fs/ocfs2/namei.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * namei.c
   4  *
   5  * Create and rename file, directory, symlinks
   6  *
   7  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
   8  *
   9  *  Portions of this code from linux/fs/ext3/dir.c
  10  *
  11  *  Copyright (C) 1992, 1993, 1994, 1995
  12  *  Remy Card ([email protected])
  13  *  Laboratoire MASI - Institut Blaise pascal
  14  *  Universite Pierre et Marie Curie (Paris VI)
  15  *
  16  *   from
  17  *
  18  *   linux/fs/minix/dir.c
  19  *
  20  *   Copyright (C) 1991, 1992 Linux Torvalds
  21  */
  22
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/quotaops.h>
  28 #include <linux/iversion.h>
  29
  30 #include <cluster/masklog.h>
  31
  32 #include "ocfs2.h"
  33
  34 #include "alloc.h"
  35 #include "dcache.h"
  36 #include "dir.h"
  37 #include "dlmglue.h"
  38 #include "extent_map.h"
  39 #include "file.h"
  40 #include "inode.h"
  41 #include "journal.h"
  42 #include "namei.h"
  43 #include "suballoc.h"
  44 #include "super.h"
  45 #include "symlink.h"
  46 #include "sysfile.h"
  47 #include "uptodate.h"
  48 #include "xattr.h"
  49 #include "acl.h"
  50 #include "ocfs2_trace.h"
  51 #include "ioctl.h"
  52
  53 #include "buffer_head_io.h"
  54
  55 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
  56                               struct inode *dir,
  57                               struct inode *inode,
  58                               dev_t dev,
  59                               struct buffer_head **new_fe_bh,
  60                               struct buffer_head *parent_fe_bh,
  61                               handle_t *handle,
  62                               struct ocfs2_alloc_context *inode_ac);
  63
  64 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
  65                                     struct inode **ret_orphan_dir,
  66                                     u64 blkno,
  67                                     char *name,
  68                                     struct ocfs2_dir_lookup_result *lookup,
  69                                     bool dio);
  70
  71 static int ocfs2_orphan_add(struct ocfs2_super *osb,
  72                             handle_t *handle,
  73                             struct inode *inode,
  74                             struct buffer_head *fe_bh,
  75                             char *name,
  76                             struct ocfs2_dir_lookup_result *lookup,
  77                             struct inode *orphan_dir_inode,
  78                             bool dio);
  79
  80 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
  81                                      handle_t *handle,
  82                                      struct inode *inode,
  83                                      const char *symname);
  84
  85 static int ocfs2_double_lock(struct ocfs2_super *osb,
  86                              struct buffer_head **bh1,
  87                              struct inode *inode1,
  88                              struct buffer_head **bh2,
  89                              struct inode *inode2,
  90                              int rename);
  91
  92 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
  93 /* An orphan dir name is an 8 byte value, printed as a hex string */
  94 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
  95
  96 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
  97                                    unsigned int flags)
  98 {
  99         int status;
 100         u64 blkno;
 101         struct inode *inode = NULL;
 102         struct dentry *ret;
 103         struct ocfs2_inode_info *oi;
 104
 105         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
 106                            dentry->d_name.name,
 107                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
 108
 109         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
 110                 ret = ERR_PTR(-ENAMETOOLONG);
 111                 goto bail;
 112         }
 113
 114         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
 115         if (status < 0) {
 116                 if (status != -ENOENT)
 117                         mlog_errno(status);
 118                 ret = ERR_PTR(status);
 119                 goto bail;
 120         }
 121
 122         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
 123                                             dentry->d_name.len, &blkno);
 124         if (status < 0)
 125                 goto bail_add;
 126
 127         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
 128         if (IS_ERR(inode)) {
 129                 ret = ERR_PTR(-EACCES);
 130                 goto bail_unlock;
 131         }
 132
 133         oi = OCFS2_I(inode);
 134         /* Clear any orphaned state... If we were able to look up the
 135          * inode from a directory, it certainly can't be orphaned. We
 136          * might have the bad state from a node which intended to
 137          * orphan this inode but crashed before it could commit the
 138          * unlink. */
 139         spin_lock(&oi->ip_lock);
 140         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
 141         spin_unlock(&oi->ip_lock);
 142
 143 bail_add:
 144         ret = d_splice_alias(inode, dentry);
 145
 146         if (inode) {
 147                 /*
 148                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
 149                  * dentry, it will d_move() it on top of ourse. The
 150                  * return value will indicate this however, so in
 151                  * those cases, we switch them around for the locking
 152                  * code.
 153                  *
 154                  * NOTE: This dentry already has ->d_op set from
 155                  * ocfs2_get_parent() and ocfs2_get_dentry()
 156                  */
 157                 if (!IS_ERR_OR_NULL(ret))
 158                         dentry = ret;
 159
 160                 status = ocfs2_dentry_attach_lock(dentry, inode,
 161                                                   OCFS2_I(dir)->ip_blkno);
 162                 if (status) {
 163                         mlog_errno(status);
 164                         ret = ERR_PTR(status);
 165                         goto bail_unlock;
 166                 }
 167         } else
 168                 ocfs2_dentry_attach_gen(dentry);
 169
 170 bail_unlock:
 171         /* Don't drop the cluster lock until *after* the d_add --
 172          * unlink on another node will message us to remove that
 173          * dentry under this lock so otherwise we can race this with
 174          * the downconvert thread and have a stale dentry. */
 175         ocfs2_inode_unlock(dir, 0);
 176
 177 bail:
 178
 179         trace_ocfs2_lookup_ret(ret);
 180
 181         return ret;
 182 }
 183
 184 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
 185 {
 186         struct inode *inode;
 187         int status;
 188
 189         inode = new_inode(dir->i_sb);
 190         if (!inode) {
 191                 mlog(ML_ERROR, "new_inode failed!\n");
 192                 return ERR_PTR(-ENOMEM);
 193         }
 194
 195         /* populate as many fields early on as possible - many of
 196          * these are used by the support functions here and in
 197          * callers. */
 198         if (S_ISDIR(mode))
 199                 set_nlink(inode, 2);
 200         mode = mode_strip_sgid(&nop_mnt_idmap, dir, mode);
 201         inode_init_owner(&nop_mnt_idmap, inode, dir, mode);
 202         status = dquot_initialize(inode);
 203         if (status)
 204                 return ERR_PTR(status);
 205
 206         return inode;
 207 }
 208
 209 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
 210                 struct dentry *dentry, struct inode *inode)
 211 {
 212         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
 213
 214         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
 215         ocfs2_lock_res_free(&dl->dl_lockres);
 216         BUG_ON(dl->dl_count != 1);
 217         spin_lock(&dentry_attach_lock);
 218         dentry->d_fsdata = NULL;
 219         spin_unlock(&dentry_attach_lock);
 220         kfree(dl);
 221         iput(inode);
 222 }
 223
 224 static int ocfs2_mknod(struct mnt_idmap *idmap,
 225                        struct inode *dir,
 226                        struct dentry *dentry,
 227                        umode_t mode,
 228                        dev_t dev)
 229 {
 230         int status = 0;
 231         struct buffer_head *parent_fe_bh = NULL;
 232         handle_t *handle = NULL;
 233         struct ocfs2_super *osb;
 234         struct ocfs2_dinode *dirfe;
 235         struct ocfs2_dinode *fe = NULL;
 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                 .name = NULL,
 246                 .enable = 1,
 247         };
 248         int did_quota_inode = 0;
 249         struct ocfs2_dir_lookup_result lookup = { NULL, };
 250         sigset_t oldset;
 251         int did_block_signals = 0;
 252         struct ocfs2_dentry_lock *dl = NULL;
 253
 254         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 255                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
 256                           (unsigned long)dev, mode);
 257
 258         status = dquot_initialize(dir);
 259         if (status) {
 260                 mlog_errno(status);
 261                 return status;
 262         }
 263
 264         /* get our super block */
 265         osb = OCFS2_SB(dir->i_sb);
 266
 267         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
 268         if (status < 0) {
 269                 if (status != -ENOENT)
 270                         mlog_errno(status);
 271                 return status;
 272         }
 273
 274         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
 275                 status = -EMLINK;
 276                 goto leave;
 277         }
 278
 279         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
 280         if (!ocfs2_read_links_count(dirfe)) {
 281                 /* can't make a file in a deleted directory. */
 282                 status = -ENOENT;
 283                 goto leave;
 284         }
 285
 286         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 287                                            dentry->d_name.len);
 288         if (status)
 289                 goto leave;
 290
 291         /* get a spot inside the dir. */
 292         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 293                                               dentry->d_name.name,
 294                                               dentry->d_name.len, &lookup);
 295         if (status < 0) {
 296                 mlog_errno(status);
 297                 goto leave;
 298         }
 299
 300         /* reserve an inode spot */
 301         status = ocfs2_reserve_new_inode(osb, &inode_ac);
 302         if (status < 0) {
 303                 if (status != -ENOSPC)
 304                         mlog_errno(status);
 305                 goto leave;
 306         }
 307
 308         inode = ocfs2_get_init_inode(dir, mode);
 309         if (IS_ERR(inode)) {
 310                 status = PTR_ERR(inode);
 311                 inode = NULL;
 312                 mlog_errno(status);
 313                 goto leave;
 314         }
 315
 316         /* get security xattr */
 317         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
 318         if (status) {
 319                 if (status == -EOPNOTSUPP)
 320                         si.enable = 0;
 321                 else {
 322                         mlog_errno(status);
 323                         goto leave;
 324                 }
 325         }
 326
 327         /* calculate meta data/clusters for setting security and acl xattr */
 328         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
 329                                        &si, &want_clusters,
 330                                        &xattr_credits, &want_meta);
 331         if (status < 0) {
 332                 mlog_errno(status);
 333                 goto leave;
 334         }
 335
 336         /* Reserve a cluster if creating an extent based directory. */
 337         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
 338                 want_clusters += 1;
 339
 340                 /* Dir indexing requires extra space as well */
 341                 if (ocfs2_supports_indexed_dirs(osb))
 342                         want_meta++;
 343         }
 344
 345         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
 346         if (status < 0) {
 347                 if (status != -ENOSPC)
 348                         mlog_errno(status);
 349                 goto leave;
 350         }
 351
 352         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
 353         if (status < 0) {
 354                 if (status != -ENOSPC)
 355                         mlog_errno(status);
 356                 goto leave;
 357         }
 358
 359         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
 360                                                             S_ISDIR(mode),
 361                                                             xattr_credits));
 362         if (IS_ERR(handle)) {
 363                 status = PTR_ERR(handle);
 364                 handle = NULL;
 365                 mlog_errno(status);
 366                 goto leave;
 367         }
 368
 369         /* Starting to change things, restart is no longer possible. */
 370         ocfs2_block_signals(&oldset);
 371         did_block_signals = 1;
 372
 373         status = dquot_alloc_inode(inode);
 374         if (status)
 375                 goto leave;
 376         did_quota_inode = 1;
 377
 378         /* do the real work now. */
 379         status = ocfs2_mknod_locked(osb, dir, inode, dev,
 380                                     &new_fe_bh, parent_fe_bh, handle,
 381                                     inode_ac);
 382         if (status < 0) {
 383                 mlog_errno(status);
 384                 goto leave;
 385         }
 386
 387         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
 388         if (S_ISDIR(mode)) {
 389                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
 390                                             new_fe_bh, data_ac, meta_ac);
 391                 if (status < 0) {
 392                         mlog_errno(status);
 393                         goto leave;
 394                 }
 395
 396                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
 397                                                  parent_fe_bh,
 398                                                  OCFS2_JOURNAL_ACCESS_WRITE);
 399                 if (status < 0) {
 400                         mlog_errno(status);
 401                         goto leave;
 402                 }
 403                 ocfs2_add_links_count(dirfe, 1);
 404                 ocfs2_journal_dirty(handle, parent_fe_bh);
 405                 inc_nlink(dir);
 406         }
 407
 408         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
 409                          meta_ac, data_ac);
 410
 411         if (status < 0) {
 412                 mlog_errno(status);
 413                 goto roll_back;
 414         }
 415
 416         if (si.enable) {
 417                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
 418                                                  meta_ac, data_ac);
 419                 if (status < 0) {
 420                         mlog_errno(status);
 421                         goto roll_back;
 422                 }
 423         }
 424
 425         /*
 426          * Do this before adding the entry to the directory. We add
 427          * also set d_op after success so that ->d_iput() will cleanup
 428          * the dentry lock even if ocfs2_add_entry() fails below.
 429          */
 430         status = ocfs2_dentry_attach_lock(dentry, inode,
 431                                           OCFS2_I(dir)->ip_blkno);
 432         if (status) {
 433                 mlog_errno(status);
 434                 goto roll_back;
 435         }
 436
 437         dl = dentry->d_fsdata;
 438
 439         status = ocfs2_add_entry(handle, dentry, inode,
 440                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
 441                                  &lookup);
 442         if (status < 0) {
 443                 mlog_errno(status);
 444                 goto roll_back;
 445         }
 446
 447         insert_inode_hash(inode);
 448         d_instantiate(dentry, inode);
 449         status = 0;
 450
 451 roll_back:
 452         if (status < 0 && S_ISDIR(mode)) {
 453                 ocfs2_add_links_count(dirfe, -1);
 454                 drop_nlink(dir);
 455         }
 456
 457 leave:
 458         if (status < 0 && did_quota_inode)
 459                 dquot_free_inode(inode);
 460         if (handle) {
 461                 if (status < 0 && fe)
 462                         ocfs2_set_links_count(fe, 0);
 463                 ocfs2_commit_trans(osb, handle);
 464         }
 465
 466         ocfs2_inode_unlock(dir, 1);
 467         if (did_block_signals)
 468                 ocfs2_unblock_signals(&oldset);
 469
 470         brelse(new_fe_bh);
 471         brelse(parent_fe_bh);
 472         kfree(si.value);
 473
 474         ocfs2_free_dir_lookup_result(&lookup);
 475
 476         if (inode_ac)
 477                 ocfs2_free_alloc_context(inode_ac);
 478
 479         if (data_ac)
 480                 ocfs2_free_alloc_context(data_ac);
 481
 482         if (meta_ac)
 483                 ocfs2_free_alloc_context(meta_ac);
 484
 485         /*
 486          * We should call iput after the i_rwsem of the bitmap been
 487          * unlocked in ocfs2_free_alloc_context, or the
 488          * ocfs2_delete_inode will mutex_lock again.
 489          */
 490         if ((status < 0) && inode) {
 491                 if (dl)
 492                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
 493
 494                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
 495                 clear_nlink(inode);
 496                 iput(inode);
 497         }
 498
 499         if (status)
 500                 mlog_errno(status);
 501
 502         return status;
 503 }
 504
 505 static int __ocfs2_mknod_locked(struct inode *dir,
 506                                 struct inode *inode,
 507                                 dev_t dev,
 508                                 struct buffer_head **new_fe_bh,
 509                                 struct buffer_head *parent_fe_bh,
 510                                 handle_t *handle,
 511                                 struct ocfs2_alloc_context *inode_ac,
 512                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
 513 {
 514         int status = 0;
 515         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 516         struct ocfs2_dinode *fe = NULL;
 517         struct ocfs2_extent_list *fel;
 518         u16 feat;
 519         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 520         struct timespec64 ts;
 521
 522         *new_fe_bh = NULL;
 523
 524         /* populate as many fields early on as possible - many of
 525          * these are used by the support functions here and in
 526          * callers. */
 527         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
 528         oi->ip_blkno = fe_blkno;
 529         spin_lock(&osb->osb_lock);
 530         inode->i_generation = osb->s_next_generation++;
 531         spin_unlock(&osb->osb_lock);
 532
 533         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
 534         if (!*new_fe_bh) {
 535                 status = -ENOMEM;
 536                 mlog_errno(status);
 537                 goto leave;
 538         }
 539         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
 540
 541         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
 542                                          *new_fe_bh,
 543                                          OCFS2_JOURNAL_ACCESS_CREATE);
 544         if (status < 0) {
 545                 mlog_errno(status);
 546                 goto leave;
 547         }
 548
 549         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
 550         memset(fe, 0, osb->sb->s_blocksize);
 551
 552         fe->i_generation = cpu_to_le32(inode->i_generation);
 553         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
 554         fe->i_blkno = cpu_to_le64(fe_blkno);
 555         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
 556         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
 557         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
 558         fe->i_uid = cpu_to_le32(i_uid_read(inode));
 559         fe->i_gid = cpu_to_le32(i_gid_read(inode));
 560         fe->i_mode = cpu_to_le16(inode->i_mode);
 561         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
 562                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
 563
 564         ocfs2_set_links_count(fe, inode->i_nlink);
 565
 566         fe->i_last_eb_blk = 0;
 567         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
 568         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
 569         ktime_get_real_ts64(&ts);
 570         fe->i_atime = fe->i_ctime = fe->i_mtime =
 571                 cpu_to_le64(ts.tv_sec);
 572         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
 573                 cpu_to_le32(ts.tv_nsec);
 574         fe->i_dtime = 0;
 575
 576         /*
 577          * If supported, directories start with inline data. If inline
 578          * isn't supported, but indexing is, we start them as indexed.
 579          */
 580         feat = le16_to_cpu(fe->i_dyn_features);
 581         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
 582                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
 583
 584                 fe->id2.i_data.id_count = cpu_to_le16(
 585                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
 586         } else {
 587                 fel = &fe->id2.i_list;
 588                 fel->l_tree_depth = 0;
 589                 fel->l_next_free_rec = 0;
 590                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
 591         }
 592
 593         ocfs2_journal_dirty(handle, *new_fe_bh);
 594
 595         ocfs2_populate_inode(inode, fe, 1);
 596         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
 597         if (!ocfs2_mount_local(osb)) {
 598                 status = ocfs2_create_new_inode_locks(inode);
 599                 if (status < 0)
 600                         mlog_errno(status);
 601         }
 602
 603         ocfs2_update_inode_fsync_trans(handle, inode, 1);
 604
 605 leave:
 606         if (status < 0) {
 607                 if (*new_fe_bh) {
 608                         brelse(*new_fe_bh);
 609                         *new_fe_bh = NULL;
 610                 }
 611         }
 612
 613         if (status)
 614                 mlog_errno(status);
 615         return status;
 616 }
 617
 618 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
 619                               struct inode *dir,
 620                               struct inode *inode,
 621                               dev_t dev,
 622                               struct buffer_head **new_fe_bh,
 623                               struct buffer_head *parent_fe_bh,
 624                               handle_t *handle,
 625                               struct ocfs2_alloc_context *inode_ac)
 626 {
 627         int status = 0;
 628         u64 suballoc_loc, fe_blkno = 0;
 629         u16 suballoc_bit;
 630
 631         *new_fe_bh = NULL;
 632
 633         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
 634                                        inode_ac, &suballoc_loc,
 635                                        &suballoc_bit, &fe_blkno);
 636         if (status < 0) {
 637                 mlog_errno(status);
 638                 return status;
 639         }
 640
 641         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
 642                                     parent_fe_bh, handle, inode_ac,
 643                                     fe_blkno, suballoc_loc, suballoc_bit);
 644 }
 645
 646 static int ocfs2_mkdir(struct mnt_idmap *idmap,
 647                        struct inode *dir,
 648                        struct dentry *dentry,
 649                        umode_t mode)
 650 {
 651         int ret;
 652
 653         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 654                           OCFS2_I(dir)->ip_blkno, mode);
 655         ret = ocfs2_mknod(&nop_mnt_idmap, dir, dentry, mode | S_IFDIR, 0);
 656         if (ret)
 657                 mlog_errno(ret);
 658
 659         return ret;
 660 }
 661
 662 static int ocfs2_create(struct mnt_idmap *idmap,
 663                         struct inode *dir,
 664                         struct dentry *dentry,
 665                         umode_t mode,
 666                         bool excl)
 667 {
 668         int ret;
 669
 670         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
 671                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
 672         ret = ocfs2_mknod(&nop_mnt_idmap, dir, dentry, mode | S_IFREG, 0);
 673         if (ret)
 674                 mlog_errno(ret);
 675
 676         return ret;
 677 }
 678
 679 static int ocfs2_link(struct dentry *old_dentry,
 680                       struct inode *dir,
 681                       struct dentry *dentry)
 682 {
 683         handle_t *handle;
 684         struct inode *inode = d_inode(old_dentry);
 685         struct inode *old_dir = d_inode(old_dentry->d_parent);
 686         int err;
 687         struct buffer_head *fe_bh = NULL;
 688         struct buffer_head *old_dir_bh = NULL;
 689         struct buffer_head *parent_fe_bh = NULL;
 690         struct ocfs2_dinode *fe = NULL;
 691         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 692         struct ocfs2_dir_lookup_result lookup = { NULL, };
 693         sigset_t oldset;
 694         u64 old_de_ino;
 695
 696         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
 697                          old_dentry->d_name.len, old_dentry->d_name.name,
 698                          dentry->d_name.len, dentry->d_name.name);
 699
 700         if (S_ISDIR(inode->i_mode))
 701                 return -EPERM;
 702
 703         err = dquot_initialize(dir);
 704         if (err) {
 705                 mlog_errno(err);
 706                 return err;
 707         }
 708
 709         err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
 710                         &parent_fe_bh, dir, 0);
 711         if (err < 0) {
 712                 if (err != -ENOENT)
 713                         mlog_errno(err);
 714                 return err;
 715         }
 716
 717         /* make sure both dirs have bhs
 718          * get an extra ref on old_dir_bh if old==new */
 719         if (!parent_fe_bh) {
 720                 if (old_dir_bh) {
 721                         parent_fe_bh = old_dir_bh;
 722                         get_bh(parent_fe_bh);
 723                 } else {
 724                         mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
 725                         err = -EIO;
 726                         goto out;
 727                 }
 728         }
 729
 730         if (!dir->i_nlink) {
 731                 err = -ENOENT;
 732                 goto out;
 733         }
 734
 735         err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
 736                         old_dentry->d_name.len, &old_de_ino);
 737         if (err) {
 738                 err = -ENOENT;
 739                 goto out;
 740         }
 741
 742         /*
 743          * Check whether another node removed the source inode while we
 744          * were in the vfs.
 745          */
 746         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
 747                 err = -ENOENT;
 748                 goto out;
 749         }
 750
 751         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
 752                                         dentry->d_name.len);
 753         if (err)
 754                 goto out;
 755
 756         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
 757                                            dentry->d_name.name,
 758                                            dentry->d_name.len, &lookup);
 759         if (err < 0) {
 760                 mlog_errno(err);
 761                 goto out;
 762         }
 763
 764         err = ocfs2_inode_lock(inode, &fe_bh, 1);
 765         if (err < 0) {
 766                 if (err != -ENOENT)
 767                         mlog_errno(err);
 768                 goto out;
 769         }
 770
 771         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 772         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
 773                 err = -EMLINK;
 774                 goto out_unlock_inode;
 775         }
 776
 777         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
 778         if (IS_ERR(handle)) {
 779                 err = PTR_ERR(handle);
 780                 handle = NULL;
 781                 mlog_errno(err);
 782                 goto out_unlock_inode;
 783         }
 784
 785         /* Starting to change things, restart is no longer possible. */
 786         ocfs2_block_signals(&oldset);
 787
 788         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 789                                       OCFS2_JOURNAL_ACCESS_WRITE);
 790         if (err < 0) {
 791                 mlog_errno(err);
 792                 goto out_commit;
 793         }
 794
 795         inc_nlink(inode);
 796         inode_set_ctime_current(inode);
 797         ocfs2_set_links_count(fe, inode->i_nlink);
 798         fe->i_ctime = cpu_to_le64(inode_get_ctime_sec(inode));
 799         fe->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(inode));
 800         ocfs2_journal_dirty(handle, fe_bh);
 801
 802         err = ocfs2_add_entry(handle, dentry, inode,
 803                               OCFS2_I(inode)->ip_blkno,
 804                               parent_fe_bh, &lookup);
 805         if (err) {
 806                 ocfs2_add_links_count(fe, -1);
 807                 drop_nlink(inode);
 808                 mlog_errno(err);
 809                 goto out_commit;
 810         }
 811
 812         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
 813         if (err) {
 814                 mlog_errno(err);
 815                 goto out_commit;
 816         }
 817
 818         ihold(inode);
 819         d_instantiate(dentry, inode);
 820
 821 out_commit:
 822         ocfs2_commit_trans(osb, handle);
 823         ocfs2_unblock_signals(&oldset);
 824 out_unlock_inode:
 825         ocfs2_inode_unlock(inode, 1);
 826
 827 out:
 828         ocfs2_double_unlock(old_dir, dir);
 829
 830         brelse(fe_bh);
 831         brelse(parent_fe_bh);
 832         brelse(old_dir_bh);
 833
 834         ocfs2_free_dir_lookup_result(&lookup);
 835
 836         if (err)
 837                 mlog_errno(err);
 838
 839         return err;
 840 }
 841
 842 /*
 843  * Takes and drops an exclusive lock on the given dentry. This will
 844  * force other nodes to drop it.
 845  */
 846 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
 847 {
 848         int ret;
 849
 850         ret = ocfs2_dentry_lock(dentry, 1);
 851         if (ret)
 852                 mlog_errno(ret);
 853         else
 854                 ocfs2_dentry_unlock(dentry, 1);
 855
 856         return ret;
 857 }
 858
 859 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
 860 {
 861         if (S_ISDIR(inode->i_mode)) {
 862                 if (inode->i_nlink == 2)
 863                         return 1;
 864                 return 0;
 865         }
 866
 867         if (inode->i_nlink == 1)
 868                 return 1;
 869         return 0;
 870 }
 871
 872 static int ocfs2_unlink(struct inode *dir,
 873                         struct dentry *dentry)
 874 {
 875         int status;
 876         int child_locked = 0;
 877         bool is_unlinkable = false;
 878         struct inode *inode = d_inode(dentry);
 879         struct inode *orphan_dir = NULL;
 880         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 881         u64 blkno;
 882         struct ocfs2_dinode *fe = NULL;
 883         struct buffer_head *fe_bh = NULL;
 884         struct buffer_head *parent_node_bh = NULL;
 885         handle_t *handle = NULL;
 886         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
 887         struct ocfs2_dir_lookup_result lookup = { NULL, };
 888         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
 889
 890         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
 891                            dentry->d_name.name,
 892                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
 893                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
 894
 895         status = dquot_initialize(dir);
 896         if (status) {
 897                 mlog_errno(status);
 898                 return status;
 899         }
 900
 901         BUG_ON(d_inode(dentry->d_parent) != dir);
 902
 903         if (inode == osb->root_inode)
 904                 return -EPERM;
 905
 906         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
 907                                          OI_LS_PARENT);
 908         if (status < 0) {
 909                 if (status != -ENOENT)
 910                         mlog_errno(status);
 911                 return status;
 912         }
 913
 914         status = ocfs2_find_files_on_disk(dentry->d_name.name,
 915                                           dentry->d_name.len, &blkno, dir,
 916                                           &lookup);
 917         if (status < 0) {
 918                 if (status != -ENOENT)
 919                         mlog_errno(status);
 920                 goto leave;
 921         }
 922
 923         if (OCFS2_I(inode)->ip_blkno != blkno) {
 924                 status = -ENOENT;
 925
 926                 trace_ocfs2_unlink_noent(
 927                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
 928                                 (unsigned long long)blkno,
 929                                 OCFS2_I(inode)->ip_flags);
 930                 goto leave;
 931         }
 932
 933         status = ocfs2_inode_lock(inode, &fe_bh, 1);
 934         if (status < 0) {
 935                 if (status != -ENOENT)
 936                         mlog_errno(status);
 937                 goto leave;
 938         }
 939         child_locked = 1;
 940
 941         if (S_ISDIR(inode->i_mode)) {
 942                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
 943                         status = -ENOTEMPTY;
 944                         goto leave;
 945                 }
 946         }
 947
 948         status = ocfs2_remote_dentry_delete(dentry);
 949         if (status < 0) {
 950                 /* This remote delete should succeed under all normal
 951                  * circumstances. */
 952                 mlog_errno(status);
 953                 goto leave;
 954         }
 955
 956         if (ocfs2_inode_is_unlinkable(inode)) {
 957                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
 958                                                   OCFS2_I(inode)->ip_blkno,
 959                                                   orphan_name, &orphan_insert,
 960                                                   false);
 961                 if (status < 0) {
 962                         mlog_errno(status);
 963                         goto leave;
 964                 }
 965                 is_unlinkable = true;
 966         }
 967
 968         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
 969         if (IS_ERR(handle)) {
 970                 status = PTR_ERR(handle);
 971                 handle = NULL;
 972                 mlog_errno(status);
 973                 goto leave;
 974         }
 975
 976         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
 977                                          OCFS2_JOURNAL_ACCESS_WRITE);
 978         if (status < 0) {
 979                 mlog_errno(status);
 980                 goto leave;
 981         }
 982
 983         fe = (struct ocfs2_dinode *) fe_bh->b_data;
 984
 985         /* delete the name from the parent dir */
 986         status = ocfs2_delete_entry(handle, dir, &lookup);
 987         if (status < 0) {
 988                 mlog_errno(status);
 989                 goto leave;
 990         }
 991
 992         if (S_ISDIR(inode->i_mode))
 993                 drop_nlink(inode);
 994         drop_nlink(inode);
 995         ocfs2_set_links_count(fe, inode->i_nlink);
 996         ocfs2_journal_dirty(handle, fe_bh);
 997
 998         inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
 999         if (S_ISDIR(inode->i_mode))
1000                 drop_nlink(dir);
1001
1002         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1003         if (status < 0) {
1004                 mlog_errno(status);
1005                 if (S_ISDIR(inode->i_mode))
1006                         inc_nlink(dir);
1007                 goto leave;
1008         }
1009
1010         if (is_unlinkable) {
1011                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1012                                 orphan_name, &orphan_insert, orphan_dir, false);
1013                 if (status < 0)
1014                         mlog_errno(status);
1015         }
1016
1017 leave:
1018         if (handle)
1019                 ocfs2_commit_trans(osb, handle);
1020
1021         if (orphan_dir) {
1022                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1023                 ocfs2_inode_unlock(orphan_dir, 1);
1024                 inode_unlock(orphan_dir);
1025                 iput(orphan_dir);
1026         }
1027
1028         if (child_locked)
1029                 ocfs2_inode_unlock(inode, 1);
1030
1031         ocfs2_inode_unlock(dir, 1);
1032
1033         brelse(fe_bh);
1034         brelse(parent_node_bh);
1035
1036         ocfs2_free_dir_lookup_result(&orphan_insert);
1037         ocfs2_free_dir_lookup_result(&lookup);
1038
1039         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1040                 mlog_errno(status);
1041
1042         return status;
1043 }
1044
1045 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1046                 u64 src_inode_no, u64 dest_inode_no)
1047 {
1048         int ret = 0, i = 0;
1049         u64 parent_inode_no = 0;
1050         u64 child_inode_no = src_inode_no;
1051         struct inode *child_inode;
1052
1053 #define MAX_LOOKUP_TIMES 32
1054         while (1) {
1055                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1056                 if (IS_ERR(child_inode)) {
1057                         ret = PTR_ERR(child_inode);
1058                         break;
1059                 }
1060
1061                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1062                 if (ret < 0) {
1063                         iput(child_inode);
1064                         if (ret != -ENOENT)
1065                                 mlog_errno(ret);
1066                         break;
1067                 }
1068
1069                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1070                                 &parent_inode_no);
1071                 ocfs2_inode_unlock(child_inode, 0);
1072                 iput(child_inode);
1073                 if (ret < 0) {
1074                         ret = -ENOENT;
1075                         break;
1076                 }
1077
1078                 if (parent_inode_no == dest_inode_no) {
1079                         ret = 1;
1080                         break;
1081                 }
1082
1083                 if (parent_inode_no == osb->root_inode->i_ino) {
1084                         ret = 0;
1085                         break;
1086                 }
1087
1088                 child_inode_no = parent_inode_no;
1089
1090                 if (++i >= MAX_LOOKUP_TIMES) {
1091                         mlog_ratelimited(ML_NOTICE, "max lookup times reached, "
1092                                         "filesystem may have nested directories, "
1093                                         "src inode: %llu, dest inode: %llu.\n",
1094                                         (unsigned long long)src_inode_no,
1095                                         (unsigned long long)dest_inode_no);
1096                         ret = 0;
1097                         break;
1098                 }
1099         }
1100
1101         return ret;
1102 }
1103
1104 /*
1105  * The only place this should be used is rename and link!
1106  * if they have the same id, then the 1st one is the only one locked.
1107  */
1108 static int ocfs2_double_lock(struct ocfs2_super *osb,
1109                              struct buffer_head **bh1,
1110                              struct inode *inode1,
1111                              struct buffer_head **bh2,
1112                              struct inode *inode2,
1113                              int rename)
1114 {
1115         int status;
1116         int inode1_is_ancestor, inode2_is_ancestor;
1117         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1118         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1119
1120         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1121                                 (unsigned long long)oi2->ip_blkno);
1122
1123         if (*bh1)
1124                 *bh1 = NULL;
1125         if (*bh2)
1126                 *bh2 = NULL;
1127
1128         /* we always want to lock the one with the lower lockid first.
1129          * and if they are nested, we lock ancestor first */
1130         if (oi1->ip_blkno != oi2->ip_blkno) {
1131                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1132                                 oi1->ip_blkno);
1133                 if (inode1_is_ancestor < 0) {
1134                         status = inode1_is_ancestor;
1135                         goto bail;
1136                 }
1137
1138                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1139                                 oi2->ip_blkno);
1140                 if (inode2_is_ancestor < 0) {
1141                         status = inode2_is_ancestor;
1142                         goto bail;
1143                 }
1144
1145                 if ((inode1_is_ancestor == 1) ||
1146                                 (oi1->ip_blkno < oi2->ip_blkno &&
1147                                 inode2_is_ancestor == 0)) {
1148                         /* switch id1 and id2 around */
1149                         swap(bh2, bh1);
1150                         swap(inode2, inode1);
1151                 }
1152                 /* lock id2 */
1153                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1154                                 rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1155                 if (status < 0) {
1156                         if (status != -ENOENT)
1157                                 mlog_errno(status);
1158                         goto bail;
1159                 }
1160         }
1161
1162         /* lock id1 */
1163         status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1164                         rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1165         if (status < 0) {
1166                 /*
1167                  * An error return must mean that no cluster locks
1168                  * were held on function exit.
1169                  */
1170                 if (oi1->ip_blkno != oi2->ip_blkno) {
1171                         ocfs2_inode_unlock(inode2, 1);
1172                         brelse(*bh2);
1173                         *bh2 = NULL;
1174                 }
1175
1176                 if (status != -ENOENT)
1177                         mlog_errno(status);
1178         }
1179
1180         trace_ocfs2_double_lock_end(
1181                         (unsigned long long)oi1->ip_blkno,
1182                         (unsigned long long)oi2->ip_blkno);
1183
1184 bail:
1185         if (status)
1186                 mlog_errno(status);
1187         return status;
1188 }
1189
1190 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1191 {
1192         ocfs2_inode_unlock(inode1, 1);
1193
1194         if (inode1 != inode2)
1195                 ocfs2_inode_unlock(inode2, 1);
1196 }
1197
1198 static int ocfs2_rename(struct mnt_idmap *idmap,
1199                         struct inode *old_dir,
1200                         struct dentry *old_dentry,
1201                         struct inode *new_dir,
1202                         struct dentry *new_dentry,
1203                         unsigned int flags)
1204 {
1205         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1206         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1207         struct inode *old_inode = d_inode(old_dentry);
1208         struct inode *new_inode = d_inode(new_dentry);
1209         struct inode *orphan_dir = NULL;
1210         struct ocfs2_dinode *newfe = NULL;
1211         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1212         struct buffer_head *newfe_bh = NULL;
1213         struct buffer_head *old_inode_bh = NULL;
1214         struct ocfs2_super *osb = NULL;
1215         u64 newfe_blkno, old_de_ino;
1216         handle_t *handle = NULL;
1217         struct buffer_head *old_dir_bh = NULL;
1218         struct buffer_head *new_dir_bh = NULL;
1219         u32 old_dir_nlink = old_dir->i_nlink;
1220         struct ocfs2_dinode *old_di;
1221         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1222         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1223         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1224         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1225         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1226         bool should_add_orphan = false;
1227
1228         if (flags)
1229                 return -EINVAL;
1230
1231         /* At some point it might be nice to break this function up a
1232          * bit. */
1233
1234         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1235                            old_dentry->d_name.len, old_dentry->d_name.name,
1236                            new_dentry->d_name.len, new_dentry->d_name.name);
1237
1238         status = dquot_initialize(old_dir);
1239         if (status) {
1240                 mlog_errno(status);
1241                 goto bail;
1242         }
1243         status = dquot_initialize(new_dir);
1244         if (status) {
1245                 mlog_errno(status);
1246                 goto bail;
1247         }
1248
1249         osb = OCFS2_SB(old_dir->i_sb);
1250
1251         if (new_inode) {
1252                 if (!igrab(new_inode))
1253                         BUG();
1254         }
1255
1256         /* Assume a directory hierarchy thusly:
1257          * a/b/c
1258          * a/d
1259          * a,b,c, and d are all directories.
1260          *
1261          * from cwd of 'a' on both nodes:
1262          * node1: mv b/c d
1263          * node2: mv d   b/c
1264          *
1265          * And that's why, just like the VFS, we need a file system
1266          * rename lock. */
1267         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1268                 status = ocfs2_rename_lock(osb);
1269                 if (status < 0) {
1270                         mlog_errno(status);
1271                         goto bail;
1272                 }
1273                 rename_lock = 1;
1274
1275                 /* here we cannot guarantee the inodes haven't just been
1276                  * changed, so check if they are nested again */
1277                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1278                                 old_inode->i_ino);
1279                 if (status < 0) {
1280                         mlog_errno(status);
1281                         goto bail;
1282                 } else if (status == 1) {
1283                         status = -EPERM;
1284                         trace_ocfs2_rename_not_permitted(
1285                                         (unsigned long long)old_inode->i_ino,
1286                                         (unsigned long long)new_dir->i_ino);
1287                         goto bail;
1288                 }
1289         }
1290
1291         /* if old and new are the same, this'll just do one lock. */
1292         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1293                                    &new_dir_bh, new_dir, 1);
1294         if (status < 0) {
1295                 mlog_errno(status);
1296                 goto bail;
1297         }
1298         parents_locked = 1;
1299
1300         if (!new_dir->i_nlink) {
1301                 status = -EACCES;
1302                 goto bail;
1303         }
1304
1305         /* make sure both dirs have bhs
1306          * get an extra ref on old_dir_bh if old==new */
1307         if (!new_dir_bh) {
1308                 if (old_dir_bh) {
1309                         new_dir_bh = old_dir_bh;
1310                         get_bh(new_dir_bh);
1311                 } else {
1312                         mlog(ML_ERROR, "no old_dir_bh!\n");
1313                         status = -EIO;
1314                         goto bail;
1315                 }
1316         }
1317
1318         /*
1319          * Aside from allowing a meta data update, the locking here
1320          * also ensures that the downconvert thread on other nodes
1321          * won't have to concurrently downconvert the inode and the
1322          * dentry locks.
1323          */
1324         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1325                                          OI_LS_PARENT);
1326         if (status < 0) {
1327                 if (status != -ENOENT)
1328                         mlog_errno(status);
1329                 goto bail;
1330         }
1331         old_child_locked = 1;
1332
1333         status = ocfs2_remote_dentry_delete(old_dentry);
1334         if (status < 0) {
1335                 mlog_errno(status);
1336                 goto bail;
1337         }
1338
1339         if (S_ISDIR(old_inode->i_mode) && new_dir != old_dir) {
1340                 u64 old_inode_parent;
1341
1342                 update_dot_dot = 1;
1343                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1344                                                   old_inode,
1345                                                   &old_inode_dot_dot_res);
1346                 if (status) {
1347                         status = -EIO;
1348                         goto bail;
1349                 }
1350
1351                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1352                         status = -EIO;
1353                         goto bail;
1354                 }
1355
1356                 if (!new_inode && new_dir->i_nlink >= ocfs2_link_max(osb)) {
1357                         status = -EMLINK;
1358                         goto bail;
1359                 }
1360         }
1361
1362         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1363                                             old_dentry->d_name.len,
1364                                             &old_de_ino);
1365         if (status) {
1366                 status = -ENOENT;
1367                 goto bail;
1368         }
1369
1370         /*
1371          *  Check for inode number is _not_ due to possible IO errors.
1372          *  We might rmdir the source, keep it as pwd of some process
1373          *  and merrily kill the link to whatever was created under the
1374          *  same name. Goodbye sticky bit ;-<
1375          */
1376         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1377                 status = -ENOENT;
1378                 goto bail;
1379         }
1380
1381         /* check if the target already exists (in which case we need
1382          * to delete it */
1383         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1384                                           new_dentry->d_name.len,
1385                                           &newfe_blkno, new_dir,
1386                                           &target_lookup_res);
1387         /* The only error we allow here is -ENOENT because the new
1388          * file not existing is perfectly valid. */
1389         if ((status < 0) && (status != -ENOENT)) {
1390                 /* If we cannot find the file specified we should just */
1391                 /* return the error... */
1392                 mlog_errno(status);
1393                 goto bail;
1394         }
1395         if (status == 0)
1396                 target_exists = 1;
1397
1398         if (!target_exists && new_inode) {
1399                 /*
1400                  * Target was unlinked by another node while we were
1401                  * waiting to get to ocfs2_rename(). There isn't
1402                  * anything we can do here to help the situation, so
1403                  * bubble up the appropriate error.
1404                  */
1405                 status = -ENOENT;
1406                 goto bail;
1407         }
1408
1409         /* In case we need to overwrite an existing file, we blow it
1410          * away first */
1411         if (target_exists) {
1412                 /* VFS didn't think there existed an inode here, but
1413                  * someone else in the cluster must have raced our
1414                  * rename to create one. Today we error cleanly, in
1415                  * the future we should consider calling iget to build
1416                  * a new struct inode for this entry. */
1417                 if (!new_inode) {
1418                         status = -EACCES;
1419
1420                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1421                                                 new_dentry->d_name.name);
1422                         goto bail;
1423                 }
1424
1425                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1426                         status = -EACCES;
1427
1428                         trace_ocfs2_rename_disagree(
1429                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1430                              (unsigned long long)newfe_blkno,
1431                              OCFS2_I(new_inode)->ip_flags);
1432                         goto bail;
1433                 }
1434
1435                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1436                 if (status < 0) {
1437                         if (status != -ENOENT)
1438                                 mlog_errno(status);
1439                         goto bail;
1440                 }
1441                 new_child_locked = 1;
1442
1443                 status = ocfs2_remote_dentry_delete(new_dentry);
1444                 if (status < 0) {
1445                         mlog_errno(status);
1446                         goto bail;
1447                 }
1448
1449                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1450
1451                 trace_ocfs2_rename_over_existing(
1452                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1453                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1454
1455                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1456                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1457                                                 OCFS2_I(new_inode)->ip_blkno,
1458                                                 orphan_name, &orphan_insert,
1459                                                 false);
1460                         if (status < 0) {
1461                                 mlog_errno(status);
1462                                 goto bail;
1463                         }
1464                         should_add_orphan = true;
1465                 }
1466         } else {
1467                 BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1468
1469                 status = ocfs2_check_dir_for_entry(new_dir,
1470                                                    new_dentry->d_name.name,
1471                                                    new_dentry->d_name.len);
1472                 if (status)
1473                         goto bail;
1474
1475                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1476                                                       new_dentry->d_name.name,
1477                                                       new_dentry->d_name.len,
1478                                                       &target_insert);
1479                 if (status < 0) {
1480                         mlog_errno(status);
1481                         goto bail;
1482                 }
1483         }
1484
1485         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1486         if (IS_ERR(handle)) {
1487                 status = PTR_ERR(handle);
1488                 handle = NULL;
1489                 mlog_errno(status);
1490                 goto bail;
1491         }
1492
1493         if (target_exists) {
1494                 if (S_ISDIR(new_inode->i_mode)) {
1495                         if (new_inode->i_nlink != 2 ||
1496                             !ocfs2_empty_dir(new_inode)) {
1497                                 status = -ENOTEMPTY;
1498                                 goto bail;
1499                         }
1500                 }
1501                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1502                                                  newfe_bh,
1503                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1504                 if (status < 0) {
1505                         mlog_errno(status);
1506                         goto bail;
1507                 }
1508
1509                 /* change the dirent to point to the correct inode */
1510                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1511                                             old_inode);
1512                 if (status < 0) {
1513                         mlog_errno(status);
1514                         goto bail;
1515                 }
1516                 inode_inc_iversion(new_dir);
1517
1518                 if (S_ISDIR(new_inode->i_mode))
1519                         ocfs2_set_links_count(newfe, 0);
1520                 else
1521                         ocfs2_add_links_count(newfe, -1);
1522                 ocfs2_journal_dirty(handle, newfe_bh);
1523                 if (should_add_orphan) {
1524                         status = ocfs2_orphan_add(osb, handle, new_inode,
1525                                         newfe_bh, orphan_name,
1526                                         &orphan_insert, orphan_dir, false);
1527                         if (status < 0) {
1528                                 mlog_errno(status);
1529                                 goto bail;
1530                         }
1531                 }
1532         } else {
1533                 /* if the name was not found in new_dir, add it now */
1534                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1535                                          OCFS2_I(old_inode)->ip_blkno,
1536                                          new_dir_bh, &target_insert);
1537                 if (status < 0) {
1538                         mlog_errno(status);
1539                         goto bail;
1540                 }
1541         }
1542
1543         inode_set_ctime_current(old_inode);
1544         mark_inode_dirty(old_inode);
1545
1546         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1547                                          old_inode_bh,
1548                                          OCFS2_JOURNAL_ACCESS_WRITE);
1549         if (status >= 0) {
1550                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1551
1552                 old_di->i_ctime = cpu_to_le64(inode_get_ctime_sec(old_inode));
1553                 old_di->i_ctime_nsec = cpu_to_le32(inode_get_ctime_nsec(old_inode));
1554                 ocfs2_journal_dirty(handle, old_inode_bh);
1555         } else
1556                 mlog_errno(status);
1557
1558         /*
1559          * Now that the name has been added to new_dir, remove the old name.
1560          *
1561          * We don't keep any directory entry context around until now
1562          * because the insert might have changed the type of directory
1563          * we're dealing with.
1564          */
1565         status = ocfs2_find_entry(old_dentry->d_name.name,
1566                                   old_dentry->d_name.len, old_dir,
1567                                   &old_entry_lookup);
1568         if (status) {
1569                 if (!is_journal_aborted(osb->journal->j_journal)) {
1570                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1571                                         "is not deleted.",
1572                                         new_dentry->d_name.len, new_dentry->d_name.name,
1573                                         old_dentry->d_name.len, old_dentry->d_name.name);
1574                 }
1575                 goto bail;
1576         }
1577
1578         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1579         if (status < 0) {
1580                 mlog_errno(status);
1581                 if (!is_journal_aborted(osb->journal->j_journal)) {
1582                         ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1583                                         "is not deleted.",
1584                                         new_dentry->d_name.len, new_dentry->d_name.name,
1585                                         old_dentry->d_name.len, old_dentry->d_name.name);
1586                 }
1587                 goto bail;
1588         }
1589
1590         if (new_inode) {
1591                 drop_nlink(new_inode);
1592                 inode_set_ctime_current(new_inode);
1593         }
1594         inode_set_mtime_to_ts(old_dir, inode_set_ctime_current(old_dir));
1595
1596         if (update_dot_dot) {
1597                 status = ocfs2_update_entry(old_inode, handle,
1598                                             &old_inode_dot_dot_res, new_dir);
1599                 if (status < 0) {
1600                         mlog_errno(status);
1601                         goto bail;
1602                 }
1603         }
1604
1605         if (S_ISDIR(old_inode->i_mode)) {
1606                 drop_nlink(old_dir);
1607                 if (new_inode) {
1608                         drop_nlink(new_inode);
1609                 } else {
1610                         inc_nlink(new_dir);
1611                         mark_inode_dirty(new_dir);
1612                 }
1613         }
1614         mark_inode_dirty(old_dir);
1615         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1616         if (new_inode) {
1617                 mark_inode_dirty(new_inode);
1618                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1619         }
1620
1621         if (old_dir != new_dir) {
1622                 /* Keep the same times on both directories.*/
1623                 inode_set_mtime_to_ts(new_dir,
1624                                       inode_set_ctime_to_ts(new_dir, inode_get_ctime(old_dir)));
1625
1626                 /*
1627                  * This will also pick up the i_nlink change from the
1628                  * block above.
1629                  */
1630                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1631         }
1632
1633         if (old_dir_nlink != old_dir->i_nlink) {
1634                 if (!old_dir_bh) {
1635                         mlog(ML_ERROR, "need to change nlink for old dir "
1636                              "%llu from %d to %d but bh is NULL!\n",
1637                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1638                              (int)old_dir_nlink, old_dir->i_nlink);
1639                 } else {
1640                         struct ocfs2_dinode *fe;
1641                         status = ocfs2_journal_access_di(handle,
1642                                                          INODE_CACHE(old_dir),
1643                                                          old_dir_bh,
1644                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1645                         if (status < 0) {
1646                                 mlog_errno(status);
1647                                 goto bail;
1648                         }
1649                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1650                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1651                         ocfs2_journal_dirty(handle, old_dir_bh);
1652                 }
1653         }
1654         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1655         status = 0;
1656 bail:
1657         if (handle)
1658                 ocfs2_commit_trans(osb, handle);
1659
1660         if (orphan_dir) {
1661                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1662                 ocfs2_inode_unlock(orphan_dir, 1);
1663                 inode_unlock(orphan_dir);
1664                 iput(orphan_dir);
1665         }
1666
1667         if (new_child_locked)
1668                 ocfs2_inode_unlock(new_inode, 1);
1669
1670         if (old_child_locked)
1671                 ocfs2_inode_unlock(old_inode, 1);
1672
1673         if (parents_locked)
1674                 ocfs2_double_unlock(old_dir, new_dir);
1675
1676         if (rename_lock)
1677                 ocfs2_rename_unlock(osb);
1678
1679         if (new_inode)
1680                 sync_mapping_buffers(old_inode->i_mapping);
1681
1682         iput(new_inode);
1683
1684         ocfs2_free_dir_lookup_result(&target_lookup_res);
1685         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1686         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1687         ocfs2_free_dir_lookup_result(&orphan_insert);
1688         ocfs2_free_dir_lookup_result(&target_insert);
1689
1690         brelse(newfe_bh);
1691         brelse(old_inode_bh);
1692         brelse(old_dir_bh);
1693         brelse(new_dir_bh);
1694
1695         if (status)
1696                 mlog_errno(status);
1697
1698         return status;
1699 }
1700
1701 /*
1702  * we expect i_size = strlen(symname). Copy symname into the file
1703  * data, including the null terminator.
1704  */
1705 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1706                                      handle_t *handle,
1707                                      struct inode *inode,
1708                                      const char *symname)
1709 {
1710         struct buffer_head **bhs = NULL;
1711         const char *c;
1712         struct super_block *sb = osb->sb;
1713         u64 p_blkno, p_blocks;
1714         int virtual, blocks, status, i, bytes_left;
1715
1716         bytes_left = i_size_read(inode) + 1;
1717         /* we can't trust i_blocks because we're actually going to
1718          * write i_size + 1 bytes. */
1719         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1720
1721         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1722                                         i_size_read(inode), blocks);
1723
1724         /* Sanity check -- make sure we're going to fit. */
1725         if (bytes_left >
1726             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1727                 status = -EIO;
1728                 mlog_errno(status);
1729                 goto bail;
1730         }
1731
1732         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1733         if (!bhs) {
1734                 status = -ENOMEM;
1735                 mlog_errno(status);
1736                 goto bail;
1737         }
1738
1739         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1740                                              NULL);
1741         if (status < 0) {
1742                 mlog_errno(status);
1743                 goto bail;
1744         }
1745
1746         /* links can never be larger than one cluster so we know this
1747          * is all going to be contiguous, but do a sanity check
1748          * anyway. */
1749         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1750                 status = -EIO;
1751                 mlog_errno(status);
1752                 goto bail;
1753         }
1754
1755         virtual = 0;
1756         while(bytes_left > 0) {
1757                 c = &symname[virtual * sb->s_blocksize];
1758
1759                 bhs[virtual] = sb_getblk(sb, p_blkno);
1760                 if (!bhs[virtual]) {
1761                         status = -ENOMEM;
1762                         mlog_errno(status);
1763                         goto bail;
1764                 }
1765                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1766                                               bhs[virtual]);
1767
1768                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1769                                               bhs[virtual],
1770                                               OCFS2_JOURNAL_ACCESS_CREATE);
1771                 if (status < 0) {
1772                         mlog_errno(status);
1773                         goto bail;
1774                 }
1775
1776                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1777
1778                 memcpy(bhs[virtual]->b_data, c,
1779                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1780                        bytes_left);
1781
1782                 ocfs2_journal_dirty(handle, bhs[virtual]);
1783
1784                 virtual++;
1785                 p_blkno++;
1786                 bytes_left -= sb->s_blocksize;
1787         }
1788
1789         status = 0;
1790 bail:
1791
1792         if (bhs) {
1793                 for(i = 0; i < blocks; i++)
1794                         brelse(bhs[i]);
1795                 kfree(bhs);
1796         }
1797
1798         if (status)
1799                 mlog_errno(status);
1800         return status;
1801 }
1802
1803 static int ocfs2_symlink(struct mnt_idmap *idmap,
1804                          struct inode *dir,
1805                          struct dentry *dentry,
1806                          const char *symname)
1807 {
1808         int status, l, credits;
1809         u64 newsize;
1810         struct ocfs2_super *osb = NULL;
1811         struct inode *inode = NULL;
1812         struct super_block *sb;
1813         struct buffer_head *new_fe_bh = NULL;
1814         struct buffer_head *parent_fe_bh = NULL;
1815         struct ocfs2_dinode *fe = NULL;
1816         struct ocfs2_dinode *dirfe;
1817         handle_t *handle = NULL;
1818         struct ocfs2_alloc_context *inode_ac = NULL;
1819         struct ocfs2_alloc_context *data_ac = NULL;
1820         struct ocfs2_alloc_context *xattr_ac = NULL;
1821         int want_clusters = 0;
1822         int xattr_credits = 0;
1823         struct ocfs2_security_xattr_info si = {
1824                 .name = NULL,
1825                 .enable = 1,
1826         };
1827         int did_quota = 0, did_quota_inode = 0;
1828         struct ocfs2_dir_lookup_result lookup = { NULL, };
1829         sigset_t oldset;
1830         int did_block_signals = 0;
1831         struct ocfs2_dentry_lock *dl = NULL;
1832
1833         trace_ocfs2_symlink_begin(dir, dentry, symname,
1834                                   dentry->d_name.len, dentry->d_name.name);
1835
1836         status = dquot_initialize(dir);
1837         if (status) {
1838                 mlog_errno(status);
1839                 goto bail;
1840         }
1841
1842         sb = dir->i_sb;
1843         osb = OCFS2_SB(sb);
1844
1845         l = strlen(symname) + 1;
1846
1847         credits = ocfs2_calc_symlink_credits(sb);
1848
1849         /* lock the parent directory */
1850         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1851         if (status < 0) {
1852                 if (status != -ENOENT)
1853                         mlog_errno(status);
1854                 return status;
1855         }
1856
1857         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1858         if (!ocfs2_read_links_count(dirfe)) {
1859                 /* can't make a file in a deleted directory. */
1860                 status = -ENOENT;
1861                 goto bail;
1862         }
1863
1864         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1865                                            dentry->d_name.len);
1866         if (status)
1867                 goto bail;
1868
1869         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1870                                               dentry->d_name.name,
1871                                               dentry->d_name.len, &lookup);
1872         if (status < 0) {
1873                 mlog_errno(status);
1874                 goto bail;
1875         }
1876
1877         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1878         if (status < 0) {
1879                 if (status != -ENOSPC)
1880                         mlog_errno(status);
1881                 goto bail;
1882         }
1883
1884         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1885         if (IS_ERR(inode)) {
1886                 status = PTR_ERR(inode);
1887                 inode = NULL;
1888                 mlog_errno(status);
1889                 goto bail;
1890         }
1891
1892         /* get security xattr */
1893         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1894         if (status) {
1895                 if (status == -EOPNOTSUPP)
1896                         si.enable = 0;
1897                 else {
1898                         mlog_errno(status);
1899                         goto bail;
1900                 }
1901         }
1902
1903         /* calculate meta data/clusters for setting security xattr */
1904         if (si.enable) {
1905                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1906                                                   &xattr_credits, &xattr_ac);
1907                 if (status < 0) {
1908                         mlog_errno(status);
1909                         goto bail;
1910                 }
1911         }
1912
1913         /* don't reserve bitmap space for fast symlinks. */
1914         if (l > ocfs2_fast_symlink_chars(sb))
1915                 want_clusters += 1;
1916
1917         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1918         if (status < 0) {
1919                 if (status != -ENOSPC)
1920                         mlog_errno(status);
1921                 goto bail;
1922         }
1923
1924         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1925         if (IS_ERR(handle)) {
1926                 status = PTR_ERR(handle);
1927                 handle = NULL;
1928                 mlog_errno(status);
1929                 goto bail;
1930         }
1931
1932         /* Starting to change things, restart is no longer possible. */
1933         ocfs2_block_signals(&oldset);
1934         did_block_signals = 1;
1935
1936         status = dquot_alloc_inode(inode);
1937         if (status)
1938                 goto bail;
1939         did_quota_inode = 1;
1940
1941         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1942                                    dentry->d_name.name,
1943                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1944                                    inode->i_mode);
1945
1946         status = ocfs2_mknod_locked(osb, dir, inode,
1947                                     0, &new_fe_bh, parent_fe_bh, handle,
1948                                     inode_ac);
1949         if (status < 0) {
1950                 mlog_errno(status);
1951                 goto bail;
1952         }
1953
1954         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1955         inode->i_rdev = 0;
1956         newsize = l - 1;
1957         inode->i_op = &ocfs2_symlink_inode_operations;
1958         inode_nohighmem(inode);
1959         if (l > ocfs2_fast_symlink_chars(sb)) {
1960                 u32 offset = 0;
1961
1962                 status = dquot_alloc_space_nodirty(inode,
1963                     ocfs2_clusters_to_bytes(osb->sb, 1));
1964                 if (status)
1965                         goto bail;
1966                 did_quota = 1;
1967                 inode->i_mapping->a_ops = &ocfs2_aops;
1968                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1969                                               new_fe_bh,
1970                                               handle, data_ac, NULL,
1971                                               NULL);
1972                 if (status < 0) {
1973                         if (status != -ENOSPC && status != -EINTR) {
1974                                 mlog(ML_ERROR,
1975                                      "Failed to extend file to %llu\n",
1976                                      (unsigned long long)newsize);
1977                                 mlog_errno(status);
1978                                 status = -ENOSPC;
1979                         }
1980                         goto bail;
1981                 }
1982                 i_size_write(inode, newsize);
1983                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1984         } else {
1985                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1986                 memcpy((char *) fe->id2.i_symlink, symname, l);
1987                 i_size_write(inode, newsize);
1988                 inode->i_blocks = 0;
1989         }
1990
1991         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1992         if (status < 0) {
1993                 mlog_errno(status);
1994                 goto bail;
1995         }
1996
1997         if (!ocfs2_inode_is_fast_symlink(inode)) {
1998                 status = ocfs2_create_symlink_data(osb, handle, inode,
1999                                                    symname);
2000                 if (status < 0) {
2001                         mlog_errno(status);
2002                         goto bail;
2003                 }
2004         }
2005
2006         if (si.enable) {
2007                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
2008                                                  xattr_ac, data_ac);
2009                 if (status < 0) {
2010                         mlog_errno(status);
2011                         goto bail;
2012                 }
2013         }
2014
2015         /*
2016          * Do this before adding the entry to the directory. We add
2017          * also set d_op after success so that ->d_iput() will cleanup
2018          * the dentry lock even if ocfs2_add_entry() fails below.
2019          */
2020         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2021         if (status) {
2022                 mlog_errno(status);
2023                 goto bail;
2024         }
2025
2026         dl = dentry->d_fsdata;
2027
2028         status = ocfs2_add_entry(handle, dentry, inode,
2029                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
2030                                  &lookup);
2031         if (status < 0) {
2032                 mlog_errno(status);
2033                 goto bail;
2034         }
2035
2036         insert_inode_hash(inode);
2037         d_instantiate(dentry, inode);
2038 bail:
2039         if (status < 0 && did_quota)
2040                 dquot_free_space_nodirty(inode,
2041                                         ocfs2_clusters_to_bytes(osb->sb, 1));
2042         if (status < 0 && did_quota_inode)
2043                 dquot_free_inode(inode);
2044         if (handle) {
2045                 if (status < 0 && fe)
2046                         ocfs2_set_links_count(fe, 0);
2047                 ocfs2_commit_trans(osb, handle);
2048         }
2049
2050         ocfs2_inode_unlock(dir, 1);
2051         if (did_block_signals)
2052                 ocfs2_unblock_signals(&oldset);
2053
2054         brelse(new_fe_bh);
2055         brelse(parent_fe_bh);
2056         kfree(si.value);
2057         ocfs2_free_dir_lookup_result(&lookup);
2058         if (inode_ac)
2059                 ocfs2_free_alloc_context(inode_ac);
2060         if (data_ac)
2061                 ocfs2_free_alloc_context(data_ac);
2062         if (xattr_ac)
2063                 ocfs2_free_alloc_context(xattr_ac);
2064         if ((status < 0) && inode) {
2065                 if (dl)
2066                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2067
2068                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2069                 clear_nlink(inode);
2070                 iput(inode);
2071         }
2072
2073         if (status)
2074                 mlog_errno(status);
2075
2076         return status;
2077 }
2078
2079 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2080 {
2081         int status, namelen;
2082
2083         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2084                            (long long)blkno);
2085         if (namelen <= 0) {
2086                 if (namelen)
2087                         status = namelen;
2088                 else
2089                         status = -EINVAL;
2090                 mlog_errno(status);
2091                 goto bail;
2092         }
2093         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2094                 status = -EINVAL;
2095                 mlog_errno(status);
2096                 goto bail;
2097         }
2098
2099         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2100
2101         status = 0;
2102 bail:
2103         if (status < 0)
2104                 mlog_errno(status);
2105         return status;
2106 }
2107
2108 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2109                                         struct inode **ret_orphan_dir,
2110                                         struct buffer_head **ret_orphan_dir_bh)
2111 {
2112         struct inode *orphan_dir_inode;
2113         struct buffer_head *orphan_dir_bh = NULL;
2114         int ret = 0;
2115
2116         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2117                                                        ORPHAN_DIR_SYSTEM_INODE,
2118                                                        osb->slot_num);
2119         if (!orphan_dir_inode) {
2120                 ret = -ENOENT;
2121                 mlog_errno(ret);
2122                 return ret;
2123         }
2124
2125         inode_lock(orphan_dir_inode);
2126
2127         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2128         if (ret < 0) {
2129                 inode_unlock(orphan_dir_inode);
2130                 iput(orphan_dir_inode);
2131
2132                 mlog_errno(ret);
2133                 return ret;
2134         }
2135
2136         *ret_orphan_dir = orphan_dir_inode;
2137         *ret_orphan_dir_bh = orphan_dir_bh;
2138
2139         return 0;
2140 }
2141
2142 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2143                                       struct buffer_head *orphan_dir_bh,
2144                                       u64 blkno,
2145                                       char *name,
2146                                       struct ocfs2_dir_lookup_result *lookup,
2147                                       bool dio)
2148 {
2149         int ret;
2150         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2151         int namelen = dio ?
2152                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2153                         OCFS2_ORPHAN_NAMELEN;
2154
2155         if (dio) {
2156                 ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2157                                 OCFS2_DIO_ORPHAN_PREFIX);
2158                 if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2159                         ret = -EINVAL;
2160                         mlog_errno(ret);
2161                         return ret;
2162                 }
2163
2164                 ret = ocfs2_blkno_stringify(blkno,
2165                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2166         } else
2167                 ret = ocfs2_blkno_stringify(blkno, name);
2168         if (ret < 0) {
2169                 mlog_errno(ret);
2170                 return ret;
2171         }
2172
2173         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2174                                            orphan_dir_bh, name,
2175                                            namelen, lookup);
2176         if (ret < 0) {
2177                 mlog_errno(ret);
2178                 return ret;
2179         }
2180
2181         return 0;
2182 }
2183
2184 /**
2185  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2186  * insertion of an orphan.
2187  * @osb: ocfs2 file system
2188  * @ret_orphan_dir: Orphan dir inode - returned locked!
2189  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2190  * @lookup: dir lookup result, to be passed back into functions like
2191  *          ocfs2_orphan_add
2192  *
2193  * Returns zero on success and the ret_orphan_dir, name and lookup
2194  * fields will be populated.
2195  *
2196  * Returns non-zero on failure.
2197  */
2198 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2199                                     struct inode **ret_orphan_dir,
2200                                     u64 blkno,
2201                                     char *name,
2202                                     struct ocfs2_dir_lookup_result *lookup,
2203                                     bool dio)
2204 {
2205         struct inode *orphan_dir_inode = NULL;
2206         struct buffer_head *orphan_dir_bh = NULL;
2207         int ret = 0;
2208
2209         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2210                                            &orphan_dir_bh);
2211         if (ret < 0) {
2212                 mlog_errno(ret);
2213                 return ret;
2214         }
2215
2216         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2217                                          blkno, name, lookup, dio);
2218         if (ret < 0) {
2219                 mlog_errno(ret);
2220                 goto out;
2221         }
2222
2223         *ret_orphan_dir = orphan_dir_inode;
2224
2225 out:
2226         brelse(orphan_dir_bh);
2227
2228         if (ret) {
2229                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2230                 inode_unlock(orphan_dir_inode);
2231                 iput(orphan_dir_inode);
2232         }
2233
2234         if (ret)
2235                 mlog_errno(ret);
2236         return ret;
2237 }
2238
2239 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2240                             handle_t *handle,
2241                             struct inode *inode,
2242                             struct buffer_head *fe_bh,
2243                             char *name,
2244                             struct ocfs2_dir_lookup_result *lookup,
2245                             struct inode *orphan_dir_inode,
2246                             bool dio)
2247 {
2248         struct buffer_head *orphan_dir_bh = NULL;
2249         int status = 0;
2250         struct ocfs2_dinode *orphan_fe;
2251         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2252         int namelen = dio ?
2253                         (OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2254                         OCFS2_ORPHAN_NAMELEN;
2255
2256         trace_ocfs2_orphan_add_begin(
2257                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2258
2259         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2260         if (status < 0) {
2261                 mlog_errno(status);
2262                 goto leave;
2263         }
2264
2265         status = ocfs2_journal_access_di(handle,
2266                                          INODE_CACHE(orphan_dir_inode),
2267                                          orphan_dir_bh,
2268                                          OCFS2_JOURNAL_ACCESS_WRITE);
2269         if (status < 0) {
2270                 mlog_errno(status);
2271                 goto leave;
2272         }
2273
2274         /*
2275          * We're going to journal the change of i_flags and i_orphaned_slot.
2276          * It's safe anyway, though some callers may duplicate the journaling.
2277          * Journaling within the func just make the logic look more
2278          * straightforward.
2279          */
2280         status = ocfs2_journal_access_di(handle,
2281                                          INODE_CACHE(inode),
2282                                          fe_bh,
2283                                          OCFS2_JOURNAL_ACCESS_WRITE);
2284         if (status < 0) {
2285                 mlog_errno(status);
2286                 goto leave;
2287         }
2288
2289         /* we're a cluster, and nlink can change on disk from
2290          * underneath us... */
2291         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2292         if (S_ISDIR(inode->i_mode))
2293                 ocfs2_add_links_count(orphan_fe, 1);
2294         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2295         ocfs2_journal_dirty(handle, orphan_dir_bh);
2296
2297         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2298                                    namelen, inode,
2299                                    OCFS2_I(inode)->ip_blkno,
2300                                    orphan_dir_bh, lookup);
2301         if (status < 0) {
2302                 mlog_errno(status);
2303                 goto rollback;
2304         }
2305
2306         if (dio) {
2307                 /* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2308                  * slot.
2309                  */
2310                 fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2311                 fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2312         } else {
2313                 fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2314                 OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2315
2316                 /* Record which orphan dir our inode now resides
2317                  * in. delete_inode will use this to determine which orphan
2318                  * dir to lock. */
2319                 fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2320         }
2321
2322         ocfs2_journal_dirty(handle, fe_bh);
2323
2324         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2325                                    osb->slot_num);
2326
2327 rollback:
2328         if (status < 0) {
2329                 if (S_ISDIR(inode->i_mode))
2330                         ocfs2_add_links_count(orphan_fe, -1);
2331                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2332         }
2333
2334 leave:
2335         brelse(orphan_dir_bh);
2336
2337         return status;
2338 }
2339
2340 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2341 int ocfs2_orphan_del(struct ocfs2_super *osb,
2342                      handle_t *handle,
2343                      struct inode *orphan_dir_inode,
2344                      struct inode *inode,
2345                      struct buffer_head *orphan_dir_bh,
2346                      bool dio)
2347 {
2348         char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2349         struct ocfs2_dinode *orphan_fe;
2350         int status = 0;
2351         struct ocfs2_dir_lookup_result lookup = { NULL, };
2352
2353         if (dio) {
2354                 status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2355                                 OCFS2_DIO_ORPHAN_PREFIX);
2356                 if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2357                         status = -EINVAL;
2358                         mlog_errno(status);
2359                         return status;
2360                 }
2361
2362                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2363                                 name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2364         } else
2365                 status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2366         if (status < 0) {
2367                 mlog_errno(status);
2368                 goto leave;
2369         }
2370
2371         trace_ocfs2_orphan_del(
2372              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2373              name, strlen(name));
2374
2375         status = ocfs2_journal_access_di(handle,
2376                                          INODE_CACHE(orphan_dir_inode),
2377                                          orphan_dir_bh,
2378                                          OCFS2_JOURNAL_ACCESS_WRITE);
2379         if (status < 0) {
2380                 mlog_errno(status);
2381                 goto leave;
2382         }
2383
2384         /* find it's spot in the orphan directory */
2385         status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2386                                   &lookup);
2387         if (status) {
2388                 mlog_errno(status);
2389                 goto leave;
2390         }
2391
2392         /* remove it from the orphan directory */
2393         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2394         if (status < 0) {
2395                 mlog_errno(status);
2396                 goto leave;
2397         }
2398
2399         /* do the i_nlink dance! :) */
2400         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2401         if (S_ISDIR(inode->i_mode))
2402                 ocfs2_add_links_count(orphan_fe, -1);
2403         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2404         ocfs2_journal_dirty(handle, orphan_dir_bh);
2405
2406 leave:
2407         ocfs2_free_dir_lookup_result(&lookup);
2408
2409         if (status)
2410                 mlog_errno(status);
2411         return status;
2412 }
2413
2414 /**
2415  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2416  * allocated file. This is different from the typical 'add to orphan dir'
2417  * operation in that the inode does not yet exist. This is a problem because
2418  * the orphan dir stringifies the inode block number to come up with it's
2419  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2420  * problem. This function works around it by calling deeper into the orphan
2421  * and suballoc code than other callers. Use this only by necessity.
2422  * @dir: The directory which this inode will ultimately wind up under - not the
2423  * orphan dir!
2424  * @dir_bh: buffer_head the @dir inode block
2425  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2426  * with the string to be used for orphan dirent. Pass back to the orphan dir
2427  * code.
2428  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2429  * dir code.
2430  * @ret_di_blkno: block number where the new inode will be allocated.
2431  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2432  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2433  *
2434  * Returns zero on success and the ret_orphan_dir, name and lookup
2435  * fields will be populated.
2436  *
2437  * Returns non-zero on failure.
2438  */
2439 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2440                                         struct buffer_head *dir_bh,
2441                                         char *orphan_name,
2442                                         struct inode **ret_orphan_dir,
2443                                         u64 *ret_di_blkno,
2444                                         struct ocfs2_dir_lookup_result *orphan_insert,
2445                                         struct ocfs2_alloc_context **ret_inode_ac)
2446 {
2447         int ret;
2448         u64 di_blkno;
2449         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2450         struct inode *orphan_dir = NULL;
2451         struct buffer_head *orphan_dir_bh = NULL;
2452         struct ocfs2_alloc_context *inode_ac = NULL;
2453
2454         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2455         if (ret < 0) {
2456                 mlog_errno(ret);
2457                 return ret;
2458         }
2459
2460         /* reserve an inode spot */
2461         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2462         if (ret < 0) {
2463                 if (ret != -ENOSPC)
2464                         mlog_errno(ret);
2465                 goto out;
2466         }
2467
2468         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2469                                        &di_blkno);
2470         if (ret) {
2471                 mlog_errno(ret);
2472                 goto out;
2473         }
2474
2475         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2476                                          di_blkno, orphan_name, orphan_insert,
2477                                          false);
2478         if (ret < 0) {
2479                 mlog_errno(ret);
2480                 goto out;
2481         }
2482
2483 out:
2484         if (ret == 0) {
2485                 *ret_orphan_dir = orphan_dir;
2486                 *ret_di_blkno = di_blkno;
2487                 *ret_inode_ac = inode_ac;
2488                 /*
2489                  * orphan_name and orphan_insert are already up to
2490                  * date via prepare_orphan_dir
2491                  */
2492         } else {
2493                 /* Unroll reserve_new_inode* */
2494                 if (inode_ac)
2495                         ocfs2_free_alloc_context(inode_ac);
2496
2497                 /* Unroll orphan dir locking */
2498                 inode_unlock(orphan_dir);
2499                 ocfs2_inode_unlock(orphan_dir, 1);
2500                 iput(orphan_dir);
2501         }
2502
2503         brelse(orphan_dir_bh);
2504
2505         return ret;
2506 }
2507
2508 int ocfs2_create_inode_in_orphan(struct inode *dir,
2509                                  int mode,
2510                                  struct inode **new_inode)
2511 {
2512         int status, did_quota_inode = 0;
2513         struct inode *inode = NULL;
2514         struct inode *orphan_dir = NULL;
2515         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2516         handle_t *handle = NULL;
2517         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2518         struct buffer_head *parent_di_bh = NULL;
2519         struct buffer_head *new_di_bh = NULL;
2520         struct ocfs2_alloc_context *inode_ac = NULL;
2521         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2522         u64 di_blkno, suballoc_loc;
2523         u16 suballoc_bit;
2524
2525         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2526         if (status < 0) {
2527                 if (status != -ENOENT)
2528                         mlog_errno(status);
2529                 return status;
2530         }
2531
2532         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2533                                               orphan_name, &orphan_dir,
2534                                               &di_blkno, &orphan_insert, &inode_ac);
2535         if (status < 0) {
2536                 if (status != -ENOSPC)
2537                         mlog_errno(status);
2538                 goto leave;
2539         }
2540
2541         inode = ocfs2_get_init_inode(dir, mode);
2542         if (IS_ERR(inode)) {
2543                 status = PTR_ERR(inode);
2544                 inode = NULL;
2545                 mlog_errno(status);
2546                 goto leave;
2547         }
2548
2549         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2550         if (IS_ERR(handle)) {
2551                 status = PTR_ERR(handle);
2552                 handle = NULL;
2553                 mlog_errno(status);
2554                 goto leave;
2555         }
2556
2557         status = dquot_alloc_inode(inode);
2558         if (status)
2559                 goto leave;
2560         did_quota_inode = 1;
2561
2562         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2563                                               &suballoc_loc,
2564                                               &suballoc_bit, di_blkno);
2565         if (status < 0) {
2566                 mlog_errno(status);
2567                 goto leave;
2568         }
2569
2570         clear_nlink(inode);
2571         /* do the real work now. */
2572         status = __ocfs2_mknod_locked(dir, inode,
2573                                       0, &new_di_bh, parent_di_bh, handle,
2574                                       inode_ac, di_blkno, suballoc_loc,
2575                                       suballoc_bit);
2576         if (status < 0) {
2577                 mlog_errno(status);
2578                 goto leave;
2579         }
2580
2581         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2582                                   &orphan_insert, orphan_dir, false);
2583         if (status < 0) {
2584                 mlog_errno(status);
2585                 goto leave;
2586         }
2587
2588         /* get open lock so that only nodes can't remove it from orphan dir. */
2589         status = ocfs2_open_lock(inode);
2590         if (status < 0)
2591                 mlog_errno(status);
2592
2593         insert_inode_hash(inode);
2594 leave:
2595         if (status < 0 && did_quota_inode)
2596                 dquot_free_inode(inode);
2597         if (handle)
2598                 ocfs2_commit_trans(osb, handle);
2599
2600         if (orphan_dir) {
2601                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2602                 ocfs2_inode_unlock(orphan_dir, 1);
2603                 inode_unlock(orphan_dir);
2604                 iput(orphan_dir);
2605         }
2606
2607         if ((status < 0) && inode) {
2608                 clear_nlink(inode);
2609                 iput(inode);
2610         }
2611
2612         if (inode_ac)
2613                 ocfs2_free_alloc_context(inode_ac);
2614
2615         brelse(new_di_bh);
2616
2617         if (!status)
2618                 *new_inode = inode;
2619
2620         ocfs2_free_dir_lookup_result(&orphan_insert);
2621
2622         ocfs2_inode_unlock(dir, 1);
2623         brelse(parent_di_bh);
2624         return status;
2625 }
2626
2627 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2628         struct inode *inode)
2629 {
2630         char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2631         struct inode *orphan_dir_inode = NULL;
2632         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2633         struct buffer_head *di_bh = NULL;
2634         int status = 0;
2635         handle_t *handle = NULL;
2636         struct ocfs2_dinode *di = NULL;
2637
2638         status = ocfs2_inode_lock(inode, &di_bh, 1);
2639         if (status < 0) {
2640                 mlog_errno(status);
2641                 goto bail;
2642         }
2643
2644         di = (struct ocfs2_dinode *) di_bh->b_data;
2645         /*
2646          * Another append dio crashed?
2647          * If so, manually recover it first.
2648          */
2649         if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2650                 status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2651                 if (status < 0) {
2652                         if (status != -ENOSPC)
2653                                 mlog_errno(status);
2654                         goto bail_unlock_inode;
2655                 }
2656
2657                 status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2658                 if (status < 0) {
2659                         mlog_errno(status);
2660                         goto bail_unlock_inode;
2661                 }
2662         }
2663
2664         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2665                         OCFS2_I(inode)->ip_blkno,
2666                         orphan_name,
2667                         &orphan_insert,
2668                         true);
2669         if (status < 0) {
2670                 mlog_errno(status);
2671                 goto bail_unlock_inode;
2672         }
2673
2674         handle = ocfs2_start_trans(osb,
2675                         OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2676         if (IS_ERR(handle)) {
2677                 status = PTR_ERR(handle);
2678                 goto bail_unlock_orphan;
2679         }
2680
2681         status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2682                         &orphan_insert, orphan_dir_inode, true);
2683         if (status)
2684                 mlog_errno(status);
2685
2686         ocfs2_commit_trans(osb, handle);
2687
2688 bail_unlock_orphan:
2689         ocfs2_inode_unlock(orphan_dir_inode, 1);
2690         inode_unlock(orphan_dir_inode);
2691         iput(orphan_dir_inode);
2692
2693         ocfs2_free_dir_lookup_result(&orphan_insert);
2694
2695 bail_unlock_inode:
2696         ocfs2_inode_unlock(inode, 1);
2697         brelse(di_bh);
2698
2699 bail:
2700         return status;
2701 }
2702
2703 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2704                 struct inode *inode, struct buffer_head *di_bh,
2705                 int update_isize, loff_t end)
2706 {
2707         struct inode *orphan_dir_inode = NULL;
2708         struct buffer_head *orphan_dir_bh = NULL;
2709         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2710         handle_t *handle = NULL;
2711         int status = 0;
2712
2713         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2714                         ORPHAN_DIR_SYSTEM_INODE,
2715                         le16_to_cpu(di->i_dio_orphaned_slot));
2716         if (!orphan_dir_inode) {
2717                 status = -ENOENT;
2718                 mlog_errno(status);
2719                 goto bail;
2720         }
2721
2722         inode_lock(orphan_dir_inode);
2723         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2724         if (status < 0) {
2725                 inode_unlock(orphan_dir_inode);
2726                 iput(orphan_dir_inode);
2727                 mlog_errno(status);
2728                 goto bail;
2729         }
2730
2731         handle = ocfs2_start_trans(osb,
2732                         OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2733         if (IS_ERR(handle)) {
2734                 status = PTR_ERR(handle);
2735                 goto bail_unlock_orphan;
2736         }
2737
2738         BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2739
2740         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2741                                 inode, orphan_dir_bh, true);
2742         if (status < 0) {
2743                 mlog_errno(status);
2744                 goto bail_commit;
2745         }
2746
2747         status = ocfs2_journal_access_di(handle,
2748                         INODE_CACHE(inode),
2749                         di_bh,
2750                         OCFS2_JOURNAL_ACCESS_WRITE);
2751         if (status < 0) {
2752                 mlog_errno(status);
2753                 goto bail_commit;
2754         }
2755
2756         di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2757         di->i_dio_orphaned_slot = 0;
2758
2759         if (update_isize) {
2760                 status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2761                 if (status)
2762                         mlog_errno(status);
2763         } else
2764                 ocfs2_journal_dirty(handle, di_bh);
2765
2766 bail_commit:
2767         ocfs2_commit_trans(osb, handle);
2768
2769 bail_unlock_orphan:
2770         ocfs2_inode_unlock(orphan_dir_inode, 1);
2771         inode_unlock(orphan_dir_inode);
2772         brelse(orphan_dir_bh);
2773         iput(orphan_dir_inode);
2774
2775 bail:
2776         return status;
2777 }
2778
2779 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2780                                    struct inode *inode,
2781                                    struct dentry *dentry)
2782 {
2783         int status = 0;
2784         struct buffer_head *parent_di_bh = NULL;
2785         handle_t *handle = NULL;
2786         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2787         struct ocfs2_dinode *dir_di, *di;
2788         struct inode *orphan_dir_inode = NULL;
2789         struct buffer_head *orphan_dir_bh = NULL;
2790         struct buffer_head *di_bh = NULL;
2791         struct ocfs2_dir_lookup_result lookup = { NULL, };
2792
2793         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2794                                 dentry->d_name.len, dentry->d_name.name,
2795                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2796                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2797
2798         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2799         if (status < 0) {
2800                 if (status != -ENOENT)
2801                         mlog_errno(status);
2802                 return status;
2803         }
2804
2805         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2806         if (!dir_di->i_links_count) {
2807                 /* can't make a file in a deleted directory. */
2808                 status = -ENOENT;
2809                 goto leave;
2810         }
2811
2812         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2813                                            dentry->d_name.len);
2814         if (status)
2815                 goto leave;
2816
2817         /* get a spot inside the dir. */
2818         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2819                                               dentry->d_name.name,
2820                                               dentry->d_name.len, &lookup);
2821         if (status < 0) {
2822                 mlog_errno(status);
2823                 goto leave;
2824         }
2825
2826         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2827                                                        ORPHAN_DIR_SYSTEM_INODE,
2828                                                        osb->slot_num);
2829         if (!orphan_dir_inode) {
2830                 status = -ENOENT;
2831                 mlog_errno(status);
2832                 goto leave;
2833         }
2834
2835         inode_lock(orphan_dir_inode);
2836
2837         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2838         if (status < 0) {
2839                 mlog_errno(status);
2840                 inode_unlock(orphan_dir_inode);
2841                 iput(orphan_dir_inode);
2842                 goto leave;
2843         }
2844
2845         status = ocfs2_read_inode_block(inode, &di_bh);
2846         if (status < 0) {
2847                 mlog_errno(status);
2848                 goto orphan_unlock;
2849         }
2850
2851         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2852         if (IS_ERR(handle)) {
2853                 status = PTR_ERR(handle);
2854                 handle = NULL;
2855                 mlog_errno(status);
2856                 goto orphan_unlock;
2857         }
2858
2859         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2860                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2861         if (status < 0) {
2862                 mlog_errno(status);
2863                 goto out_commit;
2864         }
2865
2866         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2867                                   orphan_dir_bh, false);
2868         if (status < 0) {
2869                 mlog_errno(status);
2870                 goto out_commit;
2871         }
2872
2873         di = (struct ocfs2_dinode *)di_bh->b_data;
2874         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2875         di->i_orphaned_slot = 0;
2876         set_nlink(inode, 1);
2877         ocfs2_set_links_count(di, inode->i_nlink);
2878         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2879         ocfs2_journal_dirty(handle, di_bh);
2880
2881         status = ocfs2_add_entry(handle, dentry, inode,
2882                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2883                                  &lookup);
2884         if (status < 0) {
2885                 mlog_errno(status);
2886                 goto out_commit;
2887         }
2888
2889         status = ocfs2_dentry_attach_lock(dentry, inode,
2890                                           OCFS2_I(dir)->ip_blkno);
2891         if (status) {
2892                 mlog_errno(status);
2893                 goto out_commit;
2894         }
2895
2896         d_instantiate(dentry, inode);
2897         status = 0;
2898 out_commit:
2899         ocfs2_commit_trans(osb, handle);
2900 orphan_unlock:
2901         ocfs2_inode_unlock(orphan_dir_inode, 1);
2902         inode_unlock(orphan_dir_inode);
2903         iput(orphan_dir_inode);
2904 leave:
2905
2906         ocfs2_inode_unlock(dir, 1);
2907
2908         brelse(di_bh);
2909         brelse(parent_di_bh);
2910         brelse(orphan_dir_bh);
2911
2912         ocfs2_free_dir_lookup_result(&lookup);
2913
2914         if (status)
2915                 mlog_errno(status);
2916
2917         return status;
2918 }
2919
2920 const struct inode_operations ocfs2_dir_iops = {
2921         .create         = ocfs2_create,
2922         .lookup         = ocfs2_lookup,
2923         .link           = ocfs2_link,
2924         .unlink         = ocfs2_unlink,
2925         .rmdir          = ocfs2_unlink,
2926         .symlink        = ocfs2_symlink,
2927         .mkdir          = ocfs2_mkdir,
2928         .mknod          = ocfs2_mknod,
2929         .rename         = ocfs2_rename,
2930         .setattr        = ocfs2_setattr,
2931         .getattr        = ocfs2_getattr,
2932         .permission     = ocfs2_permission,
2933         .listxattr      = ocfs2_listxattr,
2934         .fiemap         = ocfs2_fiemap,
2935         .get_inode_acl  = ocfs2_iop_get_acl,
2936         .set_acl        = ocfs2_iop_set_acl,
2937         .fileattr_get   = ocfs2_fileattr_get,
2938         .fileattr_set   = ocfs2_fileattr_set,
2939 };