d3a5a09d88ff2b9f66b0c797695df848f7ca3107
[linux-2.6.git] / fs / ocfs2 / namei.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * namei.c
5  *
6  * Create and rename file, directory, symlinks
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  *  Portions of this code from linux/fs/ext3/dir.c
11  *
12  *  Copyright (C) 1992, 1993, 1994, 1995
13  *  Remy Card (card@masi.ibp.fr)
14  *  Laboratoire MASI - Institut Blaise pascal
15  *  Universite Pierre et Marie Curie (Paris VI)
16  *
17  *   from
18  *
19  *   linux/fs/minix/dir.c
20  *
21  *   Copyright (C) 1991, 1992 Linux Torvalds
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public
25  * License as published by the Free Software Foundation; either
26  * version 2 of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
31  * General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public
34  * License along with this program; if not, write to the
35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36  * Boston, MA 021110-1307, USA.
37  */
38
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
44
45 #define MLOG_MASK_PREFIX ML_NAMEI
46 #include <cluster/masklog.h>
47
48 #include "ocfs2.h"
49
50 #include "alloc.h"
51 #include "dcache.h"
52 #include "dir.h"
53 #include "dlmglue.h"
54 #include "extent_map.h"
55 #include "file.h"
56 #include "inode.h"
57 #include "journal.h"
58 #include "namei.h"
59 #include "suballoc.h"
60 #include "super.h"
61 #include "symlink.h"
62 #include "sysfile.h"
63 #include "uptodate.h"
64 #include "xattr.h"
65 #include "acl.h"
66
67 #include "buffer_head_io.h"
68
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
70                               struct inode *dir,
71                               struct inode *inode,
72                               struct dentry *dentry,
73                               dev_t dev,
74                               struct buffer_head **new_fe_bh,
75                               struct buffer_head *parent_fe_bh,
76                               handle_t *handle,
77                               struct ocfs2_alloc_context *inode_ac);
78
79 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
80                                     struct inode **ret_orphan_dir,
81                                     struct inode *inode,
82                                     char *name,
83                                     struct ocfs2_dir_lookup_result *lookup);
84
85 static int ocfs2_orphan_add(struct ocfs2_super *osb,
86                             handle_t *handle,
87                             struct inode *inode,
88                             struct ocfs2_dinode *fe,
89                             char *name,
90                             struct ocfs2_dir_lookup_result *lookup,
91                             struct inode *orphan_dir_inode);
92
93 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
94                                      handle_t *handle,
95                                      struct inode *inode,
96                                      const char *symname);
97
98 /* An orphan dir name is an 8 byte value, printed as a hex string */
99 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
100
101 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
102                                    struct nameidata *nd)
103 {
104         int status;
105         u64 blkno;
106         struct inode *inode = NULL;
107         struct dentry *ret;
108         struct ocfs2_inode_info *oi;
109
110         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
111                    dentry->d_name.len, dentry->d_name.name);
112
113         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
114                 ret = ERR_PTR(-ENAMETOOLONG);
115                 goto bail;
116         }
117
118         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
119              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
120
121         status = ocfs2_inode_lock(dir, NULL, 0);
122         if (status < 0) {
123                 if (status != -ENOENT)
124                         mlog_errno(status);
125                 ret = ERR_PTR(status);
126                 goto bail;
127         }
128
129         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
130                                             dentry->d_name.len, &blkno);
131         if (status < 0)
132                 goto bail_add;
133
134         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
135         if (IS_ERR(inode)) {
136                 ret = ERR_PTR(-EACCES);
137                 goto bail_unlock;
138         }
139
140         oi = OCFS2_I(inode);
141         /* Clear any orphaned state... If we were able to look up the
142          * inode from a directory, it certainly can't be orphaned. We
143          * might have the bad state from a node which intended to
144          * orphan this inode but crashed before it could commit the
145          * unlink. */
146         spin_lock(&oi->ip_lock);
147         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
148         spin_unlock(&oi->ip_lock);
149
150 bail_add:
151         dentry->d_op = &ocfs2_dentry_ops;
152         ret = d_splice_alias(inode, dentry);
153
154         if (inode) {
155                 /*
156                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
157                  * dentry, it will d_move() it on top of ourse. The
158                  * return value will indicate this however, so in
159                  * those cases, we switch them around for the locking
160                  * code.
161                  *
162                  * NOTE: This dentry already has ->d_op set from
163                  * ocfs2_get_parent() and ocfs2_get_dentry()
164                  */
165                 if (ret)
166                         dentry = ret;
167
168                 status = ocfs2_dentry_attach_lock(dentry, inode,
169                                                   OCFS2_I(dir)->ip_blkno);
170                 if (status) {
171                         mlog_errno(status);
172                         ret = ERR_PTR(status);
173                         goto bail_unlock;
174                 }
175         }
176
177 bail_unlock:
178         /* Don't drop the cluster lock until *after* the d_add --
179          * unlink on another node will message us to remove that
180          * dentry under this lock so otherwise we can race this with
181          * the downconvert thread and have a stale dentry. */
182         ocfs2_inode_unlock(dir, 0);
183
184 bail:
185
186         mlog_exit_ptr(ret);
187
188         return ret;
189 }
190
191 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
192 {
193         struct inode *inode;
194
195         inode = new_inode(dir->i_sb);
196         if (!inode) {
197                 mlog(ML_ERROR, "new_inode failed!\n");
198                 return NULL;
199         }
200
201         /* populate as many fields early on as possible - many of
202          * these are used by the support functions here and in
203          * callers. */
204         if (S_ISDIR(mode))
205                 inode->i_nlink = 2;
206         else
207                 inode->i_nlink = 1;
208         inode->i_uid = current_fsuid();
209         if (dir->i_mode & S_ISGID) {
210                 inode->i_gid = dir->i_gid;
211                 if (S_ISDIR(mode))
212                         mode |= S_ISGID;
213         } else
214                 inode->i_gid = current_fsgid();
215         inode->i_mode = mode;
216         vfs_dq_init(inode);
217         return inode;
218 }
219
220 static int ocfs2_mknod(struct inode *dir,
221                        struct dentry *dentry,
222                        int mode,
223                        dev_t dev)
224 {
225         int status = 0;
226         struct buffer_head *parent_fe_bh = NULL;
227         handle_t *handle = NULL;
228         struct ocfs2_super *osb;
229         struct ocfs2_dinode *dirfe;
230         struct buffer_head *new_fe_bh = NULL;
231         struct inode *inode = NULL;
232         struct ocfs2_alloc_context *inode_ac = NULL;
233         struct ocfs2_alloc_context *data_ac = NULL;
234         struct ocfs2_alloc_context *xattr_ac = NULL;
235         int want_clusters = 0;
236         int xattr_credits = 0;
237         struct ocfs2_security_xattr_info si = {
238                 .enable = 1,
239         };
240         int did_quota_inode = 0;
241         struct ocfs2_dir_lookup_result lookup = { NULL, };
242
243         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
244                    (unsigned long)dev, dentry->d_name.len,
245                    dentry->d_name.name);
246
247         /* get our super block */
248         osb = OCFS2_SB(dir->i_sb);
249
250         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
251         if (status < 0) {
252                 if (status != -ENOENT)
253                         mlog_errno(status);
254                 return status;
255         }
256
257         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
258                 status = -EMLINK;
259                 goto leave;
260         }
261
262         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
263         if (!dirfe->i_links_count) {
264                 /* can't make a file in a deleted directory. */
265                 status = -ENOENT;
266                 goto leave;
267         }
268
269         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
270                                            dentry->d_name.len);
271         if (status)
272                 goto leave;
273
274         /* get a spot inside the dir. */
275         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
276                                               dentry->d_name.name,
277                                               dentry->d_name.len, &lookup);
278         if (status < 0) {
279                 mlog_errno(status);
280                 goto leave;
281         }
282
283         /* reserve an inode spot */
284         status = ocfs2_reserve_new_inode(osb, &inode_ac);
285         if (status < 0) {
286                 if (status != -ENOSPC)
287                         mlog_errno(status);
288                 goto leave;
289         }
290
291         inode = ocfs2_get_init_inode(dir, mode);
292         if (!inode) {
293                 status = -ENOMEM;
294                 mlog_errno(status);
295                 goto leave;
296         }
297
298         /* get security xattr */
299         status = ocfs2_init_security_get(inode, dir, &si);
300         if (status) {
301                 if (status == -EOPNOTSUPP)
302                         si.enable = 0;
303                 else {
304                         mlog_errno(status);
305                         goto leave;
306                 }
307         }
308
309         /* calculate meta data/clusters for setting security and acl xattr */
310         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
311                                         &si, &want_clusters,
312                                         &xattr_credits, &xattr_ac);
313         if (status < 0) {
314                 mlog_errno(status);
315                 goto leave;
316         }
317
318         /* Reserve a cluster if creating an extent based directory. */
319         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb))
320                 want_clusters += 1;
321
322         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
323         if (status < 0) {
324                 if (status != -ENOSPC)
325                         mlog_errno(status);
326                 goto leave;
327         }
328
329         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb) +
330                                    xattr_credits);
331         if (IS_ERR(handle)) {
332                 status = PTR_ERR(handle);
333                 handle = NULL;
334                 mlog_errno(status);
335                 goto leave;
336         }
337
338         /* We don't use standard VFS wrapper because we don't want vfs_dq_init
339          * to be called. */
340         if (sb_any_quota_active(osb->sb) &&
341             osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
342                 status = -EDQUOT;
343                 goto leave;
344         }
345         did_quota_inode = 1;
346
347         /* do the real work now. */
348         status = ocfs2_mknod_locked(osb, dir, inode, dentry, dev,
349                                     &new_fe_bh, parent_fe_bh, handle,
350                                     inode_ac);
351         if (status < 0) {
352                 mlog_errno(status);
353                 goto leave;
354         }
355
356         if (S_ISDIR(mode)) {
357                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
358                                             new_fe_bh, data_ac);
359                 if (status < 0) {
360                         mlog_errno(status);
361                         goto leave;
362                 }
363
364                 status = ocfs2_journal_access_di(handle, dir, parent_fe_bh,
365                                                  OCFS2_JOURNAL_ACCESS_WRITE);
366                 if (status < 0) {
367                         mlog_errno(status);
368                         goto leave;
369                 }
370                 le16_add_cpu(&dirfe->i_links_count, 1);
371                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
372                 if (status < 0) {
373                         mlog_errno(status);
374                         goto leave;
375                 }
376                 inc_nlink(dir);
377         }
378
379         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
380                                 xattr_ac, data_ac);
381         if (status < 0) {
382                 mlog_errno(status);
383                 goto leave;
384         }
385
386         if (si.enable) {
387                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
388                                                  xattr_ac, data_ac);
389                 if (status < 0) {
390                         mlog_errno(status);
391                         goto leave;
392                 }
393         }
394
395         status = ocfs2_add_entry(handle, dentry, inode,
396                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
397                                  &lookup);
398         if (status < 0) {
399                 mlog_errno(status);
400                 goto leave;
401         }
402
403         status = ocfs2_dentry_attach_lock(dentry, inode,
404                                           OCFS2_I(dir)->ip_blkno);
405         if (status) {
406                 mlog_errno(status);
407                 goto leave;
408         }
409
410         insert_inode_hash(inode);
411         dentry->d_op = &ocfs2_dentry_ops;
412         d_instantiate(dentry, inode);
413         status = 0;
414 leave:
415         if (status < 0 && did_quota_inode)
416                 vfs_dq_free_inode(inode);
417         if (handle)
418                 ocfs2_commit_trans(osb, handle);
419
420         ocfs2_inode_unlock(dir, 1);
421
422         if (status == -ENOSPC)
423                 mlog(0, "Disk is full\n");
424
425         brelse(new_fe_bh);
426         brelse(parent_fe_bh);
427         kfree(si.name);
428         kfree(si.value);
429
430         ocfs2_free_dir_lookup_result(&lookup);
431
432         if ((status < 0) && inode) {
433                 clear_nlink(inode);
434                 iput(inode);
435         }
436
437         if (inode_ac)
438                 ocfs2_free_alloc_context(inode_ac);
439
440         if (data_ac)
441                 ocfs2_free_alloc_context(data_ac);
442
443         if (xattr_ac)
444                 ocfs2_free_alloc_context(xattr_ac);
445
446         mlog_exit(status);
447
448         return status;
449 }
450
451 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
452                               struct inode *dir,
453                               struct inode *inode,
454                               struct dentry *dentry,
455                               dev_t dev,
456                               struct buffer_head **new_fe_bh,
457                               struct buffer_head *parent_fe_bh,
458                               handle_t *handle,
459                               struct ocfs2_alloc_context *inode_ac)
460 {
461         int status = 0;
462         struct ocfs2_dinode *fe = NULL;
463         struct ocfs2_extent_list *fel;
464         u64 fe_blkno = 0;
465         u16 suballoc_bit;
466
467         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
468                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
469                    dentry->d_name.name);
470
471         *new_fe_bh = NULL;
472
473         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
474                                        &fe_blkno);
475         if (status < 0) {
476                 mlog_errno(status);
477                 goto leave;
478         }
479
480         /* populate as many fields early on as possible - many of
481          * these are used by the support functions here and in
482          * callers. */
483         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
484         OCFS2_I(inode)->ip_blkno = fe_blkno;
485         spin_lock(&osb->osb_lock);
486         inode->i_generation = osb->s_next_generation++;
487         spin_unlock(&osb->osb_lock);
488
489         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
490         if (!*new_fe_bh) {
491                 status = -EIO;
492                 mlog_errno(status);
493                 goto leave;
494         }
495         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
496
497         status = ocfs2_journal_access_di(handle, inode, *new_fe_bh,
498                                          OCFS2_JOURNAL_ACCESS_CREATE);
499         if (status < 0) {
500                 mlog_errno(status);
501                 goto leave;
502         }
503
504         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
505         memset(fe, 0, osb->sb->s_blocksize);
506
507         fe->i_generation = cpu_to_le32(inode->i_generation);
508         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
509         fe->i_blkno = cpu_to_le64(fe_blkno);
510         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
511         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
512         fe->i_uid = cpu_to_le32(inode->i_uid);
513         fe->i_gid = cpu_to_le32(inode->i_gid);
514         fe->i_mode = cpu_to_le16(inode->i_mode);
515         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
516                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
517         fe->i_links_count = cpu_to_le16(inode->i_nlink);
518
519         fe->i_last_eb_blk = 0;
520         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
521         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
522         fe->i_atime = fe->i_ctime = fe->i_mtime =
523                 cpu_to_le64(CURRENT_TIME.tv_sec);
524         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
525                 cpu_to_le32(CURRENT_TIME.tv_nsec);
526         fe->i_dtime = 0;
527
528         /*
529          * If supported, directories start with inline data.
530          */
531         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
532                 u16 feat = le16_to_cpu(fe->i_dyn_features);
533
534                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
535
536                 fe->id2.i_data.id_count = cpu_to_le16(
537                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
538         } else {
539                 fel = &fe->id2.i_list;
540                 fel->l_tree_depth = 0;
541                 fel->l_next_free_rec = 0;
542                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
543         }
544
545         status = ocfs2_journal_dirty(handle, *new_fe_bh);
546         if (status < 0) {
547                 mlog_errno(status);
548                 goto leave;
549         }
550
551         ocfs2_populate_inode(inode, fe, 1);
552         ocfs2_inode_set_new(osb, inode);
553         if (!ocfs2_mount_local(osb)) {
554                 status = ocfs2_create_new_inode_locks(inode);
555                 if (status < 0)
556                         mlog_errno(status);
557         }
558
559         status = 0; /* error in ocfs2_create_new_inode_locks is not
560                      * critical */
561
562 leave:
563         if (status < 0) {
564                 if (*new_fe_bh) {
565                         brelse(*new_fe_bh);
566                         *new_fe_bh = NULL;
567                 }
568         }
569
570         mlog_exit(status);
571         return status;
572 }
573
574 static int ocfs2_mkdir(struct inode *dir,
575                        struct dentry *dentry,
576                        int mode)
577 {
578         int ret;
579
580         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
581                    dentry->d_name.len, dentry->d_name.name);
582         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
583         mlog_exit(ret);
584
585         return ret;
586 }
587
588 static int ocfs2_create(struct inode *dir,
589                         struct dentry *dentry,
590                         int mode,
591                         struct nameidata *nd)
592 {
593         int ret;
594
595         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
596                    dentry->d_name.len, dentry->d_name.name);
597         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
598         mlog_exit(ret);
599
600         return ret;
601 }
602
603 static int ocfs2_link(struct dentry *old_dentry,
604                       struct inode *dir,
605                       struct dentry *dentry)
606 {
607         handle_t *handle;
608         struct inode *inode = old_dentry->d_inode;
609         int err;
610         struct buffer_head *fe_bh = NULL;
611         struct buffer_head *parent_fe_bh = NULL;
612         struct ocfs2_dinode *fe = NULL;
613         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
614         struct ocfs2_dir_lookup_result lookup = { NULL, };
615
616         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
617                    old_dentry->d_name.len, old_dentry->d_name.name,
618                    dentry->d_name.len, dentry->d_name.name);
619
620         if (S_ISDIR(inode->i_mode))
621                 return -EPERM;
622
623         err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
624         if (err < 0) {
625                 if (err != -ENOENT)
626                         mlog_errno(err);
627                 return err;
628         }
629
630         if (!dir->i_nlink) {
631                 err = -ENOENT;
632                 goto out;
633         }
634
635         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
636                                         dentry->d_name.len);
637         if (err)
638                 goto out;
639
640         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
641                                            dentry->d_name.name,
642                                            dentry->d_name.len, &lookup);
643         if (err < 0) {
644                 mlog_errno(err);
645                 goto out;
646         }
647
648         err = ocfs2_inode_lock(inode, &fe_bh, 1);
649         if (err < 0) {
650                 if (err != -ENOENT)
651                         mlog_errno(err);
652                 goto out;
653         }
654
655         fe = (struct ocfs2_dinode *) fe_bh->b_data;
656         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
657                 err = -EMLINK;
658                 goto out_unlock_inode;
659         }
660
661         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
662         if (IS_ERR(handle)) {
663                 err = PTR_ERR(handle);
664                 handle = NULL;
665                 mlog_errno(err);
666                 goto out_unlock_inode;
667         }
668
669         err = ocfs2_journal_access_di(handle, inode, fe_bh,
670                                       OCFS2_JOURNAL_ACCESS_WRITE);
671         if (err < 0) {
672                 mlog_errno(err);
673                 goto out_commit;
674         }
675
676         inc_nlink(inode);
677         inode->i_ctime = CURRENT_TIME;
678         fe->i_links_count = cpu_to_le16(inode->i_nlink);
679         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
680         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
681
682         err = ocfs2_journal_dirty(handle, fe_bh);
683         if (err < 0) {
684                 le16_add_cpu(&fe->i_links_count, -1);
685                 drop_nlink(inode);
686                 mlog_errno(err);
687                 goto out_commit;
688         }
689
690         err = ocfs2_add_entry(handle, dentry, inode,
691                               OCFS2_I(inode)->ip_blkno,
692                               parent_fe_bh, &lookup);
693         if (err) {
694                 le16_add_cpu(&fe->i_links_count, -1);
695                 drop_nlink(inode);
696                 mlog_errno(err);
697                 goto out_commit;
698         }
699
700         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
701         if (err) {
702                 mlog_errno(err);
703                 goto out_commit;
704         }
705
706         atomic_inc(&inode->i_count);
707         dentry->d_op = &ocfs2_dentry_ops;
708         d_instantiate(dentry, inode);
709
710 out_commit:
711         ocfs2_commit_trans(osb, handle);
712 out_unlock_inode:
713         ocfs2_inode_unlock(inode, 1);
714
715 out:
716         ocfs2_inode_unlock(dir, 1);
717
718         brelse(fe_bh);
719         brelse(parent_fe_bh);
720
721         ocfs2_free_dir_lookup_result(&lookup);
722
723         mlog_exit(err);
724
725         return err;
726 }
727
728 /*
729  * Takes and drops an exclusive lock on the given dentry. This will
730  * force other nodes to drop it.
731  */
732 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
733 {
734         int ret;
735
736         ret = ocfs2_dentry_lock(dentry, 1);
737         if (ret)
738                 mlog_errno(ret);
739         else
740                 ocfs2_dentry_unlock(dentry, 1);
741
742         return ret;
743 }
744
745 static inline int inode_is_unlinkable(struct inode *inode)
746 {
747         if (S_ISDIR(inode->i_mode)) {
748                 if (inode->i_nlink == 2)
749                         return 1;
750                 return 0;
751         }
752
753         if (inode->i_nlink == 1)
754                 return 1;
755         return 0;
756 }
757
758 static int ocfs2_unlink(struct inode *dir,
759                         struct dentry *dentry)
760 {
761         int status;
762         int child_locked = 0;
763         struct inode *inode = dentry->d_inode;
764         struct inode *orphan_dir = NULL;
765         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
766         u64 blkno;
767         struct ocfs2_dinode *fe = NULL;
768         struct buffer_head *fe_bh = NULL;
769         struct buffer_head *parent_node_bh = NULL;
770         handle_t *handle = NULL;
771         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
772         struct ocfs2_dir_lookup_result lookup = { NULL, };
773         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
774
775         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
776                    dentry->d_name.len, dentry->d_name.name);
777
778         BUG_ON(dentry->d_parent->d_inode != dir);
779
780         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
781
782         if (inode == osb->root_inode) {
783                 mlog(0, "Cannot delete the root directory\n");
784                 return -EPERM;
785         }
786
787         status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
788         if (status < 0) {
789                 if (status != -ENOENT)
790                         mlog_errno(status);
791                 return status;
792         }
793
794         status = ocfs2_find_files_on_disk(dentry->d_name.name,
795                                           dentry->d_name.len, &blkno, dir,
796                                           &lookup);
797         if (status < 0) {
798                 if (status != -ENOENT)
799                         mlog_errno(status);
800                 goto leave;
801         }
802
803         if (OCFS2_I(inode)->ip_blkno != blkno) {
804                 status = -ENOENT;
805
806                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
807                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
808                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
809                 goto leave;
810         }
811
812         status = ocfs2_inode_lock(inode, &fe_bh, 1);
813         if (status < 0) {
814                 if (status != -ENOENT)
815                         mlog_errno(status);
816                 goto leave;
817         }
818         child_locked = 1;
819
820         if (S_ISDIR(inode->i_mode)) {
821                 if (!ocfs2_empty_dir(inode)) {
822                         status = -ENOTEMPTY;
823                         goto leave;
824                 } else if (inode->i_nlink != 2) {
825                         status = -ENOTEMPTY;
826                         goto leave;
827                 }
828         }
829
830         status = ocfs2_remote_dentry_delete(dentry);
831         if (status < 0) {
832                 /* This remote delete should succeed under all normal
833                  * circumstances. */
834                 mlog_errno(status);
835                 goto leave;
836         }
837
838         if (inode_is_unlinkable(inode)) {
839                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
840                                                   orphan_name, &orphan_insert);
841                 if (status < 0) {
842                         mlog_errno(status);
843                         goto leave;
844                 }
845         }
846
847         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
848         if (IS_ERR(handle)) {
849                 status = PTR_ERR(handle);
850                 handle = NULL;
851                 mlog_errno(status);
852                 goto leave;
853         }
854
855         status = ocfs2_journal_access_di(handle, inode, fe_bh,
856                                          OCFS2_JOURNAL_ACCESS_WRITE);
857         if (status < 0) {
858                 mlog_errno(status);
859                 goto leave;
860         }
861
862         fe = (struct ocfs2_dinode *) fe_bh->b_data;
863
864         if (inode_is_unlinkable(inode)) {
865                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
866                                           &orphan_insert, orphan_dir);
867                 if (status < 0) {
868                         mlog_errno(status);
869                         goto leave;
870                 }
871         }
872
873         /* delete the name from the parent dir */
874         status = ocfs2_delete_entry(handle, dir, &lookup);
875         if (status < 0) {
876                 mlog_errno(status);
877                 goto leave;
878         }
879
880         if (S_ISDIR(inode->i_mode))
881                 drop_nlink(inode);
882         drop_nlink(inode);
883         fe->i_links_count = cpu_to_le16(inode->i_nlink);
884
885         status = ocfs2_journal_dirty(handle, fe_bh);
886         if (status < 0) {
887                 mlog_errno(status);
888                 goto leave;
889         }
890
891         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
892         if (S_ISDIR(inode->i_mode))
893                 drop_nlink(dir);
894
895         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
896         if (status < 0) {
897                 mlog_errno(status);
898                 if (S_ISDIR(inode->i_mode))
899                         inc_nlink(dir);
900         }
901
902 leave:
903         if (handle)
904                 ocfs2_commit_trans(osb, handle);
905
906         if (child_locked)
907                 ocfs2_inode_unlock(inode, 1);
908
909         ocfs2_inode_unlock(dir, 1);
910
911         if (orphan_dir) {
912                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
913                 ocfs2_inode_unlock(orphan_dir, 1);
914                 mutex_unlock(&orphan_dir->i_mutex);
915                 iput(orphan_dir);
916         }
917
918         brelse(fe_bh);
919         brelse(parent_node_bh);
920
921         ocfs2_free_dir_lookup_result(&orphan_insert);
922         ocfs2_free_dir_lookup_result(&lookup);
923
924         mlog_exit(status);
925
926         return status;
927 }
928
929 /*
930  * The only place this should be used is rename!
931  * if they have the same id, then the 1st one is the only one locked.
932  */
933 static int ocfs2_double_lock(struct ocfs2_super *osb,
934                              struct buffer_head **bh1,
935                              struct inode *inode1,
936                              struct buffer_head **bh2,
937                              struct inode *inode2)
938 {
939         int status;
940         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
941         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
942         struct buffer_head **tmpbh;
943         struct inode *tmpinode;
944
945         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
946                    (unsigned long long)oi1->ip_blkno,
947                    (unsigned long long)oi2->ip_blkno);
948
949         if (*bh1)
950                 *bh1 = NULL;
951         if (*bh2)
952                 *bh2 = NULL;
953
954         /* we always want to lock the one with the lower lockid first. */
955         if (oi1->ip_blkno != oi2->ip_blkno) {
956                 if (oi1->ip_blkno < oi2->ip_blkno) {
957                         /* switch id1 and id2 around */
958                         mlog(0, "switching them around...\n");
959                         tmpbh = bh2;
960                         bh2 = bh1;
961                         bh1 = tmpbh;
962
963                         tmpinode = inode2;
964                         inode2 = inode1;
965                         inode1 = tmpinode;
966                 }
967                 /* lock id2 */
968                 status = ocfs2_inode_lock(inode2, bh2, 1);
969                 if (status < 0) {
970                         if (status != -ENOENT)
971                                 mlog_errno(status);
972                         goto bail;
973                 }
974         }
975
976         /* lock id1 */
977         status = ocfs2_inode_lock(inode1, bh1, 1);
978         if (status < 0) {
979                 /*
980                  * An error return must mean that no cluster locks
981                  * were held on function exit.
982                  */
983                 if (oi1->ip_blkno != oi2->ip_blkno)
984                         ocfs2_inode_unlock(inode2, 1);
985
986                 if (status != -ENOENT)
987                         mlog_errno(status);
988         }
989
990 bail:
991         mlog_exit(status);
992         return status;
993 }
994
995 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
996 {
997         ocfs2_inode_unlock(inode1, 1);
998
999         if (inode1 != inode2)
1000                 ocfs2_inode_unlock(inode2, 1);
1001 }
1002
1003 static int ocfs2_rename(struct inode *old_dir,
1004                         struct dentry *old_dentry,
1005                         struct inode *new_dir,
1006                         struct dentry *new_dentry)
1007 {
1008         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1009         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1010         struct inode *old_inode = old_dentry->d_inode;
1011         struct inode *new_inode = new_dentry->d_inode;
1012         struct inode *orphan_dir = NULL;
1013         struct ocfs2_dinode *newfe = NULL;
1014         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1015         struct buffer_head *orphan_entry_bh = NULL;
1016         struct buffer_head *newfe_bh = NULL;
1017         struct buffer_head *old_inode_bh = NULL;
1018         struct buffer_head *insert_entry_bh = NULL;
1019         struct ocfs2_super *osb = NULL;
1020         u64 newfe_blkno, old_de_ino;
1021         handle_t *handle = NULL;
1022         struct buffer_head *old_dir_bh = NULL;
1023         struct buffer_head *new_dir_bh = NULL;
1024         nlink_t old_dir_nlink = old_dir->i_nlink;
1025         struct ocfs2_dinode *old_di;
1026         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1027         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1028         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1029         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1030         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1031
1032         /* At some point it might be nice to break this function up a
1033          * bit. */
1034
1035         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1036                    old_dir, old_dentry, new_dir, new_dentry,
1037                    old_dentry->d_name.len, old_dentry->d_name.name,
1038                    new_dentry->d_name.len, new_dentry->d_name.name);
1039
1040         osb = OCFS2_SB(old_dir->i_sb);
1041
1042         if (new_inode) {
1043                 if (!igrab(new_inode))
1044                         BUG();
1045         }
1046
1047         /* Assume a directory hierarchy thusly:
1048          * a/b/c
1049          * a/d
1050          * a,b,c, and d are all directories.
1051          *
1052          * from cwd of 'a' on both nodes:
1053          * node1: mv b/c d
1054          * node2: mv d   b/c
1055          *
1056          * And that's why, just like the VFS, we need a file system
1057          * rename lock. */
1058         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1059                 status = ocfs2_rename_lock(osb);
1060                 if (status < 0) {
1061                         mlog_errno(status);
1062                         goto bail;
1063                 }
1064                 rename_lock = 1;
1065         }
1066
1067         /* if old and new are the same, this'll just do one lock. */
1068         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1069                                    &new_dir_bh, new_dir);
1070         if (status < 0) {
1071                 mlog_errno(status);
1072                 goto bail;
1073         }
1074         parents_locked = 1;
1075
1076         /* make sure both dirs have bhs
1077          * get an extra ref on old_dir_bh if old==new */
1078         if (!new_dir_bh) {
1079                 if (old_dir_bh) {
1080                         new_dir_bh = old_dir_bh;
1081                         get_bh(new_dir_bh);
1082                 } else {
1083                         mlog(ML_ERROR, "no old_dir_bh!\n");
1084                         status = -EIO;
1085                         goto bail;
1086                 }
1087         }
1088
1089         /*
1090          * Aside from allowing a meta data update, the locking here
1091          * also ensures that the downconvert thread on other nodes
1092          * won't have to concurrently downconvert the inode and the
1093          * dentry locks.
1094          */
1095         status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1096         if (status < 0) {
1097                 if (status != -ENOENT)
1098                         mlog_errno(status);
1099                 goto bail;
1100         }
1101         old_child_locked = 1;
1102
1103         status = ocfs2_remote_dentry_delete(old_dentry);
1104         if (status < 0) {
1105                 mlog_errno(status);
1106                 goto bail;
1107         }
1108
1109         if (S_ISDIR(old_inode->i_mode)) {
1110                 u64 old_inode_parent;
1111
1112                 update_dot_dot = 1;
1113                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1114                                                   old_inode,
1115                                                   &old_inode_dot_dot_res);
1116                 if (status) {
1117                         status = -EIO;
1118                         goto bail;
1119                 }
1120
1121                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1122                         status = -EIO;
1123                         goto bail;
1124                 }
1125
1126                 if (!new_inode && new_dir != old_dir &&
1127                     new_dir->i_nlink >= OCFS2_LINK_MAX) {
1128                         status = -EMLINK;
1129                         goto bail;
1130                 }
1131         }
1132
1133         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1134                                             old_dentry->d_name.len,
1135                                             &old_de_ino);
1136         if (status) {
1137                 status = -ENOENT;
1138                 goto bail;
1139         }
1140
1141         /*
1142          *  Check for inode number is _not_ due to possible IO errors.
1143          *  We might rmdir the source, keep it as pwd of some process
1144          *  and merrily kill the link to whatever was created under the
1145          *  same name. Goodbye sticky bit ;-<
1146          */
1147         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1148                 status = -ENOENT;
1149                 goto bail;
1150         }
1151
1152         /* check if the target already exists (in which case we need
1153          * to delete it */
1154         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1155                                           new_dentry->d_name.len,
1156                                           &newfe_blkno, new_dir,
1157                                           &target_lookup_res);
1158         /* The only error we allow here is -ENOENT because the new
1159          * file not existing is perfectly valid. */
1160         if ((status < 0) && (status != -ENOENT)) {
1161                 /* If we cannot find the file specified we should just */
1162                 /* return the error... */
1163                 mlog_errno(status);
1164                 goto bail;
1165         }
1166         if (status == 0)
1167                 target_exists = 1;
1168
1169         if (!target_exists && new_inode) {
1170                 /*
1171                  * Target was unlinked by another node while we were
1172                  * waiting to get to ocfs2_rename(). There isn't
1173                  * anything we can do here to help the situation, so
1174                  * bubble up the appropriate error.
1175                  */
1176                 status = -ENOENT;
1177                 goto bail;
1178         }
1179
1180         /* In case we need to overwrite an existing file, we blow it
1181          * away first */
1182         if (target_exists) {
1183                 /* VFS didn't think there existed an inode here, but
1184                  * someone else in the cluster must have raced our
1185                  * rename to create one. Today we error cleanly, in
1186                  * the future we should consider calling iget to build
1187                  * a new struct inode for this entry. */
1188                 if (!new_inode) {
1189                         status = -EACCES;
1190
1191                         mlog(0, "We found an inode for name %.*s but VFS "
1192                              "didn't give us one.\n", new_dentry->d_name.len,
1193                              new_dentry->d_name.name);
1194                         goto bail;
1195                 }
1196
1197                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1198                         status = -EACCES;
1199
1200                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1201                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1202                              (unsigned long long)newfe_blkno,
1203                              OCFS2_I(new_inode)->ip_flags);
1204                         goto bail;
1205                 }
1206
1207                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1208                 if (status < 0) {
1209                         if (status != -ENOENT)
1210                                 mlog_errno(status);
1211                         goto bail;
1212                 }
1213                 new_child_locked = 1;
1214
1215                 status = ocfs2_remote_dentry_delete(new_dentry);
1216                 if (status < 0) {
1217                         mlog_errno(status);
1218                         goto bail;
1219                 }
1220
1221                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1222
1223                 mlog(0, "aha rename over existing... new_blkno=%llu "
1224                      "newfebh=%p bhblocknr=%llu\n",
1225                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1226                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1227
1228                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1229                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1230                                                           new_inode,
1231                                                           orphan_name,
1232                                                           &orphan_insert);
1233                         if (status < 0) {
1234                                 mlog_errno(status);
1235                                 goto bail;
1236                         }
1237                 }
1238         } else {
1239                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1240
1241                 status = ocfs2_check_dir_for_entry(new_dir,
1242                                                    new_dentry->d_name.name,
1243                                                    new_dentry->d_name.len);
1244                 if (status)
1245                         goto bail;
1246
1247                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1248                                                       new_dentry->d_name.name,
1249                                                       new_dentry->d_name.len,
1250                                                       &target_insert);
1251                 if (status < 0) {
1252                         mlog_errno(status);
1253                         goto bail;
1254                 }
1255         }
1256
1257         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1258         if (IS_ERR(handle)) {
1259                 status = PTR_ERR(handle);
1260                 handle = NULL;
1261                 mlog_errno(status);
1262                 goto bail;
1263         }
1264
1265         if (target_exists) {
1266                 if (S_ISDIR(new_inode->i_mode)) {
1267                         if (!ocfs2_empty_dir(new_inode) ||
1268                             new_inode->i_nlink != 2) {
1269                                 status = -ENOTEMPTY;
1270                                 goto bail;
1271                         }
1272                 }
1273                 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1274                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1275                 if (status < 0) {
1276                         mlog_errno(status);
1277                         goto bail;
1278                 }
1279
1280                 if (S_ISDIR(new_inode->i_mode) ||
1281                     (newfe->i_links_count == cpu_to_le16(1))){
1282                         status = ocfs2_orphan_add(osb, handle, new_inode,
1283                                                   newfe, orphan_name,
1284                                                   &orphan_insert, orphan_dir);
1285                         if (status < 0) {
1286                                 mlog_errno(status);
1287                                 goto bail;
1288                         }
1289                 }
1290
1291                 /* change the dirent to point to the correct inode */
1292                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1293                                             old_inode);
1294                 if (status < 0) {
1295                         mlog_errno(status);
1296                         goto bail;
1297                 }
1298                 new_dir->i_version++;
1299
1300                 if (S_ISDIR(new_inode->i_mode))
1301                         newfe->i_links_count = 0;
1302                 else
1303                         le16_add_cpu(&newfe->i_links_count, -1);
1304
1305                 status = ocfs2_journal_dirty(handle, newfe_bh);
1306                 if (status < 0) {
1307                         mlog_errno(status);
1308                         goto bail;
1309                 }
1310         } else {
1311                 /* if the name was not found in new_dir, add it now */
1312                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1313                                          OCFS2_I(old_inode)->ip_blkno,
1314                                          new_dir_bh, &target_insert);
1315         }
1316
1317         old_inode->i_ctime = CURRENT_TIME;
1318         mark_inode_dirty(old_inode);
1319
1320         status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1321                                          OCFS2_JOURNAL_ACCESS_WRITE);
1322         if (status >= 0) {
1323                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1324
1325                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1326                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1327
1328                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1329                 if (status < 0)
1330                         mlog_errno(status);
1331         } else
1332                 mlog_errno(status);
1333
1334         /*
1335          * Now that the name has been added to new_dir, remove the old name.
1336          *
1337          * We don't keep any directory entry context around until now
1338          * because the insert might have changed the type of directory
1339          * we're dealing with.
1340          */
1341         status = ocfs2_find_entry(old_dentry->d_name.name,
1342                                   old_dentry->d_name.len, old_dir,
1343                                   &old_entry_lookup);
1344         if (status)
1345                 goto bail;
1346
1347         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1348         if (status < 0) {
1349                 mlog_errno(status);
1350                 goto bail;
1351         }
1352
1353         if (new_inode) {
1354                 new_inode->i_nlink--;
1355                 new_inode->i_ctime = CURRENT_TIME;
1356         }
1357         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1358
1359         if (update_dot_dot) {
1360                 status = ocfs2_update_entry(old_inode, handle,
1361                                             &old_inode_dot_dot_res, new_dir);
1362                 old_dir->i_nlink--;
1363                 if (new_inode) {
1364                         new_inode->i_nlink--;
1365                 } else {
1366                         inc_nlink(new_dir);
1367                         mark_inode_dirty(new_dir);
1368                 }
1369         }
1370         mark_inode_dirty(old_dir);
1371         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1372         if (new_inode) {
1373                 mark_inode_dirty(new_inode);
1374                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1375         }
1376
1377         if (old_dir != new_dir) {
1378                 /* Keep the same times on both directories.*/
1379                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1380
1381                 /*
1382                  * This will also pick up the i_nlink change from the
1383                  * block above.
1384                  */
1385                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1386         }
1387
1388         if (old_dir_nlink != old_dir->i_nlink) {
1389                 if (!old_dir_bh) {
1390                         mlog(ML_ERROR, "need to change nlink for old dir "
1391                              "%llu from %d to %d but bh is NULL!\n",
1392                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1393                              (int)old_dir_nlink, old_dir->i_nlink);
1394                 } else {
1395                         struct ocfs2_dinode *fe;
1396                         status = ocfs2_journal_access_di(handle, old_dir,
1397                                                          old_dir_bh,
1398                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1399                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1400                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1401                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1402                 }
1403         }
1404
1405         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1406         status = 0;
1407 bail:
1408         if (rename_lock)
1409                 ocfs2_rename_unlock(osb);
1410
1411         if (handle)
1412                 ocfs2_commit_trans(osb, handle);
1413
1414         if (parents_locked)
1415                 ocfs2_double_unlock(old_dir, new_dir);
1416
1417         if (old_child_locked)
1418                 ocfs2_inode_unlock(old_inode, 1);
1419
1420         if (new_child_locked)
1421                 ocfs2_inode_unlock(new_inode, 1);
1422
1423         if (orphan_dir) {
1424                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1425                 ocfs2_inode_unlock(orphan_dir, 1);
1426                 mutex_unlock(&orphan_dir->i_mutex);
1427                 iput(orphan_dir);
1428         }
1429
1430         if (new_inode)
1431                 sync_mapping_buffers(old_inode->i_mapping);
1432
1433         if (new_inode)
1434                 iput(new_inode);
1435
1436         ocfs2_free_dir_lookup_result(&target_lookup_res);
1437         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1438         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1439         ocfs2_free_dir_lookup_result(&orphan_insert);
1440         ocfs2_free_dir_lookup_result(&target_insert);
1441
1442         brelse(newfe_bh);
1443         brelse(old_inode_bh);
1444         brelse(old_dir_bh);
1445         brelse(new_dir_bh);
1446         brelse(orphan_entry_bh);
1447         brelse(insert_entry_bh);
1448
1449         mlog_exit(status);
1450
1451         return status;
1452 }
1453
1454 /*
1455  * we expect i_size = strlen(symname). Copy symname into the file
1456  * data, including the null terminator.
1457  */
1458 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1459                                      handle_t *handle,
1460                                      struct inode *inode,
1461                                      const char *symname)
1462 {
1463         struct buffer_head **bhs = NULL;
1464         const char *c;
1465         struct super_block *sb = osb->sb;
1466         u64 p_blkno, p_blocks;
1467         int virtual, blocks, status, i, bytes_left;
1468
1469         bytes_left = i_size_read(inode) + 1;
1470         /* we can't trust i_blocks because we're actually going to
1471          * write i_size + 1 bytes. */
1472         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1473
1474         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1475                         (unsigned long long)inode->i_blocks,
1476                         i_size_read(inode), blocks);
1477
1478         /* Sanity check -- make sure we're going to fit. */
1479         if (bytes_left >
1480             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1481                 status = -EIO;
1482                 mlog_errno(status);
1483                 goto bail;
1484         }
1485
1486         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1487         if (!bhs) {
1488                 status = -ENOMEM;
1489                 mlog_errno(status);
1490                 goto bail;
1491         }
1492
1493         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1494                                              NULL);
1495         if (status < 0) {
1496                 mlog_errno(status);
1497                 goto bail;
1498         }
1499
1500         /* links can never be larger than one cluster so we know this
1501          * is all going to be contiguous, but do a sanity check
1502          * anyway. */
1503         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1504                 status = -EIO;
1505                 mlog_errno(status);
1506                 goto bail;
1507         }
1508
1509         virtual = 0;
1510         while(bytes_left > 0) {
1511                 c = &symname[virtual * sb->s_blocksize];
1512
1513                 bhs[virtual] = sb_getblk(sb, p_blkno);
1514                 if (!bhs[virtual]) {
1515                         status = -ENOMEM;
1516                         mlog_errno(status);
1517                         goto bail;
1518                 }
1519                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1520
1521                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1522                                               OCFS2_JOURNAL_ACCESS_CREATE);
1523                 if (status < 0) {
1524                         mlog_errno(status);
1525                         goto bail;
1526                 }
1527
1528                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1529
1530                 memcpy(bhs[virtual]->b_data, c,
1531                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1532                        bytes_left);
1533
1534                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1535                 if (status < 0) {
1536                         mlog_errno(status);
1537                         goto bail;
1538                 }
1539
1540                 virtual++;
1541                 p_blkno++;
1542                 bytes_left -= sb->s_blocksize;
1543         }
1544
1545         status = 0;
1546 bail:
1547
1548         if (bhs) {
1549                 for(i = 0; i < blocks; i++)
1550                         brelse(bhs[i]);
1551                 kfree(bhs);
1552         }
1553
1554         mlog_exit(status);
1555         return status;
1556 }
1557
1558 static int ocfs2_symlink(struct inode *dir,
1559                          struct dentry *dentry,
1560                          const char *symname)
1561 {
1562         int status, l, credits;
1563         u64 newsize;
1564         struct ocfs2_super *osb = NULL;
1565         struct inode *inode = NULL;
1566         struct super_block *sb;
1567         struct buffer_head *new_fe_bh = NULL;
1568         struct buffer_head *parent_fe_bh = NULL;
1569         struct ocfs2_dinode *fe = NULL;
1570         struct ocfs2_dinode *dirfe;
1571         handle_t *handle = NULL;
1572         struct ocfs2_alloc_context *inode_ac = NULL;
1573         struct ocfs2_alloc_context *data_ac = NULL;
1574         struct ocfs2_alloc_context *xattr_ac = NULL;
1575         int want_clusters = 0;
1576         int xattr_credits = 0;
1577         struct ocfs2_security_xattr_info si = {
1578                 .enable = 1,
1579         };
1580         int did_quota = 0, did_quota_inode = 0;
1581         struct ocfs2_dir_lookup_result lookup = { NULL, };
1582
1583         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1584                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1585
1586         sb = dir->i_sb;
1587         osb = OCFS2_SB(sb);
1588
1589         l = strlen(symname) + 1;
1590
1591         credits = ocfs2_calc_symlink_credits(sb);
1592
1593         /* lock the parent directory */
1594         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1595         if (status < 0) {
1596                 if (status != -ENOENT)
1597                         mlog_errno(status);
1598                 return status;
1599         }
1600
1601         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1602         if (!dirfe->i_links_count) {
1603                 /* can't make a file in a deleted directory. */
1604                 status = -ENOENT;
1605                 goto bail;
1606         }
1607
1608         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1609                                            dentry->d_name.len);
1610         if (status)
1611                 goto bail;
1612
1613         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1614                                               dentry->d_name.name,
1615                                               dentry->d_name.len, &lookup);
1616         if (status < 0) {
1617                 mlog_errno(status);
1618                 goto bail;
1619         }
1620
1621         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1622         if (status < 0) {
1623                 if (status != -ENOSPC)
1624                         mlog_errno(status);
1625                 goto bail;
1626         }
1627
1628         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1629         if (!inode) {
1630                 status = -ENOMEM;
1631                 mlog_errno(status);
1632                 goto bail;
1633         }
1634
1635         /* get security xattr */
1636         status = ocfs2_init_security_get(inode, dir, &si);
1637         if (status) {
1638                 if (status == -EOPNOTSUPP)
1639                         si.enable = 0;
1640                 else {
1641                         mlog_errno(status);
1642                         goto bail;
1643                 }
1644         }
1645
1646         /* calculate meta data/clusters for setting security xattr */
1647         if (si.enable) {
1648                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1649                                                   &xattr_credits, &xattr_ac);
1650                 if (status < 0) {
1651                         mlog_errno(status);
1652                         goto bail;
1653                 }
1654         }
1655
1656         /* don't reserve bitmap space for fast symlinks. */
1657         if (l > ocfs2_fast_symlink_chars(sb))
1658                 want_clusters += 1;
1659
1660         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1661         if (status < 0) {
1662                 if (status != -ENOSPC)
1663                         mlog_errno(status);
1664                 goto bail;
1665         }
1666
1667         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1668         if (IS_ERR(handle)) {
1669                 status = PTR_ERR(handle);
1670                 handle = NULL;
1671                 mlog_errno(status);
1672                 goto bail;
1673         }
1674
1675         /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1676          * to be called. */
1677         if (sb_any_quota_active(osb->sb) &&
1678             osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1679                 status = -EDQUOT;
1680                 goto bail;
1681         }
1682         did_quota_inode = 1;
1683
1684         status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1685                                     0, &new_fe_bh, parent_fe_bh, handle,
1686                                     inode_ac);
1687         if (status < 0) {
1688                 mlog_errno(status);
1689                 goto bail;
1690         }
1691
1692         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1693         inode->i_rdev = 0;
1694         newsize = l - 1;
1695         if (l > ocfs2_fast_symlink_chars(sb)) {
1696                 u32 offset = 0;
1697
1698                 inode->i_op = &ocfs2_symlink_inode_operations;
1699                 if (vfs_dq_alloc_space_nodirty(inode,
1700                     ocfs2_clusters_to_bytes(osb->sb, 1))) {
1701                         status = -EDQUOT;
1702                         goto bail;
1703                 }
1704                 did_quota = 1;
1705                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1706                                               new_fe_bh,
1707                                               handle, data_ac, NULL,
1708                                               NULL);
1709                 if (status < 0) {
1710                         if (status != -ENOSPC && status != -EINTR) {
1711                                 mlog(ML_ERROR,
1712                                      "Failed to extend file to %llu\n",
1713                                      (unsigned long long)newsize);
1714                                 mlog_errno(status);
1715                                 status = -ENOSPC;
1716                         }
1717                         goto bail;
1718                 }
1719                 i_size_write(inode, newsize);
1720                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1721         } else {
1722                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1723                 memcpy((char *) fe->id2.i_symlink, symname, l);
1724                 i_size_write(inode, newsize);
1725                 inode->i_blocks = 0;
1726         }
1727
1728         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1729         if (status < 0) {
1730                 mlog_errno(status);
1731                 goto bail;
1732         }
1733
1734         if (!ocfs2_inode_is_fast_symlink(inode)) {
1735                 status = ocfs2_create_symlink_data(osb, handle, inode,
1736                                                    symname);
1737                 if (status < 0) {
1738                         mlog_errno(status);
1739                         goto bail;
1740                 }
1741         }
1742
1743         if (si.enable) {
1744                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1745                                                  xattr_ac, data_ac);
1746                 if (status < 0) {
1747                         mlog_errno(status);
1748                         goto bail;
1749                 }
1750         }
1751
1752         status = ocfs2_add_entry(handle, dentry, inode,
1753                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1754                                  &lookup);
1755         if (status < 0) {
1756                 mlog_errno(status);
1757                 goto bail;
1758         }
1759
1760         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1761         if (status) {
1762                 mlog_errno(status);
1763                 goto bail;
1764         }
1765
1766         insert_inode_hash(inode);
1767         dentry->d_op = &ocfs2_dentry_ops;
1768         d_instantiate(dentry, inode);
1769 bail:
1770         if (status < 0 && did_quota)
1771                 vfs_dq_free_space_nodirty(inode,
1772                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1773         if (status < 0 && did_quota_inode)
1774                 vfs_dq_free_inode(inode);
1775         if (handle)
1776                 ocfs2_commit_trans(osb, handle);
1777
1778         ocfs2_inode_unlock(dir, 1);
1779
1780         brelse(new_fe_bh);
1781         brelse(parent_fe_bh);
1782         kfree(si.name);
1783         kfree(si.value);
1784         ocfs2_free_dir_lookup_result(&lookup);
1785         if (inode_ac)
1786                 ocfs2_free_alloc_context(inode_ac);
1787         if (data_ac)
1788                 ocfs2_free_alloc_context(data_ac);
1789         if (xattr_ac)
1790                 ocfs2_free_alloc_context(xattr_ac);
1791         if ((status < 0) && inode) {
1792                 clear_nlink(inode);
1793                 iput(inode);
1794         }
1795
1796         mlog_exit(status);
1797
1798         return status;
1799 }
1800
1801 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1802 {
1803         int status, namelen;
1804
1805         mlog_entry_void();
1806
1807         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1808                            (long long)blkno);
1809         if (namelen <= 0) {
1810                 if (namelen)
1811                         status = namelen;
1812                 else
1813                         status = -EINVAL;
1814                 mlog_errno(status);
1815                 goto bail;
1816         }
1817         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1818                 status = -EINVAL;
1819                 mlog_errno(status);
1820                 goto bail;
1821         }
1822
1823         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1824              namelen);
1825
1826         status = 0;
1827 bail:
1828         mlog_exit(status);
1829         return status;
1830 }
1831
1832 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1833                                     struct inode **ret_orphan_dir,
1834                                     struct inode *inode,
1835                                     char *name,
1836                                     struct ocfs2_dir_lookup_result *lookup)
1837 {
1838         struct inode *orphan_dir_inode;
1839         struct buffer_head *orphan_dir_bh = NULL;
1840         int status = 0;
1841
1842         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1843         if (status < 0) {
1844                 mlog_errno(status);
1845                 return status;
1846         }
1847
1848         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1849                                                        ORPHAN_DIR_SYSTEM_INODE,
1850                                                        osb->slot_num);
1851         if (!orphan_dir_inode) {
1852                 status = -ENOENT;
1853                 mlog_errno(status);
1854                 return status;
1855         }
1856
1857         mutex_lock(&orphan_dir_inode->i_mutex);
1858
1859         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1860         if (status < 0) {
1861                 mlog_errno(status);
1862                 goto leave;
1863         }
1864
1865         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1866                                               orphan_dir_bh, name,
1867                                               OCFS2_ORPHAN_NAMELEN, lookup);
1868         if (status < 0) {
1869                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1870
1871                 mlog_errno(status);
1872                 goto leave;
1873         }
1874
1875         *ret_orphan_dir = orphan_dir_inode;
1876
1877 leave:
1878         if (status) {
1879                 mutex_unlock(&orphan_dir_inode->i_mutex);
1880                 iput(orphan_dir_inode);
1881         }
1882
1883         brelse(orphan_dir_bh);
1884
1885         mlog_exit(status);
1886         return status;
1887 }
1888
1889 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1890                             handle_t *handle,
1891                             struct inode *inode,
1892                             struct ocfs2_dinode *fe,
1893                             char *name,
1894                             struct ocfs2_dir_lookup_result *lookup,
1895                             struct inode *orphan_dir_inode)
1896 {
1897         struct buffer_head *orphan_dir_bh = NULL;
1898         int status = 0;
1899         struct ocfs2_dinode *orphan_fe;
1900
1901         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1902
1903         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1904         if (status < 0) {
1905                 mlog_errno(status);
1906                 goto leave;
1907         }
1908
1909         status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1910                                          OCFS2_JOURNAL_ACCESS_WRITE);
1911         if (status < 0) {
1912                 mlog_errno(status);
1913                 goto leave;
1914         }
1915
1916         /* we're a cluster, and nlink can change on disk from
1917          * underneath us... */
1918         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1919         if (S_ISDIR(inode->i_mode))
1920                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1921         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1922
1923         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1924         if (status < 0) {
1925                 mlog_errno(status);
1926                 goto leave;
1927         }
1928
1929         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1930                                    OCFS2_ORPHAN_NAMELEN, inode,
1931                                    OCFS2_I(inode)->ip_blkno,
1932                                    orphan_dir_bh, lookup);
1933         if (status < 0) {
1934                 mlog_errno(status);
1935                 goto leave;
1936         }
1937
1938         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1939
1940         /* Record which orphan dir our inode now resides
1941          * in. delete_inode will use this to determine which orphan
1942          * dir to lock. */
1943         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1944
1945         mlog(0, "Inode %llu orphaned in slot %d\n",
1946              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1947
1948 leave:
1949         brelse(orphan_dir_bh);
1950
1951         mlog_exit(status);
1952         return status;
1953 }
1954
1955 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1956 int ocfs2_orphan_del(struct ocfs2_super *osb,
1957                      handle_t *handle,
1958                      struct inode *orphan_dir_inode,
1959                      struct inode *inode,
1960                      struct buffer_head *orphan_dir_bh)
1961 {
1962         char name[OCFS2_ORPHAN_NAMELEN + 1];
1963         struct ocfs2_dinode *orphan_fe;
1964         int status = 0;
1965         struct ocfs2_dir_lookup_result lookup = { NULL, };
1966
1967         mlog_entry_void();
1968
1969         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1970         if (status < 0) {
1971                 mlog_errno(status);
1972                 goto leave;
1973         }
1974
1975         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1976              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1977              OCFS2_ORPHAN_NAMELEN);
1978
1979         /* find it's spot in the orphan directory */
1980         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
1981                                   &lookup);
1982         if (status) {
1983                 mlog_errno(status);
1984                 goto leave;
1985         }
1986
1987         /* remove it from the orphan directory */
1988         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
1989         if (status < 0) {
1990                 mlog_errno(status);
1991                 goto leave;
1992         }
1993
1994         status = ocfs2_journal_access_di(handle,orphan_dir_inode,  orphan_dir_bh,
1995                                          OCFS2_JOURNAL_ACCESS_WRITE);
1996         if (status < 0) {
1997                 mlog_errno(status);
1998                 goto leave;
1999         }
2000
2001         /* do the i_nlink dance! :) */
2002         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2003         if (S_ISDIR(inode->i_mode))
2004                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2005         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2006
2007         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2008         if (status < 0) {
2009                 mlog_errno(status);
2010                 goto leave;
2011         }
2012
2013 leave:
2014         ocfs2_free_dir_lookup_result(&lookup);
2015
2016         mlog_exit(status);
2017         return status;
2018 }
2019
2020 const struct inode_operations ocfs2_dir_iops = {
2021         .create         = ocfs2_create,
2022         .lookup         = ocfs2_lookup,
2023         .link           = ocfs2_link,
2024         .unlink         = ocfs2_unlink,
2025         .rmdir          = ocfs2_unlink,
2026         .symlink        = ocfs2_symlink,
2027         .mkdir          = ocfs2_mkdir,
2028         .mknod          = ocfs2_mknod,
2029         .rename         = ocfs2_rename,
2030         .setattr        = ocfs2_setattr,
2031         .getattr        = ocfs2_getattr,
2032         .permission     = ocfs2_permission,
2033         .setxattr       = generic_setxattr,
2034         .getxattr       = generic_getxattr,
2035         .listxattr      = ocfs2_listxattr,
2036         .removexattr    = generic_removexattr,
2037 };