40f83f53053a8eac58a027c68892aa1ed42d8392
[safe/jmp/linux-2.6] / fs / ocfs2 / namei.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * namei.c
5  *
6  * Create and rename file, directory, symlinks
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  *  Portions of this code from linux/fs/ext3/dir.c
11  *
12  *  Copyright (C) 1992, 1993, 1994, 1995
13  *  Remy Card (card@masi.ibp.fr)
14  *  Laboratoire MASI - Institut Blaise pascal
15  *  Universite Pierre et Marie Curie (Paris VI)
16  *
17  *   from
18  *
19  *   linux/fs/minix/dir.c
20  *
21  *   Copyright (C) 1991, 1992 Linux Torvalds
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public
25  * License as published by the Free Software Foundation; either
26  * version 2 of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
31  * General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public
34  * License along with this program; if not, write to the
35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36  * Boston, MA 021110-1307, USA.
37  */
38
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43
44 #define MLOG_MASK_PREFIX ML_NAMEI
45 #include <cluster/masklog.h>
46
47 #include "ocfs2.h"
48
49 #include "alloc.h"
50 #include "dcache.h"
51 #include "dir.h"
52 #include "dlmglue.h"
53 #include "extent_map.h"
54 #include "file.h"
55 #include "inode.h"
56 #include "journal.h"
57 #include "namei.h"
58 #include "suballoc.h"
59 #include "super.h"
60 #include "symlink.h"
61 #include "sysfile.h"
62 #include "uptodate.h"
63 #include "vote.h"
64
65 #include "buffer_head_io.h"
66
67 #define NAMEI_RA_CHUNKS  2
68 #define NAMEI_RA_BLOCKS  4
69 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
70 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
71
72 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
73                                         struct inode *dir,
74                                         const char *name, int namelen,
75                                         unsigned long offset,
76                                         struct ocfs2_dir_entry **res_dir);
77
78 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
79                               struct inode *dir,
80                               struct ocfs2_dir_entry *de_del,
81                               struct buffer_head *bh);
82
83 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
84                              struct inode *dir,
85                              const char *name, int namelen,
86                              struct inode *inode, u64 blkno,
87                              struct buffer_head *parent_fe_bh,
88                              struct buffer_head *insert_bh);
89
90 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
91                               struct inode *dir,
92                               struct dentry *dentry, int mode,
93                               dev_t dev,
94                               struct buffer_head **new_fe_bh,
95                               struct buffer_head *parent_fe_bh,
96                               struct ocfs2_journal_handle *handle,
97                               struct inode **ret_inode,
98                               struct ocfs2_alloc_context *inode_ac);
99
100 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
101                               struct ocfs2_journal_handle *handle,
102                               struct inode *parent,
103                               struct inode *inode,
104                               struct buffer_head *fe_bh,
105                               struct ocfs2_alloc_context *data_ac);
106
107 static int ocfs2_double_lock(struct ocfs2_super *osb,
108                              struct ocfs2_journal_handle *handle,
109                              struct buffer_head **bh1,
110                              struct inode *inode1,
111                              struct buffer_head **bh2,
112                              struct inode *inode2);
113
114 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
115                                     struct ocfs2_journal_handle *handle,
116                                     struct inode *inode,
117                                     char *name,
118                                     struct buffer_head **de_bh);
119
120 static int ocfs2_orphan_add(struct ocfs2_super *osb,
121                             struct ocfs2_journal_handle *handle,
122                             struct inode *inode,
123                             struct ocfs2_dinode *fe,
124                             char *name,
125                             struct buffer_head *de_bh);
126
127 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
128                                      struct ocfs2_journal_handle *handle,
129                                      struct inode *inode,
130                                      const char *symname);
131
132 static inline int ocfs2_add_entry(struct ocfs2_journal_handle *handle,
133                                   struct dentry *dentry,
134                                   struct inode *inode, u64 blkno,
135                                   struct buffer_head *parent_fe_bh,
136                                   struct buffer_head *insert_bh)
137 {
138         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
139                                  dentry->d_name.name, dentry->d_name.len,
140                                  inode, blkno, parent_fe_bh, insert_bh);
141 }
142
143 /* An orphan dir name is an 8 byte value, printed as a hex string */
144 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
145
146 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
147                                    struct nameidata *nd)
148 {
149         int status;
150         u64 blkno;
151         struct buffer_head *dirent_bh = NULL;
152         struct inode *inode = NULL;
153         struct dentry *ret;
154         struct ocfs2_dir_entry *dirent;
155         struct ocfs2_inode_info *oi;
156
157         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
158                    dentry->d_name.len, dentry->d_name.name);
159
160         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
161                 ret = ERR_PTR(-ENAMETOOLONG);
162                 goto bail;
163         }
164
165         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
166              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
167
168         status = ocfs2_meta_lock(dir, NULL, NULL, 0);
169         if (status < 0) {
170                 if (status != -ENOENT)
171                         mlog_errno(status);
172                 ret = ERR_PTR(status);
173                 goto bail;
174         }
175
176         status = ocfs2_find_files_on_disk(dentry->d_name.name,
177                                           dentry->d_name.len, &blkno,
178                                           dir, &dirent_bh, &dirent);
179         if (status < 0)
180                 goto bail_add;
181
182         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
183         if (IS_ERR(inode)) {
184                 mlog(ML_ERROR, "Unable to create inode %llu\n",
185                      (unsigned long long)blkno);
186                 ret = ERR_PTR(-EACCES);
187                 goto bail_unlock;
188         }
189
190         oi = OCFS2_I(inode);
191         /* Clear any orphaned state... If we were able to look up the
192          * inode from a directory, it certainly can't be orphaned. We
193          * might have the bad state from a node which intended to
194          * orphan this inode but crashed before it could commit the
195          * unlink. */
196         spin_lock(&oi->ip_lock);
197         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
198         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
199         spin_unlock(&oi->ip_lock);
200
201 bail_add:
202         dentry->d_op = &ocfs2_dentry_ops;
203         ret = d_splice_alias(inode, dentry);
204
205         if (inode) {
206                 /*
207                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
208                  * dentry, it will d_move() it on top of ourse. The
209                  * return value will indicate this however, so in
210                  * those cases, we switch them around for the locking
211                  * code.
212                  *
213                  * NOTE: This dentry already has ->d_op set from
214                  * ocfs2_get_parent() and ocfs2_get_dentry()
215                  */
216                 if (ret)
217                         dentry = ret;
218
219                 status = ocfs2_dentry_attach_lock(dentry, inode,
220                                                   OCFS2_I(dir)->ip_blkno);
221                 if (status) {
222                         mlog_errno(status);
223                         ret = ERR_PTR(status);
224                         goto bail_unlock;
225                 }
226         }
227
228 bail_unlock:
229         /* Don't drop the cluster lock until *after* the d_add --
230          * unlink on another node will message us to remove that
231          * dentry under this lock so otherwise we can race this with
232          * the vote thread and have a stale dentry. */
233         ocfs2_meta_unlock(dir, 0);
234
235 bail:
236         if (dirent_bh)
237                 brelse(dirent_bh);
238
239         mlog_exit_ptr(ret);
240
241         return ret;
242 }
243
244 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
245                               struct ocfs2_journal_handle *handle,
246                               struct inode *parent,
247                               struct inode *inode,
248                               struct buffer_head *fe_bh,
249                               struct ocfs2_alloc_context *data_ac)
250 {
251         int status;
252         struct buffer_head *new_bh = NULL;
253         struct ocfs2_dir_entry *de = NULL;
254
255         mlog_entry_void();
256
257         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
258                                      data_ac, NULL, &new_bh);
259         if (status < 0) {
260                 mlog_errno(status);
261                 goto bail;
262         }
263
264         ocfs2_set_new_buffer_uptodate(inode, new_bh);
265
266         status = ocfs2_journal_access(handle, inode, new_bh,
267                                       OCFS2_JOURNAL_ACCESS_CREATE);
268         if (status < 0) {
269                 mlog_errno(status);
270                 goto bail;
271         }
272         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
273
274         de = (struct ocfs2_dir_entry *) new_bh->b_data;
275         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
276         de->name_len = 1;
277         de->rec_len =
278                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
279         strcpy(de->name, ".");
280         ocfs2_set_de_type(de, S_IFDIR);
281         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
282         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
283         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
284                                   OCFS2_DIR_REC_LEN(1));
285         de->name_len = 2;
286         strcpy(de->name, "..");
287         ocfs2_set_de_type(de, S_IFDIR);
288
289         status = ocfs2_journal_dirty(handle, new_bh);
290         if (status < 0) {
291                 mlog_errno(status);
292                 goto bail;
293         }
294
295         i_size_write(inode, inode->i_sb->s_blocksize);
296         inode->i_nlink = 2;
297         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
298         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
299         if (status < 0) {
300                 mlog_errno(status);
301                 goto bail;
302         }
303
304         status = 0;
305 bail:
306         if (new_bh)
307                 brelse(new_bh);
308
309         mlog_exit(status);
310         return status;
311 }
312
313 static int ocfs2_mknod(struct inode *dir,
314                        struct dentry *dentry,
315                        int mode,
316                        dev_t dev)
317 {
318         int status = 0;
319         struct buffer_head *parent_fe_bh = NULL;
320         struct ocfs2_journal_handle *handle = NULL;
321         struct ocfs2_super *osb;
322         struct ocfs2_dinode *dirfe;
323         struct buffer_head *new_fe_bh = NULL;
324         struct buffer_head *de_bh = NULL;
325         struct inode *inode = NULL;
326         struct ocfs2_alloc_context *inode_ac = NULL;
327         struct ocfs2_alloc_context *data_ac = NULL;
328
329         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
330                    (unsigned long)dev, dentry->d_name.len,
331                    dentry->d_name.name);
332
333         /* get our super block */
334         osb = OCFS2_SB(dir->i_sb);
335
336         handle = ocfs2_alloc_handle(osb);
337         if (handle == NULL) {
338                 status = -ENOMEM;
339                 mlog_errno(status);
340                 goto leave;
341         }
342
343         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
344         if (status < 0) {
345                 if (status != -ENOENT)
346                         mlog_errno(status);
347                 goto leave;
348         }
349
350         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
351                 status = -EMLINK;
352                 goto leave;
353         }
354
355         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
356         if (!dirfe->i_links_count) {
357                 /* can't make a file in a deleted directory. */
358                 status = -ENOENT;
359                 goto leave;
360         }
361
362         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
363                                            dentry->d_name.len);
364         if (status)
365                 goto leave;
366
367         /* get a spot inside the dir. */
368         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
369                                               dentry->d_name.name,
370                                               dentry->d_name.len, &de_bh);
371         if (status < 0) {
372                 mlog_errno(status);
373                 goto leave;
374         }
375
376         /* reserve an inode spot */
377         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
378         if (status < 0) {
379                 if (status != -ENOSPC)
380                         mlog_errno(status);
381                 goto leave;
382         }
383
384         /* are we making a directory? If so, reserve a cluster for his
385          * 1st extent. */
386         if (S_ISDIR(mode)) {
387                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
388                 if (status < 0) {
389                         if (status != -ENOSPC)
390                                 mlog_errno(status);
391                         goto leave;
392                 }
393         }
394
395         handle = ocfs2_start_trans(osb, handle, OCFS2_MKNOD_CREDITS);
396         if (IS_ERR(handle)) {
397                 status = PTR_ERR(handle);
398                 handle = NULL;
399                 mlog_errno(status);
400                 goto leave;
401         }
402
403         /* do the real work now. */
404         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
405                                     &new_fe_bh, parent_fe_bh, handle,
406                                     &inode, inode_ac);
407         if (status < 0) {
408                 mlog_errno(status);
409                 goto leave;
410         }
411
412         if (S_ISDIR(mode)) {
413                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
414                                             new_fe_bh, data_ac);
415                 if (status < 0) {
416                         mlog_errno(status);
417                         goto leave;
418                 }
419
420                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
421                                               OCFS2_JOURNAL_ACCESS_WRITE);
422                 if (status < 0) {
423                         mlog_errno(status);
424                         goto leave;
425                 }
426                 le16_add_cpu(&dirfe->i_links_count, 1);
427                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
428                 if (status < 0) {
429                         mlog_errno(status);
430                         goto leave;
431                 }
432                 dir->i_nlink++;
433         }
434
435         status = ocfs2_add_entry(handle, dentry, inode,
436                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
437                                  de_bh);
438         if (status < 0) {
439                 mlog_errno(status);
440                 goto leave;
441         }
442
443         status = ocfs2_dentry_attach_lock(dentry, inode,
444                                           OCFS2_I(dir)->ip_blkno);
445         if (status) {
446                 mlog_errno(status);
447                 goto leave;
448         }
449
450         insert_inode_hash(inode);
451         dentry->d_op = &ocfs2_dentry_ops;
452         d_instantiate(dentry, inode);
453         status = 0;
454 leave:
455         if (handle)
456                 ocfs2_commit_trans(handle);
457
458         if (status == -ENOSPC)
459                 mlog(0, "Disk is full\n");
460
461         if (new_fe_bh)
462                 brelse(new_fe_bh);
463
464         if (de_bh)
465                 brelse(de_bh);
466
467         if (parent_fe_bh)
468                 brelse(parent_fe_bh);
469
470         if ((status < 0) && inode)
471                 iput(inode);
472
473         if (inode_ac)
474                 ocfs2_free_alloc_context(inode_ac);
475
476         if (data_ac)
477                 ocfs2_free_alloc_context(data_ac);
478
479         mlog_exit(status);
480
481         return status;
482 }
483
484 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
485                               struct inode *dir,
486                               struct dentry *dentry, int mode,
487                               dev_t dev,
488                               struct buffer_head **new_fe_bh,
489                               struct buffer_head *parent_fe_bh,
490                               struct ocfs2_journal_handle *handle,
491                               struct inode **ret_inode,
492                               struct ocfs2_alloc_context *inode_ac)
493 {
494         int status = 0;
495         struct ocfs2_dinode *fe = NULL;
496         struct ocfs2_extent_list *fel;
497         u64 fe_blkno = 0;
498         u16 suballoc_bit;
499         struct inode *inode = NULL;
500
501         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
502                    (unsigned long)dev, dentry->d_name.len,
503                    dentry->d_name.name);
504
505         *new_fe_bh = NULL;
506         *ret_inode = NULL;
507
508         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
509                                        &fe_blkno);
510         if (status < 0) {
511                 mlog_errno(status);
512                 goto leave;
513         }
514
515         inode = new_inode(dir->i_sb);
516         if (IS_ERR(inode)) {
517                 status = PTR_ERR(inode);
518                 mlog(ML_ERROR, "new_inode failed!\n");
519                 goto leave;
520         }
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         if (S_ISDIR(mode))
528                 inode->i_nlink = 2;
529         else
530                 inode->i_nlink = 1;
531         inode->i_mode = mode;
532         spin_lock(&osb->osb_lock);
533         inode->i_generation = osb->s_next_generation++;
534         spin_unlock(&osb->osb_lock);
535
536         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
537         if (!*new_fe_bh) {
538                 status = -EIO;
539                 mlog_errno(status);
540                 goto leave;
541         }
542         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
543
544         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
545                                       OCFS2_JOURNAL_ACCESS_CREATE);
546         if (status < 0) {
547                 mlog_errno(status);
548                 goto leave;
549         }
550
551         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
552         memset(fe, 0, osb->sb->s_blocksize);
553
554         fe->i_generation = cpu_to_le32(inode->i_generation);
555         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
556         fe->i_blkno = cpu_to_le64(fe_blkno);
557         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
558         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
559         fe->i_uid = cpu_to_le32(current->fsuid);
560         if (dir->i_mode & S_ISGID) {
561                 fe->i_gid = cpu_to_le32(dir->i_gid);
562                 if (S_ISDIR(mode))
563                         mode |= S_ISGID;
564         } else
565                 fe->i_gid = cpu_to_le32(current->fsgid);
566         fe->i_mode = cpu_to_le16(mode);
567         if (S_ISCHR(mode) || S_ISBLK(mode))
568                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
569
570         fe->i_links_count = cpu_to_le16(inode->i_nlink);
571
572         fe->i_last_eb_blk = 0;
573         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
574         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
575         fe->i_atime = fe->i_ctime = fe->i_mtime =
576                 cpu_to_le64(CURRENT_TIME.tv_sec);
577         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
578                 cpu_to_le32(CURRENT_TIME.tv_nsec);
579         fe->i_dtime = 0;
580
581         fel = &fe->id2.i_list;
582         fel->l_tree_depth = 0;
583         fel->l_next_free_rec = 0;
584         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
585
586         status = ocfs2_journal_dirty(handle, *new_fe_bh);
587         if (status < 0) {
588                 mlog_errno(status);
589                 goto leave;
590         }
591
592         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
593                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
594                      "i_blkno=%llu, i_ino=%lu\n",
595                      (unsigned long long) (*new_fe_bh)->b_blocknr,
596                      (unsigned long long)fe->i_blkno, inode->i_ino);
597                 BUG();
598         }
599
600         ocfs2_inode_set_new(osb, inode);
601         status = ocfs2_create_new_inode_locks(inode);
602         if (status < 0)
603                 mlog_errno(status);
604
605         status = 0; /* error in ocfs2_create_new_inode_locks is not
606                      * critical */
607
608         *ret_inode = inode;
609 leave:
610         if (status < 0) {
611                 if (*new_fe_bh) {
612                         brelse(*new_fe_bh);
613                         *new_fe_bh = NULL;
614                 }
615                 if (inode)
616                         iput(inode);
617         }
618
619         mlog_exit(status);
620         return status;
621 }
622
623 static int ocfs2_mkdir(struct inode *dir,
624                        struct dentry *dentry,
625                        int mode)
626 {
627         int ret;
628
629         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
630                    dentry->d_name.len, dentry->d_name.name);
631         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
632         mlog_exit(ret);
633
634         return ret;
635 }
636
637 static int ocfs2_create(struct inode *dir,
638                         struct dentry *dentry,
639                         int mode,
640                         struct nameidata *nd)
641 {
642         int ret;
643
644         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
645                    dentry->d_name.len, dentry->d_name.name);
646         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
647         mlog_exit(ret);
648
649         return ret;
650 }
651
652 static int ocfs2_link(struct dentry *old_dentry,
653                       struct inode *dir,
654                       struct dentry *dentry)
655 {
656         struct ocfs2_journal_handle *handle = NULL;
657         struct inode *inode = old_dentry->d_inode;
658         int err;
659         struct buffer_head *fe_bh = NULL;
660         struct buffer_head *parent_fe_bh = NULL;
661         struct buffer_head *de_bh = NULL;
662         struct ocfs2_dinode *fe = NULL;
663         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
664
665         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
666                    old_dentry->d_name.len, old_dentry->d_name.name,
667                    dentry->d_name.len, dentry->d_name.name);
668
669         if (S_ISDIR(inode->i_mode)) {
670                 err = -EPERM;
671                 goto bail;
672         }
673
674         handle = ocfs2_alloc_handle(osb);
675         if (handle == NULL) {
676                 err = -ENOMEM;
677                 goto bail;
678         }
679
680         err = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
681         if (err < 0) {
682                 if (err != -ENOENT)
683                         mlog_errno(err);
684                 goto bail;
685         }
686
687         if (!dir->i_nlink) {
688                 err = -ENOENT;
689                 goto bail;
690         }
691
692         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
693                                         dentry->d_name.len);
694         if (err)
695                 goto bail;
696
697         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
698                                            dentry->d_name.name,
699                                            dentry->d_name.len, &de_bh);
700         if (err < 0) {
701                 mlog_errno(err);
702                 goto bail;
703         }
704
705         err = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
706         if (err < 0) {
707                 if (err != -ENOENT)
708                         mlog_errno(err);
709                 goto bail;
710         }
711
712         fe = (struct ocfs2_dinode *) fe_bh->b_data;
713         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
714                 err = -EMLINK;
715                 goto bail;
716         }
717
718         handle = ocfs2_start_trans(osb, handle, OCFS2_LINK_CREDITS);
719         if (IS_ERR(handle)) {
720                 err = PTR_ERR(handle);
721                 handle = NULL;
722                 mlog_errno(err);
723                 goto bail;
724         }
725
726         err = ocfs2_journal_access(handle, inode, fe_bh,
727                                    OCFS2_JOURNAL_ACCESS_WRITE);
728         if (err < 0) {
729                 mlog_errno(err);
730                 goto bail;
731         }
732
733         inode->i_nlink++;
734         inode->i_ctime = CURRENT_TIME;
735         fe->i_links_count = cpu_to_le16(inode->i_nlink);
736         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
737         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
738
739         err = ocfs2_journal_dirty(handle, fe_bh);
740         if (err < 0) {
741                 le16_add_cpu(&fe->i_links_count, -1);
742                 drop_nlink(inode);
743                 mlog_errno(err);
744                 goto bail;
745         }
746
747         err = ocfs2_add_entry(handle, dentry, inode,
748                               OCFS2_I(inode)->ip_blkno,
749                               parent_fe_bh, de_bh);
750         if (err) {
751                 le16_add_cpu(&fe->i_links_count, -1);
752                 drop_nlink(inode);
753                 mlog_errno(err);
754                 goto bail;
755         }
756
757         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
758         if (err) {
759                 mlog_errno(err);
760                 goto bail;
761         }
762
763         atomic_inc(&inode->i_count);
764         dentry->d_op = &ocfs2_dentry_ops;
765         d_instantiate(dentry, inode);
766 bail:
767         if (handle)
768                 ocfs2_commit_trans(handle);
769         if (de_bh)
770                 brelse(de_bh);
771         if (fe_bh)
772                 brelse(fe_bh);
773         if (parent_fe_bh)
774                 brelse(parent_fe_bh);
775
776         mlog_exit(err);
777
778         return err;
779 }
780
781 /*
782  * Takes and drops an exclusive lock on the given dentry. This will
783  * force other nodes to drop it.
784  */
785 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
786 {
787         int ret;
788
789         ret = ocfs2_dentry_lock(dentry, 1);
790         if (ret)
791                 mlog_errno(ret);
792         else
793                 ocfs2_dentry_unlock(dentry, 1);
794
795         return ret;
796 }
797
798 static int ocfs2_unlink(struct inode *dir,
799                         struct dentry *dentry)
800 {
801         int status;
802         unsigned int saved_nlink = 0;
803         struct inode *inode = dentry->d_inode;
804         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
805         u64 blkno;
806         struct ocfs2_dinode *fe = NULL;
807         struct buffer_head *fe_bh = NULL;
808         struct buffer_head *parent_node_bh = NULL;
809         struct ocfs2_journal_handle *handle = NULL;
810         struct ocfs2_dir_entry *dirent = NULL;
811         struct buffer_head *dirent_bh = NULL;
812         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
813         struct buffer_head *orphan_entry_bh = NULL;
814
815         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
816                    dentry->d_name.len, dentry->d_name.name);
817
818         BUG_ON(dentry->d_parent->d_inode != dir);
819
820         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
821
822         if (inode == osb->root_inode) {
823                 mlog(0, "Cannot delete the root directory\n");
824                 status = -EPERM;
825                 goto leave;
826         }
827
828         handle = ocfs2_alloc_handle(osb);
829         if (handle == NULL) {
830                 status = -ENOMEM;
831                 mlog_errno(status);
832                 goto leave;
833         }
834
835         status = ocfs2_meta_lock(dir, handle, &parent_node_bh, 1);
836         if (status < 0) {
837                 if (status != -ENOENT)
838                         mlog_errno(status);
839                 goto leave;
840         }
841
842         status = ocfs2_find_files_on_disk(dentry->d_name.name,
843                                           dentry->d_name.len, &blkno,
844                                           dir, &dirent_bh, &dirent);
845         if (status < 0) {
846                 if (status != -ENOENT)
847                         mlog_errno(status);
848                 goto leave;
849         }
850
851         if (OCFS2_I(inode)->ip_blkno != blkno) {
852                 status = -ENOENT;
853
854                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
855                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
856                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
857                 goto leave;
858         }
859
860         status = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
861         if (status < 0) {
862                 if (status != -ENOENT)
863                         mlog_errno(status);
864                 goto leave;
865         }
866
867         if (S_ISDIR(inode->i_mode)) {
868                 if (!ocfs2_empty_dir(inode)) {
869                         status = -ENOTEMPTY;
870                         goto leave;
871                 } else if (inode->i_nlink != 2) {
872                         status = -ENOTEMPTY;
873                         goto leave;
874                 }
875         }
876
877         /* There are still a few steps left until we can consider the
878          * unlink to have succeeded. Save off nlink here before
879          * modification so we can set it back in case we hit an issue
880          * before commit. */
881         saved_nlink = inode->i_nlink;
882         if (S_ISDIR(inode->i_mode))
883                 inode->i_nlink = 0;
884         else
885                 inode->i_nlink--;
886
887         status = ocfs2_remote_dentry_delete(dentry);
888         if (status < 0) {
889                 /* This vote should succeed under all normal
890                  * circumstances. */
891                 mlog_errno(status);
892                 goto leave;
893         }
894
895         if (!inode->i_nlink) {
896                 status = ocfs2_prepare_orphan_dir(osb, handle, inode,
897                                                   orphan_name,
898                                                   &orphan_entry_bh);
899                 if (status < 0) {
900                         mlog_errno(status);
901                         goto leave;
902                 }
903         }
904
905         handle = ocfs2_start_trans(osb, handle, OCFS2_UNLINK_CREDITS);
906         if (IS_ERR(handle)) {
907                 status = PTR_ERR(handle);
908                 handle = NULL;
909                 mlog_errno(status);
910                 goto leave;
911         }
912
913         status = ocfs2_journal_access(handle, inode, fe_bh,
914                                       OCFS2_JOURNAL_ACCESS_WRITE);
915         if (status < 0) {
916                 mlog_errno(status);
917                 goto leave;
918         }
919
920         fe = (struct ocfs2_dinode *) fe_bh->b_data;
921
922         if (!inode->i_nlink) {
923                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
924                                           orphan_entry_bh);
925                 if (status < 0) {
926                         mlog_errno(status);
927                         goto leave;
928                 }
929         }
930
931         /* delete the name from the parent dir */
932         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
933         if (status < 0) {
934                 mlog_errno(status);
935                 goto leave;
936         }
937
938         /* We can set nlink on the dinode now. clear the saved version
939          * so that it doesn't get set later. */
940         fe->i_links_count = cpu_to_le16(inode->i_nlink);
941         saved_nlink = 0;
942
943         status = ocfs2_journal_dirty(handle, fe_bh);
944         if (status < 0) {
945                 mlog_errno(status);
946                 goto leave;
947         }
948
949         if (S_ISDIR(inode->i_mode)) {
950                 dir->i_nlink--;
951                 status = ocfs2_mark_inode_dirty(handle, dir,
952                                                 parent_node_bh);
953                 if (status < 0) {
954                         mlog_errno(status);
955                         dir->i_nlink++;
956                 }
957         }
958
959 leave:
960         if (status < 0 && saved_nlink)
961                 inode->i_nlink = saved_nlink;
962
963         if (handle)
964                 ocfs2_commit_trans(handle);
965
966         if (fe_bh)
967                 brelse(fe_bh);
968
969         if (dirent_bh)
970                 brelse(dirent_bh);
971
972         if (parent_node_bh)
973                 brelse(parent_node_bh);
974
975         if (orphan_entry_bh)
976                 brelse(orphan_entry_bh);
977
978         mlog_exit(status);
979
980         return status;
981 }
982
983 /*
984  * The only place this should be used is rename!
985  * if they have the same id, then the 1st one is the only one locked.
986  */
987 static int ocfs2_double_lock(struct ocfs2_super *osb,
988                              struct ocfs2_journal_handle *handle,
989                              struct buffer_head **bh1,
990                              struct inode *inode1,
991                              struct buffer_head **bh2,
992                              struct inode *inode2)
993 {
994         int status;
995         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
996         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
997         struct buffer_head **tmpbh;
998         struct inode *tmpinode;
999
1000         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1001                    (unsigned long long)oi1->ip_blkno,
1002                    (unsigned long long)oi2->ip_blkno);
1003
1004         BUG_ON(!handle);
1005
1006         if (*bh1)
1007                 *bh1 = NULL;
1008         if (*bh2)
1009                 *bh2 = NULL;
1010
1011         /* we always want to lock the one with the lower lockid first. */
1012         if (oi1->ip_blkno != oi2->ip_blkno) {
1013                 if (oi1->ip_blkno < oi2->ip_blkno) {
1014                         /* switch id1 and id2 around */
1015                         mlog(0, "switching them around...\n");
1016                         tmpbh = bh2;
1017                         bh2 = bh1;
1018                         bh1 = tmpbh;
1019
1020                         tmpinode = inode2;
1021                         inode2 = inode1;
1022                         inode1 = tmpinode;
1023                 }
1024                 /* lock id2 */
1025                 status = ocfs2_meta_lock(inode2, handle, bh2, 1);
1026                 if (status < 0) {
1027                         if (status != -ENOENT)
1028                                 mlog_errno(status);
1029                         goto bail;
1030                 }
1031         }
1032         /* lock id1 */
1033         status = ocfs2_meta_lock(inode1, handle, bh1, 1);
1034         if (status < 0) {
1035                 if (status != -ENOENT)
1036                         mlog_errno(status);
1037                 goto bail;
1038         }
1039 bail:
1040         mlog_exit(status);
1041         return status;
1042 }
1043
1044 #define PARENT_INO(buffer) \
1045         ((struct ocfs2_dir_entry *) \
1046          ((char *)buffer + \
1047           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1048
1049 static int ocfs2_rename(struct inode *old_dir,
1050                         struct dentry *old_dentry,
1051                         struct inode *new_dir,
1052                         struct dentry *new_dentry)
1053 {
1054         int status = 0, rename_lock = 0;
1055         struct inode *old_inode = old_dentry->d_inode;
1056         struct inode *new_inode = new_dentry->d_inode;
1057         struct ocfs2_dinode *newfe = NULL;
1058         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1059         struct buffer_head *orphan_entry_bh = NULL;
1060         struct buffer_head *newfe_bh = NULL;
1061         struct buffer_head *insert_entry_bh = NULL;
1062         struct ocfs2_super *osb = NULL;
1063         u64 newfe_blkno;
1064         struct ocfs2_journal_handle *handle = NULL;
1065         struct buffer_head *old_dir_bh = NULL;
1066         struct buffer_head *new_dir_bh = NULL;
1067         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1068                                                                // and new_dentry
1069         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1070         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1071                                                     // this is the 1st dirent bh
1072         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1073
1074         /* At some point it might be nice to break this function up a
1075          * bit. */
1076
1077         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1078                    old_dir, old_dentry, new_dir, new_dentry,
1079                    old_dentry->d_name.len, old_dentry->d_name.name,
1080                    new_dentry->d_name.len, new_dentry->d_name.name);
1081
1082         osb = OCFS2_SB(old_dir->i_sb);
1083
1084         if (new_inode) {
1085                 if (!igrab(new_inode))
1086                         BUG();
1087         }
1088
1089         if (atomic_read(&old_dentry->d_count) > 2) {
1090                 shrink_dcache_parent(old_dentry);
1091                 if (atomic_read(&old_dentry->d_count) > 2) {
1092                         status = -EBUSY;
1093                         goto bail;
1094                 }
1095         }
1096
1097         /* Assume a directory heirarchy thusly:
1098          * a/b/c
1099          * a/d
1100          * a,b,c, and d are all directories.
1101          *
1102          * from cwd of 'a' on both nodes:
1103          * node1: mv b/c d
1104          * node2: mv d   b/c
1105          *
1106          * And that's why, just like the VFS, we need a file system
1107          * rename lock. */
1108         if (old_dentry != new_dentry) {
1109                 status = ocfs2_rename_lock(osb);
1110                 if (status < 0) {
1111                         mlog_errno(status);
1112                         goto bail;
1113                 }
1114                 rename_lock = 1;
1115         }
1116
1117         handle = ocfs2_alloc_handle(osb);
1118         if (handle == NULL) {
1119                 status = -ENOMEM;
1120                 mlog_errno(status);
1121                 goto bail;
1122         }
1123
1124         /* if old and new are the same, this'll just do one lock. */
1125         status = ocfs2_double_lock(osb, handle,
1126                                   &old_dir_bh, old_dir,
1127                                   &new_dir_bh, new_dir);
1128         if (status < 0) {
1129                 mlog_errno(status);
1130                 goto bail;
1131         }
1132
1133         /* make sure both dirs have bhs
1134          * get an extra ref on old_dir_bh if old==new */
1135         if (!new_dir_bh) {
1136                 if (old_dir_bh) {
1137                         new_dir_bh = old_dir_bh;
1138                         get_bh(new_dir_bh);
1139                 } else {
1140                         mlog(ML_ERROR, "no old_dir_bh!\n");
1141                         status = -EIO;
1142                         goto bail;
1143                 }
1144         }
1145
1146         /*
1147          * Though we don't require an inode meta data update if
1148          * old_inode is not a directory, we lock anyway here to ensure
1149          * the vote thread on other nodes won't have to concurrently
1150          * downconvert the inode and the dentry locks.
1151          */
1152         status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
1153         if (status < 0) {
1154                 if (status != -ENOENT)
1155                         mlog_errno(status);
1156                 goto bail;
1157         }
1158
1159         status = ocfs2_remote_dentry_delete(old_dentry);
1160         if (status < 0) {
1161                 mlog_errno(status);
1162                 goto bail;
1163         }
1164
1165         if (S_ISDIR(old_inode->i_mode)) {
1166                 status = -EIO;
1167                 old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1168                 if (!old_inode_de_bh)
1169                         goto bail;
1170
1171                 status = -EIO;
1172                 if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1173                     OCFS2_I(old_dir)->ip_blkno)
1174                         goto bail;
1175                 status = -EMLINK;
1176                 if (!new_inode && new_dir!=old_dir &&
1177                     new_dir->i_nlink >= OCFS2_LINK_MAX)
1178                         goto bail;
1179         }
1180
1181         status = -ENOENT;
1182         old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1183                                      old_dentry->d_name.len,
1184                                      old_dir, &old_de);
1185         if (!old_de_bh)
1186                 goto bail;
1187
1188         /*
1189          *  Check for inode number is _not_ due to possible IO errors.
1190          *  We might rmdir the source, keep it as pwd of some process
1191          *  and merrily kill the link to whatever was created under the
1192          *  same name. Goodbye sticky bit ;-<
1193          */
1194         if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1195                 goto bail;
1196
1197         /* check if the target already exists (in which case we need
1198          * to delete it */
1199         status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1200                                           new_dentry->d_name.len,
1201                                           &newfe_blkno, new_dir, &new_de_bh,
1202                                           &new_de);
1203         /* The only error we allow here is -ENOENT because the new
1204          * file not existing is perfectly valid. */
1205         if ((status < 0) && (status != -ENOENT)) {
1206                 /* If we cannot find the file specified we should just */
1207                 /* return the error... */
1208                 mlog_errno(status);
1209                 goto bail;
1210         }
1211
1212         if (!new_de && new_inode)
1213                 mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1214                      "directory!", new_inode->i_ino);
1215
1216         /* In case we need to overwrite an existing file, we blow it
1217          * away first */
1218         if (new_de) {
1219                 /* VFS didn't think there existed an inode here, but
1220                  * someone else in the cluster must have raced our
1221                  * rename to create one. Today we error cleanly, in
1222                  * the future we should consider calling iget to build
1223                  * a new struct inode for this entry. */
1224                 if (!new_inode) {
1225                         status = -EACCES;
1226
1227                         mlog(0, "We found an inode for name %.*s but VFS "
1228                              "didn't give us one.\n", new_dentry->d_name.len,
1229                              new_dentry->d_name.name);
1230                         goto bail;
1231                 }
1232
1233                 if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1234                         status = -EACCES;
1235
1236                         mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1237                              (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1238                              (unsigned long long)newfe_blkno,
1239                              OCFS2_I(new_inode)->ip_flags);
1240                         goto bail;
1241                 }
1242
1243                 status = ocfs2_meta_lock(new_inode, handle, &newfe_bh, 1);
1244                 if (status < 0) {
1245                         if (status != -ENOENT)
1246                                 mlog_errno(status);
1247                         goto bail;
1248                 }
1249
1250                 status = ocfs2_remote_dentry_delete(new_dentry);
1251                 if (status < 0) {
1252                         mlog_errno(status);
1253                         goto bail;
1254                 }
1255
1256                 newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1257
1258                 mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1259                      "newfebh=%p bhblocknr=%llu\n", new_de,
1260                      (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1261                      (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1262
1263                 if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1264                         status = ocfs2_prepare_orphan_dir(osb, handle,
1265                                                           new_inode,
1266                                                           orphan_name,
1267                                                           &orphan_entry_bh);
1268                         if (status < 0) {
1269                                 mlog_errno(status);
1270                                 goto bail;
1271                         }
1272                 }
1273         } else {
1274                 BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1275
1276                 status = ocfs2_check_dir_for_entry(new_dir,
1277                                                    new_dentry->d_name.name,
1278                                                    new_dentry->d_name.len);
1279                 if (status)
1280                         goto bail;
1281
1282                 status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1283                                                       new_dentry->d_name.name,
1284                                                       new_dentry->d_name.len,
1285                                                       &insert_entry_bh);
1286                 if (status < 0) {
1287                         mlog_errno(status);
1288                         goto bail;
1289                 }
1290         }
1291
1292         handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1293         if (IS_ERR(handle)) {
1294                 status = PTR_ERR(handle);
1295                 handle = NULL;
1296                 mlog_errno(status);
1297                 goto bail;
1298         }
1299
1300         if (new_de) {
1301                 if (S_ISDIR(new_inode->i_mode)) {
1302                         if (!ocfs2_empty_dir(new_inode) ||
1303                             new_inode->i_nlink != 2) {
1304                                 status = -ENOTEMPTY;
1305                                 goto bail;
1306                         }
1307                 }
1308                 status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1309                                               OCFS2_JOURNAL_ACCESS_WRITE);
1310                 if (status < 0) {
1311                         mlog_errno(status);
1312                         goto bail;
1313                 }
1314
1315                 if (S_ISDIR(new_inode->i_mode) ||
1316                     (newfe->i_links_count == cpu_to_le16(1))){
1317                         status = ocfs2_orphan_add(osb, handle, new_inode,
1318                                                   newfe, orphan_name,
1319                                                   orphan_entry_bh);
1320                         if (status < 0) {
1321                                 mlog_errno(status);
1322                                 goto bail;
1323                         }
1324                 }
1325
1326                 /* change the dirent to point to the correct inode */
1327                 status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1328                                               OCFS2_JOURNAL_ACCESS_WRITE);
1329                 if (status < 0) {
1330                         mlog_errno(status);
1331                         goto bail;
1332                 }
1333                 new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1334                 new_de->file_type = old_de->file_type;
1335                 new_dir->i_version++;
1336                 status = ocfs2_journal_dirty(handle, new_de_bh);
1337                 if (status < 0) {
1338                         mlog_errno(status);
1339                         goto bail;
1340                 }
1341
1342                 if (S_ISDIR(new_inode->i_mode))
1343                         newfe->i_links_count = 0;
1344                 else
1345                         le16_add_cpu(&newfe->i_links_count, -1);
1346
1347                 status = ocfs2_journal_dirty(handle, newfe_bh);
1348                 if (status < 0) {
1349                         mlog_errno(status);
1350                         goto bail;
1351                 }
1352         } else {
1353                 /* if the name was not found in new_dir, add it now */
1354                 status = ocfs2_add_entry(handle, new_dentry, old_inode,
1355                                          OCFS2_I(old_inode)->ip_blkno,
1356                                          new_dir_bh, insert_entry_bh);
1357         }
1358
1359         old_inode->i_ctime = CURRENT_TIME;
1360         mark_inode_dirty(old_inode);
1361
1362         /* now that the name has been added to new_dir, remove the old name */
1363         status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1364         if (status < 0) {
1365                 mlog_errno(status);
1366                 goto bail;
1367         }
1368
1369         if (new_inode) {
1370                 new_inode->i_nlink--;
1371                 new_inode->i_ctime = CURRENT_TIME;
1372         }
1373         old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1374         if (old_inode_de_bh) {
1375                 status = ocfs2_journal_access(handle, old_inode,
1376                                              old_inode_de_bh,
1377                                              OCFS2_JOURNAL_ACCESS_WRITE);
1378                 PARENT_INO(old_inode_de_bh->b_data) =
1379                         cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1380                 status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1381                 old_dir->i_nlink--;
1382                 if (new_inode) {
1383                         new_inode->i_nlink--;
1384                 } else {
1385                         new_dir->i_nlink++;
1386                         mark_inode_dirty(new_dir);
1387                 }
1388         }
1389         mark_inode_dirty(old_dir);
1390         if (new_inode)
1391                 mark_inode_dirty(new_inode);
1392
1393         if (old_dir != new_dir)
1394                 if (new_dir_nlink != new_dir->i_nlink) {
1395                         if (!new_dir_bh) {
1396                                 mlog(ML_ERROR, "need to change nlink for new "
1397                                      "dir %llu from %d to %d but bh is NULL\n",
1398                                      (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1399                                      (int)new_dir_nlink, new_dir->i_nlink);
1400                         } else {
1401                                 struct ocfs2_dinode *fe;
1402                                 status = ocfs2_journal_access(handle,
1403                                                               new_dir,
1404                                                               new_dir_bh,
1405                                                               OCFS2_JOURNAL_ACCESS_WRITE);
1406                                 fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1407                                 fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1408                                 status = ocfs2_journal_dirty(handle, new_dir_bh);
1409                         }
1410                 }
1411
1412         if (old_dir_nlink != old_dir->i_nlink) {
1413                 if (!old_dir_bh) {
1414                         mlog(ML_ERROR, "need to change nlink for old dir "
1415                              "%llu from %d to %d but bh is NULL!\n",
1416                              (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1417                              (int)old_dir_nlink, old_dir->i_nlink);
1418                 } else {
1419                         struct ocfs2_dinode *fe;
1420                         status = ocfs2_journal_access(handle, old_dir,
1421                                                       old_dir_bh,
1422                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1423                         fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1424                         fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1425                         status = ocfs2_journal_dirty(handle, old_dir_bh);
1426                 }
1427         }
1428
1429         ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1430         status = 0;
1431 bail:
1432         if (rename_lock)
1433                 ocfs2_rename_unlock(osb);
1434
1435         if (handle)
1436                 ocfs2_commit_trans(handle);
1437
1438         if (new_inode)
1439                 sync_mapping_buffers(old_inode->i_mapping);
1440
1441         if (new_inode)
1442                 iput(new_inode);
1443         if (newfe_bh)
1444                 brelse(newfe_bh);
1445         if (old_dir_bh)
1446                 brelse(old_dir_bh);
1447         if (new_dir_bh)
1448                 brelse(new_dir_bh);
1449         if (new_de_bh)
1450                 brelse(new_de_bh);
1451         if (old_de_bh)
1452                 brelse(old_de_bh);
1453         if (old_inode_de_bh)
1454                 brelse(old_inode_de_bh);
1455         if (orphan_entry_bh)
1456                 brelse(orphan_entry_bh);
1457         if (insert_entry_bh)
1458                 brelse(insert_entry_bh);
1459
1460         mlog_exit(status);
1461
1462         return status;
1463 }
1464
1465 /*
1466  * we expect i_size = strlen(symname). Copy symname into the file
1467  * data, including the null terminator.
1468  */
1469 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1470                                      struct ocfs2_journal_handle *handle,
1471                                      struct inode *inode,
1472                                      const char *symname)
1473 {
1474         struct buffer_head **bhs = NULL;
1475         const char *c;
1476         struct super_block *sb = osb->sb;
1477         u64 p_blkno;
1478         int p_blocks;
1479         int virtual, blocks, status, i, bytes_left;
1480
1481         bytes_left = i_size_read(inode) + 1;
1482         /* we can't trust i_blocks because we're actually going to
1483          * write i_size + 1 bytes. */
1484         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1485
1486         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1487                         (unsigned long long)inode->i_blocks,
1488                         i_size_read(inode), blocks);
1489
1490         /* Sanity check -- make sure we're going to fit. */
1491         if (bytes_left >
1492             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1493                 status = -EIO;
1494                 mlog_errno(status);
1495                 goto bail;
1496         }
1497
1498         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1499         if (!bhs) {
1500                 status = -ENOMEM;
1501                 mlog_errno(status);
1502                 goto bail;
1503         }
1504
1505         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1506                                              &p_blocks);
1507         if (status < 0) {
1508                 mlog_errno(status);
1509                 goto bail;
1510         }
1511
1512         /* links can never be larger than one cluster so we know this
1513          * is all going to be contiguous, but do a sanity check
1514          * anyway. */
1515         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1516                 status = -EIO;
1517                 mlog_errno(status);
1518                 goto bail;
1519         }
1520
1521         virtual = 0;
1522         while(bytes_left > 0) {
1523                 c = &symname[virtual * sb->s_blocksize];
1524
1525                 bhs[virtual] = sb_getblk(sb, p_blkno);
1526                 if (!bhs[virtual]) {
1527                         status = -ENOMEM;
1528                         mlog_errno(status);
1529                         goto bail;
1530                 }
1531                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1532
1533                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1534                                               OCFS2_JOURNAL_ACCESS_CREATE);
1535                 if (status < 0) {
1536                         mlog_errno(status);
1537                         goto bail;
1538                 }
1539
1540                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1541
1542                 memcpy(bhs[virtual]->b_data, c,
1543                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1544                        bytes_left);
1545
1546                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1547                 if (status < 0) {
1548                         mlog_errno(status);
1549                         goto bail;
1550                 }
1551
1552                 virtual++;
1553                 p_blkno++;
1554                 bytes_left -= sb->s_blocksize;
1555         }
1556
1557         status = 0;
1558 bail:
1559
1560         if (bhs) {
1561                 for(i = 0; i < blocks; i++)
1562                         if (bhs[i])
1563                                 brelse(bhs[i]);
1564                 kfree(bhs);
1565         }
1566
1567         mlog_exit(status);
1568         return status;
1569 }
1570
1571 static int ocfs2_symlink(struct inode *dir,
1572                          struct dentry *dentry,
1573                          const char *symname)
1574 {
1575         int status, l, credits;
1576         u64 newsize;
1577         struct ocfs2_super *osb = NULL;
1578         struct inode *inode = NULL;
1579         struct super_block *sb;
1580         struct buffer_head *new_fe_bh = NULL;
1581         struct buffer_head *de_bh = NULL;
1582         struct buffer_head *parent_fe_bh = NULL;
1583         struct ocfs2_dinode *fe = NULL;
1584         struct ocfs2_dinode *dirfe;
1585         struct ocfs2_journal_handle *handle = NULL;
1586         struct ocfs2_alloc_context *inode_ac = NULL;
1587         struct ocfs2_alloc_context *data_ac = NULL;
1588
1589         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1590                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1591
1592         sb = dir->i_sb;
1593         osb = OCFS2_SB(sb);
1594
1595         l = strlen(symname) + 1;
1596
1597         credits = ocfs2_calc_symlink_credits(sb);
1598
1599         handle = ocfs2_alloc_handle(osb);
1600         if (handle == NULL) {
1601                 status = -ENOMEM;
1602                 mlog_errno(status);
1603                 goto bail;
1604         }
1605
1606         /* lock the parent directory */
1607         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1608         if (status < 0) {
1609                 if (status != -ENOENT)
1610                         mlog_errno(status);
1611                 goto bail;
1612         }
1613
1614         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1615         if (!dirfe->i_links_count) {
1616                 /* can't make a file in a deleted directory. */
1617                 status = -ENOENT;
1618                 goto bail;
1619         }
1620
1621         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1622                                            dentry->d_name.len);
1623         if (status)
1624                 goto bail;
1625
1626         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1627                                               dentry->d_name.name,
1628                                               dentry->d_name.len, &de_bh);
1629         if (status < 0) {
1630                 mlog_errno(status);
1631                 goto bail;
1632         }
1633
1634         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1635         if (status < 0) {
1636                 if (status != -ENOSPC)
1637                         mlog_errno(status);
1638                 goto bail;
1639         }
1640
1641         /* don't reserve bitmap space for fast symlinks. */
1642         if (l > ocfs2_fast_symlink_chars(sb)) {
1643                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1644                 if (status < 0) {
1645                         if (status != -ENOSPC)
1646                                 mlog_errno(status);
1647                         goto bail;
1648                 }
1649         }
1650
1651         handle = ocfs2_start_trans(osb, handle, credits);
1652         if (IS_ERR(handle)) {
1653                 status = PTR_ERR(handle);
1654                 handle = NULL;
1655                 mlog_errno(status);
1656                 goto bail;
1657         }
1658
1659         status = ocfs2_mknod_locked(osb, dir, dentry,
1660                                     S_IFLNK | S_IRWXUGO, 0,
1661                                     &new_fe_bh, parent_fe_bh, handle,
1662                                     &inode, inode_ac);
1663         if (status < 0) {
1664                 mlog_errno(status);
1665                 goto bail;
1666         }
1667
1668         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1669         inode->i_rdev = 0;
1670         newsize = l - 1;
1671         if (l > ocfs2_fast_symlink_chars(sb)) {
1672                 inode->i_op = &ocfs2_symlink_inode_operations;
1673                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1674                                                     handle, data_ac, NULL,
1675                                                     NULL);
1676                 if (status < 0) {
1677                         if (status != -ENOSPC && status != -EINTR) {
1678                                 mlog(ML_ERROR,
1679                                      "Failed to extend file to %llu\n",
1680                                      (unsigned long long)newsize);
1681                                 mlog_errno(status);
1682                                 status = -ENOSPC;
1683                         }
1684                         goto bail;
1685                 }
1686                 i_size_write(inode, newsize);
1687                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1688         } else {
1689                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1690                 memcpy((char *) fe->id2.i_symlink, symname, l);
1691                 i_size_write(inode, newsize);
1692                 inode->i_blocks = 0;
1693         }
1694
1695         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1696         if (status < 0) {
1697                 mlog_errno(status);
1698                 goto bail;
1699         }
1700
1701         if (!ocfs2_inode_is_fast_symlink(inode)) {
1702                 status = ocfs2_create_symlink_data(osb, handle, inode,
1703                                                    symname);
1704                 if (status < 0) {
1705                         mlog_errno(status);
1706                         goto bail;
1707                 }
1708         }
1709
1710         status = ocfs2_add_entry(handle, dentry, inode,
1711                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1712                                  de_bh);
1713         if (status < 0) {
1714                 mlog_errno(status);
1715                 goto bail;
1716         }
1717
1718         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1719         if (status) {
1720                 mlog_errno(status);
1721                 goto bail;
1722         }
1723
1724         insert_inode_hash(inode);
1725         dentry->d_op = &ocfs2_dentry_ops;
1726         d_instantiate(dentry, inode);
1727 bail:
1728         if (handle)
1729                 ocfs2_commit_trans(handle);
1730         if (new_fe_bh)
1731                 brelse(new_fe_bh);
1732         if (parent_fe_bh)
1733                 brelse(parent_fe_bh);
1734         if (de_bh)
1735                 brelse(de_bh);
1736         if (inode_ac)
1737                 ocfs2_free_alloc_context(inode_ac);
1738         if (data_ac)
1739                 ocfs2_free_alloc_context(data_ac);
1740         if ((status < 0) && inode)
1741                 iput(inode);
1742
1743         mlog_exit(status);
1744
1745         return status;
1746 }
1747
1748 int ocfs2_check_dir_entry(struct inode * dir,
1749                           struct ocfs2_dir_entry * de,
1750                           struct buffer_head * bh,
1751                           unsigned long offset)
1752 {
1753         const char *error_msg = NULL;
1754         const int rlen = le16_to_cpu(de->rec_len);
1755
1756         if (rlen < OCFS2_DIR_REC_LEN(1))
1757                 error_msg = "rec_len is smaller than minimal";
1758         else if (rlen % 4 != 0)
1759                 error_msg = "rec_len % 4 != 0";
1760         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1761                 error_msg = "rec_len is too small for name_len";
1762         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1763                 error_msg = "directory entry across blocks";
1764
1765         if (error_msg != NULL)
1766                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1767                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1768                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1769                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1770                      de->name_len);
1771         return error_msg == NULL ? 1 : 0;
1772 }
1773
1774 /* we don't always have a dentry for what we want to add, so people
1775  * like orphan dir can call this instead.
1776  *
1777  * If you pass me insert_bh, I'll skip the search of the other dir
1778  * blocks and put the record in there.
1779  */
1780 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1781                              struct inode *dir,
1782                              const char *name, int namelen,
1783                              struct inode *inode, u64 blkno,
1784                              struct buffer_head *parent_fe_bh,
1785                              struct buffer_head *insert_bh)
1786 {
1787         unsigned long offset;
1788         unsigned short rec_len;
1789         struct ocfs2_dir_entry *de, *de1;
1790         struct super_block *sb;
1791         int retval, status;
1792
1793         mlog_entry_void();
1794
1795         sb = dir->i_sb;
1796
1797         if (!namelen)
1798                 return -EINVAL;
1799
1800         rec_len = OCFS2_DIR_REC_LEN(namelen);
1801         offset = 0;
1802         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1803         while (1) {
1804                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1805                 /* These checks should've already been passed by the
1806                  * prepare function, but I guess we can leave them
1807                  * here anyway. */
1808                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1809                         retval = -ENOENT;
1810                         goto bail;
1811                 }
1812                 if (ocfs2_match(namelen, name, de)) {
1813                         retval = -EEXIST;
1814                         goto bail;
1815                 }
1816                 if (((le64_to_cpu(de->inode) == 0) &&
1817                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1818                     (le16_to_cpu(de->rec_len) >=
1819                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1820                         status = ocfs2_journal_access(handle, dir, insert_bh,
1821                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1822                         /* By now the buffer is marked for journaling */
1823                         offset += le16_to_cpu(de->rec_len);
1824                         if (le64_to_cpu(de->inode)) {
1825                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1826                                         OCFS2_DIR_REC_LEN(de->name_len));
1827                                 de1->rec_len =
1828                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1829                                         OCFS2_DIR_REC_LEN(de->name_len));
1830                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1831                                 de = de1;
1832                         }
1833                         de->file_type = OCFS2_FT_UNKNOWN;
1834                         if (blkno) {
1835                                 de->inode = cpu_to_le64(blkno);
1836                                 ocfs2_set_de_type(de, inode->i_mode);
1837                         } else
1838                                 de->inode = 0;
1839                         de->name_len = namelen;
1840                         memcpy(de->name, name, namelen);
1841
1842                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1843                         dir->i_version++;
1844                         status = ocfs2_journal_dirty(handle, insert_bh);
1845                         retval = 0;
1846                         goto bail;
1847                 }
1848                 offset += le16_to_cpu(de->rec_len);
1849                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1850         }
1851
1852         /* when you think about it, the assert above should prevent us
1853          * from ever getting here. */
1854         retval = -ENOSPC;
1855 bail:
1856
1857         mlog_exit(retval);
1858         return retval;
1859 }
1860
1861
1862 /*
1863  * ocfs2_delete_entry deletes a directory entry by merging it with the
1864  * previous entry
1865  */
1866 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1867                               struct inode *dir,
1868                               struct ocfs2_dir_entry *de_del,
1869                               struct buffer_head *bh)
1870 {
1871         struct ocfs2_dir_entry *de, *pde;
1872         int i, status = -ENOENT;
1873
1874         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1875
1876         i = 0;
1877         pde = NULL;
1878         de = (struct ocfs2_dir_entry *) bh->b_data;
1879         while (i < bh->b_size) {
1880                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1881                         status = -EIO;
1882                         mlog_errno(status);
1883                         goto bail;
1884                 }
1885                 if (de == de_del)  {
1886                         status = ocfs2_journal_access(handle, dir, bh,
1887                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1888                         if (status < 0) {
1889                                 status = -EIO;
1890                                 mlog_errno(status);
1891                                 goto bail;
1892                         }
1893                         if (pde)
1894                                 pde->rec_len =
1895                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1896                                                     le16_to_cpu(de->rec_len));
1897                         else
1898                                 de->inode = 0;
1899                         dir->i_version++;
1900                         status = ocfs2_journal_dirty(handle, bh);
1901                         goto bail;
1902                 }
1903                 i += le16_to_cpu(de->rec_len);
1904                 pde = de;
1905                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1906         }
1907 bail:
1908         mlog_exit(status);
1909         return status;
1910 }
1911
1912 /*
1913  * Returns 0 if not found, -1 on failure, and 1 on success
1914  */
1915 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1916                                         struct inode *dir,
1917                                         const char *name, int namelen,
1918                                         unsigned long offset,
1919                                         struct ocfs2_dir_entry **res_dir)
1920 {
1921         struct ocfs2_dir_entry *de;
1922         char *dlimit, *de_buf;
1923         int de_len;
1924         int ret = 0;
1925
1926         mlog_entry_void();
1927
1928         de_buf = bh->b_data;
1929         dlimit = de_buf + dir->i_sb->s_blocksize;
1930
1931         while (de_buf < dlimit) {
1932                 /* this code is executed quadratically often */
1933                 /* do minimal checking `by hand' */
1934
1935                 de = (struct ocfs2_dir_entry *) de_buf;
1936
1937                 if (de_buf + namelen <= dlimit &&
1938                     ocfs2_match(namelen, name, de)) {
1939                         /* found a match - just to be sure, do a full check */
1940                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1941                                 ret = -1;
1942                                 goto bail;
1943                         }
1944                         *res_dir = de;
1945                         ret = 1;
1946                         goto bail;
1947                 }
1948
1949                 /* prevent looping on a bad block */
1950                 de_len = le16_to_cpu(de->rec_len);
1951                 if (de_len <= 0) {
1952                         ret = -1;
1953                         goto bail;
1954                 }
1955
1956                 de_buf += de_len;
1957                 offset += de_len;
1958         }
1959
1960 bail:
1961         mlog_exit(ret);
1962         return ret;
1963 }
1964
1965 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1966                                      struct inode *dir,
1967                                      struct ocfs2_dir_entry **res_dir)
1968 {
1969         struct super_block *sb;
1970         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1971         struct buffer_head *bh, *ret = NULL;
1972         unsigned long start, block, b;
1973         int ra_max = 0;         /* Number of bh's in the readahead
1974                                    buffer, bh_use[] */
1975         int ra_ptr = 0;         /* Current index into readahead
1976                                    buffer */
1977         int num = 0;
1978         int nblocks, i, err;
1979
1980         mlog_entry_void();
1981
1982         *res_dir = NULL;
1983         sb = dir->i_sb;
1984
1985         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1986         start = OCFS2_I(dir)->ip_dir_start_lookup;
1987         if (start >= nblocks)
1988                 start = 0;
1989         block = start;
1990
1991 restart:
1992         do {
1993                 /*
1994                  * We deal with the read-ahead logic here.
1995                  */
1996                 if (ra_ptr >= ra_max) {
1997                         /* Refill the readahead buffer */
1998                         ra_ptr = 0;
1999                         b = block;
2000                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2001                                 /*
2002                                  * Terminate if we reach the end of the
2003                                  * directory and must wrap, or if our
2004                                  * search has finished at this block.
2005                                  */
2006                                 if (b >= nblocks || (num && block == start)) {
2007                                         bh_use[ra_max] = NULL;
2008                                         break;
2009                                 }
2010                                 num++;
2011
2012                                 bh = ocfs2_bread(dir, b++, &err, 1);
2013                                 bh_use[ra_max] = bh;
2014                         }
2015                 }
2016                 if ((bh = bh_use[ra_ptr++]) == NULL)
2017                         goto next;
2018                 wait_on_buffer(bh);
2019                 if (!buffer_uptodate(bh)) {
2020                         /* read error, skip block & hope for the best */
2021                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2022                                     "offset %lu\n",
2023                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2024                                     block);
2025                         brelse(bh);
2026                         goto next;
2027                 }
2028                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2029                                           block << sb->s_blocksize_bits,
2030                                           res_dir);
2031                 if (i == 1) {
2032                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2033                         ret = bh;
2034                         goto cleanup_and_exit;
2035                 } else {
2036                         brelse(bh);
2037                         if (i < 0)
2038                                 goto cleanup_and_exit;
2039                 }
2040         next:
2041                 if (++block >= nblocks)
2042                         block = 0;
2043         } while (block != start);
2044
2045         /*
2046          * If the directory has grown while we were searching, then
2047          * search the last part of the directory before giving up.
2048          */
2049         block = nblocks;
2050         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2051         if (block < nblocks) {
2052                 start = 0;
2053                 goto restart;
2054         }
2055
2056 cleanup_and_exit:
2057         /* Clean up the read-ahead blocks */
2058         for (; ra_ptr < ra_max; ra_ptr++)
2059                 brelse(bh_use[ra_ptr]);
2060
2061         mlog_exit_ptr(ret);
2062         return ret;
2063 }
2064
2065 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2066 {
2067         int status, namelen;
2068
2069         mlog_entry_void();
2070
2071         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2072                            (long long)blkno);
2073         if (namelen <= 0) {
2074                 if (namelen)
2075                         status = namelen;
2076                 else
2077                         status = -EINVAL;
2078                 mlog_errno(status);
2079                 goto bail;
2080         }
2081         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2082                 status = -EINVAL;
2083                 mlog_errno(status);
2084                 goto bail;
2085         }
2086
2087         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2088              namelen);
2089
2090         status = 0;
2091 bail:
2092         mlog_exit(status);
2093         return status;
2094 }
2095
2096 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2097                                     struct ocfs2_journal_handle *handle,
2098                                     struct inode *inode,
2099                                     char *name,
2100                                     struct buffer_head **de_bh)
2101 {
2102         struct inode *orphan_dir_inode = NULL;
2103         struct buffer_head *orphan_dir_bh = NULL;
2104         int status = 0;
2105
2106         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2107         if (status < 0) {
2108                 mlog_errno(status);
2109                 goto leave;
2110         }
2111
2112         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2113                                                        ORPHAN_DIR_SYSTEM_INODE,
2114                                                        osb->slot_num);
2115         if (!orphan_dir_inode) {
2116                 status = -ENOENT;
2117                 mlog_errno(status);
2118                 goto leave;
2119         }
2120
2121         ocfs2_handle_add_inode(handle, orphan_dir_inode);
2122         status = ocfs2_meta_lock(orphan_dir_inode, handle, &orphan_dir_bh, 1);
2123         if (status < 0) {
2124                 mlog_errno(status);
2125                 goto leave;
2126         }
2127
2128         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2129                                               orphan_dir_bh, name,
2130                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2131         if (status < 0) {
2132                 mlog_errno(status);
2133                 goto leave;
2134         }
2135
2136 leave:
2137         if (orphan_dir_inode)
2138                 iput(orphan_dir_inode);
2139
2140         if (orphan_dir_bh)
2141                 brelse(orphan_dir_bh);
2142
2143         mlog_exit(status);
2144         return status;
2145 }
2146
2147 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2148                             struct ocfs2_journal_handle *handle,
2149                             struct inode *inode,
2150                             struct ocfs2_dinode *fe,
2151                             char *name,
2152                             struct buffer_head *de_bh)
2153 {
2154         struct inode *orphan_dir_inode = NULL;
2155         struct buffer_head *orphan_dir_bh = NULL;
2156         int status = 0;
2157         struct ocfs2_dinode *orphan_fe;
2158
2159         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2160
2161         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2162                                                        ORPHAN_DIR_SYSTEM_INODE,
2163                                                        osb->slot_num);
2164         if (!orphan_dir_inode) {
2165                 status = -ENOENT;
2166                 mlog_errno(status);
2167                 goto leave;
2168         }
2169
2170         status = ocfs2_read_block(osb,
2171                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2172                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2173                                   orphan_dir_inode);
2174         if (status < 0) {
2175                 mlog_errno(status);
2176                 goto leave;
2177         }
2178
2179         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_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                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2191         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2192
2193         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2194         if (status < 0) {
2195                 mlog_errno(status);
2196                 goto leave;
2197         }
2198
2199         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2200                                    OCFS2_ORPHAN_NAMELEN, inode,
2201                                    OCFS2_I(inode)->ip_blkno,
2202                                    orphan_dir_bh, de_bh);
2203         if (status < 0) {
2204                 mlog_errno(status);
2205                 goto leave;
2206         }
2207
2208         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2209
2210         /* Record which orphan dir our inode now resides
2211          * in. delete_inode will use this to determine which orphan
2212          * dir to lock. */
2213         spin_lock(&OCFS2_I(inode)->ip_lock);
2214         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2215         spin_unlock(&OCFS2_I(inode)->ip_lock);
2216
2217         mlog(0, "Inode %llu orphaned in slot %d\n",
2218              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2219
2220 leave:
2221         if (orphan_dir_inode)
2222                 iput(orphan_dir_inode);
2223
2224         if (orphan_dir_bh)
2225                 brelse(orphan_dir_bh);
2226
2227         mlog_exit(status);
2228         return status;
2229 }
2230
2231 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2232 int ocfs2_orphan_del(struct ocfs2_super *osb,
2233                      struct ocfs2_journal_handle *handle,
2234                      struct inode *orphan_dir_inode,
2235                      struct inode *inode,
2236                      struct buffer_head *orphan_dir_bh)
2237 {
2238         char name[OCFS2_ORPHAN_NAMELEN + 1];
2239         struct ocfs2_dinode *orphan_fe;
2240         int status = 0;
2241         struct buffer_head *target_de_bh = NULL;
2242         struct ocfs2_dir_entry *target_de = NULL;
2243
2244         mlog_entry_void();
2245
2246         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2247         if (status < 0) {
2248                 mlog_errno(status);
2249                 goto leave;
2250         }
2251
2252         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2253              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2254              OCFS2_ORPHAN_NAMELEN);
2255
2256         /* find it's spot in the orphan directory */
2257         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2258                                         orphan_dir_inode, &target_de);
2259         if (!target_de_bh) {
2260                 status = -ENOENT;
2261                 mlog_errno(status);
2262                 goto leave;
2263         }
2264
2265         /* remove it from the orphan directory */
2266         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2267                                     target_de_bh);
2268         if (status < 0) {
2269                 mlog_errno(status);
2270                 goto leave;
2271         }
2272
2273         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2274                                       OCFS2_JOURNAL_ACCESS_WRITE);
2275         if (status < 0) {
2276                 mlog_errno(status);
2277                 goto leave;
2278         }
2279
2280         /* do the i_nlink dance! :) */
2281         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2282         if (S_ISDIR(inode->i_mode))
2283                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2284         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2285
2286         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2287         if (status < 0) {
2288                 mlog_errno(status);
2289                 goto leave;
2290         }
2291
2292 leave:
2293         if (target_de_bh)
2294                 brelse(target_de_bh);
2295
2296         mlog_exit(status);
2297         return status;
2298 }
2299
2300 struct inode_operations ocfs2_dir_iops = {
2301         .create         = ocfs2_create,
2302         .lookup         = ocfs2_lookup,
2303         .link           = ocfs2_link,
2304         .unlink         = ocfs2_unlink,
2305         .rmdir          = ocfs2_unlink,
2306         .symlink        = ocfs2_symlink,
2307         .mkdir          = ocfs2_mkdir,
2308         .mknod          = ocfs2_mknod,
2309         .rename         = ocfs2_rename,
2310         .setattr        = ocfs2_setattr,
2311         .getattr        = ocfs2_getattr,
2312 };