new helper: ihold()
[linux-2.6.git] / fs / ocfs2 / namei.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * namei.c
5  *
6  * Create and rename file, directory, symlinks
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  *  Portions of this code from linux/fs/ext3/dir.c
11  *
12  *  Copyright (C) 1992, 1993, 1994, 1995
13  *  Remy Card (card@masi.ibp.fr)
14  *  Laboratoire MASI - Institut Blaise pascal
15  *  Universite Pierre et Marie Curie (Paris VI)
16  *
17  *   from
18  *
19  *   linux/fs/minix/dir.c
20  *
21  *   Copyright (C) 1991, 1992 Linux Torvalds
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public
25  * License as published by the Free Software Foundation; either
26  * version 2 of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
31  * General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public
34  * License along with this program; if not, write to the
35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36  * Boston, MA 021110-1307, USA.
37  */
38
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
44
45 #define MLOG_MASK_PREFIX ML_NAMEI
46 #include <cluster/masklog.h>
47
48 #include "ocfs2.h"
49
50 #include "alloc.h"
51 #include "dcache.h"
52 #include "dir.h"
53 #include "dlmglue.h"
54 #include "extent_map.h"
55 #include "file.h"
56 #include "inode.h"
57 #include "journal.h"
58 #include "namei.h"
59 #include "suballoc.h"
60 #include "super.h"
61 #include "symlink.h"
62 #include "sysfile.h"
63 #include "uptodate.h"
64 #include "xattr.h"
65 #include "acl.h"
66
67 #include "buffer_head_io.h"
68
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
70                               struct inode *dir,
71                               struct inode *inode,
72                               dev_t dev,
73                               struct buffer_head **new_fe_bh,
74                               struct buffer_head *parent_fe_bh,
75                               handle_t *handle,
76                               struct ocfs2_alloc_context *inode_ac);
77
78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
79                                     struct inode **ret_orphan_dir,
80                                     u64 blkno,
81                                     char *name,
82                                     struct ocfs2_dir_lookup_result *lookup);
83
84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
85                             handle_t *handle,
86                             struct inode *inode,
87                             struct buffer_head *fe_bh,
88                             char *name,
89                             struct ocfs2_dir_lookup_result *lookup,
90                             struct inode *orphan_dir_inode);
91
92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
93                                      handle_t *handle,
94                                      struct inode *inode,
95                                      const char *symname);
96
97 /* An orphan dir name is an 8 byte value, printed as a hex string */
98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
99
100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
101                                    struct nameidata *nd)
102 {
103         int status;
104         u64 blkno;
105         struct inode *inode = NULL;
106         struct dentry *ret;
107         struct ocfs2_inode_info *oi;
108
109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
110                    dentry->d_name.len, dentry->d_name.name);
111
112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
113                 ret = ERR_PTR(-ENAMETOOLONG);
114                 goto bail;
115         }
116
117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
119
120         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
121         if (status < 0) {
122                 if (status != -ENOENT)
123                         mlog_errno(status);
124                 ret = ERR_PTR(status);
125                 goto bail;
126         }
127
128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
129                                             dentry->d_name.len, &blkno);
130         if (status < 0)
131                 goto bail_add;
132
133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
134         if (IS_ERR(inode)) {
135                 ret = ERR_PTR(-EACCES);
136                 goto bail_unlock;
137         }
138
139         oi = OCFS2_I(inode);
140         /* Clear any orphaned state... If we were able to look up the
141          * inode from a directory, it certainly can't be orphaned. We
142          * might have the bad state from a node which intended to
143          * orphan this inode but crashed before it could commit the
144          * unlink. */
145         spin_lock(&oi->ip_lock);
146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
147         spin_unlock(&oi->ip_lock);
148
149 bail_add:
150         dentry->d_op = &ocfs2_dentry_ops;
151         ret = d_splice_alias(inode, dentry);
152
153         if (inode) {
154                 /*
155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
156                  * dentry, it will d_move() it on top of ourse. The
157                  * return value will indicate this however, so in
158                  * those cases, we switch them around for the locking
159                  * code.
160                  *
161                  * NOTE: This dentry already has ->d_op set from
162                  * ocfs2_get_parent() and ocfs2_get_dentry()
163                  */
164                 if (ret)
165                         dentry = ret;
166
167                 status = ocfs2_dentry_attach_lock(dentry, inode,
168                                                   OCFS2_I(dir)->ip_blkno);
169                 if (status) {
170                         mlog_errno(status);
171                         ret = ERR_PTR(status);
172                         goto bail_unlock;
173                 }
174         } else
175                 ocfs2_dentry_attach_gen(dentry);
176
177 bail_unlock:
178         /* Don't drop the cluster lock until *after* the d_add --
179          * unlink on another node will message us to remove that
180          * dentry under this lock so otherwise we can race this with
181          * the downconvert thread and have a stale dentry. */
182         ocfs2_inode_unlock(dir, 0);
183
184 bail:
185
186         mlog_exit_ptr(ret);
187
188         return ret;
189 }
190
191 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
192 {
193         struct inode *inode;
194
195         inode = new_inode(dir->i_sb);
196         if (!inode) {
197                 mlog(ML_ERROR, "new_inode failed!\n");
198                 return NULL;
199         }
200
201         /* populate as many fields early on as possible - many of
202          * these are used by the support functions here and in
203          * callers. */
204         if (S_ISDIR(mode))
205                 inode->i_nlink = 2;
206         else
207                 inode->i_nlink = 1;
208         inode_init_owner(inode, dir, mode);
209         dquot_initialize(inode);
210         return inode;
211 }
212
213 static int ocfs2_mknod(struct inode *dir,
214                        struct dentry *dentry,
215                        int mode,
216                        dev_t dev)
217 {
218         int status = 0;
219         struct buffer_head *parent_fe_bh = NULL;
220         handle_t *handle = NULL;
221         struct ocfs2_super *osb;
222         struct ocfs2_dinode *dirfe;
223         struct buffer_head *new_fe_bh = NULL;
224         struct inode *inode = NULL;
225         struct ocfs2_alloc_context *inode_ac = NULL;
226         struct ocfs2_alloc_context *data_ac = NULL;
227         struct ocfs2_alloc_context *meta_ac = NULL;
228         int want_clusters = 0;
229         int want_meta = 0;
230         int xattr_credits = 0;
231         struct ocfs2_security_xattr_info si = {
232                 .enable = 1,
233         };
234         int did_quota_inode = 0;
235         struct ocfs2_dir_lookup_result lookup = { NULL, };
236         sigset_t oldset;
237         int did_block_signals = 0;
238
239         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
240                    (unsigned long)dev, dentry->d_name.len,
241                    dentry->d_name.name);
242
243         dquot_initialize(dir);
244
245         /* get our super block */
246         osb = OCFS2_SB(dir->i_sb);
247
248         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
249         if (status < 0) {
250                 if (status != -ENOENT)
251                         mlog_errno(status);
252                 return status;
253         }
254
255         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
256                 status = -EMLINK;
257                 goto leave;
258         }
259
260         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
261         if (!ocfs2_read_links_count(dirfe)) {
262                 /* can't make a file in a deleted directory. */
263                 status = -ENOENT;
264                 goto leave;
265         }
266
267         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
268                                            dentry->d_name.len);
269         if (status)
270                 goto leave;
271
272         /* get a spot inside the dir. */
273         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
274                                               dentry->d_name.name,
275                                               dentry->d_name.len, &lookup);
276         if (status < 0) {
277                 mlog_errno(status);
278                 goto leave;
279         }
280
281         /* reserve an inode spot */
282         status = ocfs2_reserve_new_inode(osb, &inode_ac);
283         if (status < 0) {
284                 if (status != -ENOSPC)
285                         mlog_errno(status);
286                 goto leave;
287         }
288
289         inode = ocfs2_get_init_inode(dir, mode);
290         if (!inode) {
291                 status = -ENOMEM;
292                 mlog_errno(status);
293                 goto leave;
294         }
295
296         /* get security xattr */
297         status = ocfs2_init_security_get(inode, dir, &si);
298         if (status) {
299                 if (status == -EOPNOTSUPP)
300                         si.enable = 0;
301                 else {
302                         mlog_errno(status);
303                         goto leave;
304                 }
305         }
306
307         /* calculate meta data/clusters for setting security and acl xattr */
308         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
309                                        &si, &want_clusters,
310                                        &xattr_credits, &want_meta);
311         if (status < 0) {
312                 mlog_errno(status);
313                 goto leave;
314         }
315
316         /* Reserve a cluster if creating an extent based directory. */
317         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
318                 want_clusters += 1;
319
320                 /* Dir indexing requires extra space as well */
321                 if (ocfs2_supports_indexed_dirs(osb))
322                         want_meta++;
323         }
324
325         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
326         if (status < 0) {
327                 if (status != -ENOSPC)
328                         mlog_errno(status);
329                 goto leave;
330         }
331
332         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
333         if (status < 0) {
334                 if (status != -ENOSPC)
335                         mlog_errno(status);
336                 goto leave;
337         }
338
339         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
340                                                             S_ISDIR(mode),
341                                                             xattr_credits));
342         if (IS_ERR(handle)) {
343                 status = PTR_ERR(handle);
344                 handle = NULL;
345                 mlog_errno(status);
346                 goto leave;
347         }
348
349         /* Starting to change things, restart is no longer possible. */
350         ocfs2_block_signals(&oldset);
351         did_block_signals = 1;
352
353         status = dquot_alloc_inode(inode);
354         if (status)
355                 goto leave;
356         did_quota_inode = 1;
357
358         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
359                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
360                    dentry->d_name.name);
361
362         /* do the real work now. */
363         status = ocfs2_mknod_locked(osb, dir, inode, dev,
364                                     &new_fe_bh, parent_fe_bh, handle,
365                                     inode_ac);
366         if (status < 0) {
367                 mlog_errno(status);
368                 goto leave;
369         }
370
371         if (S_ISDIR(mode)) {
372                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
373                                             new_fe_bh, data_ac, meta_ac);
374                 if (status < 0) {
375                         mlog_errno(status);
376                         goto leave;
377                 }
378
379                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
380                                                  parent_fe_bh,
381                                                  OCFS2_JOURNAL_ACCESS_WRITE);
382                 if (status < 0) {
383                         mlog_errno(status);
384                         goto leave;
385                 }
386                 ocfs2_add_links_count(dirfe, 1);
387                 ocfs2_journal_dirty(handle, parent_fe_bh);
388                 inc_nlink(dir);
389         }
390
391         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
392                                 meta_ac, data_ac);
393         if (status < 0) {
394                 mlog_errno(status);
395                 goto leave;
396         }
397
398         if (si.enable) {
399                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
400                                                  meta_ac, data_ac);
401                 if (status < 0) {
402                         mlog_errno(status);
403                         goto leave;
404                 }
405         }
406
407         /*
408          * Do this before adding the entry to the directory. We add
409          * also set d_op after success so that ->d_iput() will cleanup
410          * the dentry lock even if ocfs2_add_entry() fails below.
411          */
412         status = ocfs2_dentry_attach_lock(dentry, inode,
413                                           OCFS2_I(dir)->ip_blkno);
414         if (status) {
415                 mlog_errno(status);
416                 goto leave;
417         }
418         dentry->d_op = &ocfs2_dentry_ops;
419
420         status = ocfs2_add_entry(handle, dentry, inode,
421                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
422                                  &lookup);
423         if (status < 0) {
424                 mlog_errno(status);
425                 goto leave;
426         }
427
428         insert_inode_hash(inode);
429         d_instantiate(dentry, inode);
430         status = 0;
431 leave:
432         if (status < 0 && did_quota_inode)
433                 dquot_free_inode(inode);
434         if (handle)
435                 ocfs2_commit_trans(osb, handle);
436
437         ocfs2_inode_unlock(dir, 1);
438         if (did_block_signals)
439                 ocfs2_unblock_signals(&oldset);
440
441         if (status == -ENOSPC)
442                 mlog(0, "Disk is full\n");
443
444         brelse(new_fe_bh);
445         brelse(parent_fe_bh);
446         kfree(si.name);
447         kfree(si.value);
448
449         ocfs2_free_dir_lookup_result(&lookup);
450
451         if (inode_ac)
452                 ocfs2_free_alloc_context(inode_ac);
453
454         if (data_ac)
455                 ocfs2_free_alloc_context(data_ac);
456
457         if (meta_ac)
458                 ocfs2_free_alloc_context(meta_ac);
459
460         /*
461          * We should call iput after the i_mutex of the bitmap been
462          * unlocked in ocfs2_free_alloc_context, or the
463          * ocfs2_delete_inode will mutex_lock again.
464          */
465         if ((status < 0) && inode) {
466                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
467                 clear_nlink(inode);
468                 iput(inode);
469         }
470
471         mlog_exit(status);
472
473         return status;
474 }
475
476 static int __ocfs2_mknod_locked(struct inode *dir,
477                                 struct inode *inode,
478                                 dev_t dev,
479                                 struct buffer_head **new_fe_bh,
480                                 struct buffer_head *parent_fe_bh,
481                                 handle_t *handle,
482                                 struct ocfs2_alloc_context *inode_ac,
483                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
484 {
485         int status = 0;
486         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
487         struct ocfs2_dinode *fe = NULL;
488         struct ocfs2_extent_list *fel;
489         u16 feat;
490
491         *new_fe_bh = NULL;
492
493         /* populate as many fields early on as possible - many of
494          * these are used by the support functions here and in
495          * callers. */
496         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
497         OCFS2_I(inode)->ip_blkno = fe_blkno;
498         spin_lock(&osb->osb_lock);
499         inode->i_generation = osb->s_next_generation++;
500         spin_unlock(&osb->osb_lock);
501
502         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
503         if (!*new_fe_bh) {
504                 status = -EIO;
505                 mlog_errno(status);
506                 goto leave;
507         }
508         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
509
510         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
511                                          *new_fe_bh,
512                                          OCFS2_JOURNAL_ACCESS_CREATE);
513         if (status < 0) {
514                 mlog_errno(status);
515                 goto leave;
516         }
517
518         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
519         memset(fe, 0, osb->sb->s_blocksize);
520
521         fe->i_generation = cpu_to_le32(inode->i_generation);
522         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
523         fe->i_blkno = cpu_to_le64(fe_blkno);
524         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
525         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
526         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
527         fe->i_uid = cpu_to_le32(inode->i_uid);
528         fe->i_gid = cpu_to_le32(inode->i_gid);
529         fe->i_mode = cpu_to_le16(inode->i_mode);
530         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
531                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
532
533         ocfs2_set_links_count(fe, inode->i_nlink);
534
535         fe->i_last_eb_blk = 0;
536         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
537         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
538         fe->i_atime = fe->i_ctime = fe->i_mtime =
539                 cpu_to_le64(CURRENT_TIME.tv_sec);
540         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
541                 cpu_to_le32(CURRENT_TIME.tv_nsec);
542         fe->i_dtime = 0;
543
544         /*
545          * If supported, directories start with inline data. If inline
546          * isn't supported, but indexing is, we start them as indexed.
547          */
548         feat = le16_to_cpu(fe->i_dyn_features);
549         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
550                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
551
552                 fe->id2.i_data.id_count = cpu_to_le16(
553                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
554         } else {
555                 fel = &fe->id2.i_list;
556                 fel->l_tree_depth = 0;
557                 fel->l_next_free_rec = 0;
558                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
559         }
560
561         ocfs2_journal_dirty(handle, *new_fe_bh);
562
563         ocfs2_populate_inode(inode, fe, 1);
564         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
565         if (!ocfs2_mount_local(osb)) {
566                 status = ocfs2_create_new_inode_locks(inode);
567                 if (status < 0)
568                         mlog_errno(status);
569         }
570
571         status = 0; /* error in ocfs2_create_new_inode_locks is not
572                      * critical */
573
574 leave:
575         if (status < 0) {
576                 if (*new_fe_bh) {
577                         brelse(*new_fe_bh);
578                         *new_fe_bh = NULL;
579                 }
580         }
581
582         mlog_exit(status);
583         return status;
584 }
585
586 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
587                               struct inode *dir,
588                               struct inode *inode,
589                               dev_t dev,
590                               struct buffer_head **new_fe_bh,
591                               struct buffer_head *parent_fe_bh,
592                               handle_t *handle,
593                               struct ocfs2_alloc_context *inode_ac)
594 {
595         int status = 0;
596         u64 suballoc_loc, fe_blkno = 0;
597         u16 suballoc_bit;
598
599         *new_fe_bh = NULL;
600
601         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
602                                        inode_ac, &suballoc_loc,
603                                        &suballoc_bit, &fe_blkno);
604         if (status < 0) {
605                 mlog_errno(status);
606                 return status;
607         }
608
609         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
610                                     parent_fe_bh, handle, inode_ac,
611                                     fe_blkno, suballoc_loc, suballoc_bit);
612 }
613
614 static int ocfs2_mkdir(struct inode *dir,
615                        struct dentry *dentry,
616                        int mode)
617 {
618         int ret;
619
620         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
621                    dentry->d_name.len, dentry->d_name.name);
622         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
623         mlog_exit(ret);
624
625         return ret;
626 }
627
628 static int ocfs2_create(struct inode *dir,
629                         struct dentry *dentry,
630                         int mode,
631                         struct nameidata *nd)
632 {
633         int ret;
634
635         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
636                    dentry->d_name.len, dentry->d_name.name);
637         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
638         mlog_exit(ret);
639
640         return ret;
641 }
642
643 static int ocfs2_link(struct dentry *old_dentry,
644                       struct inode *dir,
645                       struct dentry *dentry)
646 {
647         handle_t *handle;
648         struct inode *inode = old_dentry->d_inode;
649         int err;
650         struct buffer_head *fe_bh = NULL;
651         struct buffer_head *parent_fe_bh = NULL;
652         struct ocfs2_dinode *fe = NULL;
653         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
654         struct ocfs2_dir_lookup_result lookup = { NULL, };
655         sigset_t oldset;
656
657         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
658                    old_dentry->d_name.len, old_dentry->d_name.name,
659                    dentry->d_name.len, dentry->d_name.name);
660
661         if (S_ISDIR(inode->i_mode))
662                 return -EPERM;
663
664         dquot_initialize(dir);
665
666         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
667         if (err < 0) {
668                 if (err != -ENOENT)
669                         mlog_errno(err);
670                 return err;
671         }
672
673         if (!dir->i_nlink) {
674                 err = -ENOENT;
675                 goto out;
676         }
677
678         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
679                                         dentry->d_name.len);
680         if (err)
681                 goto out;
682
683         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
684                                            dentry->d_name.name,
685                                            dentry->d_name.len, &lookup);
686         if (err < 0) {
687                 mlog_errno(err);
688                 goto out;
689         }
690
691         err = ocfs2_inode_lock(inode, &fe_bh, 1);
692         if (err < 0) {
693                 if (err != -ENOENT)
694                         mlog_errno(err);
695                 goto out;
696         }
697
698         fe = (struct ocfs2_dinode *) fe_bh->b_data;
699         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
700                 err = -EMLINK;
701                 goto out_unlock_inode;
702         }
703
704         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
705         if (IS_ERR(handle)) {
706                 err = PTR_ERR(handle);
707                 handle = NULL;
708                 mlog_errno(err);
709                 goto out_unlock_inode;
710         }
711
712         /* Starting to change things, restart is no longer possible. */
713         ocfs2_block_signals(&oldset);
714
715         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
716                                       OCFS2_JOURNAL_ACCESS_WRITE);
717         if (err < 0) {
718                 mlog_errno(err);
719                 goto out_commit;
720         }
721
722         inc_nlink(inode);
723         inode->i_ctime = CURRENT_TIME;
724         ocfs2_set_links_count(fe, inode->i_nlink);
725         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
726         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
727         ocfs2_journal_dirty(handle, fe_bh);
728
729         err = ocfs2_add_entry(handle, dentry, inode,
730                               OCFS2_I(inode)->ip_blkno,
731                               parent_fe_bh, &lookup);
732         if (err) {
733                 ocfs2_add_links_count(fe, -1);
734                 drop_nlink(inode);
735                 mlog_errno(err);
736                 goto out_commit;
737         }
738
739         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
740         if (err) {
741                 mlog_errno(err);
742                 goto out_commit;
743         }
744
745         ihold(inode);
746         dentry->d_op = &ocfs2_dentry_ops;
747         d_instantiate(dentry, inode);
748
749 out_commit:
750         ocfs2_commit_trans(osb, handle);
751         ocfs2_unblock_signals(&oldset);
752 out_unlock_inode:
753         ocfs2_inode_unlock(inode, 1);
754
755 out:
756         ocfs2_inode_unlock(dir, 1);
757
758         brelse(fe_bh);
759         brelse(parent_fe_bh);
760
761         ocfs2_free_dir_lookup_result(&lookup);
762
763         mlog_exit(err);
764
765         return err;
766 }
767
768 /*
769  * Takes and drops an exclusive lock on the given dentry. This will
770  * force other nodes to drop it.
771  */
772 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
773 {
774         int ret;
775
776         ret = ocfs2_dentry_lock(dentry, 1);
777         if (ret)
778                 mlog_errno(ret);
779         else
780                 ocfs2_dentry_unlock(dentry, 1);
781
782         return ret;
783 }
784
785 static inline int inode_is_unlinkable(struct inode *inode)
786 {
787         if (S_ISDIR(inode->i_mode)) {
788                 if (inode->i_nlink == 2)
789                         return 1;
790                 return 0;
791         }
792
793         if (inode->i_nlink == 1)
794                 return 1;
795         return 0;
796 }
797
798 static int ocfs2_unlink(struct inode *dir,
799                         struct dentry *dentry)
800 {
801         int status;
802         int child_locked = 0;
803         struct inode *inode = dentry->d_inode;
804         struct inode *orphan_dir = NULL;
805         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
806         u64 blkno;
807         struct ocfs2_dinode *fe = NULL;
808         struct buffer_head *fe_bh = NULL;
809         struct buffer_head *parent_node_bh = NULL;
810         handle_t *handle = NULL;
811         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
812         struct ocfs2_dir_lookup_result lookup = { NULL, };
813         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
814
815         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
816                    dentry->d_name.len, dentry->d_name.name);
817
818         dquot_initialize(dir);
819
820         BUG_ON(dentry->d_parent->d_inode != dir);
821
822         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
823
824         if (inode == osb->root_inode) {
825                 mlog(0, "Cannot delete the root directory\n");
826                 return -EPERM;
827         }
828
829         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
830                                          OI_LS_PARENT);
831         if (status < 0) {
832                 if (status != -ENOENT)
833                         mlog_errno(status);
834                 return status;
835         }
836
837         status = ocfs2_find_files_on_disk(dentry->d_name.name,
838                                           dentry->d_name.len, &blkno, dir,
839                                           &lookup);
840         if (status < 0) {
841                 if (status != -ENOENT)
842                         mlog_errno(status);
843                 goto leave;
844         }
845
846         if (OCFS2_I(inode)->ip_blkno != blkno) {
847                 status = -ENOENT;
848
849                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
850                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
851                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
852                 goto leave;
853         }
854
855         status = ocfs2_inode_lock(inode, &fe_bh, 1);
856         if (status < 0) {
857                 if (status != -ENOENT)
858                         mlog_errno(status);
859                 goto leave;
860         }
861         child_locked = 1;
862
863         if (S_ISDIR(inode->i_mode)) {
864                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
865                         status = -ENOTEMPTY;
866                         goto leave;
867                 }
868         }
869
870         status = ocfs2_remote_dentry_delete(dentry);
871         if (status < 0) {
872                 /* This remote delete should succeed under all normal
873                  * circumstances. */
874                 mlog_errno(status);
875                 goto leave;
876         }
877
878         if (inode_is_unlinkable(inode)) {
879                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
880                                                   OCFS2_I(inode)->ip_blkno,
881                                                   orphan_name, &orphan_insert);
882                 if (status < 0) {
883                         mlog_errno(status);
884                         goto leave;
885                 }
886         }
887
888         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
889         if (IS_ERR(handle)) {
890                 status = PTR_ERR(handle);
891                 handle = NULL;
892                 mlog_errno(status);
893                 goto leave;
894         }
895
896         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
897                                          OCFS2_JOURNAL_ACCESS_WRITE);
898         if (status < 0) {
899                 mlog_errno(status);
900                 goto leave;
901         }
902
903         fe = (struct ocfs2_dinode *) fe_bh->b_data;
904
905         if (inode_is_unlinkable(inode)) {
906                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
907                                           &orphan_insert, orphan_dir);
908                 if (status < 0) {
909                         mlog_errno(status);
910                         goto leave;
911                 }
912         }
913
914         /* delete the name from the parent dir */
915         status = ocfs2_delete_entry(handle, dir, &lookup);
916         if (status < 0) {
917                 mlog_errno(status);
918                 goto leave;
919         }
920
921         if (S_ISDIR(inode->i_mode))
922                 drop_nlink(inode);
923         drop_nlink(inode);
924         ocfs2_set_links_count(fe, inode->i_nlink);
925         ocfs2_journal_dirty(handle, fe_bh);
926
927         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
928         if (S_ISDIR(inode->i_mode))
929                 drop_nlink(dir);
930
931         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
932         if (status < 0) {
933                 mlog_errno(status);
934                 if (S_ISDIR(inode->i_mode))
935                         inc_nlink(dir);
936         }
937
938 leave:
939         if (handle)
940                 ocfs2_commit_trans(osb, handle);
941
942         if (child_locked)
943                 ocfs2_inode_unlock(inode, 1);
944
945         ocfs2_inode_unlock(dir, 1);
946
947         if (orphan_dir) {
948                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
949                 ocfs2_inode_unlock(orphan_dir, 1);
950                 mutex_unlock(&orphan_dir->i_mutex);
951                 iput(orphan_dir);
952         }
953
954         brelse(fe_bh);
955         brelse(parent_node_bh);
956
957         ocfs2_free_dir_lookup_result(&orphan_insert);
958         ocfs2_free_dir_lookup_result(&lookup);
959
960         mlog_exit(status);
961
962         return status;
963 }
964
965 /*
966  * The only place this should be used is rename!
967  * if they have the same id, then the 1st one is the only one locked.
968  */
969 static int ocfs2_double_lock(struct ocfs2_super *osb,
970                              struct buffer_head **bh1,
971                              struct inode *inode1,
972                              struct buffer_head **bh2,
973                              struct inode *inode2)
974 {
975         int status;
976         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
977         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
978         struct buffer_head **tmpbh;
979         struct inode *tmpinode;
980
981         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
982                    (unsigned long long)oi1->ip_blkno,
983                    (unsigned long long)oi2->ip_blkno);
984
985         if (*bh1)
986                 *bh1 = NULL;
987         if (*bh2)
988                 *bh2 = NULL;
989
990         /* we always want to lock the one with the lower lockid first. */
991         if (oi1->ip_blkno != oi2->ip_blkno) {
992                 if (oi1->ip_blkno < oi2->ip_blkno) {
993                         /* switch id1 and id2 around */
994                         mlog(0, "switching them around...\n");
995                         tmpbh = bh2;
996                         bh2 = bh1;
997                         bh1 = tmpbh;
998
999                         tmpinode = inode2;
1000                         inode2 = inode1;
1001                         inode1 = tmpinode;
1002                 }
1003                 /* lock id2 */
1004                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1005                                                  OI_LS_RENAME1);
1006                 if (status < 0) {
1007                         if (status != -ENOENT)
1008                                 mlog_errno(status);
1009                         goto bail;
1010                 }
1011         }
1012
1013         /* lock id1 */
1014         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1015         if (status < 0) {
1016                 /*
1017                  * An error return must mean that no cluster locks
1018                  * were held on function exit.
1019                  */
1020                 if (oi1->ip_blkno != oi2->ip_blkno)
1021                         ocfs2_inode_unlock(inode2, 1);
1022
1023                 if (status != -ENOENT)
1024                         mlog_errno(status);
1025         }
1026
1027 bail:
1028         mlog_exit(status);
1029         return status;
1030 }
1031
1032 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1033 {
1034         ocfs2_inode_unlock(inode1, 1);
1035
1036         if (inode1 != inode2)
1037                 ocfs2_inode_unlock(inode2, 1);
1038 }
1039
1040 static int ocfs2_rename(struct inode *old_dir,
1041                         struct dentry *old_dentry,
1042                         struct inode *new_dir,
1043                         struct dentry *new_dentry)
1044 {
1045         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1046         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1047         struct inode *old_inode = old_dentry->d_inode;
1048         struct inode *new_inode = new_dentry->d_inode;
1049         struct inode *orphan_dir = NULL;
1050         struct ocfs2_dinode *newfe = NULL;
1051         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1052         struct buffer_head *newfe_bh = NULL;
1053         struct buffer_head *old_inode_bh = NULL;
1054         struct ocfs2_super *osb = NULL;
1055         u64 newfe_blkno, old_de_ino;
1056         handle_t *handle = NULL;
1057         struct buffer_head *old_dir_bh = NULL;
1058         struct buffer_head *new_dir_bh = NULL;
1059         nlink_t old_dir_nlink = old_dir->i_nlink;
1060         struct ocfs2_dinode *old_di;
1061         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1062         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1063         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1064         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1065         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1066
1067         /* At some point it might be nice to break this function up a
1068          * bit. */
1069
1070         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1071                    old_dir, old_dentry, new_dir, new_dentry,
1072                    old_dentry->d_name.len, old_dentry->d_name.name,
1073                    new_dentry->d_name.len, new_dentry->d_name.name);
1074
1075         dquot_initialize(old_dir);
1076         dquot_initialize(new_dir);
1077
1078         osb = OCFS2_SB(old_dir->i_sb);
1079
1080         if (new_inode) {
1081                 if (!igrab(new_inode))
1082                         BUG();
1083         }
1084
1085         /* Assume a directory hierarchy thusly:
1086          * a/b/c
1087          * a/d
1088          * a,b,c, and d are all directories.
1089          *
1090          * from cwd of 'a' on both nodes:
1091          * node1: mv b/c d
1092          * node2: mv d   b/c
1093          *
1094          * And that's why, just like the VFS, we need a file system
1095          * rename lock. */
1096         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1097                 status = ocfs2_rename_lock(osb);
1098                 if (status < 0) {
1099                         mlog_errno(status);
1100                         goto bail;
1101                 }
1102                 rename_lock = 1;
1103         }
1104
1105         /* if old and new are the same, this'll just do one lock. */
1106         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1107                                    &new_dir_bh, new_dir);
1108         if (status < 0) {
1109                 mlog_errno(status);
1110                 goto bail;
1111         }
1112         parents_locked = 1;
1113
1114         /* make sure both dirs have bhs
1115          * get an extra ref on old_dir_bh if old==new */
1116         if (!new_dir_bh) {
1117                 if (old_dir_bh) {
1118                         new_dir_bh = old_dir_bh;
1119                         get_bh(new_dir_bh);
1120                 } else {
1121                         mlog(ML_ERROR, "no old_dir_bh!\n");
1122                         status = -EIO;
1123                         goto bail;
1124                 }
1125         }
1126
1127         /*
1128          * Aside from allowing a meta data update, the locking here
1129          * also ensures that the downconvert thread on other nodes
1130          * won't have to concurrently downconvert the inode and the
1131          * dentry locks.
1132          */
1133         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1134                                          OI_LS_PARENT);
1135         if (status < 0) {
1136                 if (status != -ENOENT)
1137                         mlog_errno(status);
1138                 goto bail;
1139         }
1140         old_child_locked = 1;
1141
1142         status = ocfs2_remote_dentry_delete(old_dentry);
1143         if (status < 0) {
1144                 mlog_errno(status);
1145                 goto bail;
1146         }
1147
1148         if (S_ISDIR(old_inode->i_mode)) {
1149                 u64 old_inode_parent;
1150
1151                 update_dot_dot = 1;
1152                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1153                                                   old_inode,
1154                                                   &old_inode_dot_dot_res);
1155                 if (status) {
1156                         status = -EIO;
1157                         goto bail;
1158                 }
1159
1160                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1161                         status = -EIO;
1162                         goto bail;
1163                 }
1164
1165                 if (!new_inode && new_dir != old_dir &&
1166                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1167                         status = -EMLINK;
1168                         goto bail;
1169                 }
1170         }
1171
1172         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1173                                             old_dentry->d_name.len,
1174                                             &old_de_ino);
1175         if (status) {
1176                 status = -ENOENT;
1177                 goto bail;
1178         }
1179
1180         /*
1181          *  Check for inode number is _not_ due to possible IO errors.
1182          *  We might rmdir the source, keep it as pwd of some process
1183          *  and merrily kill the link to whatever was created under the
1184          *  same name. Goodbye sticky bit ;-<
1185          */
1186         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1187                 status = -ENOENT;
1188                 goto bail;
1189         }
1190
1191         /* check if the target already exists (in which case we need
1192          * to delete it */
1193         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1194                                           new_dentry->d_name.len,
1195                                           &newfe_blkno, new_dir,
1196                                           &target_lookup_res);
1197         /* The only error we allow here is -ENOENT because the new
1198          * file not existing is perfectly valid. */
1199         if ((status < 0) && (status != -ENOENT)) {
1200                 /* If we cannot find the file specified we should just */
1201                 /* return the error... */
1202                 mlog_errno(status);
1203                 goto bail;
1204         }
1205         if (status == 0)
1206                 target_exists = 1;
1207
1208         if (!target_exists && new_inode) {
1209                 /*
1210                  * Target was unlinked by another node while we were
1211                  * waiting to get to ocfs2_rename(). There isn't
1212                  * anything we can do here to help the situation, so
1213                  * bubble up the appropriate error.
1214                  */
1215                 status = -ENOENT;
1216                 goto bail;
1217         }
1218
1219         /* In case we need to overwrite an existing file, we blow it
1220          * away first */
1221         if (target_exists) {
1222                 /* VFS didn't think there existed an inode here, but
1223                  * someone else in the cluster must have raced our
1224                  * rename to create one. Today we error cleanly, in
1225                  * the future we should consider calling iget to build
1226                  * a new struct inode for this entry. */
1227                 if (!new_inode) {
1228                         status = -EACCES;
1229
1230                         mlog(0, "We found an inode for name %.*s but VFS "
1231                              "didn't give us one.\n", new_dentry->d_name.len,
1232                              new_dentry->d_name.name);
1233                         goto bail;
1234                 }
1235
1236                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1237                         status = -EACCES;
1238
1239                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1240                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1241                              (unsigned long long)newfe_blkno,
1242                              OCFS2_I(new_inode)->ip_flags);
1243                         goto bail;
1244                 }
1245
1246                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1247                 if (status < 0) {
1248                         if (status != -ENOENT)
1249                                 mlog_errno(status);
1250                         goto bail;
1251                 }
1252                 new_child_locked = 1;
1253
1254                 status = ocfs2_remote_dentry_delete(new_dentry);
1255                 if (status < 0) {
1256                         mlog_errno(status);
1257                         goto bail;
1258                 }
1259
1260                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1261
1262                 mlog(0, "aha rename over existing... new_blkno=%llu "
1263                      "newfebh=%p bhblocknr=%llu\n",
1264                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1265                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1266
1267                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1268                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1269                                                 OCFS2_I(new_inode)->ip_blkno,
1270                                                 orphan_name, &orphan_insert);
1271                         if (status < 0) {
1272                                 mlog_errno(status);
1273                                 goto bail;
1274                         }
1275                 }
1276         } else {
1277                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1278
1279                 status = ocfs2_check_dir_for_entry(new_dir,
1280                                                    new_dentry->d_name.name,
1281                                                    new_dentry->d_name.len);
1282                 if (status)
1283                         goto bail;
1284
1285                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1286                                                       new_dentry->d_name.name,
1287                                                       new_dentry->d_name.len,
1288                                                       &target_insert);
1289                 if (status < 0) {
1290                         mlog_errno(status);
1291                         goto bail;
1292                 }
1293         }
1294
1295         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1296         if (IS_ERR(handle)) {
1297                 status = PTR_ERR(handle);
1298                 handle = NULL;
1299                 mlog_errno(status);
1300                 goto bail;
1301         }
1302
1303         if (target_exists) {
1304                 if (S_ISDIR(new_inode->i_mode)) {
1305                         if (new_inode->i_nlink != 2 ||
1306                             !ocfs2_empty_dir(new_inode)) {
1307                                 status = -ENOTEMPTY;
1308                                 goto bail;
1309                         }
1310                 }
1311                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1312                                                  newfe_bh,
1313                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1314                 if (status < 0) {
1315                         mlog_errno(status);
1316                         goto bail;
1317                 }
1318
1319                 if (S_ISDIR(new_inode->i_mode) ||
1320                     (ocfs2_read_links_count(newfe) == 1)) {
1321                         status = ocfs2_orphan_add(osb, handle, new_inode,
1322                                                   newfe_bh, orphan_name,
1323                                                   &orphan_insert, orphan_dir);
1324                         if (status < 0) {
1325                                 mlog_errno(status);
1326                                 goto bail;
1327                         }
1328                 }
1329
1330                 /* change the dirent to point to the correct inode */
1331                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1332                                             old_inode);
1333                 if (status < 0) {
1334                         mlog_errno(status);
1335                         goto bail;
1336                 }
1337                 new_dir->i_version++;
1338
1339                 if (S_ISDIR(new_inode->i_mode))
1340                         ocfs2_set_links_count(newfe, 0);
1341                 else
1342                         ocfs2_add_links_count(newfe, -1);
1343                 ocfs2_journal_dirty(handle, newfe_bh);
1344         } else {
1345                 /* if the name was not found in new_dir, add it now */
1346                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1347                                          OCFS2_I(old_inode)->ip_blkno,
1348                                          new_dir_bh, &target_insert);
1349         }
1350
1351         old_inode->i_ctime = CURRENT_TIME;
1352         mark_inode_dirty(old_inode);
1353
1354         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1355                                          old_inode_bh,
1356                                          OCFS2_JOURNAL_ACCESS_WRITE);
1357         if (status >= 0) {
1358                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1359
1360                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1361                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1362                 ocfs2_journal_dirty(handle, old_inode_bh);
1363         } else
1364                 mlog_errno(status);
1365
1366         /*
1367          * Now that the name has been added to new_dir, remove the old name.
1368          *
1369          * We don't keep any directory entry context around until now
1370          * because the insert might have changed the type of directory
1371          * we're dealing with.
1372          */
1373         status = ocfs2_find_entry(old_dentry->d_name.name,
1374                                   old_dentry->d_name.len, old_dir,
1375                                   &old_entry_lookup);
1376         if (status)
1377                 goto bail;
1378
1379         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1380         if (status < 0) {
1381                 mlog_errno(status);
1382                 goto bail;
1383         }
1384
1385         if (new_inode) {
1386                 new_inode->i_nlink--;
1387                 new_inode->i_ctime = CURRENT_TIME;
1388         }
1389         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1390
1391         if (update_dot_dot) {
1392                 status = ocfs2_update_entry(old_inode, handle,
1393                                             &old_inode_dot_dot_res, new_dir);
1394                 old_dir->i_nlink--;
1395                 if (new_inode) {
1396                         new_inode->i_nlink--;
1397                 } else {
1398                         inc_nlink(new_dir);
1399                         mark_inode_dirty(new_dir);
1400                 }
1401         }
1402         mark_inode_dirty(old_dir);
1403         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1404         if (new_inode) {
1405                 mark_inode_dirty(new_inode);
1406                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1407         }
1408
1409         if (old_dir != new_dir) {
1410                 /* Keep the same times on both directories.*/
1411                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1412
1413                 /*
1414                  * This will also pick up the i_nlink change from the
1415                  * block above.
1416                  */
1417                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1418         }
1419
1420         if (old_dir_nlink != old_dir->i_nlink) {
1421                 if (!old_dir_bh) {
1422                         mlog(ML_ERROR, "need to change nlink for old dir "
1423                              "%llu from %d to %d but bh is NULL!\n",
1424                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1425                              (int)old_dir_nlink, old_dir->i_nlink);
1426                 } else {
1427                         struct ocfs2_dinode *fe;
1428                         status = ocfs2_journal_access_di(handle,
1429                                                          INODE_CACHE(old_dir),
1430                                                          old_dir_bh,
1431                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1432                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1433                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1434                         ocfs2_journal_dirty(handle, old_dir_bh);
1435                 }
1436         }
1437         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1438         status = 0;
1439 bail:
1440         if (rename_lock)
1441                 ocfs2_rename_unlock(osb);
1442
1443         if (handle)
1444                 ocfs2_commit_trans(osb, handle);
1445
1446         if (parents_locked)
1447                 ocfs2_double_unlock(old_dir, new_dir);
1448
1449         if (old_child_locked)
1450                 ocfs2_inode_unlock(old_inode, 1);
1451
1452         if (new_child_locked)
1453                 ocfs2_inode_unlock(new_inode, 1);
1454
1455         if (orphan_dir) {
1456                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1457                 ocfs2_inode_unlock(orphan_dir, 1);
1458                 mutex_unlock(&orphan_dir->i_mutex);
1459                 iput(orphan_dir);
1460         }
1461
1462         if (new_inode)
1463                 sync_mapping_buffers(old_inode->i_mapping);
1464
1465         if (new_inode)
1466                 iput(new_inode);
1467
1468         ocfs2_free_dir_lookup_result(&target_lookup_res);
1469         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1470         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1471         ocfs2_free_dir_lookup_result(&orphan_insert);
1472         ocfs2_free_dir_lookup_result(&target_insert);
1473
1474         brelse(newfe_bh);
1475         brelse(old_inode_bh);
1476         brelse(old_dir_bh);
1477         brelse(new_dir_bh);
1478
1479         mlog_exit(status);
1480
1481         return status;
1482 }
1483
1484 /*
1485  * we expect i_size = strlen(symname). Copy symname into the file
1486  * data, including the null terminator.
1487  */
1488 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1489                                      handle_t *handle,
1490                                      struct inode *inode,
1491                                      const char *symname)
1492 {
1493         struct buffer_head **bhs = NULL;
1494         const char *c;
1495         struct super_block *sb = osb->sb;
1496         u64 p_blkno, p_blocks;
1497         int virtual, blocks, status, i, bytes_left;
1498
1499         bytes_left = i_size_read(inode) + 1;
1500         /* we can't trust i_blocks because we're actually going to
1501          * write i_size + 1 bytes. */
1502         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1503
1504         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1505                         (unsigned long long)inode->i_blocks,
1506                         i_size_read(inode), blocks);
1507
1508         /* Sanity check -- make sure we're going to fit. */
1509         if (bytes_left >
1510             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1511                 status = -EIO;
1512                 mlog_errno(status);
1513                 goto bail;
1514         }
1515
1516         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1517         if (!bhs) {
1518                 status = -ENOMEM;
1519                 mlog_errno(status);
1520                 goto bail;
1521         }
1522
1523         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1524                                              NULL);
1525         if (status < 0) {
1526                 mlog_errno(status);
1527                 goto bail;
1528         }
1529
1530         /* links can never be larger than one cluster so we know this
1531          * is all going to be contiguous, but do a sanity check
1532          * anyway. */
1533         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1534                 status = -EIO;
1535                 mlog_errno(status);
1536                 goto bail;
1537         }
1538
1539         virtual = 0;
1540         while(bytes_left > 0) {
1541                 c = &symname[virtual * sb->s_blocksize];
1542
1543                 bhs[virtual] = sb_getblk(sb, p_blkno);
1544                 if (!bhs[virtual]) {
1545                         status = -ENOMEM;
1546                         mlog_errno(status);
1547                         goto bail;
1548                 }
1549                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1550                                               bhs[virtual]);
1551
1552                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1553                                               bhs[virtual],
1554                                               OCFS2_JOURNAL_ACCESS_CREATE);
1555                 if (status < 0) {
1556                         mlog_errno(status);
1557                         goto bail;
1558                 }
1559
1560                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1561
1562                 memcpy(bhs[virtual]->b_data, c,
1563                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1564                        bytes_left);
1565
1566                 ocfs2_journal_dirty(handle, bhs[virtual]);
1567
1568                 virtual++;
1569                 p_blkno++;
1570                 bytes_left -= sb->s_blocksize;
1571         }
1572
1573         status = 0;
1574 bail:
1575
1576         if (bhs) {
1577                 for(i = 0; i < blocks; i++)
1578                         brelse(bhs[i]);
1579                 kfree(bhs);
1580         }
1581
1582         mlog_exit(status);
1583         return status;
1584 }
1585
1586 static int ocfs2_symlink(struct inode *dir,
1587                          struct dentry *dentry,
1588                          const char *symname)
1589 {
1590         int status, l, credits;
1591         u64 newsize;
1592         struct ocfs2_super *osb = NULL;
1593         struct inode *inode = NULL;
1594         struct super_block *sb;
1595         struct buffer_head *new_fe_bh = NULL;
1596         struct buffer_head *parent_fe_bh = NULL;
1597         struct ocfs2_dinode *fe = NULL;
1598         struct ocfs2_dinode *dirfe;
1599         handle_t *handle = NULL;
1600         struct ocfs2_alloc_context *inode_ac = NULL;
1601         struct ocfs2_alloc_context *data_ac = NULL;
1602         struct ocfs2_alloc_context *xattr_ac = NULL;
1603         int want_clusters = 0;
1604         int xattr_credits = 0;
1605         struct ocfs2_security_xattr_info si = {
1606                 .enable = 1,
1607         };
1608         int did_quota = 0, did_quota_inode = 0;
1609         struct ocfs2_dir_lookup_result lookup = { NULL, };
1610         sigset_t oldset;
1611         int did_block_signals = 0;
1612
1613         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1614                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1615
1616         dquot_initialize(dir);
1617
1618         sb = dir->i_sb;
1619         osb = OCFS2_SB(sb);
1620
1621         l = strlen(symname) + 1;
1622
1623         credits = ocfs2_calc_symlink_credits(sb);
1624
1625         /* lock the parent directory */
1626         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1627         if (status < 0) {
1628                 if (status != -ENOENT)
1629                         mlog_errno(status);
1630                 return status;
1631         }
1632
1633         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1634         if (!ocfs2_read_links_count(dirfe)) {
1635                 /* can't make a file in a deleted directory. */
1636                 status = -ENOENT;
1637                 goto bail;
1638         }
1639
1640         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1641                                            dentry->d_name.len);
1642         if (status)
1643                 goto bail;
1644
1645         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1646                                               dentry->d_name.name,
1647                                               dentry->d_name.len, &lookup);
1648         if (status < 0) {
1649                 mlog_errno(status);
1650                 goto bail;
1651         }
1652
1653         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1654         if (status < 0) {
1655                 if (status != -ENOSPC)
1656                         mlog_errno(status);
1657                 goto bail;
1658         }
1659
1660         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1661         if (!inode) {
1662                 status = -ENOMEM;
1663                 mlog_errno(status);
1664                 goto bail;
1665         }
1666
1667         /* get security xattr */
1668         status = ocfs2_init_security_get(inode, dir, &si);
1669         if (status) {
1670                 if (status == -EOPNOTSUPP)
1671                         si.enable = 0;
1672                 else {
1673                         mlog_errno(status);
1674                         goto bail;
1675                 }
1676         }
1677
1678         /* calculate meta data/clusters for setting security xattr */
1679         if (si.enable) {
1680                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1681                                                   &xattr_credits, &xattr_ac);
1682                 if (status < 0) {
1683                         mlog_errno(status);
1684                         goto bail;
1685                 }
1686         }
1687
1688         /* don't reserve bitmap space for fast symlinks. */
1689         if (l > ocfs2_fast_symlink_chars(sb))
1690                 want_clusters += 1;
1691
1692         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1693         if (status < 0) {
1694                 if (status != -ENOSPC)
1695                         mlog_errno(status);
1696                 goto bail;
1697         }
1698
1699         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1700         if (IS_ERR(handle)) {
1701                 status = PTR_ERR(handle);
1702                 handle = NULL;
1703                 mlog_errno(status);
1704                 goto bail;
1705         }
1706
1707         /* Starting to change things, restart is no longer possible. */
1708         ocfs2_block_signals(&oldset);
1709         did_block_signals = 1;
1710
1711         status = dquot_alloc_inode(inode);
1712         if (status)
1713                 goto bail;
1714         did_quota_inode = 1;
1715
1716         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1717                    inode->i_mode, dentry->d_name.len,
1718                    dentry->d_name.name);
1719
1720         status = ocfs2_mknod_locked(osb, dir, inode,
1721                                     0, &new_fe_bh, parent_fe_bh, handle,
1722                                     inode_ac);
1723         if (status < 0) {
1724                 mlog_errno(status);
1725                 goto bail;
1726         }
1727
1728         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1729         inode->i_rdev = 0;
1730         newsize = l - 1;
1731         if (l > ocfs2_fast_symlink_chars(sb)) {
1732                 u32 offset = 0;
1733
1734                 inode->i_op = &ocfs2_symlink_inode_operations;
1735                 status = dquot_alloc_space_nodirty(inode,
1736                     ocfs2_clusters_to_bytes(osb->sb, 1));
1737                 if (status)
1738                         goto bail;
1739                 did_quota = 1;
1740                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1741                                               new_fe_bh,
1742                                               handle, data_ac, NULL,
1743                                               NULL);
1744                 if (status < 0) {
1745                         if (status != -ENOSPC && status != -EINTR) {
1746                                 mlog(ML_ERROR,
1747                                      "Failed to extend file to %llu\n",
1748                                      (unsigned long long)newsize);
1749                                 mlog_errno(status);
1750                                 status = -ENOSPC;
1751                         }
1752                         goto bail;
1753                 }
1754                 i_size_write(inode, newsize);
1755                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1756         } else {
1757                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1758                 memcpy((char *) fe->id2.i_symlink, symname, l);
1759                 i_size_write(inode, newsize);
1760                 inode->i_blocks = 0;
1761         }
1762
1763         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1764         if (status < 0) {
1765                 mlog_errno(status);
1766                 goto bail;
1767         }
1768
1769         if (!ocfs2_inode_is_fast_symlink(inode)) {
1770                 status = ocfs2_create_symlink_data(osb, handle, inode,
1771                                                    symname);
1772                 if (status < 0) {
1773                         mlog_errno(status);
1774                         goto bail;
1775                 }
1776         }
1777
1778         if (si.enable) {
1779                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1780                                                  xattr_ac, data_ac);
1781                 if (status < 0) {
1782                         mlog_errno(status);
1783                         goto bail;
1784                 }
1785         }
1786
1787         /*
1788          * Do this before adding the entry to the directory. We add
1789          * also set d_op after success so that ->d_iput() will cleanup
1790          * the dentry lock even if ocfs2_add_entry() fails below.
1791          */
1792         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1793         if (status) {
1794                 mlog_errno(status);
1795                 goto bail;
1796         }
1797         dentry->d_op = &ocfs2_dentry_ops;
1798
1799         status = ocfs2_add_entry(handle, dentry, inode,
1800                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1801                                  &lookup);
1802         if (status < 0) {
1803                 mlog_errno(status);
1804                 goto bail;
1805         }
1806
1807         insert_inode_hash(inode);
1808         d_instantiate(dentry, inode);
1809 bail:
1810         if (status < 0 && did_quota)
1811                 dquot_free_space_nodirty(inode,
1812                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1813         if (status < 0 && did_quota_inode)
1814                 dquot_free_inode(inode);
1815         if (handle)
1816                 ocfs2_commit_trans(osb, handle);
1817
1818         ocfs2_inode_unlock(dir, 1);
1819         if (did_block_signals)
1820                 ocfs2_unblock_signals(&oldset);
1821
1822         brelse(new_fe_bh);
1823         brelse(parent_fe_bh);
1824         kfree(si.name);
1825         kfree(si.value);
1826         ocfs2_free_dir_lookup_result(&lookup);
1827         if (inode_ac)
1828                 ocfs2_free_alloc_context(inode_ac);
1829         if (data_ac)
1830                 ocfs2_free_alloc_context(data_ac);
1831         if (xattr_ac)
1832                 ocfs2_free_alloc_context(xattr_ac);
1833         if ((status < 0) && inode) {
1834                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1835                 clear_nlink(inode);
1836                 iput(inode);
1837         }
1838
1839         mlog_exit(status);
1840
1841         return status;
1842 }
1843
1844 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1845 {
1846         int status, namelen;
1847
1848         mlog_entry_void();
1849
1850         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1851                            (long long)blkno);
1852         if (namelen <= 0) {
1853                 if (namelen)
1854                         status = namelen;
1855                 else
1856                         status = -EINVAL;
1857                 mlog_errno(status);
1858                 goto bail;
1859         }
1860         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1861                 status = -EINVAL;
1862                 mlog_errno(status);
1863                 goto bail;
1864         }
1865
1866         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1867              namelen);
1868
1869         status = 0;
1870 bail:
1871         mlog_exit(status);
1872         return status;
1873 }
1874
1875 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1876                                         struct inode **ret_orphan_dir,
1877                                         struct buffer_head **ret_orphan_dir_bh)
1878 {
1879         struct inode *orphan_dir_inode;
1880         struct buffer_head *orphan_dir_bh = NULL;
1881         int ret = 0;
1882
1883         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1884                                                        ORPHAN_DIR_SYSTEM_INODE,
1885                                                        osb->slot_num);
1886         if (!orphan_dir_inode) {
1887                 ret = -ENOENT;
1888                 mlog_errno(ret);
1889                 return ret;
1890         }
1891
1892         mutex_lock(&orphan_dir_inode->i_mutex);
1893
1894         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1895         if (ret < 0) {
1896                 mutex_unlock(&orphan_dir_inode->i_mutex);
1897                 iput(orphan_dir_inode);
1898
1899                 mlog_errno(ret);
1900                 return ret;
1901         }
1902
1903         *ret_orphan_dir = orphan_dir_inode;
1904         *ret_orphan_dir_bh = orphan_dir_bh;
1905
1906         return 0;
1907 }
1908
1909 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1910                                       struct buffer_head *orphan_dir_bh,
1911                                       u64 blkno,
1912                                       char *name,
1913                                       struct ocfs2_dir_lookup_result *lookup)
1914 {
1915         int ret;
1916         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1917
1918         ret = ocfs2_blkno_stringify(blkno, name);
1919         if (ret < 0) {
1920                 mlog_errno(ret);
1921                 return ret;
1922         }
1923
1924         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1925                                            orphan_dir_bh, name,
1926                                            OCFS2_ORPHAN_NAMELEN, lookup);
1927         if (ret < 0) {
1928                 mlog_errno(ret);
1929                 return ret;
1930         }
1931
1932         return 0;
1933 }
1934
1935 /**
1936  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1937  * insertion of an orphan.
1938  * @osb: ocfs2 file system
1939  * @ret_orphan_dir: Orphan dir inode - returned locked!
1940  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1941  * @lookup: dir lookup result, to be passed back into functions like
1942  *          ocfs2_orphan_add
1943  *
1944  * Returns zero on success and the ret_orphan_dir, name and lookup
1945  * fields will be populated.
1946  *
1947  * Returns non-zero on failure. 
1948  */
1949 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1950                                     struct inode **ret_orphan_dir,
1951                                     u64 blkno,
1952                                     char *name,
1953                                     struct ocfs2_dir_lookup_result *lookup)
1954 {
1955         struct inode *orphan_dir_inode = NULL;
1956         struct buffer_head *orphan_dir_bh = NULL;
1957         int ret = 0;
1958
1959         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1960                                            &orphan_dir_bh);
1961         if (ret < 0) {
1962                 mlog_errno(ret);
1963                 return ret;
1964         }
1965
1966         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1967                                          blkno, name, lookup);
1968         if (ret < 0) {
1969                 mlog_errno(ret);
1970                 goto out;
1971         }
1972
1973         *ret_orphan_dir = orphan_dir_inode;
1974
1975 out:
1976         brelse(orphan_dir_bh);
1977
1978         if (ret) {
1979                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1980                 mutex_unlock(&orphan_dir_inode->i_mutex);
1981                 iput(orphan_dir_inode);
1982         }
1983
1984         mlog_exit(ret);
1985         return ret;
1986 }
1987
1988 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1989                             handle_t *handle,
1990                             struct inode *inode,
1991                             struct buffer_head *fe_bh,
1992                             char *name,
1993                             struct ocfs2_dir_lookup_result *lookup,
1994                             struct inode *orphan_dir_inode)
1995 {
1996         struct buffer_head *orphan_dir_bh = NULL;
1997         int status = 0;
1998         struct ocfs2_dinode *orphan_fe;
1999         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2000
2001         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2002
2003         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2004         if (status < 0) {
2005                 mlog_errno(status);
2006                 goto leave;
2007         }
2008
2009         status = ocfs2_journal_access_di(handle,
2010                                          INODE_CACHE(orphan_dir_inode),
2011                                          orphan_dir_bh,
2012                                          OCFS2_JOURNAL_ACCESS_WRITE);
2013         if (status < 0) {
2014                 mlog_errno(status);
2015                 goto leave;
2016         }
2017
2018         /* we're a cluster, and nlink can change on disk from
2019          * underneath us... */
2020         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2021         if (S_ISDIR(inode->i_mode))
2022                 ocfs2_add_links_count(orphan_fe, 1);
2023         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2024         ocfs2_journal_dirty(handle, orphan_dir_bh);
2025
2026         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2027                                    OCFS2_ORPHAN_NAMELEN, inode,
2028                                    OCFS2_I(inode)->ip_blkno,
2029                                    orphan_dir_bh, lookup);
2030         if (status < 0) {
2031                 mlog_errno(status);
2032                 goto leave;
2033         }
2034
2035         /*
2036          * We're going to journal the change of i_flags and i_orphaned_slot.
2037          * It's safe anyway, though some callers may duplicate the journaling.
2038          * Journaling within the func just make the logic look more
2039          * straightforward.
2040          */
2041         status = ocfs2_journal_access_di(handle,
2042                                          INODE_CACHE(inode),
2043                                          fe_bh,
2044                                          OCFS2_JOURNAL_ACCESS_WRITE);
2045         if (status < 0) {
2046                 mlog_errno(status);
2047                 goto leave;
2048         }
2049
2050         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2051         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2052
2053         /* Record which orphan dir our inode now resides
2054          * in. delete_inode will use this to determine which orphan
2055          * dir to lock. */
2056         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2057
2058         ocfs2_journal_dirty(handle, fe_bh);
2059
2060         mlog(0, "Inode %llu orphaned in slot %d\n",
2061              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2062
2063 leave:
2064         brelse(orphan_dir_bh);
2065
2066         mlog_exit(status);
2067         return status;
2068 }
2069
2070 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2071 int ocfs2_orphan_del(struct ocfs2_super *osb,
2072                      handle_t *handle,
2073                      struct inode *orphan_dir_inode,
2074                      struct inode *inode,
2075                      struct buffer_head *orphan_dir_bh)
2076 {
2077         char name[OCFS2_ORPHAN_NAMELEN + 1];
2078         struct ocfs2_dinode *orphan_fe;
2079         int status = 0;
2080         struct ocfs2_dir_lookup_result lookup = { NULL, };
2081
2082         mlog_entry_void();
2083
2084         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2085         if (status < 0) {
2086                 mlog_errno(status);
2087                 goto leave;
2088         }
2089
2090         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2091              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2092              OCFS2_ORPHAN_NAMELEN);
2093
2094         /* find it's spot in the orphan directory */
2095         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2096                                   &lookup);
2097         if (status) {
2098                 mlog_errno(status);
2099                 goto leave;
2100         }
2101
2102         /* remove it from the orphan directory */
2103         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2104         if (status < 0) {
2105                 mlog_errno(status);
2106                 goto leave;
2107         }
2108
2109         status = ocfs2_journal_access_di(handle,
2110                                          INODE_CACHE(orphan_dir_inode),
2111                                          orphan_dir_bh,
2112                                          OCFS2_JOURNAL_ACCESS_WRITE);
2113         if (status < 0) {
2114                 mlog_errno(status);
2115                 goto leave;
2116         }
2117
2118         /* do the i_nlink dance! :) */
2119         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2120         if (S_ISDIR(inode->i_mode))
2121                 ocfs2_add_links_count(orphan_fe, -1);
2122         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2123         ocfs2_journal_dirty(handle, orphan_dir_bh);
2124
2125 leave:
2126         ocfs2_free_dir_lookup_result(&lookup);
2127
2128         mlog_exit(status);
2129         return status;
2130 }
2131
2132 /**
2133  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to recieve a newly
2134  * allocated file. This is different from the typical 'add to orphan dir'
2135  * operation in that the inode does not yet exist. This is a problem because
2136  * the orphan dir stringifies the inode block number to come up with it's
2137  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2138  * problem. This function works around it by calling deeper into the orphan
2139  * and suballoc code than other callers. Use this only by necessity.
2140  * @dir: The directory which this inode will ultimately wind up under - not the
2141  * orphan dir!
2142  * @dir_bh: buffer_head the @dir inode block
2143  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2144  * with the string to be used for orphan dirent. Pass back to the orphan dir
2145  * code.
2146  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2147  * dir code.
2148  * @ret_di_blkno: block number where the new inode will be allocated.
2149  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2150  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2151  *
2152  * Returns zero on success and the ret_orphan_dir, name and lookup
2153  * fields will be populated.
2154  *
2155  * Returns non-zero on failure. 
2156  */
2157 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2158                                         struct buffer_head *dir_bh,
2159                                         char *orphan_name,
2160                                         struct inode **ret_orphan_dir,
2161                                         u64 *ret_di_blkno,
2162                                         struct ocfs2_dir_lookup_result *orphan_insert,
2163                                         struct ocfs2_alloc_context **ret_inode_ac)
2164 {
2165         int ret;
2166         u64 di_blkno;
2167         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2168         struct inode *orphan_dir = NULL;
2169         struct buffer_head *orphan_dir_bh = NULL;
2170         struct ocfs2_alloc_context *inode_ac = NULL;
2171
2172         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2173         if (ret < 0) {
2174                 mlog_errno(ret);
2175                 return ret;
2176         }
2177
2178         /* reserve an inode spot */
2179         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2180         if (ret < 0) {
2181                 if (ret != -ENOSPC)
2182                         mlog_errno(ret);
2183                 goto out;
2184         }
2185
2186         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2187                                        &di_blkno);
2188         if (ret) {
2189                 mlog_errno(ret);
2190                 goto out;
2191         }
2192
2193         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2194                                          di_blkno, orphan_name, orphan_insert);
2195         if (ret < 0) {
2196                 mlog_errno(ret);
2197                 goto out;
2198         }
2199
2200 out:
2201         if (ret == 0) {
2202                 *ret_orphan_dir = orphan_dir;
2203                 *ret_di_blkno = di_blkno;
2204                 *ret_inode_ac = inode_ac;
2205                 /*
2206                  * orphan_name and orphan_insert are already up to
2207                  * date via prepare_orphan_dir
2208                  */
2209         } else {
2210                 /* Unroll reserve_new_inode* */
2211                 if (inode_ac)
2212                         ocfs2_free_alloc_context(inode_ac);
2213
2214                 /* Unroll orphan dir locking */
2215                 mutex_unlock(&orphan_dir->i_mutex);
2216                 ocfs2_inode_unlock(orphan_dir, 1);
2217                 iput(orphan_dir);
2218         }
2219
2220         brelse(orphan_dir_bh);
2221
2222         return 0;
2223 }
2224
2225 int ocfs2_create_inode_in_orphan(struct inode *dir,
2226                                  int mode,
2227                                  struct inode **new_inode)
2228 {
2229         int status, did_quota_inode = 0;
2230         struct inode *inode = NULL;
2231         struct inode *orphan_dir = NULL;
2232         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2233         struct ocfs2_dinode *di = NULL;
2234         handle_t *handle = NULL;
2235         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2236         struct buffer_head *parent_di_bh = NULL;
2237         struct buffer_head *new_di_bh = NULL;
2238         struct ocfs2_alloc_context *inode_ac = NULL;
2239         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2240         u64 uninitialized_var(di_blkno), suballoc_loc;
2241         u16 suballoc_bit;
2242
2243         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2244         if (status < 0) {
2245                 if (status != -ENOENT)
2246                         mlog_errno(status);
2247                 return status;
2248         }
2249
2250         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2251                                               orphan_name, &orphan_dir,
2252                                               &di_blkno, &orphan_insert, &inode_ac);
2253         if (status < 0) {
2254                 if (status != -ENOSPC)
2255                         mlog_errno(status);
2256                 goto leave;
2257         }
2258
2259         inode = ocfs2_get_init_inode(dir, mode);
2260         if (!inode) {
2261                 status = -ENOMEM;
2262                 mlog_errno(status);
2263                 goto leave;
2264         }
2265
2266         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2267         if (IS_ERR(handle)) {
2268                 status = PTR_ERR(handle);
2269                 handle = NULL;
2270                 mlog_errno(status);
2271                 goto leave;
2272         }
2273
2274         status = dquot_alloc_inode(inode);
2275         if (status)
2276                 goto leave;
2277         did_quota_inode = 1;
2278
2279         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2280                                               &suballoc_loc,
2281                                               &suballoc_bit, di_blkno);
2282         if (status < 0) {
2283                 mlog_errno(status);
2284                 goto leave;
2285         }
2286
2287         inode->i_nlink = 0;
2288         /* do the real work now. */
2289         status = __ocfs2_mknod_locked(dir, inode,
2290                                       0, &new_di_bh, parent_di_bh, handle,
2291                                       inode_ac, di_blkno, suballoc_loc,
2292                                       suballoc_bit);
2293         if (status < 0) {
2294                 mlog_errno(status);
2295                 goto leave;
2296         }
2297
2298         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2299         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2300                                   &orphan_insert, orphan_dir);
2301         if (status < 0) {
2302                 mlog_errno(status);
2303                 goto leave;
2304         }
2305
2306         /* get open lock so that only nodes can't remove it from orphan dir. */
2307         status = ocfs2_open_lock(inode);
2308         if (status < 0)
2309                 mlog_errno(status);
2310
2311         insert_inode_hash(inode);
2312 leave:
2313         if (status < 0 && did_quota_inode)
2314                 dquot_free_inode(inode);
2315         if (handle)
2316                 ocfs2_commit_trans(osb, handle);
2317
2318         if (orphan_dir) {
2319                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2320                 ocfs2_inode_unlock(orphan_dir, 1);
2321                 mutex_unlock(&orphan_dir->i_mutex);
2322                 iput(orphan_dir);
2323         }
2324
2325         if (status == -ENOSPC)
2326                 mlog(0, "Disk is full\n");
2327
2328         if ((status < 0) && inode) {
2329                 clear_nlink(inode);
2330                 iput(inode);
2331         }
2332
2333         if (inode_ac)
2334                 ocfs2_free_alloc_context(inode_ac);
2335
2336         brelse(new_di_bh);
2337
2338         if (!status)
2339                 *new_inode = inode;
2340
2341         ocfs2_free_dir_lookup_result(&orphan_insert);
2342
2343         ocfs2_inode_unlock(dir, 1);
2344         brelse(parent_di_bh);
2345         return status;
2346 }
2347
2348 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2349                                    struct inode *inode,
2350                                    struct dentry *dentry)
2351 {
2352         int status = 0;
2353         struct buffer_head *parent_di_bh = NULL;
2354         handle_t *handle = NULL;
2355         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2356         struct ocfs2_dinode *dir_di, *di;
2357         struct inode *orphan_dir_inode = NULL;
2358         struct buffer_head *orphan_dir_bh = NULL;
2359         struct buffer_head *di_bh = NULL;
2360         struct ocfs2_dir_lookup_result lookup = { NULL, };
2361
2362         mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2363                    dentry->d_name.len, dentry->d_name.name);
2364
2365         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2366         if (status < 0) {
2367                 if (status != -ENOENT)
2368                         mlog_errno(status);
2369                 return status;
2370         }
2371
2372         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2373         if (!dir_di->i_links_count) {
2374                 /* can't make a file in a deleted directory. */
2375                 status = -ENOENT;
2376                 goto leave;
2377         }
2378
2379         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2380                                            dentry->d_name.len);
2381         if (status)
2382                 goto leave;
2383
2384         /* get a spot inside the dir. */
2385         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2386                                               dentry->d_name.name,
2387                                               dentry->d_name.len, &lookup);
2388         if (status < 0) {
2389                 mlog_errno(status);
2390                 goto leave;
2391         }
2392
2393         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2394                                                        ORPHAN_DIR_SYSTEM_INODE,
2395                                                        osb->slot_num);
2396         if (!orphan_dir_inode) {
2397                 status = -EEXIST;
2398                 mlog_errno(status);
2399                 goto leave;
2400         }
2401
2402         mutex_lock(&orphan_dir_inode->i_mutex);
2403
2404         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2405         if (status < 0) {
2406                 mlog_errno(status);
2407                 mutex_unlock(&orphan_dir_inode->i_mutex);
2408                 iput(orphan_dir_inode);
2409                 goto leave;
2410         }
2411
2412         status = ocfs2_read_inode_block(inode, &di_bh);
2413         if (status < 0) {
2414                 mlog_errno(status);
2415                 goto orphan_unlock;
2416         }
2417
2418         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2419         if (IS_ERR(handle)) {
2420                 status = PTR_ERR(handle);
2421                 handle = NULL;
2422                 mlog_errno(status);
2423                 goto orphan_unlock;
2424         }
2425
2426         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2427                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2428         if (status < 0) {
2429                 mlog_errno(status);
2430                 goto out_commit;
2431         }
2432
2433         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2434                                   orphan_dir_bh);
2435         if (status < 0) {
2436                 mlog_errno(status);
2437                 goto out_commit;
2438         }
2439
2440         di = (struct ocfs2_dinode *)di_bh->b_data;
2441         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2442         di->i_orphaned_slot = 0;
2443         inode->i_nlink = 1;
2444         ocfs2_set_links_count(di, inode->i_nlink);
2445         ocfs2_journal_dirty(handle, di_bh);
2446
2447         status = ocfs2_add_entry(handle, dentry, inode,
2448                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2449                                  &lookup);
2450         if (status < 0) {
2451                 mlog_errno(status);
2452                 goto out_commit;
2453         }
2454
2455         status = ocfs2_dentry_attach_lock(dentry, inode,
2456                                           OCFS2_I(dir)->ip_blkno);
2457         if (status) {
2458                 mlog_errno(status);
2459                 goto out_commit;
2460         }
2461
2462         dentry->d_op = &ocfs2_dentry_ops;
2463         d_instantiate(dentry, inode);
2464         status = 0;
2465 out_commit:
2466         ocfs2_commit_trans(osb, handle);
2467 orphan_unlock:
2468         ocfs2_inode_unlock(orphan_dir_inode, 1);
2469         mutex_unlock(&orphan_dir_inode->i_mutex);
2470         iput(orphan_dir_inode);
2471 leave:
2472
2473         ocfs2_inode_unlock(dir, 1);
2474
2475         brelse(di_bh);
2476         brelse(parent_di_bh);
2477         brelse(orphan_dir_bh);
2478
2479         ocfs2_free_dir_lookup_result(&lookup);
2480
2481         mlog_exit(status);
2482
2483         return status;
2484 }
2485
2486 const struct inode_operations ocfs2_dir_iops = {
2487         .create         = ocfs2_create,
2488         .lookup         = ocfs2_lookup,
2489         .link           = ocfs2_link,
2490         .unlink         = ocfs2_unlink,
2491         .rmdir          = ocfs2_unlink,
2492         .symlink        = ocfs2_symlink,
2493         .mkdir          = ocfs2_mkdir,
2494         .mknod          = ocfs2_mknod,
2495         .rename         = ocfs2_rename,
2496         .setattr        = ocfs2_setattr,
2497         .getattr        = ocfs2_getattr,
2498         .permission     = ocfs2_permission,
2499         .setxattr       = generic_setxattr,
2500         .getxattr       = generic_getxattr,
2501         .listxattr      = ocfs2_listxattr,
2502         .removexattr    = generic_removexattr,
2503         .fiemap         = ocfs2_fiemap,
2504 };