = {
1811                 .enable = 1,
1812         };
1813         int did_quota = 0, did_quota_inode = 0;
1814         struct ocfs2_dir_lookup_result lookup = { NULL, };
1815         sigset_t oldset;
1816         int did_block_signals = 0;
1817         struct ocfs2_dentry_lock *dl = NULL;
1818
1819         trace_ocfs2_symlink_begin(dir, dentry, symname,
1820                                   dentry->d_name.len, dentry->d_name.name);
1821
1822         status = dquot_initialize(dir);
1823         if (status) {
1824                 mlog_errno(status);
1825                 goto bail;
1826         }
1827
1828         sb = dir->i_sb;
1829         osb = OCFS2_SB(sb);
1830
1831         l = strlen(symname) + 1;
1832
1833         credits = ocfs2_calc_symlink_credits(sb);
1834
1835         /* lock the parent directory */
1836         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1837         if (status < 0) {
1838                 if (status != -ENOENT)
1839                         mlog_errno(status);
1840                 return status;
1841         }
1842
1843         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1844         if (!ocfs2_read_links_count(dirfe)) {
1845                 /* can't make a file in a deleted directory. */
1846                 status = -ENOENT;
1847                 goto bail;
1848         }
1849
1850         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1851                                            dentry->d_name.len);
1852         if (status)
1853                 goto bail;
1854
1855         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1856                                               dentry->d_name.name,
1857                                               dentry->d_name.len, &lookup);
1858         if (status < 0) {
1859                 mlog_errno(status);
1860                 goto bail;
1861         }
1862
1863         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1864         if (status < 0) {
1865                 if (status != -ENOSPC)
1866                         mlog_errno(status);
1867                 goto bail;
1868         }
1869
1870         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1871         if (IS_ERR(inode)) {
1872                 status = PTR_ERR(inode);
1873                 inode = NULL;
1874                 mlog_errno(status);
1875                 goto bail;
1876         }
1877
1878         /* get security xattr */
1879         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1880         if (status) {
1881                 if (status == -EOPNOTSUPP)
1882                         si.enable = 0;
1883                 else {
1884                         mlog_errno(status);
1885                         goto bail;
1886                 }
1887         }
1888
1889         /* calculate meta data/clusters for setting security xattr */
1890         if (si.enable) {
1891                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1892                                                   &xattr_credits, &xattr_ac);
1893                 if (status < 0) {
1894                         mlog_errno(status);
1895                         goto bail;
1896                 }
1897         }
1898
1899         /* don't reserve bitmap space for fast symlinks. */
1900         if (l > ocfs2_fast_symlink_chars(sb))
1901                 want_clusters += 1;
1902
1903         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1904         if (status < 0) {
1905                 if (status != -ENOSPC)
1906                         mlog_errno(status);
1907                 goto bail;
1908         }
1909
1910         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1911         if (IS_ERR(handle)) {
1912                 status = PTR_ERR(handle);
1913                 handle = NULL;
1914                 mlog_errno(status);
1915                 goto bail;
1916         }
1917
1918         /* Starting to change things, restart is no longer possible. */
1919         ocfs2_block_signals(&oldset);
1920         did_block_signals = 1;
1921
1922         status = dquot_alloc_inode(inode);
1923         if (status)
1924                 goto bail;
1925         did_quota_inode = 1;
1926
1927         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1928                                    dentry->d_name.name,
1929                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1930                                    inode->i_mode);
1931
1932         status = ocfs2_mknod_locked(osb, dir, inode,
1933                                     0, &new_fe_bh, parent_fe_bh, handle,
1934                                     inode_ac);
1935         if (status < 0) {
1936                 mlog_errno(status);
1937                 goto bail;
1938         }
1939
1940         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1941         inode->i_rdev = 0;
1942         newsize = l - 1;
1943         inode->i_op = &ocfs2_symlink_inode_operations;
1944         inode_nohighmem(inode);
1945         if (l > ocfs2_fast_symlink_chars(sb)) {
1946                 u32 offset = 0;
1947
1948                 status = dquot_alloc_space_nodirty(inode,
1949                     ocfs2_clusters_to_bytes(osb->sb, 1));
1950                 if (status)
1951                         goto bail;
1952                 did_quota = 1;
1953                 inode->i_mapping->a_ops = &ocfs2_aops;
1954                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1955                                               new_fe_bh,
1956                                               handle, data_ac, NULL,
1957                                               NULL);
1958                 if (status < 0) {
1959                         if (status != -ENOSPC && status != -EINTR) {
1960                                 mlog(ML_ERROR,
1961                                      "Failed to extend file to %llu\n",
1962                                      (unsigned long long)newsize);
1963                                 mlog_errno(status);
1964                                 status = -ENOSPC;
1965                         }
1966                         goto bail;
1967                 }
1968                 i_size_write(inode, newsize);
1969                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1970         } else {
1971                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1972                 memcpy((char *) fe->id2.i_symlink, symname, l);
1973                 i_size_write(inode, newsize);
1974                 inode->i_blocks = 0;
1975         }
1976
1977         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1978         if (status < 0) {
1979                 mlog_errno(status);
1980                 goto bail;
1981         }
1982
1983         if (!ocfs2_inode_is_fast_symlink(inode)) {
1984                 status = ocfs2_create_symlink_data(osb, handle, inode,
1985                                                    symname);
1986                 if (status < 0) {
1987                         mlog_errno(status);
1988                         goto bail;
1989                 }
1990         }
1991
1992         if (si.enable) {
1993                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1994                                                  xattr_ac, data_ac);
1995                 if (status < 0) {
1996                         mlog_errno(status);
1997                         goto bail;
1998                 }
1999         }
2000
2001         /*
2002          * Do this before adding the entry to the directory. We add
2003          * also set d_op after success so that ->d_iput() will cleanup
2004          * the dentry lock even if ocfs2_add_entry() fails below.
2005          */
2006         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2007         if (status) {
2008                 mlog_errno(status);
2009                 goto bail;
2010         }
2011
2012         dl = dentry->d_fsdata;
2013
2014         status = ocfs2_add_entry(handle, dentry, inode,
2015                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2016                                  &lookup);
2017         if (status < 0) {
2018                 mlog_errno(status);
2019                 goto bail;
2020         }
2021
2022         insert_inode_hash(inode);
2023         d_instantiate(dentry, inode);
2024 bail:
2025         if (status < 0 && did_quota)
2026                 dquot_free_space_nodirty(inode,
2027                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2028         if (status < 0 && did_quota_inode)
2029                 dquot_free_inode(inode);
2030         if (handle)
2031                 ocfs2_commit_trans(osb, handle);
2032
2033         ocfs2_inode_unlock(dir, 1);
2034         if (did_block_signals)
2035                 ocfs2_unblock_signals(&oldset);
2036
2037         brelse(new_fe_bh);
2038         brelse(parent_fe_bh);
2039         kfree(si.value);
2040         ocfs2_free_dir_lookup_result(&lookup);
2041         if (inode_ac)
2042                 ocfs2_free_alloc_context(inode_ac);
2043         if (data_ac)
2044                 ocfs2_free_alloc_context(data_ac);
2045         if (xattr_ac)
2046                 ocfs2_free_alloc_context(xattr_ac);
2047         if ((status < 0) && inode) {
2048                 if (dl)
2049                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2050
2051                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2052                 clear_nlink(inode);
2053                 iput(inode);
2054         }
2055
2056         if (status)
2057                 mlog_errno(status);
2058
2059         return status;
2060 }
2061
2062 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2063 {
2064         int status, namelen;
2065
2066         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2067                            (long long)blkno);
2068         if (namelen <= 0) {
2069                 if (namelen)
2070                         status = namelen;
2071                 else
2072                         status = -EINVAL;
2073                 mlog_errno(status);
2074                 goto bail;
2075         }
2076         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2077                 status = -EINVAL;
2078                 mlog_errno(status);
2079                 goto bail;
2080         }
2081
2082         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2083
2084         status = 0;
2085 bail:
2086         if (status < 0)
2087                 mlog_errno(status);
2088         return status;
2089 }
2090
2091 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2092                                         struct inode **ret_orphan_dir,
2093                                         struct buffer_head **ret_orphan_dir_bh)
2094 {
2095         struct inode *orphan_dir_inode;
2096         struct buffer_head *orphan_dir_bh = NULL;
2097         int ret = 0;
2098
2099         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2100                                                        ORPHAN_DIR_SYSTEM_INODE,
2101                                                        osb->slot_num);
2102         if (!orphan_dir_inode) {
2103                 ret = -ENOENT;
2104                 mlog_errno(ret);
2105                 return ret;
2106         }
2107
2108         inode_lock(orphan_dir_inode);
2109
2110         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2111         if (ret < 0) {
2112                 inode_unlock(orphan_dir_inode);
2113                 iput(orphan_dir_inode);
2114
2115                 mlog_errno(ret);
2116                 return ret;
2117         }
2118
2119         *ret_orphan_dir = orphan_dir_inode;
2120         *ret_orphan_dir_bh = orphan_dir_bh;
2121
2122         return 0;
2123 }
2124
2125 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2126                                       struct buffer_head *orphan_dir_bh,
2127                                       u64 blkno,
2128                                       char *name,
2129                                       struct ocfs2_dir_lookup_result *lookup,
2130                                       bool dio)
2131 {
2132         int ret;
2133         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2134         int namelen = dio ?
2135                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2136                         OCFS2_ORPHAN_NAMELEN;
2137
2138         if (dio) {
2139                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2140                                 OCFS2_DIO_ORPHAN_PREFIX);
2141                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2142                         ret = -EINVAL;
2143                         mlog_errno(ret);
2144                         return ret;
2145                 }
2146
2147                 ret = ocfs2_blkno_stringify(blkno,
2148                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2149         } else
2150                 ret = ocfs2_blkno_stringify(blkno, name);
2151         if (ret < 0) {
2152                 mlog_errno(ret);
2153                 return ret;
2154         }
2155
2156         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2157                                            orphan_dir_bh, name,
2158                                            namelen, lookup);
2159         if (ret < 0) {
2160                 mlog_errno(ret);
2161                 return ret;
2162         }
2163
2164         return 0;
2165 }
2166
2167 /**
2168  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2169  * insertion of an orphan.
2170  * @osb: ocfs2 file system
2171  * @ret_orphan_dir: Orphan dir inode - returned locked!
2172  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2173  * @lookup: dir lookup result, to be passed back into functions like
2174  *          ocfs2_orphan_add
2175  *
2176  * Returns zero on success and the ret_orphan_dir, name and lookup
2177  * fields will be populated.
2178  *
2179  * Returns non-zero on failure.
2180  */
2181 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2182                                     struct inode **ret_orphan_dir,
2183                                     u64 blkno,
2184                                     char *name,
2185                                     struct ocfs2_dir_lookup_result *lookup,
2186                                     bool dio)
2187 {
2188         struct inode *orphan_dir_inode = NULL;
2189         struct buffer_head *orphan_dir_bh = NULL;
2190         int ret = 0;
2191
2192         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2193                                            &orphan_dir_bh);
2194         if (ret < 0) {
2195                 mlog_errno(ret);
2196                 return ret;
2197         }
2198
2199         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2200                                          blkno, name, lookup, dio);
2201         if (ret < 0) {
2202                 mlog_errno(ret);
2203                 goto out;
2204         }
2205
2206         *ret_orphan_dir = orphan_dir_inode;
2207
2208 out:
2209         brelse(orphan_dir_bh);
2210
2211         if (ret) {
2212                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2213                 inode_unlock(orphan_dir_inode);
2214                 iput(orphan_dir_inode);
2215         }
2216
2217         if (ret)
2218                 mlog_errno(ret);
2219         return ret;
2220 }
2221
2222 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2223                             handle_t *handle,
2224                             struct inode *inode,
2225                             struct buffer_head *fe_bh,
2226                             char *name,
2227                             struct ocfs2_dir_lookup_result *lookup,
2228                             struct inode *orphan_dir_inode,
2229                             bool dio)
2230 {
2231         struct buffer_head *orphan_dir_bh = NULL;
2232         int status = 0;
2233         struct ocfs2_dinode *orphan_fe;
2234         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2235         int namelen = dio ?
2236                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2237                         OCFS2_ORPHAN_NAMELEN;
2238
2239         trace_ocfs2_orphan_add_begin(
2240                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2241
2242         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2243         if (status < 0) {
2244                 mlog_errno(status);
2245                 goto leave;
2246         }
2247
2248         status = ocfs2_journal_access_di(handle,
2249                                          INODE_CACHE(orphan_dir_inode),
2250                                          orphan_dir_bh,
2251                                          OCFS2_JOURNAL_ACCESS_WRITE);
2252         if (status < 0) {
2253                 mlog_errno(status);
2254                 goto leave;
2255         }
2256
2257         /*
2258          * We're going to journal the change of i_flags and i_orphaned_slot.
2259          * It's safe anyway, though some callers may duplicate the journaling.
2260          * Journaling within the func just make the logic look more
2261          * straightforward.
2262          */
2263         status = ocfs2_journal_access_di(handle,
2264                                          INODE_CACHE(inode),
2265                                          fe_bh,
2266                                          OCFS2_JOURNAL_ACCESS_WRITE);
2267         if (status < 0) {
2268                 mlog_errno(status);
2269                 goto leave;
2270         }
2271
2272         /* we're a cluster, and nlink can change on disk from
2273          * underneath us... */
2274         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2275         if (S_ISDIR(inode->i_mode))
2276                 ocfs2_add_links_count(orphan_fe, 1);
2277         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2278         ocfs2_journal_dirty(handle, orphan_dir_bh);
2279
2280         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2281                                    namelen, inode,
2282                                    OCFS2_I(inode)->ip_blkno,
2283                                    orphan_dir_bh, lookup);
2284         if (status < 0) {
2285                 mlog_errno(status);
2286                 goto rollback;
2287         }
2288
2289         if (dio) {
2290                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2291                  * slot.
2292                  */
2293                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2294                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2295         } else {
2296                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2297                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2298
2299                 /* Record which orphan dir our inode now resides
2300                  * in. delete_inode will use this to determine which orphan
2301                  * dir to lock. */
2302                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2303         }
2304
2305         ocfs2_journal_dirty(handle, fe_bh);
2306
2307         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2308                                    osb->slot_num);
2309
2310 rollback:
2311         if (status < 0) {
2312                 if (S_ISDIR(inode->i_mode))
2313                         ocfs2_add_links_count(orphan_fe, -1);
2314                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2315         }
2316
2317 leave:
2318         brelse(orphan_dir_bh);
2319
2320         return status;
2321 }
2322
2323 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2324 int ocfs2_orphan_del(struct ocfs2_super *osb,
2325                      handle_t *handle,
2326                      struct inode *orphan_dir_inode,
2327                      struct inode *inode,
2328                      struct buffer_head *orphan_dir_bh,
2329                      bool dio)
2330 {
2331         char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2332         struct ocfs2_dinode *orphan_fe;
2333         int status = 0;
2334         struct ocfs2_dir_lookup_result lookup = { NULL, };
2335
2336         if (dio) {
2337                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2338                                 OCFS2_DIO_ORPHAN_PREFIX);
2339                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2340                         status = -EINVAL;
2341                         mlog_errno(status);
2342                         return status;
2343                 }
2344
2345                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2346                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2347         } else
2348                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2349         if (status < 0) {
2350                 mlog_errno(status);
2351                 goto leave;
2352         }
2353
2354         trace_ocfs2_orphan_del(
2355              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2356              name, strlen(name));
2357
2358         status = ocfs2_journal_access_di(handle,
2359                                          INODE_CACHE(orphan_dir_inode),
2360                                          orphan_dir_bh,
2361                                          OCFS2_JOURNAL_ACCESS_WRITE);
2362         if (status < 0) {
2363                 mlog_errno(status);
2364                 goto leave;
2365         }
2366
2367         /* find it's spot in the orphan directory */
2368         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2369                                   &lookup);
2370         if (status) {
2371                 mlog_errno(status);
2372                 goto leave;
2373         }
2374
2375         /* remove it from the orphan directory */
2376         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2377         if (status < 0) {
2378                 mlog_errno(status);
2379                 goto leave;
2380         }
2381
2382         /* do the i_nlink dance! :) */
2383         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2384         if (S_ISDIR(inode->i_mode))
2385                 ocfs2_add_links_count(orphan_fe, -1);
2386         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2387         ocfs2_journal_dirty(handle, orphan_dir_bh);
2388
2389 leave:
2390         ocfs2_free_dir_lookup_result(&lookup);
2391
2392         if (status)
2393                 mlog_errno(status);
2394         return status;
2395 }
2396
2397 /**
2398  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2399  * allocated file. This is different from the typical 'add to orphan dir'
2400  * operation in that the inode does not yet exist. This is a problem because
2401  * the orphan dir stringifies the inode block number to come up with it's
2402  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2403  * problem. This function works around it by calling deeper into the orphan
2404  * and suballoc code than other callers. Use this only by necessity.
2405  * @dir: The directory which this inode will ultimately wind up under - not the
2406  * orphan dir!
2407  * @dir_bh: buffer_head the @dir inode block
2408  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2409  * with the string to be used for orphan dirent. Pass back to the orphan dir
2410  * code.
2411  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2412  * dir code.
2413  * @ret_di_blkno: block number where the new inode will be allocated.
2414  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2415  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2416  *
2417  * Returns zero on success and the ret_orphan_dir, name and lookup
2418  * fields will be populated.
2419  *
2420  * Returns non-zero on failure.
2421  */
2422 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2423                                         struct buffer_head *dir_bh,
2424                                         char *orphan_name,
2425                                         struct inode **ret_orphan_dir,
2426                                         u64 *ret_di_blkno,
2427                                         struct ocfs2_dir_lookup_result *orphan_insert,
2428                                         struct ocfs2_alloc_context **ret_inode_ac)
2429 {
2430         int ret;
2431         u64 di_blkno;
2432         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2433         struct inode *orphan_dir = NULL;
2434         struct buffer_head *orphan_dir_bh = NULL;
2435         struct ocfs2_alloc_context *inode_ac = NULL;
2436
2437         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2438         if (ret < 0) {
2439                 mlog_errno(ret);
2440                 return ret;
2441         }
2442
2443         /* reserve an inode spot */
2444         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2445         if (ret < 0) {
2446                 if (ret != -ENOSPC)
2447                         mlog_errno(ret);
2448                 goto out;
2449         }
2450
2451         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2452                                        &di_blkno);
2453         if (ret) {
2454                 mlog_errno(ret);
2455                 goto out;
2456         }
2457
2458         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2459                                          di_blkno, orphan_name, orphan_insert,
2460                                          false);
2461         if (ret < 0) {
2462                 mlog_errno(ret);
2463                 goto out;
2464         }
2465
2466 out:
2467         if (ret == 0) {
2468                 *ret_orphan_dir = orphan_dir;
2469                 *ret_di_blkno = di_blkno;
2470                 *ret_inode_ac = inode_ac;
2471                 /*
2472                  * orphan_name and orphan_insert are already up to
2473                  * date via prepare_orphan_dir
2474                  */
2475         } else {
2476                 /* Unroll reserve_new_inode* */
2477                 if (inode_ac)
2478                         ocfs2_free_alloc_context(inode_ac);
2479
2480                 /* Unroll orphan dir locking */
2481                 inode_unlock(orphan_dir);
2482                 ocfs2_inode_unlock(orphan_dir, 1);
2483                 iput(orphan_dir);
2484         }
2485
2486         brelse(orphan_dir_bh);
2487
2488         return ret;
2489 }
2490
2491 int ocfs2_create_inode_in_orphan(struct inode *dir,
2492                                  int mode,
2493                                  struct inode **new_inode)
2494 {
2495         int status, did_quota_inode = 0;
2496         struct inode *inode = NULL;
2497         struct inode *orphan_dir = NULL;
2498         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2499         handle_t *handle = NULL;
2500         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2501         struct buffer_head *parent_di_bh = NULL;
2502         struct buffer_head *new_di_bh = NULL;
2503         struct ocfs2_alloc_context *inode_ac = NULL;
2504         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2505         u64 di_blkno, suballoc_loc;
2506         u16 suballoc_bit;
2507
2508         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2509         if (status < 0) {
2510                 if (status != -ENOENT)
2511                         mlog_errno(status);
2512                 return status;
2513         }
2514
2515         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2516                                               orphan_name, &orphan_dir,
2517                                               &di_blkno, &orphan_insert, &inode_ac);
2518         if (status < 0) {
2519                 if (status != -ENOSPC)
2520                         mlog_errno(status);
2521                 goto leave;
2522         }
2523
2524         inode = ocfs2_get_init_inode(dir, mode);
2525         if (IS_ERR(inode)) {
2526                 status = PTR_ERR(inode);
2527                 inode = NULL;
2528                 mlog_errno(status);
2529                 goto leave;
2530         }
2531
2532         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2533         if (IS_ERR(handle)) {
2534                 status = PTR_ERR(handle);
2535                 handle = NULL;
2536                 mlog_errno(status);
2537                 goto leave;
2538         }
2539
2540         status = dquot_alloc_inode(inode);
2541         if (status)
2542                 goto leave;
2543         did_quota_inode = 1;
2544
2545         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2546                                               &suballoc_loc,
2547                                               &suballoc_bit, di_blkno);
2548         if (status < 0) {
2549                 mlog_errno(status);
2550                 goto leave;
2551         }
2552
2553         clear_nlink(inode);
2554         /* do the real work now. */
2555         status = __ocfs2_mknod_locked(dir, inode,
2556                                       0, &new_di_bh, parent_di_bh, handle,
2557                                       inode_ac, di_blkno, suballoc_loc,
2558                                       suballoc_bit);
2559         if (status < 0) {
2560                 mlog_errno(status);
2561                 goto leave;
2562         }
2563
2564         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2565                                   &orphan_insert, orphan_dir, false);
2566         if (status < 0) {
2567                 mlog_errno(status);
2568                 goto leave;
2569         }
2570
2571         /* get open lock so that only nodes can't remove it from orphan dir. */
2572         status = ocfs2_open_lock(inode);
2573         if (status < 0)
2574                 mlog_errno(status);
2575
2576         insert_inode_hash(inode);
2577 leave:
2578         if (status < 0 && did_quota_inode)
2579                 dquot_free_inode(inode);
2580         if (handle)
2581                 ocfs2_commit_trans(osb, handle);
2582
2583         if (orphan_dir) {
2584                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2585                 ocfs2_inode_unlock(orphan_dir, 1);
2586                 inode_unlock(orphan_dir);
2587                 iput(orphan_dir);
2588         }
2589
2590         if ((status < 0) && inode) {
2591                 clear_nlink(inode);
2592                 iput(inode);
2593         }
2594
2595         if (inode_ac)
2596                 ocfs2_free_alloc_context(inode_ac);
2597
2598         brelse(new_di_bh);
2599
2600         if (!status)
2601                 *new_inode = inode;
2602
2603         ocfs2_free_dir_lookup_result(&orphan_insert);
2604
2605         ocfs2_inode_unlock(dir, 1);
2606         brelse(parent_di_bh);
2607         return status;
2608 }
2609
2610 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2611         struct inode *inode)
2612 {
2613         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2614         struct inode *orphan_dir_inode = NULL;
2615         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2616         struct buffer_head *di_bh = NULL;
2617         int status = 0;
2618         handle_t *handle = NULL;
2619         struct ocfs2_dinode *di = NULL;
2620
2621         status = ocfs2_inode_lock(inode, &di_bh, 1);
2622         if (status < 0) {
2623                 mlog_errno(status);
2624                 goto bail;
2625         }
2626
2627         di = (struct ocfs2_dinode *) di_bh->b_data;
2628         /*
2629          * Another append dio crashed?
2630          * If so, manually recover it first.
2631          */
2632         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2633                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2634                 if (status < 0) {
2635                         if (status != -ENOSPC)
2636                                 mlog_errno(status);
2637                         goto bail_unlock_inode;
2638                 }
2639
2640                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2641                 if (status < 0) {
2642                         mlog_errno(status);
2643                         goto bail_unlock_inode;
2644                 }
2645         }
2646
2647         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2648                         OCFS2_I(inode)->ip_blkno,
2649                         orphan_name,
2650                         &orphan_insert,
2651                         true);
2652         if (status < 0) {
2653                 mlog_errno(status);
2654                 goto bail_unlock_inode;
2655         }
2656
2657         handle = ocfs2_start_trans(osb,
2658                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2659         if (IS_ERR(handle)) {
2660                 status = PTR_ERR(handle);
2661                 goto bail_unlock_orphan;
2662         }
2663
2664         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2665                         &orphan_insert, orphan_dir_inode, true);
2666         if (status)
2667                 mlog_errno(status);
2668
2669         ocfs2_commit_trans(osb, handle);
2670
2671 bail_unlock_orphan:
2672         ocfs2_inode_unlock(orphan_dir_inode, 1);
2673         inode_unlock(orphan_dir_inode);
2674         iput(orphan_dir_inode);
2675
2676         ocfs2_free_dir_lookup_result(&orphan_insert);
2677
2678 bail_unlock_inode:
2679         ocfs2_inode_unlock(inode, 1);
2680         brelse(di_bh);
2681
2682 bail:
2683         return status;
2684 }
2685
2686 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2687                 struct inode *inode, struct buffer_head *di_bh,
2688                 int update_isize, loff_t end)
2689 {
2690         struct inode *orphan_dir_inode = NULL;
2691         struct buffer_head *orphan_dir_bh = NULL;
2692         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2693         handle_t *handle = NULL;
2694         int status = 0;
2695
2696         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2697                         ORPHAN_DIR_SYSTEM_INODE,
2698                         le16_to_cpu(di->i_dio_orphaned_slot));
2699         if (!orphan_dir_inode) {
2700                 status = -ENOENT;
2701                 mlog_errno(status);
2702                 goto bail;
2703         }
2704
2705         inode_lock(orphan_dir_inode);
2706         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2707         if (status < 0) {
2708                 inode_unlock(orphan_dir_inode);
2709                 iput(orphan_dir_inode);
2710                 mlog_errno(status);
2711                 goto bail;
2712         }
2713
2714         handle = ocfs2_start_trans(osb,
2715                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2716         if (IS_ERR(handle)) {
2717                 status = PTR_ERR(handle);
2718                 goto bail_unlock_orphan;
2719         }
2720
2721         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2722
2723         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2724                                 inode, orphan_dir_bh, true);
2725         if (status < 0) {
2726                 mlog_errno(status);
2727                 goto bail_commit;
2728         }
2729
2730         status = ocfs2_journal_access_di(handle,
2731                         INODE_CACHE(inode),
2732                         di_bh,
2733                         OCFS2_JOURNAL_ACCESS_WRITE);
2734         if (status < 0) {
2735                 mlog_errno(status);
2736                 goto bail_commit;
2737         }
2738
2739         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2740         di->i_dio_orphaned_slot = 0;
2741
2742         if (update_isize) {
2743                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2744                 if (status)
2745                         mlog_errno(status);
2746         } else
2747                 ocfs2_journal_dirty(handle, di_bh);
2748
2749 bail_commit:
2750         ocfs2_commit_trans(osb, handle);
2751
2752 bail_unlock_orphan:
2753         ocfs2_inode_unlock(orphan_dir_inode, 1);
2754         inode_unlock(orphan_dir_inode);
2755         brelse(orphan_dir_bh);
2756         iput(orphan_dir_inode);
2757
2758 bail:
2759         return status;
2760 }
2761
2762 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2763                                    struct inode *inode,
2764                                    struct dentry *dentry)
2765 {
2766         int status = 0;
2767         struct buffer_head *parent_di_bh = NULL;
2768         handle_t *handle = NULL;
2769         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2770         struct ocfs2_dinode *dir_di, *di;
2771         struct inode *orphan_dir_inode = NULL;
2772         struct buffer_head *orphan_dir_bh = NULL;
2773         struct buffer_head *di_bh = NULL;
2774         struct ocfs2_dir_lookup_result lookup = { NULL, };
2775
2776         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2777                                 dentry->d_name.len, dentry->d_name.name,
2778                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2779                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2780
2781         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2782         if (status < 0) {
2783                 if (status != -ENOENT)
2784                         mlog_errno(status);
2785                 return status;
2786         }
2787
2788         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2789         if (!dir_di->i_links_count) {
2790                 /* can't make a file in a deleted directory. */
2791                 status = -ENOENT;
2792                 goto leave;
2793         }
2794
2795         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2796                                            dentry->d_name.len);
2797         if (status)
2798                 goto leave;
2799
2800         /* get a spot inside the dir. */
2801         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2802                                               dentry->d_name.name,
2803                                               dentry->d_name.len, &lookup);
2804         if (status < 0) {
2805                 mlog_errno(status);
2806                 goto leave;
2807         }
2808
2809         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2810                                                        ORPHAN_DIR_SYSTEM_INODE,
2811                                                        osb->slot_num);
2812         if (!orphan_dir_inode) {
2813                 status = -ENOENT;
2814                 mlog_errno(status);
2815                 goto leave;
2816         }
2817
2818         inode_lock(orphan_dir_inode);
2819
2820         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2821         if (status < 0) {
2822                 mlog_errno(status);
2823                 inode_unlock(orphan_dir_inode);
2824                 iput(orphan_dir_inode);
2825                 goto leave;
2826         }
2827
2828         status = ocfs2_read_inode_block(inode, &di_bh);
2829         if (status < 0) {
2830                 mlog_errno(status);
2831                 goto orphan_unlock;
2832         }
2833
2834         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2835         if (IS_ERR(handle)) {
2836                 status = PTR_ERR(handle);
2837                 handle = NULL;
2838                 mlog_errno(status);
2839                 goto orphan_unlock;
2840         }
2841
2842         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2843                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2844         if (status < 0) {
2845                 mlog_errno(status);
2846                 goto out_commit;
2847         }
2848
2849         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2850                                   orphan_dir_bh, false);
2851         if (status < 0) {
2852                 mlog_errno(status);
2853                 goto out_commit;
2854         }
2855
2856         di = (struct ocfs2_dinode *)di_bh->b_data;
2857         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2858         di->i_orphaned_slot = 0;
2859         set_nlink(inode, 1);
2860         ocfs2_set_links_count(di, inode->i_nlink);
2861         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2862         ocfs2_journal_dirty(handle, di_bh);
2863
2864         status = ocfs2_add_entry(handle, dentry, inode,
2865                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2866                                  &lookup);
2867         if (status < 0) {
2868                 mlog_errno(status);
2869                 goto out_commit;
2870         }
2871
2872         status = ocfs2_dentry_attach_lock(dentry, inode,
2873                                           OCFS2_I(dir)->ip_blkno);
2874         if (status) {
2875                 mlog_errno(status);
2876                 goto out_commit;
2877         }
2878
2879         d_instantiate(dentry, inode);
2880         status = 0;
2881 out_commit:
2882         ocfs2_commit_trans(osb, handle);
2883 orphan_unlock:
2884         ocfs2_inode_unlock(orphan_dir_inode, 1);
2885         inode_unlock(orphan_dir_inode);
2886         iput(orphan_dir_inode);
2887 leave:
2888
2889         ocfs2_inode_unlock(dir, 1);
2890
2891         brelse(di_bh);
2892         brelse(parent_di_bh);
2893         brelse(orphan_dir_bh);
2894
2895         ocfs2_free_dir_lookup_result(&lookup);
2896
2897         if (status)
2898                 mlog_errno(status);
2899
2900         return status;
2901 }
2902
2903 const struct inode_operations ocfs2_dir_iops = {
2904         .create         = ocfs2_create,
2905         .lookup         = ocfs2_lookup,
2906         .link           = ocfs2_link,
2907         .unlink         = ocfs2_unlink,
2908         .rmdir          = ocfs2_unlink,
2909         .symlink        = ocfs2_symlink,
2910         .mkdir          = ocfs2_mkdir,
2911         .mknod          = ocfs2_mknod,
2912         .rename         = ocfs2_rename,
2913         .setattr        = ocfs2_setattr,
2914         .getattr        = ocfs2_getattr,
2915         .permission     = ocfs2_permission,
2916         .listxattr      = ocfs2_listxattr,
2917         .fiemap         = ocfs2_fiemap,
2918         .get_acl        = ocfs2_iop_get_acl,
2919         .set_acl        = ocfs2_iop_set_acl,
2920         .fileattr_get   = ocfs2_fileattr_get,
2921         .fileattr_set   = ocfs2_fileattr_set,
2922 };