ocfs2: Access and dirty the buffer_head in mark_written.
[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 #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 buffer_head **de_bh);
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 buffer_head *de_bh,
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 buffer_head *de_bh = NULL;
232         struct inode *inode = NULL;
233         struct ocfs2_alloc_context *inode_ac = NULL;
234         struct ocfs2_alloc_context *data_ac = NULL;
235         struct ocfs2_alloc_context *xattr_ac = NULL;
236         int want_clusters = 0;
237         int xattr_credits = 0;
238         struct ocfs2_security_xattr_info si = {
239                 .enable = 1,
240         };
241         int did_quota_inode = 0;
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, &de_bh);
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                                  de_bh);
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(de_bh);
427         brelse(parent_fe_bh);
428         kfree(si.name);
429         kfree(si.value);
430
431         if ((status < 0) && inode) {
432                 clear_nlink(inode);
433                 iput(inode);
434         }
435
436         if (inode_ac)
437                 ocfs2_free_alloc_context(inode_ac);
438
439         if (data_ac)
440                 ocfs2_free_alloc_context(data_ac);
441
442         if (xattr_ac)
443                 ocfs2_free_alloc_context(xattr_ac);
444
445         mlog_exit(status);
446
447         return status;
448 }
449
450 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
451                               struct inode *dir,
452                               struct inode *inode,
453                               struct dentry *dentry,
454                               dev_t dev,
455                               struct buffer_head **new_fe_bh,
456                               struct buffer_head *parent_fe_bh,
457                               handle_t *handle,
458                               struct ocfs2_alloc_context *inode_ac)
459 {
460         int status = 0;
461         struct ocfs2_dinode *fe = NULL;
462         struct ocfs2_extent_list *fel;
463         u64 fe_blkno = 0;
464         u16 suballoc_bit;
465
466         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
467                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
468                    dentry->d_name.name);
469
470         *new_fe_bh = NULL;
471
472         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
473                                        &fe_blkno);
474         if (status < 0) {
475                 mlog_errno(status);
476                 goto leave;
477         }
478
479         /* populate as many fields early on as possible - many of
480          * these are used by the support functions here and in
481          * callers. */
482         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
483         OCFS2_I(inode)->ip_blkno = fe_blkno;
484         spin_lock(&osb->osb_lock);
485         inode->i_generation = osb->s_next_generation++;
486         spin_unlock(&osb->osb_lock);
487
488         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
489         if (!*new_fe_bh) {
490                 status = -EIO;
491                 mlog_errno(status);
492                 goto leave;
493         }
494         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
495
496         status = ocfs2_journal_access_di(handle, inode, *new_fe_bh,
497                                          OCFS2_JOURNAL_ACCESS_CREATE);
498         if (status < 0) {
499                 mlog_errno(status);
500                 goto leave;
501         }
502
503         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
504         memset(fe, 0, osb->sb->s_blocksize);
505
506         fe->i_generation = cpu_to_le32(inode->i_generation);
507         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
508         fe->i_blkno = cpu_to_le64(fe_blkno);
509         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
510         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
511         fe->i_uid = cpu_to_le32(inode->i_uid);
512         fe->i_gid = cpu_to_le32(inode->i_gid);
513         fe->i_mode = cpu_to_le16(inode->i_mode);
514         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
515                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
516         fe->i_links_count = cpu_to_le16(inode->i_nlink);
517
518         fe->i_last_eb_blk = 0;
519         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
520         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
521         fe->i_atime = fe->i_ctime = fe->i_mtime =
522                 cpu_to_le64(CURRENT_TIME.tv_sec);
523         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
524                 cpu_to_le32(CURRENT_TIME.tv_nsec);
525         fe->i_dtime = 0;
526
527         /*
528          * If supported, directories start with inline data.
529          */
530         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
531                 u16 feat = le16_to_cpu(fe->i_dyn_features);
532
533                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
534
535                 fe->id2.i_data.id_count = cpu_to_le16(ocfs2_max_inline_data(osb->sb));
536         } else {
537                 fel = &fe->id2.i_list;
538                 fel->l_tree_depth = 0;
539                 fel->l_next_free_rec = 0;
540                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
541         }
542
543         status = ocfs2_journal_dirty(handle, *new_fe_bh);
544         if (status < 0) {
545                 mlog_errno(status);
546                 goto leave;
547         }
548
549         ocfs2_populate_inode(inode, fe, 1);
550         ocfs2_inode_set_new(osb, inode);
551         if (!ocfs2_mount_local(osb)) {
552                 status = ocfs2_create_new_inode_locks(inode);
553                 if (status < 0)
554                         mlog_errno(status);
555         }
556
557         status = 0; /* error in ocfs2_create_new_inode_locks is not
558                      * critical */
559
560 leave:
561         if (status < 0) {
562                 if (*new_fe_bh) {
563                         brelse(*new_fe_bh);
564                         *new_fe_bh = NULL;
565                 }
566         }
567
568         mlog_exit(status);
569         return status;
570 }
571
572 static int ocfs2_mkdir(struct inode *dir,
573                        struct dentry *dentry,
574                        int mode)
575 {
576         int ret;
577
578         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
579                    dentry->d_name.len, dentry->d_name.name);
580         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
581         mlog_exit(ret);
582
583         return ret;
584 }
585
586 static int ocfs2_create(struct inode *dir,
587                         struct dentry *dentry,
588                         int mode,
589                         struct nameidata *nd)
590 {
591         int ret;
592
593         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
594                    dentry->d_name.len, dentry->d_name.name);
595         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
596         mlog_exit(ret);
597
598         return ret;
599 }
600
601 static int ocfs2_link(struct dentry *old_dentry,
602                       struct inode *dir,
603                       struct dentry *dentry)
604 {
605         handle_t *handle;
606         struct inode *inode = old_dentry->d_inode;
607         int err;
608         struct buffer_head *fe_bh = NULL;
609         struct buffer_head *parent_fe_bh = NULL;
610         struct buffer_head *de_bh = NULL;
611         struct ocfs2_dinode *fe = NULL;
612         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
613
614         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
615                    old_dentry->d_name.len, old_dentry->d_name.name,
616                    dentry->d_name.len, dentry->d_name.name);
617
618         if (S_ISDIR(inode->i_mode))
619                 return -EPERM;
620
621         err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
622         if (err < 0) {
623                 if (err != -ENOENT)
624                         mlog_errno(err);
625                 return err;
626         }
627
628         if (!dir->i_nlink) {
629                 err = -ENOENT;
630                 goto out;
631         }
632
633         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
634                                         dentry->d_name.len);
635         if (err)
636                 goto out;
637
638         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
639                                            dentry->d_name.name,
640                                            dentry->d_name.len, &de_bh);
641         if (err < 0) {
642                 mlog_errno(err);
643                 goto out;
644         }
645
646         err = ocfs2_inode_lock(inode, &fe_bh, 1);
647         if (err < 0) {
648                 if (err != -ENOENT)
649                         mlog_errno(err);
650                 goto out;
651         }
652
653         fe = (struct ocfs2_dinode *) fe_bh->b_data;
654         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
655                 err = -EMLINK;
656                 goto out_unlock_inode;
657         }
658
659         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
660         if (IS_ERR(handle)) {
661                 err = PTR_ERR(handle);
662                 handle = NULL;
663                 mlog_errno(err);
664                 goto out_unlock_inode;
665         }
666
667         err = ocfs2_journal_access_di(handle, inode, fe_bh,
668                                       OCFS2_JOURNAL_ACCESS_WRITE);
669         if (err < 0) {
670                 mlog_errno(err);
671                 goto out_commit;
672         }
673
674         inc_nlink(inode);
675         inode->i_ctime = CURRENT_TIME;
676         fe->i_links_count = cpu_to_le16(inode->i_nlink);
677         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
678         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
679
680         err = ocfs2_journal_dirty(handle, fe_bh);
681         if (err < 0) {
682                 le16_add_cpu(&fe->i_links_count, -1);
683                 drop_nlink(inode);
684                 mlog_errno(err);
685                 goto out_commit;
686         }
687
688         err = ocfs2_add_entry(handle, dentry, inode,
689                               OCFS2_I(inode)->ip_blkno,
690                               parent_fe_bh, de_bh);
691         if (err) {
692                 le16_add_cpu(&fe->i_links_count, -1);
693                 drop_nlink(inode);
694                 mlog_errno(err);
695                 goto out_commit;
696         }
697
698         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
699         if (err) {
700                 mlog_errno(err);
701                 goto out_commit;
702         }
703
704         atomic_inc(&inode->i_count);
705         dentry->d_op = &ocfs2_dentry_ops;
706         d_instantiate(dentry, inode);
707
708 out_commit:
709         ocfs2_commit_trans(osb, handle);
710 out_unlock_inode:
711         ocfs2_inode_unlock(inode, 1);
712
713 out:
714         ocfs2_inode_unlock(dir, 1);
715
716         brelse(de_bh);
717         brelse(fe_bh);
718         brelse(parent_fe_bh);
719
720         mlog_exit(err);
721
722         return err;
723 }
724
725 /*
726  * Takes and drops an exclusive lock on the given dentry. This will
727  * force other nodes to drop it.
728  */
729 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
730 {
731         int ret;
732
733         ret = ocfs2_dentry_lock(dentry, 1);
734         if (ret)
735                 mlog_errno(ret);
736         else
737                 ocfs2_dentry_unlock(dentry, 1);
738
739         return ret;
740 }
741
742 static inline int inode_is_unlinkable(struct inode *inode)
743 {
744         if (S_ISDIR(inode->i_mode)) {
745                 if (inode->i_nlink == 2)
746                         return 1;
747                 return 0;
748         }
749
750         if (inode->i_nlink == 1)
751                 return 1;
752         return 0;
753 }
754
755 static int ocfs2_unlink(struct inode *dir,
756                         struct dentry *dentry)
757 {
758         int status;
759         int child_locked = 0;
760         struct inode *inode = dentry->d_inode;
761         struct inode *orphan_dir = NULL;
762         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
763         u64 blkno;
764         struct ocfs2_dinode *fe = NULL;
765         struct buffer_head *fe_bh = NULL;
766         struct buffer_head *parent_node_bh = NULL;
767         handle_t *handle = NULL;
768         struct ocfs2_dir_entry *dirent = NULL;
769         struct buffer_head *dirent_bh = NULL;
770         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
771         struct buffer_head *orphan_entry_bh = NULL;
772
773         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
774                    dentry->d_name.len, dentry->d_name.name);
775
776         BUG_ON(dentry->d_parent->d_inode != dir);
777
778         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
779
780         if (inode == osb->root_inode) {
781                 mlog(0, "Cannot delete the root directory\n");
782                 return -EPERM;
783         }
784
785         status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
786         if (status < 0) {
787                 if (status != -ENOENT)
788                         mlog_errno(status);
789                 return status;
790         }
791
792         status = ocfs2_find_files_on_disk(dentry->d_name.name,
793                                           dentry->d_name.len, &blkno,
794                                           dir, &dirent_bh, &dirent);
795         if (status < 0) {
796                 if (status != -ENOENT)
797                         mlog_errno(status);
798                 goto leave;
799         }
800
801         if (OCFS2_I(inode)->ip_blkno != blkno) {
802                 status = -ENOENT;
803
804                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
805                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
806                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
807                 goto leave;
808         }
809
810         status = ocfs2_inode_lock(inode, &fe_bh, 1);
811         if (status < 0) {
812                 if (status != -ENOENT)
813                         mlog_errno(status);
814                 goto leave;
815         }
816         child_locked = 1;
817
818         if (S_ISDIR(inode->i_mode)) {
819                 if (!ocfs2_empty_dir(inode)) {
820                         status = -ENOTEMPTY;
821                         goto leave;
822                 } else if (inode->i_nlink != 2) {
823                         status = -ENOTEMPTY;
824                         goto leave;
825                 }
826         }
827
828         status = ocfs2_remote_dentry_delete(dentry);
829         if (status < 0) {
830                 /* This remote delete should succeed under all normal
831                  * circumstances. */
832                 mlog_errno(status);
833                 goto leave;
834         }
835
836         if (inode_is_unlinkable(inode)) {
837                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
838                                                   orphan_name,
839                                                   &orphan_entry_bh);
840                 if (status < 0) {
841                         mlog_errno(status);
842                         goto leave;
843                 }
844         }
845
846         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
847         if (IS_ERR(handle)) {
848                 status = PTR_ERR(handle);
849                 handle = NULL;
850                 mlog_errno(status);
851                 goto leave;
852         }
853
854         status = ocfs2_journal_access_di(handle, inode, fe_bh,
855                                          OCFS2_JOURNAL_ACCESS_WRITE);
856         if (status < 0) {
857                 mlog_errno(status);
858                 goto leave;
859         }
860
861         fe = (struct ocfs2_dinode *) fe_bh->b_data;
862
863         if (inode_is_unlinkable(inode)) {
864                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
865                                           orphan_entry_bh, orphan_dir);
866                 if (status < 0) {
867                         mlog_errno(status);
868                         goto leave;
869                 }
870         }
871
872         /* delete the name from the parent dir */
873         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
874         if (status < 0) {
875                 mlog_errno(status);
876                 goto leave;
877         }
878
879         if (S_ISDIR(inode->i_mode))
880                 drop_nlink(inode);
881         drop_nlink(inode);
882         fe->i_links_count = cpu_to_le16(inode->i_nlink);
883
884         status = ocfs2_journal_dirty(handle, fe_bh);
885         if (status < 0) {
886                 mlog_errno(status);
887                 goto leave;
888         }
889
890         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
891         if (S_ISDIR(inode->i_mode))
892                 drop_nlink(dir);
893
894         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
895         if (status < 0) {
896                 mlog_errno(status);
897                 if (S_ISDIR(inode->i_mode))
898                         inc_nlink(dir);
899         }
900
901 leave:
902         if (handle)
903                 ocfs2_commit_trans(osb, handle);
904
905         if (child_locked)
906                 ocfs2_inode_unlock(inode, 1);
907
908         ocfs2_inode_unlock(dir, 1);
909
910         if (orphan_dir) {
911                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
912                 ocfs2_inode_unlock(orphan_dir, 1);
913                 mutex_unlock(&orphan_dir->i_mutex);
914                 iput(orphan_dir);
915         }
916
917         brelse(fe_bh);
918         brelse(dirent_bh);
919         brelse(parent_node_bh);
920         brelse(orphan_entry_bh);
921
922         mlog_exit(status);
923
924         return status;
925 }
926
927 /*
928  * The only place this should be used is rename!
929  * if they have the same id, then the 1st one is the only one locked.
930  */
931 static int ocfs2_double_lock(struct ocfs2_super *osb,
932                              struct buffer_head **bh1,
933                              struct inode *inode1,
934                              struct buffer_head **bh2,
935                              struct inode *inode2)
936 {
937         int status;
938         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
939         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
940         struct buffer_head **tmpbh;
941         struct inode *tmpinode;
942
943         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
944                    (unsigned long long)oi1->ip_blkno,
945                    (unsigned long long)oi2->ip_blkno);
946
947         if (*bh1)
948                 *bh1 = NULL;
949         if (*bh2)
950                 *bh2 = NULL;
951
952         /* we always want to lock the one with the lower lockid first. */
953         if (oi1->ip_blkno != oi2->ip_blkno) {
954                 if (oi1->ip_blkno < oi2->ip_blkno) {
955                         /* switch id1 and id2 around */
956                         mlog(0, "switching them around...\n");
957                         tmpbh = bh2;
958                         bh2 = bh1;
959                         bh1 = tmpbh;
960
961                         tmpinode = inode2;
962                         inode2 = inode1;
963                         inode1 = tmpinode;
964                 }
965                 /* lock id2 */
966                 status = ocfs2_inode_lock(inode2, bh2, 1);
967                 if (status < 0) {
968                         if (status != -ENOENT)
969                                 mlog_errno(status);
970                         goto bail;
971                 }
972         }
973
974         /* lock id1 */
975         status = ocfs2_inode_lock(inode1, bh1, 1);
976         if (status < 0) {
977                 /*
978                  * An error return must mean that no cluster locks
979                  * were held on function exit.
980                  */
981                 if (oi1->ip_blkno != oi2->ip_blkno)
982                         ocfs2_inode_unlock(inode2, 1);
983
984                 if (status != -ENOENT)
985                         mlog_errno(status);
986         }
987
988 bail:
989         mlog_exit(status);
990         return status;
991 }
992
993 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
994 {
995         ocfs2_inode_unlock(inode1, 1);
996
997         if (inode1 != inode2)
998                 ocfs2_inode_unlock(inode2, 1);
999 }
1000
1001 static int ocfs2_rename(struct inode *old_dir,
1002                         struct dentry *old_dentry,
1003                         struct inode *new_dir,
1004                         struct dentry *new_dentry)
1005 {
1006         int status = 0, rename_lock = 0, parents_locked = 0;
1007         int old_child_locked = 0, new_child_locked = 0;
1008         struct inode *old_inode = old_dentry->d_inode;
1009         struct inode *new_inode = new_dentry->d_inode;
1010         struct inode *orphan_dir = NULL;
1011         struct ocfs2_dinode *newfe = NULL;
1012         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1013         struct buffer_head *orphan_entry_bh = NULL;
1014         struct buffer_head *newfe_bh = NULL;
1015         struct buffer_head *old_inode_bh = NULL;
1016         struct buffer_head *insert_entry_bh = NULL;
1017         struct ocfs2_super *osb = NULL;
1018         u64 newfe_blkno, old_de_ino;
1019         handle_t *handle = NULL;
1020         struct buffer_head *old_dir_bh = NULL;
1021         struct buffer_head *new_dir_bh = NULL;
1022         struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
1023                 *new_de = NULL;
1024         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1025         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1026                                                     // this is the 1st dirent bh
1027         nlink_t old_dir_nlink = old_dir->i_nlink;
1028         struct ocfs2_dinode *old_di;
1029
1030         /* At some point it might be nice to break this function up a
1031          * bit. */
1032
1033         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1034                    old_dir, old_dentry, new_dir, new_dentry,
1035                    old_dentry->d_name.len, old_dentry->d_name.name,
1036                    new_dentry->d_name.len, new_dentry->d_name.name);
1037
1038         osb = OCFS2_SB(old_dir->i_sb);
1039
1040         if (new_inode) {
1041                 if (!igrab(new_inode))
1042                         BUG();
1043         }
1044
1045         /* Assume a directory hierarchy thusly:
1046          * a/b/c
1047          * a/d
1048          * a,b,c, and d are all directories.
1049          *
1050          * from cwd of 'a' on both nodes:
1051          * node1: mv b/c d
1052          * node2: mv d   b/c
1053          *
1054          * And that's why, just like the VFS, we need a file system
1055          * rename lock. */
1056         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1057                 status = ocfs2_rename_lock(osb);
1058                 if (status < 0) {
1059                         mlog_errno(status);
1060                         goto bail;
1061                 }
1062                 rename_lock = 1;
1063         }
1064
1065         /* if old and new are the same, this'll just do one lock. */
1066         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1067                                    &new_dir_bh, new_dir);
1068         if (status < 0) {
1069                 mlog_errno(status);
1070                 goto bail;
1071         }
1072         parents_locked = 1;
1073
1074         /* make sure both dirs have bhs
1075          * get an extra ref on old_dir_bh if old==new */
1076         if (!new_dir_bh) {
1077                 if (old_dir_bh) {
1078                         new_dir_bh = old_dir_bh;
1079                         get_bh(new_dir_bh);
1080                 } else {
1081                         mlog(ML_ERROR, "no old_dir_bh!\n");
1082                         status = -EIO;
1083                         goto bail;
1084                 }
1085         }
1086
1087         /*
1088          * Aside from allowing a meta data update, the locking here
1089          * also ensures that the downconvert thread on other nodes
1090          * won't have to concurrently downconvert the inode and the
1091          * dentry locks.
1092          */
1093         status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1094         if (status < 0) {
1095                 if (status != -ENOENT)
1096                         mlog_errno(status);
1097                 goto bail;
1098         }
1099         old_child_locked = 1;
1100
1101         status = ocfs2_remote_dentry_delete(old_dentry);
1102         if (status < 0) {
1103                 mlog_errno(status);
1104                 goto bail;
1105         }
1106
1107         if (S_ISDIR(old_inode->i_mode)) {
1108                 u64 old_inode_parent;
1109
1110                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1111                                                   old_inode, &old_inode_de_bh,
1112                                                   &old_inode_dot_dot_de);
1113                 if (status) {
1114                         status = -EIO;
1115                         goto bail;
1116                 }
1117
1118                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1119                         status = -EIO;
1120                         goto bail;
1121                 }
1122
1123                 if (!new_inode && new_dir != old_dir &&
1124                     new_dir->i_nlink >= OCFS2_LINK_MAX) {
1125                         status = -EMLINK;
1126                         goto bail;
1127                 }
1128         }
1129
1130         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1131                                             old_dentry->d_name.len,
1132                                             &old_de_ino);
1133         if (status) {
1134                 status = -ENOENT;
1135                 goto bail;
1136         }
1137
1138         /*
1139          *  Check for inode number is _not_ due to possible IO errors.
1140          *  We might rmdir the source, keep it as pwd of some process
1141          *  and merrily kill the link to whatever was created under the
1142          *  same name. Goodbye sticky bit ;-<
1143          */
1144         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1145                 status = -ENOENT;
1146                 goto bail;
1147         }
1148
1149         /* check if the target already exists (in which case we need
1150          * to delete it */
1151         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1152                                           new_dentry->d_name.len,
1153                                           &newfe_blkno, new_dir, &new_de_bh,
1154                                           &new_de);
1155         /* The only error we allow here is -ENOENT because the new
1156          * file not existing is perfectly valid. */
1157         if ((status < 0) && (status != -ENOENT)) {
1158                 /* If we cannot find the file specified we should just */
1159                 /* return the error... */
1160                 mlog_errno(status);
1161                 goto bail;
1162         }
1163
1164         if (!new_de && new_inode) {
1165                 /*
1166                  * Target was unlinked by another node while we were
1167                  * waiting to get to ocfs2_rename(). There isn't
1168                  * anything we can do here to help the situation, so
1169                  * bubble up the appropriate error.
1170                  */
1171                 status = -ENOENT;
1172                 goto bail;
1173         }
1174
1175         /* In case we need to overwrite an existing file, we blow it
1176          * away first */
1177         if (new_de) {
1178                 /* VFS didn't think there existed an inode here, but
1179                  * someone else in the cluster must have raced our
1180                  * rename to create one. Today we error cleanly, in
1181                  * the future we should consider calling iget to build
1182                  * a new struct inode for this entry. */
1183                 if (!new_inode) {
1184                         status = -EACCES;
1185
1186                         mlog(0, "We found an inode for name %.*s but VFS "
1187                              "didn't give us one.\n", new_dentry->d_name.len,
1188                              new_dentry->d_name.name);
1189                         goto bail;
1190                 }
1191
1192                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1193                         status = -EACCES;
1194
1195                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1196                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1197                              (unsigned long long)newfe_blkno,
1198                              OCFS2_I(new_inode)->ip_flags);
1199                         goto bail;
1200                 }
1201
1202                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1203                 if (status < 0) {
1204                         if (status != -ENOENT)
1205                                 mlog_errno(status);
1206                         goto bail;
1207                 }
1208                 new_child_locked = 1;
1209
1210                 status = ocfs2_remote_dentry_delete(new_dentry);
1211                 if (status < 0) {
1212                         mlog_errno(status);
1213                         goto bail;
1214                 }
1215
1216                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1217
1218                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1219                      "newfebh=%p bhblocknr=%llu\n", new_de,
1220                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1221                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1222
1223                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1224                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1225                                                           new_inode,
1226                                                           orphan_name,
1227                                                           &orphan_entry_bh);
1228                         if (status < 0) {
1229                                 mlog_errno(status);
1230                                 goto bail;
1231                         }
1232                 }
1233         } else {
1234                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1235
1236                 status = ocfs2_check_dir_for_entry(new_dir,
1237                                                    new_dentry->d_name.name,
1238                                                    new_dentry->d_name.len);
1239                 if (status)
1240                         goto bail;
1241
1242                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1243                                                       new_dentry->d_name.name,
1244                                                       new_dentry->d_name.len,
1245                                                       &insert_entry_bh);
1246                 if (status < 0) {
1247                         mlog_errno(status);
1248                         goto bail;
1249                 }
1250         }
1251
1252         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1253         if (IS_ERR(handle)) {
1254                 status = PTR_ERR(handle);
1255                 handle = NULL;
1256                 mlog_errno(status);
1257                 goto bail;
1258         }
1259
1260         if (new_de) {
1261                 if (S_ISDIR(new_inode->i_mode)) {
1262                         if (!ocfs2_empty_dir(new_inode) ||
1263                             new_inode->i_nlink != 2) {
1264                                 status = -ENOTEMPTY;
1265                                 goto bail;
1266                         }
1267                 }
1268                 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1269                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1270                 if (status < 0) {
1271                         mlog_errno(status);
1272                         goto bail;
1273                 }
1274
1275                 if (S_ISDIR(new_inode->i_mode) ||
1276                     (newfe->i_links_count == cpu_to_le16(1))){
1277                         status = ocfs2_orphan_add(osb, handle, new_inode,
1278                                                   newfe, orphan_name,
1279                                                   orphan_entry_bh, orphan_dir);
1280                         if (status < 0) {
1281                                 mlog_errno(status);
1282                                 goto bail;
1283                         }
1284                 }
1285
1286                 /* change the dirent to point to the correct inode */
1287                 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1288                                             new_de, old_inode);
1289                 if (status < 0) {
1290                         mlog_errno(status);
1291                         goto bail;
1292                 }
1293                 new_dir->i_version++;
1294
1295                 if (S_ISDIR(new_inode->i_mode))
1296                         newfe->i_links_count = 0;
1297                 else
1298                         le16_add_cpu(&newfe->i_links_count, -1);
1299
1300                 status = ocfs2_journal_dirty(handle, newfe_bh);
1301                 if (status < 0) {
1302                         mlog_errno(status);
1303                         goto bail;
1304                 }
1305         } else {
1306                 /* if the name was not found in new_dir, add it now */
1307                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1308                                          OCFS2_I(old_inode)->ip_blkno,
1309                                          new_dir_bh, insert_entry_bh);
1310         }
1311
1312         old_inode->i_ctime = CURRENT_TIME;
1313         mark_inode_dirty(old_inode);
1314
1315         status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1316                                          OCFS2_JOURNAL_ACCESS_WRITE);
1317         if (status >= 0) {
1318                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1319
1320                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1321                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1322
1323                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1324                 if (status < 0)
1325                         mlog_errno(status);
1326         } else
1327                 mlog_errno(status);
1328
1329         /*
1330          * Now that the name has been added to new_dir, remove the old name.
1331          *
1332          * We don't keep any directory entry context around until now
1333          * because the insert might have changed the type of directory
1334          * we're dealing with.
1335          */
1336         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1337                                      old_dentry->d_name.len,
1338                                      old_dir, &old_de);
1339         if (!old_de_bh) {
1340                 status = -EIO;
1341                 goto bail;
1342         }
1343
1344         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1345         if (status < 0) {
1346                 mlog_errno(status);
1347                 goto bail;
1348         }
1349
1350         if (new_inode) {
1351                 new_inode->i_nlink--;
1352                 new_inode->i_ctime = CURRENT_TIME;
1353         }
1354         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1355         if (old_inode_de_bh) {
1356                 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1357                                             old_inode_dot_dot_de, new_dir);
1358                 old_dir->i_nlink--;
1359                 if (new_inode) {
1360                         new_inode->i_nlink--;
1361                 } else {
1362                         inc_nlink(new_dir);
1363                         mark_inode_dirty(new_dir);
1364                 }
1365         }
1366         mark_inode_dirty(old_dir);
1367         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1368         if (new_inode) {
1369                 mark_inode_dirty(new_inode);
1370                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1371         }
1372
1373         if (old_dir != new_dir) {
1374                 /* Keep the same times on both directories.*/
1375                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1376
1377                 /*
1378                  * This will also pick up the i_nlink change from the
1379                  * block above.
1380                  */
1381                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1382         }
1383
1384         if (old_dir_nlink != old_dir->i_nlink) {
1385                 if (!old_dir_bh) {
1386                         mlog(ML_ERROR, "need to change nlink for old dir "
1387                              "%llu from %d to %d but bh is NULL!\n",
1388                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1389                              (int)old_dir_nlink, old_dir->i_nlink);
1390                 } else {
1391                         struct ocfs2_dinode *fe;
1392                         status = ocfs2_journal_access_di(handle, old_dir,
1393                                                          old_dir_bh,
1394                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1395                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1396                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1397                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1398                 }
1399         }
1400
1401         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1402         status = 0;
1403 bail:
1404         if (rename_lock)
1405                 ocfs2_rename_unlock(osb);
1406
1407         if (handle)
1408                 ocfs2_commit_trans(osb, handle);
1409
1410         if (parents_locked)
1411                 ocfs2_double_unlock(old_dir, new_dir);
1412
1413         if (old_child_locked)
1414                 ocfs2_inode_unlock(old_inode, 1);
1415
1416         if (new_child_locked)
1417                 ocfs2_inode_unlock(new_inode, 1);
1418
1419         if (orphan_dir) {
1420                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1421                 ocfs2_inode_unlock(orphan_dir, 1);
1422                 mutex_unlock(&orphan_dir->i_mutex);
1423                 iput(orphan_dir);
1424         }
1425
1426         if (new_inode)
1427                 sync_mapping_buffers(old_inode->i_mapping);
1428
1429         if (new_inode)
1430                 iput(new_inode);
1431         brelse(newfe_bh);
1432         brelse(old_inode_bh);
1433         brelse(old_dir_bh);
1434         brelse(new_dir_bh);
1435         brelse(new_de_bh);
1436         brelse(old_de_bh);
1437         brelse(old_inode_de_bh);
1438         brelse(orphan_entry_bh);
1439         brelse(insert_entry_bh);
1440
1441         mlog_exit(status);
1442
1443         return status;
1444 }
1445
1446 /*
1447  * we expect i_size = strlen(symname). Copy symname into the file
1448  * data, including the null terminator.
1449  */
1450 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1451                                      handle_t *handle,
1452                                      struct inode *inode,
1453                                      const char *symname)
1454 {
1455         struct buffer_head **bhs = NULL;
1456         const char *c;
1457         struct super_block *sb = osb->sb;
1458         u64 p_blkno, p_blocks;
1459         int virtual, blocks, status, i, bytes_left;
1460
1461         bytes_left = i_size_read(inode) + 1;
1462         /* we can't trust i_blocks because we're actually going to
1463          * write i_size + 1 bytes. */
1464         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1465
1466         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1467                         (unsigned long long)inode->i_blocks,
1468                         i_size_read(inode), blocks);
1469
1470         /* Sanity check -- make sure we're going to fit. */
1471         if (bytes_left >
1472             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1473                 status = -EIO;
1474                 mlog_errno(status);
1475                 goto bail;
1476         }
1477
1478         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1479         if (!bhs) {
1480                 status = -ENOMEM;
1481                 mlog_errno(status);
1482                 goto bail;
1483         }
1484
1485         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1486                                              NULL);
1487         if (status < 0) {
1488                 mlog_errno(status);
1489                 goto bail;
1490         }
1491
1492         /* links can never be larger than one cluster so we know this
1493          * is all going to be contiguous, but do a sanity check
1494          * anyway. */
1495         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1496                 status = -EIO;
1497                 mlog_errno(status);
1498                 goto bail;
1499         }
1500
1501         virtual = 0;
1502         while(bytes_left > 0) {
1503                 c = &symname[virtual * sb->s_blocksize];
1504
1505                 bhs[virtual] = sb_getblk(sb, p_blkno);
1506                 if (!bhs[virtual]) {
1507                         status = -ENOMEM;
1508                         mlog_errno(status);
1509                         goto bail;
1510                 }
1511                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1512
1513                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1514                                               OCFS2_JOURNAL_ACCESS_CREATE);
1515                 if (status < 0) {
1516                         mlog_errno(status);
1517                         goto bail;
1518                 }
1519
1520                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1521
1522                 memcpy(bhs[virtual]->b_data, c,
1523                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1524                        bytes_left);
1525
1526                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1527                 if (status < 0) {
1528                         mlog_errno(status);
1529                         goto bail;
1530                 }
1531
1532                 virtual++;
1533                 p_blkno++;
1534                 bytes_left -= sb->s_blocksize;
1535         }
1536
1537         status = 0;
1538 bail:
1539
1540         if (bhs) {
1541                 for(i = 0; i < blocks; i++)
1542                         brelse(bhs[i]);
1543                 kfree(bhs);
1544         }
1545
1546         mlog_exit(status);
1547         return status;
1548 }
1549
1550 static int ocfs2_symlink(struct inode *dir,
1551                          struct dentry *dentry,
1552                          const char *symname)
1553 {
1554         int status, l, credits;
1555         u64 newsize;
1556         struct ocfs2_super *osb = NULL;
1557         struct inode *inode = NULL;
1558         struct super_block *sb;
1559         struct buffer_head *new_fe_bh = NULL;
1560         struct buffer_head *de_bh = NULL;
1561         struct buffer_head *parent_fe_bh = NULL;
1562         struct ocfs2_dinode *fe = NULL;
1563         struct ocfs2_dinode *dirfe;
1564         handle_t *handle = NULL;
1565         struct ocfs2_alloc_context *inode_ac = NULL;
1566         struct ocfs2_alloc_context *data_ac = NULL;
1567         struct ocfs2_alloc_context *xattr_ac = NULL;
1568         int want_clusters = 0;
1569         int xattr_credits = 0;
1570         struct ocfs2_security_xattr_info si = {
1571                 .enable = 1,
1572         };
1573         int did_quota = 0, did_quota_inode = 0;
1574
1575         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1576                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1577
1578         sb = dir->i_sb;
1579         osb = OCFS2_SB(sb);
1580
1581         l = strlen(symname) + 1;
1582
1583         credits = ocfs2_calc_symlink_credits(sb);
1584
1585         /* lock the parent directory */
1586         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1587         if (status < 0) {
1588                 if (status != -ENOENT)
1589                         mlog_errno(status);
1590                 return status;
1591         }
1592
1593         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1594         if (!dirfe->i_links_count) {
1595                 /* can't make a file in a deleted directory. */
1596                 status = -ENOENT;
1597                 goto bail;
1598         }
1599
1600         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1601                                            dentry->d_name.len);
1602         if (status)
1603                 goto bail;
1604
1605         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1606                                               dentry->d_name.name,
1607                                               dentry->d_name.len, &de_bh);
1608         if (status < 0) {
1609                 mlog_errno(status);
1610                 goto bail;
1611         }
1612
1613         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1614         if (status < 0) {
1615                 if (status != -ENOSPC)
1616                         mlog_errno(status);
1617                 goto bail;
1618         }
1619
1620         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1621         if (!inode) {
1622                 status = -ENOMEM;
1623                 mlog_errno(status);
1624                 goto bail;
1625         }
1626
1627         /* get security xattr */
1628         status = ocfs2_init_security_get(inode, dir, &si);
1629         if (status) {
1630                 if (status == -EOPNOTSUPP)
1631                         si.enable = 0;
1632                 else {
1633                         mlog_errno(status);
1634                         goto bail;
1635                 }
1636         }
1637
1638         /* calculate meta data/clusters for setting security xattr */
1639         if (si.enable) {
1640                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1641                                                   &xattr_credits, &xattr_ac);
1642                 if (status < 0) {
1643                         mlog_errno(status);
1644                         goto bail;
1645                 }
1646         }
1647
1648         /* don't reserve bitmap space for fast symlinks. */
1649         if (l > ocfs2_fast_symlink_chars(sb))
1650                 want_clusters += 1;
1651
1652         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1653         if (status < 0) {
1654                 if (status != -ENOSPC)
1655                         mlog_errno(status);
1656                 goto bail;
1657         }
1658
1659         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1660         if (IS_ERR(handle)) {
1661                 status = PTR_ERR(handle);
1662                 handle = NULL;
1663                 mlog_errno(status);
1664                 goto bail;
1665         }
1666
1667         /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1668          * to be called. */
1669         if (sb_any_quota_active(osb->sb) &&
1670             osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1671                 status = -EDQUOT;
1672                 goto bail;
1673         }
1674         did_quota_inode = 1;
1675
1676         status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1677                                     0, &new_fe_bh, parent_fe_bh, handle,
1678                                     inode_ac);
1679         if (status < 0) {
1680                 mlog_errno(status);
1681                 goto bail;
1682         }
1683
1684         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1685         inode->i_rdev = 0;
1686         newsize = l - 1;
1687         if (l > ocfs2_fast_symlink_chars(sb)) {
1688                 u32 offset = 0;
1689
1690                 inode->i_op = &ocfs2_symlink_inode_operations;
1691                 if (vfs_dq_alloc_space_nodirty(inode,
1692                     ocfs2_clusters_to_bytes(osb->sb, 1))) {
1693                         status = -EDQUOT;
1694                         goto bail;
1695                 }
1696                 did_quota = 1;
1697                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1698                                               new_fe_bh,
1699                                               handle, data_ac, NULL,
1700                                               NULL);
1701                 if (status < 0) {
1702                         if (status != -ENOSPC && status != -EINTR) {
1703                                 mlog(ML_ERROR,
1704                                      "Failed to extend file to %llu\n",
1705                                      (unsigned long long)newsize);
1706                                 mlog_errno(status);
1707                                 status = -ENOSPC;
1708                         }
1709                         goto bail;
1710                 }
1711                 i_size_write(inode, newsize);
1712                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1713         } else {
1714                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1715                 memcpy((char *) fe->id2.i_symlink, symname, l);
1716                 i_size_write(inode, newsize);
1717                 inode->i_blocks = 0;
1718         }
1719
1720         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1721         if (status < 0) {
1722                 mlog_errno(status);
1723                 goto bail;
1724         }
1725
1726         if (!ocfs2_inode_is_fast_symlink(inode)) {
1727                 status = ocfs2_create_symlink_data(osb, handle, inode,
1728                                                    symname);
1729                 if (status < 0) {
1730                         mlog_errno(status);
1731                         goto bail;
1732                 }
1733         }
1734
1735         if (si.enable) {
1736                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1737                                                  xattr_ac, data_ac);
1738                 if (status < 0) {
1739                         mlog_errno(status);
1740                         goto bail;
1741                 }
1742         }
1743
1744         status = ocfs2_add_entry(handle, dentry, inode,
1745                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1746                                  de_bh);
1747         if (status < 0) {
1748                 mlog_errno(status);
1749                 goto bail;
1750         }
1751
1752         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1753         if (status) {
1754                 mlog_errno(status);
1755                 goto bail;
1756         }
1757
1758         insert_inode_hash(inode);
1759         dentry->d_op = &ocfs2_dentry_ops;
1760         d_instantiate(dentry, inode);
1761 bail:
1762         if (status < 0 && did_quota)
1763                 vfs_dq_free_space_nodirty(inode,
1764                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1765         if (status < 0 && did_quota_inode)
1766                 vfs_dq_free_inode(inode);
1767         if (handle)
1768                 ocfs2_commit_trans(osb, handle);
1769
1770         ocfs2_inode_unlock(dir, 1);
1771
1772         brelse(new_fe_bh);
1773         brelse(parent_fe_bh);
1774         brelse(de_bh);
1775         kfree(si.name);
1776         kfree(si.value);
1777         if (inode_ac)
1778                 ocfs2_free_alloc_context(inode_ac);
1779         if (data_ac)
1780                 ocfs2_free_alloc_context(data_ac);
1781         if (xattr_ac)
1782                 ocfs2_free_alloc_context(xattr_ac);
1783         if ((status < 0) && inode) {
1784                 clear_nlink(inode);
1785                 iput(inode);
1786         }
1787
1788         mlog_exit(status);
1789
1790         return status;
1791 }
1792
1793 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1794 {
1795         int status, namelen;
1796
1797         mlog_entry_void();
1798
1799         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1800                            (long long)blkno);
1801         if (namelen <= 0) {
1802                 if (namelen)
1803                         status = namelen;
1804                 else
1805                         status = -EINVAL;
1806                 mlog_errno(status);
1807                 goto bail;
1808         }
1809         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1810                 status = -EINVAL;
1811                 mlog_errno(status);
1812                 goto bail;
1813         }
1814
1815         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1816              namelen);
1817
1818         status = 0;
1819 bail:
1820         mlog_exit(status);
1821         return status;
1822 }
1823
1824 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1825                                     struct inode **ret_orphan_dir,
1826                                     struct inode *inode,
1827                                     char *name,
1828                                     struct buffer_head **de_bh)
1829 {
1830         struct inode *orphan_dir_inode;
1831         struct buffer_head *orphan_dir_bh = NULL;
1832         int status = 0;
1833
1834         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1835         if (status < 0) {
1836                 mlog_errno(status);
1837                 return status;
1838         }
1839
1840         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1841                                                        ORPHAN_DIR_SYSTEM_INODE,
1842                                                        osb->slot_num);
1843         if (!orphan_dir_inode) {
1844                 status = -ENOENT;
1845                 mlog_errno(status);
1846                 return status;
1847         }
1848
1849         mutex_lock(&orphan_dir_inode->i_mutex);
1850
1851         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1852         if (status < 0) {
1853                 mlog_errno(status);
1854                 goto leave;
1855         }
1856
1857         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1858                                               orphan_dir_bh, name,
1859                                               OCFS2_ORPHAN_NAMELEN, de_bh);
1860         if (status < 0) {
1861                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1862
1863                 mlog_errno(status);
1864                 goto leave;
1865         }
1866
1867         *ret_orphan_dir = orphan_dir_inode;
1868
1869 leave:
1870         if (status) {
1871                 mutex_unlock(&orphan_dir_inode->i_mutex);
1872                 iput(orphan_dir_inode);
1873         }
1874
1875         brelse(orphan_dir_bh);
1876
1877         mlog_exit(status);
1878         return status;
1879 }
1880
1881 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1882                             handle_t *handle,
1883                             struct inode *inode,
1884                             struct ocfs2_dinode *fe,
1885                             char *name,
1886                             struct buffer_head *de_bh,
1887                             struct inode *orphan_dir_inode)
1888 {
1889         struct buffer_head *orphan_dir_bh = NULL;
1890         int status = 0;
1891         struct ocfs2_dinode *orphan_fe;
1892
1893         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1894
1895         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1896         if (status < 0) {
1897                 mlog_errno(status);
1898                 goto leave;
1899         }
1900
1901         status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1902                                          OCFS2_JOURNAL_ACCESS_WRITE);
1903         if (status < 0) {
1904                 mlog_errno(status);
1905                 goto leave;
1906         }
1907
1908         /* we're a cluster, and nlink can change on disk from
1909          * underneath us... */
1910         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1911         if (S_ISDIR(inode->i_mode))
1912                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1913         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1914
1915         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1916         if (status < 0) {
1917                 mlog_errno(status);
1918                 goto leave;
1919         }
1920
1921         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1922                                    OCFS2_ORPHAN_NAMELEN, inode,
1923                                    OCFS2_I(inode)->ip_blkno,
1924                                    orphan_dir_bh, de_bh);
1925         if (status < 0) {
1926                 mlog_errno(status);
1927                 goto leave;
1928         }
1929
1930         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1931
1932         /* Record which orphan dir our inode now resides
1933          * in. delete_inode will use this to determine which orphan
1934          * dir to lock. */
1935         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1936
1937         mlog(0, "Inode %llu orphaned in slot %d\n",
1938              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1939
1940 leave:
1941         brelse(orphan_dir_bh);
1942
1943         mlog_exit(status);
1944         return status;
1945 }
1946
1947 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1948 int ocfs2_orphan_del(struct ocfs2_super *osb,
1949                      handle_t *handle,
1950                      struct inode *orphan_dir_inode,
1951                      struct inode *inode,
1952                      struct buffer_head *orphan_dir_bh)
1953 {
1954         char name[OCFS2_ORPHAN_NAMELEN + 1];
1955         struct ocfs2_dinode *orphan_fe;
1956         int status = 0;
1957         struct buffer_head *target_de_bh = NULL;
1958         struct ocfs2_dir_entry *target_de = NULL;
1959
1960         mlog_entry_void();
1961
1962         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1963         if (status < 0) {
1964                 mlog_errno(status);
1965                 goto leave;
1966         }
1967
1968         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1969              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1970              OCFS2_ORPHAN_NAMELEN);
1971
1972         /* find it's spot in the orphan directory */
1973         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1974                                         orphan_dir_inode, &target_de);
1975         if (!target_de_bh) {
1976                 status = -ENOENT;
1977                 mlog_errno(status);
1978                 goto leave;
1979         }
1980
1981         /* remove it from the orphan directory */
1982         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1983                                     target_de_bh);
1984         if (status < 0) {
1985                 mlog_errno(status);
1986                 goto leave;
1987         }
1988
1989         status = ocfs2_journal_access_di(handle,orphan_dir_inode,  orphan_dir_bh,
1990                                          OCFS2_JOURNAL_ACCESS_WRITE);
1991         if (status < 0) {
1992                 mlog_errno(status);
1993                 goto leave;
1994         }
1995
1996         /* do the i_nlink dance! :) */
1997         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1998         if (S_ISDIR(inode->i_mode))
1999                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2000         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2001
2002         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2003         if (status < 0) {
2004                 mlog_errno(status);
2005                 goto leave;
2006         }
2007
2008 leave:
2009         brelse(target_de_bh);
2010
2011         mlog_exit(status);
2012         return status;
2013 }
2014
2015 const struct inode_operations ocfs2_dir_iops = {
2016         .create         = ocfs2_create,
2017         .lookup         = ocfs2_lookup,
2018         .link           = ocfs2_link,
2019         .unlink         = ocfs2_unlink,
2020         .rmdir          = ocfs2_unlink,
2021         .symlink        = ocfs2_symlink,
2022         .mkdir          = ocfs2_mkdir,
2023         .mknod          = ocfs2_mknod,
2024         .rename         = ocfs2_rename,
2025         .setattr        = ocfs2_setattr,
2026         .getattr        = ocfs2_getattr,
2027         .permission     = ocfs2_permission,
2028         .setxattr       = generic_setxattr,
2029         .getxattr       = generic_getxattr,
2030         .listxattr      = ocfs2_listxattr,
2031         .removexattr    = generic_removexattr,
2032 };