f64cff0ceed8f9be4d28dd66aa718a55d0143f12
[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 inode **ret_orphan_dir,
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                             struct inode *orphan_dir_inode);
127
128 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
129                                      struct ocfs2_journal_handle *handle,
130                                      struct inode *inode,
131                                      const char *symname);
132
133 static inline int ocfs2_add_entry(struct ocfs2_journal_handle *handle,
134                                   struct dentry *dentry,
135                                   struct inode *inode, u64 blkno,
136                                   struct buffer_head *parent_fe_bh,
137                                   struct buffer_head *insert_bh)
138 {
139         return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
140                                  dentry->d_name.name, dentry->d_name.len,
141                                  inode, blkno, parent_fe_bh, insert_bh);
142 }
143
144 /* An orphan dir name is an 8 byte value, printed as a hex string */
145 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
146
147 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
148                                    struct nameidata *nd)
149 {
150         int status;
151         u64 blkno;
152         struct buffer_head *dirent_bh = NULL;
153         struct inode *inode = NULL;
154         struct dentry *ret;
155         struct ocfs2_dir_entry *dirent;
156         struct ocfs2_inode_info *oi;
157
158         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
159                    dentry->d_name.len, dentry->d_name.name);
160
161         if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
162                 ret = ERR_PTR(-ENAMETOOLONG);
163                 goto bail;
164         }
165
166         mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
167              dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
168
169         status = ocfs2_meta_lock(dir, NULL, NULL, 0);
170         if (status < 0) {
171                 if (status != -ENOENT)
172                         mlog_errno(status);
173                 ret = ERR_PTR(status);
174                 goto bail;
175         }
176
177         status = ocfs2_find_files_on_disk(dentry->d_name.name,
178                                           dentry->d_name.len, &blkno,
179                                           dir, &dirent_bh, &dirent);
180         if (status < 0)
181                 goto bail_add;
182
183         inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
184         if (IS_ERR(inode)) {
185                 mlog(ML_ERROR, "Unable to create inode %llu\n",
186                      (unsigned long long)blkno);
187                 ret = ERR_PTR(-EACCES);
188                 goto bail_unlock;
189         }
190
191         oi = OCFS2_I(inode);
192         /* Clear any orphaned state... If we were able to look up the
193          * inode from a directory, it certainly can't be orphaned. We
194          * might have the bad state from a node which intended to
195          * orphan this inode but crashed before it could commit the
196          * unlink. */
197         spin_lock(&oi->ip_lock);
198         oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
199         oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
200         spin_unlock(&oi->ip_lock);
201
202 bail_add:
203         dentry->d_op = &ocfs2_dentry_ops;
204         ret = d_splice_alias(inode, dentry);
205
206         if (inode) {
207                 /*
208                  * If d_splice_alias() finds a DCACHE_DISCONNECTED
209                  * dentry, it will d_move() it on top of ourse. The
210                  * return value will indicate this however, so in
211                  * those cases, we switch them around for the locking
212                  * code.
213                  *
214                  * NOTE: This dentry already has ->d_op set from
215                  * ocfs2_get_parent() and ocfs2_get_dentry()
216                  */
217                 if (ret)
218                         dentry = ret;
219
220                 status = ocfs2_dentry_attach_lock(dentry, inode,
221                                                   OCFS2_I(dir)->ip_blkno);
222                 if (status) {
223                         mlog_errno(status);
224                         ret = ERR_PTR(status);
225                         goto bail_unlock;
226                 }
227         }
228
229 bail_unlock:
230         /* Don't drop the cluster lock until *after* the d_add --
231          * unlink on another node will message us to remove that
232          * dentry under this lock so otherwise we can race this with
233          * the vote thread and have a stale dentry. */
234         ocfs2_meta_unlock(dir, 0);
235
236 bail:
237         if (dirent_bh)
238                 brelse(dirent_bh);
239
240         mlog_exit_ptr(ret);
241
242         return ret;
243 }
244
245 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
246                               struct ocfs2_journal_handle *handle,
247                               struct inode *parent,
248                               struct inode *inode,
249                               struct buffer_head *fe_bh,
250                               struct ocfs2_alloc_context *data_ac)
251 {
252         int status;
253         struct buffer_head *new_bh = NULL;
254         struct ocfs2_dir_entry *de = NULL;
255
256         mlog_entry_void();
257
258         status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
259                                      data_ac, NULL, &new_bh);
260         if (status < 0) {
261                 mlog_errno(status);
262                 goto bail;
263         }
264
265         ocfs2_set_new_buffer_uptodate(inode, new_bh);
266
267         status = ocfs2_journal_access(handle, inode, new_bh,
268                                       OCFS2_JOURNAL_ACCESS_CREATE);
269         if (status < 0) {
270                 mlog_errno(status);
271                 goto bail;
272         }
273         memset(new_bh->b_data, 0, osb->sb->s_blocksize);
274
275         de = (struct ocfs2_dir_entry *) new_bh->b_data;
276         de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
277         de->name_len = 1;
278         de->rec_len =
279                 cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
280         strcpy(de->name, ".");
281         ocfs2_set_de_type(de, S_IFDIR);
282         de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
283         de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
284         de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
285                                   OCFS2_DIR_REC_LEN(1));
286         de->name_len = 2;
287         strcpy(de->name, "..");
288         ocfs2_set_de_type(de, S_IFDIR);
289
290         status = ocfs2_journal_dirty(handle, new_bh);
291         if (status < 0) {
292                 mlog_errno(status);
293                 goto bail;
294         }
295
296         i_size_write(inode, inode->i_sb->s_blocksize);
297         inode->i_nlink = 2;
298         inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
299         status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
300         if (status < 0) {
301                 mlog_errno(status);
302                 goto bail;
303         }
304
305         status = 0;
306 bail:
307         if (new_bh)
308                 brelse(new_bh);
309
310         mlog_exit(status);
311         return status;
312 }
313
314 static int ocfs2_mknod(struct inode *dir,
315                        struct dentry *dentry,
316                        int mode,
317                        dev_t dev)
318 {
319         int status = 0;
320         struct buffer_head *parent_fe_bh = NULL;
321         struct ocfs2_journal_handle *handle = NULL;
322         struct ocfs2_super *osb;
323         struct ocfs2_dinode *dirfe;
324         struct buffer_head *new_fe_bh = NULL;
325         struct buffer_head *de_bh = NULL;
326         struct inode *inode = NULL;
327         struct ocfs2_alloc_context *inode_ac = NULL;
328         struct ocfs2_alloc_context *data_ac = NULL;
329
330         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
331                    (unsigned long)dev, dentry->d_name.len,
332                    dentry->d_name.name);
333
334         /* get our super block */
335         osb = OCFS2_SB(dir->i_sb);
336
337         status = ocfs2_meta_lock(dir, NULL, &parent_fe_bh, 1);
338         if (status < 0) {
339                 if (status != -ENOENT)
340                         mlog_errno(status);
341                 return status;
342         }
343
344         handle = ocfs2_alloc_handle(osb);
345         if (handle == NULL) {
346                 status = -ENOMEM;
347                 mlog_errno(status);
348                 goto leave;
349         }
350
351         if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
352                 status = -EMLINK;
353                 goto leave;
354         }
355
356         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
357         if (!dirfe->i_links_count) {
358                 /* can't make a file in a deleted directory. */
359                 status = -ENOENT;
360                 goto leave;
361         }
362
363         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
364                                            dentry->d_name.len);
365         if (status)
366                 goto leave;
367
368         /* get a spot inside the dir. */
369         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
370                                               dentry->d_name.name,
371                                               dentry->d_name.len, &de_bh);
372         if (status < 0) {
373                 mlog_errno(status);
374                 goto leave;
375         }
376
377         /* reserve an inode spot */
378         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
379         if (status < 0) {
380                 if (status != -ENOSPC)
381                         mlog_errno(status);
382                 goto leave;
383         }
384
385         /* are we making a directory? If so, reserve a cluster for his
386          * 1st extent. */
387         if (S_ISDIR(mode)) {
388                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
389                 if (status < 0) {
390                         if (status != -ENOSPC)
391                                 mlog_errno(status);
392                         goto leave;
393                 }
394         }
395
396         handle = ocfs2_start_trans(osb, handle, OCFS2_MKNOD_CREDITS);
397         if (IS_ERR(handle)) {
398                 status = PTR_ERR(handle);
399                 handle = NULL;
400                 mlog_errno(status);
401                 goto leave;
402         }
403
404         /* do the real work now. */
405         status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
406                                     &new_fe_bh, parent_fe_bh, handle,
407                                     &inode, inode_ac);
408         if (status < 0) {
409                 mlog_errno(status);
410                 goto leave;
411         }
412
413         if (S_ISDIR(mode)) {
414                 status = ocfs2_fill_new_dir(osb, handle, dir, inode,
415                                             new_fe_bh, data_ac);
416                 if (status < 0) {
417                         mlog_errno(status);
418                         goto leave;
419                 }
420
421                 status = ocfs2_journal_access(handle, dir, parent_fe_bh,
422                                               OCFS2_JOURNAL_ACCESS_WRITE);
423                 if (status < 0) {
424                         mlog_errno(status);
425                         goto leave;
426                 }
427                 le16_add_cpu(&dirfe->i_links_count, 1);
428                 status = ocfs2_journal_dirty(handle, parent_fe_bh);
429                 if (status < 0) {
430                         mlog_errno(status);
431                         goto leave;
432                 }
433                 inc_nlink(dir);
434         }
435
436         status = ocfs2_add_entry(handle, dentry, inode,
437                                  OCFS2_I(inode)->ip_blkno, parent_fe_bh,
438                                  de_bh);
439         if (status < 0) {
440                 mlog_errno(status);
441                 goto leave;
442         }
443
444         status = ocfs2_dentry_attach_lock(dentry, inode,
445                                           OCFS2_I(dir)->ip_blkno);
446         if (status) {
447                 mlog_errno(status);
448                 goto leave;
449         }
450
451         insert_inode_hash(inode);
452         dentry->d_op = &ocfs2_dentry_ops;
453         d_instantiate(dentry, inode);
454         status = 0;
455 leave:
456         if (handle)
457                 ocfs2_commit_trans(handle);
458
459         ocfs2_meta_unlock(dir, 1);
460
461         if (status == -ENOSPC)
462                 mlog(0, "Disk is full\n");
463
464         if (new_fe_bh)
465                 brelse(new_fe_bh);
466
467         if (de_bh)
468                 brelse(de_bh);
469
470         if (parent_fe_bh)
471                 brelse(parent_fe_bh);
472
473         if ((status < 0) && inode)
474                 iput(inode);
475
476         if (inode_ac)
477                 ocfs2_free_alloc_context(inode_ac);
478
479         if (data_ac)
480                 ocfs2_free_alloc_context(data_ac);
481
482         mlog_exit(status);
483
484         return status;
485 }
486
487 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
488                               struct inode *dir,
489                               struct dentry *dentry, int mode,
490                               dev_t dev,
491                               struct buffer_head **new_fe_bh,
492                               struct buffer_head *parent_fe_bh,
493                               struct ocfs2_journal_handle *handle,
494                               struct inode **ret_inode,
495                               struct ocfs2_alloc_context *inode_ac)
496 {
497         int status = 0;
498         struct ocfs2_dinode *fe = NULL;
499         struct ocfs2_extent_list *fel;
500         u64 fe_blkno = 0;
501         u16 suballoc_bit;
502         struct inode *inode = NULL;
503
504         mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
505                    (unsigned long)dev, dentry->d_name.len,
506                    dentry->d_name.name);
507
508         *new_fe_bh = NULL;
509         *ret_inode = NULL;
510
511         status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
512                                        &fe_blkno);
513         if (status < 0) {
514                 mlog_errno(status);
515                 goto leave;
516         }
517
518         inode = new_inode(dir->i_sb);
519         if (IS_ERR(inode)) {
520                 status = PTR_ERR(inode);
521                 mlog(ML_ERROR, "new_inode failed!\n");
522                 goto leave;
523         }
524
525         /* populate as many fields early on as possible - many of
526          * these are used by the support functions here and in
527          * callers. */
528         inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
529         OCFS2_I(inode)->ip_blkno = fe_blkno;
530         if (S_ISDIR(mode))
531                 inode->i_nlink = 2;
532         else
533                 inode->i_nlink = 1;
534         inode->i_mode = mode;
535         spin_lock(&osb->osb_lock);
536         inode->i_generation = osb->s_next_generation++;
537         spin_unlock(&osb->osb_lock);
538
539         *new_fe_bh = sb_getblk(osb->sb, fe_blkno);
540         if (!*new_fe_bh) {
541                 status = -EIO;
542                 mlog_errno(status);
543                 goto leave;
544         }
545         ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
546
547         status = ocfs2_journal_access(handle, inode, *new_fe_bh,
548                                       OCFS2_JOURNAL_ACCESS_CREATE);
549         if (status < 0) {
550                 mlog_errno(status);
551                 goto leave;
552         }
553
554         fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
555         memset(fe, 0, osb->sb->s_blocksize);
556
557         fe->i_generation = cpu_to_le32(inode->i_generation);
558         fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
559         fe->i_blkno = cpu_to_le64(fe_blkno);
560         fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
561         fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
562         fe->i_uid = cpu_to_le32(current->fsuid);
563         if (dir->i_mode & S_ISGID) {
564                 fe->i_gid = cpu_to_le32(dir->i_gid);
565                 if (S_ISDIR(mode))
566                         mode |= S_ISGID;
567         } else
568                 fe->i_gid = cpu_to_le32(current->fsgid);
569         fe->i_mode = cpu_to_le16(mode);
570         if (S_ISCHR(mode) || S_ISBLK(mode))
571                 fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
572
573         fe->i_links_count = cpu_to_le16(inode->i_nlink);
574
575         fe->i_last_eb_blk = 0;
576         strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
577         le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
578         fe->i_atime = fe->i_ctime = fe->i_mtime =
579                 cpu_to_le64(CURRENT_TIME.tv_sec);
580         fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
581                 cpu_to_le32(CURRENT_TIME.tv_nsec);
582         fe->i_dtime = 0;
583
584         fel = &fe->id2.i_list;
585         fel->l_tree_depth = 0;
586         fel->l_next_free_rec = 0;
587         fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
588
589         status = ocfs2_journal_dirty(handle, *new_fe_bh);
590         if (status < 0) {
591                 mlog_errno(status);
592                 goto leave;
593         }
594
595         if (ocfs2_populate_inode(inode, fe, 1) < 0) {
596                 mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
597                      "i_blkno=%llu, i_ino=%lu\n",
598                      (unsigned long long) (*new_fe_bh)->b_blocknr,
599                      (unsigned long long)fe->i_blkno, inode->i_ino);
600                 BUG();
601         }
602
603         ocfs2_inode_set_new(osb, inode);
604         status = ocfs2_create_new_inode_locks(inode);
605         if (status < 0)
606                 mlog_errno(status);
607
608         status = 0; /* error in ocfs2_create_new_inode_locks is not
609                      * critical */
610
611         *ret_inode = inode;
612 leave:
613         if (status < 0) {
614                 if (*new_fe_bh) {
615                         brelse(*new_fe_bh);
616                         *new_fe_bh = NULL;
617                 }
618                 if (inode)
619                         iput(inode);
620         }
621
622         mlog_exit(status);
623         return status;
624 }
625
626 static int ocfs2_mkdir(struct inode *dir,
627                        struct dentry *dentry,
628                        int mode)
629 {
630         int ret;
631
632         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
633                    dentry->d_name.len, dentry->d_name.name);
634         ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
635         mlog_exit(ret);
636
637         return ret;
638 }
639
640 static int ocfs2_create(struct inode *dir,
641                         struct dentry *dentry,
642                         int mode,
643                         struct nameidata *nd)
644 {
645         int ret;
646
647         mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
648                    dentry->d_name.len, dentry->d_name.name);
649         ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
650         mlog_exit(ret);
651
652         return ret;
653 }
654
655 static int ocfs2_link(struct dentry *old_dentry,
656                       struct inode *dir,
657                       struct dentry *dentry)
658 {
659         struct ocfs2_journal_handle *handle;
660         struct inode *inode = old_dentry->d_inode;
661         int err;
662         struct buffer_head *fe_bh = NULL;
663         struct buffer_head *parent_fe_bh = NULL;
664         struct buffer_head *de_bh = NULL;
665         struct ocfs2_dinode *fe = NULL;
666         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
667
668         mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
669                    old_dentry->d_name.len, old_dentry->d_name.name,
670                    dentry->d_name.len, dentry->d_name.name);
671
672         if (S_ISDIR(inode->i_mode))
673                 return -EPERM;
674
675         err = ocfs2_meta_lock(dir, NULL, &parent_fe_bh, 1);
676         if (err < 0) {
677                 if (err != -ENOENT)
678                         mlog_errno(err);
679                 return err;
680         }
681
682         if (!dir->i_nlink) {
683                 err = -ENOENT;
684                 goto out;
685         }
686
687         err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
688                                         dentry->d_name.len);
689         if (err)
690                 goto out;
691
692         err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
693                                            dentry->d_name.name,
694                                            dentry->d_name.len, &de_bh);
695         if (err < 0) {
696                 mlog_errno(err);
697                 goto out;
698         }
699
700         err = ocfs2_meta_lock(inode, NULL, &fe_bh, 1);
701         if (err < 0) {
702                 if (err != -ENOENT)
703                         mlog_errno(err);
704                 goto out;
705         }
706
707         fe = (struct ocfs2_dinode *) fe_bh->b_data;
708         if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
709                 err = -EMLINK;
710                 goto out_unlock_inode;
711         }
712
713         handle = ocfs2_start_trans(osb, NULL, OCFS2_LINK_CREDITS);
714         if (IS_ERR(handle)) {
715                 err = PTR_ERR(handle);
716                 handle = NULL;
717                 mlog_errno(err);
718                 goto out_unlock_inode;
719         }
720
721         err = ocfs2_journal_access(handle, inode, fe_bh,
722                                    OCFS2_JOURNAL_ACCESS_WRITE);
723         if (err < 0) {
724                 mlog_errno(err);
725                 goto out_commit;
726         }
727
728         inc_nlink(inode);
729         inode->i_ctime = CURRENT_TIME;
730         fe->i_links_count = cpu_to_le16(inode->i_nlink);
731         fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
732         fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
733
734         err = ocfs2_journal_dirty(handle, fe_bh);
735         if (err < 0) {
736                 le16_add_cpu(&fe->i_links_count, -1);
737                 drop_nlink(inode);
738                 mlog_errno(err);
739                 goto out_commit;
740         }
741
742         err = ocfs2_add_entry(handle, dentry, inode,
743                               OCFS2_I(inode)->ip_blkno,
744                               parent_fe_bh, de_bh);
745         if (err) {
746                 le16_add_cpu(&fe->i_links_count, -1);
747                 drop_nlink(inode);
748                 mlog_errno(err);
749                 goto out_commit;
750         }
751
752         err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
753         if (err) {
754                 mlog_errno(err);
755                 goto out_commit;
756         }
757
758         atomic_inc(&inode->i_count);
759         dentry->d_op = &ocfs2_dentry_ops;
760         d_instantiate(dentry, inode);
761
762 out_commit:
763         ocfs2_commit_trans(handle);
764 out_unlock_inode:
765         ocfs2_meta_unlock(inode, 1);
766
767 out:
768         ocfs2_meta_unlock(dir, 1);
769
770         if (de_bh)
771                 brelse(de_bh);
772         if (fe_bh)
773                 brelse(fe_bh);
774         if (parent_fe_bh)
775                 brelse(parent_fe_bh);
776
777         mlog_exit(err);
778
779         return err;
780 }
781
782 /*
783  * Takes and drops an exclusive lock on the given dentry. This will
784  * force other nodes to drop it.
785  */
786 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
787 {
788         int ret;
789
790         ret = ocfs2_dentry_lock(dentry, 1);
791         if (ret)
792                 mlog_errno(ret);
793         else
794                 ocfs2_dentry_unlock(dentry, 1);
795
796         return ret;
797 }
798
799 static inline int inode_is_unlinkable(struct inode *inode)
800 {
801         if (S_ISDIR(inode->i_mode)) {
802                 if (inode->i_nlink == 2)
803                         return 1;
804                 return 0;
805         }
806
807         if (inode->i_nlink == 1)
808                 return 1;
809         return 0;
810 }
811
812 static int ocfs2_unlink(struct inode *dir,
813                         struct dentry *dentry)
814 {
815         int status;
816         int child_locked = 0;
817         struct inode *inode = dentry->d_inode;
818         struct inode *orphan_dir = NULL;
819         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
820         u64 blkno;
821         struct ocfs2_dinode *fe = NULL;
822         struct buffer_head *fe_bh = NULL;
823         struct buffer_head *parent_node_bh = NULL;
824         struct ocfs2_journal_handle *handle = NULL;
825         struct ocfs2_dir_entry *dirent = NULL;
826         struct buffer_head *dirent_bh = NULL;
827         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
828         struct buffer_head *orphan_entry_bh = NULL;
829
830         mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
831                    dentry->d_name.len, dentry->d_name.name);
832
833         BUG_ON(dentry->d_parent->d_inode != dir);
834
835         mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
836
837         if (inode == osb->root_inode) {
838                 mlog(0, "Cannot delete the root directory\n");
839                 return -EPERM;
840         }
841
842         status = ocfs2_meta_lock(dir, NULL, &parent_node_bh, 1);
843         if (status < 0) {
844                 if (status != -ENOENT)
845                         mlog_errno(status);
846                 return status;
847         }
848
849         status = ocfs2_find_files_on_disk(dentry->d_name.name,
850                                           dentry->d_name.len, &blkno,
851                                           dir, &dirent_bh, &dirent);
852         if (status < 0) {
853                 if (status != -ENOENT)
854                         mlog_errno(status);
855                 goto leave;
856         }
857
858         if (OCFS2_I(inode)->ip_blkno != blkno) {
859                 status = -ENOENT;
860
861                 mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
862                      (unsigned long long)OCFS2_I(inode)->ip_blkno,
863                      (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
864                 goto leave;
865         }
866
867         status = ocfs2_meta_lock(inode, NULL, &fe_bh, 1);
868         if (status < 0) {
869                 if (status != -ENOENT)
870                         mlog_errno(status);
871                 goto leave;
872         }
873         child_locked = 1;
874
875         if (S_ISDIR(inode->i_mode)) {
876                 if (!ocfs2_empty_dir(inode)) {
877                         status = -ENOTEMPTY;
878                         goto leave;
879                 } else if (inode->i_nlink != 2) {
880                         status = -ENOTEMPTY;
881                         goto leave;
882                 }
883         }
884
885         status = ocfs2_remote_dentry_delete(dentry);
886         if (status < 0) {
887                 /* This vote should succeed under all normal
888                  * circumstances. */
889                 mlog_errno(status);
890                 goto leave;
891         }
892
893         if (inode_is_unlinkable(inode)) {
894                 status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
895                                                   orphan_name,
896                                                   &orphan_entry_bh);
897                 if (status < 0) {
898                         mlog_errno(status);
899                         goto leave;
900                 }
901         }
902
903         handle = ocfs2_start_trans(osb, NULL, OCFS2_UNLINK_CREDITS);
904         if (IS_ERR(handle)) {
905                 status = PTR_ERR(handle);
906                 handle = NULL;
907                 mlog_errno(status);
908                 goto leave;
909         }
910
911         status = ocfs2_journal_access(handle, inode, fe_bh,
912                                       OCFS2_JOURNAL_ACCESS_WRITE);
913         if (status < 0) {
914                 mlog_errno(status);
915                 goto leave;
916         }
917
918         fe = (struct ocfs2_dinode *) fe_bh->b_data;
919
920         if (inode_is_unlinkable(inode)) {
921                 status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
922                                           orphan_entry_bh, orphan_dir);
923                 if (status < 0) {
924                         mlog_errno(status);
925                         goto leave;
926                 }
927         }
928
929         /* delete the name from the parent dir */
930         status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
931         if (status < 0) {
932                 mlog_errno(status);
933                 goto leave;
934         }
935
936         if (S_ISDIR(inode->i_mode))
937                 drop_nlink(inode);
938         drop_nlink(inode);
939         fe->i_links_count = cpu_to_le16(inode->i_nlink);
940
941         status = ocfs2_journal_dirty(handle, fe_bh);
942         if (status < 0) {
943                 mlog_errno(status);
944                 goto leave;
945         }
946
947         if (S_ISDIR(inode->i_mode)) {
948                 drop_nlink(dir);
949                 status = ocfs2_mark_inode_dirty(handle, dir,
950                                                 parent_node_bh);
951                 if (status < 0) {
952                         mlog_errno(status);
953                         inc_nlink(dir);
954                 }
955         }
956
957 leave:
958         if (handle)
959                 ocfs2_commit_trans(handle);
960
961         if (child_locked)
962                 ocfs2_meta_unlock(inode, 1);
963
964         ocfs2_meta_unlock(dir, 1);
965
966         if (orphan_dir) {
967                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
968                 ocfs2_meta_unlock(orphan_dir, 1);
969                 mutex_unlock(&orphan_dir->i_mutex);
970                 iput(orphan_dir);
971         }
972
973         if (fe_bh)
974                 brelse(fe_bh);
975
976         if (dirent_bh)
977                 brelse(dirent_bh);
978
979         if (parent_node_bh)
980                 brelse(parent_node_bh);
981
982         if (orphan_entry_bh)
983                 brelse(orphan_entry_bh);
984
985         mlog_exit(status);
986
987         return status;
988 }
989
990 /*
991  * The only place this should be used is rename!
992  * if they have the same id, then the 1st one is the only one locked.
993  */
994 static int ocfs2_double_lock(struct ocfs2_super *osb,
995                              struct ocfs2_journal_handle *handle,
996                              struct buffer_head **bh1,
997                              struct inode *inode1,
998                              struct buffer_head **bh2,
999                              struct inode *inode2)
1000 {
1001         int status;
1002         struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1003         struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1004         struct buffer_head **tmpbh;
1005         struct inode *tmpinode;
1006
1007         mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
1008                    (unsigned long long)oi1->ip_blkno,
1009                    (unsigned long long)oi2->ip_blkno);
1010
1011         BUG_ON(!handle);
1012
1013         if (*bh1)
1014                 *bh1 = NULL;
1015         if (*bh2)
1016                 *bh2 = NULL;
1017
1018         /* we always want to lock the one with the lower lockid first. */
1019         if (oi1->ip_blkno != oi2->ip_blkno) {
1020                 if (oi1->ip_blkno < oi2->ip_blkno) {
1021                         /* switch id1 and id2 around */
1022                         mlog(0, "switching them around...\n");
1023                         tmpbh = bh2;
1024                         bh2 = bh1;
1025                         bh1 = tmpbh;
1026
1027                         tmpinode = inode2;
1028                         inode2 = inode1;
1029                         inode1 = tmpinode;
1030                 }
1031                 /* lock id2 */
1032                 status = ocfs2_meta_lock(inode2, handle, bh2, 1);
1033                 if (status < 0) {
1034                         if (status != -ENOENT)
1035                                 mlog_errno(status);
1036                         goto bail;
1037                 }
1038         }
1039         /* lock id1 */
1040         status = ocfs2_meta_lock(inode1, handle, bh1, 1);
1041         if (status < 0) {
1042                 if (status != -ENOENT)
1043                         mlog_errno(status);
1044                 goto bail;
1045         }
1046 bail:
1047         mlog_exit(status);
1048         return status;
1049 }
1050
1051 #define PARENT_INO(buffer) \
1052         ((struct ocfs2_dir_entry *) \
1053          ((char *)buffer + \
1054           le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1055
1056 static int ocfs2_rename(struct inode *old_dir,
1057                         struct dentry *old_dentry,
1058                         struct inode *new_dir,
1059                         struct dentry *new_dentry)
1060 {
1061         int status = 0, rename_lock = 0;
1062         struct inode *old_inode = old_dentry->d_inode;
1063         struct inode *new_inode = new_dentry->d_inode;
1064         struct inode *orphan_dir = NULL;
1065         struct ocfs2_dinode *newfe = NULL;
1066         char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1067         struct buffer_head *orphan_entry_bh = NULL;
1068         struct buffer_head *newfe_bh = NULL;
1069         struct buffer_head *insert_entry_bh = NULL;
1070         struct ocfs2_super *osb = NULL;
1071         u64 newfe_blkno;
1072         struct ocfs2_journal_handle *handle = NULL;
1073         struct buffer_head *old_dir_bh = NULL;
1074         struct buffer_head *new_dir_bh = NULL;
1075         struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1076                                                                // and new_dentry
1077         struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1078         struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1079                                                     // this is the 1st dirent bh
1080         nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1081
1082         /* At some point it might be nice to break this function up a
1083          * bit. */
1084
1085         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1086                    old_dir, old_dentry, new_dir, new_dentry,
1087                    old_dentry->d_name.len, old_dentry->d_name.name,
1088                    new_dentry->d_name.len, new_dentry->d_name.name);
1089
1090         osb = OCFS2_SB(old_dir->i_sb);
1091
1092         if (new_inode) {
1093                 if (!igrab(new_inode))
1094                         BUG();
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, &orphan_dir,
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, orphan_dir);
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                         inc_nlink(new_dir);
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 (orphan_dir) {
1439                 /* This was locked for us in ocfs2_prepare_orphan_dir() */
1440                 ocfs2_meta_unlock(orphan_dir, 1);
1441                 mutex_unlock(&orphan_dir->i_mutex);
1442                 iput(orphan_dir);
1443         }
1444
1445         if (new_inode)
1446                 sync_mapping_buffers(old_inode->i_mapping);
1447
1448         if (new_inode)
1449                 iput(new_inode);
1450         if (newfe_bh)
1451                 brelse(newfe_bh);
1452         if (old_dir_bh)
1453                 brelse(old_dir_bh);
1454         if (new_dir_bh)
1455                 brelse(new_dir_bh);
1456         if (new_de_bh)
1457                 brelse(new_de_bh);
1458         if (old_de_bh)
1459                 brelse(old_de_bh);
1460         if (old_inode_de_bh)
1461                 brelse(old_inode_de_bh);
1462         if (orphan_entry_bh)
1463                 brelse(orphan_entry_bh);
1464         if (insert_entry_bh)
1465                 brelse(insert_entry_bh);
1466
1467         mlog_exit(status);
1468
1469         return status;
1470 }
1471
1472 /*
1473  * we expect i_size = strlen(symname). Copy symname into the file
1474  * data, including the null terminator.
1475  */
1476 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1477                                      struct ocfs2_journal_handle *handle,
1478                                      struct inode *inode,
1479                                      const char *symname)
1480 {
1481         struct buffer_head **bhs = NULL;
1482         const char *c;
1483         struct super_block *sb = osb->sb;
1484         u64 p_blkno;
1485         int p_blocks;
1486         int virtual, blocks, status, i, bytes_left;
1487
1488         bytes_left = i_size_read(inode) + 1;
1489         /* we can't trust i_blocks because we're actually going to
1490          * write i_size + 1 bytes. */
1491         blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1492
1493         mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1494                         (unsigned long long)inode->i_blocks,
1495                         i_size_read(inode), blocks);
1496
1497         /* Sanity check -- make sure we're going to fit. */
1498         if (bytes_left >
1499             ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1500                 status = -EIO;
1501                 mlog_errno(status);
1502                 goto bail;
1503         }
1504
1505         bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1506         if (!bhs) {
1507                 status = -ENOMEM;
1508                 mlog_errno(status);
1509                 goto bail;
1510         }
1511
1512         status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1513                                              &p_blocks);
1514         if (status < 0) {
1515                 mlog_errno(status);
1516                 goto bail;
1517         }
1518
1519         /* links can never be larger than one cluster so we know this
1520          * is all going to be contiguous, but do a sanity check
1521          * anyway. */
1522         if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1523                 status = -EIO;
1524                 mlog_errno(status);
1525                 goto bail;
1526         }
1527
1528         virtual = 0;
1529         while(bytes_left > 0) {
1530                 c = &symname[virtual * sb->s_blocksize];
1531
1532                 bhs[virtual] = sb_getblk(sb, p_blkno);
1533                 if (!bhs[virtual]) {
1534                         status = -ENOMEM;
1535                         mlog_errno(status);
1536                         goto bail;
1537                 }
1538                 ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1539
1540                 status = ocfs2_journal_access(handle, inode, bhs[virtual],
1541                                               OCFS2_JOURNAL_ACCESS_CREATE);
1542                 if (status < 0) {
1543                         mlog_errno(status);
1544                         goto bail;
1545                 }
1546
1547                 memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1548
1549                 memcpy(bhs[virtual]->b_data, c,
1550                        (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1551                        bytes_left);
1552
1553                 status = ocfs2_journal_dirty(handle, bhs[virtual]);
1554                 if (status < 0) {
1555                         mlog_errno(status);
1556                         goto bail;
1557                 }
1558
1559                 virtual++;
1560                 p_blkno++;
1561                 bytes_left -= sb->s_blocksize;
1562         }
1563
1564         status = 0;
1565 bail:
1566
1567         if (bhs) {
1568                 for(i = 0; i < blocks; i++)
1569                         if (bhs[i])
1570                                 brelse(bhs[i]);
1571                 kfree(bhs);
1572         }
1573
1574         mlog_exit(status);
1575         return status;
1576 }
1577
1578 static int ocfs2_symlink(struct inode *dir,
1579                          struct dentry *dentry,
1580                          const char *symname)
1581 {
1582         int status, l, credits;
1583         u64 newsize;
1584         struct ocfs2_super *osb = NULL;
1585         struct inode *inode = NULL;
1586         struct super_block *sb;
1587         struct buffer_head *new_fe_bh = NULL;
1588         struct buffer_head *de_bh = NULL;
1589         struct buffer_head *parent_fe_bh = NULL;
1590         struct ocfs2_dinode *fe = NULL;
1591         struct ocfs2_dinode *dirfe;
1592         struct ocfs2_journal_handle *handle = NULL;
1593         struct ocfs2_alloc_context *inode_ac = NULL;
1594         struct ocfs2_alloc_context *data_ac = NULL;
1595
1596         mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1597                    dentry, symname, dentry->d_name.len, dentry->d_name.name);
1598
1599         sb = dir->i_sb;
1600         osb = OCFS2_SB(sb);
1601
1602         l = strlen(symname) + 1;
1603
1604         credits = ocfs2_calc_symlink_credits(sb);
1605
1606         handle = ocfs2_alloc_handle(osb);
1607         if (handle == NULL) {
1608                 status = -ENOMEM;
1609                 mlog_errno(status);
1610                 goto bail;
1611         }
1612
1613         /* lock the parent directory */
1614         status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1615         if (status < 0) {
1616                 if (status != -ENOENT)
1617                         mlog_errno(status);
1618                 goto bail;
1619         }
1620
1621         dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1622         if (!dirfe->i_links_count) {
1623                 /* can't make a file in a deleted directory. */
1624                 status = -ENOENT;
1625                 goto bail;
1626         }
1627
1628         status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1629                                            dentry->d_name.len);
1630         if (status)
1631                 goto bail;
1632
1633         status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1634                                               dentry->d_name.name,
1635                                               dentry->d_name.len, &de_bh);
1636         if (status < 0) {
1637                 mlog_errno(status);
1638                 goto bail;
1639         }
1640
1641         status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1642         if (status < 0) {
1643                 if (status != -ENOSPC)
1644                         mlog_errno(status);
1645                 goto bail;
1646         }
1647
1648         /* don't reserve bitmap space for fast symlinks. */
1649         if (l > ocfs2_fast_symlink_chars(sb)) {
1650                 status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1651                 if (status < 0) {
1652                         if (status != -ENOSPC)
1653                                 mlog_errno(status);
1654                         goto bail;
1655                 }
1656         }
1657
1658         handle = ocfs2_start_trans(osb, handle, credits);
1659         if (IS_ERR(handle)) {
1660                 status = PTR_ERR(handle);
1661                 handle = NULL;
1662                 mlog_errno(status);
1663                 goto bail;
1664         }
1665
1666         status = ocfs2_mknod_locked(osb, dir, dentry,
1667                                     S_IFLNK | S_IRWXUGO, 0,
1668                                     &new_fe_bh, parent_fe_bh, handle,
1669                                     &inode, inode_ac);
1670         if (status < 0) {
1671                 mlog_errno(status);
1672                 goto bail;
1673         }
1674
1675         fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1676         inode->i_rdev = 0;
1677         newsize = l - 1;
1678         if (l > ocfs2_fast_symlink_chars(sb)) {
1679                 inode->i_op = &ocfs2_symlink_inode_operations;
1680                 status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1681                                                     handle, data_ac, NULL,
1682                                                     NULL);
1683                 if (status < 0) {
1684                         if (status != -ENOSPC && status != -EINTR) {
1685                                 mlog(ML_ERROR,
1686                                      "Failed to extend file to %llu\n",
1687                                      (unsigned long long)newsize);
1688                                 mlog_errno(status);
1689                                 status = -ENOSPC;
1690                         }
1691                         goto bail;
1692                 }
1693                 i_size_write(inode, newsize);
1694                 inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1695         } else {
1696                 inode->i_op = &ocfs2_fast_symlink_inode_operations;
1697                 memcpy((char *) fe->id2.i_symlink, symname, l);
1698                 i_size_write(inode, newsize);
1699                 inode->i_blocks = 0;
1700         }
1701
1702         status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1703         if (status < 0) {
1704                 mlog_errno(status);
1705                 goto bail;
1706         }
1707
1708         if (!ocfs2_inode_is_fast_symlink(inode)) {
1709                 status = ocfs2_create_symlink_data(osb, handle, inode,
1710                                                    symname);
1711                 if (status < 0) {
1712                         mlog_errno(status);
1713                         goto bail;
1714                 }
1715         }
1716
1717         status = ocfs2_add_entry(handle, dentry, inode,
1718                                  le64_to_cpu(fe->i_blkno), parent_fe_bh,
1719                                  de_bh);
1720         if (status < 0) {
1721                 mlog_errno(status);
1722                 goto bail;
1723         }
1724
1725         status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1726         if (status) {
1727                 mlog_errno(status);
1728                 goto bail;
1729         }
1730
1731         insert_inode_hash(inode);
1732         dentry->d_op = &ocfs2_dentry_ops;
1733         d_instantiate(dentry, inode);
1734 bail:
1735         if (handle)
1736                 ocfs2_commit_trans(handle);
1737         if (new_fe_bh)
1738                 brelse(new_fe_bh);
1739         if (parent_fe_bh)
1740                 brelse(parent_fe_bh);
1741         if (de_bh)
1742                 brelse(de_bh);
1743         if (inode_ac)
1744                 ocfs2_free_alloc_context(inode_ac);
1745         if (data_ac)
1746                 ocfs2_free_alloc_context(data_ac);
1747         if ((status < 0) && inode)
1748                 iput(inode);
1749
1750         mlog_exit(status);
1751
1752         return status;
1753 }
1754
1755 int ocfs2_check_dir_entry(struct inode * dir,
1756                           struct ocfs2_dir_entry * de,
1757                           struct buffer_head * bh,
1758                           unsigned long offset)
1759 {
1760         const char *error_msg = NULL;
1761         const int rlen = le16_to_cpu(de->rec_len);
1762
1763         if (rlen < OCFS2_DIR_REC_LEN(1))
1764                 error_msg = "rec_len is smaller than minimal";
1765         else if (rlen % 4 != 0)
1766                 error_msg = "rec_len % 4 != 0";
1767         else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1768                 error_msg = "rec_len is too small for name_len";
1769         else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1770                 error_msg = "directory entry across blocks";
1771
1772         if (error_msg != NULL)
1773                 mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1774                      "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1775                      (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1776                      offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1777                      de->name_len);
1778         return error_msg == NULL ? 1 : 0;
1779 }
1780
1781 /* we don't always have a dentry for what we want to add, so people
1782  * like orphan dir can call this instead.
1783  *
1784  * If you pass me insert_bh, I'll skip the search of the other dir
1785  * blocks and put the record in there.
1786  */
1787 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1788                              struct inode *dir,
1789                              const char *name, int namelen,
1790                              struct inode *inode, u64 blkno,
1791                              struct buffer_head *parent_fe_bh,
1792                              struct buffer_head *insert_bh)
1793 {
1794         unsigned long offset;
1795         unsigned short rec_len;
1796         struct ocfs2_dir_entry *de, *de1;
1797         struct super_block *sb;
1798         int retval, status;
1799
1800         mlog_entry_void();
1801
1802         sb = dir->i_sb;
1803
1804         if (!namelen)
1805                 return -EINVAL;
1806
1807         rec_len = OCFS2_DIR_REC_LEN(namelen);
1808         offset = 0;
1809         de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1810         while (1) {
1811                 BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1812                 /* These checks should've already been passed by the
1813                  * prepare function, but I guess we can leave them
1814                  * here anyway. */
1815                 if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1816                         retval = -ENOENT;
1817                         goto bail;
1818                 }
1819                 if (ocfs2_match(namelen, name, de)) {
1820                         retval = -EEXIST;
1821                         goto bail;
1822                 }
1823                 if (((le64_to_cpu(de->inode) == 0) &&
1824                      (le16_to_cpu(de->rec_len) >= rec_len)) ||
1825                     (le16_to_cpu(de->rec_len) >=
1826                      (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1827                         status = ocfs2_journal_access(handle, dir, insert_bh,
1828                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1829                         /* By now the buffer is marked for journaling */
1830                         offset += le16_to_cpu(de->rec_len);
1831                         if (le64_to_cpu(de->inode)) {
1832                                 de1 = (struct ocfs2_dir_entry *)((char *) de +
1833                                         OCFS2_DIR_REC_LEN(de->name_len));
1834                                 de1->rec_len =
1835                                         cpu_to_le16(le16_to_cpu(de->rec_len) -
1836                                         OCFS2_DIR_REC_LEN(de->name_len));
1837                                 de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1838                                 de = de1;
1839                         }
1840                         de->file_type = OCFS2_FT_UNKNOWN;
1841                         if (blkno) {
1842                                 de->inode = cpu_to_le64(blkno);
1843                                 ocfs2_set_de_type(de, inode->i_mode);
1844                         } else
1845                                 de->inode = 0;
1846                         de->name_len = namelen;
1847                         memcpy(de->name, name, namelen);
1848
1849                         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1850                         dir->i_version++;
1851                         status = ocfs2_journal_dirty(handle, insert_bh);
1852                         retval = 0;
1853                         goto bail;
1854                 }
1855                 offset += le16_to_cpu(de->rec_len);
1856                 de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1857         }
1858
1859         /* when you think about it, the assert above should prevent us
1860          * from ever getting here. */
1861         retval = -ENOSPC;
1862 bail:
1863
1864         mlog_exit(retval);
1865         return retval;
1866 }
1867
1868
1869 /*
1870  * ocfs2_delete_entry deletes a directory entry by merging it with the
1871  * previous entry
1872  */
1873 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1874                               struct inode *dir,
1875                               struct ocfs2_dir_entry *de_del,
1876                               struct buffer_head *bh)
1877 {
1878         struct ocfs2_dir_entry *de, *pde;
1879         int i, status = -ENOENT;
1880
1881         mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1882
1883         i = 0;
1884         pde = NULL;
1885         de = (struct ocfs2_dir_entry *) bh->b_data;
1886         while (i < bh->b_size) {
1887                 if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1888                         status = -EIO;
1889                         mlog_errno(status);
1890                         goto bail;
1891                 }
1892                 if (de == de_del)  {
1893                         status = ocfs2_journal_access(handle, dir, bh,
1894                                                       OCFS2_JOURNAL_ACCESS_WRITE);
1895                         if (status < 0) {
1896                                 status = -EIO;
1897                                 mlog_errno(status);
1898                                 goto bail;
1899                         }
1900                         if (pde)
1901                                 pde->rec_len =
1902                                         cpu_to_le16(le16_to_cpu(pde->rec_len) +
1903                                                     le16_to_cpu(de->rec_len));
1904                         else
1905                                 de->inode = 0;
1906                         dir->i_version++;
1907                         status = ocfs2_journal_dirty(handle, bh);
1908                         goto bail;
1909                 }
1910                 i += le16_to_cpu(de->rec_len);
1911                 pde = de;
1912                 de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1913         }
1914 bail:
1915         mlog_exit(status);
1916         return status;
1917 }
1918
1919 /*
1920  * Returns 0 if not found, -1 on failure, and 1 on success
1921  */
1922 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1923                                         struct inode *dir,
1924                                         const char *name, int namelen,
1925                                         unsigned long offset,
1926                                         struct ocfs2_dir_entry **res_dir)
1927 {
1928         struct ocfs2_dir_entry *de;
1929         char *dlimit, *de_buf;
1930         int de_len;
1931         int ret = 0;
1932
1933         mlog_entry_void();
1934
1935         de_buf = bh->b_data;
1936         dlimit = de_buf + dir->i_sb->s_blocksize;
1937
1938         while (de_buf < dlimit) {
1939                 /* this code is executed quadratically often */
1940                 /* do minimal checking `by hand' */
1941
1942                 de = (struct ocfs2_dir_entry *) de_buf;
1943
1944                 if (de_buf + namelen <= dlimit &&
1945                     ocfs2_match(namelen, name, de)) {
1946                         /* found a match - just to be sure, do a full check */
1947                         if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1948                                 ret = -1;
1949                                 goto bail;
1950                         }
1951                         *res_dir = de;
1952                         ret = 1;
1953                         goto bail;
1954                 }
1955
1956                 /* prevent looping on a bad block */
1957                 de_len = le16_to_cpu(de->rec_len);
1958                 if (de_len <= 0) {
1959                         ret = -1;
1960                         goto bail;
1961                 }
1962
1963                 de_buf += de_len;
1964                 offset += de_len;
1965         }
1966
1967 bail:
1968         mlog_exit(ret);
1969         return ret;
1970 }
1971
1972 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1973                                      struct inode *dir,
1974                                      struct ocfs2_dir_entry **res_dir)
1975 {
1976         struct super_block *sb;
1977         struct buffer_head *bh_use[NAMEI_RA_SIZE];
1978         struct buffer_head *bh, *ret = NULL;
1979         unsigned long start, block, b;
1980         int ra_max = 0;         /* Number of bh's in the readahead
1981                                    buffer, bh_use[] */
1982         int ra_ptr = 0;         /* Current index into readahead
1983                                    buffer */
1984         int num = 0;
1985         int nblocks, i, err;
1986
1987         mlog_entry_void();
1988
1989         *res_dir = NULL;
1990         sb = dir->i_sb;
1991
1992         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1993         start = OCFS2_I(dir)->ip_dir_start_lookup;
1994         if (start >= nblocks)
1995                 start = 0;
1996         block = start;
1997
1998 restart:
1999         do {
2000                 /*
2001                  * We deal with the read-ahead logic here.
2002                  */
2003                 if (ra_ptr >= ra_max) {
2004                         /* Refill the readahead buffer */
2005                         ra_ptr = 0;
2006                         b = block;
2007                         for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2008                                 /*
2009                                  * Terminate if we reach the end of the
2010                                  * directory and must wrap, or if our
2011                                  * search has finished at this block.
2012                                  */
2013                                 if (b >= nblocks || (num && block == start)) {
2014                                         bh_use[ra_max] = NULL;
2015                                         break;
2016                                 }
2017                                 num++;
2018
2019                                 bh = ocfs2_bread(dir, b++, &err, 1);
2020                                 bh_use[ra_max] = bh;
2021                         }
2022                 }
2023                 if ((bh = bh_use[ra_ptr++]) == NULL)
2024                         goto next;
2025                 wait_on_buffer(bh);
2026                 if (!buffer_uptodate(bh)) {
2027                         /* read error, skip block & hope for the best */
2028                         ocfs2_error(dir->i_sb, "reading directory %llu, "
2029                                     "offset %lu\n",
2030                                     (unsigned long long)OCFS2_I(dir)->ip_blkno,
2031                                     block);
2032                         brelse(bh);
2033                         goto next;
2034                 }
2035                 i = ocfs2_search_dirblock(bh, dir, name, namelen,
2036                                           block << sb->s_blocksize_bits,
2037                                           res_dir);
2038                 if (i == 1) {
2039                         OCFS2_I(dir)->ip_dir_start_lookup = block;
2040                         ret = bh;
2041                         goto cleanup_and_exit;
2042                 } else {
2043                         brelse(bh);
2044                         if (i < 0)
2045                                 goto cleanup_and_exit;
2046                 }
2047         next:
2048                 if (++block >= nblocks)
2049                         block = 0;
2050         } while (block != start);
2051
2052         /*
2053          * If the directory has grown while we were searching, then
2054          * search the last part of the directory before giving up.
2055          */
2056         block = nblocks;
2057         nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2058         if (block < nblocks) {
2059                 start = 0;
2060                 goto restart;
2061         }
2062
2063 cleanup_and_exit:
2064         /* Clean up the read-ahead blocks */
2065         for (; ra_ptr < ra_max; ra_ptr++)
2066                 brelse(bh_use[ra_ptr]);
2067
2068         mlog_exit_ptr(ret);
2069         return ret;
2070 }
2071
2072 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2073 {
2074         int status, namelen;
2075
2076         mlog_entry_void();
2077
2078         namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2079                            (long long)blkno);
2080         if (namelen <= 0) {
2081                 if (namelen)
2082                         status = namelen;
2083                 else
2084                         status = -EINVAL;
2085                 mlog_errno(status);
2086                 goto bail;
2087         }
2088         if (namelen != OCFS2_ORPHAN_NAMELEN) {
2089                 status = -EINVAL;
2090                 mlog_errno(status);
2091                 goto bail;
2092         }
2093
2094         mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2095              namelen);
2096
2097         status = 0;
2098 bail:
2099         mlog_exit(status);
2100         return status;
2101 }
2102
2103 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2104                                     struct inode **ret_orphan_dir,
2105                                     struct inode *inode,
2106                                     char *name,
2107                                     struct buffer_head **de_bh)
2108 {
2109         struct inode *orphan_dir_inode;
2110         struct buffer_head *orphan_dir_bh = NULL;
2111         int status = 0;
2112
2113         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2114         if (status < 0) {
2115                 mlog_errno(status);
2116                 return status;
2117         }
2118
2119         orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2120                                                        ORPHAN_DIR_SYSTEM_INODE,
2121                                                        osb->slot_num);
2122         if (!orphan_dir_inode) {
2123                 status = -ENOENT;
2124                 mlog_errno(status);
2125                 return status;
2126         }
2127
2128         mutex_lock(&orphan_dir_inode->i_mutex);
2129
2130         status = ocfs2_meta_lock(orphan_dir_inode, NULL, &orphan_dir_bh, 1);
2131         if (status < 0) {
2132                 mlog_errno(status);
2133                 goto leave;
2134         }
2135
2136         status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2137                                               orphan_dir_bh, name,
2138                                               OCFS2_ORPHAN_NAMELEN, de_bh);
2139         if (status < 0) {
2140                 ocfs2_meta_unlock(orphan_dir_inode, 1);
2141
2142                 mlog_errno(status);
2143                 goto leave;
2144         }
2145
2146         *ret_orphan_dir = orphan_dir_inode;
2147
2148 leave:
2149         if (status) {
2150                 mutex_unlock(&orphan_dir_inode->i_mutex);
2151                 iput(orphan_dir_inode);
2152         }
2153
2154         if (orphan_dir_bh)
2155                 brelse(orphan_dir_bh);
2156
2157         mlog_exit(status);
2158         return status;
2159 }
2160
2161 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2162                             struct ocfs2_journal_handle *handle,
2163                             struct inode *inode,
2164                             struct ocfs2_dinode *fe,
2165                             char *name,
2166                             struct buffer_head *de_bh,
2167                             struct inode *orphan_dir_inode)
2168 {
2169         struct buffer_head *orphan_dir_bh = NULL;
2170         int status = 0;
2171         struct ocfs2_dinode *orphan_fe;
2172
2173         mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2174
2175         status = ocfs2_read_block(osb,
2176                                   OCFS2_I(orphan_dir_inode)->ip_blkno,
2177                                   &orphan_dir_bh, OCFS2_BH_CACHED,
2178                                   orphan_dir_inode);
2179         if (status < 0) {
2180                 mlog_errno(status);
2181                 goto leave;
2182         }
2183
2184         status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2185                                       OCFS2_JOURNAL_ACCESS_WRITE);
2186         if (status < 0) {
2187                 mlog_errno(status);
2188                 goto leave;
2189         }
2190
2191         /* we're a cluster, and nlink can change on disk from
2192          * underneath us... */
2193         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2194         if (S_ISDIR(inode->i_mode))
2195                 le16_add_cpu(&orphan_fe->i_links_count, 1);
2196         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2197
2198         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2199         if (status < 0) {
2200                 mlog_errno(status);
2201                 goto leave;
2202         }
2203
2204         status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2205                                    OCFS2_ORPHAN_NAMELEN, inode,
2206                                    OCFS2_I(inode)->ip_blkno,
2207                                    orphan_dir_bh, de_bh);
2208         if (status < 0) {
2209                 mlog_errno(status);
2210                 goto leave;
2211         }
2212
2213         le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2214
2215         /* Record which orphan dir our inode now resides
2216          * in. delete_inode will use this to determine which orphan
2217          * dir to lock. */
2218         spin_lock(&OCFS2_I(inode)->ip_lock);
2219         OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2220         spin_unlock(&OCFS2_I(inode)->ip_lock);
2221
2222         mlog(0, "Inode %llu orphaned in slot %d\n",
2223              (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2224
2225 leave:
2226         if (orphan_dir_bh)
2227                 brelse(orphan_dir_bh);
2228
2229         mlog_exit(status);
2230         return status;
2231 }
2232
2233 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2234 int ocfs2_orphan_del(struct ocfs2_super *osb,
2235                      struct ocfs2_journal_handle *handle,
2236                      struct inode *orphan_dir_inode,
2237                      struct inode *inode,
2238                      struct buffer_head *orphan_dir_bh)
2239 {
2240         char name[OCFS2_ORPHAN_NAMELEN + 1];
2241         struct ocfs2_dinode *orphan_fe;
2242         int status = 0;
2243         struct buffer_head *target_de_bh = NULL;
2244         struct ocfs2_dir_entry *target_de = NULL;
2245
2246         mlog_entry_void();
2247
2248         status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2249         if (status < 0) {
2250                 mlog_errno(status);
2251                 goto leave;
2252         }
2253
2254         mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2255              name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2256              OCFS2_ORPHAN_NAMELEN);
2257
2258         /* find it's spot in the orphan directory */
2259         target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2260                                         orphan_dir_inode, &target_de);
2261         if (!target_de_bh) {
2262                 status = -ENOENT;
2263                 mlog_errno(status);
2264                 goto leave;
2265         }
2266
2267         /* remove it from the orphan directory */
2268         status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2269                                     target_de_bh);
2270         if (status < 0) {
2271                 mlog_errno(status);
2272                 goto leave;
2273         }
2274
2275         status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2276                                       OCFS2_JOURNAL_ACCESS_WRITE);
2277         if (status < 0) {
2278                 mlog_errno(status);
2279                 goto leave;
2280         }
2281
2282         /* do the i_nlink dance! :) */
2283         orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2284         if (S_ISDIR(inode->i_mode))
2285                 le16_add_cpu(&orphan_fe->i_links_count, -1);
2286         orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2287
2288         status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2289         if (status < 0) {
2290                 mlog_errno(status);
2291                 goto leave;
2292         }
2293
2294 leave:
2295         if (target_de_bh)
2296                 brelse(target_de_bh);
2297
2298         mlog_exit(status);
2299         return status;
2300 }
2301
2302 struct inode_operations ocfs2_dir_iops = {
2303         .create         = ocfs2_create,
2304         .lookup         = ocfs2_lookup,
2305         .link           = ocfs2_link,
2306         .unlink         = ocfs2_unlink,
2307         .rmdir          = ocfs2_unlink,
2308         .symlink        = ocfs2_symlink,
2309         .mkdir          = ocfs2_mkdir,
2310         .mknod          = ocfs2_mknod,
2311         .rename         = ocfs2_rename,
2312         .setattr        = ocfs2_setattr,
2313         .getattr        = ocfs2_getattr,
2314 };