fs/ocfs2/namei.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* -*- mode: c; c-basic-offset: 8; -*-
   3  * vim: noexpandtab sw=8 ts=8 sts=0:
   4  *
   5  * namei.c
   6  *
   7  * Create and rename file, directory, symlinks
   8  *
   9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
  10  *
  11  *  Portions of this code from linux/fs/ext3/dir.c
  12  *
  13  *  Copyright (C) 1992, 1993, 1994, 1995
  14  *  Remy Card (card@masi.ibp.fr)
  15  *  Laboratoire MASI - Institut Blaise pascal
  16  *  Universite Pierre et Marie Curie (Paris VI)
  17  *
  18  *   from
  19  *
  20  *   linux/fs/minix/dir.c
  21  *
  22  *   Copyright (C) 1991, 1992 Linux Torvalds
  23  */
  24
  25 #include <linux/fs.h>
  26 #include <linux/types.h>
  27 #include <linux/slab.h>
  28 #include <linux/highmem.h>
  29 #include <linux/quotaops.h>
  30 #include <linux/iversion.h>
  31
  32 #include <cluster/masklog.h>
  33
  34 #include "ocfs2.h"
  35
  36 #include "alloc.h"
  37 #include "dcache.h"
  38 #include "dir.h"
  39 #include "dlmglue.h"
  40 #include "extent_map.h"
  41 #include "file.h"
  42 #include "inode.h"
  43 #include "journal.h"
  44 #include "namei.h"
  45 #include "suballoc.h"
  46 #include "super.h"
  47 #include "symlink.h"
  48 #include "sysfile.h"
  49 #include "uptodate.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "ocfs2_trace.h"
  53 #include "ioctl.h"
  54
  55 #include "buffer_head_io.h"
  56
  57 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  58                               struct inode *dir,
  59                               struct inode *inode,
  60                               dev_t dev,
  61                               struct buffer_head **new_fe_bh,
  62                               struct buffer_head *parent_fe_bh,
  63                               handle_t *handle,
  64                               struct ocfs2_alloc_context *inode_ac);
  65
  66 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  67                                     struct inode **ret_orphan_dir,
  68                                     u64 blkno,
  69                                     char *name,
  70                                     struct ocfs2_dir_lookup_result *lookup,
  71                                     bool dio);
  72
  73 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  74                             handle_t *handle,
  75                             struct inode *inode,
  76                             struct buffer_head *fe_bh,
  77                             char *name,
  78                             struct ocfs2_dir_lookup_result *lookup,
  79                             struct inode *orphan_dir_inode,
  80                             bool dio);
  81
  82 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  83                                      handle_t *handle,
  84                                      struct inode *inode,
  85                                      const char *symname);
  86
  87 static int ocfs2_double_lock(struct ocfs2_super *osb,
  88                              struct buffer_head **bh1,
  89                              struct inode *inode1,
  90                              struct buffer_head **bh2,
  91                              struct inode *inode2,
  92                              int rename);
  93
  94 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
  95 /* An orphan dir name is an 8 byte value, printed as a hex string */
  96 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  97
  98 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
  99                                    unsigned int flags)
 100 {
 101         int status;
 102         u64 blkno;
 103         struct inode *inode = NULL;
 104         struct dentry *ret;
 105         struct ocfs2_inode_info *oi;
 106
 107         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 108                            dentry->d_name.name,
 109                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 110
 111         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 112                 ret = ERR_PTR(-ENAMETOOLONG);
 113                 goto bail;
 114         }
 115
 116         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 117         if (status < 0) {
 118                 if (status != -ENOENT)
 119                         mlog_errno(status);
 120                 ret = ERR_PTR(status);
 121                 goto bail;
 122         }
 123
 124         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 125                                             dentry->d_name.len, &blkno);
 126         if (status < 0)
 127                 goto bail_add;
 128
 129         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 130         if (IS_ERR(inode)) {
 131                 ret = ERR_PTR(-EACCES);
 132                 goto bail_unlock;
 133         }
 134
 135         oi = OCFS2_I(inode);
 136         /* Clear any orphaned state... If we were able to look up the
 137          * inode from a directory, it certainly can't be orphaned. We
 138          * might have the bad state from a node which intended to
 139          * orphan this inode but crashed before it could commit the
 140          * unlink. */
 141         spin_lock(&oi->ip_lock);
 142         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 143         spin_unlock(&oi->ip_lock);
 144
 145 bail_add:
 146         ret = d_splice_alias(inode, dentry);
 147
 148         if (inode) {
 149                 /*
 150                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 151                  * dentry, it will d_move() it on top of ourse. The
 152                  * return value will indicate this however, so in
 153                  * those cases, we switch them around for the locking
 154                  * code.
 155                  *
 156                  * NOTE: This dentry already has ->d_op set from
 157                  * ocfs2_get_parent() and ocfs2_get_dentry()
 158                  */
 159                 if (!IS_ERR_OR_NULL(ret))
 160                         dentry = ret;
 161
 162                 status = ocfs2_dentry_attach_lock(dentry, inode,
 163                                                   OCFS2_I(dir)->ip_blkno);
 164                 if (status) {
 165                         mlog_errno(status);
 166                         ret = ERR_PTR(status);
 167                         goto bail_unlock;
 168                 }
 169         } else
 170                 ocfs2_dentry_attach_gen(dentry);
 171
 172 bail_unlock:
 173         /* Don't drop the cluster lock until *after* the d_add --
 174          * unlink on another node will message us to remove that
 175          * dentry under this lock so otherwise we can race this with
 176          * the downconvert thread and have a stale dentry. */
 177         ocfs2_inode_unlock(dir, 0);
 178
 179 bail:
 180
 181         trace_ocfs2_lookup_ret(ret);
 182
 183         return ret;
 184 }
 185
 186 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 187 {
 188         struct inode *inode;
 189         int status;
 190
 191         inode = new_inode(dir->i_sb);
 192         if (!inode) {
 193                 mlog(ML_ERROR, "new_inode failed!\n");
 194                 return ERR_PTR(-ENOMEM);
 195         }
 196
 197         /* populate as many fields early on as possible - many of
 198          * these are used by the support functions here and in
 199          * callers. */
 200         if (S_ISDIR(mode))
 201                 set_nlink(inode, 2);
 202         inode_init_owner(&init_user_ns, inode, dir, mode);
 203         status = dquot_initialize(inode);
 204         if (status)
 205                 return ERR_PTR(status);
 206
 207         return inode;
 208 }
 209
 210 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 211                 struct dentry *dentry, struct inode *inode)
 212 {
 213         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 214
 215         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 216         ocfs2_lock_res_free(&dl->dl_lockres);
 217         BUG_ON(dl->dl_count != 1);
 218         spin_lock(&dentry_attach_lock);
 219         dentry->d_fsdata = NULL;
 220         spin_unlock(&dentry_attach_lock);
 221         kfree(dl);
 222         iput(inode);
 223 }
 224
 225 static int ocfs2_mknod(struct user_namespace *mnt_userns,
 226                        struct inode *dir,
 227                        struct dentry *dentry,
 228                        umode_t mode,
 229                        dev_t dev)
 230 {
 231         int status = 0;
 232         struct buffer_head *parent_fe_bh = NULL;
 233         handle_t *handle = NULL;
 234         struct ocfs2_super *osb;
 235         struct ocfs2_dinode *dirfe;
 236         struct buffer_head *new_fe_bh = NULL;
 237         struct inode *inode = NULL;
 238         struct ocfs2_alloc_context *inode_ac = NULL;
 239         struct ocfs2_alloc_context *data_ac = NULL;
 240         struct ocfs2_alloc_context *meta_ac = NULL;
 241         int want_clusters = 0;
 242         int want_meta = 0;
 243         int xattr_credits = 0;
 244         struct ocfs2_security_xattr_info si = {
 245                 .enable = 1,
 246         };
 247         int did_quota_inode = 0;
 248         struct ocfs2_dir_lookup_result lookup = { NULL, };
 249         sigset_t oldset;
 250         int did_block_signals = 0;
 251         struct ocfs2_dentry_lock *dl = NULL;
 252
 253         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 254                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 255                           (unsigned long)dev, mode);
 256
 257         status = dquot_initialize(dir);
 258         if (status) {
 259                 mlog_errno(status);
 260                 return status;
 261         }
 262
 263         /* get our super block */
 264         osb = OCFS2_SB(dir->i_sb);
 265
 266         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 267         if (status < 0) {
 268                 if (status != -ENOENT)
 269                         mlog_errno(status);
 270                 return status;
 271         }
 272
 273         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 274                 status = -EMLINK;
 275                 goto leave;
 276         }
 277
 278         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 279         if (!ocfs2_read_links_count(dirfe)) {
 280                 /* can't make a file in a deleted directory. */
 281                 status = -ENOENT;
 282                 goto leave;
 283         }
 284
 285         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 286                                            dentry->d_name.len);
 287         if (status)
 288                 goto leave;
 289
 290         /* get a spot inside the dir. */
 291         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 292                                               dentry->d_name.name,
 293                                               dentry->d_name.len, &lookup);
 294         if (status < 0) {
 295                 mlog_errno(status);
 296                 goto leave;
 297         }
 298
 299         /* reserve an inode spot */
 300         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 301         if (status < 0) {
 302                 if (status != -ENOSPC)
 303                         mlog_errno(status);
 304                 goto leave;
 305         }
 306
 307         inode = ocfs2_get_init_inode(dir, mode);
 308         if (IS_ERR(inode)) {
 309                 status = PTR_ERR(inode);
 310                 inode = NULL;
 311                 mlog_errno(status);
 312                 goto leave;
 313         }
 314
 315         /* get security xattr */
 316         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 317         if (status) {
 318                 if (status == -EOPNOTSUPP)
 319                         si.enable = 0;
 320                 else {
 321                         mlog_errno(status);
 322                         goto leave;
 323                 }
 324         }
 325
 326         /* calculate meta data/clusters for setting security and acl xattr */
 327         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 328                                        &si, &want_clusters,
 329                                        &xattr_credits, &want_meta);
 330         if (status < 0) {
 331                 mlog_errno(status);
 332                 goto leave;
 333         }
 334
 335         /* Reserve a cluster if creating an extent based directory. */
 336         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 337                 want_clusters += 1;
 338
 339                 /* Dir indexing requires extra space as well */
 340                 if (ocfs2_supports_indexed_dirs(osb))
 341                         want_meta++;
 342         }
 343
 344         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 345         if (status < 0) {
 346                 if (status != -ENOSPC)
 347                         mlog_errno(status);
 348                 goto leave;
 349         }
 350
 351         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 352         if (status < 0) {
 353                 if (status != -ENOSPC)
 354                         mlog_errno(status);
 355                 goto leave;
 356         }
 357
 358         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 359                                                             S_ISDIR(mode),
 360                                                             xattr_credits));
 361         if (IS_ERR(handle)) {
 362                 status = PTR_ERR(handle);
 363                 handle = NULL;
 364                 mlog_errno(status);
 365                 goto leave;
 366         }
 367
 368         /* Starting to change things, restart is no longer possible. */
 369         ocfs2_block_signals(&oldset);
 370         did_block_signals = 1;
 371
 372         status = dquot_alloc_inode(inode);
 373         if (status)
 374                 goto leave;
 375         did_quota_inode = 1;
 376
 377         /* do the real work now. */
 378         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 379                                     &new_fe_bh, parent_fe_bh, handle,
 380                                     inode_ac);
 381         if (status < 0) {
 382                 mlog_errno(status);
 383                 goto leave;
 384         }
 385
 386         if (S_ISDIR(mode)) {
 387                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 388                                             new_fe_bh, data_ac, meta_ac);
 389                 if (status < 0) {
 390                         mlog_errno(status);
 391                         goto leave;
 392                 }
 393
 394                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 395                                                  parent_fe_bh,
 396                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 397                 if (status < 0) {
 398                         mlog_errno(status);
 399                         goto leave;
 400                 }
 401                 ocfs2_add_links_count(dirfe, 1);
 402                 ocfs2_journal_dirty(handle, parent_fe_bh);
 403                 inc_nlink(dir);
 404         }
 405
 406         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 407                          meta_ac, data_ac);
 408
 409         if (status < 0) {
 410                 mlog_errno(status);
 411                 goto roll_back;
 412         }
 413
 414         if (si.enable) {
 415                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 416                                                  meta_ac, data_ac);
 417                 if (status < 0) {
 418                         mlog_errno(status);
 419                         goto roll_back;
 420                 }
 421         }
 422
 423         /*
 424          * Do this before adding the entry to the directory. We add
 425          * also set d_op after success so that ->d_iput() will cleanup
 426          * the dentry lock even if ocfs2_add_entry() fails below.
 427          */
 428         status = ocfs2_dentry_attach_lock(dentry, inode,
 429                                           OCFS2_I(dir)->ip_blkno);
 430         if (status) {
 431                 mlog_errno(status);
 432                 goto roll_back;
 433         }
 434
 435         dl = dentry->d_fsdata;
 436
 437         status = ocfs2_add_entry(handle, dentry, inode,
 438                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 439                                  &lookup);
 440         if (status < 0) {
 441                 mlog_errno(status);
 442                 goto roll_back;
 443         }
 444
 445         insert_inode_hash(inode);
 446         d_instantiate(dentry, inode);
 447         status = 0;
 448
 449 roll_back:
 450         if (status < 0 && S_ISDIR(mode)) {
 451                 ocfs2_add_links_count(dirfe, -1);
 452                 drop_nlink(dir);
 453         }
 454
 455 leave:
 456         if (status < 0 && did_quota_inode)
 457                 dquot_free_inode(inode);
 458         if (handle)
 459                 ocfs2_commit_trans(osb, handle);
 460
 461         ocfs2_inode_unlock(dir, 1);
 462         if (did_block_signals)
 463                 ocfs2_unblock_signals(&oldset);
 464
 465         brelse(new_fe_bh);
 466         brelse(parent_fe_bh);
 467         kfree(si.value);
 468
 469         ocfs2_free_dir_lookup_result(&lookup);
 470
 471         if (inode_ac)
 472                 ocfs2_free_alloc_context(inode_ac);
 473
 474         if (data_ac)
 475                 ocfs2_free_alloc_context(data_ac);
 476
 477         if (meta_ac)
 478                 ocfs2_free_alloc_context(meta_ac);
 479
 480         /*
 481          * We should call iput after the i_mutex of the bitmap been
 482          * unlocked in ocfs2_free_alloc_context, or the
 483          * ocfs2_delete_inode will mutex_lock again.
 484          */
 485         if ((status < 0) && inode) {
 486                 if (dl)
 487                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 488
 489                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 490                 clear_nlink(inode);
 491                 iput(inode);
 492         }
 493
 494         if (status)
 495                 mlog_errno(status);
 496
 497         return status;
 498 }
 499
 500 static int __ocfs2_mknod_locked(struct inode *dir,
 501                                 struct inode *inode,
 502                                 dev_t dev,
 503                                 struct buffer_head **new_fe_bh,
 504                                 struct buffer_head *parent_fe_bh,
 505                                 handle_t *handle,
 506                                 struct ocfs2_alloc_context *inode_ac,
 507                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 508 {
 509         int status = 0;
 510         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 511         struct ocfs2_dinode *fe = NULL;
 512         struct ocfs2_extent_list *fel;
 513         u16 feat;
 514         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 515         struct timespec64 ts;
 516
 517         *new_fe_bh = NULL;
 518
 519         /* populate as many fields early on as possible - many of
 520          * these are used by the support functions here and in
 521          * callers. */
 522         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 523         oi->ip_blkno = fe_blkno;
 524         spin_lock(&osb->osb_lock);
 525         inode->i_generation = osb->s_next_generation++;
 526         spin_unlock(&osb->osb_lock);
 527
 528         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 529         if (!*new_fe_bh) {
 530                 status = -ENOMEM;
 531                 mlog_errno(status);
 532                 goto leave;
 533         }
 534         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 535
 536         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 537                                          *new_fe_bh,
 538                                          OCFS2_JOURNAL_ACCESS_CREATE);
 539         if (status < 0) {
 540                 mlog_errno(status);
 541                 goto leave;
 542         }
 543
 544         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 545         memset(fe, 0, osb->sb->s_blocksize);
 546
 547         fe->i_generation = cpu_to_le32(inode->i_generation);
 548         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 549         fe->i_blkno = cpu_to_le64(fe_blkno);
 550         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 551         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 552         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 553         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 554         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 555         fe->i_mode = cpu_to_le16(inode->i_mode);
 556         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 557                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 558
 559         ocfs2_set_links_count(fe, inode->i_nlink);
 560
 561         fe->i_last_eb_blk = 0;
 562         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 563         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 564         ktime_get_real_ts64(&ts);
 565         fe->i_atime = fe->i_ctime = fe->i_mtime =
 566                 cpu_to_le64(ts.tv_sec);
 567         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 568                 cpu_to_le32(ts.tv_nsec);
 569         fe->i_dtime = 0;
 570
 571         /*
 572          * If supported, directories start with inline data. If inline
 573          * isn't supported, but indexing is, we start them as indexed.
 574          */
 575         feat = le16_to_cpu(fe->i_dyn_features);
 576         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 577                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 578
 579                 fe->id2.i_data.id_count = cpu_to_le16(
 580                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 581         } else {
 582                 fel = &fe->id2.i_list;
 583                 fel->l_tree_depth = 0;
 584                 fel->l_next_free_rec = 0;
 585                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 586         }
 587
 588         ocfs2_journal_dirty(handle, *new_fe_bh);
 589
 590         ocfs2_populate_inode(inode, fe, 1);
 591         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 592         if (!ocfs2_mount_local(osb)) {
 593                 status = ocfs2_create_new_inode_locks(inode);
 594                 if (status < 0)
 595                         mlog_errno(status);
 596         }
 597
 598         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 599
 600 leave:
 601         if (status < 0) {
 602                 if (*new_fe_bh) {
 603                         brelse(*new_fe_bh);
 604                         *new_fe_bh = NULL;
 605                 }
 606         }
 607
 608         if (status)
 609                 mlog_errno(status);
 610         return status;
 611 }
 612
 613 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 614                               struct inode *dir,
 615                               struct inode *inode,
 616                               dev_t dev,
 617                               struct buffer_head **new_fe_bh,
 618                               struct buffer_head *parent_fe_bh,
 619                               handle_t *handle,
 620                               struct ocfs2_alloc_context *inode_ac)
 621 {
 622         int status = 0;
 623         u64 suballoc_loc, fe_blkno = 0;
 624         u16 suballoc_bit;
 625
 626         *new_fe_bh = NULL;
 627
 628         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 629                                        inode_ac, &suballoc_loc,
 630                                        &suballoc_bit, &fe_blkno);
 631         if (status < 0) {
 632                 mlog_errno(status);
 633                 return status;
 634         }
 635
 636         status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 637                                     parent_fe_bh, handle, inode_ac,
 638                                     fe_blkno, suballoc_loc, suballoc_bit);
 639         if (status < 0) {
 640                 u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
 641                 int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
 642                                 inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
 643                 if (tmp)
 644                         mlog_errno(tmp);
 645         }
 646
 647         return status;
 648 }
 649
 650 static int ocfs2_mkdir(struct user_namespace *mnt_userns,
 651                        struct inode *dir,
 652                        struct dentry *dentry,
 653                        umode_t mode)
 654 {
 655         int ret;
 656
 657         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 658                           OCFS2_I(dir)->ip_blkno, mode);
 659         ret = ocfs2_mknod(&init_user_ns, dir, dentry, mode | S_IFDIR, 0);
 660         if (ret)
 661                 mlog_errno(ret);
 662
 663         return ret;
 664 }
 665
 666 static int ocfs2_create(struct user_namespace *mnt_userns,
 667                         struct inode *dir,
 668                         struct dentry *dentry,
 669                         umode_t mode,
 670                         bool excl)
 671 {
 672         int ret;
 673
 674         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 675                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 676         ret = ocfs2_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0);
 677         if (ret)
 678                 mlog_errno(ret);
 679
 680         return ret;
 681 }
 682
 683 static int ocfs2_link(struct dentry *old_dentry,
 684                       struct inode *dir,
 685                       struct dentry *dentry)
 686 {
 687         handle_t *handle;
 688         struct inode *inode = d_inode(old_dentry);
 689         struct inode *old_dir = d_inode(old_dentry->d_parent);
 690         int err;
 691         struct buffer_head *fe_bh = NULL;
 692         struct buffer_head *old_dir_bh = NULL;
 693         struct buffer_head *parent_fe_bh = NULL;
 694         struct ocfs2_dinode *fe = NULL;
 695         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 696         struct ocfs2_dir_lookup_result lookup = { NULL, };
 697         sigset_t oldset;
 698         u64 old_de_ino;
 699
 700         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 701                          old_dentry->d_name.len, old_dentry->d_name.name,
 702                          dentry->d_name.len, dentry->d_name.name);
 703
 704         if (S_ISDIR(inode->i_mode))
 705                 return -EPERM;
 706
 707         err = dquot_initialize(dir);
 708         if (err) {
 709                 mlog_errno(err);
 710                 return err;
 711         }
 712
 713         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 714                         &parent_fe_bh, dir, 0);
 715         if (err < 0) {
 716                 if (err != -ENOENT)
 717                         mlog_errno(err);
 718                 return err;
 719         }
 720
 721         /* make sure both dirs have bhs
 722          * get an extra ref on old_dir_bh if old==new */
 723         if (!parent_fe_bh) {
 724                 if (old_dir_bh) {
 725                         parent_fe_bh = old_dir_bh;
 726                         get_bh(parent_fe_bh);
 727                 } else {
 728                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 729                         err = -EIO;
 730                         goto out;
 731                 }
 732         }
 733
 734         if (!dir->i_nlink) {
 735                 err = -ENOENT;
 736                 goto out;
 737         }
 738
 739         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 740                         old_dentry->d_name.len, &old_de_ino);
 741         if (err) {
 742                 err = -ENOENT;
 743                 goto out;
 744         }
 745
 746         /*
 747          * Check whether another node removed the source inode while we
 748          * were in the vfs.
 749          */
 750         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 751                 err = -ENOENT;
 752                 goto out;
 753         }
 754
 755         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 756                                         dentry->d_name.len);
 757         if (err)
 758                 goto out;
 759
 760         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 761                                            dentry->d_name.name,
 762                                            dentry->d_name.len, &lookup);
 763         if (err < 0) {
 764                 mlog_errno(err);
 765                 goto out;
 766         }
 767
 768         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 769         if (err < 0) {
 770                 if (err != -ENOENT)
 771                         mlog_errno(err);
 772                 goto out;
 773         }
 774
 775         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 776         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 777                 err = -EMLINK;
 778                 goto out_unlock_inode;
 779         }
 780
 781         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 782         if (IS_ERR(handle)) {
 783                 err = PTR_ERR(handle);
 784                 handle = NULL;
 785                 mlog_errno(err);
 786                 goto out_unlock_inode;
 787         }
 788
 789         /* Starting to change things, restart is no longer possible. */
 790         ocfs2_block_signals(&oldset);
 791
 792         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 793                                       OCFS2_JOURNAL_ACCESS_WRITE);
 794         if (err < 0) {
 795                 mlog_errno(err);
 796                 goto out_commit;
 797         }
 798
 799         inc_nlink(inode);
 800         inode->i_ctime = current_time(inode);
 801         ocfs2_set_links_count(fe, inode->i_nlink);
 802         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 803         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 804         ocfs2_journal_dirty(handle, fe_bh);
 805
 806         err = ocfs2_add_entry(handle, dentry, inode,
 807                               OCFS2_I(inode)->ip_blkno,
 808                               parent_fe_bh, &lookup);
 809         if (err) {
 810                 ocfs2_add_links_count(fe, -1);
 811                 drop_nlink(inode);
 812                 mlog_errno(err);
 813                 goto out_commit;
 814         }
 815
 816         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 817         if (err) {
 818                 mlog_errno(err);
 819                 goto out_commit;
 820         }
 821
 822         ihold(inode);
 823         d_instantiate(dentry, inode);
 824
 825 out_commit:
 826         ocfs2_commit_trans(osb, handle);
 827         ocfs2_unblock_signals(&oldset);
 828 out_unlock_inode:
 829         ocfs2_inode_unlock(inode, 1);
 830
 831 out:
 832         ocfs2_double_unlock(old_dir, dir);
 833
 834         brelse(fe_bh);
 835         brelse(parent_fe_bh);
 836         brelse(old_dir_bh);
 837
 838         ocfs2_free_dir_lookup_result(&lookup);
 839
 840         if (err)
 841                 mlog_errno(err);
 842
 843         return err;
 844 }
 845
 846 /*
 847  * Takes and drops an exclusive lock on the given dentry. This will
 848  * force other nodes to drop it.
 849  */
 850 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 851 {
 852         int ret;
 853
 854         ret = ocfs2_dentry_lock(dentry, 1);
 855         if (ret)
 856                 mlog_errno(ret);
 857         else
 858                 ocfs2_dentry_unlock(dentry, 1);
 859
 860         return ret;
 861 }
 862
 863 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 864 {
 865         if (S_ISDIR(inode->i_mode)) {
 866                 if (inode->i_nlink == 2)
 867                         return 1;
 868                 return 0;
 869         }
 870
 871         if (inode->i_nlink == 1)
 872                 return 1;
 873         return 0;
 874 }
 875
 876 static int ocfs2_unlink(struct inode *dir,
 877                         struct dentry *dentry)
 878 {
 879         int status;
 880         int child_locked = 0;
 881         bool is_unlinkable = false;
 882         struct inode *inode = d_inode(dentry);
 883         struct inode *orphan_dir = NULL;
 884         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 885         u64 blkno;
 886         struct ocfs2_dinode *fe = NULL;
 887         struct buffer_head *fe_bh = NULL;
 888         struct buffer_head *parent_node_bh = NULL;
 889         handle_t *handle = NULL;
 890         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 891         struct ocfs2_dir_lookup_result lookup = { NULL, };
 892         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 893
 894         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 895                            dentry->d_name.name,
 896                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 897                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 898
 899         status = dquot_initialize(dir);
 900         if (status) {
 901                 mlog_errno(status);
 902                 return status;
 903         }
 904
 905         BUG_ON(d_inode(dentry->d_parent) != dir);
 906
 907         if (inode == osb->root_inode)
 908                 return -EPERM;
 909
 910         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 911                                          OI_LS_PARENT);
 912         if (status < 0) {
 913                 if (status != -ENOENT)
 914                         mlog_errno(status);
 915                 return status;
 916         }
 917
 918         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 919                                           dentry->d_name.len, &blkno, dir,
 920                                           &lookup);
 921         if (status < 0) {
 922                 if (status != -ENOENT)
 923                         mlog_errno(status);
 924                 goto leave;
 925         }
 926
 927         if (OCFS2_I(inode)->ip_blkno != blkno) {
 928                 status = -ENOENT;
 929
 930                 trace_ocfs2_unlink_noent(
 931                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 932                                 (unsigned long long)blkno,
 933                                 OCFS2_I(inode)->ip_flags);
 934                 goto leave;
 935         }
 936
 937         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 938         if (status < 0) {
 939                 if (status != -ENOENT)
 940                         mlog_errno(status);
 941                 goto leave;
 942         }
 943         child_locked = 1;
 944
 945         if (S_ISDIR(inode->i_mode)) {
 946                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 947                         status = -ENOTEMPTY;
 948                         goto leave;
 949                 }
 950         }
 951
 952         status = ocfs2_remote_dentry_delete(dentry);
 953         if (status < 0) {
 954                 /* This remote delete should succeed under all normal
 955                  * circumstances. */
 956                 mlog_errno(status);
 957                 goto leave;
 958         }
 959
 960         if (ocfs2_inode_is_unlinkable(inode)) {
 961                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 962                                                   OCFS2_I(inode)->ip_blkno,
 963                                                   orphan_name, &orphan_insert,
 964                                                   false);
 965                 if (status < 0) {
 966                         mlog_errno(status);
 967                         goto leave;
 968                 }
 969                 is_unlinkable = true;
 970         }
 971
 972         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 973         if (IS_ERR(handle)) {
 974                 status = PTR_ERR(handle);
 975                 handle = NULL;
 976                 mlog_errno(status);
 977                 goto leave;
 978         }
 979
 980         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 981                                          OCFS2_JOURNAL_ACCESS_WRITE);
 982         if (status < 0) {
 983                 mlog_errno(status);
 984                 goto leave;
 985         }
 986
 987         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 988
 989         /* delete the name from the parent dir */
 990         status = ocfs2_delete_entry(handle, dir, &lookup);
 991         if (status < 0) {
 992                 mlog_errno(status);
 993                 goto leave;
 994         }
 995
 996         if (S_ISDIR(inode->i_mode))
 997                 drop_nlink(inode);
 998         drop_nlink(inode);
 999         ocfs2_set_links_count(fe, inode->i_nlink);
1000         ocfs2_journal_dirty(handle, fe_bh);
1001
1002         dir->i_ctime = dir->i_mtime = current_time(dir);
1003         if (S_ISDIR(inode->i_mode))
1004                 drop_nlink(dir);
1005
1006         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1007         if (status < 0) {
1008                 mlog_errno(status);
1009                 if (S_ISDIR(inode->i_mode))
1010                         inc_nlink(dir);
1011                 goto leave;
1012         }
1013
1014         if (is_unlinkable) {
1015                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1016                                 orphan_name, &orphan_insert, orphan_dir, false);
1017                 if (status < 0)
1018                         mlog_errno(status);
1019         }
1020
1021 leave:
1022         if (handle)
1023                 ocfs2_commit_trans(osb, handle);
1024
1025         if (orphan_dir) {
1026                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1027                 ocfs2_inode_unlock(orphan_dir, 1);
1028                 inode_unlock(orphan_dir);
1029                 iput(orphan_dir);
1030         }
1031
1032         if (child_locked)
1033                 ocfs2_inode_unlock(inode, 1);
1034
1035         ocfs2_inode_unlock(dir, 1);
1036
1037         brelse(fe_bh);
1038         brelse(parent_node_bh);
1039
1040         ocfs2_free_dir_lookup_result(&orphan_insert);
1041         ocfs2_free_dir_lookup_result(&lookup);
1042
1043         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1044                 mlog_errno(status);
1045
1046         return status;
1047 }
1048
1049 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1050                 u64 src_inode_no, u64 dest_inode_no)
1051 {
1052         int ret = 0, i = 0;
1053         u64 parent_inode_no = 0;
1054         u64 child_inode_no = src_inode_no;
1055         struct inode *child_inode;
1056
1057 #define MAX_LOOKUP_TIMES 32
1058         while (1) {
1059                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1060                 if (IS_ERR(child_inode)) {
1061                         ret = PTR_ERR(child_inode);
1062                         break;
1063                 }
1064
1065                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1066                 if (ret < 0) {
1067                         iput(child_inode);
1068                         if (ret != -ENOENT)
1069                                 mlog_errno(ret);
1070                         break;
1071                 }
1072
1073                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1074                                 &parent_inode_no);
1075                 ocfs2_inode_unlock(child_inode, 0);
1076                 iput(child_inode);
1077                 if (ret < 0) {
1078                         ret = -ENOENT;
1079                         break;
1080                 }
1081
1082                 if (parent_inode_no == dest_inode_no) {
1083                         ret = 1;
1084                         break;
1085                 }
1086
1087                 if (parent_inode_no == osb->root_inode->i_ino) {
1088                         ret = 0;
1089                         break;
1090                 }
1091
1092                 child_inode_no = parent_inode_no;
1093
1094                 if (++i >= MAX_LOOKUP_TIMES) {
1095                         mlog_ratelimited(ML_NOTICE, "max lookup times reached, "
1096                                         "filesystem may have nested directories, "
1097                                         "src inode: %llu, dest inode: %llu.\n",
1098                                         (unsigned long long)src_inode_no,
1099                                         (unsigned long long)dest_inode_no);
1100                         ret = 0;
1101                         break;
1102                 }
1103         }
1104
1105         return ret;
1106 }
1107
1108 /*
1109  * The only place this should be used is rename and link!
1110  * if they have the same id, then the 1st one is the only one locked.
1111  */
1112 static int ocfs2_double_lock(struct ocfs2_super *osb,
1113                              struct buffer_head **bh1,
1114                              struct inode *inode1,
1115                              struct buffer_head **bh2,
1116                              struct inode *inode2,
1117                              int rename)
1118 {
1119         int status;
1120         int inode1_is_ancestor, inode2_is_ancestor;
1121         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1122         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1123
1124         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1125                                 (unsigned long long)oi2->ip_blkno);
1126
1127         if (*bh1)
1128                 *bh1 = NULL;
1129         if (*bh2)
1130                 *bh2 = NULL;
1131
1132         /* we always want to lock the one with the lower lockid first.
1133          * and if they are nested, we lock ancestor first */
1134         if (oi1->ip_blkno != oi2->ip_blkno) {
1135                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1136                                 oi1->ip_blkno);
1137                 if (inode1_is_ancestor < 0) {
1138                         status = inode1_is_ancestor;
1139                         goto bail;
1140                 }
1141
1142                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1143                                 oi2->ip_blkno);
1144                 if (inode2_is_ancestor < 0) {
1145                         status = inode2_is_ancestor;
1146                         goto bail;
1147                 }
1148
1149                 if ((inode1_is_ancestor == 1) ||
1150                                 (oi1->ip_blkno < oi2->ip_blkno &&
1151                                 inode2_is_ancestor == 0)) {
1152                         /* switch id1 and id2 around */
1153                         swap(bh2, bh1);
1154                         swap(inode2, inode1);
1155                 }
1156                 /* lock id2 */
1157                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1158                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1159                 if (status < 0) {
1160                         if (status != -ENOENT)
1161                                 mlog_errno(status);
1162                         goto bail;
1163                 }
1164         }
1165
1166         /* lock id1 */
1167         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1168                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1169         if (status < 0) {
1170                 /*
1171                  * An error return must mean that no cluster locks
1172                  * were held on function exit.
1173                  */
1174                 if (oi1->ip_blkno != oi2->ip_blkno) {
1175                         ocfs2_inode_unlock(inode2, 1);
1176                         brelse(*bh2);
1177                         *bh2 = NULL;
1178                 }
1179
1180                 if (status != -ENOENT)
1181                         mlog_errno(status);
1182         }
1183
1184         trace_ocfs2_double_lock_end(
1185                         (unsigned long long)oi1->ip_blkno,
1186                         (unsigned long long)oi2->ip_blkno);
1187
1188 bail:
1189         if (status)
1190                 mlog_errno(status);
1191         return status;
1192 }
1193
1194 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1195 {
1196         ocfs2_inode_unlock(inode1, 1);
1197
1198         if (inode1 != inode2)
1199                 ocfs2_inode_unlock(inode2, 1);
1200 }
1201
1202 static int ocfs2_rename(struct user_namespace *mnt_userns,
1203                         struct inode *old_dir,
1204                         struct dentry *old_dentry,
1205                         struct inode *new_dir,
1206                         struct dentry *new_dentry,
1207                         unsigned int flags)
1208 {
1209         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1210         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1211         struct inode *old_inode = d_inode(old_dentry);
1212         struct inode *new_inode = d_inode(new_dentry);
1213         struct inode *orphan_dir = NULL;
1214         struct ocfs2_dinode *newfe = NULL;
1215         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1216         struct buffer_head *newfe_bh = NULL;
1217         struct buffer_head *old_inode_bh = NULL;
1218         struct ocfs2_super *osb = NULL;
1219         u64 newfe_blkno, old_de_ino;
1220         handle_t *handle = NULL;
1221         struct buffer_head *old_dir_bh = NULL;
1222         struct buffer_head *new_dir_bh = NULL;
1223         u32 old_dir_nlink = old_dir->i_nlink;
1224         struct ocfs2_dinode *old_di;
1225         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1226         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1227         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1228         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1229         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1230         bool should_add_orphan = false;
1231
1232         if (flags)
1233                 return -EINVAL;
1234
1235         /* At some point it might be nice to break this function up a
1236          * bit. */
1237
1238         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1239                            old_dentry->d_name.len, old_dentry->d_name.name,
1240                            new_dentry->d_name.len, new_dentry->d_name.name);
1241
1242         status = dquot_initialize(old_dir);
1243         if (status) {
1244                 mlog_errno(status);
1245                 goto bail;
1246         }
1247         status = dquot_initialize(new_dir);
1248         if (status) {
1249                 mlog_errno(status);
1250                 goto bail;
1251         }
1252
1253         osb = OCFS2_SB(old_dir->i_sb);
1254
1255         if (new_inode) {
1256                 if (!igrab(new_inode))
1257                         BUG();
1258         }
1259
1260         /* Assume a directory hierarchy thusly:
1261          * a/b/c
1262          * a/d
1263          * a,b,c, and d are all directories.
1264          *
1265          * from cwd of 'a' on both nodes:
1266          * node1: mv b/c d
1267          * node2: mv d   b/c
1268          *
1269          * And that's why, just like the VFS, we need a file system
1270          * rename lock. */
1271         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1272                 status = ocfs2_rename_lock(osb);
1273                 if (status < 0) {
1274                         mlog_errno(status);
1275                         goto bail;
1276                 }
1277                 rename_lock = 1;
1278
1279                 /* here we cannot guarantee the inodes haven't just been
1280                  * changed, so check if they are nested again */
1281                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1282                                 old_inode->i_ino);
1283                 if (status < 0) {
1284                         mlog_errno(status);
1285                         goto bail;
1286                 } else if (status == 1) {
1287                         status = -EPERM;
1288                         trace_ocfs2_rename_not_permitted(
1289                                         (unsigned long long)old_inode->i_ino,
1290                                         (unsigned long long)new_dir->i_ino);
1291                         goto bail;
1292                 }
1293         }
1294
1295         /* if old and new are the same, this'll just do one lock. */
1296         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1297                                    &new_dir_bh, new_dir, 1);
1298         if (status < 0) {
1299                 mlog_errno(status);
1300                 goto bail;
1301         }
1302         parents_locked = 1;
1303
1304         if (!new_dir->i_nlink) {
1305                 status = -EACCES;
1306                 goto bail;
1307         }
1308
1309         /* make sure both dirs have bhs
1310          * get an extra ref on old_dir_bh if old==new */
1311         if (!new_dir_bh) {
1312                 if (old_dir_bh) {
1313                         new_dir_bh = old_dir_bh;
1314                         get_bh(new_dir_bh);
1315                 } else {
1316                         mlog(ML_ERROR, "no old_dir_bh!\n");
1317                         status = -EIO;
1318                         goto bail;
1319                 }
1320         }
1321
1322         /*
1323          * Aside from allowing a meta data update, the locking here
1324          * also ensures that the downconvert thread on other nodes
1325          * won't have to concurrently downconvert the inode and the
1326          * dentry locks.
1327          */
1328         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1329                                          OI_LS_PARENT);
1330         if (status < 0) {
1331                 if (status != -ENOENT)
1332                         mlog_errno(status);
1333                 goto bail;
1334         }
1335         old_child_locked = 1;
1336
1337         status = ocfs2_remote_dentry_delete(old_dentry);
1338         if (status < 0) {
1339                 mlog_errno(status);
1340                 goto bail;
1341         }
1342
1343         if (S_ISDIR(old_inode->i_mode)) {
1344                 u64 old_inode_parent;
1345
1346                 update_dot_dot = 1;
1347                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1348                                                   old_inode,
1349                                                   &old_inode_dot_dot_res);
1350                 if (status) {
1351                         status = -EIO;
1352                         goto bail;
1353                 }
1354
1355                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1356                         status = -EIO;
1357                         goto bail;
1358                 }
1359
1360                 if (!new_inode && new_dir != old_dir &&
1361                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1362                         status = -EMLINK;
1363                         goto bail;
1364                 }
1365         }
1366
1367         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1368                                             old_dentry->d_name.len,
1369                                             &old_de_ino);
1370         if (status) {
1371                 status = -ENOENT;
1372                 goto bail;
1373         }
1374
1375         /*
1376          *  Check for inode number is _not_ due to possible IO errors.
1377          *  We might rmdir the source, keep it as pwd of some process
1378          *  and merrily kill the link to whatever was created under the
1379          *  same name. Goodbye sticky bit ;-<
1380          */
1381         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1382                 status = -ENOENT;
1383                 goto bail;
1384         }
1385
1386         /* check if the target already exists (in which case we need
1387          * to delete it */
1388         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1389                                           new_dentry->d_name.len,
1390                                           &newfe_blkno, new_dir,
1391                                           &target_lookup_res);
1392         /* The only error we allow here is -ENOENT because the new
1393          * file not existing is perfectly valid. */
1394         if ((status < 0) && (status != -ENOENT)) {
1395                 /* If we cannot find the file specified we should just */
1396                 /* return the error... */
1397                 mlog_errno(status);
1398                 goto bail;
1399         }
1400         if (status == 0)
1401                 target_exists = 1;
1402
1403         if (!target_exists && new_inode) {
1404                 /*
1405                  * Target was unlinked by another node while we were
1406                  * waiting to get to ocfs2_rename(). There isn't
1407                  * anything we can do here to help the situation, so
1408                  * bubble up the appropriate error.
1409                  */
1410                 status = -ENOENT;
1411                 goto bail;
1412         }
1413
1414         /* In case we need to overwrite an existing file, we blow it
1415          * away first */
1416         if (target_exists) {
1417                 /* VFS didn't think there existed an inode here, but
1418                  * someone else in the cluster must have raced our
1419                  * rename to create one. Today we error cleanly, in
1420                  * the future we should consider calling iget to build
1421                  * a new struct inode for this entry. */
1422                 if (!new_inode) {
1423                         status = -EACCES;
1424
1425                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1426                                                 new_dentry->d_name.name);
1427                         goto bail;
1428                 }
1429
1430                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1431                         status = -EACCES;
1432
1433                         trace_ocfs2_rename_disagree(
1434                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1435                              (unsigned long long)newfe_blkno,
1436                              OCFS2_I(new_inode)->ip_flags);
1437                         goto bail;
1438                 }
1439
1440                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1441                 if (status < 0) {
1442                         if (status != -ENOENT)
1443                                 mlog_errno(status);
1444                         goto bail;
1445                 }
1446                 new_child_locked = 1;
1447
1448                 status = ocfs2_remote_dentry_delete(new_dentry);
1449                 if (status < 0) {
1450                         mlog_errno(status);
1451                         goto bail;
1452                 }
1453
1454                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1455
1456                 trace_ocfs2_rename_over_existing(
1457                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1458                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1459
1460                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1461                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1462                                                 OCFS2_I(new_inode)->ip_blkno,
1463                                                 orphan_name, &orphan_insert,
1464                                                 false);
1465                         if (status < 0) {
1466                                 mlog_errno(status);
1467                                 goto bail;
1468                         }
1469                         should_add_orphan = true;
1470                 }
1471         } else {
1472                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1473
1474                 status = ocfs2_check_dir_for_entry(new_dir,
1475                                                    new_dentry->d_name.name,
1476                                                    new_dentry->d_name.len);
1477                 if (status)
1478                         goto bail;
1479
1480                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1481                                                       new_dentry->d_name.name,
1482                                                       new_dentry->d_name.len,
1483                                                       &target_insert);
1484                 if (status < 0) {
1485                         mlog_errno(status);
1486                         goto bail;
1487                 }
1488         }
1489
1490         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1491         if (IS_ERR(handle)) {
1492                 status = PTR_ERR(handle);
1493                 handle = NULL;
1494                 mlog_errno(status);
1495                 goto bail;
1496         }
1497
1498         if (target_exists) {
1499                 if (S_ISDIR(new_inode->i_mode)) {
1500                         if (new_inode->i_nlink != 2 ||
1501                             !ocfs2_empty_dir(new_inode)) {
1502                                 status = -ENOTEMPTY;
1503                                 goto bail;
1504                         }
1505                 }
1506                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1507                                                  newfe_bh,
1508                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1509                 if (status < 0) {
1510                         mlog_errno(status);
1511                         goto bail;
1512                 }
1513
1514                 /* change the dirent to point to the correct inode */
1515                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1516                                             old_inode);
1517                 if (status < 0) {
1518                         mlog_errno(status);
1519                         goto bail;
1520                 }
1521                 inode_inc_iversion(new_dir);
1522
1523                 if (S_ISDIR(new_inode->i_mode))
1524                         ocfs2_set_links_count(newfe, 0);
1525                 else
1526                         ocfs2_add_links_count(newfe, -1);
1527                 ocfs2_journal_dirty(handle, newfe_bh);
1528                 if (should_add_orphan) {
1529                         status = ocfs2_orphan_add(osb, handle, new_inode,
1530                                         newfe_bh, orphan_name,
1531                                         &orphan_insert, orphan_dir, false);
1532                         if (status < 0) {
1533                                 mlog_errno(status);
1534                                 goto bail;
1535                         }
1536                 }
1537         } else {
1538                 /* if the name was not found in new_dir, add it now */
1539                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1540                                          OCFS2_I(old_inode)->ip_blkno,
1541                                          new_dir_bh, &target_insert);
1542         }
1543
1544         old_inode->i_ctime = current_time(old_inode);
1545         mark_inode_dirty(old_inode);
1546
1547         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1548                                          old_inode_bh,
1549                                          OCFS2_JOURNAL_ACCESS_WRITE);
1550         if (status >= 0) {
1551                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1552
1553                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1554                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1555                 ocfs2_journal_dirty(handle, old_inode_bh);
1556         } else
1557                 mlog_errno(status);
1558
1559         /*
1560          * Now that the name has been added to new_dir, remove the old name.
1561          *
1562          * We don't keep any directory entry context around until now
1563          * because the insert might have changed the type of directory
1564          * we're dealing with.
1565          */
1566         status = ocfs2_find_entry(old_dentry->d_name.name,
1567                                   old_dentry->d_name.len, old_dir,
1568                                   &old_entry_lookup);
1569         if (status) {
1570                 if (!is_journal_aborted(osb->journal->j_journal)) {
1571                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1572                                         "is not deleted.",
1573                                         new_dentry->d_name.len, new_dentry->d_name.name,
1574                                         old_dentry->d_name.len, old_dentry->d_name.name);
1575                 }
1576                 goto bail;
1577         }
1578
1579         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1580         if (status < 0) {
1581                 mlog_errno(status);
1582                 if (!is_journal_aborted(osb->journal->j_journal)) {
1583                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1584                                         "is not deleted.",
1585                                         new_dentry->d_name.len, new_dentry->d_name.name,
1586                                         old_dentry->d_name.len, old_dentry->d_name.name);
1587                 }
1588                 goto bail;
1589         }
1590
1591         if (new_inode) {
1592                 drop_nlink(new_inode);
1593                 new_inode->i_ctime = current_time(new_inode);
1594         }
1595         old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1596
1597         if (update_dot_dot) {
1598                 status = ocfs2_update_entry(old_inode, handle,
1599                                             &old_inode_dot_dot_res, new_dir);
1600                 drop_nlink(old_dir);
1601                 if (new_inode) {
1602                         drop_nlink(new_inode);
1603                 } else {
1604                         inc_nlink(new_dir);
1605                         mark_inode_dirty(new_dir);
1606                 }
1607         }
1608         mark_inode_dirty(old_dir);
1609         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1610         if (new_inode) {
1611                 mark_inode_dirty(new_inode);
1612                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1613         }
1614
1615         if (old_dir != new_dir) {
1616                 /* Keep the same times on both directories.*/
1617                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1618
1619                 /*
1620                  * This will also pick up the i_nlink change from the
1621                  * block above.
1622                  */
1623                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1624         }
1625
1626         if (old_dir_nlink != old_dir->i_nlink) {
1627                 if (!old_dir_bh) {
1628                         mlog(ML_ERROR, "need to change nlink for old dir "
1629                              "%llu from %d to %d but bh is NULL!\n",
1630                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1631                              (int)old_dir_nlink, old_dir->i_nlink);
1632                 } else {
1633                         struct ocfs2_dinode *fe;
1634                         status = ocfs2_journal_access_di(handle,
1635                                                          INODE_CACHE(old_dir),
1636                                                          old_dir_bh,
1637                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1638                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1639                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1640                         ocfs2_journal_dirty(handle, old_dir_bh);
1641                 }
1642         }
1643         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1644         status = 0;
1645 bail:
1646         if (handle)
1647                 ocfs2_commit_trans(osb, handle);
1648
1649         if (orphan_dir) {
1650                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1651                 ocfs2_inode_unlock(orphan_dir, 1);
1652                 inode_unlock(orphan_dir);
1653                 iput(orphan_dir);
1654         }
1655
1656         if (new_child_locked)
1657                 ocfs2_inode_unlock(new_inode, 1);
1658
1659         if (old_child_locked)
1660                 ocfs2_inode_unlock(old_inode, 1);
1661
1662         if (parents_locked)
1663                 ocfs2_double_unlock(old_dir, new_dir);
1664
1665         if (rename_lock)
1666                 ocfs2_rename_unlock(osb);
1667
1668         if (new_inode)
1669                 sync_mapping_buffers(old_inode->i_mapping);
1670
1671         iput(new_inode);
1672
1673         ocfs2_free_dir_lookup_result(&target_lookup_res);
1674         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1675         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1676         ocfs2_free_dir_lookup_result(&orphan_insert);
1677         ocfs2_free_dir_lookup_result(&target_insert);
1678
1679         brelse(newfe_bh);
1680         brelse(old_inode_bh);
1681         brelse(old_dir_bh);
1682         brelse(new_dir_bh);
1683
1684         if (status)
1685                 mlog_errno(status);
1686
1687         return status;
1688 }
1689
1690 /*
1691  * we expect i_size = strlen(symname). Copy symname into the file
1692  * data, including the null terminator.
1693  */
1694 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1695                                      handle_t *handle,
1696                                      struct inode *inode,
1697                                      const char *symname)
1698 {
1699         struct buffer_head **bhs = NULL;
1700         const char *c;
1701         struct super_block *sb = osb->sb;
1702         u64 p_blkno, p_blocks;
1703         int virtual, blocks, status, i, bytes_left;
1704
1705         bytes_left = i_size_read(inode) + 1;
1706         /* we can't trust i_blocks because we're actually going to
1707          * write i_size + 1 bytes. */
1708         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1709
1710         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1711                                         i_size_read(inode), blocks);
1712
1713         /* Sanity check -- make sure we're going to fit. */
1714         if (bytes_left >
1715             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1716                 status = -EIO;
1717                 mlog_errno(status);
1718                 goto bail;
1719         }
1720
1721         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1722         if (!bhs) {
1723                 status = -ENOMEM;
1724                 mlog_errno(status);
1725                 goto bail;
1726         }
1727
1728         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1729                                              NULL);
1730         if (status < 0) {
1731                 mlog_errno(status);
1732                 goto bail;
1733         }
1734
1735         /* links can never be larger than one cluster so we know this
1736          * is all going to be contiguous, but do a sanity check
1737          * anyway. */
1738         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1739                 status = -EIO;
1740                 mlog_errno(status);
1741                 goto bail;
1742         }
1743
1744         virtual = 0;
1745         while(bytes_left > 0) {
1746                 c = &symname[virtual * sb->s_blocksize];
1747
1748                 bhs[virtual] = sb_getblk(sb, p_blkno);
1749                 if (!bhs[virtual]) {
1750                         status = -ENOMEM;
1751                         mlog_errno(status);
1752                         goto bail;
1753                 }
1754                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1755                                               bhs[virtual]);
1756
1757                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1758                                               bhs[virtual],
1759                                               OCFS2_JOURNAL_ACCESS_CREATE);
1760                 if (status < 0) {
1761                         mlog_errno(status);
1762                         goto bail;
1763                 }
1764
1765                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1766
1767                 memcpy(bhs[virtual]->b_data, c,
1768                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1769                        bytes_left);
1770
1771                 ocfs2_journal_dirty(handle, bhs[virtual]);
1772
1773                 virtual++;
1774                 p_blkno++;
1775                 bytes_left -= sb->s_blocksize;
1776         }
1777
1778         status = 0;
1779 bail:
1780
1781         if (bhs) {
1782                 for(i = 0; i < blocks; i++)
1783                         brelse(bhs[i]);
1784                 kfree(bhs);
1785         }
1786
1787         if (status)
1788                 mlog_errno(status);
1789         return status;
1790 }
1791
1792 static int ocfs2_symlink(struct user_namespace *mnt_userns,
1793                          struct inode *dir,
1794                          struct dentry *dentry,
1795                          const char *symname)
1796 {
1797         int status, l, credits;
1798         u64 newsize;
1799         struct ocfs2_super *osb = NULL;
1800         struct inode *inode = NULL;
1801         struct super_block *sb;
1802         struct buffer_head *new_fe_bh = NULL;
1803         struct buffer_head *parent_fe_bh = NULL;
1804         struct ocfs2_dinode *fe = NULL;
1805         struct ocfs2_dinode *dirfe;
1806         handle_t *handle = NULL;
1807         struct ocfs2_alloc_context *inode_ac = NULL;
1808         struct ocfs2_alloc_context *data_ac = NULL;
1809         struct ocfs2_alloc_context *xattr_ac = NULL;
1810         int want_clusters = 0;
1811         int xattr_credits = 0;
1812         struct ocfs2_security_xattr_info si = {
1813                 .enable = 1,
1814         };
1815         int did_quota = 0, did_quota_inode = 0;
1816         struct ocfs2_dir_lookup_result lookup = { NULL, };
1817         sigset_t oldset;
1818         int did_block_signals = 0;
1819         struct ocfs2_dentry_lock *dl = NULL;
1820
1821         trace_ocfs2_symlink_begin(dir, dentry, symname,
1822                                   dentry->d_name.len, dentry->d_name.name);
1823
1824         status = dquot_initialize(dir);
1825         if (status) {
1826                 mlog_errno(status);
1827                 goto bail;
1828         }
1829
1830         sb = dir->i_sb;
1831         osb = OCFS2_SB(sb);
1832
1833         l = strlen(symname) + 1;
1834
1835         credits = ocfs2_calc_symlink_credits(sb);
1836
1837         /* lock the parent directory */
1838         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1839         if (status < 0) {
1840                 if (status != -ENOENT)
1841                         mlog_errno(status);
1842                 return status;
1843         }
1844
1845         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1846         if (!ocfs2_read_links_count(dirfe)) {
1847                 /* can't make a file in a deleted directory. */
1848                 status = -ENOENT;
1849                 goto bail;
1850         }
1851
1852         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1853                                            dentry->d_name.len);
1854         if (status)
1855                 goto bail;
1856
1857         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1858                                               dentry->d_name.name,
1859                                               dentry->d_name.len, &lookup);
1860         if (status < 0) {
1861                 mlog_errno(status);
1862                 goto bail;
1863         }
1864
1865         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1866         if (status < 0) {
1867                 if (status != -ENOSPC)
1868                         mlog_errno(status);
1869                 goto bail;
1870         }
1871
1872         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1873         if (IS_ERR(inode)) {
1874                 status = PTR_ERR(inode);
1875                 inode = NULL;
1876                 mlog_errno(status);
1877                 goto bail;
1878         }
1879
1880         /* get security xattr */
1881         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1882         if (status) {
1883                 if (status == -EOPNOTSUPP)
1884                         si.enable = 0;
1885                 else {
1886                         mlog_errno(status);
1887                         goto bail;
1888                 }
1889         }
1890
1891         /* calculate meta data/clusters for setting security xattr */
1892         if (si.enable) {
1893                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1894                                                   &xattr_credits, &xattr_ac);
1895                 if (status < 0) {
1896                         mlog_errno(status);
1897                         goto bail;
1898                 }
1899         }
1900
1901         /* don't reserve bitmap space for fast symlinks. */
1902         if (l > ocfs2_fast_symlink_chars(sb))
1903                 want_clusters += 1;
1904
1905         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1906         if (status < 0) {
1907                 if (status != -ENOSPC)
1908                         mlog_errno(status);
1909                 goto bail;
1910         }
1911
1912         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1913         if (IS_ERR(handle)) {
1914                 status = PTR_ERR(handle);
1915                 handle = NULL;
1916                 mlog_errno(status);
1917                 goto bail;
1918         }
1919
1920         /* Starting to change things, restart is no longer possible. */
1921         ocfs2_block_signals(&oldset);
1922         did_block_signals = 1;
1923
1924         status = dquot_alloc_inode(inode);
1925         if (status)
1926                 goto bail;
1927         did_quota_inode = 1;
1928
1929         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1930                                    dentry->d_name.name,
1931                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1932                                    inode->i_mode);
1933
1934         status = ocfs2_mknod_locked(osb, dir, inode,
1935                                     0, &new_fe_bh, parent_fe_bh, handle,
1936                                     inode_ac);
1937         if (status < 0) {
1938                 mlog_errno(status);
1939                 goto bail;
1940         }
1941
1942         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1943         inode->i_rdev = 0;
1944         newsize = l - 1;
1945         inode->i_op = &ocfs2_symlink_inode_operations;
1946         inode_nohighmem(inode);
1947         if (l > ocfs2_fast_symlink_chars(sb)) {
1948                 u32 offset = 0;
1949
1950                 status = dquot_alloc_space_nodirty(inode,
1951                     ocfs2_clusters_to_bytes(osb->sb, 1));
1952                 if (status)
1953                         goto bail;
1954                 did_quota = 1;
1955                 inode->i_mapping->a_ops = &ocfs2_aops;
1956                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1957                                               new_fe_bh,
1958                                               handle, data_ac, NULL,
1959                                               NULL);
1960                 if (status < 0) {
1961                         if (status != -ENOSPC && status != -EINTR) {
1962                                 mlog(ML_ERROR,
1963                                      "Failed to extend file to %llu\n",
1964                                      (unsigned long long)newsize);
1965                                 mlog_errno(status);
1966                                 status = -ENOSPC;
1967                         }
1968                         goto bail;
1969                 }
1970                 i_size_write(inode, newsize);
1971                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1972         } else {
1973                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1974                 memcpy((char *) fe->id2.i_symlink, symname, l);
1975                 i_size_write(inode, newsize);
1976                 inode->i_blocks = 0;
1977         }
1978
1979         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1980         if (status < 0) {
1981                 mlog_errno(status);
1982                 goto bail;
1983         }
1984
1985         if (!ocfs2_inode_is_fast_symlink(inode)) {
1986                 status = ocfs2_create_symlink_data(osb, handle, inode,
1987                                                    symname);
1988                 if (status < 0) {
1989                         mlog_errno(status);
1990                         goto bail;
1991                 }
1992         }
1993
1994         if (si.enable) {
1995                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1996                                                  xattr_ac, data_ac);
1997                 if (status < 0) {
1998                         mlog_errno(status);
1999                         goto bail;
2000                 }
2001         }
2002
2003         /*
2004          * Do this before adding the entry to the directory. We add
2005          * also set d_op after success so that ->d_iput() will cleanup
2006          * the dentry lock even if ocfs2_add_entry() fails below.
2007          */
2008         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2009         if (status) {
2010                 mlog_errno(status);
2011                 goto bail;
2012         }
2013
2014         dl = dentry->d_fsdata;
2015
2016         status = ocfs2_add_entry(handle, dentry, inode,
2017                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2018                                  &lookup);
2019         if (status < 0) {
2020                 mlog_errno(status);
2021                 goto bail;
2022         }
2023
2024         insert_inode_hash(inode);
2025         d_instantiate(dentry, inode);
2026 bail:
2027         if (status < 0 && did_quota)
2028                 dquot_free_space_nodirty(inode,
2029                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2030         if (status < 0 && did_quota_inode)
2031                 dquot_free_inode(inode);
2032         if (handle)
2033                 ocfs2_commit_trans(osb, handle);
2034
2035         ocfs2_inode_unlock(dir, 1);
2036         if (did_block_signals)
2037                 ocfs2_unblock_signals(&oldset);
2038
2039         brelse(new_fe_bh);
2040         brelse(parent_fe_bh);
2041         kfree(si.value);
2042         ocfs2_free_dir_lookup_result(&lookup);
2043         if (inode_ac)
2044                 ocfs2_free_alloc_context(inode_ac);
2045         if (data_ac)
2046                 ocfs2_free_alloc_context(data_ac);
2047         if (xattr_ac)
2048                 ocfs2_free_alloc_context(xattr_ac);
2049         if ((status < 0) && inode) {
2050                 if (dl)
2051                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2052
2053                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2054                 clear_nlink(inode);
2055                 iput(inode);
2056         }
2057
2058         if (status)
2059                 mlog_errno(status);
2060
2061         return status;
2062 }
2063
2064 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2065 {
2066         int status, namelen;
2067
2068         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2069                            (long long)blkno);
2070         if (namelen <= 0) {
2071                 if (namelen)
2072                         status = namelen;
2073                 else
2074                         status = -EINVAL;
2075                 mlog_errno(status);
2076                 goto bail;
2077         }
2078         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2079                 status = -EINVAL;
2080                 mlog_errno(status);
2081                 goto bail;
2082         }
2083
2084         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2085
2086         status = 0;
2087 bail:
2088         if (status < 0)
2089                 mlog_errno(status);
2090         return status;
2091 }
2092
2093 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2094                                         struct inode **ret_orphan_dir,
2095                                         struct buffer_head **ret_orphan_dir_bh)
2096 {
2097         struct inode *orphan_dir_inode;
2098         struct buffer_head *orphan_dir_bh = NULL;
2099         int ret = 0;
2100
2101         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2102                                                        ORPHAN_DIR_SYSTEM_INODE,
2103                                                        osb->slot_num);
2104         if (!orphan_dir_inode) {
2105                 ret = -ENOENT;
2106                 mlog_errno(ret);
2107                 return ret;
2108         }
2109
2110         inode_lock(orphan_dir_inode);
2111
2112         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2113         if (ret < 0) {
2114                 inode_unlock(orphan_dir_inode);
2115                 iput(orphan_dir_inode);
2116
2117                 mlog_errno(ret);
2118                 return ret;
2119         }
2120
2121         *ret_orphan_dir = orphan_dir_inode;
2122         *ret_orphan_dir_bh = orphan_dir_bh;
2123
2124         return 0;
2125 }
2126
2127 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2128                                       struct buffer_head *orphan_dir_bh,
2129                                       u64 blkno,
2130                                       char *name,
2131                                       struct ocfs2_dir_lookup_result *lookup,
2132                                       bool dio)
2133 {
2134         int ret;
2135         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2136         int namelen = dio ?
2137                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2138                         OCFS2_ORPHAN_NAMELEN;
2139
2140         if (dio) {
2141                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2142                                 OCFS2_DIO_ORPHAN_PREFIX);
2143                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2144                         ret = -EINVAL;
2145                         mlog_errno(ret);
2146                         return ret;
2147                 }
2148
2149                 ret = ocfs2_blkno_stringify(blkno,
2150                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2151         } else
2152                 ret = ocfs2_blkno_stringify(blkno, name);
2153         if (ret < 0) {
2154                 mlog_errno(ret);
2155                 return ret;
2156         }
2157
2158         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2159                                            orphan_dir_bh, name,
2160                                            namelen, lookup);
2161         if (ret < 0) {
2162                 mlog_errno(ret);
2163                 return ret;
2164         }
2165
2166         return 0;
2167 }
2168
2169 /**
2170  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2171  * insertion of an orphan.
2172  * @osb: ocfs2 file system
2173  * @ret_orphan_dir: Orphan dir inode - returned locked!
2174  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2175  * @lookup: dir lookup result, to be passed back into functions like
2176  *          ocfs2_orphan_add
2177  *
2178  * Returns zero on success and the ret_orphan_dir, name and lookup
2179  * fields will be populated.
2180  *
2181  * Returns non-zero on failure.
2182  */
2183 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2184                                     struct inode **ret_orphan_dir,
2185                                     u64 blkno,
2186                                     char *name,
2187                                     struct ocfs2_dir_lookup_result *lookup,
2188                                     bool dio)
2189 {
2190         struct inode *orphan_dir_inode = NULL;
2191         struct buffer_head *orphan_dir_bh = NULL;
2192         int ret = 0;
2193
2194         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2195                                            &orphan_dir_bh);
2196         if (ret < 0) {
2197                 mlog_errno(ret);
2198                 return ret;
2199         }
2200
2201         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2202                                          blkno, name, lookup, dio);
2203         if (ret < 0) {
2204                 mlog_errno(ret);
2205                 goto out;
2206         }
2207
2208         *ret_orphan_dir = orphan_dir_inode;
2209
2210 out:
2211         brelse(orphan_dir_bh);
2212
2213         if (ret) {
2214                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2215                 inode_unlock(orphan_dir_inode);
2216                 iput(orphan_dir_inode);
2217         }
2218
2219         if (ret)
2220                 mlog_errno(ret);
2221         return ret;
2222 }
2223
2224 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2225                             handle_t *handle,
2226                             struct inode *inode,
2227                             struct buffer_head *fe_bh,
2228                             char *name,
2229                             struct ocfs2_dir_lookup_result *lookup,
2230                             struct inode *orphan_dir_inode,
2231                             bool dio)
2232 {
2233         struct buffer_head *orphan_dir_bh = NULL;
2234         int status = 0;
2235         struct ocfs2_dinode *orphan_fe;
2236         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2237         int namelen = dio ?
2238                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2239                         OCFS2_ORPHAN_NAMELEN;
2240
2241         trace_ocfs2_orphan_add_begin(
2242                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2243
2244         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2245         if (status < 0) {
2246                 mlog_errno(status);
2247                 goto leave;
2248         }
2249
2250         status = ocfs2_journal_access_di(handle,
2251                                          INODE_CACHE(orphan_dir_inode),
2252                                          orphan_dir_bh,
2253                                          OCFS2_JOURNAL_ACCESS_WRITE);
2254         if (status < 0) {
2255                 mlog_errno(status);
2256                 goto leave;
2257         }
2258
2259         /*
2260          * We're going to journal the change of i_flags and i_orphaned_slot.
2261          * It's safe anyway, though some callers may duplicate the journaling.
2262          * Journaling within the func just make the logic look more
2263          * straightforward.
2264          */
2265         status = ocfs2_journal_access_di(handle,
2266                                          INODE_CACHE(inode),
2267                                          fe_bh,
2268                                          OCFS2_JOURNAL_ACCESS_WRITE);
2269         if (status < 0) {
2270                 mlog_errno(status);
2271                 goto leave;
2272         }
2273
2274         /* we're a cluster, and nlink can change on disk from
2275          * underneath us... */
2276         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2277         if (S_ISDIR(inode->i_mode))
2278                 ocfs2_add_links_count(orphan_fe, 1);
2279         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2280         ocfs2_journal_dirty(handle, orphan_dir_bh);
2281
2282         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2283                                    namelen, inode,
2284                                    OCFS2_I(inode)->ip_blkno,
2285                                    orphan_dir_bh, lookup);
2286         if (status < 0) {
2287                 mlog_errno(status);
2288                 goto rollback;
2289         }
2290
2291         if (dio) {
2292                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2293                  * slot.
2294                  */
2295                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2296                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2297         } else {
2298                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2299                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2300
2301                 /* Record which orphan dir our inode now resides
2302                  * in. delete_inode will use this to determine which orphan
2303                  * dir to lock. */
2304                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2305         }
2306
2307         ocfs2_journal_dirty(handle, fe_bh);
2308
2309         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2310                                    osb->slot_num);
2311
2312 rollback:
2313         if (status < 0) {
2314                 if (S_ISDIR(inode->i_mode))
2315                         ocfs2_add_links_count(orphan_fe, -1);
2316                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2317         }
2318
2319 leave:
2320         brelse(orphan_dir_bh);
2321
2322         return status;
2323 }
2324
2325 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2326 int ocfs2_orphan_del(struct ocfs2_super *osb,
2327                      handle_t *handle,
2328                      struct inode *orphan_dir_inode,
2329                      struct inode *inode,
2330                      struct buffer_head *orphan_dir_bh,
2331                      bool dio)
2332 {
2333         char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2334         struct ocfs2_dinode *orphan_fe;
2335         int status = 0;
2336         struct ocfs2_dir_lookup_result lookup = { NULL, };
2337
2338         if (dio) {
2339                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2340                                 OCFS2_DIO_ORPHAN_PREFIX);
2341                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2342                         status = -EINVAL;
2343                         mlog_errno(status);
2344                         return status;
2345                 }
2346
2347                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2348                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2349         } else
2350                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2351         if (status < 0) {
2352                 mlog_errno(status);
2353                 goto leave;
2354         }
2355
2356         trace_ocfs2_orphan_del(
2357              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2358              name, strlen(name));
2359
2360         status = ocfs2_journal_access_di(handle,
2361                                          INODE_CACHE(orphan_dir_inode),
2362                                          orphan_dir_bh,
2363                                          OCFS2_JOURNAL_ACCESS_WRITE);
2364         if (status < 0) {
2365                 mlog_errno(status);
2366                 goto leave;
2367         }
2368
2369         /* find it's spot in the orphan directory */
2370         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2371                                   &lookup);
2372         if (status) {
2373                 mlog_errno(status);
2374                 goto leave;
2375         }
2376
2377         /* remove it from the orphan directory */
2378         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2379         if (status < 0) {
2380                 mlog_errno(status);
2381                 goto leave;
2382         }
2383
2384         /* do the i_nlink dance! :) */
2385         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2386         if (S_ISDIR(inode->i_mode))
2387                 ocfs2_add_links_count(orphan_fe, -1);
2388         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2389         ocfs2_journal_dirty(handle, orphan_dir_bh);
2390
2391 leave:
2392         ocfs2_free_dir_lookup_result(&lookup);
2393
2394         if (status)
2395                 mlog_errno(status);
2396         return status;
2397 }
2398
2399 /**
2400  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2401  * allocated file. This is different from the typical 'add to orphan dir'
2402  * operation in that the inode does not yet exist. This is a problem because
2403  * the orphan dir stringifies the inode block number to come up with it's
2404  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2405  * problem. This function works around it by calling deeper into the orphan
2406  * and suballoc code than other callers. Use this only by necessity.
2407  * @dir: The directory which this inode will ultimately wind up under - not the
2408  * orphan dir!
2409  * @dir_bh: buffer_head the @dir inode block
2410  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2411  * with the string to be used for orphan dirent. Pass back to the orphan dir
2412  * code.
2413  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2414  * dir code.
2415  * @ret_di_blkno: block number where the new inode will be allocated.
2416  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2417  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2418  *
2419  * Returns zero on success and the ret_orphan_dir, name and lookup
2420  * fields will be populated.
2421  *
2422  * Returns non-zero on failure.
2423  */
2424 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2425                                         struct buffer_head *dir_bh,
2426                                         char *orphan_name,
2427                                         struct inode **ret_orphan_dir,
2428                                         u64 *ret_di_blkno,
2429                                         struct ocfs2_dir_lookup_result *orphan_insert,
2430                                         struct ocfs2_alloc_context **ret_inode_ac)
2431 {
2432         int ret;
2433         u64 di_blkno;
2434         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2435         struct inode *orphan_dir = NULL;
2436         struct buffer_head *orphan_dir_bh = NULL;
2437         struct ocfs2_alloc_context *inode_ac = NULL;
2438
2439         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2440         if (ret < 0) {
2441                 mlog_errno(ret);
2442                 return ret;
2443         }
2444
2445         /* reserve an inode spot */
2446         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2447         if (ret < 0) {
2448                 if (ret != -ENOSPC)
2449                         mlog_errno(ret);
2450                 goto out;
2451         }
2452
2453         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2454                                        &di_blkno);
2455         if (ret) {
2456                 mlog_errno(ret);
2457                 goto out;
2458         }
2459
2460         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2461                                          di_blkno, orphan_name, orphan_insert,
2462                                          false);
2463         if (ret < 0) {
2464                 mlog_errno(ret);
2465                 goto out;
2466         }
2467
2468 out:
2469         if (ret == 0) {
2470                 *ret_orphan_dir = orphan_dir;
2471                 *ret_di_blkno = di_blkno;
2472                 *ret_inode_ac = inode_ac;
2473                 /*
2474                  * orphan_name and orphan_insert are already up to
2475                  * date via prepare_orphan_dir
2476                  */
2477         } else {
2478                 /* Unroll reserve_new_inode* */
2479                 if (inode_ac)
2480                         ocfs2_free_alloc_context(inode_ac);
2481
2482                 /* Unroll orphan dir locking */
2483                 inode_unlock(orphan_dir);
2484                 ocfs2_inode_unlock(orphan_dir, 1);
2485                 iput(orphan_dir);
2486         }
2487
2488         brelse(orphan_dir_bh);
2489
2490         return ret;
2491 }
2492
2493 int ocfs2_create_inode_in_orphan(struct inode *dir,
2494                                  int mode,
2495                                  struct inode **new_inode)
2496 {
2497         int status, did_quota_inode = 0;
2498         struct inode *inode = NULL;
2499         struct inode *orphan_dir = NULL;
2500         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2501         handle_t *handle = NULL;
2502         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2503         struct buffer_head *parent_di_bh = NULL;
2504         struct buffer_head *new_di_bh = NULL;
2505         struct ocfs2_alloc_context *inode_ac = NULL;
2506         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2507         u64 di_blkno, suballoc_loc;
2508         u16 suballoc_bit;
2509
2510         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2511         if (status < 0) {
2512                 if (status != -ENOENT)
2513                         mlog_errno(status);
2514                 return status;
2515         }
2516
2517         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2518                                               orphan_name, &orphan_dir,
2519                                               &di_blkno, &orphan_insert, &inode_ac);
2520         if (status < 0) {
2521                 if (status != -ENOSPC)
2522                         mlog_errno(status);
2523                 goto leave;
2524         }
2525
2526         inode = ocfs2_get_init_inode(dir, mode);
2527         if (IS_ERR(inode)) {
2528                 status = PTR_ERR(inode);
2529                 inode = NULL;
2530                 mlog_errno(status);
2531                 goto leave;
2532         }
2533
2534         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2535         if (IS_ERR(handle)) {
2536                 status = PTR_ERR(handle);
2537                 handle = NULL;
2538                 mlog_errno(status);
2539                 goto leave;
2540         }
2541
2542         status = dquot_alloc_inode(inode);
2543         if (status)
2544                 goto leave;
2545         did_quota_inode = 1;
2546
2547         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2548                                               &suballoc_loc,
2549                                               &suballoc_bit, di_blkno);
2550         if (status < 0) {
2551                 mlog_errno(status);
2552                 goto leave;
2553         }
2554
2555         clear_nlink(inode);
2556         /* do the real work now. */
2557         status = __ocfs2_mknod_locked(dir, inode,
2558                                       0, &new_di_bh, parent_di_bh, handle,
2559                                       inode_ac, di_blkno, suballoc_loc,
2560                                       suballoc_bit);
2561         if (status < 0) {
2562                 mlog_errno(status);
2563                 goto leave;
2564         }
2565
2566         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2567                                   &orphan_insert, orphan_dir, false);
2568         if (status < 0) {
2569                 mlog_errno(status);
2570                 goto leave;
2571         }
2572
2573         /* get open lock so that only nodes can't remove it from orphan dir. */
2574         status = ocfs2_open_lock(inode);
2575         if (status < 0)
2576                 mlog_errno(status);
2577
2578         insert_inode_hash(inode);
2579 leave:
2580         if (status < 0 && did_quota_inode)
2581                 dquot_free_inode(inode);
2582         if (handle)
2583                 ocfs2_commit_trans(osb, handle);
2584
2585         if (orphan_dir) {
2586                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2587                 ocfs2_inode_unlock(orphan_dir, 1);
2588                 inode_unlock(orphan_dir);
2589                 iput(orphan_dir);
2590         }
2591
2592         if ((status < 0) && inode) {
2593                 clear_nlink(inode);
2594                 iput(inode);
2595         }
2596
2597         if (inode_ac)
2598                 ocfs2_free_alloc_context(inode_ac);
2599
2600         brelse(new_di_bh);
2601
2602         if (!status)
2603                 *new_inode = inode;
2604
2605         ocfs2_free_dir_lookup_result(&orphan_insert);
2606
2607         ocfs2_inode_unlock(dir, 1);
2608         brelse(parent_di_bh);
2609         return status;
2610 }
2611
2612 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2613         struct inode *inode)
2614 {
2615         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2616         struct inode *orphan_dir_inode = NULL;
2617         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2618         struct buffer_head *di_bh = NULL;
2619         int status = 0;
2620         handle_t *handle = NULL;
2621         struct ocfs2_dinode *di = NULL;
2622
2623         status = ocfs2_inode_lock(inode, &di_bh, 1);
2624         if (status < 0) {
2625                 mlog_errno(status);
2626                 goto bail;
2627         }
2628
2629         di = (struct ocfs2_dinode *) di_bh->b_data;
2630         /*
2631          * Another append dio crashed?
2632          * If so, manually recover it first.
2633          */
2634         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2635                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2636                 if (status < 0) {
2637                         if (status != -ENOSPC)
2638                                 mlog_errno(status);
2639                         goto bail_unlock_inode;
2640                 }
2641
2642                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2643                 if (status < 0) {
2644                         mlog_errno(status);
2645                         goto bail_unlock_inode;
2646                 }
2647         }
2648
2649         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2650                         OCFS2_I(inode)->ip_blkno,
2651                         orphan_name,
2652                         &orphan_insert,
2653                         true);
2654         if (status < 0) {
2655                 mlog_errno(status);
2656                 goto bail_unlock_inode;
2657         }
2658
2659         handle = ocfs2_start_trans(osb,
2660                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2661         if (IS_ERR(handle)) {
2662                 status = PTR_ERR(handle);
2663                 goto bail_unlock_orphan;
2664         }
2665
2666         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2667                         &orphan_insert, orphan_dir_inode, true);
2668         if (status)
2669                 mlog_errno(status);
2670
2671         ocfs2_commit_trans(osb, handle);
2672
2673 bail_unlock_orphan:
2674         ocfs2_inode_unlock(orphan_dir_inode, 1);
2675         inode_unlock(orphan_dir_inode);
2676         iput(orphan_dir_inode);
2677
2678         ocfs2_free_dir_lookup_result(&orphan_insert);
2679
2680 bail_unlock_inode:
2681         ocfs2_inode_unlock(inode, 1);
2682         brelse(di_bh);
2683
2684 bail:
2685         return status;
2686 }
2687
2688 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2689                 struct inode *inode, struct buffer_head *di_bh,
2690                 int update_isize, loff_t end)
2691 {
2692         struct inode *orphan_dir_inode = NULL;
2693         struct buffer_head *orphan_dir_bh = NULL;
2694         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2695         handle_t *handle = NULL;
2696         int status = 0;
2697
2698         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2699                         ORPHAN_DIR_SYSTEM_INODE,
2700                         le16_to_cpu(di->i_dio_orphaned_slot));
2701         if (!orphan_dir_inode) {
2702                 status = -ENOENT;
2703                 mlog_errno(status);
2704                 goto bail;
2705         }
2706
2707         inode_lock(orphan_dir_inode);
2708         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2709         if (status < 0) {
2710                 inode_unlock(orphan_dir_inode);
2711                 iput(orphan_dir_inode);
2712                 mlog_errno(status);
2713                 goto bail;
2714         }
2715
2716         handle = ocfs2_start_trans(osb,
2717                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2718         if (IS_ERR(handle)) {
2719                 status = PTR_ERR(handle);
2720                 goto bail_unlock_orphan;
2721         }
2722
2723         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2724
2725         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2726                                 inode, orphan_dir_bh, true);
2727         if (status < 0) {
2728                 mlog_errno(status);
2729                 goto bail_commit;
2730         }
2731
2732         status = ocfs2_journal_access_di(handle,
2733                         INODE_CACHE(inode),
2734                         di_bh,
2735                         OCFS2_JOURNAL_ACCESS_WRITE);
2736         if (status < 0) {
2737                 mlog_errno(status);
2738                 goto bail_commit;
2739         }
2740
2741         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2742         di->i_dio_orphaned_slot = 0;
2743
2744         if (update_isize) {
2745                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2746                 if (status)
2747                         mlog_errno(status);
2748         } else
2749                 ocfs2_journal_dirty(handle, di_bh);
2750
2751 bail_commit:
2752         ocfs2_commit_trans(osb, handle);
2753
2754 bail_unlock_orphan:
2755         ocfs2_inode_unlock(orphan_dir_inode, 1);
2756         inode_unlock(orphan_dir_inode);
2757         brelse(orphan_dir_bh);
2758         iput(orphan_dir_inode);
2759
2760 bail:
2761         return status;
2762 }
2763
2764 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2765                                    struct inode *inode,
2766                                    struct dentry *dentry)
2767 {
2768         int status = 0;
2769         struct buffer_head *parent_di_bh = NULL;
2770         handle_t *handle = NULL;
2771         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2772         struct ocfs2_dinode *dir_di, *di;
2773         struct inode *orphan_dir_inode = NULL;
2774         struct buffer_head *orphan_dir_bh = NULL;
2775         struct buffer_head *di_bh = NULL;
2776         struct ocfs2_dir_lookup_result lookup = { NULL, };
2777
2778         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2779                                 dentry->d_name.len, dentry->d_name.name,
2780                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2781                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2782
2783         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2784         if (status < 0) {
2785                 if (status != -ENOENT)
2786                         mlog_errno(status);
2787                 return status;
2788         }
2789
2790         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2791         if (!dir_di->i_links_count) {
2792                 /* can't make a file in a deleted directory. */
2793                 status = -ENOENT;
2794                 goto leave;
2795         }
2796
2797         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2798                                            dentry->d_name.len);
2799         if (status)
2800                 goto leave;
2801
2802         /* get a spot inside the dir. */
2803         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2804                                               dentry->d_name.name,
2805                                               dentry->d_name.len, &lookup);
2806         if (status < 0) {
2807                 mlog_errno(status);
2808                 goto leave;
2809         }
2810
2811         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2812                                                        ORPHAN_DIR_SYSTEM_INODE,
2813                                                        osb->slot_num);
2814         if (!orphan_dir_inode) {
2815                 status = -ENOENT;
2816                 mlog_errno(status);
2817                 goto leave;
2818         }
2819
2820         inode_lock(orphan_dir_inode);
2821
2822         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2823         if (status < 0) {
2824                 mlog_errno(status);
2825                 inode_unlock(orphan_dir_inode);
2826                 iput(orphan_dir_inode);
2827                 goto leave;
2828         }
2829
2830         status = ocfs2_read_inode_block(inode, &di_bh);
2831         if (status < 0) {
2832                 mlog_errno(status);
2833                 goto orphan_unlock;
2834         }
2835
2836         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2837         if (IS_ERR(handle)) {
2838                 status = PTR_ERR(handle);
2839                 handle = NULL;
2840                 mlog_errno(status);
2841                 goto orphan_unlock;
2842         }
2843
2844         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2845                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2846         if (status < 0) {
2847                 mlog_errno(status);
2848                 goto out_commit;
2849         }
2850
2851         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2852                                   orphan_dir_bh, false);
2853         if (status < 0) {
2854                 mlog_errno(status);
2855                 goto out_commit;
2856         }
2857
2858         di = (struct ocfs2_dinode *)di_bh->b_data;
2859         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2860         di->i_orphaned_slot = 0;
2861         set_nlink(inode, 1);
2862         ocfs2_set_links_count(di, inode->i_nlink);
2863         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2864         ocfs2_journal_dirty(handle, di_bh);
2865
2866         status = ocfs2_add_entry(handle, dentry, inode,
2867                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2868                                  &lookup);
2869         if (status < 0) {
2870                 mlog_errno(status);
2871                 goto out_commit;
2872         }
2873
2874         status = ocfs2_dentry_attach_lock(dentry, inode,
2875                                           OCFS2_I(dir)->ip_blkno);
2876         if (status) {
2877                 mlog_errno(status);
2878                 goto out_commit;
2879         }
2880
2881         d_instantiate(dentry, inode);
2882         status = 0;
2883 out_commit:
2884         ocfs2_commit_trans(osb, handle);
2885 orphan_unlock:
2886         ocfs2_inode_unlock(orphan_dir_inode, 1);
2887         inode_unlock(orphan_dir_inode);
2888         iput(orphan_dir_inode);
2889 leave:
2890
2891         ocfs2_inode_unlock(dir, 1);
2892
2893         brelse(di_bh);
2894         brelse(parent_di_bh);
2895         brelse(orphan_dir_bh);
2896
2897         ocfs2_free_dir_lookup_result(&lookup);
2898
2899         if (status)
2900                 mlog_errno(status);
2901
2902         return status;
2903 }
2904
2905 const struct inode_operations ocfs2_dir_iops = {
2906         .create         = ocfs2_create,
2907         .lookup         = ocfs2_lookup,
2908         .link           = ocfs2_link,
2909         .unlink         = ocfs2_unlink,
2910         .rmdir          = ocfs2_unlink,
2911         .symlink        = ocfs2_symlink,
2912         .mkdir          = ocfs2_mkdir,
2913         .mknod          = ocfs2_mknod,
2914         .rename         = ocfs2_rename,
2915         .setattr        = ocfs2_setattr,
2916         .getattr        = ocfs2_getattr,
2917         .permission     = ocfs2_permission,
2918         .listxattr      = ocfs2_listxattr,
2919         .fiemap         = ocfs2_fiemap,
2920         .get_acl        = ocfs2_iop_get_acl,
2921         .set_acl        = ocfs2_iop_set_acl,
2922         .fileattr_get   = ocfs2_fileattr_get,
2923         .fileattr_set   = ocfs2_fileattr_set,
2924 };