xref: /linux/fs/ocfs2/namei.c (revision f3d9478b2ce468c3115b02ecae7e975990697f15)
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 "symlink.h"
60 #include "sysfile.h"
61 #include "uptodate.h"
62 #include "vote.h"
63 
64 #include "buffer_head_io.h"
65 
66 #define NAMEI_RA_CHUNKS  2
67 #define NAMEI_RA_BLOCKS  4
68 #define NAMEI_RA_SIZE        (NAMEI_RA_CHUNKS * NAMEI_RA_BLOCKS)
69 #define NAMEI_RA_INDEX(c,b)  (((c) * NAMEI_RA_BLOCKS) + (b))
70 
71 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
72 					struct inode *dir,
73 					const char *name, int namelen,
74 					unsigned long offset,
75 					struct ocfs2_dir_entry **res_dir);
76 
77 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
78 			      struct inode *dir,
79 			      struct ocfs2_dir_entry *de_del,
80 			      struct buffer_head *bh);
81 
82 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
83 			     struct inode *dir,
84 			     const char *name, int namelen,
85 			     struct inode *inode, u64 blkno,
86 			     struct buffer_head *parent_fe_bh,
87 			     struct buffer_head *insert_bh);
88 
89 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
90 			      struct inode *dir,
91 			      struct dentry *dentry, int mode,
92 			      dev_t dev,
93 			      struct buffer_head **new_fe_bh,
94 			      struct buffer_head *parent_fe_bh,
95 			      struct ocfs2_journal_handle *handle,
96 			      struct inode **ret_inode,
97 			      struct ocfs2_alloc_context *inode_ac);
98 
99 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
100 			      struct ocfs2_journal_handle *handle,
101 			      struct inode *parent,
102 			      struct inode *inode,
103 			      struct buffer_head *fe_bh,
104 			      struct ocfs2_alloc_context *data_ac);
105 
106 static int ocfs2_double_lock(struct ocfs2_super *osb,
107 			     struct ocfs2_journal_handle *handle,
108 			     struct buffer_head **bh1,
109 			     struct inode *inode1,
110 			     struct buffer_head **bh2,
111 			     struct inode *inode2);
112 
113 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
114 				    struct ocfs2_journal_handle *handle,
115 				    struct inode *inode,
116 				    char *name,
117 				    struct buffer_head **de_bh);
118 
119 static int ocfs2_orphan_add(struct ocfs2_super *osb,
120 			    struct ocfs2_journal_handle *handle,
121 			    struct inode *inode,
122 			    struct ocfs2_dinode *fe,
123 			    char *name,
124 			    struct buffer_head *de_bh);
125 
126 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
127 				     struct ocfs2_journal_handle *handle,
128 				     struct inode *inode,
129 				     const char *symname);
130 
131 static inline int ocfs2_add_entry(struct ocfs2_journal_handle *handle,
132 				  struct dentry *dentry,
133 				  struct inode *inode, u64 blkno,
134 				  struct buffer_head *parent_fe_bh,
135 				  struct buffer_head *insert_bh)
136 {
137 	return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
138 				 dentry->d_name.name, dentry->d_name.len,
139 				 inode, blkno, parent_fe_bh, insert_bh);
140 }
141 
142 /* An orphan dir name is an 8 byte value, printed as a hex string */
143 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
144 
145 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
146 				   struct nameidata *nd)
147 {
148 	int status;
149 	u64 blkno;
150 	struct buffer_head *dirent_bh = NULL;
151 	struct inode *inode = NULL;
152 	struct dentry *ret;
153 	struct ocfs2_dir_entry *dirent;
154 	struct ocfs2_inode_info *oi;
155 
156 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
157 		   dentry->d_name.len, dentry->d_name.name);
158 
159 	if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
160 		ret = ERR_PTR(-ENAMETOOLONG);
161 		goto bail;
162 	}
163 
164 	mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
165 	     dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
166 
167 	status = ocfs2_meta_lock(dir, NULL, NULL, 0);
168 	if (status < 0) {
169 		if (status != -ENOENT)
170 			mlog_errno(status);
171 		ret = ERR_PTR(status);
172 		goto bail;
173 	}
174 
175 	status = ocfs2_find_files_on_disk(dentry->d_name.name,
176 					  dentry->d_name.len, &blkno,
177 					  dir, &dirent_bh, &dirent);
178 	if (status < 0)
179 		goto bail_add;
180 
181 	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno);
182 	if (IS_ERR(inode)) {
183 		mlog(ML_ERROR, "Unable to create inode %llu\n",
184 		     (unsigned long long)blkno);
185 		ret = ERR_PTR(-EACCES);
186 		goto bail_unlock;
187 	}
188 
189 	oi = OCFS2_I(inode);
190 	/* Clear any orphaned state... If we were able to look up the
191 	 * inode from a directory, it certainly can't be orphaned. We
192 	 * might have the bad state from a node which intended to
193 	 * orphan this inode but crashed before it could commit the
194 	 * unlink. */
195 	spin_lock(&oi->ip_lock);
196 	oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
197 	oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
198 	spin_unlock(&oi->ip_lock);
199 
200 bail_add:
201 
202 	dentry->d_op = &ocfs2_dentry_ops;
203 	ret = d_splice_alias(inode, dentry);
204 
205 bail_unlock:
206 	/* Don't drop the cluster lock until *after* the d_add --
207 	 * unlink on another node will message us to remove that
208 	 * dentry under this lock so otherwise we can race this with
209 	 * the vote thread and have a stale dentry. */
210 	ocfs2_meta_unlock(dir, 0);
211 
212 bail:
213 	if (dirent_bh)
214 		brelse(dirent_bh);
215 
216 	mlog_exit_ptr(ret);
217 
218 	return ret;
219 }
220 
221 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
222 			      struct ocfs2_journal_handle *handle,
223 			      struct inode *parent,
224 			      struct inode *inode,
225 			      struct buffer_head *fe_bh,
226 			      struct ocfs2_alloc_context *data_ac)
227 {
228 	int status;
229 	struct buffer_head *new_bh = NULL;
230 	struct ocfs2_dir_entry *de = NULL;
231 
232 	mlog_entry_void();
233 
234 	status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
235 				     data_ac, NULL, &new_bh);
236 	if (status < 0) {
237 		mlog_errno(status);
238 		goto bail;
239 	}
240 
241 	ocfs2_set_new_buffer_uptodate(inode, new_bh);
242 
243 	status = ocfs2_journal_access(handle, inode, new_bh,
244 				      OCFS2_JOURNAL_ACCESS_CREATE);
245 	if (status < 0) {
246 		mlog_errno(status);
247 		goto bail;
248 	}
249 	memset(new_bh->b_data, 0, osb->sb->s_blocksize);
250 
251 	de = (struct ocfs2_dir_entry *) new_bh->b_data;
252 	de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
253 	de->name_len = 1;
254 	de->rec_len =
255 		cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
256 	strcpy(de->name, ".");
257 	ocfs2_set_de_type(de, S_IFDIR);
258 	de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
259 	de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
260 	de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
261 				  OCFS2_DIR_REC_LEN(1));
262 	de->name_len = 2;
263 	strcpy(de->name, "..");
264 	ocfs2_set_de_type(de, S_IFDIR);
265 
266 	status = ocfs2_journal_dirty(handle, new_bh);
267 	if (status < 0) {
268 		mlog_errno(status);
269 		goto bail;
270 	}
271 
272 	i_size_write(inode, inode->i_sb->s_blocksize);
273 	inode->i_nlink = 2;
274 	inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
275 	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
276 	if (status < 0) {
277 		mlog_errno(status);
278 		goto bail;
279 	}
280 
281 	status = 0;
282 bail:
283 	if (new_bh)
284 		brelse(new_bh);
285 
286 	mlog_exit(status);
287 	return status;
288 }
289 
290 static int ocfs2_mknod(struct inode *dir,
291 		       struct dentry *dentry,
292 		       int mode,
293 		       dev_t dev)
294 {
295 	int status = 0;
296 	struct buffer_head *parent_fe_bh = NULL;
297 	struct ocfs2_journal_handle *handle = NULL;
298 	struct ocfs2_super *osb;
299 	struct ocfs2_dinode *dirfe;
300 	struct buffer_head *new_fe_bh = NULL;
301 	struct buffer_head *de_bh = NULL;
302 	struct inode *inode = NULL;
303 	struct ocfs2_alloc_context *inode_ac = NULL;
304 	struct ocfs2_alloc_context *data_ac = NULL;
305 
306 	mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
307 		   (unsigned long)dev, dentry->d_name.len,
308 		   dentry->d_name.name);
309 
310 	/* get our super block */
311 	osb = OCFS2_SB(dir->i_sb);
312 
313 	if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
314 		mlog(ML_ERROR, "inode %llu has i_nlink of %u\n",
315 		     (unsigned long long)OCFS2_I(dir)->ip_blkno, dir->i_nlink);
316 		status = -EMLINK;
317 		goto leave;
318 	}
319 
320 	handle = ocfs2_alloc_handle(osb);
321 	if (handle == NULL) {
322 		status = -ENOMEM;
323 		mlog_errno(status);
324 		goto leave;
325 	}
326 
327 	status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
328 	if (status < 0) {
329 		if (status != -ENOENT)
330 			mlog_errno(status);
331 		goto leave;
332 	}
333 
334 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
335 	if (!dirfe->i_links_count) {
336 		/* can't make a file in a deleted directory. */
337 		status = -ENOENT;
338 		goto leave;
339 	}
340 
341 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
342 					   dentry->d_name.len);
343 	if (status)
344 		goto leave;
345 
346 	/* get a spot inside the dir. */
347 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
348 					      dentry->d_name.name,
349 					      dentry->d_name.len, &de_bh);
350 	if (status < 0) {
351 		mlog_errno(status);
352 		goto leave;
353 	}
354 
355 	/* reserve an inode spot */
356 	status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
357 	if (status < 0) {
358 		if (status != -ENOSPC)
359 			mlog_errno(status);
360 		goto leave;
361 	}
362 
363 	/* are we making a directory? If so, reserve a cluster for his
364 	 * 1st extent. */
365 	if (S_ISDIR(mode)) {
366 		status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
367 		if (status < 0) {
368 			if (status != -ENOSPC)
369 				mlog_errno(status);
370 			goto leave;
371 		}
372 	}
373 
374 	handle = ocfs2_start_trans(osb, handle, OCFS2_MKNOD_CREDITS);
375 	if (IS_ERR(handle)) {
376 		status = PTR_ERR(handle);
377 		handle = NULL;
378 		mlog_errno(status);
379 		goto leave;
380 	}
381 
382 	/* do the real work now. */
383 	status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
384 				    &new_fe_bh, parent_fe_bh, handle,
385 				    &inode, inode_ac);
386 	if (status < 0) {
387 		mlog_errno(status);
388 		goto leave;
389 	}
390 
391 	if (S_ISDIR(mode)) {
392 		status = ocfs2_fill_new_dir(osb, handle, dir, inode,
393 					    new_fe_bh, data_ac);
394 		if (status < 0) {
395 			mlog_errno(status);
396 			goto leave;
397 		}
398 
399 		status = ocfs2_journal_access(handle, dir, parent_fe_bh,
400 					      OCFS2_JOURNAL_ACCESS_WRITE);
401 		if (status < 0) {
402 			mlog_errno(status);
403 			goto leave;
404 		}
405 		le16_add_cpu(&dirfe->i_links_count, 1);
406 		status = ocfs2_journal_dirty(handle, parent_fe_bh);
407 		if (status < 0) {
408 			mlog_errno(status);
409 			goto leave;
410 		}
411 		dir->i_nlink++;
412 	}
413 
414 	status = ocfs2_add_entry(handle, dentry, inode,
415 				 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
416 				 de_bh);
417 	if (status < 0) {
418 		mlog_errno(status);
419 		goto leave;
420 	}
421 
422 	insert_inode_hash(inode);
423 	dentry->d_op = &ocfs2_dentry_ops;
424 	d_instantiate(dentry, inode);
425 	status = 0;
426 leave:
427 	if (handle)
428 		ocfs2_commit_trans(handle);
429 
430 	if (status == -ENOSPC)
431 		mlog(0, "Disk is full\n");
432 
433 	if (new_fe_bh)
434 		brelse(new_fe_bh);
435 
436 	if (de_bh)
437 		brelse(de_bh);
438 
439 	if (parent_fe_bh)
440 		brelse(parent_fe_bh);
441 
442 	if ((status < 0) && inode)
443 		iput(inode);
444 
445 	if (inode_ac)
446 		ocfs2_free_alloc_context(inode_ac);
447 
448 	if (data_ac)
449 		ocfs2_free_alloc_context(data_ac);
450 
451 	mlog_exit(status);
452 
453 	return status;
454 }
455 
456 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
457 			      struct inode *dir,
458 			      struct dentry *dentry, int mode,
459 			      dev_t dev,
460 			      struct buffer_head **new_fe_bh,
461 			      struct buffer_head *parent_fe_bh,
462 			      struct ocfs2_journal_handle *handle,
463 			      struct inode **ret_inode,
464 			      struct ocfs2_alloc_context *inode_ac)
465 {
466 	int status = 0;
467 	struct ocfs2_dinode *fe = NULL;
468 	struct ocfs2_extent_list *fel;
469 	u64 fe_blkno = 0;
470 	u16 suballoc_bit;
471 	struct inode *inode = NULL;
472 
473 	mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
474 		   (unsigned long)dev, dentry->d_name.len,
475 		   dentry->d_name.name);
476 
477 	*new_fe_bh = NULL;
478 	*ret_inode = NULL;
479 
480 	status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
481 				       &fe_blkno);
482 	if (status < 0) {
483 		mlog_errno(status);
484 		goto leave;
485 	}
486 
487 	inode = new_inode(dir->i_sb);
488 	if (IS_ERR(inode)) {
489 		status = PTR_ERR(inode);
490 		mlog(ML_ERROR, "new_inode failed!\n");
491 		goto leave;
492 	}
493 
494 	/* populate as many fields early on as possible - many of
495 	 * these are used by the support functions here and in
496 	 * callers. */
497 	inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
498 	OCFS2_I(inode)->ip_blkno = fe_blkno;
499 	if (S_ISDIR(mode))
500 		inode->i_nlink = 2;
501 	else
502 		inode->i_nlink = 1;
503 	inode->i_mode = mode;
504 	spin_lock(&osb->osb_lock);
505 	inode->i_generation = osb->s_next_generation++;
506 	spin_unlock(&osb->osb_lock);
507 
508 	*new_fe_bh = sb_getblk(osb->sb, fe_blkno);
509 	if (!*new_fe_bh) {
510 		status = -EIO;
511 		mlog_errno(status);
512 		goto leave;
513 	}
514 	ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
515 
516 	status = ocfs2_journal_access(handle, inode, *new_fe_bh,
517 				      OCFS2_JOURNAL_ACCESS_CREATE);
518 	if (status < 0) {
519 		mlog_errno(status);
520 		goto leave;
521 	}
522 
523 	fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
524 	memset(fe, 0, osb->sb->s_blocksize);
525 
526 	fe->i_generation = cpu_to_le32(inode->i_generation);
527 	fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
528 	fe->i_blkno = cpu_to_le64(fe_blkno);
529 	fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
530 	fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
531 	fe->i_uid = cpu_to_le32(current->fsuid);
532 	if (dir->i_mode & S_ISGID) {
533 		fe->i_gid = cpu_to_le32(dir->i_gid);
534 		if (S_ISDIR(mode))
535 			mode |= S_ISGID;
536 	} else
537 		fe->i_gid = cpu_to_le32(current->fsgid);
538 	fe->i_mode = cpu_to_le16(mode);
539 	if (S_ISCHR(mode) || S_ISBLK(mode))
540 		fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
541 
542 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
543 
544 	fe->i_last_eb_blk = 0;
545 	strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
546 	le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
547 	fe->i_atime = fe->i_ctime = fe->i_mtime =
548 		cpu_to_le64(CURRENT_TIME.tv_sec);
549 	fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
550 		cpu_to_le32(CURRENT_TIME.tv_nsec);
551 	fe->i_dtime = 0;
552 
553 	fel = &fe->id2.i_list;
554 	fel->l_tree_depth = 0;
555 	fel->l_next_free_rec = 0;
556 	fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
557 
558 	status = ocfs2_journal_dirty(handle, *new_fe_bh);
559 	if (status < 0) {
560 		mlog_errno(status);
561 		goto leave;
562 	}
563 
564 	if (ocfs2_populate_inode(inode, fe, 1) < 0) {
565 		mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
566 		     "i_blkno=%llu, i_ino=%lu\n",
567 		     (unsigned long long) (*new_fe_bh)->b_blocknr,
568 		     (unsigned long long)fe->i_blkno, inode->i_ino);
569 		BUG();
570 	}
571 
572 	ocfs2_inode_set_new(osb, inode);
573 	status = ocfs2_create_new_inode_locks(inode);
574 	if (status < 0)
575 		mlog_errno(status);
576 
577 	status = 0; /* error in ocfs2_create_new_inode_locks is not
578 		     * critical */
579 
580 	*ret_inode = inode;
581 leave:
582 	if (status < 0) {
583 		if (*new_fe_bh) {
584 			brelse(*new_fe_bh);
585 			*new_fe_bh = NULL;
586 		}
587 		if (inode)
588 			iput(inode);
589 	}
590 
591 	mlog_exit(status);
592 	return status;
593 }
594 
595 static int ocfs2_mkdir(struct inode *dir,
596 		       struct dentry *dentry,
597 		       int mode)
598 {
599 	int ret;
600 
601 	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
602 		   dentry->d_name.len, dentry->d_name.name);
603 	ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
604 	mlog_exit(ret);
605 
606 	return ret;
607 }
608 
609 static int ocfs2_create(struct inode *dir,
610 			struct dentry *dentry,
611 			int mode,
612 			struct nameidata *nd)
613 {
614 	int ret;
615 
616 	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
617 		   dentry->d_name.len, dentry->d_name.name);
618 	ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
619 	mlog_exit(ret);
620 
621 	return ret;
622 }
623 
624 static int ocfs2_link(struct dentry *old_dentry,
625 		      struct inode *dir,
626 		      struct dentry *dentry)
627 {
628 	struct ocfs2_journal_handle *handle = NULL;
629 	struct inode *inode = old_dentry->d_inode;
630 	int err;
631 	struct buffer_head *fe_bh = NULL;
632 	struct buffer_head *parent_fe_bh = NULL;
633 	struct buffer_head *de_bh = NULL;
634 	struct ocfs2_dinode *fe = NULL;
635 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
636 
637 	mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
638 		   old_dentry->d_name.len, old_dentry->d_name.name,
639 		   dentry->d_name.len, dentry->d_name.name);
640 
641 	if (S_ISDIR(inode->i_mode)) {
642 		err = -EPERM;
643 		goto bail;
644 	}
645 
646 	if (inode->i_nlink >= OCFS2_LINK_MAX) {
647 		err = -EMLINK;
648 		goto bail;
649 	}
650 
651 	handle = ocfs2_alloc_handle(osb);
652 	if (handle == NULL) {
653 		err = -ENOMEM;
654 		goto bail;
655 	}
656 
657 	err = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
658 	if (err < 0) {
659 		if (err != -ENOENT)
660 			mlog_errno(err);
661 		goto bail;
662 	}
663 
664 	err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
665 					dentry->d_name.len);
666 	if (err)
667 		goto bail;
668 
669 	err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
670 					   dentry->d_name.name,
671 					   dentry->d_name.len, &de_bh);
672 	if (err < 0) {
673 		mlog_errno(err);
674 		goto bail;
675 	}
676 
677 	err = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
678 	if (err < 0) {
679 		if (err != -ENOENT)
680 			mlog_errno(err);
681 		goto bail;
682 	}
683 
684 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
685 	if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
686 		err = -EMLINK;
687 		goto bail;
688 	}
689 
690 	handle = ocfs2_start_trans(osb, handle, OCFS2_LINK_CREDITS);
691 	if (IS_ERR(handle)) {
692 		err = PTR_ERR(handle);
693 		handle = NULL;
694 		mlog_errno(err);
695 		goto bail;
696 	}
697 
698 	err = ocfs2_journal_access(handle, inode, fe_bh,
699 				   OCFS2_JOURNAL_ACCESS_WRITE);
700 	if (err < 0) {
701 		mlog_errno(err);
702 		goto bail;
703 	}
704 
705 	inode->i_nlink++;
706 	inode->i_ctime = CURRENT_TIME;
707 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
708 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
709 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
710 
711 	err = ocfs2_journal_dirty(handle, fe_bh);
712 	if (err < 0) {
713 		le16_add_cpu(&fe->i_links_count, -1);
714 		inode->i_nlink--;
715 		mlog_errno(err);
716 		goto bail;
717 	}
718 
719 	err = ocfs2_add_entry(handle, dentry, inode,
720 			      OCFS2_I(inode)->ip_blkno,
721 			      parent_fe_bh, de_bh);
722 	if (err) {
723 		le16_add_cpu(&fe->i_links_count, -1);
724 		inode->i_nlink--;
725 		mlog_errno(err);
726 		goto bail;
727 	}
728 
729 	atomic_inc(&inode->i_count);
730 	dentry->d_op = &ocfs2_dentry_ops;
731 	d_instantiate(dentry, inode);
732 bail:
733 	if (handle)
734 		ocfs2_commit_trans(handle);
735 	if (de_bh)
736 		brelse(de_bh);
737 	if (fe_bh)
738 		brelse(fe_bh);
739 	if (parent_fe_bh)
740 		brelse(parent_fe_bh);
741 
742 	mlog_exit(err);
743 
744 	return err;
745 }
746 
747 static int ocfs2_unlink(struct inode *dir,
748 			struct dentry *dentry)
749 {
750 	int status;
751 	unsigned int saved_nlink = 0;
752 	struct inode *inode = dentry->d_inode;
753 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
754 	u64 blkno;
755 	struct ocfs2_dinode *fe = NULL;
756 	struct buffer_head *fe_bh = NULL;
757 	struct buffer_head *parent_node_bh = NULL;
758 	struct ocfs2_journal_handle *handle = NULL;
759 	struct ocfs2_dir_entry *dirent = NULL;
760 	struct buffer_head *dirent_bh = NULL;
761 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
762 	struct buffer_head *orphan_entry_bh = NULL;
763 
764 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
765 		   dentry->d_name.len, dentry->d_name.name);
766 
767 	BUG_ON(dentry->d_parent->d_inode != dir);
768 
769 	mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
770 
771 	if (inode == osb->root_inode) {
772 		mlog(0, "Cannot delete the root directory\n");
773 		status = -EPERM;
774 		goto leave;
775 	}
776 
777 	handle = ocfs2_alloc_handle(osb);
778 	if (handle == NULL) {
779 		status = -ENOMEM;
780 		mlog_errno(status);
781 		goto leave;
782 	}
783 
784 	status = ocfs2_meta_lock(dir, handle, &parent_node_bh, 1);
785 	if (status < 0) {
786 		if (status != -ENOENT)
787 			mlog_errno(status);
788 		goto leave;
789 	}
790 
791 	status = ocfs2_find_files_on_disk(dentry->d_name.name,
792 					  dentry->d_name.len, &blkno,
793 					  dir, &dirent_bh, &dirent);
794 	if (status < 0) {
795 		if (status != -ENOENT)
796 			mlog_errno(status);
797 		goto leave;
798 	}
799 
800 	if (OCFS2_I(inode)->ip_blkno != blkno) {
801 		status = -ENOENT;
802 
803 		mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
804 		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
805 		     (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
806 		goto leave;
807 	}
808 
809 	status = ocfs2_meta_lock(inode, handle, &fe_bh, 1);
810 	if (status < 0) {
811 		if (status != -ENOENT)
812 			mlog_errno(status);
813 		goto leave;
814 	}
815 
816 	if (S_ISDIR(inode->i_mode)) {
817 	       	if (!ocfs2_empty_dir(inode)) {
818 			status = -ENOTEMPTY;
819 			goto leave;
820 		} else if (inode->i_nlink != 2) {
821 			status = -ENOTEMPTY;
822 			goto leave;
823 		}
824 	}
825 
826 	/* There are still a few steps left until we can consider the
827 	 * unlink to have succeeded. Save off nlink here before
828 	 * modification so we can set it back in case we hit an issue
829 	 * before commit. */
830 	saved_nlink = inode->i_nlink;
831 	if (S_ISDIR(inode->i_mode))
832 		inode->i_nlink = 0;
833 	else
834 		inode->i_nlink--;
835 
836 	status = ocfs2_request_unlink_vote(inode, dentry,
837 					   (unsigned int) inode->i_nlink);
838 	if (status < 0) {
839 		/* This vote should succeed under all normal
840 		 * circumstances. */
841 		mlog_errno(status);
842 		goto leave;
843 	}
844 
845 	if (!inode->i_nlink) {
846 		status = ocfs2_prepare_orphan_dir(osb, handle, inode,
847 						  orphan_name,
848 						  &orphan_entry_bh);
849 		if (status < 0) {
850 			mlog_errno(status);
851 			goto leave;
852 		}
853 	}
854 
855 	handle = ocfs2_start_trans(osb, handle, OCFS2_UNLINK_CREDITS);
856 	if (IS_ERR(handle)) {
857 		status = PTR_ERR(handle);
858 		handle = NULL;
859 		mlog_errno(status);
860 		goto leave;
861 	}
862 
863 	status = ocfs2_journal_access(handle, inode, fe_bh,
864 				      OCFS2_JOURNAL_ACCESS_WRITE);
865 	if (status < 0) {
866 		mlog_errno(status);
867 		goto leave;
868 	}
869 
870 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
871 
872 	if (!inode->i_nlink) {
873 		status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
874 					  orphan_entry_bh);
875 		if (status < 0) {
876 			mlog_errno(status);
877 			goto leave;
878 		}
879 	}
880 
881 	/* delete the name from the parent dir */
882 	status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
883 	if (status < 0) {
884 		mlog_errno(status);
885 		goto leave;
886 	}
887 
888 	/* We can set nlink on the dinode now. clear the saved version
889 	 * so that it doesn't get set later. */
890 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
891 	saved_nlink = 0;
892 
893 	status = ocfs2_journal_dirty(handle, fe_bh);
894 	if (status < 0) {
895 		mlog_errno(status);
896 		goto leave;
897 	}
898 
899 	if (S_ISDIR(inode->i_mode)) {
900 		dir->i_nlink--;
901 		status = ocfs2_mark_inode_dirty(handle, dir,
902 						parent_node_bh);
903 		if (status < 0) {
904 			mlog_errno(status);
905 			dir->i_nlink++;
906 		}
907 	}
908 
909 leave:
910 	if (status < 0 && saved_nlink)
911 		inode->i_nlink = saved_nlink;
912 
913 	if (handle)
914 		ocfs2_commit_trans(handle);
915 
916 	if (fe_bh)
917 		brelse(fe_bh);
918 
919 	if (dirent_bh)
920 		brelse(dirent_bh);
921 
922 	if (parent_node_bh)
923 		brelse(parent_node_bh);
924 
925 	if (orphan_entry_bh)
926 		brelse(orphan_entry_bh);
927 
928 	mlog_exit(status);
929 
930 	return status;
931 }
932 
933 /*
934  * The only place this should be used is rename!
935  * if they have the same id, then the 1st one is the only one locked.
936  */
937 static int ocfs2_double_lock(struct ocfs2_super *osb,
938 			     struct ocfs2_journal_handle *handle,
939 			     struct buffer_head **bh1,
940 			     struct inode *inode1,
941 			     struct buffer_head **bh2,
942 			     struct inode *inode2)
943 {
944 	int status;
945 	struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
946 	struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
947 	struct buffer_head **tmpbh;
948 	struct inode *tmpinode;
949 
950 	mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
951 		   (unsigned long long)oi1->ip_blkno,
952 		   (unsigned long long)oi2->ip_blkno);
953 
954 	BUG_ON(!handle);
955 
956 	if (*bh1)
957 		*bh1 = NULL;
958 	if (*bh2)
959 		*bh2 = NULL;
960 
961 	/* we always want to lock the one with the lower lockid first. */
962 	if (oi1->ip_blkno != oi2->ip_blkno) {
963 		if (oi1->ip_blkno < oi2->ip_blkno) {
964 			/* switch id1 and id2 around */
965 			mlog(0, "switching them around...\n");
966 			tmpbh = bh2;
967 			bh2 = bh1;
968 			bh1 = tmpbh;
969 
970 			tmpinode = inode2;
971 			inode2 = inode1;
972 			inode1 = tmpinode;
973 		}
974 		/* lock id2 */
975 		status = ocfs2_meta_lock(inode2, handle, bh2, 1);
976 		if (status < 0) {
977 			if (status != -ENOENT)
978 				mlog_errno(status);
979 			goto bail;
980 		}
981 	}
982 	/* lock id1 */
983 	status = ocfs2_meta_lock(inode1, handle, bh1, 1);
984 	if (status < 0) {
985 		if (status != -ENOENT)
986 			mlog_errno(status);
987 		goto bail;
988 	}
989 bail:
990 	mlog_exit(status);
991 	return status;
992 }
993 
994 #define PARENT_INO(buffer) \
995 	((struct ocfs2_dir_entry *) \
996 	 ((char *)buffer + \
997 	  le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
998 
999 static int ocfs2_rename(struct inode *old_dir,
1000 			struct dentry *old_dentry,
1001 			struct inode *new_dir,
1002 			struct dentry *new_dentry)
1003 {
1004 	int status = 0, rename_lock = 0;
1005 	struct inode *old_inode = old_dentry->d_inode;
1006 	struct inode *new_inode = new_dentry->d_inode;
1007 	struct ocfs2_dinode *newfe = NULL;
1008 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1009 	struct buffer_head *orphan_entry_bh = NULL;
1010 	struct buffer_head *newfe_bh = NULL;
1011 	struct buffer_head *insert_entry_bh = NULL;
1012 	struct ocfs2_super *osb = NULL;
1013 	u64 newfe_blkno;
1014 	struct ocfs2_journal_handle *handle = NULL;
1015 	struct buffer_head *old_dir_bh = NULL;
1016 	struct buffer_head *new_dir_bh = NULL;
1017 	struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1018 							       // and new_dentry
1019 	struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1020 	struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1021 						    // this is the 1st dirent bh
1022 	nlink_t old_dir_nlink = old_dir->i_nlink, new_dir_nlink = new_dir->i_nlink;
1023 	unsigned int links_count;
1024 
1025 	/* At some point it might be nice to break this function up a
1026 	 * bit. */
1027 
1028 	mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1029 		   old_dir, old_dentry, new_dir, new_dentry,
1030 		   old_dentry->d_name.len, old_dentry->d_name.name,
1031 		   new_dentry->d_name.len, new_dentry->d_name.name);
1032 
1033 	osb = OCFS2_SB(old_dir->i_sb);
1034 
1035 	if (new_inode) {
1036 		if (!igrab(new_inode))
1037 			BUG();
1038 	}
1039 
1040 	if (atomic_read(&old_dentry->d_count) > 2) {
1041 		shrink_dcache_parent(old_dentry);
1042 		if (atomic_read(&old_dentry->d_count) > 2) {
1043 			status = -EBUSY;
1044 			goto bail;
1045 		}
1046 	}
1047 
1048 	/* Assume a directory heirarchy thusly:
1049 	 * a/b/c
1050 	 * a/d
1051 	 * a,b,c, and d are all directories.
1052 	 *
1053 	 * from cwd of 'a' on both nodes:
1054 	 * node1: mv b/c d
1055 	 * node2: mv d   b/c
1056 	 *
1057 	 * And that's why, just like the VFS, we need a file system
1058 	 * rename lock. */
1059 	if (old_dentry != new_dentry) {
1060 		status = ocfs2_rename_lock(osb);
1061 		if (status < 0) {
1062 			mlog_errno(status);
1063 			goto bail;
1064 		}
1065 		rename_lock = 1;
1066 	}
1067 
1068 	handle = ocfs2_alloc_handle(osb);
1069 	if (handle == NULL) {
1070 		status = -ENOMEM;
1071 		mlog_errno(status);
1072 		goto bail;
1073 	}
1074 
1075 	/* if old and new are the same, this'll just do one lock. */
1076 	status = ocfs2_double_lock(osb, handle,
1077 				  &old_dir_bh, old_dir,
1078 				  &new_dir_bh, new_dir);
1079 	if (status < 0) {
1080 		mlog_errno(status);
1081 		goto bail;
1082 	}
1083 
1084 	/* make sure both dirs have bhs
1085 	 * get an extra ref on old_dir_bh if old==new */
1086 	if (!new_dir_bh) {
1087 		if (old_dir_bh) {
1088 			new_dir_bh = old_dir_bh;
1089 			get_bh(new_dir_bh);
1090 		} else {
1091 			mlog(ML_ERROR, "no old_dir_bh!\n");
1092 			status = -EIO;
1093 			goto bail;
1094 		}
1095 	}
1096 
1097 	if (S_ISDIR(old_inode->i_mode)) {
1098 		/* Directories actually require metadata updates to
1099 		 * the directory info so we can't get away with not
1100 		 * doing node locking on it. */
1101 		status = ocfs2_meta_lock(old_inode, handle, NULL, 1);
1102 		if (status < 0) {
1103 			if (status != -ENOENT)
1104 				mlog_errno(status);
1105 			goto bail;
1106 		}
1107 
1108 		status = ocfs2_request_rename_vote(old_inode, old_dentry);
1109 		if (status < 0) {
1110 			mlog_errno(status);
1111 			goto bail;
1112 		}
1113 
1114 		status = -EIO;
1115 		old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1116 		if (!old_inode_de_bh)
1117 			goto bail;
1118 
1119 		status = -EIO;
1120 		if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1121 		    OCFS2_I(old_dir)->ip_blkno)
1122 			goto bail;
1123 		status = -EMLINK;
1124 		if (!new_inode && new_dir!=old_dir &&
1125 		    new_dir->i_nlink >= OCFS2_LINK_MAX)
1126 			goto bail;
1127 	} else {
1128 		/* Ah, the simple case - we're a file so just send a
1129 		 * message. */
1130 		status = ocfs2_request_rename_vote(old_inode, old_dentry);
1131 		if (status < 0) {
1132 			mlog_errno(status);
1133 			goto bail;
1134 		}
1135 	}
1136 
1137 	status = -ENOENT;
1138 	old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1139 				     old_dentry->d_name.len,
1140 				     old_dir, &old_de);
1141 	if (!old_de_bh)
1142 		goto bail;
1143 
1144 	/*
1145 	 *  Check for inode number is _not_ due to possible IO errors.
1146 	 *  We might rmdir the source, keep it as pwd of some process
1147 	 *  and merrily kill the link to whatever was created under the
1148 	 *  same name. Goodbye sticky bit ;-<
1149 	 */
1150 	if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1151 		goto bail;
1152 
1153 	/* check if the target already exists (in which case we need
1154 	 * to delete it */
1155 	status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1156 					  new_dentry->d_name.len,
1157 					  &newfe_blkno, new_dir, &new_de_bh,
1158 					  &new_de);
1159 	/* The only error we allow here is -ENOENT because the new
1160 	 * file not existing is perfectly valid. */
1161 	if ((status < 0) && (status != -ENOENT)) {
1162 		/* If we cannot find the file specified we should just */
1163 		/* return the error... */
1164 		mlog_errno(status);
1165 		goto bail;
1166 	}
1167 
1168 	if (!new_de && new_inode)
1169 		mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1170 		     "directory!", new_inode->i_ino);
1171 
1172 	/* In case we need to overwrite an existing file, we blow it
1173 	 * away first */
1174 	if (new_de) {
1175 		/* VFS didn't think there existed an inode here, but
1176 		 * someone else in the cluster must have raced our
1177 		 * rename to create one. Today we error cleanly, in
1178 		 * the future we should consider calling iget to build
1179 		 * a new struct inode for this entry. */
1180 		if (!new_inode) {
1181 			status = -EACCES;
1182 
1183 			mlog(0, "We found an inode for name %.*s but VFS "
1184 			     "didn't give us one.\n", new_dentry->d_name.len,
1185 			     new_dentry->d_name.name);
1186 			goto bail;
1187 		}
1188 
1189 		if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1190 			status = -EACCES;
1191 
1192 			mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1193 			     (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1194 			     (unsigned long long)newfe_blkno,
1195 			     OCFS2_I(new_inode)->ip_flags);
1196 			goto bail;
1197 		}
1198 
1199 		status = ocfs2_meta_lock(new_inode, handle, &newfe_bh, 1);
1200 		if (status < 0) {
1201 			if (status != -ENOENT)
1202 				mlog_errno(status);
1203 			goto bail;
1204 		}
1205 
1206 		if (S_ISDIR(new_inode->i_mode))
1207 			links_count = 0;
1208 		else
1209 			links_count = (unsigned int) (new_inode->i_nlink - 1);
1210 
1211 		status = ocfs2_request_unlink_vote(new_inode, new_dentry,
1212 						   links_count);
1213 		if (status < 0) {
1214 			mlog_errno(status);
1215 			goto bail;
1216 		}
1217 
1218 		newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1219 
1220 		mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1221 		     "newfebh=%p bhblocknr=%llu\n", new_de,
1222 		     (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1223 		     (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1224 
1225 		if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1226 			status = ocfs2_prepare_orphan_dir(osb, handle,
1227 							  new_inode,
1228 							  orphan_name,
1229 							  &orphan_entry_bh);
1230 			if (status < 0) {
1231 				mlog_errno(status);
1232 				goto bail;
1233 			}
1234 		}
1235 	} else {
1236 		BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1237 
1238 		status = ocfs2_check_dir_for_entry(new_dir,
1239 						   new_dentry->d_name.name,
1240 						   new_dentry->d_name.len);
1241 		if (status)
1242 			goto bail;
1243 
1244 		status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1245 						      new_dentry->d_name.name,
1246 						      new_dentry->d_name.len,
1247 						      &insert_entry_bh);
1248 		if (status < 0) {
1249 			mlog_errno(status);
1250 			goto bail;
1251 		}
1252 	}
1253 
1254 	handle = ocfs2_start_trans(osb, handle, OCFS2_RENAME_CREDITS);
1255 	if (IS_ERR(handle)) {
1256 		status = PTR_ERR(handle);
1257 		handle = NULL;
1258 		mlog_errno(status);
1259 		goto bail;
1260 	}
1261 
1262 	if (new_de) {
1263 		if (S_ISDIR(new_inode->i_mode)) {
1264 			if (!ocfs2_empty_dir(new_inode) ||
1265 			    new_inode->i_nlink != 2) {
1266 				status = -ENOTEMPTY;
1267 				goto bail;
1268 			}
1269 		}
1270 		status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1271 					      OCFS2_JOURNAL_ACCESS_WRITE);
1272 		if (status < 0) {
1273 			mlog_errno(status);
1274 			goto bail;
1275 		}
1276 
1277 		if (S_ISDIR(new_inode->i_mode) ||
1278 		    (newfe->i_links_count == cpu_to_le16(1))){
1279 			status = ocfs2_orphan_add(osb, handle, new_inode,
1280 						  newfe, orphan_name,
1281 						  orphan_entry_bh);
1282 			if (status < 0) {
1283 				mlog_errno(status);
1284 				goto bail;
1285 			}
1286 		}
1287 
1288 		/* change the dirent to point to the correct inode */
1289 		status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1290 					      OCFS2_JOURNAL_ACCESS_WRITE);
1291 		if (status < 0) {
1292 			mlog_errno(status);
1293 			goto bail;
1294 		}
1295 		new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1296 		new_de->file_type = old_de->file_type;
1297 		new_dir->i_version++;
1298 		status = ocfs2_journal_dirty(handle, new_de_bh);
1299 		if (status < 0) {
1300 			mlog_errno(status);
1301 			goto bail;
1302 		}
1303 
1304 		if (S_ISDIR(new_inode->i_mode))
1305 			newfe->i_links_count = 0;
1306 		else
1307 			le16_add_cpu(&newfe->i_links_count, -1);
1308 
1309 		status = ocfs2_journal_dirty(handle, newfe_bh);
1310 		if (status < 0) {
1311 			mlog_errno(status);
1312 			goto bail;
1313 		}
1314 	} else {
1315 		/* if the name was not found in new_dir, add it now */
1316 		status = ocfs2_add_entry(handle, new_dentry, old_inode,
1317 					 OCFS2_I(old_inode)->ip_blkno,
1318 					 new_dir_bh, insert_entry_bh);
1319 	}
1320 
1321 	old_inode->i_ctime = CURRENT_TIME;
1322 	mark_inode_dirty(old_inode);
1323 
1324 	/* now that the name has been added to new_dir, remove the old name */
1325 	status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1326 	if (status < 0) {
1327 		mlog_errno(status);
1328 		goto bail;
1329 	}
1330 
1331 	if (new_inode) {
1332 		new_inode->i_nlink--;
1333 		new_inode->i_ctime = CURRENT_TIME;
1334 	}
1335 	old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1336 	if (old_inode_de_bh) {
1337 		status = ocfs2_journal_access(handle, old_inode,
1338 					     old_inode_de_bh,
1339 					     OCFS2_JOURNAL_ACCESS_WRITE);
1340 		PARENT_INO(old_inode_de_bh->b_data) =
1341 			cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1342 		status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1343 		old_dir->i_nlink--;
1344 		if (new_inode) {
1345 			new_inode->i_nlink--;
1346 		} else {
1347 			new_dir->i_nlink++;
1348 			mark_inode_dirty(new_dir);
1349 		}
1350 	}
1351 	mark_inode_dirty(old_dir);
1352 	if (new_inode)
1353 		mark_inode_dirty(new_inode);
1354 
1355 	if (old_dir != new_dir)
1356 		if (new_dir_nlink != new_dir->i_nlink) {
1357 			if (!new_dir_bh) {
1358 				mlog(ML_ERROR, "need to change nlink for new "
1359 				     "dir %llu from %d to %d but bh is NULL\n",
1360 				     (unsigned long long)OCFS2_I(new_dir)->ip_blkno,
1361 				     (int)new_dir_nlink, new_dir->i_nlink);
1362 			} else {
1363 				struct ocfs2_dinode *fe;
1364 				status = ocfs2_journal_access(handle,
1365 							      new_dir,
1366 							      new_dir_bh,
1367 							      OCFS2_JOURNAL_ACCESS_WRITE);
1368 				fe = (struct ocfs2_dinode *) new_dir_bh->b_data;
1369 				fe->i_links_count = cpu_to_le16(new_dir->i_nlink);
1370 				status = ocfs2_journal_dirty(handle, new_dir_bh);
1371 			}
1372 		}
1373 
1374 	if (old_dir_nlink != old_dir->i_nlink) {
1375 		if (!old_dir_bh) {
1376 			mlog(ML_ERROR, "need to change nlink for old dir "
1377 			     "%llu from %d to %d but bh is NULL!\n",
1378 			     (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1379 			     (int)old_dir_nlink, old_dir->i_nlink);
1380 		} else {
1381 			struct ocfs2_dinode *fe;
1382 			status = ocfs2_journal_access(handle, old_dir,
1383 						      old_dir_bh,
1384 						      OCFS2_JOURNAL_ACCESS_WRITE);
1385 			fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1386 			fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1387 			status = ocfs2_journal_dirty(handle, old_dir_bh);
1388 		}
1389 	}
1390 
1391 	status = 0;
1392 bail:
1393 	if (rename_lock)
1394 		ocfs2_rename_unlock(osb);
1395 
1396 	if (handle)
1397 		ocfs2_commit_trans(handle);
1398 
1399 	if (new_inode)
1400 		sync_mapping_buffers(old_inode->i_mapping);
1401 
1402 	if (new_inode)
1403 		iput(new_inode);
1404 	if (newfe_bh)
1405 		brelse(newfe_bh);
1406 	if (old_dir_bh)
1407 		brelse(old_dir_bh);
1408 	if (new_dir_bh)
1409 		brelse(new_dir_bh);
1410 	if (new_de_bh)
1411 		brelse(new_de_bh);
1412 	if (old_de_bh)
1413 		brelse(old_de_bh);
1414 	if (old_inode_de_bh)
1415 		brelse(old_inode_de_bh);
1416 	if (orphan_entry_bh)
1417 		brelse(orphan_entry_bh);
1418 	if (insert_entry_bh)
1419 		brelse(insert_entry_bh);
1420 
1421 	mlog_exit(status);
1422 
1423 	return status;
1424 }
1425 
1426 /*
1427  * we expect i_size = strlen(symname). Copy symname into the file
1428  * data, including the null terminator.
1429  */
1430 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1431 				     struct ocfs2_journal_handle *handle,
1432 				     struct inode *inode,
1433 				     const char *symname)
1434 {
1435 	struct buffer_head **bhs = NULL;
1436 	const char *c;
1437 	struct super_block *sb = osb->sb;
1438 	u64 p_blkno;
1439 	int p_blocks;
1440 	int virtual, blocks, status, i, bytes_left;
1441 
1442 	bytes_left = i_size_read(inode) + 1;
1443 	/* we can't trust i_blocks because we're actually going to
1444 	 * write i_size + 1 bytes. */
1445 	blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1446 
1447 	mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1448 			(unsigned long long)inode->i_blocks,
1449 			i_size_read(inode), blocks);
1450 
1451 	/* Sanity check -- make sure we're going to fit. */
1452 	if (bytes_left >
1453 	    ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1454 		status = -EIO;
1455 		mlog_errno(status);
1456 		goto bail;
1457 	}
1458 
1459 	bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1460 	if (!bhs) {
1461 		status = -ENOMEM;
1462 		mlog_errno(status);
1463 		goto bail;
1464 	}
1465 
1466 	status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1467 					     &p_blocks);
1468 	if (status < 0) {
1469 		mlog_errno(status);
1470 		goto bail;
1471 	}
1472 
1473 	/* links can never be larger than one cluster so we know this
1474 	 * is all going to be contiguous, but do a sanity check
1475 	 * anyway. */
1476 	if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1477 		status = -EIO;
1478 		mlog_errno(status);
1479 		goto bail;
1480 	}
1481 
1482 	virtual = 0;
1483 	while(bytes_left > 0) {
1484 		c = &symname[virtual * sb->s_blocksize];
1485 
1486 		bhs[virtual] = sb_getblk(sb, p_blkno);
1487 		if (!bhs[virtual]) {
1488 			status = -ENOMEM;
1489 			mlog_errno(status);
1490 			goto bail;
1491 		}
1492 		ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1493 
1494 		status = ocfs2_journal_access(handle, inode, bhs[virtual],
1495 					      OCFS2_JOURNAL_ACCESS_CREATE);
1496 		if (status < 0) {
1497 			mlog_errno(status);
1498 			goto bail;
1499 		}
1500 
1501 		memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1502 
1503 		memcpy(bhs[virtual]->b_data, c,
1504 		       (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1505 		       bytes_left);
1506 
1507 		status = ocfs2_journal_dirty(handle, bhs[virtual]);
1508 		if (status < 0) {
1509 			mlog_errno(status);
1510 			goto bail;
1511 		}
1512 
1513 		virtual++;
1514 		p_blkno++;
1515 		bytes_left -= sb->s_blocksize;
1516 	}
1517 
1518 	status = 0;
1519 bail:
1520 
1521 	if (bhs) {
1522 		for(i = 0; i < blocks; i++)
1523 			if (bhs[i])
1524 				brelse(bhs[i]);
1525 		kfree(bhs);
1526 	}
1527 
1528 	mlog_exit(status);
1529 	return status;
1530 }
1531 
1532 static int ocfs2_symlink(struct inode *dir,
1533 			 struct dentry *dentry,
1534 			 const char *symname)
1535 {
1536 	int status, l, credits;
1537 	u64 newsize;
1538 	struct ocfs2_super *osb = NULL;
1539 	struct inode *inode = NULL;
1540 	struct super_block *sb;
1541 	struct buffer_head *new_fe_bh = NULL;
1542 	struct buffer_head *de_bh = NULL;
1543 	struct buffer_head *parent_fe_bh = NULL;
1544 	struct ocfs2_dinode *fe = NULL;
1545 	struct ocfs2_dinode *dirfe;
1546 	struct ocfs2_journal_handle *handle = NULL;
1547 	struct ocfs2_alloc_context *inode_ac = NULL;
1548 	struct ocfs2_alloc_context *data_ac = NULL;
1549 
1550 	mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1551 		   dentry, symname, dentry->d_name.len, dentry->d_name.name);
1552 
1553 	sb = dir->i_sb;
1554 	osb = OCFS2_SB(sb);
1555 
1556 	l = strlen(symname) + 1;
1557 
1558 	credits = ocfs2_calc_symlink_credits(sb);
1559 
1560 	handle = ocfs2_alloc_handle(osb);
1561 	if (handle == NULL) {
1562 		status = -ENOMEM;
1563 		mlog_errno(status);
1564 		goto bail;
1565 	}
1566 
1567 	/* lock the parent directory */
1568 	status = ocfs2_meta_lock(dir, handle, &parent_fe_bh, 1);
1569 	if (status < 0) {
1570 		if (status != -ENOENT)
1571 			mlog_errno(status);
1572 		goto bail;
1573 	}
1574 
1575 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1576 	if (!dirfe->i_links_count) {
1577 		/* can't make a file in a deleted directory. */
1578 		status = -ENOENT;
1579 		goto bail;
1580 	}
1581 
1582 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1583 					   dentry->d_name.len);
1584 	if (status)
1585 		goto bail;
1586 
1587 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1588 					      dentry->d_name.name,
1589 					      dentry->d_name.len, &de_bh);
1590 	if (status < 0) {
1591 		mlog_errno(status);
1592 		goto bail;
1593 	}
1594 
1595 	status = ocfs2_reserve_new_inode(osb, handle, &inode_ac);
1596 	if (status < 0) {
1597 		if (status != -ENOSPC)
1598 			mlog_errno(status);
1599 		goto bail;
1600 	}
1601 
1602 	/* don't reserve bitmap space for fast symlinks. */
1603 	if (l > ocfs2_fast_symlink_chars(sb)) {
1604 		status = ocfs2_reserve_clusters(osb, handle, 1, &data_ac);
1605 		if (status < 0) {
1606 			if (status != -ENOSPC)
1607 				mlog_errno(status);
1608 			goto bail;
1609 		}
1610 	}
1611 
1612 	handle = ocfs2_start_trans(osb, handle, credits);
1613 	if (IS_ERR(handle)) {
1614 		status = PTR_ERR(handle);
1615 		handle = NULL;
1616 		mlog_errno(status);
1617 		goto bail;
1618 	}
1619 
1620 	status = ocfs2_mknod_locked(osb, dir, dentry,
1621 				    S_IFLNK | S_IRWXUGO, 0,
1622 				    &new_fe_bh, parent_fe_bh, handle,
1623 				    &inode, inode_ac);
1624 	if (status < 0) {
1625 		mlog_errno(status);
1626 		goto bail;
1627 	}
1628 
1629 	fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1630 	inode->i_rdev = 0;
1631 	newsize = l - 1;
1632 	if (l > ocfs2_fast_symlink_chars(sb)) {
1633 		inode->i_op = &ocfs2_symlink_inode_operations;
1634 		status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1635 						    handle, data_ac, NULL,
1636 						    NULL);
1637 		if (status < 0) {
1638 			if (status != -ENOSPC && status != -EINTR) {
1639 				mlog(ML_ERROR,
1640 				     "Failed to extend file to %llu\n",
1641 				     (unsigned long long)newsize);
1642 				mlog_errno(status);
1643 				status = -ENOSPC;
1644 			}
1645 			goto bail;
1646 		}
1647 		i_size_write(inode, newsize);
1648 		inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1649 	} else {
1650 		inode->i_op = &ocfs2_fast_symlink_inode_operations;
1651 		memcpy((char *) fe->id2.i_symlink, symname, l);
1652 		i_size_write(inode, newsize);
1653 		inode->i_blocks = 0;
1654 	}
1655 
1656 	status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1657 	if (status < 0) {
1658 		mlog_errno(status);
1659 		goto bail;
1660 	}
1661 
1662 	if (!ocfs2_inode_is_fast_symlink(inode)) {
1663 		status = ocfs2_create_symlink_data(osb, handle, inode,
1664 						   symname);
1665 		if (status < 0) {
1666 			mlog_errno(status);
1667 			goto bail;
1668 		}
1669 	}
1670 
1671 	status = ocfs2_add_entry(handle, dentry, inode,
1672 				 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1673 				 de_bh);
1674 	if (status < 0) {
1675 		mlog_errno(status);
1676 		goto bail;
1677 	}
1678 
1679 	insert_inode_hash(inode);
1680 	dentry->d_op = &ocfs2_dentry_ops;
1681 	d_instantiate(dentry, inode);
1682 bail:
1683 	if (handle)
1684 		ocfs2_commit_trans(handle);
1685 	if (new_fe_bh)
1686 		brelse(new_fe_bh);
1687 	if (parent_fe_bh)
1688 		brelse(parent_fe_bh);
1689 	if (de_bh)
1690 		brelse(de_bh);
1691 	if (inode_ac)
1692 		ocfs2_free_alloc_context(inode_ac);
1693 	if (data_ac)
1694 		ocfs2_free_alloc_context(data_ac);
1695 	if ((status < 0) && inode)
1696 		iput(inode);
1697 
1698 	mlog_exit(status);
1699 
1700 	return status;
1701 }
1702 
1703 int ocfs2_check_dir_entry(struct inode * dir,
1704 			  struct ocfs2_dir_entry * de,
1705 			  struct buffer_head * bh,
1706 			  unsigned long offset)
1707 {
1708 	const char *error_msg = NULL;
1709 	const int rlen = le16_to_cpu(de->rec_len);
1710 
1711 	if (rlen < OCFS2_DIR_REC_LEN(1))
1712 		error_msg = "rec_len is smaller than minimal";
1713 	else if (rlen % 4 != 0)
1714 		error_msg = "rec_len % 4 != 0";
1715 	else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1716 		error_msg = "rec_len is too small for name_len";
1717 	else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1718 		error_msg = "directory entry across blocks";
1719 
1720 	if (error_msg != NULL)
1721 		mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1722 		     "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1723 		     (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1724 		     offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1725 		     de->name_len);
1726 	return error_msg == NULL ? 1 : 0;
1727 }
1728 
1729 /* we don't always have a dentry for what we want to add, so people
1730  * like orphan dir can call this instead.
1731  *
1732  * If you pass me insert_bh, I'll skip the search of the other dir
1733  * blocks and put the record in there.
1734  */
1735 static int __ocfs2_add_entry(struct ocfs2_journal_handle *handle,
1736 			     struct inode *dir,
1737 			     const char *name, int namelen,
1738 			     struct inode *inode, u64 blkno,
1739 			     struct buffer_head *parent_fe_bh,
1740 			     struct buffer_head *insert_bh)
1741 {
1742 	unsigned long offset;
1743 	unsigned short rec_len;
1744 	struct ocfs2_dir_entry *de, *de1;
1745 	struct super_block *sb;
1746 	int retval, status;
1747 
1748 	mlog_entry_void();
1749 
1750 	sb = dir->i_sb;
1751 
1752 	if (!namelen)
1753 		return -EINVAL;
1754 
1755 	rec_len = OCFS2_DIR_REC_LEN(namelen);
1756 	offset = 0;
1757 	de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1758 	while (1) {
1759 		BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1760 		/* These checks should've already been passed by the
1761 		 * prepare function, but I guess we can leave them
1762 		 * here anyway. */
1763 		if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1764 			retval = -ENOENT;
1765 			goto bail;
1766 		}
1767 		if (ocfs2_match(namelen, name, de)) {
1768 			retval = -EEXIST;
1769 			goto bail;
1770 		}
1771 		if (((le64_to_cpu(de->inode) == 0) &&
1772 		     (le16_to_cpu(de->rec_len) >= rec_len)) ||
1773 		    (le16_to_cpu(de->rec_len) >=
1774 		     (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1775 			status = ocfs2_journal_access(handle, dir, insert_bh,
1776 						      OCFS2_JOURNAL_ACCESS_WRITE);
1777 			/* By now the buffer is marked for journaling */
1778 			offset += le16_to_cpu(de->rec_len);
1779 			if (le64_to_cpu(de->inode)) {
1780 				de1 = (struct ocfs2_dir_entry *)((char *) de +
1781 					OCFS2_DIR_REC_LEN(de->name_len));
1782 				de1->rec_len =
1783 					cpu_to_le16(le16_to_cpu(de->rec_len) -
1784 					OCFS2_DIR_REC_LEN(de->name_len));
1785 				de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1786 				de = de1;
1787 			}
1788 			de->file_type = OCFS2_FT_UNKNOWN;
1789 			if (blkno) {
1790 				de->inode = cpu_to_le64(blkno);
1791 				ocfs2_set_de_type(de, inode->i_mode);
1792 			} else
1793 				de->inode = 0;
1794 			de->name_len = namelen;
1795 			memcpy(de->name, name, namelen);
1796 
1797 			dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1798 			dir->i_version++;
1799 			status = ocfs2_journal_dirty(handle, insert_bh);
1800 			retval = 0;
1801 			goto bail;
1802 		}
1803 		offset += le16_to_cpu(de->rec_len);
1804 		de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1805 	}
1806 
1807 	/* when you think about it, the assert above should prevent us
1808 	 * from ever getting here. */
1809 	retval = -ENOSPC;
1810 bail:
1811 
1812 	mlog_exit(retval);
1813 	return retval;
1814 }
1815 
1816 
1817 /*
1818  * ocfs2_delete_entry deletes a directory entry by merging it with the
1819  * previous entry
1820  */
1821 static int ocfs2_delete_entry(struct ocfs2_journal_handle *handle,
1822 			      struct inode *dir,
1823 			      struct ocfs2_dir_entry *de_del,
1824 			      struct buffer_head *bh)
1825 {
1826 	struct ocfs2_dir_entry *de, *pde;
1827 	int i, status = -ENOENT;
1828 
1829 	mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1830 
1831 	i = 0;
1832 	pde = NULL;
1833 	de = (struct ocfs2_dir_entry *) bh->b_data;
1834 	while (i < bh->b_size) {
1835 		if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1836 			status = -EIO;
1837 			mlog_errno(status);
1838 			goto bail;
1839 		}
1840 		if (de == de_del)  {
1841 			status = ocfs2_journal_access(handle, dir, bh,
1842 						      OCFS2_JOURNAL_ACCESS_WRITE);
1843 			if (status < 0) {
1844 				status = -EIO;
1845 				mlog_errno(status);
1846 				goto bail;
1847 			}
1848 			if (pde)
1849 				pde->rec_len =
1850 					cpu_to_le16(le16_to_cpu(pde->rec_len) +
1851 						    le16_to_cpu(de->rec_len));
1852 			else
1853 				de->inode = 0;
1854 			dir->i_version++;
1855 			status = ocfs2_journal_dirty(handle, bh);
1856 			goto bail;
1857 		}
1858 		i += le16_to_cpu(de->rec_len);
1859 		pde = de;
1860 		de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1861 	}
1862 bail:
1863 	mlog_exit(status);
1864 	return status;
1865 }
1866 
1867 /*
1868  * Returns 0 if not found, -1 on failure, and 1 on success
1869  */
1870 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1871 					struct inode *dir,
1872 					const char *name, int namelen,
1873 					unsigned long offset,
1874 					struct ocfs2_dir_entry **res_dir)
1875 {
1876 	struct ocfs2_dir_entry *de;
1877 	char *dlimit, *de_buf;
1878 	int de_len;
1879 	int ret = 0;
1880 
1881 	mlog_entry_void();
1882 
1883 	de_buf = bh->b_data;
1884 	dlimit = de_buf + dir->i_sb->s_blocksize;
1885 
1886 	while (de_buf < dlimit) {
1887 		/* this code is executed quadratically often */
1888 		/* do minimal checking `by hand' */
1889 
1890 		de = (struct ocfs2_dir_entry *) de_buf;
1891 
1892 		if (de_buf + namelen <= dlimit &&
1893 		    ocfs2_match(namelen, name, de)) {
1894 			/* found a match - just to be sure, do a full check */
1895 			if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1896 				ret = -1;
1897 				goto bail;
1898 			}
1899 			*res_dir = de;
1900 			ret = 1;
1901 			goto bail;
1902 		}
1903 
1904 		/* prevent looping on a bad block */
1905 		de_len = le16_to_cpu(de->rec_len);
1906 		if (de_len <= 0) {
1907 			ret = -1;
1908 			goto bail;
1909 		}
1910 
1911 		de_buf += de_len;
1912 		offset += de_len;
1913 	}
1914 
1915 bail:
1916 	mlog_exit(ret);
1917 	return ret;
1918 }
1919 
1920 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1921 				     struct inode *dir,
1922 				     struct ocfs2_dir_entry **res_dir)
1923 {
1924 	struct super_block *sb;
1925 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1926 	struct buffer_head *bh, *ret = NULL;
1927 	unsigned long start, block, b;
1928 	int ra_max = 0;		/* Number of bh's in the readahead
1929 				   buffer, bh_use[] */
1930 	int ra_ptr = 0;		/* Current index into readahead
1931 				   buffer */
1932 	int num = 0;
1933 	int nblocks, i, err;
1934 
1935 	mlog_entry_void();
1936 
1937 	*res_dir = NULL;
1938 	sb = dir->i_sb;
1939 
1940 	nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
1941 	start = OCFS2_I(dir)->ip_dir_start_lookup;
1942 	if (start >= nblocks)
1943 		start = 0;
1944 	block = start;
1945 
1946 restart:
1947 	do {
1948 		/*
1949 		 * We deal with the read-ahead logic here.
1950 		 */
1951 		if (ra_ptr >= ra_max) {
1952 			/* Refill the readahead buffer */
1953 			ra_ptr = 0;
1954 			b = block;
1955 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
1956 				/*
1957 				 * Terminate if we reach the end of the
1958 				 * directory and must wrap, or if our
1959 				 * search has finished at this block.
1960 				 */
1961 				if (b >= nblocks || (num && block == start)) {
1962 					bh_use[ra_max] = NULL;
1963 					break;
1964 				}
1965 				num++;
1966 
1967 				/* XXX: questionable readahead stuff here */
1968 				bh = ocfs2_bread(dir, b++, &err, 1);
1969 				bh_use[ra_max] = bh;
1970 #if 0		// ???
1971 				if (bh)
1972 					ll_rw_block(READ, 1, &bh);
1973 #endif
1974 			}
1975 		}
1976 		if ((bh = bh_use[ra_ptr++]) == NULL)
1977 			goto next;
1978 		wait_on_buffer(bh);
1979 		if (!buffer_uptodate(bh)) {
1980 			/* read error, skip block & hope for the best */
1981 			brelse(bh);
1982 			goto next;
1983 		}
1984 		i = ocfs2_search_dirblock(bh, dir, name, namelen,
1985 					  block << sb->s_blocksize_bits,
1986 					  res_dir);
1987 		if (i == 1) {
1988 			OCFS2_I(dir)->ip_dir_start_lookup = block;
1989 			ret = bh;
1990 			goto cleanup_and_exit;
1991 		} else {
1992 			brelse(bh);
1993 			if (i < 0)
1994 				goto cleanup_and_exit;
1995 		}
1996 	next:
1997 		if (++block >= nblocks)
1998 			block = 0;
1999 	} while (block != start);
2000 
2001 	/*
2002 	 * If the directory has grown while we were searching, then
2003 	 * search the last part of the directory before giving up.
2004 	 */
2005 	block = nblocks;
2006 	nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2007 	if (block < nblocks) {
2008 		start = 0;
2009 		goto restart;
2010 	}
2011 
2012 cleanup_and_exit:
2013 	/* Clean up the read-ahead blocks */
2014 	for (; ra_ptr < ra_max; ra_ptr++)
2015 		brelse(bh_use[ra_ptr]);
2016 
2017 	mlog_exit_ptr(ret);
2018 	return ret;
2019 }
2020 
2021 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2022 {
2023 	int status, namelen;
2024 
2025 	mlog_entry_void();
2026 
2027 	namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2028 			   (long long)blkno);
2029 	if (namelen <= 0) {
2030 		if (namelen)
2031 			status = namelen;
2032 		else
2033 			status = -EINVAL;
2034 		mlog_errno(status);
2035 		goto bail;
2036 	}
2037 	if (namelen != OCFS2_ORPHAN_NAMELEN) {
2038 		status = -EINVAL;
2039 		mlog_errno(status);
2040 		goto bail;
2041 	}
2042 
2043 	mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2044 	     namelen);
2045 
2046 	status = 0;
2047 bail:
2048 	mlog_exit(status);
2049 	return status;
2050 }
2051 
2052 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2053 				    struct ocfs2_journal_handle *handle,
2054 				    struct inode *inode,
2055 				    char *name,
2056 				    struct buffer_head **de_bh)
2057 {
2058 	struct inode *orphan_dir_inode = NULL;
2059 	struct buffer_head *orphan_dir_bh = NULL;
2060 	int status = 0;
2061 
2062 	status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2063 	if (status < 0) {
2064 		mlog_errno(status);
2065 		goto leave;
2066 	}
2067 
2068 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2069 						       ORPHAN_DIR_SYSTEM_INODE,
2070 						       osb->slot_num);
2071 	if (!orphan_dir_inode) {
2072 		status = -ENOENT;
2073 		mlog_errno(status);
2074 		goto leave;
2075 	}
2076 
2077 	ocfs2_handle_add_inode(handle, orphan_dir_inode);
2078 	status = ocfs2_meta_lock(orphan_dir_inode, handle, &orphan_dir_bh, 1);
2079 	if (status < 0) {
2080 		mlog_errno(status);
2081 		goto leave;
2082 	}
2083 
2084 	status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2085 					      orphan_dir_bh, name,
2086 					      OCFS2_ORPHAN_NAMELEN, de_bh);
2087 	if (status < 0) {
2088 		mlog_errno(status);
2089 		goto leave;
2090 	}
2091 
2092 leave:
2093 	if (orphan_dir_inode)
2094 		iput(orphan_dir_inode);
2095 
2096 	if (orphan_dir_bh)
2097 		brelse(orphan_dir_bh);
2098 
2099 	mlog_exit(status);
2100 	return status;
2101 }
2102 
2103 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2104 			    struct ocfs2_journal_handle *handle,
2105 			    struct inode *inode,
2106 			    struct ocfs2_dinode *fe,
2107 			    char *name,
2108 			    struct buffer_head *de_bh)
2109 {
2110 	struct inode *orphan_dir_inode = NULL;
2111 	struct buffer_head *orphan_dir_bh = NULL;
2112 	int status = 0;
2113 	struct ocfs2_dinode *orphan_fe;
2114 
2115 	mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2116 
2117 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2118 						       ORPHAN_DIR_SYSTEM_INODE,
2119 						       osb->slot_num);
2120 	if (!orphan_dir_inode) {
2121 		status = -ENOENT;
2122 		mlog_errno(status);
2123 		goto leave;
2124 	}
2125 
2126 	status = ocfs2_read_block(osb,
2127 				  OCFS2_I(orphan_dir_inode)->ip_blkno,
2128 				  &orphan_dir_bh, OCFS2_BH_CACHED,
2129 				  orphan_dir_inode);
2130 	if (status < 0) {
2131 		mlog_errno(status);
2132 		goto leave;
2133 	}
2134 
2135 	status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2136 				      OCFS2_JOURNAL_ACCESS_WRITE);
2137 	if (status < 0) {
2138 		mlog_errno(status);
2139 		goto leave;
2140 	}
2141 
2142 	/* we're a cluster, and nlink can change on disk from
2143 	 * underneath us... */
2144 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2145 	if (S_ISDIR(inode->i_mode))
2146 		le16_add_cpu(&orphan_fe->i_links_count, 1);
2147 	orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2148 
2149 	status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2150 	if (status < 0) {
2151 		mlog_errno(status);
2152 		goto leave;
2153 	}
2154 
2155 	status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2156 				   OCFS2_ORPHAN_NAMELEN, inode,
2157 				   OCFS2_I(inode)->ip_blkno,
2158 				   orphan_dir_bh, de_bh);
2159 	if (status < 0) {
2160 		mlog_errno(status);
2161 		goto leave;
2162 	}
2163 
2164 	le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2165 
2166 	/* Record which orphan dir our inode now resides
2167 	 * in. delete_inode will use this to determine which orphan
2168 	 * dir to lock. */
2169 	spin_lock(&OCFS2_I(inode)->ip_lock);
2170 	OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2171 	spin_unlock(&OCFS2_I(inode)->ip_lock);
2172 
2173 	mlog(0, "Inode %llu orphaned in slot %d\n",
2174 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2175 
2176 leave:
2177 	if (orphan_dir_inode)
2178 		iput(orphan_dir_inode);
2179 
2180 	if (orphan_dir_bh)
2181 		brelse(orphan_dir_bh);
2182 
2183 	mlog_exit(status);
2184 	return status;
2185 }
2186 
2187 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2188 int ocfs2_orphan_del(struct ocfs2_super *osb,
2189 		     struct ocfs2_journal_handle *handle,
2190 		     struct inode *orphan_dir_inode,
2191 		     struct inode *inode,
2192 		     struct buffer_head *orphan_dir_bh)
2193 {
2194 	char name[OCFS2_ORPHAN_NAMELEN + 1];
2195 	struct ocfs2_dinode *orphan_fe;
2196 	int status = 0;
2197 	struct buffer_head *target_de_bh = NULL;
2198 	struct ocfs2_dir_entry *target_de = NULL;
2199 
2200 	mlog_entry_void();
2201 
2202 	status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2203 	if (status < 0) {
2204 		mlog_errno(status);
2205 		goto leave;
2206 	}
2207 
2208 	mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2209 	     name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2210 	     OCFS2_ORPHAN_NAMELEN);
2211 
2212 	/* find it's spot in the orphan directory */
2213 	target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2214 					orphan_dir_inode, &target_de);
2215 	if (!target_de_bh) {
2216 		status = -ENOENT;
2217 		mlog_errno(status);
2218 		goto leave;
2219 	}
2220 
2221 	/* remove it from the orphan directory */
2222 	status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2223 				    target_de_bh);
2224 	if (status < 0) {
2225 		mlog_errno(status);
2226 		goto leave;
2227 	}
2228 
2229 	status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2230 				      OCFS2_JOURNAL_ACCESS_WRITE);
2231 	if (status < 0) {
2232 		mlog_errno(status);
2233 		goto leave;
2234 	}
2235 
2236 	/* do the i_nlink dance! :) */
2237 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2238 	if (S_ISDIR(inode->i_mode))
2239 		le16_add_cpu(&orphan_fe->i_links_count, -1);
2240 	orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2241 
2242 	status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2243 	if (status < 0) {
2244 		mlog_errno(status);
2245 		goto leave;
2246 	}
2247 
2248 leave:
2249 	if (target_de_bh)
2250 		brelse(target_de_bh);
2251 
2252 	mlog_exit(status);
2253 	return status;
2254 }
2255 
2256 struct inode_operations ocfs2_dir_iops = {
2257 	.create		= ocfs2_create,
2258 	.lookup		= ocfs2_lookup,
2259 	.link		= ocfs2_link,
2260 	.unlink		= ocfs2_unlink,
2261 	.rmdir		= ocfs2_unlink,
2262 	.symlink	= ocfs2_symlink,
2263 	.mkdir		= ocfs2_mkdir,
2264 	.mknod		= ocfs2_mknod,
2265 	.rename		= ocfs2_rename,
2266 	.setattr	= ocfs2_setattr,
2267 	.getattr	= ocfs2_getattr,
2268 };
2269