ocfs2: add ocfs2_init_acl in mknod
[linux-2.6] / 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
44 #define MLOG_MASK_PREFIX ML_NAMEI
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
66 #include "buffer_head_io.h"
67
68 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
69                               struct inode *dir,
70                               struct inode *inode,
71                               struct dentry *dentry,
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                                     struct inode *inode,
81                                     char *name,
82                                     struct buffer_head **de_bh);
83
84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
85                             handle_t *handle,
86                             struct inode *inode,
87                             struct ocfs2_dinode *fe,
88                             char *name,
89                             struct buffer_head *de_bh,
90                             struct inode *orphan_dir_inode);
91
92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
93                                      handle_t *handle,
94                                      struct inode *inode,
95                                      const char *symname);
96
97 /* An orphan dir name is an 8 byte value, printed as a hex string */
98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
99
100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
101                                    struct nameidata *nd)
102 {
103         int status;
104         u64 blkno;
105         struct inode *inode = NULL;
106         struct dentry *ret;
107         struct ocfs2_inode_info *oi;
108
109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
110                    dentry->d_name.len, dentry->d_name.name);
111
112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
113                 ret = ERR_PTR(-ENAMETOOLONG);
114                 goto bail;
115         }
116
117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
119
120         status = ocfs2_inode_lock(dir, NULL, 0);
121         if (status < 0) {
122                 if (status != -ENOENT)
123                         mlog_errno(status);
124                 ret = ERR_PTR(status);
125                 goto bail;
126         }
127
128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
129                                             dentry->d_name.len, &blkno);
130         if (status < 0)
131                 goto bail_add;
132
133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
134         if (IS_ERR(inode)) {
135                 ret = ERR_PTR(-EACCES);
136                 goto bail_unlock;
137         }
138
139         oi = OCFS2_I(inode);
140         /* Clear any orphaned state... If we were able to look up the
141          * inode from a directory, it certainly can't be orphaned. We
142          * might have the bad state from a node which intended to
143          * orphan this inode but crashed before it could commit the
144          * unlink. */
145         spin_lock(&oi->ip_lock);
146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
147         spin_unlock(&oi->ip_lock);
148
149 bail_add:
150         dentry->d_op = &ocfs2_dentry_ops;
151         ret = d_splice_alias(inode, dentry);
152
153         if (inode) {
154                 /*
155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
156                  * dentry, it will d_move() it on top of ourse. The
157                  * return value will indicate this however, so in
158                  * those cases, we switch them around for the locking
159                  * code.
160                  *
161                  * NOTE: This dentry already has ->d_op set from
162                  * ocfs2_get_parent() and ocfs2_get_dentry()
163                  */
164                 if (ret)
165                         dentry = ret;
166
167                 status = ocfs2_dentry_attach_lock(dentry, inode,
168                                                   OCFS2_I(dir)->ip_blkno);
169                 if (status) {
170                         mlog_errno(status);
171                         ret = ERR_PTR(status);
172                         goto bail_unlock;
173                 }
174         }
175
176 bail_unlock:
177         /* Don't drop the cluster lock until *after* the d_add --
178          * unlink on another node will message us to remove that
179          * dentry under this lock so otherwise we can race this with
180          * the downconvert thread and have a stale dentry. */
181         ocfs2_inode_unlock(dir, 0);
182
183 bail:
184
185         mlog_exit_ptr(ret);
186
187         return ret;
188 }
189
190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
191 {
192         struct inode *inode;
193
194         inode = new_inode(dir->i_sb);
195         if (!inode) {
196                 mlog(ML_ERROR, "new_inode failed!\n");
197                 return NULL;
198         }
199
200         /* populate as many fields early on as possible - many of
201          * these are used by the support functions here and in
202          * callers. */
203         if (S_ISDIR(mode))
204                 inode->i_nlink = 2;
205         else
206                 inode->i_nlink = 1;
207         inode->i_uid = current_fsuid();
208         if (dir->i_mode & S_ISGID) {
209                 inode->i_gid = dir->i_gid;
210                 if (S_ISDIR(mode))
211                         mode |= S_ISGID;
212         } else
213                 inode->i_gid = current_fsgid();
214         inode->i_mode = mode;
215         return inode;
216 }
217
218 static int ocfs2_mknod(struct inode *dir,
219                        struct dentry *dentry,
220                        int mode,
221                        dev_t dev)
222 {
223         int status = 0;
224         struct buffer_head *parent_fe_bh = NULL;
225         handle_t *handle = NULL;
226         struct ocfs2_super *osb;
227         struct ocfs2_dinode *dirfe;
228         struct buffer_head *new_fe_bh = NULL;
229         struct buffer_head *de_bh = NULL;
230         struct inode *inode = NULL;
231         struct ocfs2_alloc_context *inode_ac = NULL;
232         struct ocfs2_alloc_context *data_ac = NULL;
233         struct ocfs2_alloc_context *xattr_ac = NULL;
234         int want_clusters = 0;
235         int xattr_credits = 0;
236         struct ocfs2_security_xattr_info si = {
237                 .enable = 1,
238         };
239
240         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
241                    (unsigned long)dev, dentry->d_name.len,
242                    dentry->d_name.name);
243
244         /* get our super block */
245         osb = OCFS2_SB(dir->i_sb);
246
247         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
248         if (status < 0) {
249                 if (status != -ENOENT)
250                         mlog_errno(status);
251                 return status;
252         }
253
254         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
255                 status = -EMLINK;
256                 goto leave;
257         }
258
259         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
260         if (!dirfe->i_links_count) {
261                 /* can't make a file in a deleted directory. */
262                 status = -ENOENT;
263                 goto leave;
264         }
265
266         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
267                                            dentry->d_name.len);
268         if (status)
269                 goto leave;
270
271         /* get a spot inside the dir. */
272         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
273                                               dentry->d_name.name,
274                                               dentry->d_name.len, &de_bh);
275         if (status < 0) {
276                 mlog_errno(status);
277                 goto leave;
278         }
279
280         /* reserve an inode spot */
281         status = ocfs2_reserve_new_inode(osb, &inode_ac);
282         if (status < 0) {
283                 if (status != -ENOSPC)
284                         mlog_errno(status);
285                 goto leave;
286         }
287
288         inode = ocfs2_get_init_inode(dir, mode);
289         if (!inode) {
290                 status = -ENOMEM;
291                 mlog_errno(status);
292                 goto leave;
293         }
294
295         /* get security xattr */
296         status = ocfs2_init_security_get(inode, dir, &si);
297         if (status) {
298                 if (status == -EOPNOTSUPP)
299                         si.enable = 0;
300                 else {
301                         mlog_errno(status);
302                         goto leave;
303                 }
304         }
305
306         /* calculate meta data/clusters for setting security and acl xattr */
307         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
308                                         &si, &want_clusters,
309                                         &xattr_credits, &xattr_ac);
310         if (status < 0) {
311                 mlog_errno(status);
312                 goto leave;
313         }
314
315         /* Reserve a cluster if creating an extent based directory. */
316         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb))
317                 want_clusters += 1;
318
319         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
320         if (status < 0) {
321                 if (status != -ENOSPC)
322                         mlog_errno(status);
323                 goto leave;
324         }
325
326         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS + xattr_credits);
327         if (IS_ERR(handle)) {
328                 status = PTR_ERR(handle);
329                 handle = NULL;
330                 mlog_errno(status);
331                 goto leave;
332         }
333
334         /* do the real work now. */
335         status = ocfs2_mknod_locked(osb, dir, inode, dentry, dev,
336                                     &new_fe_bh, parent_fe_bh, handle,
337                                     inode_ac);
338         if (status < 0) {
339                 mlog_errno(status);
340                 goto leave;
341         }
342
343         if (S_ISDIR(mode)) {
344                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
345                                             new_fe_bh, data_ac);
346                 if (status < 0) {
347                         mlog_errno(status);
348                         goto leave;
349                 }
350
351                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
352                                               OCFS2_JOURNAL_ACCESS_WRITE);
353                 if (status < 0) {
354                         mlog_errno(status);
355                         goto leave;
356                 }
357                 le16_add_cpu(&dirfe->i_links_count, 1);
358                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
359                 if (status < 0) {
360                         mlog_errno(status);
361                         goto leave;
362                 }
363                 inc_nlink(dir);
364         }
365
366         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
367                                 xattr_ac, data_ac);
368         if (status < 0) {
369                 mlog_errno(status);
370                 goto leave;
371         }
372
373         if (si.enable) {
374                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
375                                                  xattr_ac, data_ac);
376                 if (status < 0) {
377                         mlog_errno(status);
378                         goto leave;
379                 }
380         }
381
382         status = ocfs2_add_entry(handle, dentry, inode,
383                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
384                                  de_bh);
385         if (status < 0) {
386                 mlog_errno(status);
387                 goto leave;
388         }
389
390         status = ocfs2_dentry_attach_lock(dentry, inode,
391                                           OCFS2_I(dir)->ip_blkno);
392         if (status) {
393                 mlog_errno(status);
394                 goto leave;
395         }
396
397         insert_inode_hash(inode);
398         dentry->d_op = &ocfs2_dentry_ops;
399         d_instantiate(dentry, inode);
400         status = 0;
401 leave:
402         if (handle)
403                 ocfs2_commit_trans(osb, handle);
404
405         ocfs2_inode_unlock(dir, 1);
406
407         if (status == -ENOSPC)
408                 mlog(0, "Disk is full\n");
409
410         brelse(new_fe_bh);
411         brelse(de_bh);
412         brelse(parent_fe_bh);
413         kfree(si.name);
414         kfree(si.value);
415
416         if ((status < 0) && inode) {
417                 clear_nlink(inode);
418                 iput(inode);
419         }
420
421         if (inode_ac)
422                 ocfs2_free_alloc_context(inode_ac);
423
424         if (data_ac)
425                 ocfs2_free_alloc_context(data_ac);
426
427         if (xattr_ac)
428                 ocfs2_free_alloc_context(xattr_ac);
429
430         mlog_exit(status);
431
432         return status;
433 }
434
435 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
436                               struct inode *dir,
437                               struct inode *inode,
438                               struct dentry *dentry,
439                               dev_t dev,
440                               struct buffer_head **new_fe_bh,
441                               struct buffer_head *parent_fe_bh,
442                               handle_t *handle,
443                               struct ocfs2_alloc_context *inode_ac)
444 {
445         int status = 0;
446         struct ocfs2_dinode *fe = NULL;
447         struct ocfs2_extent_list *fel;
448         u64 fe_blkno = 0;
449         u16 suballoc_bit;
450
451         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
452                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
453                    dentry->d_name.name);
454
455         *new_fe_bh = NULL;
456
457         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
458                                        &fe_blkno);
459         if (status < 0) {
460                 mlog_errno(status);
461                 goto leave;
462         }
463
464         /* populate as many fields early on as possible - many of
465          * these are used by the support functions here and in
466          * callers. */
467         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
468         OCFS2_I(inode)->ip_blkno = fe_blkno;
469         spin_lock(&osb->osb_lock);
470         inode->i_generation = osb->s_next_generation++;
471         spin_unlock(&osb->osb_lock);
472
473         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
474         if (!*new_fe_bh) {
475                 status = -EIO;
476                 mlog_errno(status);
477                 goto leave;
478         }
479         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
480
481         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
482                                       OCFS2_JOURNAL_ACCESS_CREATE);
483         if (status < 0) {
484                 mlog_errno(status);
485                 goto leave;
486         }
487
488         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
489         memset(fe, 0, osb->sb->s_blocksize);
490
491         fe->i_generation = cpu_to_le32(inode->i_generation);
492         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
493         fe->i_blkno = cpu_to_le64(fe_blkno);
494         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
495         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
496         fe->i_uid = cpu_to_le32(inode->i_uid);
497         fe->i_gid = cpu_to_le32(inode->i_gid);
498         fe->i_mode = cpu_to_le16(inode->i_mode);
499         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
500                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
501         fe->i_links_count = cpu_to_le16(inode->i_nlink);
502
503         fe->i_last_eb_blk = 0;
504         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
505         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
506         fe->i_atime = fe->i_ctime = fe->i_mtime =
507                 cpu_to_le64(CURRENT_TIME.tv_sec);
508         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
509                 cpu_to_le32(CURRENT_TIME.tv_nsec);
510         fe->i_dtime = 0;
511
512         /*
513          * If supported, directories start with inline data.
514          */
515         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
516                 u16 feat = le16_to_cpu(fe->i_dyn_features);
517
518                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
519
520                 fe->id2.i_data.id_count = cpu_to_le16(ocfs2_max_inline_data(osb->sb));
521         } else {
522                 fel = &fe->id2.i_list;
523                 fel->l_tree_depth = 0;
524                 fel->l_next_free_rec = 0;
525                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
526         }
527
528         status = ocfs2_journal_dirty(handle, *new_fe_bh);
529         if (status < 0) {
530                 mlog_errno(status);
531                 goto leave;
532         }
533
534         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
535                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
536                      "i_blkno=%llu, i_ino=%lu\n",
537                      (unsigned long long)(*new_fe_bh)->b_blocknr,
538                      (unsigned long long)le64_to_cpu(fe->i_blkno),
539                      inode->i_ino);
540                 BUG();
541         }
542
543         ocfs2_inode_set_new(osb, inode);
544         if (!ocfs2_mount_local(osb)) {
545                 status = ocfs2_create_new_inode_locks(inode);
546                 if (status < 0)
547                         mlog_errno(status);
548         }
549
550         status = 0; /* error in ocfs2_create_new_inode_locks is not
551                      * critical */
552
553 leave:
554         if (status < 0) {
555                 if (*new_fe_bh) {
556                         brelse(*new_fe_bh);
557                         *new_fe_bh = NULL;
558                 }
559         }
560
561         mlog_exit(status);
562         return status;
563 }
564
565 static int ocfs2_mkdir(struct inode *dir,
566                        struct dentry *dentry,
567                        int mode)
568 {
569         int ret;
570
571         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
572                    dentry->d_name.len, dentry->d_name.name);
573         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
574         mlog_exit(ret);
575
576         return ret;
577 }
578
579 static int ocfs2_create(struct inode *dir,
580                         struct dentry *dentry,
581                         int mode,
582                         struct nameidata *nd)
583 {
584         int ret;
585
586         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
587                    dentry->d_name.len, dentry->d_name.name);
588         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
589         mlog_exit(ret);
590
591         return ret;
592 }
593
594 static int ocfs2_link(struct dentry *old_dentry,
595                       struct inode *dir,
596                       struct dentry *dentry)
597 {
598         handle_t *handle;
599         struct inode *inode = old_dentry->d_inode;
600         int err;
601         struct buffer_head *fe_bh = NULL;
602         struct buffer_head *parent_fe_bh = NULL;
603         struct buffer_head *de_bh = NULL;
604         struct ocfs2_dinode *fe = NULL;
605         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
606
607         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
608                    old_dentry->d_name.len, old_dentry->d_name.name,
609                    dentry->d_name.len, dentry->d_name.name);
610
611         if (S_ISDIR(inode->i_mode))
612                 return -EPERM;
613
614         err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
615         if (err < 0) {
616                 if (err != -ENOENT)
617                         mlog_errno(err);
618                 return err;
619         }
620
621         if (!dir->i_nlink) {
622                 err = -ENOENT;
623                 goto out;
624         }
625
626         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
627                                         dentry->d_name.len);
628         if (err)
629                 goto out;
630
631         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
632                                            dentry->d_name.name,
633                                            dentry->d_name.len, &de_bh);
634         if (err < 0) {
635                 mlog_errno(err);
636                 goto out;
637         }
638
639         err = ocfs2_inode_lock(inode, &fe_bh, 1);
640         if (err < 0) {
641                 if (err != -ENOENT)
642                         mlog_errno(err);
643                 goto out;
644         }
645
646         fe = (struct ocfs2_dinode *) fe_bh->b_data;
647         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
648                 err = -EMLINK;
649                 goto out_unlock_inode;
650         }
651
652         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
653         if (IS_ERR(handle)) {
654                 err = PTR_ERR(handle);
655                 handle = NULL;
656                 mlog_errno(err);
657                 goto out_unlock_inode;
658         }
659
660         err = ocfs2_journal_access(handle, inode, fe_bh,
661                                    OCFS2_JOURNAL_ACCESS_WRITE);
662         if (err < 0) {
663                 mlog_errno(err);
664                 goto out_commit;
665         }
666
667         inc_nlink(inode);
668         inode->i_ctime = CURRENT_TIME;
669         fe->i_links_count = cpu_to_le16(inode->i_nlink);
670         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
671         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
672
673         err = ocfs2_journal_dirty(handle, fe_bh);
674         if (err < 0) {
675                 le16_add_cpu(&fe->i_links_count, -1);
676                 drop_nlink(inode);
677                 mlog_errno(err);
678                 goto out_commit;
679         }
680
681         err = ocfs2_add_entry(handle, dentry, inode,
682                               OCFS2_I(inode)->ip_blkno,
683                               parent_fe_bh, de_bh);
684         if (err) {
685                 le16_add_cpu(&fe->i_links_count, -1);
686                 drop_nlink(inode);
687                 mlog_errno(err);
688                 goto out_commit;
689         }
690
691         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
692         if (err) {
693                 mlog_errno(err);
694                 goto out_commit;
695         }
696
697         atomic_inc(&inode->i_count);
698         dentry->d_op = &ocfs2_dentry_ops;
699         d_instantiate(dentry, inode);
700
701 out_commit:
702         ocfs2_commit_trans(osb, handle);
703 out_unlock_inode:
704         ocfs2_inode_unlock(inode, 1);
705
706 out:
707         ocfs2_inode_unlock(dir, 1);
708
709         brelse(de_bh);
710         brelse(fe_bh);
711         brelse(parent_fe_bh);
712
713         mlog_exit(err);
714
715         return err;
716 }
717
718 /*
719  * Takes and drops an exclusive lock on the given dentry. This will
720  * force other nodes to drop it.
721  */
722 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
723 {
724         int ret;
725
726         ret = ocfs2_dentry_lock(dentry, 1);
727         if (ret)
728                 mlog_errno(ret);
729         else
730                 ocfs2_dentry_unlock(dentry, 1);
731
732         return ret;
733 }
734
735 static inline int inode_is_unlinkable(struct inode *inode)
736 {
737         if (S_ISDIR(inode->i_mode)) {
738                 if (inode->i_nlink == 2)
739                         return 1;
740                 return 0;
741         }
742
743         if (inode->i_nlink == 1)
744                 return 1;
745         return 0;
746 }
747
748 static int ocfs2_unlink(struct inode *dir,
749                         struct dentry *dentry)
750 {
751         int status;
752         int child_locked = 0;
753         struct inode *inode = dentry->d_inode;
754         struct inode *orphan_dir = NULL;
755         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
756         u64 blkno;
757         struct ocfs2_dinode *fe = NULL;
758         struct buffer_head *fe_bh = NULL;
759         struct buffer_head *parent_node_bh = NULL;
760         handle_t *handle = NULL;
761         struct ocfs2_dir_entry *dirent = NULL;
762         struct buffer_head *dirent_bh = NULL;
763         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
764         struct buffer_head *orphan_entry_bh = NULL;
765
766         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
767                    dentry->d_name.len, dentry->d_name.name);
768
769         BUG_ON(dentry->d_parent->d_inode != dir);
770
771         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
772
773         if (inode == osb->root_inode) {
774                 mlog(0, "Cannot delete the root directory\n");
775                 return -EPERM;
776         }
777
778         status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
779         if (status < 0) {
780                 if (status != -ENOENT)
781                         mlog_errno(status);
782                 return status;
783         }
784
785         status = ocfs2_find_files_on_disk(dentry->d_name.name,
786                                           dentry->d_name.len, &blkno,
787                                           dir, &dirent_bh, &dirent);
788         if (status < 0) {
789                 if (status != -ENOENT)
790                         mlog_errno(status);
791                 goto leave;
792         }
793
794         if (OCFS2_I(inode)->ip_blkno != blkno) {
795                 status = -ENOENT;
796
797                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
798                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
799                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
800                 goto leave;
801         }
802
803         status = ocfs2_inode_lock(inode, &fe_bh, 1);
804         if (status < 0) {
805                 if (status != -ENOENT)
806                         mlog_errno(status);
807                 goto leave;
808         }
809         child_locked = 1;
810
811         if (S_ISDIR(inode->i_mode)) {
812                 if (!ocfs2_empty_dir(inode)) {
813                         status = -ENOTEMPTY;
814                         goto leave;
815                 } else if (inode->i_nlink != 2) {
816                         status = -ENOTEMPTY;
817                         goto leave;
818                 }
819         }
820
821         status = ocfs2_remote_dentry_delete(dentry);
822         if (status < 0) {
823                 /* This remote delete should succeed under all normal
824                  * circumstances. */
825                 mlog_errno(status);
826                 goto leave;
827         }
828
829         if (inode_is_unlinkable(inode)) {
830                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
831                                                   orphan_name,
832                                                   &orphan_entry_bh);
833                 if (status < 0) {
834                         mlog_errno(status);
835                         goto leave;
836                 }
837         }
838
839         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
840         if (IS_ERR(handle)) {
841                 status = PTR_ERR(handle);
842                 handle = NULL;
843                 mlog_errno(status);
844                 goto leave;
845         }
846
847         status = ocfs2_journal_access(handle, inode, fe_bh,
848                                       OCFS2_JOURNAL_ACCESS_WRITE);
849         if (status < 0) {
850                 mlog_errno(status);
851                 goto leave;
852         }
853
854         fe = (struct ocfs2_dinode *) fe_bh->b_data;
855
856         if (inode_is_unlinkable(inode)) {
857                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
858                                           orphan_entry_bh, orphan_dir);
859                 if (status < 0) {
860                         mlog_errno(status);
861                         goto leave;
862                 }
863         }
864
865         /* delete the name from the parent dir */
866         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
867         if (status < 0) {
868                 mlog_errno(status);
869                 goto leave;
870         }
871
872         if (S_ISDIR(inode->i_mode))
873                 drop_nlink(inode);
874         drop_nlink(inode);
875         fe->i_links_count = cpu_to_le16(inode->i_nlink);
876
877         status = ocfs2_journal_dirty(handle, fe_bh);
878         if (status < 0) {
879                 mlog_errno(status);
880                 goto leave;
881         }
882
883         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
884         if (S_ISDIR(inode->i_mode))
885                 drop_nlink(dir);
886
887         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
888         if (status < 0) {
889                 mlog_errno(status);
890                 if (S_ISDIR(inode->i_mode))
891                         inc_nlink(dir);
892         }
893
894 leave:
895         if (handle)
896                 ocfs2_commit_trans(osb, handle);
897
898         if (child_locked)
899                 ocfs2_inode_unlock(inode, 1);
900
901         ocfs2_inode_unlock(dir, 1);
902
903         if (orphan_dir) {
904                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
905                 ocfs2_inode_unlock(orphan_dir, 1);
906                 mutex_unlock(&orphan_dir->i_mutex);
907                 iput(orphan_dir);
908         }
909
910         brelse(fe_bh);
911         brelse(dirent_bh);
912         brelse(parent_node_bh);
913         brelse(orphan_entry_bh);
914
915         mlog_exit(status);
916
917         return status;
918 }
919
920 /*
921  * The only place this should be used is rename!
922  * if they have the same id, then the 1st one is the only one locked.
923  */
924 static int ocfs2_double_lock(struct ocfs2_super *osb,
925                              struct buffer_head **bh1,
926                              struct inode *inode1,
927                              struct buffer_head **bh2,
928                              struct inode *inode2)
929 {
930         int status;
931         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
932         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
933         struct buffer_head **tmpbh;
934         struct inode *tmpinode;
935
936         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
937                    (unsigned long long)oi1->ip_blkno,
938                    (unsigned long long)oi2->ip_blkno);
939
940         if (*bh1)
941                 *bh1 = NULL;
942         if (*bh2)
943                 *bh2 = NULL;
944
945         /* we always want to lock the one with the lower lockid first. */
946         if (oi1->ip_blkno != oi2->ip_blkno) {
947                 if (oi1->ip_blkno < oi2->ip_blkno) {
948                         /* switch id1 and id2 around */
949                         mlog(0, "switching them around...\n");
950                         tmpbh = bh2;
951                         bh2 = bh1;
952                         bh1 = tmpbh;
953
954                         tmpinode = inode2;
955                         inode2 = inode1;
956                         inode1 = tmpinode;
957                 }
958                 /* lock id2 */
959                 status = ocfs2_inode_lock(inode2, bh2, 1);
960                 if (status < 0) {
961                         if (status != -ENOENT)
962                                 mlog_errno(status);
963                         goto bail;
964                 }
965         }
966
967         /* lock id1 */
968         status = ocfs2_inode_lock(inode1, bh1, 1);
969         if (status < 0) {
970                 /*
971                  * An error return must mean that no cluster locks
972                  * were held on function exit.
973                  */
974                 if (oi1->ip_blkno != oi2->ip_blkno)
975                         ocfs2_inode_unlock(inode2, 1);
976
977                 if (status != -ENOENT)
978                         mlog_errno(status);
979         }
980
981 bail:
982         mlog_exit(status);
983         return status;
984 }
985
986 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
987 {
988         ocfs2_inode_unlock(inode1, 1);
989
990         if (inode1 != inode2)
991                 ocfs2_inode_unlock(inode2, 1);
992 }
993
994 static int ocfs2_rename(struct inode *old_dir,
995                         struct dentry *old_dentry,
996                         struct inode *new_dir,
997                         struct dentry *new_dentry)
998 {
999         int status = 0, rename_lock = 0, parents_locked = 0;
1000         int old_child_locked = 0, new_child_locked = 0;
1001         struct inode *old_inode = old_dentry->d_inode;
1002         struct inode *new_inode = new_dentry->d_inode;
1003         struct inode *orphan_dir = NULL;
1004         struct ocfs2_dinode *newfe = NULL;
1005         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1006         struct buffer_head *orphan_entry_bh = NULL;
1007         struct buffer_head *newfe_bh = NULL;
1008         struct buffer_head *old_inode_bh = NULL;
1009         struct buffer_head *insert_entry_bh = NULL;
1010         struct ocfs2_super *osb = NULL;
1011         u64 newfe_blkno, old_de_ino;
1012         handle_t *handle = NULL;
1013         struct buffer_head *old_dir_bh = NULL;
1014         struct buffer_head *new_dir_bh = NULL;
1015         struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
1016                 *new_de = NULL;
1017         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1018         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1019                                                     // this is the 1st dirent bh
1020         nlink_t old_dir_nlink = old_dir->i_nlink;
1021         struct ocfs2_dinode *old_di;
1022
1023         /* At some point it might be nice to break this function up a
1024          * bit. */
1025
1026         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1027                    old_dir, old_dentry, new_dir, new_dentry,
1028                    old_dentry->d_name.len, old_dentry->d_name.name,
1029                    new_dentry->d_name.len, new_dentry->d_name.name);
1030
1031         osb = OCFS2_SB(old_dir->i_sb);
1032
1033         if (new_inode) {
1034                 if (!igrab(new_inode))
1035                         BUG();
1036         }
1037
1038         /* Assume a directory hierarchy thusly:
1039          * a/b/c
1040          * a/d
1041          * a,b,c, and d are all directories.
1042          *
1043          * from cwd of 'a' on both nodes:
1044          * node1: mv b/c d
1045          * node2: mv d   b/c
1046          *
1047          * And that's why, just like the VFS, we need a file system
1048          * rename lock. */
1049         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1050                 status = ocfs2_rename_lock(osb);
1051                 if (status < 0) {
1052                         mlog_errno(status);
1053                         goto bail;
1054                 }
1055                 rename_lock = 1;
1056         }
1057
1058         /* if old and new are the same, this'll just do one lock. */
1059         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1060                                    &new_dir_bh, new_dir);
1061         if (status < 0) {
1062                 mlog_errno(status);
1063                 goto bail;
1064         }
1065         parents_locked = 1;
1066
1067         /* make sure both dirs have bhs
1068          * get an extra ref on old_dir_bh if old==new */
1069         if (!new_dir_bh) {
1070                 if (old_dir_bh) {
1071                         new_dir_bh = old_dir_bh;
1072                         get_bh(new_dir_bh);
1073                 } else {
1074                         mlog(ML_ERROR, "no old_dir_bh!\n");
1075                         status = -EIO;
1076                         goto bail;
1077                 }
1078         }
1079
1080         /*
1081          * Aside from allowing a meta data update, the locking here
1082          * also ensures that the downconvert thread on other nodes
1083          * won't have to concurrently downconvert the inode and the
1084          * dentry locks.
1085          */
1086         status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1087         if (status < 0) {
1088                 if (status != -ENOENT)
1089                         mlog_errno(status);
1090                 goto bail;
1091         }
1092         old_child_locked = 1;
1093
1094         status = ocfs2_remote_dentry_delete(old_dentry);
1095         if (status < 0) {
1096                 mlog_errno(status);
1097                 goto bail;
1098         }
1099
1100         if (S_ISDIR(old_inode->i_mode)) {
1101                 u64 old_inode_parent;
1102
1103                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1104                                                   old_inode, &old_inode_de_bh,
1105                                                   &old_inode_dot_dot_de);
1106                 if (status) {
1107                         status = -EIO;
1108                         goto bail;
1109                 }
1110
1111                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1112                         status = -EIO;
1113                         goto bail;
1114                 }
1115
1116                 if (!new_inode && new_dir != old_dir &&
1117                     new_dir->i_nlink >= OCFS2_LINK_MAX) {
1118                         status = -EMLINK;
1119                         goto bail;
1120                 }
1121         }
1122
1123         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1124                                             old_dentry->d_name.len,
1125                                             &old_de_ino);
1126         if (status) {
1127                 status = -ENOENT;
1128                 goto bail;
1129         }
1130
1131         /*
1132          *  Check for inode number is _not_ due to possible IO errors.
1133          *  We might rmdir the source, keep it as pwd of some process
1134          *  and merrily kill the link to whatever was created under the
1135          *  same name. Goodbye sticky bit ;-<
1136          */
1137         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1138                 status = -ENOENT;
1139                 goto bail;
1140         }
1141
1142         /* check if the target already exists (in which case we need
1143          * to delete it */
1144         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1145                                           new_dentry->d_name.len,
1146                                           &newfe_blkno, new_dir, &new_de_bh,
1147                                           &new_de);
1148         /* The only error we allow here is -ENOENT because the new
1149          * file not existing is perfectly valid. */
1150         if ((status < 0) && (status != -ENOENT)) {
1151                 /* If we cannot find the file specified we should just */
1152                 /* return the error... */
1153                 mlog_errno(status);
1154                 goto bail;
1155         }
1156
1157         if (!new_de && new_inode) {
1158                 /*
1159                  * Target was unlinked by another node while we were
1160                  * waiting to get to ocfs2_rename(). There isn't
1161                  * anything we can do here to help the situation, so
1162                  * bubble up the appropriate error.
1163                  */
1164                 status = -ENOENT;
1165                 goto bail;
1166         }
1167
1168         /* In case we need to overwrite an existing file, we blow it
1169          * away first */
1170         if (new_de) {
1171                 /* VFS didn't think there existed an inode here, but
1172                  * someone else in the cluster must have raced our
1173                  * rename to create one. Today we error cleanly, in
1174                  * the future we should consider calling iget to build
1175                  * a new struct inode for this entry. */
1176                 if (!new_inode) {
1177                         status = -EACCES;
1178
1179                         mlog(0, "We found an inode for name %.*s but VFS "
1180                              "didn't give us one.\n", new_dentry->d_name.len,
1181                              new_dentry->d_name.name);
1182                         goto bail;
1183                 }
1184
1185                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1186                         status = -EACCES;
1187
1188                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1189                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1190                              (unsigned long long)newfe_blkno,
1191                              OCFS2_I(new_inode)->ip_flags);
1192                         goto bail;
1193                 }
1194
1195                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1196                 if (status < 0) {
1197                         if (status != -ENOENT)
1198                                 mlog_errno(status);
1199                         goto bail;
1200                 }
1201                 new_child_locked = 1;
1202
1203                 status = ocfs2_remote_dentry_delete(new_dentry);
1204                 if (status < 0) {
1205                         mlog_errno(status);
1206                         goto bail;
1207                 }
1208
1209                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1210
1211                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1212                      "newfebh=%p bhblocknr=%llu\n", new_de,
1213                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1214                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1215
1216                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1217                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1218                                                           new_inode,
1219                                                           orphan_name,
1220                                                           &orphan_entry_bh);
1221                         if (status < 0) {
1222                                 mlog_errno(status);
1223                                 goto bail;
1224                         }
1225                 }
1226         } else {
1227                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1228
1229                 status = ocfs2_check_dir_for_entry(new_dir,
1230                                                    new_dentry->d_name.name,
1231                                                    new_dentry->d_name.len);
1232                 if (status)
1233                         goto bail;
1234
1235                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1236                                                       new_dentry->d_name.name,
1237                                                       new_dentry->d_name.len,
1238                                                       &insert_entry_bh);
1239                 if (status < 0) {
1240                         mlog_errno(status);
1241                         goto bail;
1242                 }
1243         }
1244
1245         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1246         if (IS_ERR(handle)) {
1247                 status = PTR_ERR(handle);
1248                 handle = NULL;
1249                 mlog_errno(status);
1250                 goto bail;
1251         }
1252
1253         if (new_de) {
1254                 if (S_ISDIR(new_inode->i_mode)) {
1255                         if (!ocfs2_empty_dir(new_inode) ||
1256                             new_inode->i_nlink != 2) {
1257                                 status = -ENOTEMPTY;
1258                                 goto bail;
1259                         }
1260                 }
1261                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1262                                               OCFS2_JOURNAL_ACCESS_WRITE);
1263                 if (status < 0) {
1264                         mlog_errno(status);
1265                         goto bail;
1266                 }
1267
1268                 if (S_ISDIR(new_inode->i_mode) ||
1269                     (newfe->i_links_count == cpu_to_le16(1))){
1270                         status = ocfs2_orphan_add(osb, handle, new_inode,
1271                                                   newfe, orphan_name,
1272                                                   orphan_entry_bh, orphan_dir);
1273                         if (status < 0) {
1274                                 mlog_errno(status);
1275                                 goto bail;
1276                         }
1277                 }
1278
1279                 /* change the dirent to point to the correct inode */
1280                 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1281                                             new_de, old_inode);
1282                 if (status < 0) {
1283                         mlog_errno(status);
1284                         goto bail;
1285                 }
1286                 new_dir->i_version++;
1287
1288                 if (S_ISDIR(new_inode->i_mode))
1289                         newfe->i_links_count = 0;
1290                 else
1291                         le16_add_cpu(&newfe->i_links_count, -1);
1292
1293                 status = ocfs2_journal_dirty(handle, newfe_bh);
1294                 if (status < 0) {
1295                         mlog_errno(status);
1296                         goto bail;
1297                 }
1298         } else {
1299                 /* if the name was not found in new_dir, add it now */
1300                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1301                                          OCFS2_I(old_inode)->ip_blkno,
1302                                          new_dir_bh, insert_entry_bh);
1303         }
1304
1305         old_inode->i_ctime = CURRENT_TIME;
1306         mark_inode_dirty(old_inode);
1307
1308         status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1309                                       OCFS2_JOURNAL_ACCESS_WRITE);
1310         if (status >= 0) {
1311                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1312
1313                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1314                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1315
1316                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1317                 if (status < 0)
1318                         mlog_errno(status);
1319         } else
1320                 mlog_errno(status);
1321
1322         /*
1323          * Now that the name has been added to new_dir, remove the old name.
1324          *
1325          * We don't keep any directory entry context around until now
1326          * because the insert might have changed the type of directory
1327          * we're dealing with.
1328          */
1329         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1330                                      old_dentry->d_name.len,
1331                                      old_dir, &old_de);
1332         if (!old_de_bh) {
1333                 status = -EIO;
1334                 goto bail;
1335         }
1336
1337         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1338         if (status < 0) {
1339                 mlog_errno(status);
1340                 goto bail;
1341         }
1342
1343         if (new_inode) {
1344                 new_inode->i_nlink--;
1345                 new_inode->i_ctime = CURRENT_TIME;
1346         }
1347         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1348         if (old_inode_de_bh) {
1349                 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1350                                             old_inode_dot_dot_de, new_dir);
1351                 old_dir->i_nlink--;
1352                 if (new_inode) {
1353                         new_inode->i_nlink--;
1354                 } else {
1355                         inc_nlink(new_dir);
1356                         mark_inode_dirty(new_dir);
1357                 }
1358         }
1359         mark_inode_dirty(old_dir);
1360         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1361         if (new_inode) {
1362                 mark_inode_dirty(new_inode);
1363                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1364         }
1365
1366         if (old_dir != new_dir) {
1367                 /* Keep the same times on both directories.*/
1368                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1369
1370                 /*
1371                  * This will also pick up the i_nlink change from the
1372                  * block above.
1373                  */
1374                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1375         }
1376
1377         if (old_dir_nlink != old_dir->i_nlink) {
1378                 if (!old_dir_bh) {
1379                         mlog(ML_ERROR, "need to change nlink for old dir "
1380                              "%llu from %d to %d but bh is NULL!\n",
1381                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1382                              (int)old_dir_nlink, old_dir->i_nlink);
1383                 } else {
1384                         struct ocfs2_dinode *fe;
1385                         status = ocfs2_journal_access(handle, old_dir,
1386                                                       old_dir_bh,
1387                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1388                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1389                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1390                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1391                 }
1392         }
1393
1394         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1395         status = 0;
1396 bail:
1397         if (rename_lock)
1398                 ocfs2_rename_unlock(osb);
1399
1400         if (handle)
1401                 ocfs2_commit_trans(osb, handle);
1402
1403         if (parents_locked)
1404                 ocfs2_double_unlock(old_dir, new_dir);
1405
1406         if (old_child_locked)
1407                 ocfs2_inode_unlock(old_inode, 1);
1408
1409         if (new_child_locked)
1410                 ocfs2_inode_unlock(new_inode, 1);
1411
1412         if (orphan_dir) {
1413                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1414                 ocfs2_inode_unlock(orphan_dir, 1);
1415                 mutex_unlock(&orphan_dir->i_mutex);
1416                 iput(orphan_dir);
1417         }
1418
1419         if (new_inode)
1420                 sync_mapping_buffers(old_inode->i_mapping);
1421
1422         if (new_inode)
1423                 iput(new_inode);
1424         brelse(newfe_bh);
1425         brelse(old_inode_bh);
1426         brelse(old_dir_bh);
1427         brelse(new_dir_bh);
1428         brelse(new_de_bh);
1429         brelse(old_de_bh);
1430         brelse(old_inode_de_bh);
1431         brelse(orphan_entry_bh);
1432         brelse(insert_entry_bh);
1433
1434         mlog_exit(status);
1435
1436         return status;
1437 }
1438
1439 /*
1440  * we expect i_size = strlen(symname). Copy symname into the file
1441  * data, including the null terminator.
1442  */
1443 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1444                                      handle_t *handle,
1445                                      struct inode *inode,
1446                                      const char *symname)
1447 {
1448         struct buffer_head **bhs = NULL;
1449         const char *c;
1450         struct super_block *sb = osb->sb;
1451         u64 p_blkno, p_blocks;
1452         int virtual, blocks, status, i, bytes_left;
1453
1454         bytes_left = i_size_read(inode) + 1;
1455         /* we can't trust i_blocks because we're actually going to
1456          * write i_size + 1 bytes. */
1457         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1458
1459         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1460                         (unsigned long long)inode->i_blocks,
1461                         i_size_read(inode), blocks);
1462
1463         /* Sanity check -- make sure we're going to fit. */
1464         if (bytes_left >
1465             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1466                 status = -EIO;
1467                 mlog_errno(status);
1468                 goto bail;
1469         }
1470
1471         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1472         if (!bhs) {
1473                 status = -ENOMEM;
1474                 mlog_errno(status);
1475                 goto bail;
1476         }
1477
1478         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1479                                              NULL);
1480         if (status < 0) {
1481                 mlog_errno(status);
1482                 goto bail;
1483         }
1484
1485         /* links can never be larger than one cluster so we know this
1486          * is all going to be contiguous, but do a sanity check
1487          * anyway. */
1488         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1489                 status = -EIO;
1490                 mlog_errno(status);
1491                 goto bail;
1492         }
1493
1494         virtual = 0;
1495         while(bytes_left > 0) {
1496                 c = &symname[virtual * sb->s_blocksize];
1497
1498                 bhs[virtual] = sb_getblk(sb, p_blkno);
1499                 if (!bhs[virtual]) {
1500                         status = -ENOMEM;
1501                         mlog_errno(status);
1502                         goto bail;
1503                 }
1504                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1505
1506                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1507                                               OCFS2_JOURNAL_ACCESS_CREATE);
1508                 if (status < 0) {
1509                         mlog_errno(status);
1510                         goto bail;
1511                 }
1512
1513                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1514
1515                 memcpy(bhs[virtual]->b_data, c,
1516                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1517                        bytes_left);
1518
1519                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1520                 if (status < 0) {
1521                         mlog_errno(status);
1522                         goto bail;
1523                 }
1524
1525                 virtual++;
1526                 p_blkno++;
1527                 bytes_left -= sb->s_blocksize;
1528         }
1529
1530         status = 0;
1531 bail:
1532
1533         if (bhs) {
1534                 for(i = 0; i < blocks; i++)
1535                         brelse(bhs[i]);
1536                 kfree(bhs);
1537         }
1538
1539         mlog_exit(status);
1540         return status;
1541 }
1542
1543 static int ocfs2_symlink(struct inode *dir,
1544                          struct dentry *dentry,
1545                          const char *symname)
1546 {
1547         int status, l, credits;
1548         u64 newsize;
1549         struct ocfs2_super *osb = NULL;
1550         struct inode *inode = NULL;
1551         struct super_block *sb;
1552         struct buffer_head *new_fe_bh = NULL;
1553         struct buffer_head *de_bh = NULL;
1554         struct buffer_head *parent_fe_bh = NULL;
1555         struct ocfs2_dinode *fe = NULL;
1556         struct ocfs2_dinode *dirfe;
1557         handle_t *handle = NULL;
1558         struct ocfs2_alloc_context *inode_ac = NULL;
1559         struct ocfs2_alloc_context *data_ac = NULL;
1560         struct ocfs2_alloc_context *xattr_ac = NULL;
1561         int want_clusters = 0;
1562         int xattr_credits = 0;
1563         struct ocfs2_security_xattr_info si = {
1564                 .enable = 1,
1565         };
1566
1567         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1568                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1569
1570         sb = dir->i_sb;
1571         osb = OCFS2_SB(sb);
1572
1573         l = strlen(symname) + 1;
1574
1575         credits = ocfs2_calc_symlink_credits(sb);
1576
1577         /* lock the parent directory */
1578         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1579         if (status < 0) {
1580                 if (status != -ENOENT)
1581                         mlog_errno(status);
1582                 return status;
1583         }
1584
1585         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1586         if (!dirfe->i_links_count) {
1587                 /* can't make a file in a deleted directory. */
1588                 status = -ENOENT;
1589                 goto bail;
1590         }
1591
1592         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1593                                            dentry->d_name.len);
1594         if (status)
1595                 goto bail;
1596
1597         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1598                                               dentry->d_name.name,
1599                                               dentry->d_name.len, &de_bh);
1600         if (status < 0) {
1601                 mlog_errno(status);
1602                 goto bail;
1603         }
1604
1605         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1606         if (status < 0) {
1607                 if (status != -ENOSPC)
1608                         mlog_errno(status);
1609                 goto bail;
1610         }
1611
1612         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1613         if (!inode) {
1614                 status = -ENOMEM;
1615                 mlog_errno(status);
1616                 goto bail;
1617         }
1618
1619         /* get security xattr */
1620         status = ocfs2_init_security_get(inode, dir, &si);
1621         if (status) {
1622                 if (status == -EOPNOTSUPP)
1623                         si.enable = 0;
1624                 else {
1625                         mlog_errno(status);
1626                         goto bail;
1627                 }
1628         }
1629
1630         /* calculate meta data/clusters for setting security xattr */
1631         if (si.enable) {
1632                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1633                                                   &xattr_credits, &xattr_ac);
1634                 if (status < 0) {
1635                         mlog_errno(status);
1636                         goto bail;
1637                 }
1638         }
1639
1640         /* don't reserve bitmap space for fast symlinks. */
1641         if (l > ocfs2_fast_symlink_chars(sb))
1642                 want_clusters += 1;
1643
1644         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1645         if (status < 0) {
1646                 if (status != -ENOSPC)
1647                         mlog_errno(status);
1648                 goto bail;
1649         }
1650
1651         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1652         if (IS_ERR(handle)) {
1653                 status = PTR_ERR(handle);
1654                 handle = NULL;
1655                 mlog_errno(status);
1656                 goto bail;
1657         }
1658
1659         status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1660                                     0, &new_fe_bh, parent_fe_bh, handle,
1661                                     inode_ac);
1662         if (status < 0) {
1663                 mlog_errno(status);
1664                 goto bail;
1665         }
1666
1667         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1668         inode->i_rdev = 0;
1669         newsize = l - 1;
1670         if (l > ocfs2_fast_symlink_chars(sb)) {
1671                 u32 offset = 0;
1672
1673                 inode->i_op = &ocfs2_symlink_inode_operations;
1674                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1675                                               new_fe_bh,
1676                                               handle, data_ac, NULL,
1677                                               NULL);
1678                 if (status < 0) {
1679                         if (status != -ENOSPC && status != -EINTR) {
1680                                 mlog(ML_ERROR,
1681                                      "Failed to extend file to %llu\n",
1682                                      (unsigned long long)newsize);
1683                                 mlog_errno(status);
1684                                 status = -ENOSPC;
1685                         }
1686                         goto bail;
1687                 }
1688                 i_size_write(inode, newsize);
1689                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1690         } else {
1691                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1692                 memcpy((char *) fe->id2.i_symlink, symname, l);
1693                 i_size_write(inode, newsize);
1694                 inode->i_blocks = 0;
1695         }
1696
1697         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1698         if (status < 0) {
1699                 mlog_errno(status);
1700                 goto bail;
1701         }
1702
1703         if (!ocfs2_inode_is_fast_symlink(inode)) {
1704                 status = ocfs2_create_symlink_data(osb, handle, inode,
1705                                                    symname);
1706                 if (status < 0) {
1707                         mlog_errno(status);
1708                         goto bail;
1709                 }
1710         }
1711
1712         if (si.enable) {
1713                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1714                                                  xattr_ac, data_ac);
1715                 if (status < 0) {
1716                         mlog_errno(status);
1717                         goto bail;
1718                 }
1719         }
1720
1721         status = ocfs2_add_entry(handle, dentry, inode,
1722                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1723                                  de_bh);
1724         if (status < 0) {
1725                 mlog_errno(status);
1726                 goto bail;
1727         }
1728
1729         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1730         if (status) {
1731                 mlog_errno(status);
1732                 goto bail;
1733         }
1734
1735         insert_inode_hash(inode);
1736         dentry->d_op = &ocfs2_dentry_ops;
1737         d_instantiate(dentry, inode);
1738 bail:
1739         if (handle)
1740                 ocfs2_commit_trans(osb, handle);
1741
1742         ocfs2_inode_unlock(dir, 1);
1743
1744         brelse(new_fe_bh);
1745         brelse(parent_fe_bh);
1746         brelse(de_bh);
1747         kfree(si.name);
1748         kfree(si.value);
1749         if (inode_ac)
1750                 ocfs2_free_alloc_context(inode_ac);
1751         if (data_ac)
1752                 ocfs2_free_alloc_context(data_ac);
1753         if (xattr_ac)
1754                 ocfs2_free_alloc_context(xattr_ac);
1755         if ((status < 0) && inode) {
1756                 clear_nlink(inode);
1757                 iput(inode);
1758         }
1759
1760         mlog_exit(status);
1761
1762         return status;
1763 }
1764
1765 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1766 {
1767         int status, namelen;
1768
1769         mlog_entry_void();
1770
1771         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1772                            (long long)blkno);
1773         if (namelen <= 0) {
1774                 if (namelen)
1775                         status = namelen;
1776                 else
1777                         status = -EINVAL;
1778                 mlog_errno(status);
1779                 goto bail;
1780         }
1781         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1782                 status = -EINVAL;
1783                 mlog_errno(status);
1784                 goto bail;
1785         }
1786
1787         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1788              namelen);
1789
1790         status = 0;
1791 bail:
1792         mlog_exit(status);
1793         return status;
1794 }
1795
1796 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1797                                     struct inode **ret_orphan_dir,
1798                                     struct inode *inode,
1799                                     char *name,
1800                                     struct buffer_head **de_bh)
1801 {
1802         struct inode *orphan_dir_inode;
1803         struct buffer_head *orphan_dir_bh = NULL;
1804         int status = 0;
1805
1806         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1807         if (status < 0) {
1808                 mlog_errno(status);
1809                 return status;
1810         }
1811
1812         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1813                                                        ORPHAN_DIR_SYSTEM_INODE,
1814                                                        osb->slot_num);
1815         if (!orphan_dir_inode) {
1816                 status = -ENOENT;
1817                 mlog_errno(status);
1818                 return status;
1819         }
1820
1821         mutex_lock(&orphan_dir_inode->i_mutex);
1822
1823         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1824         if (status < 0) {
1825                 mlog_errno(status);
1826                 goto leave;
1827         }
1828
1829         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1830                                               orphan_dir_bh, name,
1831                                               OCFS2_ORPHAN_NAMELEN, de_bh);
1832         if (status < 0) {
1833                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1834
1835                 mlog_errno(status);
1836                 goto leave;
1837         }
1838
1839         *ret_orphan_dir = orphan_dir_inode;
1840
1841 leave:
1842         if (status) {
1843                 mutex_unlock(&orphan_dir_inode->i_mutex);
1844                 iput(orphan_dir_inode);
1845         }
1846
1847         brelse(orphan_dir_bh);
1848
1849         mlog_exit(status);
1850         return status;
1851 }
1852
1853 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1854                             handle_t *handle,
1855                             struct inode *inode,
1856                             struct ocfs2_dinode *fe,
1857                             char *name,
1858                             struct buffer_head *de_bh,
1859                             struct inode *orphan_dir_inode)
1860 {
1861         struct buffer_head *orphan_dir_bh = NULL;
1862         int status = 0;
1863         struct ocfs2_dinode *orphan_fe;
1864
1865         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1866
1867         status = ocfs2_read_block(orphan_dir_inode,
1868                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
1869                                   &orphan_dir_bh);
1870         if (status < 0) {
1871                 mlog_errno(status);
1872                 goto leave;
1873         }
1874
1875         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1876                                       OCFS2_JOURNAL_ACCESS_WRITE);
1877         if (status < 0) {
1878                 mlog_errno(status);
1879                 goto leave;
1880         }
1881
1882         /* we're a cluster, and nlink can change on disk from
1883          * underneath us... */
1884         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1885         if (S_ISDIR(inode->i_mode))
1886                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1887         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1888
1889         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1890         if (status < 0) {
1891                 mlog_errno(status);
1892                 goto leave;
1893         }
1894
1895         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1896                                    OCFS2_ORPHAN_NAMELEN, inode,
1897                                    OCFS2_I(inode)->ip_blkno,
1898                                    orphan_dir_bh, de_bh);
1899         if (status < 0) {
1900                 mlog_errno(status);
1901                 goto leave;
1902         }
1903
1904         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1905
1906         /* Record which orphan dir our inode now resides
1907          * in. delete_inode will use this to determine which orphan
1908          * dir to lock. */
1909         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1910
1911         mlog(0, "Inode %llu orphaned in slot %d\n",
1912              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1913
1914 leave:
1915         brelse(orphan_dir_bh);
1916
1917         mlog_exit(status);
1918         return status;
1919 }
1920
1921 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1922 int ocfs2_orphan_del(struct ocfs2_super *osb,
1923                      handle_t *handle,
1924                      struct inode *orphan_dir_inode,
1925                      struct inode *inode,
1926                      struct buffer_head *orphan_dir_bh)
1927 {
1928         char name[OCFS2_ORPHAN_NAMELEN + 1];
1929         struct ocfs2_dinode *orphan_fe;
1930         int status = 0;
1931         struct buffer_head *target_de_bh = NULL;
1932         struct ocfs2_dir_entry *target_de = NULL;
1933
1934         mlog_entry_void();
1935
1936         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1937         if (status < 0) {
1938                 mlog_errno(status);
1939                 goto leave;
1940         }
1941
1942         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1943              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1944              OCFS2_ORPHAN_NAMELEN);
1945
1946         /* find it's spot in the orphan directory */
1947         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1948                                         orphan_dir_inode, &target_de);
1949         if (!target_de_bh) {
1950                 status = -ENOENT;
1951                 mlog_errno(status);
1952                 goto leave;
1953         }
1954
1955         /* remove it from the orphan directory */
1956         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1957                                     target_de_bh);
1958         if (status < 0) {
1959                 mlog_errno(status);
1960                 goto leave;
1961         }
1962
1963         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
1964                                       OCFS2_JOURNAL_ACCESS_WRITE);
1965         if (status < 0) {
1966                 mlog_errno(status);
1967                 goto leave;
1968         }
1969
1970         /* do the i_nlink dance! :) */
1971         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1972         if (S_ISDIR(inode->i_mode))
1973                 le16_add_cpu(&orphan_fe->i_links_count, -1);
1974         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1975
1976         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1977         if (status < 0) {
1978                 mlog_errno(status);
1979                 goto leave;
1980         }
1981
1982 leave:
1983         brelse(target_de_bh);
1984
1985         mlog_exit(status);
1986         return status;
1987 }
1988
1989 const struct inode_operations ocfs2_dir_iops = {
1990         .create         = ocfs2_create,
1991         .lookup         = ocfs2_lookup,
1992         .link           = ocfs2_link,
1993         .unlink         = ocfs2_unlink,
1994         .rmdir          = ocfs2_unlink,
1995         .symlink        = ocfs2_symlink,
1996         .mkdir          = ocfs2_mkdir,
1997         .mknod          = ocfs2_mknod,
1998         .rename         = ocfs2_rename,
1999         .setattr        = ocfs2_setattr,
2000         .getattr        = ocfs2_getattr,
2001         .permission     = ocfs2_permission,
2002         .setxattr       = generic_setxattr,
2003         .getxattr       = generic_getxattr,
2004         .listxattr      = ocfs2_listxattr,
2005         .removexattr    = generic_removexattr,
2006 };