]> www.pilppa.org Git - linux-2.6-omap-h63xx.git/blob - fs/ocfs2/namei.c
OMAP: dmtimer: enable all timers to be wakeup events
[linux-2.6-omap-h63xx.git] / fs / ocfs2 / namei.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * namei.c
5  *
6  * Create and rename file, directory, symlinks
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  *  Portions of this code from linux/fs/ext3/dir.c
11  *
12  *  Copyright (C) 1992, 1993, 1994, 1995
13  *  Remy Card (card@masi.ibp.fr)
14  *  Laboratoire MASI - Institut Blaise pascal
15  *  Universite Pierre et Marie Curie (Paris VI)
16  *
17  *   from
18  *
19  *   linux/fs/minix/dir.c
20  *
21  *   Copyright (C) 1991, 1992 Linux Torvalds
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public
25  * License as published by the Free Software Foundation; either
26  * version 2 of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
31  * General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public
34  * License along with this program; if not, write to the
35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36  * Boston, MA 021110-1307, USA.
37  */
38
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
44
45 #define MLOG_MASK_PREFIX ML_NAMEI
46 #include <cluster/masklog.h>
47
48 #include "ocfs2.h"
49
50 #include "alloc.h"
51 #include "dcache.h"
52 #include "dir.h"
53 #include "dlmglue.h"
54 #include "extent_map.h"
55 #include "file.h"
56 #include "inode.h"
57 #include "journal.h"
58 #include "namei.h"
59 #include "suballoc.h"
60 #include "super.h"
61 #include "symlink.h"
62 #include "sysfile.h"
63 #include "uptodate.h"
64 #include "xattr.h"
65 #include "acl.h"
66
67 #include "buffer_head_io.h"
68
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
70                               struct inode *dir,
71                               struct inode *inode,
72                               struct dentry *dentry,
73                               dev_t dev,
74                               struct buffer_head **new_fe_bh,
75                               struct buffer_head *parent_fe_bh,
76                               handle_t *handle,
77                               struct ocfs2_alloc_context *inode_ac);
78
79 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
80                                     struct inode **ret_orphan_dir,
81                                     struct inode *inode,
82                                     char *name,
83                                     struct 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(
536                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
537         } else {
538                 fel = &fe->id2.i_list;
539                 fel->l_tree_depth = 0;
540                 fel->l_next_free_rec = 0;
541                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
542         }
543
544         status = ocfs2_journal_dirty(handle, *new_fe_bh);
545         if (status < 0) {
546                 mlog_errno(status);
547                 goto leave;
548         }
549
550         ocfs2_populate_inode(inode, fe, 1);
551         ocfs2_inode_set_new(osb, inode);
552         if (!ocfs2_mount_local(osb)) {
553                 status = ocfs2_create_new_inode_locks(inode);
554                 if (status < 0)
555                         mlog_errno(status);
556         }
557
558         status = 0; /* error in ocfs2_create_new_inode_locks is not
559                      * critical */
560
561 leave:
562         if (status < 0) {
563                 if (*new_fe_bh) {
564                         brelse(*new_fe_bh);
565                         *new_fe_bh = NULL;
566                 }
567         }
568
569         mlog_exit(status);
570         return status;
571 }
572
573 static int ocfs2_mkdir(struct inode *dir,
574                        struct dentry *dentry,
575                        int mode)
576 {
577         int ret;
578
579         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
580                    dentry->d_name.len, dentry->d_name.name);
581         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
582         mlog_exit(ret);
583
584         return ret;
585 }
586
587 static int ocfs2_create(struct inode *dir,
588                         struct dentry *dentry,
589                         int mode,
590                         struct nameidata *nd)
591 {
592         int ret;
593
594         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
595                    dentry->d_name.len, dentry->d_name.name);
596         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
597         mlog_exit(ret);
598
599         return ret;
600 }
601
602 static int ocfs2_link(struct dentry *old_dentry,
603                       struct inode *dir,
604                       struct dentry *dentry)
605 {
606         handle_t *handle;
607         struct inode *inode = old_dentry->d_inode;
608         int err;
609         struct buffer_head *fe_bh = NULL;
610         struct buffer_head *parent_fe_bh = NULL;
611         struct buffer_head *de_bh = NULL;
612         struct ocfs2_dinode *fe = NULL;
613         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
614
615         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
616                    old_dentry->d_name.len, old_dentry->d_name.name,
617                    dentry->d_name.len, dentry->d_name.name);
618
619         if (S_ISDIR(inode->i_mode))
620                 return -EPERM;
621
622         err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
623         if (err < 0) {
624                 if (err != -ENOENT)
625                         mlog_errno(err);
626                 return err;
627         }
628
629         if (!dir->i_nlink) {
630                 err = -ENOENT;
631                 goto out;
632         }
633
634         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
635                                         dentry->d_name.len);
636         if (err)
637                 goto out;
638
639         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
640                                            dentry->d_name.name,
641                                            dentry->d_name.len, &de_bh);
642         if (err < 0) {
643                 mlog_errno(err);
644                 goto out;
645         }
646
647         err = ocfs2_inode_lock(inode, &fe_bh, 1);
648         if (err < 0) {
649                 if (err != -ENOENT)
650                         mlog_errno(err);
651                 goto out;
652         }
653
654         fe = (struct ocfs2_dinode *) fe_bh->b_data;
655         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
656                 err = -EMLINK;
657                 goto out_unlock_inode;
658         }
659
660         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
661         if (IS_ERR(handle)) {
662                 err = PTR_ERR(handle);
663                 handle = NULL;
664                 mlog_errno(err);
665                 goto out_unlock_inode;
666         }
667
668         err = ocfs2_journal_access_di(handle, inode, fe_bh,
669                                       OCFS2_JOURNAL_ACCESS_WRITE);
670         if (err < 0) {
671                 mlog_errno(err);
672                 goto out_commit;
673         }
674
675         inc_nlink(inode);
676         inode->i_ctime = CURRENT_TIME;
677         fe->i_links_count = cpu_to_le16(inode->i_nlink);
678         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
679         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
680
681         err = ocfs2_journal_dirty(handle, fe_bh);
682         if (err < 0) {
683                 le16_add_cpu(&fe->i_links_count, -1);
684                 drop_nlink(inode);
685                 mlog_errno(err);
686                 goto out_commit;
687         }
688
689         err = ocfs2_add_entry(handle, dentry, inode,
690                               OCFS2_I(inode)->ip_blkno,
691                               parent_fe_bh, de_bh);
692         if (err) {
693                 le16_add_cpu(&fe->i_links_count, -1);
694                 drop_nlink(inode);
695                 mlog_errno(err);
696                 goto out_commit;
697         }
698
699         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
700         if (err) {
701                 mlog_errno(err);
702                 goto out_commit;
703         }
704
705         atomic_inc(&inode->i_count);
706         dentry->d_op = &ocfs2_dentry_ops;
707         d_instantiate(dentry, inode);
708
709 out_commit:
710         ocfs2_commit_trans(osb, handle);
711 out_unlock_inode:
712         ocfs2_inode_unlock(inode, 1);
713
714 out:
715         ocfs2_inode_unlock(dir, 1);
716
717         brelse(de_bh);
718         brelse(fe_bh);
719         brelse(parent_fe_bh);
720
721         mlog_exit(err);
722
723         return err;
724 }
725
726 /*
727  * Takes and drops an exclusive lock on the given dentry. This will
728  * force other nodes to drop it.
729  */
730 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
731 {
732         int ret;
733
734         ret = ocfs2_dentry_lock(dentry, 1);
735         if (ret)
736                 mlog_errno(ret);
737         else
738                 ocfs2_dentry_unlock(dentry, 1);
739
740         return ret;
741 }
742
743 static inline int inode_is_unlinkable(struct inode *inode)
744 {
745         if (S_ISDIR(inode->i_mode)) {
746                 if (inode->i_nlink == 2)
747                         return 1;
748                 return 0;
749         }
750
751         if (inode->i_nlink == 1)
752                 return 1;
753         return 0;
754 }
755
756 static int ocfs2_unlink(struct inode *dir,
757                         struct dentry *dentry)
758 {
759         int status;
760         int child_locked = 0;
761         struct inode *inode = dentry->d_inode;
762         struct inode *orphan_dir = NULL;
763         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
764         u64 blkno;
765         struct ocfs2_dinode *fe = NULL;
766         struct buffer_head *fe_bh = NULL;
767         struct buffer_head *parent_node_bh = NULL;
768         handle_t *handle = NULL;
769         struct ocfs2_dir_entry *dirent = NULL;
770         struct buffer_head *dirent_bh = NULL;
771         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
772         struct buffer_head *orphan_entry_bh = NULL;
773
774         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
775                    dentry->d_name.len, dentry->d_name.name);
776
777         BUG_ON(dentry->d_parent->d_inode != dir);
778
779         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
780
781         if (inode == osb->root_inode) {
782                 mlog(0, "Cannot delete the root directory\n");
783                 return -EPERM;
784         }
785
786         status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
787         if (status < 0) {
788                 if (status != -ENOENT)
789                         mlog_errno(status);
790                 return status;
791         }
792
793         status = ocfs2_find_files_on_disk(dentry->d_name.name,
794                                           dentry->d_name.len, &blkno,
795                                           dir, &dirent_bh, &dirent);
796         if (status < 0) {
797                 if (status != -ENOENT)
798                         mlog_errno(status);
799                 goto leave;
800         }
801
802         if (OCFS2_I(inode)->ip_blkno != blkno) {
803                 status = -ENOENT;
804
805                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
806                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
807                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
808                 goto leave;
809         }
810
811         status = ocfs2_inode_lock(inode, &fe_bh, 1);
812         if (status < 0) {
813                 if (status != -ENOENT)
814                         mlog_errno(status);
815                 goto leave;
816         }
817         child_locked = 1;
818
819         if (S_ISDIR(inode->i_mode)) {
820                 if (!ocfs2_empty_dir(inode)) {
821                         status = -ENOTEMPTY;
822                         goto leave;
823                 } else if (inode->i_nlink != 2) {
824                         status = -ENOTEMPTY;
825                         goto leave;
826                 }
827         }
828
829         status = ocfs2_remote_dentry_delete(dentry);
830         if (status < 0) {
831                 /* This remote delete should succeed under all normal
832                  * circumstances. */
833                 mlog_errno(status);
834                 goto leave;
835         }
836
837         if (inode_is_unlinkable(inode)) {
838                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
839                                                   orphan_name,
840                                                   &orphan_entry_bh);
841                 if (status < 0) {
842                         mlog_errno(status);
843                         goto leave;
844                 }
845         }
846
847         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
848         if (IS_ERR(handle)) {
849                 status = PTR_ERR(handle);
850                 handle = NULL;
851                 mlog_errno(status);
852                 goto leave;
853         }
854
855         status = ocfs2_journal_access_di(handle, inode, fe_bh,
856                                          OCFS2_JOURNAL_ACCESS_WRITE);
857         if (status < 0) {
858                 mlog_errno(status);
859                 goto leave;
860         }
861
862         fe = (struct ocfs2_dinode *) fe_bh->b_data;
863
864         if (inode_is_unlinkable(inode)) {
865                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
866                                           orphan_entry_bh, orphan_dir);
867                 if (status < 0) {
868                         mlog_errno(status);
869                         goto leave;
870                 }
871         }
872
873         /* delete the name from the parent dir */
874         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
875         if (status < 0) {
876                 mlog_errno(status);
877                 goto leave;
878         }
879
880         if (S_ISDIR(inode->i_mode))
881                 drop_nlink(inode);
882         drop_nlink(inode);
883         fe->i_links_count = cpu_to_le16(inode->i_nlink);
884
885         status = ocfs2_journal_dirty(handle, fe_bh);
886         if (status < 0) {
887                 mlog_errno(status);
888                 goto leave;
889         }
890
891         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
892         if (S_ISDIR(inode->i_mode))
893                 drop_nlink(dir);
894
895         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
896         if (status < 0) {
897                 mlog_errno(status);
898                 if (S_ISDIR(inode->i_mode))
899                         inc_nlink(dir);
900         }
901
902 leave:
903         if (handle)
904                 ocfs2_commit_trans(osb, handle);
905
906         if (child_locked)
907                 ocfs2_inode_unlock(inode, 1);
908
909         ocfs2_inode_unlock(dir, 1);
910
911         if (orphan_dir) {
912                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
913                 ocfs2_inode_unlock(orphan_dir, 1);
914                 mutex_unlock(&orphan_dir->i_mutex);
915                 iput(orphan_dir);
916         }
917
918         brelse(fe_bh);
919         brelse(dirent_bh);
920         brelse(parent_node_bh);
921         brelse(orphan_entry_bh);
922
923         mlog_exit(status);
924
925         return status;
926 }
927
928 /*
929  * The only place this should be used is rename!
930  * if they have the same id, then the 1st one is the only one locked.
931  */
932 static int ocfs2_double_lock(struct ocfs2_super *osb,
933                              struct buffer_head **bh1,
934                              struct inode *inode1,
935                              struct buffer_head **bh2,
936                              struct inode *inode2)
937 {
938         int status;
939         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
940         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
941         struct buffer_head **tmpbh;
942         struct inode *tmpinode;
943
944         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
945                    (unsigned long long)oi1->ip_blkno,
946                    (unsigned long long)oi2->ip_blkno);
947
948         if (*bh1)
949                 *bh1 = NULL;
950         if (*bh2)
951                 *bh2 = NULL;
952
953         /* we always want to lock the one with the lower lockid first. */
954         if (oi1->ip_blkno != oi2->ip_blkno) {
955                 if (oi1->ip_blkno < oi2->ip_blkno) {
956                         /* switch id1 and id2 around */
957                         mlog(0, "switching them around...\n");
958                         tmpbh = bh2;
959                         bh2 = bh1;
960                         bh1 = tmpbh;
961
962                         tmpinode = inode2;
963                         inode2 = inode1;
964                         inode1 = tmpinode;
965                 }
966                 /* lock id2 */
967                 status = ocfs2_inode_lock(inode2, bh2, 1);
968                 if (status < 0) {
969                         if (status != -ENOENT)
970                                 mlog_errno(status);
971                         goto bail;
972                 }
973         }
974
975         /* lock id1 */
976         status = ocfs2_inode_lock(inode1, bh1, 1);
977         if (status < 0) {
978                 /*
979                  * An error return must mean that no cluster locks
980                  * were held on function exit.
981                  */
982                 if (oi1->ip_blkno != oi2->ip_blkno)
983                         ocfs2_inode_unlock(inode2, 1);
984
985                 if (status != -ENOENT)
986                         mlog_errno(status);
987         }
988
989 bail:
990         mlog_exit(status);
991         return status;
992 }
993
994 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
995 {
996         ocfs2_inode_unlock(inode1, 1);
997
998         if (inode1 != inode2)
999                 ocfs2_inode_unlock(inode2, 1);
1000 }
1001
1002 static int ocfs2_rename(struct inode *old_dir,
1003                         struct dentry *old_dentry,
1004                         struct inode *new_dir,
1005                         struct dentry *new_dentry)
1006 {
1007         int status = 0, rename_lock = 0, parents_locked = 0;
1008         int old_child_locked = 0, new_child_locked = 0;
1009         struct inode *old_inode = old_dentry->d_inode;
1010         struct inode *new_inode = new_dentry->d_inode;
1011         struct inode *orphan_dir = NULL;
1012         struct ocfs2_dinode *newfe = NULL;
1013         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1014         struct buffer_head *orphan_entry_bh = NULL;
1015         struct buffer_head *newfe_bh = NULL;
1016         struct buffer_head *old_inode_bh = NULL;
1017         struct buffer_head *insert_entry_bh = NULL;
1018         struct ocfs2_super *osb = NULL;
1019         u64 newfe_blkno, old_de_ino;
1020         handle_t *handle = NULL;
1021         struct buffer_head *old_dir_bh = NULL;
1022         struct buffer_head *new_dir_bh = NULL;
1023         struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
1024                 *new_de = NULL;
1025         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1026         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1027                                                     // this is the 1st dirent bh
1028         nlink_t old_dir_nlink = old_dir->i_nlink;
1029         struct ocfs2_dinode *old_di;
1030
1031         /* At some point it might be nice to break this function up a
1032          * bit. */
1033
1034         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1035                    old_dir, old_dentry, new_dir, new_dentry,
1036                    old_dentry->d_name.len, old_dentry->d_name.name,
1037                    new_dentry->d_name.len, new_dentry->d_name.name);
1038
1039         osb = OCFS2_SB(old_dir->i_sb);
1040
1041         if (new_inode) {
1042                 if (!igrab(new_inode))
1043                         BUG();
1044         }
1045
1046         /* Assume a directory hierarchy thusly:
1047          * a/b/c
1048          * a/d
1049          * a,b,c, and d are all directories.
1050          *
1051          * from cwd of 'a' on both nodes:
1052          * node1: mv b/c d
1053          * node2: mv d   b/c
1054          *
1055          * And that's why, just like the VFS, we need a file system
1056          * rename lock. */
1057         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1058                 status = ocfs2_rename_lock(osb);
1059                 if (status < 0) {
1060                         mlog_errno(status);
1061                         goto bail;
1062                 }
1063                 rename_lock = 1;
1064         }
1065
1066         /* if old and new are the same, this'll just do one lock. */
1067         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1068                                    &new_dir_bh, new_dir);
1069         if (status < 0) {
1070                 mlog_errno(status);
1071                 goto bail;
1072         }
1073         parents_locked = 1;
1074
1075         /* make sure both dirs have bhs
1076          * get an extra ref on old_dir_bh if old==new */
1077         if (!new_dir_bh) {
1078                 if (old_dir_bh) {
1079                         new_dir_bh = old_dir_bh;
1080                         get_bh(new_dir_bh);
1081                 } else {
1082                         mlog(ML_ERROR, "no old_dir_bh!\n");
1083                         status = -EIO;
1084                         goto bail;
1085                 }
1086         }
1087
1088         /*
1089          * Aside from allowing a meta data update, the locking here
1090          * also ensures that the downconvert thread on other nodes
1091          * won't have to concurrently downconvert the inode and the
1092          * dentry locks.
1093          */
1094         status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1095         if (status < 0) {
1096                 if (status != -ENOENT)
1097                         mlog_errno(status);
1098                 goto bail;
1099         }
1100         old_child_locked = 1;
1101
1102         status = ocfs2_remote_dentry_delete(old_dentry);
1103         if (status < 0) {
1104                 mlog_errno(status);
1105                 goto bail;
1106         }
1107
1108         if (S_ISDIR(old_inode->i_mode)) {
1109                 u64 old_inode_parent;
1110
1111                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1112                                                   old_inode, &old_inode_de_bh,
1113                                                   &old_inode_dot_dot_de);
1114                 if (status) {
1115                         status = -EIO;
1116                         goto bail;
1117                 }
1118
1119                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1120                         status = -EIO;
1121                         goto bail;
1122                 }
1123
1124                 if (!new_inode && new_dir != old_dir &&
1125                     new_dir->i_nlink >= OCFS2_LINK_MAX) {
1126                         status = -EMLINK;
1127                         goto bail;
1128                 }
1129         }
1130
1131         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1132                                             old_dentry->d_name.len,
1133                                             &old_de_ino);
1134         if (status) {
1135                 status = -ENOENT;
1136                 goto bail;
1137         }
1138
1139         /*
1140          *  Check for inode number is _not_ due to possible IO errors.
1141          *  We might rmdir the source, keep it as pwd of some process
1142          *  and merrily kill the link to whatever was created under the
1143          *  same name. Goodbye sticky bit ;-<
1144          */
1145         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1146                 status = -ENOENT;
1147                 goto bail;
1148         }
1149
1150         /* check if the target already exists (in which case we need
1151          * to delete it */
1152         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1153                                           new_dentry->d_name.len,
1154                                           &newfe_blkno, new_dir, &new_de_bh,
1155                                           &new_de);
1156         /* The only error we allow here is -ENOENT because the new
1157          * file not existing is perfectly valid. */
1158         if ((status < 0) && (status != -ENOENT)) {
1159                 /* If we cannot find the file specified we should just */
1160                 /* return the error... */
1161                 mlog_errno(status);
1162                 goto bail;
1163         }
1164
1165         if (!new_de && new_inode) {
1166                 /*
1167                  * Target was unlinked by another node while we were
1168                  * waiting to get to ocfs2_rename(). There isn't
1169                  * anything we can do here to help the situation, so
1170                  * bubble up the appropriate error.
1171                  */
1172                 status = -ENOENT;
1173                 goto bail;
1174         }
1175
1176         /* In case we need to overwrite an existing file, we blow it
1177          * away first */
1178         if (new_de) {
1179                 /* VFS didn't think there existed an inode here, but
1180                  * someone else in the cluster must have raced our
1181                  * rename to create one. Today we error cleanly, in
1182                  * the future we should consider calling iget to build
1183                  * a new struct inode for this entry. */
1184                 if (!new_inode) {
1185                         status = -EACCES;
1186
1187                         mlog(0, "We found an inode for name %.*s but VFS "
1188                              "didn't give us one.\n", new_dentry->d_name.len,
1189                              new_dentry->d_name.name);
1190                         goto bail;
1191                 }
1192
1193                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1194                         status = -EACCES;
1195
1196                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1197                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1198                              (unsigned long long)newfe_blkno,
1199                              OCFS2_I(new_inode)->ip_flags);
1200                         goto bail;
1201                 }
1202
1203                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1204                 if (status < 0) {
1205                         if (status != -ENOENT)
1206                                 mlog_errno(status);
1207                         goto bail;
1208                 }
1209                 new_child_locked = 1;
1210
1211                 status = ocfs2_remote_dentry_delete(new_dentry);
1212                 if (status < 0) {
1213                         mlog_errno(status);
1214                         goto bail;
1215                 }
1216
1217                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1218
1219                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1220                      "newfebh=%p bhblocknr=%llu\n", new_de,
1221                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1222                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1223
1224                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1225                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1226                                                           new_inode,
1227                                                           orphan_name,
1228                                                           &orphan_entry_bh);
1229                         if (status < 0) {
1230                                 mlog_errno(status);
1231                                 goto bail;
1232                         }
1233                 }
1234         } else {
1235                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1236
1237                 status = ocfs2_check_dir_for_entry(new_dir,
1238                                                    new_dentry->d_name.name,
1239                                                    new_dentry->d_name.len);
1240                 if (status)
1241                         goto bail;
1242
1243                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1244                                                       new_dentry->d_name.name,
1245                                                       new_dentry->d_name.len,
1246                                                       &insert_entry_bh);
1247                 if (status < 0) {
1248                         mlog_errno(status);
1249                         goto bail;
1250                 }
1251         }
1252
1253         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1254         if (IS_ERR(handle)) {
1255                 status = PTR_ERR(handle);
1256                 handle = NULL;
1257                 mlog_errno(status);
1258                 goto bail;
1259         }
1260
1261         if (new_de) {
1262                 if (S_ISDIR(new_inode->i_mode)) {
1263                         if (!ocfs2_empty_dir(new_inode) ||
1264                             new_inode->i_nlink != 2) {
1265                                 status = -ENOTEMPTY;
1266                                 goto bail;
1267                         }
1268                 }
1269                 status = ocfs2_journal_access_di(handle, new_inode, newfe_bh,
1270                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1271                 if (status < 0) {
1272                         mlog_errno(status);
1273                         goto bail;
1274                 }
1275
1276                 if (S_ISDIR(new_inode->i_mode) ||
1277                     (newfe->i_links_count == cpu_to_le16(1))){
1278                         status = ocfs2_orphan_add(osb, handle, new_inode,
1279                                                   newfe, orphan_name,
1280                                                   orphan_entry_bh, orphan_dir);
1281                         if (status < 0) {
1282                                 mlog_errno(status);
1283                                 goto bail;
1284                         }
1285                 }
1286
1287                 /* change the dirent to point to the correct inode */
1288                 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1289                                             new_de, old_inode);
1290                 if (status < 0) {
1291                         mlog_errno(status);
1292                         goto bail;
1293                 }
1294                 new_dir->i_version++;
1295
1296                 if (S_ISDIR(new_inode->i_mode))
1297                         newfe->i_links_count = 0;
1298                 else
1299                         le16_add_cpu(&newfe->i_links_count, -1);
1300
1301                 status = ocfs2_journal_dirty(handle, newfe_bh);
1302                 if (status < 0) {
1303                         mlog_errno(status);
1304                         goto bail;
1305                 }
1306         } else {
1307                 /* if the name was not found in new_dir, add it now */
1308                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1309                                          OCFS2_I(old_inode)->ip_blkno,
1310                                          new_dir_bh, insert_entry_bh);
1311         }
1312
1313         old_inode->i_ctime = CURRENT_TIME;
1314         mark_inode_dirty(old_inode);
1315
1316         status = ocfs2_journal_access_di(handle, old_inode, old_inode_bh,
1317                                          OCFS2_JOURNAL_ACCESS_WRITE);
1318         if (status >= 0) {
1319                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1320
1321                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1322                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1323
1324                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1325                 if (status < 0)
1326                         mlog_errno(status);
1327         } else
1328                 mlog_errno(status);
1329
1330         /*
1331          * Now that the name has been added to new_dir, remove the old name.
1332          *
1333          * We don't keep any directory entry context around until now
1334          * because the insert might have changed the type of directory
1335          * we're dealing with.
1336          */
1337         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1338                                      old_dentry->d_name.len,
1339                                      old_dir, &old_de);
1340         if (!old_de_bh) {
1341                 status = -EIO;
1342                 goto bail;
1343         }
1344
1345         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1346         if (status < 0) {
1347                 mlog_errno(status);
1348                 goto bail;
1349         }
1350
1351         if (new_inode) {
1352                 new_inode->i_nlink--;
1353                 new_inode->i_ctime = CURRENT_TIME;
1354         }
1355         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1356         if (old_inode_de_bh) {
1357                 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1358                                             old_inode_dot_dot_de, new_dir);
1359                 old_dir->i_nlink--;
1360                 if (new_inode) {
1361                         new_inode->i_nlink--;
1362                 } else {
1363                         inc_nlink(new_dir);
1364                         mark_inode_dirty(new_dir);
1365                 }
1366         }
1367         mark_inode_dirty(old_dir);
1368         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1369         if (new_inode) {
1370                 mark_inode_dirty(new_inode);
1371                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1372         }
1373
1374         if (old_dir != new_dir) {
1375                 /* Keep the same times on both directories.*/
1376                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1377
1378                 /*
1379                  * This will also pick up the i_nlink change from the
1380                  * block above.
1381                  */
1382                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1383         }
1384
1385         if (old_dir_nlink != old_dir->i_nlink) {
1386                 if (!old_dir_bh) {
1387                         mlog(ML_ERROR, "need to change nlink for old dir "
1388                              "%llu from %d to %d but bh is NULL!\n",
1389                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1390                              (int)old_dir_nlink, old_dir->i_nlink);
1391                 } else {
1392                         struct ocfs2_dinode *fe;
1393                         status = ocfs2_journal_access_di(handle, old_dir,
1394                                                          old_dir_bh,
1395                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1396                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1397                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1398                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1399                 }
1400         }
1401
1402         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1403         status = 0;
1404 bail:
1405         if (rename_lock)
1406                 ocfs2_rename_unlock(osb);
1407
1408         if (handle)
1409                 ocfs2_commit_trans(osb, handle);
1410
1411         if (parents_locked)
1412                 ocfs2_double_unlock(old_dir, new_dir);
1413
1414         if (old_child_locked)
1415                 ocfs2_inode_unlock(old_inode, 1);
1416
1417         if (new_child_locked)
1418                 ocfs2_inode_unlock(new_inode, 1);
1419
1420         if (orphan_dir) {
1421                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1422                 ocfs2_inode_unlock(orphan_dir, 1);
1423                 mutex_unlock(&orphan_dir->i_mutex);
1424                 iput(orphan_dir);
1425         }
1426
1427         if (new_inode)
1428                 sync_mapping_buffers(old_inode->i_mapping);
1429
1430         if (new_inode)
1431                 iput(new_inode);
1432         brelse(newfe_bh);
1433         brelse(old_inode_bh);
1434         brelse(old_dir_bh);
1435         brelse(new_dir_bh);
1436         brelse(new_de_bh);
1437         brelse(old_de_bh);
1438         brelse(old_inode_de_bh);
1439         brelse(orphan_entry_bh);
1440         brelse(insert_entry_bh);
1441
1442         mlog_exit(status);
1443
1444         return status;
1445 }
1446
1447 /*
1448  * we expect i_size = strlen(symname). Copy symname into the file
1449  * data, including the null terminator.
1450  */
1451 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1452                                      handle_t *handle,
1453                                      struct inode *inode,
1454                                      const char *symname)
1455 {
1456         struct buffer_head **bhs = NULL;
1457         const char *c;
1458         struct super_block *sb = osb->sb;
1459         u64 p_blkno, p_blocks;
1460         int virtual, blocks, status, i, bytes_left;
1461
1462         bytes_left = i_size_read(inode) + 1;
1463         /* we can't trust i_blocks because we're actually going to
1464          * write i_size + 1 bytes. */
1465         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1466
1467         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1468                         (unsigned long long)inode->i_blocks,
1469                         i_size_read(inode), blocks);
1470
1471         /* Sanity check -- make sure we're going to fit. */
1472         if (bytes_left >
1473             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1474                 status = -EIO;
1475                 mlog_errno(status);
1476                 goto bail;
1477         }
1478
1479         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1480         if (!bhs) {
1481                 status = -ENOMEM;
1482                 mlog_errno(status);
1483                 goto bail;
1484         }
1485
1486         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1487                                              NULL);
1488         if (status < 0) {
1489                 mlog_errno(status);
1490                 goto bail;
1491         }
1492
1493         /* links can never be larger than one cluster so we know this
1494          * is all going to be contiguous, but do a sanity check
1495          * anyway. */
1496         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1497                 status = -EIO;
1498                 mlog_errno(status);
1499                 goto bail;
1500         }
1501
1502         virtual = 0;
1503         while(bytes_left > 0) {
1504                 c = &symname[virtual * sb->s_blocksize];
1505
1506                 bhs[virtual] = sb_getblk(sb, p_blkno);
1507                 if (!bhs[virtual]) {
1508                         status = -ENOMEM;
1509                         mlog_errno(status);
1510                         goto bail;
1511                 }
1512                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1513
1514                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1515                                               OCFS2_JOURNAL_ACCESS_CREATE);
1516                 if (status < 0) {
1517                         mlog_errno(status);
1518                         goto bail;
1519                 }
1520
1521                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1522
1523                 memcpy(bhs[virtual]->b_data, c,
1524                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1525                        bytes_left);
1526
1527                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1528                 if (status < 0) {
1529                         mlog_errno(status);
1530                         goto bail;
1531                 }
1532
1533                 virtual++;
1534                 p_blkno++;
1535                 bytes_left -= sb->s_blocksize;
1536         }
1537
1538         status = 0;
1539 bail:
1540
1541         if (bhs) {
1542                 for(i = 0; i < blocks; i++)
1543                         brelse(bhs[i]);
1544                 kfree(bhs);
1545         }
1546
1547         mlog_exit(status);
1548         return status;
1549 }
1550
1551 static int ocfs2_symlink(struct inode *dir,
1552                          struct dentry *dentry,
1553                          const char *symname)
1554 {
1555         int status, l, credits;
1556         u64 newsize;
1557         struct ocfs2_super *osb = NULL;
1558         struct inode *inode = NULL;
1559         struct super_block *sb;
1560         struct buffer_head *new_fe_bh = NULL;
1561         struct buffer_head *de_bh = NULL;
1562         struct buffer_head *parent_fe_bh = NULL;
1563         struct ocfs2_dinode *fe = NULL;
1564         struct ocfs2_dinode *dirfe;
1565         handle_t *handle = NULL;
1566         struct ocfs2_alloc_context *inode_ac = NULL;
1567         struct ocfs2_alloc_context *data_ac = NULL;
1568         struct ocfs2_alloc_context *xattr_ac = NULL;
1569         int want_clusters = 0;
1570         int xattr_credits = 0;
1571         struct ocfs2_security_xattr_info si = {
1572                 .enable = 1,
1573         };
1574         int did_quota = 0, did_quota_inode = 0;
1575
1576         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1577                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1578
1579         sb = dir->i_sb;
1580         osb = OCFS2_SB(sb);
1581
1582         l = strlen(symname) + 1;
1583
1584         credits = ocfs2_calc_symlink_credits(sb);
1585
1586         /* lock the parent directory */
1587         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1588         if (status < 0) {
1589                 if (status != -ENOENT)
1590                         mlog_errno(status);
1591                 return status;
1592         }
1593
1594         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1595         if (!dirfe->i_links_count) {
1596                 /* can't make a file in a deleted directory. */
1597                 status = -ENOENT;
1598                 goto bail;
1599         }
1600
1601         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1602                                            dentry->d_name.len);
1603         if (status)
1604                 goto bail;
1605
1606         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1607                                               dentry->d_name.name,
1608                                               dentry->d_name.len, &de_bh);
1609         if (status < 0) {
1610                 mlog_errno(status);
1611                 goto bail;
1612         }
1613
1614         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1615         if (status < 0) {
1616                 if (status != -ENOSPC)
1617                         mlog_errno(status);
1618                 goto bail;
1619         }
1620
1621         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1622         if (!inode) {
1623                 status = -ENOMEM;
1624                 mlog_errno(status);
1625                 goto bail;
1626         }
1627
1628         /* get security xattr */
1629         status = ocfs2_init_security_get(inode, dir, &si);
1630         if (status) {
1631                 if (status == -EOPNOTSUPP)
1632                         si.enable = 0;
1633                 else {
1634                         mlog_errno(status);
1635                         goto bail;
1636                 }
1637         }
1638
1639         /* calculate meta data/clusters for setting security xattr */
1640         if (si.enable) {
1641                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1642                                                   &xattr_credits, &xattr_ac);
1643                 if (status < 0) {
1644                         mlog_errno(status);
1645                         goto bail;
1646                 }
1647         }
1648
1649         /* don't reserve bitmap space for fast symlinks. */
1650         if (l > ocfs2_fast_symlink_chars(sb))
1651                 want_clusters += 1;
1652
1653         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1654         if (status < 0) {
1655                 if (status != -ENOSPC)
1656                         mlog_errno(status);
1657                 goto bail;
1658         }
1659
1660         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1661         if (IS_ERR(handle)) {
1662                 status = PTR_ERR(handle);
1663                 handle = NULL;
1664                 mlog_errno(status);
1665                 goto bail;
1666         }
1667
1668         /* We don't use standard VFS wrapper because we don't want vfs_dq_init
1669          * to be called. */
1670         if (sb_any_quota_active(osb->sb) &&
1671             osb->sb->dq_op->alloc_inode(inode, 1) == NO_QUOTA) {
1672                 status = -EDQUOT;
1673                 goto bail;
1674         }
1675         did_quota_inode = 1;
1676
1677         status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1678                                     0, &new_fe_bh, parent_fe_bh, handle,
1679                                     inode_ac);
1680         if (status < 0) {
1681                 mlog_errno(status);
1682                 goto bail;
1683         }
1684
1685         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1686         inode->i_rdev = 0;
1687         newsize = l - 1;
1688         if (l > ocfs2_fast_symlink_chars(sb)) {
1689                 u32 offset = 0;
1690
1691                 inode->i_op = &ocfs2_symlink_inode_operations;
1692                 if (vfs_dq_alloc_space_nodirty(inode,
1693                     ocfs2_clusters_to_bytes(osb->sb, 1))) {
1694                         status = -EDQUOT;
1695                         goto bail;
1696                 }
1697                 did_quota = 1;
1698                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1699                                               new_fe_bh,
1700                                               handle, data_ac, NULL,
1701                                               NULL);
1702                 if (status < 0) {
1703                         if (status != -ENOSPC && status != -EINTR) {
1704                                 mlog(ML_ERROR,
1705                                      "Failed to extend file to %llu\n",
1706                                      (unsigned long long)newsize);
1707                                 mlog_errno(status);
1708                                 status = -ENOSPC;
1709                         }
1710                         goto bail;
1711                 }
1712                 i_size_write(inode, newsize);
1713                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1714         } else {
1715                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1716                 memcpy((char *) fe->id2.i_symlink, symname, l);
1717                 i_size_write(inode, newsize);
1718                 inode->i_blocks = 0;
1719         }
1720
1721         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1722         if (status < 0) {
1723                 mlog_errno(status);
1724                 goto bail;
1725         }
1726
1727         if (!ocfs2_inode_is_fast_symlink(inode)) {
1728                 status = ocfs2_create_symlink_data(osb, handle, inode,
1729                                                    symname);
1730                 if (status < 0) {
1731                         mlog_errno(status);
1732                         goto bail;
1733                 }
1734         }
1735
1736         if (si.enable) {
1737                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1738                                                  xattr_ac, data_ac);
1739                 if (status < 0) {
1740                         mlog_errno(status);
1741                         goto bail;
1742                 }
1743         }
1744
1745         status = ocfs2_add_entry(handle, dentry, inode,
1746                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1747                                  de_bh);
1748         if (status < 0) {
1749                 mlog_errno(status);
1750                 goto bail;
1751         }
1752
1753         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1754         if (status) {
1755                 mlog_errno(status);
1756                 goto bail;
1757         }
1758
1759         insert_inode_hash(inode);
1760         dentry->d_op = &ocfs2_dentry_ops;
1761         d_instantiate(dentry, inode);
1762 bail:
1763         if (status < 0 && did_quota)
1764                 vfs_dq_free_space_nodirty(inode,
1765                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1766         if (status < 0 && did_quota_inode)
1767                 vfs_dq_free_inode(inode);
1768         if (handle)
1769                 ocfs2_commit_trans(osb, handle);
1770
1771         ocfs2_inode_unlock(dir, 1);
1772
1773         brelse(new_fe_bh);
1774         brelse(parent_fe_bh);
1775         brelse(de_bh);
1776         kfree(si.name);
1777         kfree(si.value);
1778         if (inode_ac)
1779                 ocfs2_free_alloc_context(inode_ac);
1780         if (data_ac)
1781                 ocfs2_free_alloc_context(data_ac);
1782         if (xattr_ac)
1783                 ocfs2_free_alloc_context(xattr_ac);
1784         if ((status < 0) && inode) {
1785                 clear_nlink(inode);
1786                 iput(inode);
1787         }
1788
1789         mlog_exit(status);
1790
1791         return status;
1792 }
1793
1794 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1795 {
1796         int status, namelen;
1797
1798         mlog_entry_void();
1799
1800         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1801                            (long long)blkno);
1802         if (namelen <= 0) {
1803                 if (namelen)
1804                         status = namelen;
1805                 else
1806                         status = -EINVAL;
1807                 mlog_errno(status);
1808                 goto bail;
1809         }
1810         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1811                 status = -EINVAL;
1812                 mlog_errno(status);
1813                 goto bail;
1814         }
1815
1816         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1817              namelen);
1818
1819         status = 0;
1820 bail:
1821         mlog_exit(status);
1822         return status;
1823 }
1824
1825 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1826                                     struct inode **ret_orphan_dir,
1827                                     struct inode *inode,
1828                                     char *name,
1829                                     struct buffer_head **de_bh)
1830 {
1831         struct inode *orphan_dir_inode;
1832         struct buffer_head *orphan_dir_bh = NULL;
1833         int status = 0;
1834
1835         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1836         if (status < 0) {
1837                 mlog_errno(status);
1838                 return status;
1839         }
1840
1841         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1842                                                        ORPHAN_DIR_SYSTEM_INODE,
1843                                                        osb->slot_num);
1844         if (!orphan_dir_inode) {
1845                 status = -ENOENT;
1846                 mlog_errno(status);
1847                 return status;
1848         }
1849
1850         mutex_lock(&orphan_dir_inode->i_mutex);
1851
1852         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1853         if (status < 0) {
1854                 mlog_errno(status);
1855                 goto leave;
1856         }
1857
1858         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1859                                               orphan_dir_bh, name,
1860                                               OCFS2_ORPHAN_NAMELEN, de_bh);
1861         if (status < 0) {
1862                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1863
1864                 mlog_errno(status);
1865                 goto leave;
1866         }
1867
1868         *ret_orphan_dir = orphan_dir_inode;
1869
1870 leave:
1871         if (status) {
1872                 mutex_unlock(&orphan_dir_inode->i_mutex);
1873                 iput(orphan_dir_inode);
1874         }
1875
1876         brelse(orphan_dir_bh);
1877
1878         mlog_exit(status);
1879         return status;
1880 }
1881
1882 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1883                             handle_t *handle,
1884                             struct inode *inode,
1885                             struct ocfs2_dinode *fe,
1886                             char *name,
1887                             struct buffer_head *de_bh,
1888                             struct inode *orphan_dir_inode)
1889 {
1890         struct buffer_head *orphan_dir_bh = NULL;
1891         int status = 0;
1892         struct ocfs2_dinode *orphan_fe;
1893
1894         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1895
1896         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1897         if (status < 0) {
1898                 mlog_errno(status);
1899                 goto leave;
1900         }
1901
1902         status = ocfs2_journal_access_di(handle, orphan_dir_inode, orphan_dir_bh,
1903                                          OCFS2_JOURNAL_ACCESS_WRITE);
1904         if (status < 0) {
1905                 mlog_errno(status);
1906                 goto leave;
1907         }
1908
1909         /* we're a cluster, and nlink can change on disk from
1910          * underneath us... */
1911         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1912         if (S_ISDIR(inode->i_mode))
1913                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1914         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1915
1916         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1917         if (status < 0) {
1918                 mlog_errno(status);
1919                 goto leave;
1920         }
1921
1922         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1923                                    OCFS2_ORPHAN_NAMELEN, inode,
1924                                    OCFS2_I(inode)->ip_blkno,
1925                                    orphan_dir_bh, de_bh);
1926         if (status < 0) {
1927                 mlog_errno(status);
1928                 goto leave;
1929         }
1930
1931         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1932
1933         /* Record which orphan dir our inode now resides
1934          * in. delete_inode will use this to determine which orphan
1935          * dir to lock. */
1936         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1937
1938         mlog(0, "Inode %llu orphaned in slot %d\n",
1939              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1940
1941 leave:
1942         brelse(orphan_dir_bh);
1943
1944         mlog_exit(status);
1945         return status;
1946 }
1947
1948 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1949 int ocfs2_orphan_del(struct ocfs2_super *osb,
1950                      handle_t *handle,
1951                      struct inode *orphan_dir_inode,
1952                      struct inode *inode,
1953                      struct buffer_head *orphan_dir_bh)
1954 {
1955         char name[OCFS2_ORPHAN_NAMELEN + 1];
1956         struct ocfs2_dinode *orphan_fe;
1957         int status = 0;
1958         struct buffer_head *target_de_bh = NULL;
1959         struct ocfs2_dir_entry *target_de = NULL;
1960
1961         mlog_entry_void();
1962
1963         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1964         if (status < 0) {
1965                 mlog_errno(status);
1966                 goto leave;
1967         }
1968
1969         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1970              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1971              OCFS2_ORPHAN_NAMELEN);
1972
1973         /* find it's spot in the orphan directory */
1974         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1975                                         orphan_dir_inode, &target_de);
1976         if (!target_de_bh) {
1977                 status = -ENOENT;
1978                 mlog_errno(status);
1979                 goto leave;
1980         }
1981
1982         /* remove it from the orphan directory */
1983         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1984                                     target_de_bh);
1985         if (status < 0) {
1986                 mlog_errno(status);
1987                 goto leave;
1988         }
1989
1990         status = ocfs2_journal_access_di(handle,orphan_dir_inode,  orphan_dir_bh,
1991                                          OCFS2_JOURNAL_ACCESS_WRITE);
1992         if (status < 0) {
1993                 mlog_errno(status);
1994                 goto leave;
1995         }
1996
1997         /* do the i_nlink dance! :) */
1998         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1999         if (S_ISDIR(inode->i_mode))
2000                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2001         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2002
2003         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2004         if (status < 0) {
2005                 mlog_errno(status);
2006                 goto leave;
2007         }
2008
2009 leave:
2010         brelse(target_de_bh);
2011
2012         mlog_exit(status);
2013         return status;
2014 }
2015
2016 const struct inode_operations ocfs2_dir_iops = {
2017         .create         = ocfs2_create,
2018         .lookup         = ocfs2_lookup,
2019         .link           = ocfs2_link,
2020         .unlink         = ocfs2_unlink,
2021         .rmdir          = ocfs2_unlink,
2022         .symlink        = ocfs2_symlink,
2023         .mkdir          = ocfs2_mkdir,
2024         .mknod          = ocfs2_mknod,
2025         .rename         = ocfs2_rename,
2026         .setattr        = ocfs2_setattr,
2027         .getattr        = ocfs2_getattr,
2028         .permission     = ocfs2_permission,
2029         .setxattr       = generic_setxattr,
2030         .getxattr       = generic_getxattr,
2031         .listxattr      = ocfs2_listxattr,
2032         .removexattr    = generic_removexattr,
2033 };