vfs: push dentry_unhash on rename_dir into file systems
[linux-3.10.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 #include <cluster/masklog.h>
46
47 #include "ocfs2.h"
48
49 #include "alloc.h"
50 #include "dcache.h"
51 #include "dir.h"
52 #include "dlmglue.h"
53 #include "extent_map.h"
54 #include "file.h"
55 #include "inode.h"
56 #include "journal.h"
57 #include "namei.h"
58 #include "suballoc.h"
59 #include "super.h"
60 #include "symlink.h"
61 #include "sysfile.h"
62 #include "uptodate.h"
63 #include "xattr.h"
64 #include "acl.h"
65 #include "ocfs2_trace.h"
66
67 #include "buffer_head_io.h"
68
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
70                               struct inode *dir,
71                               struct inode *inode,
72                               dev_t dev,
73                               struct buffer_head **new_fe_bh,
74                               struct buffer_head *parent_fe_bh,
75                               handle_t *handle,
76                               struct ocfs2_alloc_context *inode_ac);
77
78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
79                                     struct inode **ret_orphan_dir,
80                                     u64 blkno,
81                                     char *name,
82                                     struct ocfs2_dir_lookup_result *lookup);
83
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         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
110                            dentry->d_name.name,
111                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
112
113         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
114                 ret = ERR_PTR(-ENAMETOOLONG);
115                 goto bail;
116         }
117
118         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
119         if (status < 0) {
120                 if (status != -ENOENT)
121                         mlog_errno(status);
122                 ret = ERR_PTR(status);
123                 goto bail;
124         }
125
126         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
127                                             dentry->d_name.len, &blkno);
128         if (status < 0)
129                 goto bail_add;
130
131         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
132         if (IS_ERR(inode)) {
133                 ret = ERR_PTR(-EACCES);
134                 goto bail_unlock;
135         }
136
137         oi = OCFS2_I(inode);
138         /* Clear any orphaned state... If we were able to look up the
139          * inode from a directory, it certainly can't be orphaned. We
140          * might have the bad state from a node which intended to
141          * orphan this inode but crashed before it could commit the
142          * unlink. */
143         spin_lock(&oi->ip_lock);
144         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
145         spin_unlock(&oi->ip_lock);
146
147 bail_add:
148         ret = d_splice_alias(inode, dentry);
149
150         if (inode) {
151                 /*
152                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
153                  * dentry, it will d_move() it on top of ourse. The
154                  * return value will indicate this however, so in
155                  * those cases, we switch them around for the locking
156                  * code.
157                  *
158                  * NOTE: This dentry already has ->d_op set from
159                  * ocfs2_get_parent() and ocfs2_get_dentry()
160                  */
161                 if (ret)
162                         dentry = ret;
163
164                 status = ocfs2_dentry_attach_lock(dentry, inode,
165                                                   OCFS2_I(dir)->ip_blkno);
166                 if (status) {
167                         mlog_errno(status);
168                         ret = ERR_PTR(status);
169                         goto bail_unlock;
170                 }
171         } else
172                 ocfs2_dentry_attach_gen(dentry);
173
174 bail_unlock:
175         /* Don't drop the cluster lock until *after* the d_add --
176          * unlink on another node will message us to remove that
177          * dentry under this lock so otherwise we can race this with
178          * the downconvert thread and have a stale dentry. */
179         ocfs2_inode_unlock(dir, 0);
180
181 bail:
182
183         trace_ocfs2_lookup_ret(ret);
184
185         return ret;
186 }
187
188 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
189 {
190         struct inode *inode;
191
192         inode = new_inode(dir->i_sb);
193         if (!inode) {
194                 mlog(ML_ERROR, "new_inode failed!\n");
195                 return NULL;
196         }
197
198         /* populate as many fields early on as possible - many of
199          * these are used by the support functions here and in
200          * callers. */
201         if (S_ISDIR(mode))
202                 inode->i_nlink = 2;
203         else
204                 inode->i_nlink = 1;
205         inode_init_owner(inode, dir, mode);
206         dquot_initialize(inode);
207         return inode;
208 }
209
210 static int ocfs2_mknod(struct inode *dir,
211                        struct dentry *dentry,
212                        int mode,
213                        dev_t dev)
214 {
215         int status = 0;
216         struct buffer_head *parent_fe_bh = NULL;
217         handle_t *handle = NULL;
218         struct ocfs2_super *osb;
219         struct ocfs2_dinode *dirfe;
220         struct buffer_head *new_fe_bh = NULL;
221         struct inode *inode = NULL;
222         struct ocfs2_alloc_context *inode_ac = NULL;
223         struct ocfs2_alloc_context *data_ac = NULL;
224         struct ocfs2_alloc_context *meta_ac = NULL;
225         int want_clusters = 0;
226         int want_meta = 0;
227         int xattr_credits = 0;
228         struct ocfs2_security_xattr_info si = {
229                 .enable = 1,
230         };
231         int did_quota_inode = 0;
232         struct ocfs2_dir_lookup_result lookup = { NULL, };
233         sigset_t oldset;
234         int did_block_signals = 0;
235
236         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
237                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
238                           (unsigned long)dev, mode);
239
240         dquot_initialize(dir);
241
242         /* get our super block */
243         osb = OCFS2_SB(dir->i_sb);
244
245         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
246         if (status < 0) {
247                 if (status != -ENOENT)
248                         mlog_errno(status);
249                 return status;
250         }
251
252         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
253                 status = -EMLINK;
254                 goto leave;
255         }
256
257         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
258         if (!ocfs2_read_links_count(dirfe)) {
259                 /* can't make a file in a deleted directory. */
260                 status = -ENOENT;
261                 goto leave;
262         }
263
264         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
265                                            dentry->d_name.len);
266         if (status)
267                 goto leave;
268
269         /* get a spot inside the dir. */
270         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
271                                               dentry->d_name.name,
272                                               dentry->d_name.len, &lookup);
273         if (status < 0) {
274                 mlog_errno(status);
275                 goto leave;
276         }
277
278         /* reserve an inode spot */
279         status = ocfs2_reserve_new_inode(osb, &inode_ac);
280         if (status < 0) {
281                 if (status != -ENOSPC)
282                         mlog_errno(status);
283                 goto leave;
284         }
285
286         inode = ocfs2_get_init_inode(dir, mode);
287         if (!inode) {
288                 status = -ENOMEM;
289                 mlog_errno(status);
290                 goto leave;
291         }
292
293         /* get security xattr */
294         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
295         if (status) {
296                 if (status == -EOPNOTSUPP)
297                         si.enable = 0;
298                 else {
299                         mlog_errno(status);
300                         goto leave;
301                 }
302         }
303
304         /* calculate meta data/clusters for setting security and acl xattr */
305         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
306                                        &si, &want_clusters,
307                                        &xattr_credits, &want_meta);
308         if (status < 0) {
309                 mlog_errno(status);
310                 goto leave;
311         }
312
313         /* Reserve a cluster if creating an extent based directory. */
314         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
315                 want_clusters += 1;
316
317                 /* Dir indexing requires extra space as well */
318                 if (ocfs2_supports_indexed_dirs(osb))
319                         want_meta++;
320         }
321
322         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
323         if (status < 0) {
324                 if (status != -ENOSPC)
325                         mlog_errno(status);
326                 goto leave;
327         }
328
329         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
330         if (status < 0) {
331                 if (status != -ENOSPC)
332                         mlog_errno(status);
333                 goto leave;
334         }
335
336         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
337                                                             S_ISDIR(mode),
338                                                             xattr_credits));
339         if (IS_ERR(handle)) {
340                 status = PTR_ERR(handle);
341                 handle = NULL;
342                 mlog_errno(status);
343                 goto leave;
344         }
345
346         /* Starting to change things, restart is no longer possible. */
347         ocfs2_block_signals(&oldset);
348         did_block_signals = 1;
349
350         status = dquot_alloc_inode(inode);
351         if (status)
352                 goto leave;
353         did_quota_inode = 1;
354
355         /* do the real work now. */
356         status = ocfs2_mknod_locked(osb, dir, inode, dev,
357                                     &new_fe_bh, parent_fe_bh, handle,
358                                     inode_ac);
359         if (status < 0) {
360                 mlog_errno(status);
361                 goto leave;
362         }
363
364         if (S_ISDIR(mode)) {
365                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
366                                             new_fe_bh, data_ac, meta_ac);
367                 if (status < 0) {
368                         mlog_errno(status);
369                         goto leave;
370                 }
371
372                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
373                                                  parent_fe_bh,
374                                                  OCFS2_JOURNAL_ACCESS_WRITE);
375                 if (status < 0) {
376                         mlog_errno(status);
377                         goto leave;
378                 }
379                 ocfs2_add_links_count(dirfe, 1);
380                 ocfs2_journal_dirty(handle, parent_fe_bh);
381                 inc_nlink(dir);
382         }
383
384         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
385                                 meta_ac, data_ac);
386         if (status < 0) {
387                 mlog_errno(status);
388                 goto leave;
389         }
390
391         if (si.enable) {
392                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
393                                                  meta_ac, data_ac);
394                 if (status < 0) {
395                         mlog_errno(status);
396                         goto leave;
397                 }
398         }
399
400         /*
401          * Do this before adding the entry to the directory. We add
402          * also set d_op after success so that ->d_iput() will cleanup
403          * the dentry lock even if ocfs2_add_entry() fails below.
404          */
405         status = ocfs2_dentry_attach_lock(dentry, inode,
406                                           OCFS2_I(dir)->ip_blkno);
407         if (status) {
408                 mlog_errno(status);
409                 goto leave;
410         }
411
412         status = ocfs2_add_entry(handle, dentry, inode,
413                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
414                                  &lookup);
415         if (status < 0) {
416                 mlog_errno(status);
417                 goto leave;
418         }
419
420         insert_inode_hash(inode);
421         d_instantiate(dentry, inode);
422         status = 0;
423 leave:
424         if (status < 0 && did_quota_inode)
425                 dquot_free_inode(inode);
426         if (handle)
427                 ocfs2_commit_trans(osb, handle);
428
429         ocfs2_inode_unlock(dir, 1);
430         if (did_block_signals)
431                 ocfs2_unblock_signals(&oldset);
432
433         brelse(new_fe_bh);
434         brelse(parent_fe_bh);
435         kfree(si.name);
436         kfree(si.value);
437
438         ocfs2_free_dir_lookup_result(&lookup);
439
440         if (inode_ac)
441                 ocfs2_free_alloc_context(inode_ac);
442
443         if (data_ac)
444                 ocfs2_free_alloc_context(data_ac);
445
446         if (meta_ac)
447                 ocfs2_free_alloc_context(meta_ac);
448
449         /*
450          * We should call iput after the i_mutex of the bitmap been
451          * unlocked in ocfs2_free_alloc_context, or the
452          * ocfs2_delete_inode will mutex_lock again.
453          */
454         if ((status < 0) && inode) {
455                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
456                 clear_nlink(inode);
457                 iput(inode);
458         }
459
460         if (status)
461                 mlog_errno(status);
462
463         return status;
464 }
465
466 static int __ocfs2_mknod_locked(struct inode *dir,
467                                 struct inode *inode,
468                                 dev_t dev,
469                                 struct buffer_head **new_fe_bh,
470                                 struct buffer_head *parent_fe_bh,
471                                 handle_t *handle,
472                                 struct ocfs2_alloc_context *inode_ac,
473                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
474 {
475         int status = 0;
476         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
477         struct ocfs2_dinode *fe = NULL;
478         struct ocfs2_extent_list *fel;
479         u16 feat;
480
481         *new_fe_bh = NULL;
482
483         /* populate as many fields early on as possible - many of
484          * these are used by the support functions here and in
485          * callers. */
486         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
487         OCFS2_I(inode)->ip_blkno = fe_blkno;
488         spin_lock(&osb->osb_lock);
489         inode->i_generation = osb->s_next_generation++;
490         spin_unlock(&osb->osb_lock);
491
492         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
493         if (!*new_fe_bh) {
494                 status = -EIO;
495                 mlog_errno(status);
496                 goto leave;
497         }
498         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
499
500         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
501                                          *new_fe_bh,
502                                          OCFS2_JOURNAL_ACCESS_CREATE);
503         if (status < 0) {
504                 mlog_errno(status);
505                 goto leave;
506         }
507
508         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
509         memset(fe, 0, osb->sb->s_blocksize);
510
511         fe->i_generation = cpu_to_le32(inode->i_generation);
512         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
513         fe->i_blkno = cpu_to_le64(fe_blkno);
514         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
515         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
516         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
517         fe->i_uid = cpu_to_le32(inode->i_uid);
518         fe->i_gid = cpu_to_le32(inode->i_gid);
519         fe->i_mode = cpu_to_le16(inode->i_mode);
520         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
521                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
522
523         ocfs2_set_links_count(fe, inode->i_nlink);
524
525         fe->i_last_eb_blk = 0;
526         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
527         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
528         fe->i_atime = fe->i_ctime = fe->i_mtime =
529                 cpu_to_le64(CURRENT_TIME.tv_sec);
530         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
531                 cpu_to_le32(CURRENT_TIME.tv_nsec);
532         fe->i_dtime = 0;
533
534         /*
535          * If supported, directories start with inline data. If inline
536          * isn't supported, but indexing is, we start them as indexed.
537          */
538         feat = le16_to_cpu(fe->i_dyn_features);
539         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
540                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
541
542                 fe->id2.i_data.id_count = cpu_to_le16(
543                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
544         } else {
545                 fel = &fe->id2.i_list;
546                 fel->l_tree_depth = 0;
547                 fel->l_next_free_rec = 0;
548                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
549         }
550
551         ocfs2_journal_dirty(handle, *new_fe_bh);
552
553         ocfs2_populate_inode(inode, fe, 1);
554         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
555         if (!ocfs2_mount_local(osb)) {
556                 status = ocfs2_create_new_inode_locks(inode);
557                 if (status < 0)
558                         mlog_errno(status);
559         }
560
561         status = 0; /* error in ocfs2_create_new_inode_locks is not
562                      * critical */
563
564 leave:
565         if (status < 0) {
566                 if (*new_fe_bh) {
567                         brelse(*new_fe_bh);
568                         *new_fe_bh = NULL;
569                 }
570         }
571
572         if (status)
573                 mlog_errno(status);
574         return status;
575 }
576
577 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
578                               struct inode *dir,
579                               struct inode *inode,
580                               dev_t dev,
581                               struct buffer_head **new_fe_bh,
582                               struct buffer_head *parent_fe_bh,
583                               handle_t *handle,
584                               struct ocfs2_alloc_context *inode_ac)
585 {
586         int status = 0;
587         u64 suballoc_loc, fe_blkno = 0;
588         u16 suballoc_bit;
589
590         *new_fe_bh = NULL;
591
592         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
593                                        inode_ac, &suballoc_loc,
594                                        &suballoc_bit, &fe_blkno);
595         if (status < 0) {
596                 mlog_errno(status);
597                 return status;
598         }
599
600         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
601                                     parent_fe_bh, handle, inode_ac,
602                                     fe_blkno, suballoc_loc, suballoc_bit);
603 }
604
605 static int ocfs2_mkdir(struct inode *dir,
606                        struct dentry *dentry,
607                        int mode)
608 {
609         int ret;
610
611         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
612                           OCFS2_I(dir)->ip_blkno, mode);
613         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
614         if (ret)
615                 mlog_errno(ret);
616
617         return ret;
618 }
619
620 static int ocfs2_create(struct inode *dir,
621                         struct dentry *dentry,
622                         int mode,
623                         struct nameidata *nd)
624 {
625         int ret;
626
627         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
628                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
629         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
630         if (ret)
631                 mlog_errno(ret);
632
633         return ret;
634 }
635
636 static int ocfs2_link(struct dentry *old_dentry,
637                       struct inode *dir,
638                       struct dentry *dentry)
639 {
640         handle_t *handle;
641         struct inode *inode = old_dentry->d_inode;
642         int err;
643         struct buffer_head *fe_bh = NULL;
644         struct buffer_head *parent_fe_bh = NULL;
645         struct ocfs2_dinode *fe = NULL;
646         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
647         struct ocfs2_dir_lookup_result lookup = { NULL, };
648         sigset_t oldset;
649
650         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
651                          old_dentry->d_name.len, old_dentry->d_name.name,
652                          dentry->d_name.len, dentry->d_name.name);
653
654         if (S_ISDIR(inode->i_mode))
655                 return -EPERM;
656
657         dquot_initialize(dir);
658
659         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
660         if (err < 0) {
661                 if (err != -ENOENT)
662                         mlog_errno(err);
663                 return err;
664         }
665
666         if (!dir->i_nlink) {
667                 err = -ENOENT;
668                 goto out;
669         }
670
671         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
672                                         dentry->d_name.len);
673         if (err)
674                 goto out;
675
676         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
677                                            dentry->d_name.name,
678                                            dentry->d_name.len, &lookup);
679         if (err < 0) {
680                 mlog_errno(err);
681                 goto out;
682         }
683
684         err = ocfs2_inode_lock(inode, &fe_bh, 1);
685         if (err < 0) {
686                 if (err != -ENOENT)
687                         mlog_errno(err);
688                 goto out;
689         }
690
691         fe = (struct ocfs2_dinode *) fe_bh->b_data;
692         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
693                 err = -EMLINK;
694                 goto out_unlock_inode;
695         }
696
697         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
698         if (IS_ERR(handle)) {
699                 err = PTR_ERR(handle);
700                 handle = NULL;
701                 mlog_errno(err);
702                 goto out_unlock_inode;
703         }
704
705         /* Starting to change things, restart is no longer possible. */
706         ocfs2_block_signals(&oldset);
707
708         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
709                                       OCFS2_JOURNAL_ACCESS_WRITE);
710         if (err < 0) {
711                 mlog_errno(err);
712                 goto out_commit;
713         }
714
715         inc_nlink(inode);
716         inode->i_ctime = CURRENT_TIME;
717         ocfs2_set_links_count(fe, inode->i_nlink);
718         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
719         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
720         ocfs2_journal_dirty(handle, fe_bh);
721
722         err = ocfs2_add_entry(handle, dentry, inode,
723                               OCFS2_I(inode)->ip_blkno,
724                               parent_fe_bh, &lookup);
725         if (err) {
726                 ocfs2_add_links_count(fe, -1);
727                 drop_nlink(inode);
728                 mlog_errno(err);
729                 goto out_commit;
730         }
731
732         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
733         if (err) {
734                 mlog_errno(err);
735                 goto out_commit;
736         }
737
738         ihold(inode);
739         d_instantiate(dentry, inode);
740
741 out_commit:
742         ocfs2_commit_trans(osb, handle);
743         ocfs2_unblock_signals(&oldset);
744 out_unlock_inode:
745         ocfs2_inode_unlock(inode, 1);
746
747 out:
748         ocfs2_inode_unlock(dir, 1);
749
750         brelse(fe_bh);
751         brelse(parent_fe_bh);
752
753         ocfs2_free_dir_lookup_result(&lookup);
754
755         if (err)
756                 mlog_errno(err);
757
758         return err;
759 }
760
761 /*
762  * Takes and drops an exclusive lock on the given dentry. This will
763  * force other nodes to drop it.
764  */
765 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
766 {
767         int ret;
768
769         ret = ocfs2_dentry_lock(dentry, 1);
770         if (ret)
771                 mlog_errno(ret);
772         else
773                 ocfs2_dentry_unlock(dentry, 1);
774
775         return ret;
776 }
777
778 static inline int inode_is_unlinkable(struct inode *inode)
779 {
780         if (S_ISDIR(inode->i_mode)) {
781                 if (inode->i_nlink == 2)
782                         return 1;
783                 return 0;
784         }
785
786         if (inode->i_nlink == 1)
787                 return 1;
788         return 0;
789 }
790
791 static int ocfs2_unlink(struct inode *dir,
792                         struct dentry *dentry)
793 {
794         int status;
795         int child_locked = 0;
796         struct inode *inode = dentry->d_inode;
797         struct inode *orphan_dir = NULL;
798         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
799         u64 blkno;
800         struct ocfs2_dinode *fe = NULL;
801         struct buffer_head *fe_bh = NULL;
802         struct buffer_head *parent_node_bh = NULL;
803         handle_t *handle = NULL;
804         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
805         struct ocfs2_dir_lookup_result lookup = { NULL, };
806         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
807
808         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
809                            dentry->d_name.name,
810                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
811                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
812
813         if (S_ISDIR(inode->i_mode))
814                 dentry_unhash(dentry);
815
816         dquot_initialize(dir);
817
818         BUG_ON(dentry->d_parent->d_inode != dir);
819
820         if (inode == osb->root_inode)
821                 return -EPERM;
822
823         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
824                                          OI_LS_PARENT);
825         if (status < 0) {
826                 if (status != -ENOENT)
827                         mlog_errno(status);
828                 return status;
829         }
830
831         status = ocfs2_find_files_on_disk(dentry->d_name.name,
832                                           dentry->d_name.len, &blkno, dir,
833                                           &lookup);
834         if (status < 0) {
835                 if (status != -ENOENT)
836                         mlog_errno(status);
837                 goto leave;
838         }
839
840         if (OCFS2_I(inode)->ip_blkno != blkno) {
841                 status = -ENOENT;
842
843                 trace_ocfs2_unlink_noent(
844                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
845                                 (unsigned long long)blkno,
846                                 OCFS2_I(inode)->ip_flags);
847                 goto leave;
848         }
849
850         status = ocfs2_inode_lock(inode, &fe_bh, 1);
851         if (status < 0) {
852                 if (status != -ENOENT)
853                         mlog_errno(status);
854                 goto leave;
855         }
856         child_locked = 1;
857
858         if (S_ISDIR(inode->i_mode)) {
859                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
860                         status = -ENOTEMPTY;
861                         goto leave;
862                 }
863         }
864
865         status = ocfs2_remote_dentry_delete(dentry);
866         if (status < 0) {
867                 /* This remote delete should succeed under all normal
868                  * circumstances. */
869                 mlog_errno(status);
870                 goto leave;
871         }
872
873         if (inode_is_unlinkable(inode)) {
874                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
875                                                   OCFS2_I(inode)->ip_blkno,
876                                                   orphan_name, &orphan_insert);
877                 if (status < 0) {
878                         mlog_errno(status);
879                         goto leave;
880                 }
881         }
882
883         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
884         if (IS_ERR(handle)) {
885                 status = PTR_ERR(handle);
886                 handle = NULL;
887                 mlog_errno(status);
888                 goto leave;
889         }
890
891         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
892                                          OCFS2_JOURNAL_ACCESS_WRITE);
893         if (status < 0) {
894                 mlog_errno(status);
895                 goto leave;
896         }
897
898         fe = (struct ocfs2_dinode *) fe_bh->b_data;
899
900         if (inode_is_unlinkable(inode)) {
901                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
902                                           &orphan_insert, orphan_dir);
903                 if (status < 0) {
904                         mlog_errno(status);
905                         goto leave;
906                 }
907         }
908
909         /* delete the name from the parent dir */
910         status = ocfs2_delete_entry(handle, dir, &lookup);
911         if (status < 0) {
912                 mlog_errno(status);
913                 goto leave;
914         }
915
916         if (S_ISDIR(inode->i_mode))
917                 drop_nlink(inode);
918         drop_nlink(inode);
919         ocfs2_set_links_count(fe, inode->i_nlink);
920         ocfs2_journal_dirty(handle, fe_bh);
921
922         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
923         if (S_ISDIR(inode->i_mode))
924                 drop_nlink(dir);
925
926         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
927         if (status < 0) {
928                 mlog_errno(status);
929                 if (S_ISDIR(inode->i_mode))
930                         inc_nlink(dir);
931         }
932
933 leave:
934         if (handle)
935                 ocfs2_commit_trans(osb, handle);
936
937         if (child_locked)
938                 ocfs2_inode_unlock(inode, 1);
939
940         ocfs2_inode_unlock(dir, 1);
941
942         if (orphan_dir) {
943                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
944                 ocfs2_inode_unlock(orphan_dir, 1);
945                 mutex_unlock(&orphan_dir->i_mutex);
946                 iput(orphan_dir);
947         }
948
949         brelse(fe_bh);
950         brelse(parent_node_bh);
951
952         ocfs2_free_dir_lookup_result(&orphan_insert);
953         ocfs2_free_dir_lookup_result(&lookup);
954
955         if (status)
956                 mlog_errno(status);
957
958         return status;
959 }
960
961 /*
962  * The only place this should be used is rename!
963  * if they have the same id, then the 1st one is the only one locked.
964  */
965 static int ocfs2_double_lock(struct ocfs2_super *osb,
966                              struct buffer_head **bh1,
967                              struct inode *inode1,
968                              struct buffer_head **bh2,
969                              struct inode *inode2)
970 {
971         int status;
972         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
973         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
974         struct buffer_head **tmpbh;
975         struct inode *tmpinode;
976
977         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
978                                 (unsigned long long)oi2->ip_blkno);
979
980         if (*bh1)
981                 *bh1 = NULL;
982         if (*bh2)
983                 *bh2 = NULL;
984
985         /* we always want to lock the one with the lower lockid first. */
986         if (oi1->ip_blkno != oi2->ip_blkno) {
987                 if (oi1->ip_blkno < oi2->ip_blkno) {
988                         /* switch id1 and id2 around */
989                         tmpbh = bh2;
990                         bh2 = bh1;
991                         bh1 = tmpbh;
992
993                         tmpinode = inode2;
994                         inode2 = inode1;
995                         inode1 = tmpinode;
996                 }
997                 /* lock id2 */
998                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
999                                                  OI_LS_RENAME1);
1000                 if (status < 0) {
1001                         if (status != -ENOENT)
1002                                 mlog_errno(status);
1003                         goto bail;
1004                 }
1005         }
1006
1007         /* lock id1 */
1008         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1009         if (status < 0) {
1010                 /*
1011                  * An error return must mean that no cluster locks
1012                  * were held on function exit.
1013                  */
1014                 if (oi1->ip_blkno != oi2->ip_blkno) {
1015                         ocfs2_inode_unlock(inode2, 1);
1016                         brelse(*bh2);
1017                         *bh2 = NULL;
1018                 }
1019
1020                 if (status != -ENOENT)
1021                         mlog_errno(status);
1022         }
1023
1024         trace_ocfs2_double_lock_end(
1025                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1026                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1027
1028 bail:
1029         if (status)
1030                 mlog_errno(status);
1031         return status;
1032 }
1033
1034 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1035 {
1036         ocfs2_inode_unlock(inode1, 1);
1037
1038         if (inode1 != inode2)
1039                 ocfs2_inode_unlock(inode2, 1);
1040 }
1041
1042 static int ocfs2_rename(struct inode *old_dir,
1043                         struct dentry *old_dentry,
1044                         struct inode *new_dir,
1045                         struct dentry *new_dentry)
1046 {
1047         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1048         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1049         struct inode *old_inode = old_dentry->d_inode;
1050         struct inode *new_inode = new_dentry->d_inode;
1051         struct inode *orphan_dir = NULL;
1052         struct ocfs2_dinode *newfe = NULL;
1053         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1054         struct buffer_head *newfe_bh = NULL;
1055         struct buffer_head *old_inode_bh = NULL;
1056         struct ocfs2_super *osb = NULL;
1057         u64 newfe_blkno, old_de_ino;
1058         handle_t *handle = NULL;
1059         struct buffer_head *old_dir_bh = NULL;
1060         struct buffer_head *new_dir_bh = NULL;
1061         nlink_t old_dir_nlink = old_dir->i_nlink;
1062         struct ocfs2_dinode *old_di;
1063         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1064         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1065         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1066         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1067         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1068
1069         if (new_inode && S_ISDIR(new_inode->i_mode))
1070                 dentry_unhash(new_dentry);
1071
1072         /* At some point it might be nice to break this function up a
1073          * bit. */
1074
1075         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1076                            old_dentry->d_name.len, old_dentry->d_name.name,
1077                            new_dentry->d_name.len, new_dentry->d_name.name);
1078
1079         dquot_initialize(old_dir);
1080         dquot_initialize(new_dir);
1081
1082         osb = OCFS2_SB(old_dir->i_sb);
1083
1084         if (new_inode) {
1085                 if (!igrab(new_inode))
1086                         BUG();
1087         }
1088
1089         /* Assume a directory hierarchy thusly:
1090          * a/b/c
1091          * a/d
1092          * a,b,c, and d are all directories.
1093          *
1094          * from cwd of 'a' on both nodes:
1095          * node1: mv b/c d
1096          * node2: mv d   b/c
1097          *
1098          * And that's why, just like the VFS, we need a file system
1099          * rename lock. */
1100         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1101                 status = ocfs2_rename_lock(osb);
1102                 if (status < 0) {
1103                         mlog_errno(status);
1104                         goto bail;
1105                 }
1106                 rename_lock = 1;
1107         }
1108
1109         /* if old and new are the same, this'll just do one lock. */
1110         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1111                                    &new_dir_bh, new_dir);
1112         if (status < 0) {
1113                 mlog_errno(status);
1114                 goto bail;
1115         }
1116         parents_locked = 1;
1117
1118         /* make sure both dirs have bhs
1119          * get an extra ref on old_dir_bh if old==new */
1120         if (!new_dir_bh) {
1121                 if (old_dir_bh) {
1122                         new_dir_bh = old_dir_bh;
1123                         get_bh(new_dir_bh);
1124                 } else {
1125                         mlog(ML_ERROR, "no old_dir_bh!\n");
1126                         status = -EIO;
1127                         goto bail;
1128                 }
1129         }
1130
1131         /*
1132          * Aside from allowing a meta data update, the locking here
1133          * also ensures that the downconvert thread on other nodes
1134          * won't have to concurrently downconvert the inode and the
1135          * dentry locks.
1136          */
1137         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1138                                          OI_LS_PARENT);
1139         if (status < 0) {
1140                 if (status != -ENOENT)
1141                         mlog_errno(status);
1142                 goto bail;
1143         }
1144         old_child_locked = 1;
1145
1146         status = ocfs2_remote_dentry_delete(old_dentry);
1147         if (status < 0) {
1148                 mlog_errno(status);
1149                 goto bail;
1150         }
1151
1152         if (S_ISDIR(old_inode->i_mode)) {
1153                 u64 old_inode_parent;
1154
1155                 update_dot_dot = 1;
1156                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1157                                                   old_inode,
1158                                                   &old_inode_dot_dot_res);
1159                 if (status) {
1160                         status = -EIO;
1161                         goto bail;
1162                 }
1163
1164                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1165                         status = -EIO;
1166                         goto bail;
1167                 }
1168
1169                 if (!new_inode && new_dir != old_dir &&
1170                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1171                         status = -EMLINK;
1172                         goto bail;
1173                 }
1174         }
1175
1176         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1177                                             old_dentry->d_name.len,
1178                                             &old_de_ino);
1179         if (status) {
1180                 status = -ENOENT;
1181                 goto bail;
1182         }
1183
1184         /*
1185          *  Check for inode number is _not_ due to possible IO errors.
1186          *  We might rmdir the source, keep it as pwd of some process
1187          *  and merrily kill the link to whatever was created under the
1188          *  same name. Goodbye sticky bit ;-<
1189          */
1190         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1191                 status = -ENOENT;
1192                 goto bail;
1193         }
1194
1195         /* check if the target already exists (in which case we need
1196          * to delete it */
1197         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1198                                           new_dentry->d_name.len,
1199                                           &newfe_blkno, new_dir,
1200                                           &target_lookup_res);
1201         /* The only error we allow here is -ENOENT because the new
1202          * file not existing is perfectly valid. */
1203         if ((status < 0) && (status != -ENOENT)) {
1204                 /* If we cannot find the file specified we should just */
1205                 /* return the error... */
1206                 mlog_errno(status);
1207                 goto bail;
1208         }
1209         if (status == 0)
1210                 target_exists = 1;
1211
1212         if (!target_exists && new_inode) {
1213                 /*
1214                  * Target was unlinked by another node while we were
1215                  * waiting to get to ocfs2_rename(). There isn't
1216                  * anything we can do here to help the situation, so
1217                  * bubble up the appropriate error.
1218                  */
1219                 status = -ENOENT;
1220                 goto bail;
1221         }
1222
1223         /* In case we need to overwrite an existing file, we blow it
1224          * away first */
1225         if (target_exists) {
1226                 /* VFS didn't think there existed an inode here, but
1227                  * someone else in the cluster must have raced our
1228                  * rename to create one. Today we error cleanly, in
1229                  * the future we should consider calling iget to build
1230                  * a new struct inode for this entry. */
1231                 if (!new_inode) {
1232                         status = -EACCES;
1233
1234                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1235                                                 new_dentry->d_name.name);
1236                         goto bail;
1237                 }
1238
1239                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1240                         status = -EACCES;
1241
1242                         trace_ocfs2_rename_disagree(
1243                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1244                              (unsigned long long)newfe_blkno,
1245                              OCFS2_I(new_inode)->ip_flags);
1246                         goto bail;
1247                 }
1248
1249                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1250                 if (status < 0) {
1251                         if (status != -ENOENT)
1252                                 mlog_errno(status);
1253                         goto bail;
1254                 }
1255                 new_child_locked = 1;
1256
1257                 status = ocfs2_remote_dentry_delete(new_dentry);
1258                 if (status < 0) {
1259                         mlog_errno(status);
1260                         goto bail;
1261                 }
1262
1263                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1264
1265                 trace_ocfs2_rename_over_existing(
1266                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1267                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1268
1269                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1270                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1271                                                 OCFS2_I(new_inode)->ip_blkno,
1272                                                 orphan_name, &orphan_insert);
1273                         if (status < 0) {
1274                                 mlog_errno(status);
1275                                 goto bail;
1276                         }
1277                 }
1278         } else {
1279                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1280
1281                 status = ocfs2_check_dir_for_entry(new_dir,
1282                                                    new_dentry->d_name.name,
1283                                                    new_dentry->d_name.len);
1284                 if (status)
1285                         goto bail;
1286
1287                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1288                                                       new_dentry->d_name.name,
1289                                                       new_dentry->d_name.len,
1290                                                       &target_insert);
1291                 if (status < 0) {
1292                         mlog_errno(status);
1293                         goto bail;
1294                 }
1295         }
1296
1297         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1298         if (IS_ERR(handle)) {
1299                 status = PTR_ERR(handle);
1300                 handle = NULL;
1301                 mlog_errno(status);
1302                 goto bail;
1303         }
1304
1305         if (target_exists) {
1306                 if (S_ISDIR(new_inode->i_mode)) {
1307                         if (new_inode->i_nlink != 2 ||
1308                             !ocfs2_empty_dir(new_inode)) {
1309                                 status = -ENOTEMPTY;
1310                                 goto bail;
1311                         }
1312                 }
1313                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1314                                                  newfe_bh,
1315                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1316                 if (status < 0) {
1317                         mlog_errno(status);
1318                         goto bail;
1319                 }
1320
1321                 if (S_ISDIR(new_inode->i_mode) ||
1322                     (ocfs2_read_links_count(newfe) == 1)) {
1323                         status = ocfs2_orphan_add(osb, handle, new_inode,
1324                                                   newfe_bh, orphan_name,
1325                                                   &orphan_insert, orphan_dir);
1326                         if (status < 0) {
1327                                 mlog_errno(status);
1328                                 goto bail;
1329                         }
1330                 }
1331
1332                 /* change the dirent to point to the correct inode */
1333                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1334                                             old_inode);
1335                 if (status < 0) {
1336                         mlog_errno(status);
1337                         goto bail;
1338                 }
1339                 new_dir->i_version++;
1340
1341                 if (S_ISDIR(new_inode->i_mode))
1342                         ocfs2_set_links_count(newfe, 0);
1343                 else
1344                         ocfs2_add_links_count(newfe, -1);
1345                 ocfs2_journal_dirty(handle, newfe_bh);
1346         } else {
1347                 /* if the name was not found in new_dir, add it now */
1348                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1349                                          OCFS2_I(old_inode)->ip_blkno,
1350                                          new_dir_bh, &target_insert);
1351         }
1352
1353         old_inode->i_ctime = CURRENT_TIME;
1354         mark_inode_dirty(old_inode);
1355
1356         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1357                                          old_inode_bh,
1358                                          OCFS2_JOURNAL_ACCESS_WRITE);
1359         if (status >= 0) {
1360                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1361
1362                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1363                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1364                 ocfs2_journal_dirty(handle, old_inode_bh);
1365         } else
1366                 mlog_errno(status);
1367
1368         /*
1369          * Now that the name has been added to new_dir, remove the old name.
1370          *
1371          * We don't keep any directory entry context around until now
1372          * because the insert might have changed the type of directory
1373          * we're dealing with.
1374          */
1375         status = ocfs2_find_entry(old_dentry->d_name.name,
1376                                   old_dentry->d_name.len, old_dir,
1377                                   &old_entry_lookup);
1378         if (status)
1379                 goto bail;
1380
1381         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1382         if (status < 0) {
1383                 mlog_errno(status);
1384                 goto bail;
1385         }
1386
1387         if (new_inode) {
1388                 new_inode->i_nlink--;
1389                 new_inode->i_ctime = CURRENT_TIME;
1390         }
1391         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1392
1393         if (update_dot_dot) {
1394                 status = ocfs2_update_entry(old_inode, handle,
1395                                             &old_inode_dot_dot_res, new_dir);
1396                 old_dir->i_nlink--;
1397                 if (new_inode) {
1398                         new_inode->i_nlink--;
1399                 } else {
1400                         inc_nlink(new_dir);
1401                         mark_inode_dirty(new_dir);
1402                 }
1403         }
1404         mark_inode_dirty(old_dir);
1405         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1406         if (new_inode) {
1407                 mark_inode_dirty(new_inode);
1408                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1409         }
1410
1411         if (old_dir != new_dir) {
1412                 /* Keep the same times on both directories.*/
1413                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1414
1415                 /*
1416                  * This will also pick up the i_nlink change from the
1417                  * block above.
1418                  */
1419                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1420         }
1421
1422         if (old_dir_nlink != old_dir->i_nlink) {
1423                 if (!old_dir_bh) {
1424                         mlog(ML_ERROR, "need to change nlink for old dir "
1425                              "%llu from %d to %d but bh is NULL!\n",
1426                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1427                              (int)old_dir_nlink, old_dir->i_nlink);
1428                 } else {
1429                         struct ocfs2_dinode *fe;
1430                         status = ocfs2_journal_access_di(handle,
1431                                                          INODE_CACHE(old_dir),
1432                                                          old_dir_bh,
1433                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1434                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1435                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1436                         ocfs2_journal_dirty(handle, old_dir_bh);
1437                 }
1438         }
1439         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1440         status = 0;
1441 bail:
1442         if (rename_lock)
1443                 ocfs2_rename_unlock(osb);
1444
1445         if (handle)
1446                 ocfs2_commit_trans(osb, handle);
1447
1448         if (parents_locked)
1449                 ocfs2_double_unlock(old_dir, new_dir);
1450
1451         if (old_child_locked)
1452                 ocfs2_inode_unlock(old_inode, 1);
1453
1454         if (new_child_locked)
1455                 ocfs2_inode_unlock(new_inode, 1);
1456
1457         if (orphan_dir) {
1458                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1459                 ocfs2_inode_unlock(orphan_dir, 1);
1460                 mutex_unlock(&orphan_dir->i_mutex);
1461                 iput(orphan_dir);
1462         }
1463
1464         if (new_inode)
1465                 sync_mapping_buffers(old_inode->i_mapping);
1466
1467         if (new_inode)
1468                 iput(new_inode);
1469
1470         ocfs2_free_dir_lookup_result(&target_lookup_res);
1471         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1472         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1473         ocfs2_free_dir_lookup_result(&orphan_insert);
1474         ocfs2_free_dir_lookup_result(&target_insert);
1475
1476         brelse(newfe_bh);
1477         brelse(old_inode_bh);
1478         brelse(old_dir_bh);
1479         brelse(new_dir_bh);
1480
1481         if (status)
1482                 mlog_errno(status);
1483
1484         return status;
1485 }
1486
1487 /*
1488  * we expect i_size = strlen(symname). Copy symname into the file
1489  * data, including the null terminator.
1490  */
1491 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1492                                      handle_t *handle,
1493                                      struct inode *inode,
1494                                      const char *symname)
1495 {
1496         struct buffer_head **bhs = NULL;
1497         const char *c;
1498         struct super_block *sb = osb->sb;
1499         u64 p_blkno, p_blocks;
1500         int virtual, blocks, status, i, bytes_left;
1501
1502         bytes_left = i_size_read(inode) + 1;
1503         /* we can't trust i_blocks because we're actually going to
1504          * write i_size + 1 bytes. */
1505         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1506
1507         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1508                                         i_size_read(inode), blocks);
1509
1510         /* Sanity check -- make sure we're going to fit. */
1511         if (bytes_left >
1512             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1513                 status = -EIO;
1514                 mlog_errno(status);
1515                 goto bail;
1516         }
1517
1518         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1519         if (!bhs) {
1520                 status = -ENOMEM;
1521                 mlog_errno(status);
1522                 goto bail;
1523         }
1524
1525         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1526                                              NULL);
1527         if (status < 0) {
1528                 mlog_errno(status);
1529                 goto bail;
1530         }
1531
1532         /* links can never be larger than one cluster so we know this
1533          * is all going to be contiguous, but do a sanity check
1534          * anyway. */
1535         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1536                 status = -EIO;
1537                 mlog_errno(status);
1538                 goto bail;
1539         }
1540
1541         virtual = 0;
1542         while(bytes_left > 0) {
1543                 c = &symname[virtual * sb->s_blocksize];
1544
1545                 bhs[virtual] = sb_getblk(sb, p_blkno);
1546                 if (!bhs[virtual]) {
1547                         status = -ENOMEM;
1548                         mlog_errno(status);
1549                         goto bail;
1550                 }
1551                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1552                                               bhs[virtual]);
1553
1554                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1555                                               bhs[virtual],
1556                                               OCFS2_JOURNAL_ACCESS_CREATE);
1557                 if (status < 0) {
1558                         mlog_errno(status);
1559                         goto bail;
1560                 }
1561
1562                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1563
1564                 memcpy(bhs[virtual]->b_data, c,
1565                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1566                        bytes_left);
1567
1568                 ocfs2_journal_dirty(handle, bhs[virtual]);
1569
1570                 virtual++;
1571                 p_blkno++;
1572                 bytes_left -= sb->s_blocksize;
1573         }
1574
1575         status = 0;
1576 bail:
1577
1578         if (bhs) {
1579                 for(i = 0; i < blocks; i++)
1580                         brelse(bhs[i]);
1581                 kfree(bhs);
1582         }
1583
1584         if (status)
1585                 mlog_errno(status);
1586         return status;
1587 }
1588
1589 static int ocfs2_symlink(struct inode *dir,
1590                          struct dentry *dentry,
1591                          const char *symname)
1592 {
1593         int status, l, credits;
1594         u64 newsize;
1595         struct ocfs2_super *osb = NULL;
1596         struct inode *inode = NULL;
1597         struct super_block *sb;
1598         struct buffer_head *new_fe_bh = NULL;
1599         struct buffer_head *parent_fe_bh = NULL;
1600         struct ocfs2_dinode *fe = NULL;
1601         struct ocfs2_dinode *dirfe;
1602         handle_t *handle = NULL;
1603         struct ocfs2_alloc_context *inode_ac = NULL;
1604         struct ocfs2_alloc_context *data_ac = NULL;
1605         struct ocfs2_alloc_context *xattr_ac = NULL;
1606         int want_clusters = 0;
1607         int xattr_credits = 0;
1608         struct ocfs2_security_xattr_info si = {
1609                 .enable = 1,
1610         };
1611         int did_quota = 0, did_quota_inode = 0;
1612         struct ocfs2_dir_lookup_result lookup = { NULL, };
1613         sigset_t oldset;
1614         int did_block_signals = 0;
1615
1616         trace_ocfs2_symlink_begin(dir, dentry, symname,
1617                                   dentry->d_name.len, dentry->d_name.name);
1618
1619         dquot_initialize(dir);
1620
1621         sb = dir->i_sb;
1622         osb = OCFS2_SB(sb);
1623
1624         l = strlen(symname) + 1;
1625
1626         credits = ocfs2_calc_symlink_credits(sb);
1627
1628         /* lock the parent directory */
1629         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1630         if (status < 0) {
1631                 if (status != -ENOENT)
1632                         mlog_errno(status);
1633                 return status;
1634         }
1635
1636         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1637         if (!ocfs2_read_links_count(dirfe)) {
1638                 /* can't make a file in a deleted directory. */
1639                 status = -ENOENT;
1640                 goto bail;
1641         }
1642
1643         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1644                                            dentry->d_name.len);
1645         if (status)
1646                 goto bail;
1647
1648         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1649                                               dentry->d_name.name,
1650                                               dentry->d_name.len, &lookup);
1651         if (status < 0) {
1652                 mlog_errno(status);
1653                 goto bail;
1654         }
1655
1656         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1657         if (status < 0) {
1658                 if (status != -ENOSPC)
1659                         mlog_errno(status);
1660                 goto bail;
1661         }
1662
1663         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1664         if (!inode) {
1665                 status = -ENOMEM;
1666                 mlog_errno(status);
1667                 goto bail;
1668         }
1669
1670         /* get security xattr */
1671         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1672         if (status) {
1673                 if (status == -EOPNOTSUPP)
1674                         si.enable = 0;
1675                 else {
1676                         mlog_errno(status);
1677                         goto bail;
1678                 }
1679         }
1680
1681         /* calculate meta data/clusters for setting security xattr */
1682         if (si.enable) {
1683                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1684                                                   &xattr_credits, &xattr_ac);
1685                 if (status < 0) {
1686                         mlog_errno(status);
1687                         goto bail;
1688                 }
1689         }
1690
1691         /* don't reserve bitmap space for fast symlinks. */
1692         if (l > ocfs2_fast_symlink_chars(sb))
1693                 want_clusters += 1;
1694
1695         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1696         if (status < 0) {
1697                 if (status != -ENOSPC)
1698                         mlog_errno(status);
1699                 goto bail;
1700         }
1701
1702         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1703         if (IS_ERR(handle)) {
1704                 status = PTR_ERR(handle);
1705                 handle = NULL;
1706                 mlog_errno(status);
1707                 goto bail;
1708         }
1709
1710         /* Starting to change things, restart is no longer possible. */
1711         ocfs2_block_signals(&oldset);
1712         did_block_signals = 1;
1713
1714         status = dquot_alloc_inode(inode);
1715         if (status)
1716                 goto bail;
1717         did_quota_inode = 1;
1718
1719         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1720                                    dentry->d_name.name,
1721                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1722                                    inode->i_mode);
1723
1724         status = ocfs2_mknod_locked(osb, dir, inode,
1725                                     0, &new_fe_bh, parent_fe_bh, handle,
1726                                     inode_ac);
1727         if (status < 0) {
1728                 mlog_errno(status);
1729                 goto bail;
1730         }
1731
1732         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1733         inode->i_rdev = 0;
1734         newsize = l - 1;
1735         if (l > ocfs2_fast_symlink_chars(sb)) {
1736                 u32 offset = 0;
1737
1738                 inode->i_op = &ocfs2_symlink_inode_operations;
1739                 status = dquot_alloc_space_nodirty(inode,
1740                     ocfs2_clusters_to_bytes(osb->sb, 1));
1741                 if (status)
1742                         goto bail;
1743                 did_quota = 1;
1744                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1745                                               new_fe_bh,
1746                                               handle, data_ac, NULL,
1747                                               NULL);
1748                 if (status < 0) {
1749                         if (status != -ENOSPC && status != -EINTR) {
1750                                 mlog(ML_ERROR,
1751                                      "Failed to extend file to %llu\n",
1752                                      (unsigned long long)newsize);
1753                                 mlog_errno(status);
1754                                 status = -ENOSPC;
1755                         }
1756                         goto bail;
1757                 }
1758                 i_size_write(inode, newsize);
1759                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1760         } else {
1761                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1762                 memcpy((char *) fe->id2.i_symlink, symname, l);
1763                 i_size_write(inode, newsize);
1764                 inode->i_blocks = 0;
1765         }
1766
1767         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1768         if (status < 0) {
1769                 mlog_errno(status);
1770                 goto bail;
1771         }
1772
1773         if (!ocfs2_inode_is_fast_symlink(inode)) {
1774                 status = ocfs2_create_symlink_data(osb, handle, inode,
1775                                                    symname);
1776                 if (status < 0) {
1777                         mlog_errno(status);
1778                         goto bail;
1779                 }
1780         }
1781
1782         if (si.enable) {
1783                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1784                                                  xattr_ac, data_ac);
1785                 if (status < 0) {
1786                         mlog_errno(status);
1787                         goto bail;
1788                 }
1789         }
1790
1791         /*
1792          * Do this before adding the entry to the directory. We add
1793          * also set d_op after success so that ->d_iput() will cleanup
1794          * the dentry lock even if ocfs2_add_entry() fails below.
1795          */
1796         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1797         if (status) {
1798                 mlog_errno(status);
1799                 goto bail;
1800         }
1801
1802         status = ocfs2_add_entry(handle, dentry, inode,
1803                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1804                                  &lookup);
1805         if (status < 0) {
1806                 mlog_errno(status);
1807                 goto bail;
1808         }
1809
1810         insert_inode_hash(inode);
1811         d_instantiate(dentry, inode);
1812 bail:
1813         if (status < 0 && did_quota)
1814                 dquot_free_space_nodirty(inode,
1815                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1816         if (status < 0 && did_quota_inode)
1817                 dquot_free_inode(inode);
1818         if (handle)
1819                 ocfs2_commit_trans(osb, handle);
1820
1821         ocfs2_inode_unlock(dir, 1);
1822         if (did_block_signals)
1823                 ocfs2_unblock_signals(&oldset);
1824
1825         brelse(new_fe_bh);
1826         brelse(parent_fe_bh);
1827         kfree(si.name);
1828         kfree(si.value);
1829         ocfs2_free_dir_lookup_result(&lookup);
1830         if (inode_ac)
1831                 ocfs2_free_alloc_context(inode_ac);
1832         if (data_ac)
1833                 ocfs2_free_alloc_context(data_ac);
1834         if (xattr_ac)
1835                 ocfs2_free_alloc_context(xattr_ac);
1836         if ((status < 0) && inode) {
1837                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1838                 clear_nlink(inode);
1839                 iput(inode);
1840         }
1841
1842         if (status)
1843                 mlog_errno(status);
1844
1845         return status;
1846 }
1847
1848 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1849 {
1850         int status, namelen;
1851
1852         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1853                            (long long)blkno);
1854         if (namelen <= 0) {
1855                 if (namelen)
1856                         status = namelen;
1857                 else
1858                         status = -EINVAL;
1859                 mlog_errno(status);
1860                 goto bail;
1861         }
1862         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1863                 status = -EINVAL;
1864                 mlog_errno(status);
1865                 goto bail;
1866         }
1867
1868         trace_ocfs2_blkno_stringify(blkno, name, namelen);
1869
1870         status = 0;
1871 bail:
1872         if (status < 0)
1873                 mlog_errno(status);
1874         return status;
1875 }
1876
1877 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1878                                         struct inode **ret_orphan_dir,
1879                                         struct buffer_head **ret_orphan_dir_bh)
1880 {
1881         struct inode *orphan_dir_inode;
1882         struct buffer_head *orphan_dir_bh = NULL;
1883         int ret = 0;
1884
1885         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1886                                                        ORPHAN_DIR_SYSTEM_INODE,
1887                                                        osb->slot_num);
1888         if (!orphan_dir_inode) {
1889                 ret = -ENOENT;
1890                 mlog_errno(ret);
1891                 return ret;
1892         }
1893
1894         mutex_lock(&orphan_dir_inode->i_mutex);
1895
1896         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1897         if (ret < 0) {
1898                 mutex_unlock(&orphan_dir_inode->i_mutex);
1899                 iput(orphan_dir_inode);
1900
1901                 mlog_errno(ret);
1902                 return ret;
1903         }
1904
1905         *ret_orphan_dir = orphan_dir_inode;
1906         *ret_orphan_dir_bh = orphan_dir_bh;
1907
1908         return 0;
1909 }
1910
1911 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1912                                       struct buffer_head *orphan_dir_bh,
1913                                       u64 blkno,
1914                                       char *name,
1915                                       struct ocfs2_dir_lookup_result *lookup)
1916 {
1917         int ret;
1918         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1919
1920         ret = ocfs2_blkno_stringify(blkno, name);
1921         if (ret < 0) {
1922                 mlog_errno(ret);
1923                 return ret;
1924         }
1925
1926         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1927                                            orphan_dir_bh, name,
1928                                            OCFS2_ORPHAN_NAMELEN, lookup);
1929         if (ret < 0) {
1930                 mlog_errno(ret);
1931                 return ret;
1932         }
1933
1934         return 0;
1935 }
1936
1937 /**
1938  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1939  * insertion of an orphan.
1940  * @osb: ocfs2 file system
1941  * @ret_orphan_dir: Orphan dir inode - returned locked!
1942  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1943  * @lookup: dir lookup result, to be passed back into functions like
1944  *          ocfs2_orphan_add
1945  *
1946  * Returns zero on success and the ret_orphan_dir, name and lookup
1947  * fields will be populated.
1948  *
1949  * Returns non-zero on failure. 
1950  */
1951 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1952                                     struct inode **ret_orphan_dir,
1953                                     u64 blkno,
1954                                     char *name,
1955                                     struct ocfs2_dir_lookup_result *lookup)
1956 {
1957         struct inode *orphan_dir_inode = NULL;
1958         struct buffer_head *orphan_dir_bh = NULL;
1959         int ret = 0;
1960
1961         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1962                                            &orphan_dir_bh);
1963         if (ret < 0) {
1964                 mlog_errno(ret);
1965                 return ret;
1966         }
1967
1968         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1969                                          blkno, name, lookup);
1970         if (ret < 0) {
1971                 mlog_errno(ret);
1972                 goto out;
1973         }
1974
1975         *ret_orphan_dir = orphan_dir_inode;
1976
1977 out:
1978         brelse(orphan_dir_bh);
1979
1980         if (ret) {
1981                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1982                 mutex_unlock(&orphan_dir_inode->i_mutex);
1983                 iput(orphan_dir_inode);
1984         }
1985
1986         if (ret)
1987                 mlog_errno(ret);
1988         return ret;
1989 }
1990
1991 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1992                             handle_t *handle,
1993                             struct inode *inode,
1994                             struct buffer_head *fe_bh,
1995                             char *name,
1996                             struct ocfs2_dir_lookup_result *lookup,
1997                             struct inode *orphan_dir_inode)
1998 {
1999         struct buffer_head *orphan_dir_bh = NULL;
2000         int status = 0;
2001         struct ocfs2_dinode *orphan_fe;
2002         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2003
2004         trace_ocfs2_orphan_add_begin(
2005                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2006
2007         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2008         if (status < 0) {
2009                 mlog_errno(status);
2010                 goto leave;
2011         }
2012
2013         status = ocfs2_journal_access_di(handle,
2014                                          INODE_CACHE(orphan_dir_inode),
2015                                          orphan_dir_bh,
2016                                          OCFS2_JOURNAL_ACCESS_WRITE);
2017         if (status < 0) {
2018                 mlog_errno(status);
2019                 goto leave;
2020         }
2021
2022         /* we're a cluster, and nlink can change on disk from
2023          * underneath us... */
2024         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2025         if (S_ISDIR(inode->i_mode))
2026                 ocfs2_add_links_count(orphan_fe, 1);
2027         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2028         ocfs2_journal_dirty(handle, orphan_dir_bh);
2029
2030         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2031                                    OCFS2_ORPHAN_NAMELEN, inode,
2032                                    OCFS2_I(inode)->ip_blkno,
2033                                    orphan_dir_bh, lookup);
2034         if (status < 0) {
2035                 mlog_errno(status);
2036                 goto leave;
2037         }
2038
2039         /*
2040          * We're going to journal the change of i_flags and i_orphaned_slot.
2041          * It's safe anyway, though some callers may duplicate the journaling.
2042          * Journaling within the func just make the logic look more
2043          * straightforward.
2044          */
2045         status = ocfs2_journal_access_di(handle,
2046                                          INODE_CACHE(inode),
2047                                          fe_bh,
2048                                          OCFS2_JOURNAL_ACCESS_WRITE);
2049         if (status < 0) {
2050                 mlog_errno(status);
2051                 goto leave;
2052         }
2053
2054         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2055         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2056
2057         /* Record which orphan dir our inode now resides
2058          * in. delete_inode will use this to determine which orphan
2059          * dir to lock. */
2060         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2061
2062         ocfs2_journal_dirty(handle, fe_bh);
2063
2064         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2065                                    osb->slot_num);
2066
2067 leave:
2068         brelse(orphan_dir_bh);
2069
2070         if (status)
2071                 mlog_errno(status);
2072         return status;
2073 }
2074
2075 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2076 int ocfs2_orphan_del(struct ocfs2_super *osb,
2077                      handle_t *handle,
2078                      struct inode *orphan_dir_inode,
2079                      struct inode *inode,
2080                      struct buffer_head *orphan_dir_bh)
2081 {
2082         char name[OCFS2_ORPHAN_NAMELEN + 1];
2083         struct ocfs2_dinode *orphan_fe;
2084         int status = 0;
2085         struct ocfs2_dir_lookup_result lookup = { NULL, };
2086
2087         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2088         if (status < 0) {
2089                 mlog_errno(status);
2090                 goto leave;
2091         }
2092
2093         trace_ocfs2_orphan_del(
2094              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2095              name, OCFS2_ORPHAN_NAMELEN);
2096
2097         /* find it's spot in the orphan directory */
2098         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2099                                   &lookup);
2100         if (status) {
2101                 mlog_errno(status);
2102                 goto leave;
2103         }
2104
2105         /* remove it from the orphan directory */
2106         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2107         if (status < 0) {
2108                 mlog_errno(status);
2109                 goto leave;
2110         }
2111
2112         status = ocfs2_journal_access_di(handle,
2113                                          INODE_CACHE(orphan_dir_inode),
2114                                          orphan_dir_bh,
2115                                          OCFS2_JOURNAL_ACCESS_WRITE);
2116         if (status < 0) {
2117                 mlog_errno(status);
2118                 goto leave;
2119         }
2120
2121         /* do the i_nlink dance! :) */
2122         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2123         if (S_ISDIR(inode->i_mode))
2124                 ocfs2_add_links_count(orphan_fe, -1);
2125         orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2126         ocfs2_journal_dirty(handle, orphan_dir_bh);
2127
2128 leave:
2129         ocfs2_free_dir_lookup_result(&lookup);
2130
2131         if (status)
2132                 mlog_errno(status);
2133         return status;
2134 }
2135
2136 /**
2137  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2138  * allocated file. This is different from the typical 'add to orphan dir'
2139  * operation in that the inode does not yet exist. This is a problem because
2140  * the orphan dir stringifies the inode block number to come up with it's
2141  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2142  * problem. This function works around it by calling deeper into the orphan
2143  * and suballoc code than other callers. Use this only by necessity.
2144  * @dir: The directory which this inode will ultimately wind up under - not the
2145  * orphan dir!
2146  * @dir_bh: buffer_head the @dir inode block
2147  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2148  * with the string to be used for orphan dirent. Pass back to the orphan dir
2149  * code.
2150  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2151  * dir code.
2152  * @ret_di_blkno: block number where the new inode will be allocated.
2153  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2154  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2155  *
2156  * Returns zero on success and the ret_orphan_dir, name and lookup
2157  * fields will be populated.
2158  *
2159  * Returns non-zero on failure. 
2160  */
2161 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2162                                         struct buffer_head *dir_bh,
2163                                         char *orphan_name,
2164                                         struct inode **ret_orphan_dir,
2165                                         u64 *ret_di_blkno,
2166                                         struct ocfs2_dir_lookup_result *orphan_insert,
2167                                         struct ocfs2_alloc_context **ret_inode_ac)
2168 {
2169         int ret;
2170         u64 di_blkno;
2171         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2172         struct inode *orphan_dir = NULL;
2173         struct buffer_head *orphan_dir_bh = NULL;
2174         struct ocfs2_alloc_context *inode_ac = NULL;
2175
2176         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2177         if (ret < 0) {
2178                 mlog_errno(ret);
2179                 return ret;
2180         }
2181
2182         /* reserve an inode spot */
2183         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2184         if (ret < 0) {
2185                 if (ret != -ENOSPC)
2186                         mlog_errno(ret);
2187                 goto out;
2188         }
2189
2190         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2191                                        &di_blkno);
2192         if (ret) {
2193                 mlog_errno(ret);
2194                 goto out;
2195         }
2196
2197         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2198                                          di_blkno, orphan_name, orphan_insert);
2199         if (ret < 0) {
2200                 mlog_errno(ret);
2201                 goto out;
2202         }
2203
2204 out:
2205         if (ret == 0) {
2206                 *ret_orphan_dir = orphan_dir;
2207                 *ret_di_blkno = di_blkno;
2208                 *ret_inode_ac = inode_ac;
2209                 /*
2210                  * orphan_name and orphan_insert are already up to
2211                  * date via prepare_orphan_dir
2212                  */
2213         } else {
2214                 /* Unroll reserve_new_inode* */
2215                 if (inode_ac)
2216                         ocfs2_free_alloc_context(inode_ac);
2217
2218                 /* Unroll orphan dir locking */
2219                 mutex_unlock(&orphan_dir->i_mutex);
2220                 ocfs2_inode_unlock(orphan_dir, 1);
2221                 iput(orphan_dir);
2222         }
2223
2224         brelse(orphan_dir_bh);
2225
2226         return 0;
2227 }
2228
2229 int ocfs2_create_inode_in_orphan(struct inode *dir,
2230                                  int mode,
2231                                  struct inode **new_inode)
2232 {
2233         int status, did_quota_inode = 0;
2234         struct inode *inode = NULL;
2235         struct inode *orphan_dir = NULL;
2236         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2237         struct ocfs2_dinode *di = NULL;
2238         handle_t *handle = NULL;
2239         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2240         struct buffer_head *parent_di_bh = NULL;
2241         struct buffer_head *new_di_bh = NULL;
2242         struct ocfs2_alloc_context *inode_ac = NULL;
2243         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2244         u64 uninitialized_var(di_blkno), suballoc_loc;
2245         u16 suballoc_bit;
2246
2247         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2248         if (status < 0) {
2249                 if (status != -ENOENT)
2250                         mlog_errno(status);
2251                 return status;
2252         }
2253
2254         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2255                                               orphan_name, &orphan_dir,
2256                                               &di_blkno, &orphan_insert, &inode_ac);
2257         if (status < 0) {
2258                 if (status != -ENOSPC)
2259                         mlog_errno(status);
2260                 goto leave;
2261         }
2262
2263         inode = ocfs2_get_init_inode(dir, mode);
2264         if (!inode) {
2265                 status = -ENOMEM;
2266                 mlog_errno(status);
2267                 goto leave;
2268         }
2269
2270         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2271         if (IS_ERR(handle)) {
2272                 status = PTR_ERR(handle);
2273                 handle = NULL;
2274                 mlog_errno(status);
2275                 goto leave;
2276         }
2277
2278         status = dquot_alloc_inode(inode);
2279         if (status)
2280                 goto leave;
2281         did_quota_inode = 1;
2282
2283         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2284                                               &suballoc_loc,
2285                                               &suballoc_bit, di_blkno);
2286         if (status < 0) {
2287                 mlog_errno(status);
2288                 goto leave;
2289         }
2290
2291         inode->i_nlink = 0;
2292         /* do the real work now. */
2293         status = __ocfs2_mknod_locked(dir, inode,
2294                                       0, &new_di_bh, parent_di_bh, handle,
2295                                       inode_ac, di_blkno, suballoc_loc,
2296                                       suballoc_bit);
2297         if (status < 0) {
2298                 mlog_errno(status);
2299                 goto leave;
2300         }
2301
2302         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2303         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2304                                   &orphan_insert, orphan_dir);
2305         if (status < 0) {
2306                 mlog_errno(status);
2307                 goto leave;
2308         }
2309
2310         /* get open lock so that only nodes can't remove it from orphan dir. */
2311         status = ocfs2_open_lock(inode);
2312         if (status < 0)
2313                 mlog_errno(status);
2314
2315         insert_inode_hash(inode);
2316 leave:
2317         if (status < 0 && did_quota_inode)
2318                 dquot_free_inode(inode);
2319         if (handle)
2320                 ocfs2_commit_trans(osb, handle);
2321
2322         if (orphan_dir) {
2323                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2324                 ocfs2_inode_unlock(orphan_dir, 1);
2325                 mutex_unlock(&orphan_dir->i_mutex);
2326                 iput(orphan_dir);
2327         }
2328
2329         if ((status < 0) && inode) {
2330                 clear_nlink(inode);
2331                 iput(inode);
2332         }
2333
2334         if (inode_ac)
2335                 ocfs2_free_alloc_context(inode_ac);
2336
2337         brelse(new_di_bh);
2338
2339         if (!status)
2340                 *new_inode = inode;
2341
2342         ocfs2_free_dir_lookup_result(&orphan_insert);
2343
2344         ocfs2_inode_unlock(dir, 1);
2345         brelse(parent_di_bh);
2346         return status;
2347 }
2348
2349 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2350                                    struct inode *inode,
2351                                    struct dentry *dentry)
2352 {
2353         int status = 0;
2354         struct buffer_head *parent_di_bh = NULL;
2355         handle_t *handle = NULL;
2356         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2357         struct ocfs2_dinode *dir_di, *di;
2358         struct inode *orphan_dir_inode = NULL;
2359         struct buffer_head *orphan_dir_bh = NULL;
2360         struct buffer_head *di_bh = NULL;
2361         struct ocfs2_dir_lookup_result lookup = { NULL, };
2362
2363         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2364                                 dentry->d_name.len, dentry->d_name.name,
2365                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2366                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2367
2368         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2369         if (status < 0) {
2370                 if (status != -ENOENT)
2371                         mlog_errno(status);
2372                 return status;
2373         }
2374
2375         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2376         if (!dir_di->i_links_count) {
2377                 /* can't make a file in a deleted directory. */
2378                 status = -ENOENT;
2379                 goto leave;
2380         }
2381
2382         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2383                                            dentry->d_name.len);
2384         if (status)
2385                 goto leave;
2386
2387         /* get a spot inside the dir. */
2388         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2389                                               dentry->d_name.name,
2390                                               dentry->d_name.len, &lookup);
2391         if (status < 0) {
2392                 mlog_errno(status);
2393                 goto leave;
2394         }
2395
2396         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2397                                                        ORPHAN_DIR_SYSTEM_INODE,
2398                                                        osb->slot_num);
2399         if (!orphan_dir_inode) {
2400                 status = -EEXIST;
2401                 mlog_errno(status);
2402                 goto leave;
2403         }
2404
2405         mutex_lock(&orphan_dir_inode->i_mutex);
2406
2407         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2408         if (status < 0) {
2409                 mlog_errno(status);
2410                 mutex_unlock(&orphan_dir_inode->i_mutex);
2411                 iput(orphan_dir_inode);
2412                 goto leave;
2413         }
2414
2415         status = ocfs2_read_inode_block(inode, &di_bh);
2416         if (status < 0) {
2417                 mlog_errno(status);
2418                 goto orphan_unlock;
2419         }
2420
2421         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2422         if (IS_ERR(handle)) {
2423                 status = PTR_ERR(handle);
2424                 handle = NULL;
2425                 mlog_errno(status);
2426                 goto orphan_unlock;
2427         }
2428
2429         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2430                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2431         if (status < 0) {
2432                 mlog_errno(status);
2433                 goto out_commit;
2434         }
2435
2436         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2437                                   orphan_dir_bh);
2438         if (status < 0) {
2439                 mlog_errno(status);
2440                 goto out_commit;
2441         }
2442
2443         di = (struct ocfs2_dinode *)di_bh->b_data;
2444         le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2445         di->i_orphaned_slot = 0;
2446         inode->i_nlink = 1;
2447         ocfs2_set_links_count(di, inode->i_nlink);
2448         ocfs2_journal_dirty(handle, di_bh);
2449
2450         status = ocfs2_add_entry(handle, dentry, inode,
2451                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2452                                  &lookup);
2453         if (status < 0) {
2454                 mlog_errno(status);
2455                 goto out_commit;
2456         }
2457
2458         status = ocfs2_dentry_attach_lock(dentry, inode,
2459                                           OCFS2_I(dir)->ip_blkno);
2460         if (status) {
2461                 mlog_errno(status);
2462                 goto out_commit;
2463         }
2464
2465         d_instantiate(dentry, inode);
2466         status = 0;
2467 out_commit:
2468         ocfs2_commit_trans(osb, handle);
2469 orphan_unlock:
2470         ocfs2_inode_unlock(orphan_dir_inode, 1);
2471         mutex_unlock(&orphan_dir_inode->i_mutex);
2472         iput(orphan_dir_inode);
2473 leave:
2474
2475         ocfs2_inode_unlock(dir, 1);
2476
2477         brelse(di_bh);
2478         brelse(parent_di_bh);
2479         brelse(orphan_dir_bh);
2480
2481         ocfs2_free_dir_lookup_result(&lookup);
2482
2483         if (status)
2484                 mlog_errno(status);
2485
2486         return status;
2487 }
2488
2489 const struct inode_operations ocfs2_dir_iops = {
2490         .create         = ocfs2_create,
2491         .lookup         = ocfs2_lookup,
2492         .link           = ocfs2_link,
2493         .unlink         = ocfs2_unlink,
2494         .rmdir          = ocfs2_unlink,
2495         .symlink        = ocfs2_symlink,
2496         .mkdir          = ocfs2_mkdir,
2497         .mknod          = ocfs2_mknod,
2498         .rename         = ocfs2_rename,
2499         .setattr        = ocfs2_setattr,
2500         .getattr        = ocfs2_getattr,
2501         .permission     = ocfs2_permission,
2502         .setxattr       = generic_setxattr,
2503         .getxattr       = generic_getxattr,
2504         .listxattr      = ocfs2_listxattr,
2505         .removexattr    = generic_removexattr,
2506         .fiemap         = ocfs2_fiemap,
2507 };