ocfs2: Wrap inode block reads in a dedicated function.
[linux-2.6] / fs / ocfs2 / namei.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * namei.c
5  *
6  * Create and rename file, directory, symlinks
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  *  Portions of this code from linux/fs/ext3/dir.c
11  *
12  *  Copyright (C) 1992, 1993, 1994, 1995
13  *  Remy Card (card@masi.ibp.fr)
14  *  Laboratoire MASI - Institut Blaise pascal
15  *  Universite Pierre et Marie Curie (Paris VI)
16  *
17  *   from
18  *
19  *   linux/fs/minix/dir.c
20  *
21  *   Copyright (C) 1991, 1992 Linux Torvalds
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public
25  * License as published by the Free Software Foundation; either
26  * version 2 of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
31  * General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public
34  * License along with this program; if not, write to the
35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36  * Boston, MA 021110-1307, USA.
37  */
38
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43
44 #define MLOG_MASK_PREFIX ML_NAMEI
45 #include <cluster/masklog.h>
46
47 #include "ocfs2.h"
48
49 #include "alloc.h"
50 #include "dcache.h"
51 #include "dir.h"
52 #include "dlmglue.h"
53 #include "extent_map.h"
54 #include "file.h"
55 #include "inode.h"
56 #include "journal.h"
57 #include "namei.h"
58 #include "suballoc.h"
59 #include "super.h"
60 #include "symlink.h"
61 #include "sysfile.h"
62 #include "uptodate.h"
63 #include "xattr.h"
64 #include "acl.h"
65
66 #include "buffer_head_io.h"
67
68 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
69                               struct inode *dir,
70                               struct inode *inode,
71                               struct dentry *dentry,
72                               dev_t dev,
73                               struct buffer_head **new_fe_bh,
74                               struct buffer_head *parent_fe_bh,
75                               handle_t *handle,
76                               struct ocfs2_alloc_context *inode_ac);
77
78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
79                                     struct inode **ret_orphan_dir,
80                                     struct inode *inode,
81                                     char *name,
82                                     struct buffer_head **de_bh);
83
84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
85                             handle_t *handle,
86                             struct inode *inode,
87                             struct ocfs2_dinode *fe,
88                             char *name,
89                             struct buffer_head *de_bh,
90                             struct inode *orphan_dir_inode);
91
92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
93                                      handle_t *handle,
94                                      struct inode *inode,
95                                      const char *symname);
96
97 /* An orphan dir name is an 8 byte value, printed as a hex string */
98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
99
100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
101                                    struct nameidata *nd)
102 {
103         int status;
104         u64 blkno;
105         struct inode *inode = NULL;
106         struct dentry *ret;
107         struct ocfs2_inode_info *oi;
108
109         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
110                    dentry->d_name.len, dentry->d_name.name);
111
112         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
113                 ret = ERR_PTR(-ENAMETOOLONG);
114                 goto bail;
115         }
116
117         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
118              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
119
120         status = ocfs2_inode_lock(dir, NULL, 0);
121         if (status < 0) {
122                 if (status != -ENOENT)
123                         mlog_errno(status);
124                 ret = ERR_PTR(status);
125                 goto bail;
126         }
127
128         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
129                                             dentry->d_name.len, &blkno);
130         if (status < 0)
131                 goto bail_add;
132
133         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
134         if (IS_ERR(inode)) {
135                 ret = ERR_PTR(-EACCES);
136                 goto bail_unlock;
137         }
138
139         oi = OCFS2_I(inode);
140         /* Clear any orphaned state... If we were able to look up the
141          * inode from a directory, it certainly can't be orphaned. We
142          * might have the bad state from a node which intended to
143          * orphan this inode but crashed before it could commit the
144          * unlink. */
145         spin_lock(&oi->ip_lock);
146         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
147         spin_unlock(&oi->ip_lock);
148
149 bail_add:
150         dentry->d_op = &ocfs2_dentry_ops;
151         ret = d_splice_alias(inode, dentry);
152
153         if (inode) {
154                 /*
155                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
156                  * dentry, it will d_move() it on top of ourse. The
157                  * return value will indicate this however, so in
158                  * those cases, we switch them around for the locking
159                  * code.
160                  *
161                  * NOTE: This dentry already has ->d_op set from
162                  * ocfs2_get_parent() and ocfs2_get_dentry()
163                  */
164                 if (ret)
165                         dentry = ret;
166
167                 status = ocfs2_dentry_attach_lock(dentry, inode,
168                                                   OCFS2_I(dir)->ip_blkno);
169                 if (status) {
170                         mlog_errno(status);
171                         ret = ERR_PTR(status);
172                         goto bail_unlock;
173                 }
174         }
175
176 bail_unlock:
177         /* Don't drop the cluster lock until *after* the d_add --
178          * unlink on another node will message us to remove that
179          * dentry under this lock so otherwise we can race this with
180          * the downconvert thread and have a stale dentry. */
181         ocfs2_inode_unlock(dir, 0);
182
183 bail:
184
185         mlog_exit_ptr(ret);
186
187         return ret;
188 }
189
190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
191 {
192         struct inode *inode;
193
194         inode = new_inode(dir->i_sb);
195         if (!inode) {
196                 mlog(ML_ERROR, "new_inode failed!\n");
197                 return NULL;
198         }
199
200         /* populate as many fields early on as possible - many of
201          * these are used by the support functions here and in
202          * callers. */
203         if (S_ISDIR(mode))
204                 inode->i_nlink = 2;
205         else
206                 inode->i_nlink = 1;
207         inode->i_uid = current_fsuid();
208         if (dir->i_mode & S_ISGID) {
209                 inode->i_gid = dir->i_gid;
210                 if (S_ISDIR(mode))
211                         mode |= S_ISGID;
212         } else
213                 inode->i_gid = current_fsgid();
214         inode->i_mode = mode;
215         return inode;
216 }
217
218 static int ocfs2_mknod(struct inode *dir,
219                        struct dentry *dentry,
220                        int mode,
221                        dev_t dev)
222 {
223         int status = 0;
224         struct buffer_head *parent_fe_bh = NULL;
225         handle_t *handle = NULL;
226         struct ocfs2_super *osb;
227         struct ocfs2_dinode *dirfe;
228         struct buffer_head *new_fe_bh = NULL;
229         struct buffer_head *de_bh = NULL;
230         struct inode *inode = NULL;
231         struct ocfs2_alloc_context *inode_ac = NULL;
232         struct ocfs2_alloc_context *data_ac = NULL;
233         struct ocfs2_alloc_context *xattr_ac = NULL;
234         int want_clusters = 0;
235         int xattr_credits = 0;
236         struct ocfs2_security_xattr_info si = {
237                 .enable = 1,
238         };
239
240         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
241                    (unsigned long)dev, dentry->d_name.len,
242                    dentry->d_name.name);
243
244         /* get our super block */
245         osb = OCFS2_SB(dir->i_sb);
246
247         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
248         if (status < 0) {
249                 if (status != -ENOENT)
250                         mlog_errno(status);
251                 return status;
252         }
253
254         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
255                 status = -EMLINK;
256                 goto leave;
257         }
258
259         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
260         if (!dirfe->i_links_count) {
261                 /* can't make a file in a deleted directory. */
262                 status = -ENOENT;
263                 goto leave;
264         }
265
266         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
267                                            dentry->d_name.len);
268         if (status)
269                 goto leave;
270
271         /* get a spot inside the dir. */
272         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
273                                               dentry->d_name.name,
274                                               dentry->d_name.len, &de_bh);
275         if (status < 0) {
276                 mlog_errno(status);
277                 goto leave;
278         }
279
280         /* reserve an inode spot */
281         status = ocfs2_reserve_new_inode(osb, &inode_ac);
282         if (status < 0) {
283                 if (status != -ENOSPC)
284                         mlog_errno(status);
285                 goto leave;
286         }
287
288         inode = ocfs2_get_init_inode(dir, mode);
289         if (!inode) {
290                 status = -ENOMEM;
291                 mlog_errno(status);
292                 goto leave;
293         }
294
295         /* get security xattr */
296         status = ocfs2_init_security_get(inode, dir, &si);
297         if (status) {
298                 if (status == -EOPNOTSUPP)
299                         si.enable = 0;
300                 else {
301                         mlog_errno(status);
302                         goto leave;
303                 }
304         }
305
306         /* calculate meta data/clusters for setting security and acl xattr */
307         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
308                                         &si, &want_clusters,
309                                         &xattr_credits, &xattr_ac);
310         if (status < 0) {
311                 mlog_errno(status);
312                 goto leave;
313         }
314
315         /* Reserve a cluster if creating an extent based directory. */
316         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb))
317                 want_clusters += 1;
318
319         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
320         if (status < 0) {
321                 if (status != -ENOSPC)
322                         mlog_errno(status);
323                 goto leave;
324         }
325
326         handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS + xattr_credits);
327         if (IS_ERR(handle)) {
328                 status = PTR_ERR(handle);
329                 handle = NULL;
330                 mlog_errno(status);
331                 goto leave;
332         }
333
334         /* do the real work now. */
335         status = ocfs2_mknod_locked(osb, dir, inode, dentry, dev,
336                                     &new_fe_bh, parent_fe_bh, handle,
337                                     inode_ac);
338         if (status < 0) {
339                 mlog_errno(status);
340                 goto leave;
341         }
342
343         if (S_ISDIR(mode)) {
344                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
345                                             new_fe_bh, data_ac);
346                 if (status < 0) {
347                         mlog_errno(status);
348                         goto leave;
349                 }
350
351                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
352                                               OCFS2_JOURNAL_ACCESS_WRITE);
353                 if (status < 0) {
354                         mlog_errno(status);
355                         goto leave;
356                 }
357                 le16_add_cpu(&dirfe->i_links_count, 1);
358                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
359                 if (status < 0) {
360                         mlog_errno(status);
361                         goto leave;
362                 }
363                 inc_nlink(dir);
364         }
365
366         status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
367                                 xattr_ac, data_ac);
368         if (status < 0) {
369                 mlog_errno(status);
370                 goto leave;
371         }
372
373         if (si.enable) {
374                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
375                                                  xattr_ac, data_ac);
376                 if (status < 0) {
377                         mlog_errno(status);
378                         goto leave;
379                 }
380         }
381
382         status = ocfs2_add_entry(handle, dentry, inode,
383                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
384                                  de_bh);
385         if (status < 0) {
386                 mlog_errno(status);
387                 goto leave;
388         }
389
390         status = ocfs2_dentry_attach_lock(dentry, inode,
391                                           OCFS2_I(dir)->ip_blkno);
392         if (status) {
393                 mlog_errno(status);
394                 goto leave;
395         }
396
397         insert_inode_hash(inode);
398         dentry->d_op = &ocfs2_dentry_ops;
399         d_instantiate(dentry, inode);
400         status = 0;
401 leave:
402         if (handle)
403                 ocfs2_commit_trans(osb, handle);
404
405         ocfs2_inode_unlock(dir, 1);
406
407         if (status == -ENOSPC)
408                 mlog(0, "Disk is full\n");
409
410         brelse(new_fe_bh);
411         brelse(de_bh);
412         brelse(parent_fe_bh);
413         kfree(si.name);
414         kfree(si.value);
415
416         if ((status < 0) && inode) {
417                 clear_nlink(inode);
418                 iput(inode);
419         }
420
421         if (inode_ac)
422                 ocfs2_free_alloc_context(inode_ac);
423
424         if (data_ac)
425                 ocfs2_free_alloc_context(data_ac);
426
427         if (xattr_ac)
428                 ocfs2_free_alloc_context(xattr_ac);
429
430         mlog_exit(status);
431
432         return status;
433 }
434
435 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
436                               struct inode *dir,
437                               struct inode *inode,
438                               struct dentry *dentry,
439                               dev_t dev,
440                               struct buffer_head **new_fe_bh,
441                               struct buffer_head *parent_fe_bh,
442                               handle_t *handle,
443                               struct ocfs2_alloc_context *inode_ac)
444 {
445         int status = 0;
446         struct ocfs2_dinode *fe = NULL;
447         struct ocfs2_extent_list *fel;
448         u64 fe_blkno = 0;
449         u16 suballoc_bit;
450
451         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
452                    inode->i_mode, (unsigned long)dev, dentry->d_name.len,
453                    dentry->d_name.name);
454
455         *new_fe_bh = NULL;
456
457         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
458                                        &fe_blkno);
459         if (status < 0) {
460                 mlog_errno(status);
461                 goto leave;
462         }
463
464         /* populate as many fields early on as possible - many of
465          * these are used by the support functions here and in
466          * callers. */
467         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
468         OCFS2_I(inode)->ip_blkno = fe_blkno;
469         spin_lock(&osb->osb_lock);
470         inode->i_generation = osb->s_next_generation++;
471         spin_unlock(&osb->osb_lock);
472
473         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
474         if (!*new_fe_bh) {
475                 status = -EIO;
476                 mlog_errno(status);
477                 goto leave;
478         }
479         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
480
481         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
482                                       OCFS2_JOURNAL_ACCESS_CREATE);
483         if (status < 0) {
484                 mlog_errno(status);
485                 goto leave;
486         }
487
488         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
489         memset(fe, 0, osb->sb->s_blocksize);
490
491         fe->i_generation = cpu_to_le32(inode->i_generation);
492         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
493         fe->i_blkno = cpu_to_le64(fe_blkno);
494         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
495         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
496         fe->i_uid = cpu_to_le32(inode->i_uid);
497         fe->i_gid = cpu_to_le32(inode->i_gid);
498         fe->i_mode = cpu_to_le16(inode->i_mode);
499         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
500                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
501         fe->i_links_count = cpu_to_le16(inode->i_nlink);
502
503         fe->i_last_eb_blk = 0;
504         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
505         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
506         fe->i_atime = fe->i_ctime = fe->i_mtime =
507                 cpu_to_le64(CURRENT_TIME.tv_sec);
508         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
509                 cpu_to_le32(CURRENT_TIME.tv_nsec);
510         fe->i_dtime = 0;
511
512         /*
513          * If supported, directories start with inline data.
514          */
515         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
516                 u16 feat = le16_to_cpu(fe->i_dyn_features);
517
518                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
519
520                 fe->id2.i_data.id_count = cpu_to_le16(ocfs2_max_inline_data(osb->sb));
521         } else {
522                 fel = &fe->id2.i_list;
523                 fel->l_tree_depth = 0;
524                 fel->l_next_free_rec = 0;
525                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
526         }
527
528         status = ocfs2_journal_dirty(handle, *new_fe_bh);
529         if (status < 0) {
530                 mlog_errno(status);
531                 goto leave;
532         }
533
534         ocfs2_populate_inode(inode, fe, 1);
535         ocfs2_inode_set_new(osb, inode);
536         if (!ocfs2_mount_local(osb)) {
537                 status = ocfs2_create_new_inode_locks(inode);
538                 if (status < 0)
539                         mlog_errno(status);
540         }
541
542         status = 0; /* error in ocfs2_create_new_inode_locks is not
543                      * critical */
544
545 leave:
546         if (status < 0) {
547                 if (*new_fe_bh) {
548                         brelse(*new_fe_bh);
549                         *new_fe_bh = NULL;
550                 }
551         }
552
553         mlog_exit(status);
554         return status;
555 }
556
557 static int ocfs2_mkdir(struct inode *dir,
558                        struct dentry *dentry,
559                        int mode)
560 {
561         int ret;
562
563         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
564                    dentry->d_name.len, dentry->d_name.name);
565         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
566         mlog_exit(ret);
567
568         return ret;
569 }
570
571 static int ocfs2_create(struct inode *dir,
572                         struct dentry *dentry,
573                         int mode,
574                         struct nameidata *nd)
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_IFREG, 0);
581         mlog_exit(ret);
582
583         return ret;
584 }
585
586 static int ocfs2_link(struct dentry *old_dentry,
587                       struct inode *dir,
588                       struct dentry *dentry)
589 {
590         handle_t *handle;
591         struct inode *inode = old_dentry->d_inode;
592         int err;
593         struct buffer_head *fe_bh = NULL;
594         struct buffer_head *parent_fe_bh = NULL;
595         struct buffer_head *de_bh = NULL;
596         struct ocfs2_dinode *fe = NULL;
597         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
598
599         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
600                    old_dentry->d_name.len, old_dentry->d_name.name,
601                    dentry->d_name.len, dentry->d_name.name);
602
603         if (S_ISDIR(inode->i_mode))
604                 return -EPERM;
605
606         err = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
607         if (err < 0) {
608                 if (err != -ENOENT)
609                         mlog_errno(err);
610                 return err;
611         }
612
613         if (!dir->i_nlink) {
614                 err = -ENOENT;
615                 goto out;
616         }
617
618         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
619                                         dentry->d_name.len);
620         if (err)
621                 goto out;
622
623         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
624                                            dentry->d_name.name,
625                                            dentry->d_name.len, &de_bh);
626         if (err < 0) {
627                 mlog_errno(err);
628                 goto out;
629         }
630
631         err = ocfs2_inode_lock(inode, &fe_bh, 1);
632         if (err < 0) {
633                 if (err != -ENOENT)
634                         mlog_errno(err);
635                 goto out;
636         }
637
638         fe = (struct ocfs2_dinode *) fe_bh->b_data;
639         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
640                 err = -EMLINK;
641                 goto out_unlock_inode;
642         }
643
644         handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
645         if (IS_ERR(handle)) {
646                 err = PTR_ERR(handle);
647                 handle = NULL;
648                 mlog_errno(err);
649                 goto out_unlock_inode;
650         }
651
652         err = ocfs2_journal_access(handle, inode, fe_bh,
653                                    OCFS2_JOURNAL_ACCESS_WRITE);
654         if (err < 0) {
655                 mlog_errno(err);
656                 goto out_commit;
657         }
658
659         inc_nlink(inode);
660         inode->i_ctime = CURRENT_TIME;
661         fe->i_links_count = cpu_to_le16(inode->i_nlink);
662         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
663         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
664
665         err = ocfs2_journal_dirty(handle, fe_bh);
666         if (err < 0) {
667                 le16_add_cpu(&fe->i_links_count, -1);
668                 drop_nlink(inode);
669                 mlog_errno(err);
670                 goto out_commit;
671         }
672
673         err = ocfs2_add_entry(handle, dentry, inode,
674                               OCFS2_I(inode)->ip_blkno,
675                               parent_fe_bh, de_bh);
676         if (err) {
677                 le16_add_cpu(&fe->i_links_count, -1);
678                 drop_nlink(inode);
679                 mlog_errno(err);
680                 goto out_commit;
681         }
682
683         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
684         if (err) {
685                 mlog_errno(err);
686                 goto out_commit;
687         }
688
689         atomic_inc(&inode->i_count);
690         dentry->d_op = &ocfs2_dentry_ops;
691         d_instantiate(dentry, inode);
692
693 out_commit:
694         ocfs2_commit_trans(osb, handle);
695 out_unlock_inode:
696         ocfs2_inode_unlock(inode, 1);
697
698 out:
699         ocfs2_inode_unlock(dir, 1);
700
701         brelse(de_bh);
702         brelse(fe_bh);
703         brelse(parent_fe_bh);
704
705         mlog_exit(err);
706
707         return err;
708 }
709
710 /*
711  * Takes and drops an exclusive lock on the given dentry. This will
712  * force other nodes to drop it.
713  */
714 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
715 {
716         int ret;
717
718         ret = ocfs2_dentry_lock(dentry, 1);
719         if (ret)
720                 mlog_errno(ret);
721         else
722                 ocfs2_dentry_unlock(dentry, 1);
723
724         return ret;
725 }
726
727 static inline int inode_is_unlinkable(struct inode *inode)
728 {
729         if (S_ISDIR(inode->i_mode)) {
730                 if (inode->i_nlink == 2)
731                         return 1;
732                 return 0;
733         }
734
735         if (inode->i_nlink == 1)
736                 return 1;
737         return 0;
738 }
739
740 static int ocfs2_unlink(struct inode *dir,
741                         struct dentry *dentry)
742 {
743         int status;
744         int child_locked = 0;
745         struct inode *inode = dentry->d_inode;
746         struct inode *orphan_dir = NULL;
747         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
748         u64 blkno;
749         struct ocfs2_dinode *fe = NULL;
750         struct buffer_head *fe_bh = NULL;
751         struct buffer_head *parent_node_bh = NULL;
752         handle_t *handle = NULL;
753         struct ocfs2_dir_entry *dirent = NULL;
754         struct buffer_head *dirent_bh = NULL;
755         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
756         struct buffer_head *orphan_entry_bh = NULL;
757
758         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
759                    dentry->d_name.len, dentry->d_name.name);
760
761         BUG_ON(dentry->d_parent->d_inode != dir);
762
763         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
764
765         if (inode == osb->root_inode) {
766                 mlog(0, "Cannot delete the root directory\n");
767                 return -EPERM;
768         }
769
770         status = ocfs2_inode_lock(dir, &parent_node_bh, 1);
771         if (status < 0) {
772                 if (status != -ENOENT)
773                         mlog_errno(status);
774                 return status;
775         }
776
777         status = ocfs2_find_files_on_disk(dentry->d_name.name,
778                                           dentry->d_name.len, &blkno,
779                                           dir, &dirent_bh, &dirent);
780         if (status < 0) {
781                 if (status != -ENOENT)
782                         mlog_errno(status);
783                 goto leave;
784         }
785
786         if (OCFS2_I(inode)->ip_blkno != blkno) {
787                 status = -ENOENT;
788
789                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
790                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
791                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
792                 goto leave;
793         }
794
795         status = ocfs2_inode_lock(inode, &fe_bh, 1);
796         if (status < 0) {
797                 if (status != -ENOENT)
798                         mlog_errno(status);
799                 goto leave;
800         }
801         child_locked = 1;
802
803         if (S_ISDIR(inode->i_mode)) {
804                 if (!ocfs2_empty_dir(inode)) {
805                         status = -ENOTEMPTY;
806                         goto leave;
807                 } else if (inode->i_nlink != 2) {
808                         status = -ENOTEMPTY;
809                         goto leave;
810                 }
811         }
812
813         status = ocfs2_remote_dentry_delete(dentry);
814         if (status < 0) {
815                 /* This remote delete should succeed under all normal
816                  * circumstances. */
817                 mlog_errno(status);
818                 goto leave;
819         }
820
821         if (inode_is_unlinkable(inode)) {
822                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
823                                                   orphan_name,
824                                                   &orphan_entry_bh);
825                 if (status < 0) {
826                         mlog_errno(status);
827                         goto leave;
828                 }
829         }
830
831         handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
832         if (IS_ERR(handle)) {
833                 status = PTR_ERR(handle);
834                 handle = NULL;
835                 mlog_errno(status);
836                 goto leave;
837         }
838
839         status = ocfs2_journal_access(handle, inode, fe_bh,
840                                       OCFS2_JOURNAL_ACCESS_WRITE);
841         if (status < 0) {
842                 mlog_errno(status);
843                 goto leave;
844         }
845
846         fe = (struct ocfs2_dinode *) fe_bh->b_data;
847
848         if (inode_is_unlinkable(inode)) {
849                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
850                                           orphan_entry_bh, orphan_dir);
851                 if (status < 0) {
852                         mlog_errno(status);
853                         goto leave;
854                 }
855         }
856
857         /* delete the name from the parent dir */
858         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
859         if (status < 0) {
860                 mlog_errno(status);
861                 goto leave;
862         }
863
864         if (S_ISDIR(inode->i_mode))
865                 drop_nlink(inode);
866         drop_nlink(inode);
867         fe->i_links_count = cpu_to_le16(inode->i_nlink);
868
869         status = ocfs2_journal_dirty(handle, fe_bh);
870         if (status < 0) {
871                 mlog_errno(status);
872                 goto leave;
873         }
874
875         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
876         if (S_ISDIR(inode->i_mode))
877                 drop_nlink(dir);
878
879         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
880         if (status < 0) {
881                 mlog_errno(status);
882                 if (S_ISDIR(inode->i_mode))
883                         inc_nlink(dir);
884         }
885
886 leave:
887         if (handle)
888                 ocfs2_commit_trans(osb, handle);
889
890         if (child_locked)
891                 ocfs2_inode_unlock(inode, 1);
892
893         ocfs2_inode_unlock(dir, 1);
894
895         if (orphan_dir) {
896                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
897                 ocfs2_inode_unlock(orphan_dir, 1);
898                 mutex_unlock(&orphan_dir->i_mutex);
899                 iput(orphan_dir);
900         }
901
902         brelse(fe_bh);
903         brelse(dirent_bh);
904         brelse(parent_node_bh);
905         brelse(orphan_entry_bh);
906
907         mlog_exit(status);
908
909         return status;
910 }
911
912 /*
913  * The only place this should be used is rename!
914  * if they have the same id, then the 1st one is the only one locked.
915  */
916 static int ocfs2_double_lock(struct ocfs2_super *osb,
917                              struct buffer_head **bh1,
918                              struct inode *inode1,
919                              struct buffer_head **bh2,
920                              struct inode *inode2)
921 {
922         int status;
923         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
924         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
925         struct buffer_head **tmpbh;
926         struct inode *tmpinode;
927
928         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
929                    (unsigned long long)oi1->ip_blkno,
930                    (unsigned long long)oi2->ip_blkno);
931
932         if (*bh1)
933                 *bh1 = NULL;
934         if (*bh2)
935                 *bh2 = NULL;
936
937         /* we always want to lock the one with the lower lockid first. */
938         if (oi1->ip_blkno != oi2->ip_blkno) {
939                 if (oi1->ip_blkno < oi2->ip_blkno) {
940                         /* switch id1 and id2 around */
941                         mlog(0, "switching them around...\n");
942                         tmpbh = bh2;
943                         bh2 = bh1;
944                         bh1 = tmpbh;
945
946                         tmpinode = inode2;
947                         inode2 = inode1;
948                         inode1 = tmpinode;
949                 }
950                 /* lock id2 */
951                 status = ocfs2_inode_lock(inode2, bh2, 1);
952                 if (status < 0) {
953                         if (status != -ENOENT)
954                                 mlog_errno(status);
955                         goto bail;
956                 }
957         }
958
959         /* lock id1 */
960         status = ocfs2_inode_lock(inode1, bh1, 1);
961         if (status < 0) {
962                 /*
963                  * An error return must mean that no cluster locks
964                  * were held on function exit.
965                  */
966                 if (oi1->ip_blkno != oi2->ip_blkno)
967                         ocfs2_inode_unlock(inode2, 1);
968
969                 if (status != -ENOENT)
970                         mlog_errno(status);
971         }
972
973 bail:
974         mlog_exit(status);
975         return status;
976 }
977
978 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
979 {
980         ocfs2_inode_unlock(inode1, 1);
981
982         if (inode1 != inode2)
983                 ocfs2_inode_unlock(inode2, 1);
984 }
985
986 static int ocfs2_rename(struct inode *old_dir,
987                         struct dentry *old_dentry,
988                         struct inode *new_dir,
989                         struct dentry *new_dentry)
990 {
991         int status = 0, rename_lock = 0, parents_locked = 0;
992         int old_child_locked = 0, new_child_locked = 0;
993         struct inode *old_inode = old_dentry->d_inode;
994         struct inode *new_inode = new_dentry->d_inode;
995         struct inode *orphan_dir = NULL;
996         struct ocfs2_dinode *newfe = NULL;
997         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
998         struct buffer_head *orphan_entry_bh = NULL;
999         struct buffer_head *newfe_bh = NULL;
1000         struct buffer_head *old_inode_bh = NULL;
1001         struct buffer_head *insert_entry_bh = NULL;
1002         struct ocfs2_super *osb = NULL;
1003         u64 newfe_blkno, old_de_ino;
1004         handle_t *handle = NULL;
1005         struct buffer_head *old_dir_bh = NULL;
1006         struct buffer_head *new_dir_bh = NULL;
1007         struct ocfs2_dir_entry *old_inode_dot_dot_de = NULL, *old_de = NULL,
1008                 *new_de = NULL;
1009         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1010         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1011                                                     // this is the 1st dirent bh
1012         nlink_t old_dir_nlink = old_dir->i_nlink;
1013         struct ocfs2_dinode *old_di;
1014
1015         /* At some point it might be nice to break this function up a
1016          * bit. */
1017
1018         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1019                    old_dir, old_dentry, new_dir, new_dentry,
1020                    old_dentry->d_name.len, old_dentry->d_name.name,
1021                    new_dentry->d_name.len, new_dentry->d_name.name);
1022
1023         osb = OCFS2_SB(old_dir->i_sb);
1024
1025         if (new_inode) {
1026                 if (!igrab(new_inode))
1027                         BUG();
1028         }
1029
1030         /* Assume a directory hierarchy thusly:
1031          * a/b/c
1032          * a/d
1033          * a,b,c, and d are all directories.
1034          *
1035          * from cwd of 'a' on both nodes:
1036          * node1: mv b/c d
1037          * node2: mv d   b/c
1038          *
1039          * And that's why, just like the VFS, we need a file system
1040          * rename lock. */
1041         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1042                 status = ocfs2_rename_lock(osb);
1043                 if (status < 0) {
1044                         mlog_errno(status);
1045                         goto bail;
1046                 }
1047                 rename_lock = 1;
1048         }
1049
1050         /* if old and new are the same, this'll just do one lock. */
1051         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1052                                    &new_dir_bh, new_dir);
1053         if (status < 0) {
1054                 mlog_errno(status);
1055                 goto bail;
1056         }
1057         parents_locked = 1;
1058
1059         /* make sure both dirs have bhs
1060          * get an extra ref on old_dir_bh if old==new */
1061         if (!new_dir_bh) {
1062                 if (old_dir_bh) {
1063                         new_dir_bh = old_dir_bh;
1064                         get_bh(new_dir_bh);
1065                 } else {
1066                         mlog(ML_ERROR, "no old_dir_bh!\n");
1067                         status = -EIO;
1068                         goto bail;
1069                 }
1070         }
1071
1072         /*
1073          * Aside from allowing a meta data update, the locking here
1074          * also ensures that the downconvert thread on other nodes
1075          * won't have to concurrently downconvert the inode and the
1076          * dentry locks.
1077          */
1078         status = ocfs2_inode_lock(old_inode, &old_inode_bh, 1);
1079         if (status < 0) {
1080                 if (status != -ENOENT)
1081                         mlog_errno(status);
1082                 goto bail;
1083         }
1084         old_child_locked = 1;
1085
1086         status = ocfs2_remote_dentry_delete(old_dentry);
1087         if (status < 0) {
1088                 mlog_errno(status);
1089                 goto bail;
1090         }
1091
1092         if (S_ISDIR(old_inode->i_mode)) {
1093                 u64 old_inode_parent;
1094
1095                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1096                                                   old_inode, &old_inode_de_bh,
1097                                                   &old_inode_dot_dot_de);
1098                 if (status) {
1099                         status = -EIO;
1100                         goto bail;
1101                 }
1102
1103                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1104                         status = -EIO;
1105                         goto bail;
1106                 }
1107
1108                 if (!new_inode && new_dir != old_dir &&
1109                     new_dir->i_nlink >= OCFS2_LINK_MAX) {
1110                         status = -EMLINK;
1111                         goto bail;
1112                 }
1113         }
1114
1115         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1116                                             old_dentry->d_name.len,
1117                                             &old_de_ino);
1118         if (status) {
1119                 status = -ENOENT;
1120                 goto bail;
1121         }
1122
1123         /*
1124          *  Check for inode number is _not_ due to possible IO errors.
1125          *  We might rmdir the source, keep it as pwd of some process
1126          *  and merrily kill the link to whatever was created under the
1127          *  same name. Goodbye sticky bit ;-<
1128          */
1129         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1130                 status = -ENOENT;
1131                 goto bail;
1132         }
1133
1134         /* check if the target already exists (in which case we need
1135          * to delete it */
1136         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1137                                           new_dentry->d_name.len,
1138                                           &newfe_blkno, new_dir, &new_de_bh,
1139                                           &new_de);
1140         /* The only error we allow here is -ENOENT because the new
1141          * file not existing is perfectly valid. */
1142         if ((status < 0) && (status != -ENOENT)) {
1143                 /* If we cannot find the file specified we should just */
1144                 /* return the error... */
1145                 mlog_errno(status);
1146                 goto bail;
1147         }
1148
1149         if (!new_de && new_inode) {
1150                 /*
1151                  * Target was unlinked by another node while we were
1152                  * waiting to get to ocfs2_rename(). There isn't
1153                  * anything we can do here to help the situation, so
1154                  * bubble up the appropriate error.
1155                  */
1156                 status = -ENOENT;
1157                 goto bail;
1158         }
1159
1160         /* In case we need to overwrite an existing file, we blow it
1161          * away first */
1162         if (new_de) {
1163                 /* VFS didn't think there existed an inode here, but
1164                  * someone else in the cluster must have raced our
1165                  * rename to create one. Today we error cleanly, in
1166                  * the future we should consider calling iget to build
1167                  * a new struct inode for this entry. */
1168                 if (!new_inode) {
1169                         status = -EACCES;
1170
1171                         mlog(0, "We found an inode for name %.*s but VFS "
1172                              "didn't give us one.\n", new_dentry->d_name.len,
1173                              new_dentry->d_name.name);
1174                         goto bail;
1175                 }
1176
1177                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1178                         status = -EACCES;
1179
1180                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1181                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1182                              (unsigned long long)newfe_blkno,
1183                              OCFS2_I(new_inode)->ip_flags);
1184                         goto bail;
1185                 }
1186
1187                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1188                 if (status < 0) {
1189                         if (status != -ENOENT)
1190                                 mlog_errno(status);
1191                         goto bail;
1192                 }
1193                 new_child_locked = 1;
1194
1195                 status = ocfs2_remote_dentry_delete(new_dentry);
1196                 if (status < 0) {
1197                         mlog_errno(status);
1198                         goto bail;
1199                 }
1200
1201                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1202
1203                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1204                      "newfebh=%p bhblocknr=%llu\n", new_de,
1205                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1206                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1207
1208                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1209                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1210                                                           new_inode,
1211                                                           orphan_name,
1212                                                           &orphan_entry_bh);
1213                         if (status < 0) {
1214                                 mlog_errno(status);
1215                                 goto bail;
1216                         }
1217                 }
1218         } else {
1219                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1220
1221                 status = ocfs2_check_dir_for_entry(new_dir,
1222                                                    new_dentry->d_name.name,
1223                                                    new_dentry->d_name.len);
1224                 if (status)
1225                         goto bail;
1226
1227                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1228                                                       new_dentry->d_name.name,
1229                                                       new_dentry->d_name.len,
1230                                                       &insert_entry_bh);
1231                 if (status < 0) {
1232                         mlog_errno(status);
1233                         goto bail;
1234                 }
1235         }
1236
1237         handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1238         if (IS_ERR(handle)) {
1239                 status = PTR_ERR(handle);
1240                 handle = NULL;
1241                 mlog_errno(status);
1242                 goto bail;
1243         }
1244
1245         if (new_de) {
1246                 if (S_ISDIR(new_inode->i_mode)) {
1247                         if (!ocfs2_empty_dir(new_inode) ||
1248                             new_inode->i_nlink != 2) {
1249                                 status = -ENOTEMPTY;
1250                                 goto bail;
1251                         }
1252                 }
1253                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1254                                               OCFS2_JOURNAL_ACCESS_WRITE);
1255                 if (status < 0) {
1256                         mlog_errno(status);
1257                         goto bail;
1258                 }
1259
1260                 if (S_ISDIR(new_inode->i_mode) ||
1261                     (newfe->i_links_count == cpu_to_le16(1))){
1262                         status = ocfs2_orphan_add(osb, handle, new_inode,
1263                                                   newfe, orphan_name,
1264                                                   orphan_entry_bh, orphan_dir);
1265                         if (status < 0) {
1266                                 mlog_errno(status);
1267                                 goto bail;
1268                         }
1269                 }
1270
1271                 /* change the dirent to point to the correct inode */
1272                 status = ocfs2_update_entry(new_dir, handle, new_de_bh,
1273                                             new_de, old_inode);
1274                 if (status < 0) {
1275                         mlog_errno(status);
1276                         goto bail;
1277                 }
1278                 new_dir->i_version++;
1279
1280                 if (S_ISDIR(new_inode->i_mode))
1281                         newfe->i_links_count = 0;
1282                 else
1283                         le16_add_cpu(&newfe->i_links_count, -1);
1284
1285                 status = ocfs2_journal_dirty(handle, newfe_bh);
1286                 if (status < 0) {
1287                         mlog_errno(status);
1288                         goto bail;
1289                 }
1290         } else {
1291                 /* if the name was not found in new_dir, add it now */
1292                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1293                                          OCFS2_I(old_inode)->ip_blkno,
1294                                          new_dir_bh, insert_entry_bh);
1295         }
1296
1297         old_inode->i_ctime = CURRENT_TIME;
1298         mark_inode_dirty(old_inode);
1299
1300         status = ocfs2_journal_access(handle, old_inode, old_inode_bh,
1301                                       OCFS2_JOURNAL_ACCESS_WRITE);
1302         if (status >= 0) {
1303                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1304
1305                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1306                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1307
1308                 status = ocfs2_journal_dirty(handle, old_inode_bh);
1309                 if (status < 0)
1310                         mlog_errno(status);
1311         } else
1312                 mlog_errno(status);
1313
1314         /*
1315          * Now that the name has been added to new_dir, remove the old name.
1316          *
1317          * We don't keep any directory entry context around until now
1318          * because the insert might have changed the type of directory
1319          * we're dealing with.
1320          */
1321         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1322                                      old_dentry->d_name.len,
1323                                      old_dir, &old_de);
1324         if (!old_de_bh) {
1325                 status = -EIO;
1326                 goto bail;
1327         }
1328
1329         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1330         if (status < 0) {
1331                 mlog_errno(status);
1332                 goto bail;
1333         }
1334
1335         if (new_inode) {
1336                 new_inode->i_nlink--;
1337                 new_inode->i_ctime = CURRENT_TIME;
1338         }
1339         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1340         if (old_inode_de_bh) {
1341                 status = ocfs2_update_entry(old_inode, handle, old_inode_de_bh,
1342                                             old_inode_dot_dot_de, new_dir);
1343                 old_dir->i_nlink--;
1344                 if (new_inode) {
1345                         new_inode->i_nlink--;
1346                 } else {
1347                         inc_nlink(new_dir);
1348                         mark_inode_dirty(new_dir);
1349                 }
1350         }
1351         mark_inode_dirty(old_dir);
1352         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1353         if (new_inode) {
1354                 mark_inode_dirty(new_inode);
1355                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1356         }
1357
1358         if (old_dir != new_dir) {
1359                 /* Keep the same times on both directories.*/
1360                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1361
1362                 /*
1363                  * This will also pick up the i_nlink change from the
1364                  * block above.
1365                  */
1366                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1367         }
1368
1369         if (old_dir_nlink != old_dir->i_nlink) {
1370                 if (!old_dir_bh) {
1371                         mlog(ML_ERROR, "need to change nlink for old dir "
1372                              "%llu from %d to %d but bh is NULL!\n",
1373                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1374                              (int)old_dir_nlink, old_dir->i_nlink);
1375                 } else {
1376                         struct ocfs2_dinode *fe;
1377                         status = ocfs2_journal_access(handle, old_dir,
1378                                                       old_dir_bh,
1379                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1380                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1381                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1382                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1383                 }
1384         }
1385
1386         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1387         status = 0;
1388 bail:
1389         if (rename_lock)
1390                 ocfs2_rename_unlock(osb);
1391
1392         if (handle)
1393                 ocfs2_commit_trans(osb, handle);
1394
1395         if (parents_locked)
1396                 ocfs2_double_unlock(old_dir, new_dir);
1397
1398         if (old_child_locked)
1399                 ocfs2_inode_unlock(old_inode, 1);
1400
1401         if (new_child_locked)
1402                 ocfs2_inode_unlock(new_inode, 1);
1403
1404         if (orphan_dir) {
1405                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1406                 ocfs2_inode_unlock(orphan_dir, 1);
1407                 mutex_unlock(&orphan_dir->i_mutex);
1408                 iput(orphan_dir);
1409         }
1410
1411         if (new_inode)
1412                 sync_mapping_buffers(old_inode->i_mapping);
1413
1414         if (new_inode)
1415                 iput(new_inode);
1416         brelse(newfe_bh);
1417         brelse(old_inode_bh);
1418         brelse(old_dir_bh);
1419         brelse(new_dir_bh);
1420         brelse(new_de_bh);
1421         brelse(old_de_bh);
1422         brelse(old_inode_de_bh);
1423         brelse(orphan_entry_bh);
1424         brelse(insert_entry_bh);
1425
1426         mlog_exit(status);
1427
1428         return status;
1429 }
1430
1431 /*
1432  * we expect i_size = strlen(symname). Copy symname into the file
1433  * data, including the null terminator.
1434  */
1435 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1436                                      handle_t *handle,
1437                                      struct inode *inode,
1438                                      const char *symname)
1439 {
1440         struct buffer_head **bhs = NULL;
1441         const char *c;
1442         struct super_block *sb = osb->sb;
1443         u64 p_blkno, p_blocks;
1444         int virtual, blocks, status, i, bytes_left;
1445
1446         bytes_left = i_size_read(inode) + 1;
1447         /* we can't trust i_blocks because we're actually going to
1448          * write i_size + 1 bytes. */
1449         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1450
1451         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1452                         (unsigned long long)inode->i_blocks,
1453                         i_size_read(inode), blocks);
1454
1455         /* Sanity check -- make sure we're going to fit. */
1456         if (bytes_left >
1457             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1458                 status = -EIO;
1459                 mlog_errno(status);
1460                 goto bail;
1461         }
1462
1463         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1464         if (!bhs) {
1465                 status = -ENOMEM;
1466                 mlog_errno(status);
1467                 goto bail;
1468         }
1469
1470         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1471                                              NULL);
1472         if (status < 0) {
1473                 mlog_errno(status);
1474                 goto bail;
1475         }
1476
1477         /* links can never be larger than one cluster so we know this
1478          * is all going to be contiguous, but do a sanity check
1479          * anyway. */
1480         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1481                 status = -EIO;
1482                 mlog_errno(status);
1483                 goto bail;
1484         }
1485
1486         virtual = 0;
1487         while(bytes_left > 0) {
1488                 c = &symname[virtual * sb->s_blocksize];
1489
1490                 bhs[virtual] = sb_getblk(sb, p_blkno);
1491                 if (!bhs[virtual]) {
1492                         status = -ENOMEM;
1493                         mlog_errno(status);
1494                         goto bail;
1495                 }
1496                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1497
1498                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1499                                               OCFS2_JOURNAL_ACCESS_CREATE);
1500                 if (status < 0) {
1501                         mlog_errno(status);
1502                         goto bail;
1503                 }
1504
1505                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1506
1507                 memcpy(bhs[virtual]->b_data, c,
1508                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1509                        bytes_left);
1510
1511                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1512                 if (status < 0) {
1513                         mlog_errno(status);
1514                         goto bail;
1515                 }
1516
1517                 virtual++;
1518                 p_blkno++;
1519                 bytes_left -= sb->s_blocksize;
1520         }
1521
1522         status = 0;
1523 bail:
1524
1525         if (bhs) {
1526                 for(i = 0; i < blocks; i++)
1527                         brelse(bhs[i]);
1528                 kfree(bhs);
1529         }
1530
1531         mlog_exit(status);
1532         return status;
1533 }
1534
1535 static int ocfs2_symlink(struct inode *dir,
1536                          struct dentry *dentry,
1537                          const char *symname)
1538 {
1539         int status, l, credits;
1540         u64 newsize;
1541         struct ocfs2_super *osb = NULL;
1542         struct inode *inode = NULL;
1543         struct super_block *sb;
1544         struct buffer_head *new_fe_bh = NULL;
1545         struct buffer_head *de_bh = NULL;
1546         struct buffer_head *parent_fe_bh = NULL;
1547         struct ocfs2_dinode *fe = NULL;
1548         struct ocfs2_dinode *dirfe;
1549         handle_t *handle = NULL;
1550         struct ocfs2_alloc_context *inode_ac = NULL;
1551         struct ocfs2_alloc_context *data_ac = NULL;
1552         struct ocfs2_alloc_context *xattr_ac = NULL;
1553         int want_clusters = 0;
1554         int xattr_credits = 0;
1555         struct ocfs2_security_xattr_info si = {
1556                 .enable = 1,
1557         };
1558
1559         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1560                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1561
1562         sb = dir->i_sb;
1563         osb = OCFS2_SB(sb);
1564
1565         l = strlen(symname) + 1;
1566
1567         credits = ocfs2_calc_symlink_credits(sb);
1568
1569         /* lock the parent directory */
1570         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1571         if (status < 0) {
1572                 if (status != -ENOENT)
1573                         mlog_errno(status);
1574                 return status;
1575         }
1576
1577         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1578         if (!dirfe->i_links_count) {
1579                 /* can't make a file in a deleted directory. */
1580                 status = -ENOENT;
1581                 goto bail;
1582         }
1583
1584         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1585                                            dentry->d_name.len);
1586         if (status)
1587                 goto bail;
1588
1589         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1590                                               dentry->d_name.name,
1591                                               dentry->d_name.len, &de_bh);
1592         if (status < 0) {
1593                 mlog_errno(status);
1594                 goto bail;
1595         }
1596
1597         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1598         if (status < 0) {
1599                 if (status != -ENOSPC)
1600                         mlog_errno(status);
1601                 goto bail;
1602         }
1603
1604         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1605         if (!inode) {
1606                 status = -ENOMEM;
1607                 mlog_errno(status);
1608                 goto bail;
1609         }
1610
1611         /* get security xattr */
1612         status = ocfs2_init_security_get(inode, dir, &si);
1613         if (status) {
1614                 if (status == -EOPNOTSUPP)
1615                         si.enable = 0;
1616                 else {
1617                         mlog_errno(status);
1618                         goto bail;
1619                 }
1620         }
1621
1622         /* calculate meta data/clusters for setting security xattr */
1623         if (si.enable) {
1624                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1625                                                   &xattr_credits, &xattr_ac);
1626                 if (status < 0) {
1627                         mlog_errno(status);
1628                         goto bail;
1629                 }
1630         }
1631
1632         /* don't reserve bitmap space for fast symlinks. */
1633         if (l > ocfs2_fast_symlink_chars(sb))
1634                 want_clusters += 1;
1635
1636         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1637         if (status < 0) {
1638                 if (status != -ENOSPC)
1639                         mlog_errno(status);
1640                 goto bail;
1641         }
1642
1643         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1644         if (IS_ERR(handle)) {
1645                 status = PTR_ERR(handle);
1646                 handle = NULL;
1647                 mlog_errno(status);
1648                 goto bail;
1649         }
1650
1651         status = ocfs2_mknod_locked(osb, dir, inode, dentry,
1652                                     0, &new_fe_bh, parent_fe_bh, handle,
1653                                     inode_ac);
1654         if (status < 0) {
1655                 mlog_errno(status);
1656                 goto bail;
1657         }
1658
1659         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1660         inode->i_rdev = 0;
1661         newsize = l - 1;
1662         if (l > ocfs2_fast_symlink_chars(sb)) {
1663                 u32 offset = 0;
1664
1665                 inode->i_op = &ocfs2_symlink_inode_operations;
1666                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1667                                               new_fe_bh,
1668                                               handle, data_ac, NULL,
1669                                               NULL);
1670                 if (status < 0) {
1671                         if (status != -ENOSPC && status != -EINTR) {
1672                                 mlog(ML_ERROR,
1673                                      "Failed to extend file to %llu\n",
1674                                      (unsigned long long)newsize);
1675                                 mlog_errno(status);
1676                                 status = -ENOSPC;
1677                         }
1678                         goto bail;
1679                 }
1680                 i_size_write(inode, newsize);
1681                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1682         } else {
1683                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1684                 memcpy((char *) fe->id2.i_symlink, symname, l);
1685                 i_size_write(inode, newsize);
1686                 inode->i_blocks = 0;
1687         }
1688
1689         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1690         if (status < 0) {
1691                 mlog_errno(status);
1692                 goto bail;
1693         }
1694
1695         if (!ocfs2_inode_is_fast_symlink(inode)) {
1696                 status = ocfs2_create_symlink_data(osb, handle, inode,
1697                                                    symname);
1698                 if (status < 0) {
1699                         mlog_errno(status);
1700                         goto bail;
1701                 }
1702         }
1703
1704         if (si.enable) {
1705                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1706                                                  xattr_ac, data_ac);
1707                 if (status < 0) {
1708                         mlog_errno(status);
1709                         goto bail;
1710                 }
1711         }
1712
1713         status = ocfs2_add_entry(handle, dentry, inode,
1714                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1715                                  de_bh);
1716         if (status < 0) {
1717                 mlog_errno(status);
1718                 goto bail;
1719         }
1720
1721         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1722         if (status) {
1723                 mlog_errno(status);
1724                 goto bail;
1725         }
1726
1727         insert_inode_hash(inode);
1728         dentry->d_op = &ocfs2_dentry_ops;
1729         d_instantiate(dentry, inode);
1730 bail:
1731         if (handle)
1732                 ocfs2_commit_trans(osb, handle);
1733
1734         ocfs2_inode_unlock(dir, 1);
1735
1736         brelse(new_fe_bh);
1737         brelse(parent_fe_bh);
1738         brelse(de_bh);
1739         kfree(si.name);
1740         kfree(si.value);
1741         if (inode_ac)
1742                 ocfs2_free_alloc_context(inode_ac);
1743         if (data_ac)
1744                 ocfs2_free_alloc_context(data_ac);
1745         if (xattr_ac)
1746                 ocfs2_free_alloc_context(xattr_ac);
1747         if ((status < 0) && inode) {
1748                 clear_nlink(inode);
1749                 iput(inode);
1750         }
1751
1752         mlog_exit(status);
1753
1754         return status;
1755 }
1756
1757 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1758 {
1759         int status, namelen;
1760
1761         mlog_entry_void();
1762
1763         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1764                            (long long)blkno);
1765         if (namelen <= 0) {
1766                 if (namelen)
1767                         status = namelen;
1768                 else
1769                         status = -EINVAL;
1770                 mlog_errno(status);
1771                 goto bail;
1772         }
1773         if (namelen != OCFS2_ORPHAN_NAMELEN) {
1774                 status = -EINVAL;
1775                 mlog_errno(status);
1776                 goto bail;
1777         }
1778
1779         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1780              namelen);
1781
1782         status = 0;
1783 bail:
1784         mlog_exit(status);
1785         return status;
1786 }
1787
1788 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1789                                     struct inode **ret_orphan_dir,
1790                                     struct inode *inode,
1791                                     char *name,
1792                                     struct buffer_head **de_bh)
1793 {
1794         struct inode *orphan_dir_inode;
1795         struct buffer_head *orphan_dir_bh = NULL;
1796         int status = 0;
1797
1798         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1799         if (status < 0) {
1800                 mlog_errno(status);
1801                 return status;
1802         }
1803
1804         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1805                                                        ORPHAN_DIR_SYSTEM_INODE,
1806                                                        osb->slot_num);
1807         if (!orphan_dir_inode) {
1808                 status = -ENOENT;
1809                 mlog_errno(status);
1810                 return status;
1811         }
1812
1813         mutex_lock(&orphan_dir_inode->i_mutex);
1814
1815         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1816         if (status < 0) {
1817                 mlog_errno(status);
1818                 goto leave;
1819         }
1820
1821         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1822                                               orphan_dir_bh, name,
1823                                               OCFS2_ORPHAN_NAMELEN, de_bh);
1824         if (status < 0) {
1825                 ocfs2_inode_unlock(orphan_dir_inode, 1);
1826
1827                 mlog_errno(status);
1828                 goto leave;
1829         }
1830
1831         *ret_orphan_dir = orphan_dir_inode;
1832
1833 leave:
1834         if (status) {
1835                 mutex_unlock(&orphan_dir_inode->i_mutex);
1836                 iput(orphan_dir_inode);
1837         }
1838
1839         brelse(orphan_dir_bh);
1840
1841         mlog_exit(status);
1842         return status;
1843 }
1844
1845 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1846                             handle_t *handle,
1847                             struct inode *inode,
1848                             struct ocfs2_dinode *fe,
1849                             char *name,
1850                             struct buffer_head *de_bh,
1851                             struct inode *orphan_dir_inode)
1852 {
1853         struct buffer_head *orphan_dir_bh = NULL;
1854         int status = 0;
1855         struct ocfs2_dinode *orphan_fe;
1856
1857         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
1858
1859         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
1860         if (status < 0) {
1861                 mlog_errno(status);
1862                 goto leave;
1863         }
1864
1865         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
1866                                       OCFS2_JOURNAL_ACCESS_WRITE);
1867         if (status < 0) {
1868                 mlog_errno(status);
1869                 goto leave;
1870         }
1871
1872         /* we're a cluster, and nlink can change on disk from
1873          * underneath us... */
1874         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1875         if (S_ISDIR(inode->i_mode))
1876                 le16_add_cpu(&orphan_fe->i_links_count, 1);
1877         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1878
1879         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1880         if (status < 0) {
1881                 mlog_errno(status);
1882                 goto leave;
1883         }
1884
1885         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
1886                                    OCFS2_ORPHAN_NAMELEN, inode,
1887                                    OCFS2_I(inode)->ip_blkno,
1888                                    orphan_dir_bh, de_bh);
1889         if (status < 0) {
1890                 mlog_errno(status);
1891                 goto leave;
1892         }
1893
1894         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
1895
1896         /* Record which orphan dir our inode now resides
1897          * in. delete_inode will use this to determine which orphan
1898          * dir to lock. */
1899         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
1900
1901         mlog(0, "Inode %llu orphaned in slot %d\n",
1902              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
1903
1904 leave:
1905         brelse(orphan_dir_bh);
1906
1907         mlog_exit(status);
1908         return status;
1909 }
1910
1911 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
1912 int ocfs2_orphan_del(struct ocfs2_super *osb,
1913                      handle_t *handle,
1914                      struct inode *orphan_dir_inode,
1915                      struct inode *inode,
1916                      struct buffer_head *orphan_dir_bh)
1917 {
1918         char name[OCFS2_ORPHAN_NAMELEN + 1];
1919         struct ocfs2_dinode *orphan_fe;
1920         int status = 0;
1921         struct buffer_head *target_de_bh = NULL;
1922         struct ocfs2_dir_entry *target_de = NULL;
1923
1924         mlog_entry_void();
1925
1926         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
1927         if (status < 0) {
1928                 mlog_errno(status);
1929                 goto leave;
1930         }
1931
1932         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
1933              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
1934              OCFS2_ORPHAN_NAMELEN);
1935
1936         /* find it's spot in the orphan directory */
1937         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
1938                                         orphan_dir_inode, &target_de);
1939         if (!target_de_bh) {
1940                 status = -ENOENT;
1941                 mlog_errno(status);
1942                 goto leave;
1943         }
1944
1945         /* remove it from the orphan directory */
1946         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
1947                                     target_de_bh);
1948         if (status < 0) {
1949                 mlog_errno(status);
1950                 goto leave;
1951         }
1952
1953         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
1954                                       OCFS2_JOURNAL_ACCESS_WRITE);
1955         if (status < 0) {
1956                 mlog_errno(status);
1957                 goto leave;
1958         }
1959
1960         /* do the i_nlink dance! :) */
1961         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
1962         if (S_ISDIR(inode->i_mode))
1963                 le16_add_cpu(&orphan_fe->i_links_count, -1);
1964         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
1965
1966         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
1967         if (status < 0) {
1968                 mlog_errno(status);
1969                 goto leave;
1970         }
1971
1972 leave:
1973         brelse(target_de_bh);
1974
1975         mlog_exit(status);
1976         return status;
1977 }
1978
1979 const struct inode_operations ocfs2_dir_iops = {
1980         .create         = ocfs2_create,
1981         .lookup         = ocfs2_lookup,
1982         .link           = ocfs2_link,
1983         .unlink         = ocfs2_unlink,
1984         .rmdir          = ocfs2_unlink,
1985         .symlink        = ocfs2_symlink,
1986         .mkdir          = ocfs2_mkdir,
1987         .mknod          = ocfs2_mknod,
1988         .rename         = ocfs2_rename,
1989         .setattr        = ocfs2_setattr,
1990         .getattr        = ocfs2_getattr,
1991         .permission     = ocfs2_permission,
1992         .setxattr       = generic_setxattr,
1993         .getxattr       = generic_getxattr,
1994         .listxattr      = ocfs2_listxattr,
1995         .removexattr    = generic_removexattr,
1996 };