ARM: shmobile: Remove FSF address from copyright headers
[cascardo/linux.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 #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 #include "ocfs2_trace.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                               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                                     u64 blkno,
81                                     char *name,
82                                     struct ocfs2_dir_lookup_result *lookup);
83
84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
85                             handle_t *handle,
86                             struct inode *inode,
87                             struct buffer_head *fe_bh,
88                             char *name,
89                             struct ocfs2_dir_lookup_result *lookup,
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                                    unsigned int flags)
102 {
103         int status;
104         u64 blkno;
105         struct inode *inode = NULL;
106         struct dentry *ret;
107         struct ocfs2_inode_info *oi;
108
109         trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
110                            dentry->d_name.name,
111                            (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
112
113         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
114                 ret = ERR_PTR(-ENAMETOOLONG);
115                 goto bail;
116         }
117
118         status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
119         if (status < 0) {
120                 if (status != -ENOENT)
121                         mlog_errno(status);
122                 ret = ERR_PTR(status);
123                 goto bail;
124         }
125
126         status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
127                                             dentry->d_name.len, &blkno);
128         if (status < 0)
129                 goto bail_add;
130
131         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
132         if (IS_ERR(inode)) {
133                 ret = ERR_PTR(-EACCES);
134                 goto bail_unlock;
135         }
136
137         oi = OCFS2_I(inode);
138         /* Clear any orphaned state... If we were able to look up the
139          * inode from a directory, it certainly can't be orphaned. We
140          * might have the bad state from a node which intended to
141          * orphan this inode but crashed before it could commit the
142          * unlink. */
143         spin_lock(&oi->ip_lock);
144         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
145         spin_unlock(&oi->ip_lock);
146
147 bail_add:
148         ret = d_splice_alias(inode, dentry);
149
150         if (inode) {
151                 /*
152                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
153                  * dentry, it will d_move() it on top of ourse. The
154                  * return value will indicate this however, so in
155                  * those cases, we switch them around for the locking
156                  * code.
157                  *
158                  * NOTE: This dentry already has ->d_op set from
159                  * ocfs2_get_parent() and ocfs2_get_dentry()
160                  */
161                 if (ret)
162                         dentry = ret;
163
164                 status = ocfs2_dentry_attach_lock(dentry, inode,
165                                                   OCFS2_I(dir)->ip_blkno);
166                 if (status) {
167                         mlog_errno(status);
168                         ret = ERR_PTR(status);
169                         goto bail_unlock;
170                 }
171         } else
172                 ocfs2_dentry_attach_gen(dentry);
173
174 bail_unlock:
175         /* Don't drop the cluster lock until *after* the d_add --
176          * unlink on another node will message us to remove that
177          * dentry under this lock so otherwise we can race this with
178          * the downconvert thread and have a stale dentry. */
179         ocfs2_inode_unlock(dir, 0);
180
181 bail:
182
183         trace_ocfs2_lookup_ret(ret);
184
185         return ret;
186 }
187
188 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
189 {
190         struct inode *inode;
191
192         inode = new_inode(dir->i_sb);
193         if (!inode) {
194                 mlog(ML_ERROR, "new_inode failed!\n");
195                 return NULL;
196         }
197
198         /* populate as many fields early on as possible - many of
199          * these are used by the support functions here and in
200          * callers. */
201         if (S_ISDIR(mode))
202                 set_nlink(inode, 2);
203         inode_init_owner(inode, dir, mode);
204         dquot_initialize(inode);
205         return inode;
206 }
207
208 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
209                 struct dentry *dentry, struct inode *inode)
210 {
211         struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
212
213         ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
214         ocfs2_lock_res_free(&dl->dl_lockres);
215         BUG_ON(dl->dl_count != 1);
216         spin_lock(&dentry_attach_lock);
217         dentry->d_fsdata = NULL;
218         spin_unlock(&dentry_attach_lock);
219         kfree(dl);
220         iput(inode);
221 }
222
223 static int ocfs2_mknod(struct inode *dir,
224                        struct dentry *dentry,
225                        umode_t mode,
226                        dev_t dev)
227 {
228         int status = 0;
229         struct buffer_head *parent_fe_bh = NULL;
230         handle_t *handle = NULL;
231         struct ocfs2_super *osb;
232         struct ocfs2_dinode *dirfe;
233         struct buffer_head *new_fe_bh = NULL;
234         struct inode *inode = NULL;
235         struct ocfs2_alloc_context *inode_ac = NULL;
236         struct ocfs2_alloc_context *data_ac = NULL;
237         struct ocfs2_alloc_context *meta_ac = NULL;
238         int want_clusters = 0;
239         int want_meta = 0;
240         int xattr_credits = 0;
241         struct ocfs2_security_xattr_info si = {
242                 .enable = 1,
243         };
244         int did_quota_inode = 0;
245         struct ocfs2_dir_lookup_result lookup = { NULL, };
246         sigset_t oldset;
247         int did_block_signals = 0;
248         struct posix_acl *default_acl = NULL, *acl = NULL;
249         struct ocfs2_dentry_lock *dl = NULL;
250
251         trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
252                           (unsigned long long)OCFS2_I(dir)->ip_blkno,
253                           (unsigned long)dev, mode);
254
255         dquot_initialize(dir);
256
257         /* get our super block */
258         osb = OCFS2_SB(dir->i_sb);
259
260         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
261         if (status < 0) {
262                 if (status != -ENOENT)
263                         mlog_errno(status);
264                 return status;
265         }
266
267         if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
268                 status = -EMLINK;
269                 goto leave;
270         }
271
272         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
273         if (!ocfs2_read_links_count(dirfe)) {
274                 /* can't make a file in a deleted directory. */
275                 status = -ENOENT;
276                 goto leave;
277         }
278
279         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
280                                            dentry->d_name.len);
281         if (status)
282                 goto leave;
283
284         /* get a spot inside the dir. */
285         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
286                                               dentry->d_name.name,
287                                               dentry->d_name.len, &lookup);
288         if (status < 0) {
289                 mlog_errno(status);
290                 goto leave;
291         }
292
293         /* reserve an inode spot */
294         status = ocfs2_reserve_new_inode(osb, &inode_ac);
295         if (status < 0) {
296                 if (status != -ENOSPC)
297                         mlog_errno(status);
298                 goto leave;
299         }
300
301         inode = ocfs2_get_init_inode(dir, mode);
302         if (!inode) {
303                 status = -ENOMEM;
304                 mlog_errno(status);
305                 goto leave;
306         }
307
308         /* get security xattr */
309         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
310         if (status) {
311                 if (status == -EOPNOTSUPP)
312                         si.enable = 0;
313                 else {
314                         mlog_errno(status);
315                         goto leave;
316                 }
317         }
318
319         /* calculate meta data/clusters for setting security and acl xattr */
320         status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
321                                        &si, &want_clusters,
322                                        &xattr_credits, &want_meta);
323         if (status < 0) {
324                 mlog_errno(status);
325                 goto leave;
326         }
327
328         /* Reserve a cluster if creating an extent based directory. */
329         if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
330                 want_clusters += 1;
331
332                 /* Dir indexing requires extra space as well */
333                 if (ocfs2_supports_indexed_dirs(osb))
334                         want_meta++;
335         }
336
337         status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
338         if (status < 0) {
339                 if (status != -ENOSPC)
340                         mlog_errno(status);
341                 goto leave;
342         }
343
344         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
345         if (status < 0) {
346                 if (status != -ENOSPC)
347                         mlog_errno(status);
348                 goto leave;
349         }
350
351         status = posix_acl_create(dir, &mode, &default_acl, &acl);
352         if (status) {
353                 mlog_errno(status);
354                 goto leave;
355         }
356
357         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
358                                                             S_ISDIR(mode),
359                                                             xattr_credits));
360         if (IS_ERR(handle)) {
361                 status = PTR_ERR(handle);
362                 handle = NULL;
363                 mlog_errno(status);
364                 goto leave;
365         }
366
367         /* Starting to change things, restart is no longer possible. */
368         ocfs2_block_signals(&oldset);
369         did_block_signals = 1;
370
371         status = dquot_alloc_inode(inode);
372         if (status)
373                 goto leave;
374         did_quota_inode = 1;
375
376         /* do the real work now. */
377         status = ocfs2_mknod_locked(osb, dir, inode, dev,
378                                     &new_fe_bh, parent_fe_bh, handle,
379                                     inode_ac);
380         if (status < 0) {
381                 mlog_errno(status);
382                 goto leave;
383         }
384
385         if (S_ISDIR(mode)) {
386                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
387                                             new_fe_bh, data_ac, meta_ac);
388                 if (status < 0) {
389                         mlog_errno(status);
390                         goto leave;
391                 }
392
393                 status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
394                                                  parent_fe_bh,
395                                                  OCFS2_JOURNAL_ACCESS_WRITE);
396                 if (status < 0) {
397                         mlog_errno(status);
398                         goto leave;
399                 }
400                 ocfs2_add_links_count(dirfe, 1);
401                 ocfs2_journal_dirty(handle, parent_fe_bh);
402                 inc_nlink(dir);
403         }
404
405         if (default_acl) {
406                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
407                                        ACL_TYPE_DEFAULT, default_acl,
408                                        meta_ac, data_ac);
409         }
410         if (!status && acl) {
411                 status = ocfs2_set_acl(handle, inode, new_fe_bh,
412                                        ACL_TYPE_ACCESS, acl,
413                                        meta_ac, data_ac);
414         }
415
416         if (status < 0) {
417                 mlog_errno(status);
418                 goto leave;
419         }
420
421         if (si.enable) {
422                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
423                                                  meta_ac, data_ac);
424                 if (status < 0) {
425                         mlog_errno(status);
426                         goto leave;
427                 }
428         }
429
430         /*
431          * Do this before adding the entry to the directory. We add
432          * also set d_op after success so that ->d_iput() will cleanup
433          * the dentry lock even if ocfs2_add_entry() fails below.
434          */
435         status = ocfs2_dentry_attach_lock(dentry, inode,
436                                           OCFS2_I(dir)->ip_blkno);
437         if (status) {
438                 mlog_errno(status);
439                 goto leave;
440         }
441
442         dl = dentry->d_fsdata;
443
444         status = ocfs2_add_entry(handle, dentry, inode,
445                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
446                                  &lookup);
447         if (status < 0) {
448                 mlog_errno(status);
449                 goto leave;
450         }
451
452         insert_inode_hash(inode);
453         d_instantiate(dentry, inode);
454         status = 0;
455 leave:
456         if (default_acl)
457                 posix_acl_release(default_acl);
458         if (acl)
459                 posix_acl_release(acl);
460         if (status < 0 && did_quota_inode)
461                 dquot_free_inode(inode);
462         if (handle)
463                 ocfs2_commit_trans(osb, handle);
464
465         ocfs2_inode_unlock(dir, 1);
466         if (did_block_signals)
467                 ocfs2_unblock_signals(&oldset);
468
469         brelse(new_fe_bh);
470         brelse(parent_fe_bh);
471         kfree(si.value);
472
473         ocfs2_free_dir_lookup_result(&lookup);
474
475         if (inode_ac)
476                 ocfs2_free_alloc_context(inode_ac);
477
478         if (data_ac)
479                 ocfs2_free_alloc_context(data_ac);
480
481         if (meta_ac)
482                 ocfs2_free_alloc_context(meta_ac);
483
484         /*
485          * We should call iput after the i_mutex of the bitmap been
486          * unlocked in ocfs2_free_alloc_context, or the
487          * ocfs2_delete_inode will mutex_lock again.
488          */
489         if ((status < 0) && inode) {
490                 if (dl)
491                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
492
493                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
494                 clear_nlink(inode);
495                 iput(inode);
496         }
497
498         if (status)
499                 mlog_errno(status);
500
501         return status;
502 }
503
504 static int __ocfs2_mknod_locked(struct inode *dir,
505                                 struct inode *inode,
506                                 dev_t dev,
507                                 struct buffer_head **new_fe_bh,
508                                 struct buffer_head *parent_fe_bh,
509                                 handle_t *handle,
510                                 struct ocfs2_alloc_context *inode_ac,
511                                 u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
512 {
513         int status = 0;
514         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
515         struct ocfs2_dinode *fe = NULL;
516         struct ocfs2_extent_list *fel;
517         u16 feat;
518         struct ocfs2_inode_info *oi = OCFS2_I(inode);
519
520         *new_fe_bh = NULL;
521
522         /* populate as many fields early on as possible - many of
523          * these are used by the support functions here and in
524          * callers. */
525         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
526         OCFS2_I(inode)->ip_blkno = fe_blkno;
527         spin_lock(&osb->osb_lock);
528         inode->i_generation = osb->s_next_generation++;
529         spin_unlock(&osb->osb_lock);
530
531         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
532         if (!*new_fe_bh) {
533                 status = -ENOMEM;
534                 mlog_errno(status);
535                 goto leave;
536         }
537         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
538
539         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
540                                          *new_fe_bh,
541                                          OCFS2_JOURNAL_ACCESS_CREATE);
542         if (status < 0) {
543                 mlog_errno(status);
544                 goto leave;
545         }
546
547         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
548         memset(fe, 0, osb->sb->s_blocksize);
549
550         fe->i_generation = cpu_to_le32(inode->i_generation);
551         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
552         fe->i_blkno = cpu_to_le64(fe_blkno);
553         fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
554         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
555         fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
556         fe->i_uid = cpu_to_le32(i_uid_read(inode));
557         fe->i_gid = cpu_to_le32(i_gid_read(inode));
558         fe->i_mode = cpu_to_le16(inode->i_mode);
559         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
560                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
561
562         ocfs2_set_links_count(fe, inode->i_nlink);
563
564         fe->i_last_eb_blk = 0;
565         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
566         fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
567         fe->i_atime = fe->i_ctime = fe->i_mtime =
568                 cpu_to_le64(CURRENT_TIME.tv_sec);
569         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
570                 cpu_to_le32(CURRENT_TIME.tv_nsec);
571         fe->i_dtime = 0;
572
573         /*
574          * If supported, directories start with inline data. If inline
575          * isn't supported, but indexing is, we start them as indexed.
576          */
577         feat = le16_to_cpu(fe->i_dyn_features);
578         if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
579                 fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
580
581                 fe->id2.i_data.id_count = cpu_to_le16(
582                                 ocfs2_max_inline_data_with_xattr(osb->sb, fe));
583         } else {
584                 fel = &fe->id2.i_list;
585                 fel->l_tree_depth = 0;
586                 fel->l_next_free_rec = 0;
587                 fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
588         }
589
590         ocfs2_journal_dirty(handle, *new_fe_bh);
591
592         ocfs2_populate_inode(inode, fe, 1);
593         ocfs2_ci_set_new(osb, INODE_CACHE(inode));
594         if (!ocfs2_mount_local(osb)) {
595                 status = ocfs2_create_new_inode_locks(inode);
596                 if (status < 0)
597                         mlog_errno(status);
598         }
599
600         oi->i_sync_tid = handle->h_transaction->t_tid;
601         oi->i_datasync_tid = handle->h_transaction->t_tid;
602
603 leave:
604         if (status < 0) {
605                 if (*new_fe_bh) {
606                         brelse(*new_fe_bh);
607                         *new_fe_bh = NULL;
608                 }
609         }
610
611         if (status)
612                 mlog_errno(status);
613         return status;
614 }
615
616 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
617                               struct inode *dir,
618                               struct inode *inode,
619                               dev_t dev,
620                               struct buffer_head **new_fe_bh,
621                               struct buffer_head *parent_fe_bh,
622                               handle_t *handle,
623                               struct ocfs2_alloc_context *inode_ac)
624 {
625         int status = 0;
626         u64 suballoc_loc, fe_blkno = 0;
627         u16 suballoc_bit;
628
629         *new_fe_bh = NULL;
630
631         status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
632                                        inode_ac, &suballoc_loc,
633                                        &suballoc_bit, &fe_blkno);
634         if (status < 0) {
635                 mlog_errno(status);
636                 return status;
637         }
638
639         return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
640                                     parent_fe_bh, handle, inode_ac,
641                                     fe_blkno, suballoc_loc, suballoc_bit);
642 }
643
644 static int ocfs2_mkdir(struct inode *dir,
645                        struct dentry *dentry,
646                        umode_t mode)
647 {
648         int ret;
649
650         trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
651                           OCFS2_I(dir)->ip_blkno, mode);
652         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
653         if (ret)
654                 mlog_errno(ret);
655
656         return ret;
657 }
658
659 static int ocfs2_create(struct inode *dir,
660                         struct dentry *dentry,
661                         umode_t mode,
662                         bool excl)
663 {
664         int ret;
665
666         trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
667                            (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
668         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
669         if (ret)
670                 mlog_errno(ret);
671
672         return ret;
673 }
674
675 static int ocfs2_link(struct dentry *old_dentry,
676                       struct inode *dir,
677                       struct dentry *dentry)
678 {
679         handle_t *handle;
680         struct inode *inode = old_dentry->d_inode;
681         int err;
682         struct buffer_head *fe_bh = NULL;
683         struct buffer_head *parent_fe_bh = NULL;
684         struct ocfs2_dinode *fe = NULL;
685         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
686         struct ocfs2_dir_lookup_result lookup = { NULL, };
687         sigset_t oldset;
688         u64 old_de_ino;
689
690         trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
691                          old_dentry->d_name.len, old_dentry->d_name.name,
692                          dentry->d_name.len, dentry->d_name.name);
693
694         if (S_ISDIR(inode->i_mode))
695                 return -EPERM;
696
697         dquot_initialize(dir);
698
699         err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
700         if (err < 0) {
701                 if (err != -ENOENT)
702                         mlog_errno(err);
703                 return err;
704         }
705
706         if (!dir->i_nlink) {
707                 err = -ENOENT;
708                 goto out;
709         }
710
711         err = ocfs2_lookup_ino_from_name(dir, old_dentry->d_name.name,
712                         old_dentry->d_name.len, &old_de_ino);
713         if (err) {
714                 err = -ENOENT;
715                 goto out;
716         }
717
718         /*
719          * Check whether another node removed the source inode while we
720          * were in the vfs.
721          */
722         if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
723                 err = -ENOENT;
724                 goto out;
725         }
726
727         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
728                                         dentry->d_name.len);
729         if (err)
730                 goto out;
731
732         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
733                                            dentry->d_name.name,
734                                            dentry->d_name.len, &lookup);
735         if (err < 0) {
736                 mlog_errno(err);
737                 goto out;
738         }
739
740         err = ocfs2_inode_lock(inode, &fe_bh, 1);
741         if (err < 0) {
742                 if (err != -ENOENT)
743                         mlog_errno(err);
744                 goto out;
745         }
746
747         fe = (struct ocfs2_dinode *) fe_bh->b_data;
748         if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
749                 err = -EMLINK;
750                 goto out_unlock_inode;
751         }
752
753         handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
754         if (IS_ERR(handle)) {
755                 err = PTR_ERR(handle);
756                 handle = NULL;
757                 mlog_errno(err);
758                 goto out_unlock_inode;
759         }
760
761         /* Starting to change things, restart is no longer possible. */
762         ocfs2_block_signals(&oldset);
763
764         err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
765                                       OCFS2_JOURNAL_ACCESS_WRITE);
766         if (err < 0) {
767                 mlog_errno(err);
768                 goto out_commit;
769         }
770
771         inc_nlink(inode);
772         inode->i_ctime = CURRENT_TIME;
773         ocfs2_set_links_count(fe, inode->i_nlink);
774         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
775         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
776         ocfs2_journal_dirty(handle, fe_bh);
777
778         err = ocfs2_add_entry(handle, dentry, inode,
779                               OCFS2_I(inode)->ip_blkno,
780                               parent_fe_bh, &lookup);
781         if (err) {
782                 ocfs2_add_links_count(fe, -1);
783                 drop_nlink(inode);
784                 mlog_errno(err);
785                 goto out_commit;
786         }
787
788         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
789         if (err) {
790                 mlog_errno(err);
791                 goto out_commit;
792         }
793
794         ihold(inode);
795         d_instantiate(dentry, inode);
796
797 out_commit:
798         ocfs2_commit_trans(osb, handle);
799         ocfs2_unblock_signals(&oldset);
800 out_unlock_inode:
801         ocfs2_inode_unlock(inode, 1);
802
803 out:
804         ocfs2_inode_unlock(dir, 1);
805
806         brelse(fe_bh);
807         brelse(parent_fe_bh);
808
809         ocfs2_free_dir_lookup_result(&lookup);
810
811         if (err)
812                 mlog_errno(err);
813
814         return err;
815 }
816
817 /*
818  * Takes and drops an exclusive lock on the given dentry. This will
819  * force other nodes to drop it.
820  */
821 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
822 {
823         int ret;
824
825         ret = ocfs2_dentry_lock(dentry, 1);
826         if (ret)
827                 mlog_errno(ret);
828         else
829                 ocfs2_dentry_unlock(dentry, 1);
830
831         return ret;
832 }
833
834 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
835 {
836         if (S_ISDIR(inode->i_mode)) {
837                 if (inode->i_nlink == 2)
838                         return 1;
839                 return 0;
840         }
841
842         if (inode->i_nlink == 1)
843                 return 1;
844         return 0;
845 }
846
847 static int ocfs2_unlink(struct inode *dir,
848                         struct dentry *dentry)
849 {
850         int status;
851         int child_locked = 0;
852         bool is_unlinkable = false;
853         struct inode *inode = dentry->d_inode;
854         struct inode *orphan_dir = NULL;
855         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
856         u64 blkno;
857         struct ocfs2_dinode *fe = NULL;
858         struct buffer_head *fe_bh = NULL;
859         struct buffer_head *parent_node_bh = NULL;
860         handle_t *handle = NULL;
861         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
862         struct ocfs2_dir_lookup_result lookup = { NULL, };
863         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
864
865         trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
866                            dentry->d_name.name,
867                            (unsigned long long)OCFS2_I(dir)->ip_blkno,
868                            (unsigned long long)OCFS2_I(inode)->ip_blkno);
869
870         dquot_initialize(dir);
871
872         BUG_ON(dentry->d_parent->d_inode != dir);
873
874         if (inode == osb->root_inode)
875                 return -EPERM;
876
877         status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
878                                          OI_LS_PARENT);
879         if (status < 0) {
880                 if (status != -ENOENT)
881                         mlog_errno(status);
882                 return status;
883         }
884
885         status = ocfs2_find_files_on_disk(dentry->d_name.name,
886                                           dentry->d_name.len, &blkno, dir,
887                                           &lookup);
888         if (status < 0) {
889                 if (status != -ENOENT)
890                         mlog_errno(status);
891                 goto leave;
892         }
893
894         if (OCFS2_I(inode)->ip_blkno != blkno) {
895                 status = -ENOENT;
896
897                 trace_ocfs2_unlink_noent(
898                                 (unsigned long long)OCFS2_I(inode)->ip_blkno,
899                                 (unsigned long long)blkno,
900                                 OCFS2_I(inode)->ip_flags);
901                 goto leave;
902         }
903
904         status = ocfs2_inode_lock(inode, &fe_bh, 1);
905         if (status < 0) {
906                 if (status != -ENOENT)
907                         mlog_errno(status);
908                 goto leave;
909         }
910         child_locked = 1;
911
912         if (S_ISDIR(inode->i_mode)) {
913                 if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
914                         status = -ENOTEMPTY;
915                         goto leave;
916                 }
917         }
918
919         status = ocfs2_remote_dentry_delete(dentry);
920         if (status < 0) {
921                 /* This remote delete should succeed under all normal
922                  * circumstances. */
923                 mlog_errno(status);
924                 goto leave;
925         }
926
927         if (ocfs2_inode_is_unlinkable(inode)) {
928                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
929                                                   OCFS2_I(inode)->ip_blkno,
930                                                   orphan_name, &orphan_insert);
931                 if (status < 0) {
932                         mlog_errno(status);
933                         goto leave;
934                 }
935                 is_unlinkable = true;
936         }
937
938         handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
939         if (IS_ERR(handle)) {
940                 status = PTR_ERR(handle);
941                 handle = NULL;
942                 mlog_errno(status);
943                 goto leave;
944         }
945
946         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
947                                          OCFS2_JOURNAL_ACCESS_WRITE);
948         if (status < 0) {
949                 mlog_errno(status);
950                 goto leave;
951         }
952
953         fe = (struct ocfs2_dinode *) fe_bh->b_data;
954
955         /* delete the name from the parent dir */
956         status = ocfs2_delete_entry(handle, dir, &lookup);
957         if (status < 0) {
958                 mlog_errno(status);
959                 goto leave;
960         }
961
962         if (S_ISDIR(inode->i_mode))
963                 drop_nlink(inode);
964         drop_nlink(inode);
965         ocfs2_set_links_count(fe, inode->i_nlink);
966         ocfs2_journal_dirty(handle, fe_bh);
967
968         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
969         if (S_ISDIR(inode->i_mode))
970                 drop_nlink(dir);
971
972         status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
973         if (status < 0) {
974                 mlog_errno(status);
975                 if (S_ISDIR(inode->i_mode))
976                         inc_nlink(dir);
977                 goto leave;
978         }
979
980         if (is_unlinkable) {
981                 status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
982                                 orphan_name, &orphan_insert, orphan_dir);
983                 if (status < 0)
984                         mlog_errno(status);
985         }
986
987 leave:
988         if (handle)
989                 ocfs2_commit_trans(osb, handle);
990
991         if (child_locked)
992                 ocfs2_inode_unlock(inode, 1);
993
994         ocfs2_inode_unlock(dir, 1);
995
996         if (orphan_dir) {
997                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
998                 ocfs2_inode_unlock(orphan_dir, 1);
999                 mutex_unlock(&orphan_dir->i_mutex);
1000                 iput(orphan_dir);
1001         }
1002
1003         brelse(fe_bh);
1004         brelse(parent_node_bh);
1005
1006         ocfs2_free_dir_lookup_result(&orphan_insert);
1007         ocfs2_free_dir_lookup_result(&lookup);
1008
1009         if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1010                 mlog_errno(status);
1011
1012         return status;
1013 }
1014
1015 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1016                 u64 src_inode_no, u64 dest_inode_no)
1017 {
1018         int ret = 0, i = 0;
1019         u64 parent_inode_no = 0;
1020         u64 child_inode_no = src_inode_no;
1021         struct inode *child_inode;
1022
1023 #define MAX_LOOKUP_TIMES 32
1024         while (1) {
1025                 child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1026                 if (IS_ERR(child_inode)) {
1027                         ret = PTR_ERR(child_inode);
1028                         break;
1029                 }
1030
1031                 ret = ocfs2_inode_lock(child_inode, NULL, 0);
1032                 if (ret < 0) {
1033                         iput(child_inode);
1034                         if (ret != -ENOENT)
1035                                 mlog_errno(ret);
1036                         break;
1037                 }
1038
1039                 ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1040                                 &parent_inode_no);
1041                 ocfs2_inode_unlock(child_inode, 0);
1042                 iput(child_inode);
1043                 if (ret < 0) {
1044                         ret = -ENOENT;
1045                         break;
1046                 }
1047
1048                 if (parent_inode_no == dest_inode_no) {
1049                         ret = 1;
1050                         break;
1051                 }
1052
1053                 if (parent_inode_no == osb->root_inode->i_ino) {
1054                         ret = 0;
1055                         break;
1056                 }
1057
1058                 child_inode_no = parent_inode_no;
1059
1060                 if (++i >= MAX_LOOKUP_TIMES) {
1061                         mlog(ML_NOTICE, "max lookup times reached, filesystem "
1062                                         "may have nested directories, "
1063                                         "src inode: %llu, dest inode: %llu.\n",
1064                                         (unsigned long long)src_inode_no,
1065                                         (unsigned long long)dest_inode_no);
1066                         ret = 0;
1067                         break;
1068                 }
1069         }
1070
1071         return ret;
1072 }
1073
1074 /*
1075  * The only place this should be used is rename!
1076  * if they have the same id, then the 1st one is the only one locked.
1077  */
1078 static int ocfs2_double_lock(struct ocfs2_super *osb,
1079                              struct buffer_head **bh1,
1080                              struct inode *inode1,
1081                              struct buffer_head **bh2,
1082                              struct inode *inode2)
1083 {
1084         int status;
1085         int inode1_is_ancestor, inode2_is_ancestor;
1086         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1087         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1088         struct buffer_head **tmpbh;
1089         struct inode *tmpinode;
1090
1091         trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1092                                 (unsigned long long)oi2->ip_blkno);
1093
1094         if (*bh1)
1095                 *bh1 = NULL;
1096         if (*bh2)
1097                 *bh2 = NULL;
1098
1099         /* we always want to lock the one with the lower lockid first.
1100          * and if they are nested, we lock ancestor first */
1101         if (oi1->ip_blkno != oi2->ip_blkno) {
1102                 inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1103                                 oi1->ip_blkno);
1104                 if (inode1_is_ancestor < 0) {
1105                         status = inode1_is_ancestor;
1106                         goto bail;
1107                 }
1108
1109                 inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1110                                 oi2->ip_blkno);
1111                 if (inode2_is_ancestor < 0) {
1112                         status = inode2_is_ancestor;
1113                         goto bail;
1114                 }
1115
1116                 if ((inode1_is_ancestor == 1) ||
1117                                 (oi1->ip_blkno < oi2->ip_blkno &&
1118                                 inode2_is_ancestor == 0)) {
1119                         /* switch id1 and id2 around */
1120                         tmpbh = bh2;
1121                         bh2 = bh1;
1122                         bh1 = tmpbh;
1123
1124                         tmpinode = inode2;
1125                         inode2 = inode1;
1126                         inode1 = tmpinode;
1127                 }
1128                 /* lock id2 */
1129                 status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1130                                                  OI_LS_RENAME1);
1131                 if (status < 0) {
1132                         if (status != -ENOENT)
1133                                 mlog_errno(status);
1134                         goto bail;
1135                 }
1136         }
1137
1138         /* lock id1 */
1139         status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1140         if (status < 0) {
1141                 /*
1142                  * An error return must mean that no cluster locks
1143                  * were held on function exit.
1144                  */
1145                 if (oi1->ip_blkno != oi2->ip_blkno) {
1146                         ocfs2_inode_unlock(inode2, 1);
1147                         brelse(*bh2);
1148                         *bh2 = NULL;
1149                 }
1150
1151                 if (status != -ENOENT)
1152                         mlog_errno(status);
1153         }
1154
1155         trace_ocfs2_double_lock_end(
1156                         (unsigned long long)OCFS2_I(inode1)->ip_blkno,
1157                         (unsigned long long)OCFS2_I(inode2)->ip_blkno);
1158
1159 bail:
1160         if (status)
1161                 mlog_errno(status);
1162         return status;
1163 }
1164
1165 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1166 {
1167         ocfs2_inode_unlock(inode1, 1);
1168
1169         if (inode1 != inode2)
1170                 ocfs2_inode_unlock(inode2, 1);
1171 }
1172
1173 static int ocfs2_rename(struct inode *old_dir,
1174                         struct dentry *old_dentry,
1175                         struct inode *new_dir,
1176                         struct dentry *new_dentry)
1177 {
1178         int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1179         int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1180         struct inode *old_inode = old_dentry->d_inode;
1181         struct inode *new_inode = new_dentry->d_inode;
1182         struct inode *orphan_dir = NULL;
1183         struct ocfs2_dinode *newfe = NULL;
1184         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1185         struct buffer_head *newfe_bh = NULL;
1186         struct buffer_head *old_inode_bh = NULL;
1187         struct ocfs2_super *osb = NULL;
1188         u64 newfe_blkno, old_de_ino;
1189         handle_t *handle = NULL;
1190         struct buffer_head *old_dir_bh = NULL;
1191         struct buffer_head *new_dir_bh = NULL;
1192         u32 old_dir_nlink = old_dir->i_nlink;
1193         struct ocfs2_dinode *old_di;
1194         struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1195         struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1196         struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1197         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1198         struct ocfs2_dir_lookup_result target_insert = { NULL, };
1199         bool should_add_orphan = false;
1200
1201         /* At some point it might be nice to break this function up a
1202          * bit. */
1203
1204         trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1205                            old_dentry->d_name.len, old_dentry->d_name.name,
1206                            new_dentry->d_name.len, new_dentry->d_name.name);
1207
1208         dquot_initialize(old_dir);
1209         dquot_initialize(new_dir);
1210
1211         osb = OCFS2_SB(old_dir->i_sb);
1212
1213         if (new_inode) {
1214                 if (!igrab(new_inode))
1215                         BUG();
1216         }
1217
1218         /* Assume a directory hierarchy thusly:
1219          * a/b/c
1220          * a/d
1221          * a,b,c, and d are all directories.
1222          *
1223          * from cwd of 'a' on both nodes:
1224          * node1: mv b/c d
1225          * node2: mv d   b/c
1226          *
1227          * And that's why, just like the VFS, we need a file system
1228          * rename lock. */
1229         if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1230                 status = ocfs2_rename_lock(osb);
1231                 if (status < 0) {
1232                         mlog_errno(status);
1233                         goto bail;
1234                 }
1235                 rename_lock = 1;
1236
1237                 /* here we cannot guarantee the inodes haven't just been
1238                  * changed, so check if they are nested again */
1239                 status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1240                                 old_inode->i_ino);
1241                 if (status < 0) {
1242                         mlog_errno(status);
1243                         goto bail;
1244                 } else if (status == 1) {
1245                         status = -EPERM;
1246                         trace_ocfs2_rename_not_permitted(
1247                                         (unsigned long long)old_inode->i_ino,
1248                                         (unsigned long long)new_dir->i_ino);
1249                         goto bail;
1250                 }
1251         }
1252
1253         /* if old and new are the same, this'll just do one lock. */
1254         status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1255                                    &new_dir_bh, new_dir);
1256         if (status < 0) {
1257                 mlog_errno(status);
1258                 goto bail;
1259         }
1260         parents_locked = 1;
1261
1262         /* make sure both dirs have bhs
1263          * get an extra ref on old_dir_bh if old==new */
1264         if (!new_dir_bh) {
1265                 if (old_dir_bh) {
1266                         new_dir_bh = old_dir_bh;
1267                         get_bh(new_dir_bh);
1268                 } else {
1269                         mlog(ML_ERROR, "no old_dir_bh!\n");
1270                         status = -EIO;
1271                         goto bail;
1272                 }
1273         }
1274
1275         /*
1276          * Aside from allowing a meta data update, the locking here
1277          * also ensures that the downconvert thread on other nodes
1278          * won't have to concurrently downconvert the inode and the
1279          * dentry locks.
1280          */
1281         status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1282                                          OI_LS_PARENT);
1283         if (status < 0) {
1284                 if (status != -ENOENT)
1285                         mlog_errno(status);
1286                 goto bail;
1287         }
1288         old_child_locked = 1;
1289
1290         status = ocfs2_remote_dentry_delete(old_dentry);
1291         if (status < 0) {
1292                 mlog_errno(status);
1293                 goto bail;
1294         }
1295
1296         if (S_ISDIR(old_inode->i_mode)) {
1297                 u64 old_inode_parent;
1298
1299                 update_dot_dot = 1;
1300                 status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1301                                                   old_inode,
1302                                                   &old_inode_dot_dot_res);
1303                 if (status) {
1304                         status = -EIO;
1305                         goto bail;
1306                 }
1307
1308                 if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1309                         status = -EIO;
1310                         goto bail;
1311                 }
1312
1313                 if (!new_inode && new_dir != old_dir &&
1314                     new_dir->i_nlink >= ocfs2_link_max(osb)) {
1315                         status = -EMLINK;
1316                         goto bail;
1317                 }
1318         }
1319
1320         status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1321                                             old_dentry->d_name.len,
1322                                             &old_de_ino);
1323         if (status) {
1324                 status = -ENOENT;
1325                 goto bail;
1326         }
1327
1328         /*
1329          *  Check for inode number is _not_ due to possible IO errors.
1330          *  We might rmdir the source, keep it as pwd of some process
1331          *  and merrily kill the link to whatever was created under the
1332          *  same name. Goodbye sticky bit ;-<
1333          */
1334         if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1335                 status = -ENOENT;
1336                 goto bail;
1337         }
1338
1339         /* check if the target already exists (in which case we need
1340          * to delete it */
1341         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1342                                           new_dentry->d_name.len,
1343                                           &newfe_blkno, new_dir,
1344                                           &target_lookup_res);
1345         /* The only error we allow here is -ENOENT because the new
1346          * file not existing is perfectly valid. */
1347         if ((status < 0) && (status != -ENOENT)) {
1348                 /* If we cannot find the file specified we should just */
1349                 /* return the error... */
1350                 mlog_errno(status);
1351                 goto bail;
1352         }
1353         if (status == 0)
1354                 target_exists = 1;
1355
1356         if (!target_exists && new_inode) {
1357                 /*
1358                  * Target was unlinked by another node while we were
1359                  * waiting to get to ocfs2_rename(). There isn't
1360                  * anything we can do here to help the situation, so
1361                  * bubble up the appropriate error.
1362                  */
1363                 status = -ENOENT;
1364                 goto bail;
1365         }
1366
1367         /* In case we need to overwrite an existing file, we blow it
1368          * away first */
1369         if (target_exists) {
1370                 /* VFS didn't think there existed an inode here, but
1371                  * someone else in the cluster must have raced our
1372                  * rename to create one. Today we error cleanly, in
1373                  * the future we should consider calling iget to build
1374                  * a new struct inode for this entry. */
1375                 if (!new_inode) {
1376                         status = -EACCES;
1377
1378                         trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1379                                                 new_dentry->d_name.name);
1380                         goto bail;
1381                 }
1382
1383                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1384                         status = -EACCES;
1385
1386                         trace_ocfs2_rename_disagree(
1387                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1388                              (unsigned long long)newfe_blkno,
1389                              OCFS2_I(new_inode)->ip_flags);
1390                         goto bail;
1391                 }
1392
1393                 status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1394                 if (status < 0) {
1395                         if (status != -ENOENT)
1396                                 mlog_errno(status);
1397                         goto bail;
1398                 }
1399                 new_child_locked = 1;
1400
1401                 status = ocfs2_remote_dentry_delete(new_dentry);
1402                 if (status < 0) {
1403                         mlog_errno(status);
1404                         goto bail;
1405                 }
1406
1407                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1408
1409                 trace_ocfs2_rename_over_existing(
1410                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1411                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1412
1413                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1414                         status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1415                                                 OCFS2_I(new_inode)->ip_blkno,
1416                                                 orphan_name, &orphan_insert);
1417                         if (status < 0) {
1418                                 mlog_errno(status);
1419                                 goto bail;
1420                         }
1421                         should_add_orphan = true;
1422                 }
1423         } else {
1424                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1425
1426                 status = ocfs2_check_dir_for_entry(new_dir,
1427                                                    new_dentry->d_name.name,
1428                                                    new_dentry->d_name.len);
1429                 if (status)
1430                         goto bail;
1431
1432                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1433                                                       new_dentry->d_name.name,
1434                                                       new_dentry->d_name.len,
1435                                                       &target_insert);
1436                 if (status < 0) {
1437                         mlog_errno(status);
1438                         goto bail;
1439                 }
1440         }
1441
1442         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1443         if (IS_ERR(handle)) {
1444                 status = PTR_ERR(handle);
1445                 handle = NULL;
1446                 mlog_errno(status);
1447                 goto bail;
1448         }
1449
1450         if (target_exists) {
1451                 if (S_ISDIR(new_inode->i_mode)) {
1452                         if (new_inode->i_nlink != 2 ||
1453                             !ocfs2_empty_dir(new_inode)) {
1454                                 status = -ENOTEMPTY;
1455                                 goto bail;
1456                         }
1457                 }
1458                 status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1459                                                  newfe_bh,
1460                                                  OCFS2_JOURNAL_ACCESS_WRITE);
1461                 if (status < 0) {
1462                         mlog_errno(status);
1463                         goto bail;
1464                 }
1465
1466                 /* change the dirent to point to the correct inode */
1467                 status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1468                                             old_inode);
1469                 if (status < 0) {
1470                         mlog_errno(status);
1471                         goto bail;
1472                 }
1473                 new_dir->i_version++;
1474
1475                 if (S_ISDIR(new_inode->i_mode))
1476                         ocfs2_set_links_count(newfe, 0);
1477                 else
1478                         ocfs2_add_links_count(newfe, -1);
1479                 ocfs2_journal_dirty(handle, newfe_bh);
1480                 if (should_add_orphan) {
1481                         status = ocfs2_orphan_add(osb, handle, new_inode,
1482                                         newfe_bh, orphan_name,
1483                                         &orphan_insert, orphan_dir);
1484                         if (status < 0) {
1485                                 mlog_errno(status);
1486                                 goto bail;
1487                         }
1488                 }
1489         } else {
1490                 /* if the name was not found in new_dir, add it now */
1491                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1492                                          OCFS2_I(old_inode)->ip_blkno,
1493                                          new_dir_bh, &target_insert);
1494         }
1495
1496         old_inode->i_ctime = CURRENT_TIME;
1497         mark_inode_dirty(old_inode);
1498
1499         status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1500                                          old_inode_bh,
1501                                          OCFS2_JOURNAL_ACCESS_WRITE);
1502         if (status >= 0) {
1503                 old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1504
1505                 old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1506                 old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1507                 ocfs2_journal_dirty(handle, old_inode_bh);
1508         } else
1509                 mlog_errno(status);
1510
1511         /*
1512          * Now that the name has been added to new_dir, remove the old name.
1513          *
1514          * We don't keep any directory entry context around until now
1515          * because the insert might have changed the type of directory
1516          * we're dealing with.
1517          */
1518         status = ocfs2_find_entry(old_dentry->d_name.name,
1519                                   old_dentry->d_name.len, old_dir,
1520                                   &old_entry_lookup);
1521         if (status)
1522                 goto bail;
1523
1524         status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1525         if (status < 0) {
1526                 mlog_errno(status);
1527                 goto bail;
1528         }
1529
1530         if (new_inode) {
1531                 drop_nlink(new_inode);
1532                 new_inode->i_ctime = CURRENT_TIME;
1533         }
1534         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1535
1536         if (update_dot_dot) {
1537                 status = ocfs2_update_entry(old_inode, handle,
1538                                             &old_inode_dot_dot_res, new_dir);
1539                 drop_nlink(old_dir);
1540                 if (new_inode) {
1541                         drop_nlink(new_inode);
1542                 } else {
1543                         inc_nlink(new_dir);
1544                         mark_inode_dirty(new_dir);
1545                 }
1546         }
1547         mark_inode_dirty(old_dir);
1548         ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1549         if (new_inode) {
1550                 mark_inode_dirty(new_inode);
1551                 ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1552         }
1553
1554         if (old_dir != new_dir) {
1555                 /* Keep the same times on both directories.*/
1556                 new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1557
1558                 /*
1559                  * This will also pick up the i_nlink change from the
1560                  * block above.
1561                  */
1562                 ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1563         }
1564
1565         if (old_dir_nlink != old_dir->i_nlink) {
1566                 if (!old_dir_bh) {
1567                         mlog(ML_ERROR, "need to change nlink for old dir "
1568                              "%llu from %d to %d but bh is NULL!\n",
1569                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1570                              (int)old_dir_nlink, old_dir->i_nlink);
1571                 } else {
1572                         struct ocfs2_dinode *fe;
1573                         status = ocfs2_journal_access_di(handle,
1574                                                          INODE_CACHE(old_dir),
1575                                                          old_dir_bh,
1576                                                          OCFS2_JOURNAL_ACCESS_WRITE);
1577                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1578                         ocfs2_set_links_count(fe, old_dir->i_nlink);
1579                         ocfs2_journal_dirty(handle, old_dir_bh);
1580                 }
1581         }
1582         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1583         status = 0;
1584 bail:
1585         if (rename_lock)
1586                 ocfs2_rename_unlock(osb);
1587
1588         if (handle)
1589                 ocfs2_commit_trans(osb, handle);
1590
1591         if (parents_locked)
1592                 ocfs2_double_unlock(old_dir, new_dir);
1593
1594         if (old_child_locked)
1595                 ocfs2_inode_unlock(old_inode, 1);
1596
1597         if (new_child_locked)
1598                 ocfs2_inode_unlock(new_inode, 1);
1599
1600         if (orphan_dir) {
1601                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1602                 ocfs2_inode_unlock(orphan_dir, 1);
1603                 mutex_unlock(&orphan_dir->i_mutex);
1604                 iput(orphan_dir);
1605         }
1606
1607         if (new_inode)
1608                 sync_mapping_buffers(old_inode->i_mapping);
1609
1610         if (new_inode)
1611                 iput(new_inode);
1612
1613         ocfs2_free_dir_lookup_result(&target_lookup_res);
1614         ocfs2_free_dir_lookup_result(&old_entry_lookup);
1615         ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1616         ocfs2_free_dir_lookup_result(&orphan_insert);
1617         ocfs2_free_dir_lookup_result(&target_insert);
1618
1619         brelse(newfe_bh);
1620         brelse(old_inode_bh);
1621         brelse(old_dir_bh);
1622         brelse(new_dir_bh);
1623
1624         if (status)
1625                 mlog_errno(status);
1626
1627         return status;
1628 }
1629
1630 /*
1631  * we expect i_size = strlen(symname). Copy symname into the file
1632  * data, including the null terminator.
1633  */
1634 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1635                                      handle_t *handle,
1636                                      struct inode *inode,
1637                                      const char *symname)
1638 {
1639         struct buffer_head **bhs = NULL;
1640         const char *c;
1641         struct super_block *sb = osb->sb;
1642         u64 p_blkno, p_blocks;
1643         int virtual, blocks, status, i, bytes_left;
1644
1645         bytes_left = i_size_read(inode) + 1;
1646         /* we can't trust i_blocks because we're actually going to
1647          * write i_size + 1 bytes. */
1648         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1649
1650         trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1651                                         i_size_read(inode), blocks);
1652
1653         /* Sanity check -- make sure we're going to fit. */
1654         if (bytes_left >
1655             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1656                 status = -EIO;
1657                 mlog_errno(status);
1658                 goto bail;
1659         }
1660
1661         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1662         if (!bhs) {
1663                 status = -ENOMEM;
1664                 mlog_errno(status);
1665                 goto bail;
1666         }
1667
1668         status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1669                                              NULL);
1670         if (status < 0) {
1671                 mlog_errno(status);
1672                 goto bail;
1673         }
1674
1675         /* links can never be larger than one cluster so we know this
1676          * is all going to be contiguous, but do a sanity check
1677          * anyway. */
1678         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1679                 status = -EIO;
1680                 mlog_errno(status);
1681                 goto bail;
1682         }
1683
1684         virtual = 0;
1685         while(bytes_left > 0) {
1686                 c = &symname[virtual * sb->s_blocksize];
1687
1688                 bhs[virtual] = sb_getblk(sb, p_blkno);
1689                 if (!bhs[virtual]) {
1690                         status = -ENOMEM;
1691                         mlog_errno(status);
1692                         goto bail;
1693                 }
1694                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1695                                               bhs[virtual]);
1696
1697                 status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1698                                               bhs[virtual],
1699                                               OCFS2_JOURNAL_ACCESS_CREATE);
1700                 if (status < 0) {
1701                         mlog_errno(status);
1702                         goto bail;
1703                 }
1704
1705                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1706
1707                 memcpy(bhs[virtual]->b_data, c,
1708                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1709                        bytes_left);
1710
1711                 ocfs2_journal_dirty(handle, bhs[virtual]);
1712
1713                 virtual++;
1714                 p_blkno++;
1715                 bytes_left -= sb->s_blocksize;
1716         }
1717
1718         status = 0;
1719 bail:
1720
1721         if (bhs) {
1722                 for(i = 0; i < blocks; i++)
1723                         brelse(bhs[i]);
1724                 kfree(bhs);
1725         }
1726
1727         if (status)
1728                 mlog_errno(status);
1729         return status;
1730 }
1731
1732 static int ocfs2_symlink(struct inode *dir,
1733                          struct dentry *dentry,
1734                          const char *symname)
1735 {
1736         int status, l, credits;
1737         u64 newsize;
1738         struct ocfs2_super *osb = NULL;
1739         struct inode *inode = NULL;
1740         struct super_block *sb;
1741         struct buffer_head *new_fe_bh = NULL;
1742         struct buffer_head *parent_fe_bh = NULL;
1743         struct ocfs2_dinode *fe = NULL;
1744         struct ocfs2_dinode *dirfe;
1745         handle_t *handle = NULL;
1746         struct ocfs2_alloc_context *inode_ac = NULL;
1747         struct ocfs2_alloc_context *data_ac = NULL;
1748         struct ocfs2_alloc_context *xattr_ac = NULL;
1749         int want_clusters = 0;
1750         int xattr_credits = 0;
1751         struct ocfs2_security_xattr_info si = {
1752                 .enable = 1,
1753         };
1754         int did_quota = 0, did_quota_inode = 0;
1755         struct ocfs2_dir_lookup_result lookup = { NULL, };
1756         sigset_t oldset;
1757         int did_block_signals = 0;
1758         struct ocfs2_dentry_lock *dl = NULL;
1759
1760         trace_ocfs2_symlink_begin(dir, dentry, symname,
1761                                   dentry->d_name.len, dentry->d_name.name);
1762
1763         dquot_initialize(dir);
1764
1765         sb = dir->i_sb;
1766         osb = OCFS2_SB(sb);
1767
1768         l = strlen(symname) + 1;
1769
1770         credits = ocfs2_calc_symlink_credits(sb);
1771
1772         /* lock the parent directory */
1773         status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1774         if (status < 0) {
1775                 if (status != -ENOENT)
1776                         mlog_errno(status);
1777                 return status;
1778         }
1779
1780         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1781         if (!ocfs2_read_links_count(dirfe)) {
1782                 /* can't make a file in a deleted directory. */
1783                 status = -ENOENT;
1784                 goto bail;
1785         }
1786
1787         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1788                                            dentry->d_name.len);
1789         if (status)
1790                 goto bail;
1791
1792         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1793                                               dentry->d_name.name,
1794                                               dentry->d_name.len, &lookup);
1795         if (status < 0) {
1796                 mlog_errno(status);
1797                 goto bail;
1798         }
1799
1800         status = ocfs2_reserve_new_inode(osb, &inode_ac);
1801         if (status < 0) {
1802                 if (status != -ENOSPC)
1803                         mlog_errno(status);
1804                 goto bail;
1805         }
1806
1807         inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1808         if (!inode) {
1809                 status = -ENOMEM;
1810                 mlog_errno(status);
1811                 goto bail;
1812         }
1813
1814         /* get security xattr */
1815         status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1816         if (status) {
1817                 if (status == -EOPNOTSUPP)
1818                         si.enable = 0;
1819                 else {
1820                         mlog_errno(status);
1821                         goto bail;
1822                 }
1823         }
1824
1825         /* calculate meta data/clusters for setting security xattr */
1826         if (si.enable) {
1827                 status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1828                                                   &xattr_credits, &xattr_ac);
1829                 if (status < 0) {
1830                         mlog_errno(status);
1831                         goto bail;
1832                 }
1833         }
1834
1835         /* don't reserve bitmap space for fast symlinks. */
1836         if (l > ocfs2_fast_symlink_chars(sb))
1837                 want_clusters += 1;
1838
1839         status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1840         if (status < 0) {
1841                 if (status != -ENOSPC)
1842                         mlog_errno(status);
1843                 goto bail;
1844         }
1845
1846         handle = ocfs2_start_trans(osb, credits + xattr_credits);
1847         if (IS_ERR(handle)) {
1848                 status = PTR_ERR(handle);
1849                 handle = NULL;
1850                 mlog_errno(status);
1851                 goto bail;
1852         }
1853
1854         /* Starting to change things, restart is no longer possible. */
1855         ocfs2_block_signals(&oldset);
1856         did_block_signals = 1;
1857
1858         status = dquot_alloc_inode(inode);
1859         if (status)
1860                 goto bail;
1861         did_quota_inode = 1;
1862
1863         trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1864                                    dentry->d_name.name,
1865                                    (unsigned long long)OCFS2_I(dir)->ip_blkno,
1866                                    inode->i_mode);
1867
1868         status = ocfs2_mknod_locked(osb, dir, inode,
1869                                     0, &new_fe_bh, parent_fe_bh, handle,
1870                                     inode_ac);
1871         if (status < 0) {
1872                 mlog_errno(status);
1873                 goto bail;
1874         }
1875
1876         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1877         inode->i_rdev = 0;
1878         newsize = l - 1;
1879         inode->i_op = &ocfs2_symlink_inode_operations;
1880         if (l > ocfs2_fast_symlink_chars(sb)) {
1881                 u32 offset = 0;
1882
1883                 status = dquot_alloc_space_nodirty(inode,
1884                     ocfs2_clusters_to_bytes(osb->sb, 1));
1885                 if (status)
1886                         goto bail;
1887                 did_quota = 1;
1888                 inode->i_mapping->a_ops = &ocfs2_aops;
1889                 status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1890                                               new_fe_bh,
1891                                               handle, data_ac, NULL,
1892                                               NULL);
1893                 if (status < 0) {
1894                         if (status != -ENOSPC && status != -EINTR) {
1895                                 mlog(ML_ERROR,
1896                                      "Failed to extend file to %llu\n",
1897                                      (unsigned long long)newsize);
1898                                 mlog_errno(status);
1899                                 status = -ENOSPC;
1900                         }
1901                         goto bail;
1902                 }
1903                 i_size_write(inode, newsize);
1904                 inode->i_blocks = ocfs2_inode_sector_count(inode);
1905         } else {
1906                 inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1907                 memcpy((char *) fe->id2.i_symlink, symname, l);
1908                 i_size_write(inode, newsize);
1909                 inode->i_blocks = 0;
1910         }
1911
1912         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1913         if (status < 0) {
1914                 mlog_errno(status);
1915                 goto bail;
1916         }
1917
1918         if (!ocfs2_inode_is_fast_symlink(inode)) {
1919                 status = ocfs2_create_symlink_data(osb, handle, inode,
1920                                                    symname);
1921                 if (status < 0) {
1922                         mlog_errno(status);
1923                         goto bail;
1924                 }
1925         }
1926
1927         if (si.enable) {
1928                 status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1929                                                  xattr_ac, data_ac);
1930                 if (status < 0) {
1931                         mlog_errno(status);
1932                         goto bail;
1933                 }
1934         }
1935
1936         /*
1937          * Do this before adding the entry to the directory. We add
1938          * also set d_op after success so that ->d_iput() will cleanup
1939          * the dentry lock even if ocfs2_add_entry() fails below.
1940          */
1941         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1942         if (status) {
1943                 mlog_errno(status);
1944                 goto bail;
1945         }
1946
1947         dl = dentry->d_fsdata;
1948
1949         status = ocfs2_add_entry(handle, dentry, inode,
1950                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1951                                  &lookup);
1952         if (status < 0) {
1953                 mlog_errno(status);
1954                 goto bail;
1955         }
1956
1957         insert_inode_hash(inode);
1958         d_instantiate(dentry, inode);
1959 bail:
1960         if (status < 0 && did_quota)
1961                 dquot_free_space_nodirty(inode,
1962                                         ocfs2_clusters_to_bytes(osb->sb, 1));
1963         if (status < 0 && did_quota_inode)
1964                 dquot_free_inode(inode);
1965         if (handle)
1966                 ocfs2_commit_trans(osb, handle);
1967
1968         ocfs2_inode_unlock(dir, 1);
1969         if (did_block_signals)
1970                 ocfs2_unblock_signals(&oldset);
1971
1972         brelse(new_fe_bh);
1973         brelse(parent_fe_bh);
1974         kfree(si.value);
1975         ocfs2_free_dir_lookup_result(&lookup);
1976         if (inode_ac)
1977                 ocfs2_free_alloc_context(inode_ac);
1978         if (data_ac)
1979                 ocfs2_free_alloc_context(data_ac);
1980         if (xattr_ac)
1981                 ocfs2_free_alloc_context(xattr_ac);
1982         if ((status < 0) && inode) {
1983                 if (dl)
1984                         ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
1985
1986                 OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1987                 clear_nlink(inode);
1988                 iput(inode);
1989         }
1990
1991         if (status)
1992                 mlog_errno(status);
1993
1994         return status;
1995 }
1996
1997 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1998 {
1999         int status, namelen;
2000
2001         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2002                            (long long)blkno);
2003         if (namelen <= 0) {
2004                 if (namelen)
2005                         status = namelen;
2006                 else
2007                         status = -EINVAL;
2008                 mlog_errno(status);
2009                 goto bail;
2010         }
2011         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2012                 status = -EINVAL;
2013                 mlog_errno(status);
2014                 goto bail;
2015         }
2016
2017         trace_ocfs2_blkno_stringify(blkno, name, namelen);
2018
2019         status = 0;
2020 bail:
2021         if (status < 0)
2022                 mlog_errno(status);
2023         return status;
2024 }
2025
2026 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2027                                         struct inode **ret_orphan_dir,
2028                                         struct buffer_head **ret_orphan_dir_bh)
2029 {
2030         struct inode *orphan_dir_inode;
2031         struct buffer_head *orphan_dir_bh = NULL;
2032         int ret = 0;
2033
2034         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2035                                                        ORPHAN_DIR_SYSTEM_INODE,
2036                                                        osb->slot_num);
2037         if (!orphan_dir_inode) {
2038                 ret = -ENOENT;
2039                 mlog_errno(ret);
2040                 return ret;
2041         }
2042
2043         mutex_lock(&orphan_dir_inode->i_mutex);
2044
2045         ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2046         if (ret < 0) {
2047                 mutex_unlock(&orphan_dir_inode->i_mutex);
2048                 iput(orphan_dir_inode);
2049
2050                 mlog_errno(ret);
2051                 return ret;
2052         }
2053
2054         *ret_orphan_dir = orphan_dir_inode;
2055         *ret_orphan_dir_bh = orphan_dir_bh;
2056
2057         return 0;
2058 }
2059
2060 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2061                                       struct buffer_head *orphan_dir_bh,
2062                                       u64 blkno,
2063                                       char *name,
2064                                       struct ocfs2_dir_lookup_result *lookup)
2065 {
2066         int ret;
2067         struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2068
2069         ret = ocfs2_blkno_stringify(blkno, name);
2070         if (ret < 0) {
2071                 mlog_errno(ret);
2072                 return ret;
2073         }
2074
2075         ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2076                                            orphan_dir_bh, name,
2077                                            OCFS2_ORPHAN_NAMELEN, lookup);
2078         if (ret < 0) {
2079                 mlog_errno(ret);
2080                 return ret;
2081         }
2082
2083         return 0;
2084 }
2085
2086 /**
2087  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2088  * insertion of an orphan.
2089  * @osb: ocfs2 file system
2090  * @ret_orphan_dir: Orphan dir inode - returned locked!
2091  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2092  * @lookup: dir lookup result, to be passed back into functions like
2093  *          ocfs2_orphan_add
2094  *
2095  * Returns zero on success and the ret_orphan_dir, name and lookup
2096  * fields will be populated.
2097  *
2098  * Returns non-zero on failure. 
2099  */
2100 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2101                                     struct inode **ret_orphan_dir,
2102                                     u64 blkno,
2103                                     char *name,
2104                                     struct ocfs2_dir_lookup_result *lookup)
2105 {
2106         struct inode *orphan_dir_inode = NULL;
2107         struct buffer_head *orphan_dir_bh = NULL;
2108         int ret = 0;
2109
2110         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2111                                            &orphan_dir_bh);
2112         if (ret < 0) {
2113                 mlog_errno(ret);
2114                 return ret;
2115         }
2116
2117         ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2118                                          blkno, name, lookup);
2119         if (ret < 0) {
2120                 mlog_errno(ret);
2121                 goto out;
2122         }
2123
2124         *ret_orphan_dir = orphan_dir_inode;
2125
2126 out:
2127         brelse(orphan_dir_bh);
2128
2129         if (ret) {
2130                 ocfs2_inode_unlock(orphan_dir_inode, 1);
2131                 mutex_unlock(&orphan_dir_inode->i_mutex);
2132                 iput(orphan_dir_inode);
2133         }
2134
2135         if (ret)
2136                 mlog_errno(ret);
2137         return ret;
2138 }
2139
2140 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2141                             handle_t *handle,
2142                             struct inode *inode,
2143                             struct buffer_head *fe_bh,
2144                             char *name,
2145                             struct ocfs2_dir_lookup_result *lookup,
2146                             struct inode *orphan_dir_inode)
2147 {
2148         struct buffer_head *orphan_dir_bh = NULL;
2149         int status = 0;
2150         struct ocfs2_dinode *orphan_fe;
2151         struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2152
2153         trace_ocfs2_orphan_add_begin(
2154                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2155
2156         status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2157         if (status < 0) {
2158                 mlog_errno(status);
2159                 goto leave;
2160         }
2161
2162         status = ocfs2_journal_access_di(handle,
2163                                          INODE_CACHE(orphan_dir_inode),
2164                                          orphan_dir_bh,
2165                                          OCFS2_JOURNAL_ACCESS_WRITE);
2166         if (status < 0) {
2167                 mlog_errno(status);
2168                 goto leave;
2169         }
2170
2171         /*
2172          * We're going to journal the change of i_flags and i_orphaned_slot.
2173          * It's safe anyway, though some callers may duplicate the journaling.
2174          * Journaling within the func just make the logic look more
2175          * straightforward.
2176          */
2177         status = ocfs2_journal_access_di(handle,
2178                                          INODE_CACHE(inode),
2179                                          fe_bh,
2180                                          OCFS2_JOURNAL_ACCESS_WRITE);
2181         if (status < 0) {
2182                 mlog_errno(status);
2183                 goto leave;
2184         }
2185
2186         /* we're a cluster, and nlink can change on disk from
2187          * underneath us... */
2188         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2189         if (S_ISDIR(inode->i_mode))
2190                 ocfs2_add_links_count(orphan_fe, 1);
2191         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2192         ocfs2_journal_dirty(handle, orphan_dir_bh);
2193
2194         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2195                                    OCFS2_ORPHAN_NAMELEN, inode,
2196                                    OCFS2_I(inode)->ip_blkno,
2197                                    orphan_dir_bh, lookup);
2198         if (status < 0) {
2199                 mlog_errno(status);
2200                 goto rollback;
2201         }
2202
2203         fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2204         OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2205
2206         /* Record which orphan dir our inode now resides
2207          * in. delete_inode will use this to determine which orphan
2208          * dir to lock. */
2209         fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2210
2211         ocfs2_journal_dirty(handle, fe_bh);
2212
2213         trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2214                                    osb->slot_num);
2215
2216 rollback:
2217         if (status < 0) {
2218                 if (S_ISDIR(inode->i_mode))
2219                         ocfs2_add_links_count(orphan_fe, -1);
2220                 set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2221         }
2222
2223 leave:
2224         brelse(orphan_dir_bh);
2225
2226         return status;
2227 }
2228
2229 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2230 int ocfs2_orphan_del(struct ocfs2_super *osb,
2231                      handle_t *handle,
2232                      struct inode *orphan_dir_inode,
2233                      struct inode *inode,
2234                      struct buffer_head *orphan_dir_bh)
2235 {
2236         char name[OCFS2_ORPHAN_NAMELEN + 1];
2237         struct ocfs2_dinode *orphan_fe;
2238         int status = 0;
2239         struct ocfs2_dir_lookup_result lookup = { NULL, };
2240
2241         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2242         if (status < 0) {
2243                 mlog_errno(status);
2244                 goto leave;
2245         }
2246
2247         trace_ocfs2_orphan_del(
2248              (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2249              name, OCFS2_ORPHAN_NAMELEN);
2250
2251         /* find it's spot in the orphan directory */
2252         status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2253                                   &lookup);
2254         if (status) {
2255                 mlog_errno(status);
2256                 goto leave;
2257         }
2258
2259         /* remove it from the orphan directory */
2260         status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2261         if (status < 0) {
2262                 mlog_errno(status);
2263                 goto leave;
2264         }
2265
2266         status = ocfs2_journal_access_di(handle,
2267                                          INODE_CACHE(orphan_dir_inode),
2268                                          orphan_dir_bh,
2269                                          OCFS2_JOURNAL_ACCESS_WRITE);
2270         if (status < 0) {
2271                 mlog_errno(status);
2272                 goto leave;
2273         }
2274
2275         /* do the i_nlink dance! :) */
2276         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2277         if (S_ISDIR(inode->i_mode))
2278                 ocfs2_add_links_count(orphan_fe, -1);
2279         set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2280         ocfs2_journal_dirty(handle, orphan_dir_bh);
2281
2282 leave:
2283         ocfs2_free_dir_lookup_result(&lookup);
2284
2285         if (status)
2286                 mlog_errno(status);
2287         return status;
2288 }
2289
2290 /**
2291  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2292  * allocated file. This is different from the typical 'add to orphan dir'
2293  * operation in that the inode does not yet exist. This is a problem because
2294  * the orphan dir stringifies the inode block number to come up with it's
2295  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2296  * problem. This function works around it by calling deeper into the orphan
2297  * and suballoc code than other callers. Use this only by necessity.
2298  * @dir: The directory which this inode will ultimately wind up under - not the
2299  * orphan dir!
2300  * @dir_bh: buffer_head the @dir inode block
2301  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2302  * with the string to be used for orphan dirent. Pass back to the orphan dir
2303  * code.
2304  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2305  * dir code.
2306  * @ret_di_blkno: block number where the new inode will be allocated.
2307  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2308  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2309  *
2310  * Returns zero on success and the ret_orphan_dir, name and lookup
2311  * fields will be populated.
2312  *
2313  * Returns non-zero on failure. 
2314  */
2315 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2316                                         struct buffer_head *dir_bh,
2317                                         char *orphan_name,
2318                                         struct inode **ret_orphan_dir,
2319                                         u64 *ret_di_blkno,
2320                                         struct ocfs2_dir_lookup_result *orphan_insert,
2321                                         struct ocfs2_alloc_context **ret_inode_ac)
2322 {
2323         int ret;
2324         u64 di_blkno;
2325         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2326         struct inode *orphan_dir = NULL;
2327         struct buffer_head *orphan_dir_bh = NULL;
2328         struct ocfs2_alloc_context *inode_ac = NULL;
2329
2330         ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2331         if (ret < 0) {
2332                 mlog_errno(ret);
2333                 return ret;
2334         }
2335
2336         /* reserve an inode spot */
2337         ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2338         if (ret < 0) {
2339                 if (ret != -ENOSPC)
2340                         mlog_errno(ret);
2341                 goto out;
2342         }
2343
2344         ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2345                                        &di_blkno);
2346         if (ret) {
2347                 mlog_errno(ret);
2348                 goto out;
2349         }
2350
2351         ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2352                                          di_blkno, orphan_name, orphan_insert);
2353         if (ret < 0) {
2354                 mlog_errno(ret);
2355                 goto out;
2356         }
2357
2358 out:
2359         if (ret == 0) {
2360                 *ret_orphan_dir = orphan_dir;
2361                 *ret_di_blkno = di_blkno;
2362                 *ret_inode_ac = inode_ac;
2363                 /*
2364                  * orphan_name and orphan_insert are already up to
2365                  * date via prepare_orphan_dir
2366                  */
2367         } else {
2368                 /* Unroll reserve_new_inode* */
2369                 if (inode_ac)
2370                         ocfs2_free_alloc_context(inode_ac);
2371
2372                 /* Unroll orphan dir locking */
2373                 mutex_unlock(&orphan_dir->i_mutex);
2374                 ocfs2_inode_unlock(orphan_dir, 1);
2375                 iput(orphan_dir);
2376         }
2377
2378         brelse(orphan_dir_bh);
2379
2380         return ret;
2381 }
2382
2383 int ocfs2_create_inode_in_orphan(struct inode *dir,
2384                                  int mode,
2385                                  struct inode **new_inode)
2386 {
2387         int status, did_quota_inode = 0;
2388         struct inode *inode = NULL;
2389         struct inode *orphan_dir = NULL;
2390         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2391         struct ocfs2_dinode *di = NULL;
2392         handle_t *handle = NULL;
2393         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2394         struct buffer_head *parent_di_bh = NULL;
2395         struct buffer_head *new_di_bh = NULL;
2396         struct ocfs2_alloc_context *inode_ac = NULL;
2397         struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2398         u64 uninitialized_var(di_blkno), suballoc_loc;
2399         u16 suballoc_bit;
2400
2401         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2402         if (status < 0) {
2403                 if (status != -ENOENT)
2404                         mlog_errno(status);
2405                 return status;
2406         }
2407
2408         status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2409                                               orphan_name, &orphan_dir,
2410                                               &di_blkno, &orphan_insert, &inode_ac);
2411         if (status < 0) {
2412                 if (status != -ENOSPC)
2413                         mlog_errno(status);
2414                 goto leave;
2415         }
2416
2417         inode = ocfs2_get_init_inode(dir, mode);
2418         if (!inode) {
2419                 status = -ENOMEM;
2420                 mlog_errno(status);
2421                 goto leave;
2422         }
2423
2424         handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2425         if (IS_ERR(handle)) {
2426                 status = PTR_ERR(handle);
2427                 handle = NULL;
2428                 mlog_errno(status);
2429                 goto leave;
2430         }
2431
2432         status = dquot_alloc_inode(inode);
2433         if (status)
2434                 goto leave;
2435         did_quota_inode = 1;
2436
2437         status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2438                                               &suballoc_loc,
2439                                               &suballoc_bit, di_blkno);
2440         if (status < 0) {
2441                 mlog_errno(status);
2442                 goto leave;
2443         }
2444
2445         clear_nlink(inode);
2446         /* do the real work now. */
2447         status = __ocfs2_mknod_locked(dir, inode,
2448                                       0, &new_di_bh, parent_di_bh, handle,
2449                                       inode_ac, di_blkno, suballoc_loc,
2450                                       suballoc_bit);
2451         if (status < 0) {
2452                 mlog_errno(status);
2453                 goto leave;
2454         }
2455
2456         di = (struct ocfs2_dinode *)new_di_bh->b_data;
2457         status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2458                                   &orphan_insert, orphan_dir);
2459         if (status < 0) {
2460                 mlog_errno(status);
2461                 goto leave;
2462         }
2463
2464         /* get open lock so that only nodes can't remove it from orphan dir. */
2465         status = ocfs2_open_lock(inode);
2466         if (status < 0)
2467                 mlog_errno(status);
2468
2469         insert_inode_hash(inode);
2470 leave:
2471         if (status < 0 && did_quota_inode)
2472                 dquot_free_inode(inode);
2473         if (handle)
2474                 ocfs2_commit_trans(osb, handle);
2475
2476         if (orphan_dir) {
2477                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
2478                 ocfs2_inode_unlock(orphan_dir, 1);
2479                 mutex_unlock(&orphan_dir->i_mutex);
2480                 iput(orphan_dir);
2481         }
2482
2483         if ((status < 0) && inode) {
2484                 clear_nlink(inode);
2485                 iput(inode);
2486         }
2487
2488         if (inode_ac)
2489                 ocfs2_free_alloc_context(inode_ac);
2490
2491         brelse(new_di_bh);
2492
2493         if (!status)
2494                 *new_inode = inode;
2495
2496         ocfs2_free_dir_lookup_result(&orphan_insert);
2497
2498         ocfs2_inode_unlock(dir, 1);
2499         brelse(parent_di_bh);
2500         return status;
2501 }
2502
2503 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2504                                    struct inode *inode,
2505                                    struct dentry *dentry)
2506 {
2507         int status = 0;
2508         struct buffer_head *parent_di_bh = NULL;
2509         handle_t *handle = NULL;
2510         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2511         struct ocfs2_dinode *dir_di, *di;
2512         struct inode *orphan_dir_inode = NULL;
2513         struct buffer_head *orphan_dir_bh = NULL;
2514         struct buffer_head *di_bh = NULL;
2515         struct ocfs2_dir_lookup_result lookup = { NULL, };
2516
2517         trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2518                                 dentry->d_name.len, dentry->d_name.name,
2519                                 (unsigned long long)OCFS2_I(dir)->ip_blkno,
2520                                 (unsigned long long)OCFS2_I(inode)->ip_blkno);
2521
2522         status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2523         if (status < 0) {
2524                 if (status != -ENOENT)
2525                         mlog_errno(status);
2526                 return status;
2527         }
2528
2529         dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2530         if (!dir_di->i_links_count) {
2531                 /* can't make a file in a deleted directory. */
2532                 status = -ENOENT;
2533                 goto leave;
2534         }
2535
2536         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2537                                            dentry->d_name.len);
2538         if (status)
2539                 goto leave;
2540
2541         /* get a spot inside the dir. */
2542         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2543                                               dentry->d_name.name,
2544                                               dentry->d_name.len, &lookup);
2545         if (status < 0) {
2546                 mlog_errno(status);
2547                 goto leave;
2548         }
2549
2550         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2551                                                        ORPHAN_DIR_SYSTEM_INODE,
2552                                                        osb->slot_num);
2553         if (!orphan_dir_inode) {
2554                 status = -EEXIST;
2555                 mlog_errno(status);
2556                 goto leave;
2557         }
2558
2559         mutex_lock(&orphan_dir_inode->i_mutex);
2560
2561         status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2562         if (status < 0) {
2563                 mlog_errno(status);
2564                 mutex_unlock(&orphan_dir_inode->i_mutex);
2565                 iput(orphan_dir_inode);
2566                 goto leave;
2567         }
2568
2569         status = ocfs2_read_inode_block(inode, &di_bh);
2570         if (status < 0) {
2571                 mlog_errno(status);
2572                 goto orphan_unlock;
2573         }
2574
2575         handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2576         if (IS_ERR(handle)) {
2577                 status = PTR_ERR(handle);
2578                 handle = NULL;
2579                 mlog_errno(status);
2580                 goto orphan_unlock;
2581         }
2582
2583         status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2584                                          di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2585         if (status < 0) {
2586                 mlog_errno(status);
2587                 goto out_commit;
2588         }
2589
2590         status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2591                                   orphan_dir_bh);
2592         if (status < 0) {
2593                 mlog_errno(status);
2594                 goto out_commit;
2595         }
2596
2597         di = (struct ocfs2_dinode *)di_bh->b_data;
2598         di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2599         di->i_orphaned_slot = 0;
2600         set_nlink(inode, 1);
2601         ocfs2_set_links_count(di, inode->i_nlink);
2602         ocfs2_update_inode_fsync_trans(handle, inode, 1);
2603         ocfs2_journal_dirty(handle, di_bh);
2604
2605         status = ocfs2_add_entry(handle, dentry, inode,
2606                                  OCFS2_I(inode)->ip_blkno, parent_di_bh,
2607                                  &lookup);
2608         if (status < 0) {
2609                 mlog_errno(status);
2610                 goto out_commit;
2611         }
2612
2613         status = ocfs2_dentry_attach_lock(dentry, inode,
2614                                           OCFS2_I(dir)->ip_blkno);
2615         if (status) {
2616                 mlog_errno(status);
2617                 goto out_commit;
2618         }
2619
2620         d_instantiate(dentry, inode);
2621         status = 0;
2622 out_commit:
2623         ocfs2_commit_trans(osb, handle);
2624 orphan_unlock:
2625         ocfs2_inode_unlock(orphan_dir_inode, 1);
2626         mutex_unlock(&orphan_dir_inode->i_mutex);
2627         iput(orphan_dir_inode);
2628 leave:
2629
2630         ocfs2_inode_unlock(dir, 1);
2631
2632         brelse(di_bh);
2633         brelse(parent_di_bh);
2634         brelse(orphan_dir_bh);
2635
2636         ocfs2_free_dir_lookup_result(&lookup);
2637
2638         if (status)
2639                 mlog_errno(status);
2640
2641         return status;
2642 }
2643
2644 const struct inode_operations ocfs2_dir_iops = {
2645         .create         = ocfs2_create,
2646         .lookup         = ocfs2_lookup,
2647         .link           = ocfs2_link,
2648         .unlink         = ocfs2_unlink,
2649         .rmdir          = ocfs2_unlink,
2650         .symlink        = ocfs2_symlink,
2651         .mkdir          = ocfs2_mkdir,
2652         .mknod          = ocfs2_mknod,
2653         .rename         = ocfs2_rename,
2654         .setattr        = ocfs2_setattr,
2655         .getattr        = ocfs2_getattr,
2656         .permission     = ocfs2_permission,
2657         .setxattr       = generic_setxattr,
2658         .getxattr       = generic_getxattr,
2659         .listxattr      = ocfs2_listxattr,
2660         .removexattr    = generic_removexattr,
2661         .fiemap         = ocfs2_fiemap,
2662         .get_acl        = ocfs2_iop_get_acl,
2663         .set_acl        = ocfs2_iop_set_acl,
2664 };