xref: /linux/fs/ocfs2/namei.c (revision c537b994505099b7197e7d3125b942ecbcc51eb6)
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(handle_t *handle,
79 			      struct inode *dir,
80 			      struct ocfs2_dir_entry *de_del,
81 			      struct buffer_head *bh);
82 
83 static int __ocfs2_add_entry(handle_t *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 			      handle_t *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 			      handle_t *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_prepare_orphan_dir(struct ocfs2_super *osb,
108 				    struct inode **ret_orphan_dir,
109 				    struct inode *inode,
110 				    char *name,
111 				    struct buffer_head **de_bh);
112 
113 static int ocfs2_orphan_add(struct ocfs2_super *osb,
114 			    handle_t *handle,
115 			    struct inode *inode,
116 			    struct ocfs2_dinode *fe,
117 			    char *name,
118 			    struct buffer_head *de_bh,
119 			    struct inode *orphan_dir_inode);
120 
121 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
122 				     handle_t *handle,
123 				     struct inode *inode,
124 				     const char *symname);
125 
126 static inline int ocfs2_add_entry(handle_t *handle,
127 				  struct dentry *dentry,
128 				  struct inode *inode, u64 blkno,
129 				  struct buffer_head *parent_fe_bh,
130 				  struct buffer_head *insert_bh)
131 {
132 	return __ocfs2_add_entry(handle, dentry->d_parent->d_inode,
133 				 dentry->d_name.name, dentry->d_name.len,
134 				 inode, blkno, parent_fe_bh, insert_bh);
135 }
136 
137 /* An orphan dir name is an 8 byte value, printed as a hex string */
138 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
139 
140 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
141 				   struct nameidata *nd)
142 {
143 	int status;
144 	u64 blkno;
145 	struct buffer_head *dirent_bh = NULL;
146 	struct inode *inode = NULL;
147 	struct dentry *ret;
148 	struct ocfs2_dir_entry *dirent;
149 	struct ocfs2_inode_info *oi;
150 
151 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
152 		   dentry->d_name.len, dentry->d_name.name);
153 
154 	if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
155 		ret = ERR_PTR(-ENAMETOOLONG);
156 		goto bail;
157 	}
158 
159 	mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
160 	     dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
161 
162 	status = ocfs2_meta_lock(dir, NULL, 0);
163 	if (status < 0) {
164 		if (status != -ENOENT)
165 			mlog_errno(status);
166 		ret = ERR_PTR(status);
167 		goto bail;
168 	}
169 
170 	status = ocfs2_find_files_on_disk(dentry->d_name.name,
171 					  dentry->d_name.len, &blkno,
172 					  dir, &dirent_bh, &dirent);
173 	if (status < 0)
174 		goto bail_add;
175 
176 	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0);
177 	if (IS_ERR(inode)) {
178 		mlog(ML_ERROR, "Unable to create inode %llu\n",
179 		     (unsigned long long)blkno);
180 		ret = ERR_PTR(-EACCES);
181 		goto bail_unlock;
182 	}
183 
184 	oi = OCFS2_I(inode);
185 	/* Clear any orphaned state... If we were able to look up the
186 	 * inode from a directory, it certainly can't be orphaned. We
187 	 * might have the bad state from a node which intended to
188 	 * orphan this inode but crashed before it could commit the
189 	 * unlink. */
190 	spin_lock(&oi->ip_lock);
191 	oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
192 	oi->ip_orphaned_slot = OCFS2_INVALID_SLOT;
193 	spin_unlock(&oi->ip_lock);
194 
195 bail_add:
196 	dentry->d_op = &ocfs2_dentry_ops;
197 	ret = d_splice_alias(inode, dentry);
198 
199 	if (inode) {
200 		/*
201 		 * If d_splice_alias() finds a DCACHE_DISCONNECTED
202 		 * dentry, it will d_move() it on top of ourse. The
203 		 * return value will indicate this however, so in
204 		 * those cases, we switch them around for the locking
205 		 * code.
206 		 *
207 		 * NOTE: This dentry already has ->d_op set from
208 		 * ocfs2_get_parent() and ocfs2_get_dentry()
209 		 */
210 		if (ret)
211 			dentry = ret;
212 
213 		status = ocfs2_dentry_attach_lock(dentry, inode,
214 						  OCFS2_I(dir)->ip_blkno);
215 		if (status) {
216 			mlog_errno(status);
217 			ret = ERR_PTR(status);
218 			goto bail_unlock;
219 		}
220 	}
221 
222 bail_unlock:
223 	/* Don't drop the cluster lock until *after* the d_add --
224 	 * unlink on another node will message us to remove that
225 	 * dentry under this lock so otherwise we can race this with
226 	 * the vote thread and have a stale dentry. */
227 	ocfs2_meta_unlock(dir, 0);
228 
229 bail:
230 	if (dirent_bh)
231 		brelse(dirent_bh);
232 
233 	mlog_exit_ptr(ret);
234 
235 	return ret;
236 }
237 
238 static int ocfs2_fill_new_dir(struct ocfs2_super *osb,
239 			      handle_t *handle,
240 			      struct inode *parent,
241 			      struct inode *inode,
242 			      struct buffer_head *fe_bh,
243 			      struct ocfs2_alloc_context *data_ac)
244 {
245 	int status;
246 	struct buffer_head *new_bh = NULL;
247 	struct ocfs2_dir_entry *de = NULL;
248 
249 	mlog_entry_void();
250 
251 	status = ocfs2_do_extend_dir(osb->sb, handle, inode, fe_bh,
252 				     data_ac, NULL, &new_bh);
253 	if (status < 0) {
254 		mlog_errno(status);
255 		goto bail;
256 	}
257 
258 	ocfs2_set_new_buffer_uptodate(inode, new_bh);
259 
260 	status = ocfs2_journal_access(handle, inode, new_bh,
261 				      OCFS2_JOURNAL_ACCESS_CREATE);
262 	if (status < 0) {
263 		mlog_errno(status);
264 		goto bail;
265 	}
266 	memset(new_bh->b_data, 0, osb->sb->s_blocksize);
267 
268 	de = (struct ocfs2_dir_entry *) new_bh->b_data;
269 	de->inode = cpu_to_le64(OCFS2_I(inode)->ip_blkno);
270 	de->name_len = 1;
271 	de->rec_len =
272 		cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
273 	strcpy(de->name, ".");
274 	ocfs2_set_de_type(de, S_IFDIR);
275 	de = (struct ocfs2_dir_entry *) ((char *)de + le16_to_cpu(de->rec_len));
276 	de->inode = cpu_to_le64(OCFS2_I(parent)->ip_blkno);
277 	de->rec_len = cpu_to_le16(inode->i_sb->s_blocksize -
278 				  OCFS2_DIR_REC_LEN(1));
279 	de->name_len = 2;
280 	strcpy(de->name, "..");
281 	ocfs2_set_de_type(de, S_IFDIR);
282 
283 	status = ocfs2_journal_dirty(handle, new_bh);
284 	if (status < 0) {
285 		mlog_errno(status);
286 		goto bail;
287 	}
288 
289 	i_size_write(inode, inode->i_sb->s_blocksize);
290 	inode->i_nlink = 2;
291 	inode->i_blocks = ocfs2_align_bytes_to_sectors(inode->i_sb->s_blocksize);
292 	status = ocfs2_mark_inode_dirty(handle, inode, fe_bh);
293 	if (status < 0) {
294 		mlog_errno(status);
295 		goto bail;
296 	}
297 
298 	status = 0;
299 bail:
300 	if (new_bh)
301 		brelse(new_bh);
302 
303 	mlog_exit(status);
304 	return status;
305 }
306 
307 static int ocfs2_mknod(struct inode *dir,
308 		       struct dentry *dentry,
309 		       int mode,
310 		       dev_t dev)
311 {
312 	int status = 0;
313 	struct buffer_head *parent_fe_bh = NULL;
314 	handle_t *handle = NULL;
315 	struct ocfs2_super *osb;
316 	struct ocfs2_dinode *dirfe;
317 	struct buffer_head *new_fe_bh = NULL;
318 	struct buffer_head *de_bh = NULL;
319 	struct inode *inode = NULL;
320 	struct ocfs2_alloc_context *inode_ac = NULL;
321 	struct ocfs2_alloc_context *data_ac = NULL;
322 
323 	mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
324 		   (unsigned long)dev, dentry->d_name.len,
325 		   dentry->d_name.name);
326 
327 	/* get our super block */
328 	osb = OCFS2_SB(dir->i_sb);
329 
330 	status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
331 	if (status < 0) {
332 		if (status != -ENOENT)
333 			mlog_errno(status);
334 		return status;
335 	}
336 
337 	if (S_ISDIR(mode) && (dir->i_nlink >= OCFS2_LINK_MAX)) {
338 		status = -EMLINK;
339 		goto leave;
340 	}
341 
342 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
343 	if (!dirfe->i_links_count) {
344 		/* can't make a file in a deleted directory. */
345 		status = -ENOENT;
346 		goto leave;
347 	}
348 
349 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
350 					   dentry->d_name.len);
351 	if (status)
352 		goto leave;
353 
354 	/* get a spot inside the dir. */
355 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
356 					      dentry->d_name.name,
357 					      dentry->d_name.len, &de_bh);
358 	if (status < 0) {
359 		mlog_errno(status);
360 		goto leave;
361 	}
362 
363 	/* reserve an inode spot */
364 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
365 	if (status < 0) {
366 		if (status != -ENOSPC)
367 			mlog_errno(status);
368 		goto leave;
369 	}
370 
371 	/* are we making a directory? If so, reserve a cluster for his
372 	 * 1st extent. */
373 	if (S_ISDIR(mode)) {
374 		status = ocfs2_reserve_clusters(osb, 1, &data_ac);
375 		if (status < 0) {
376 			if (status != -ENOSPC)
377 				mlog_errno(status);
378 			goto leave;
379 		}
380 	}
381 
382 	handle = ocfs2_start_trans(osb, OCFS2_MKNOD_CREDITS);
383 	if (IS_ERR(handle)) {
384 		status = PTR_ERR(handle);
385 		handle = NULL;
386 		mlog_errno(status);
387 		goto leave;
388 	}
389 
390 	/* do the real work now. */
391 	status = ocfs2_mknod_locked(osb, dir, dentry, mode, dev,
392 				    &new_fe_bh, parent_fe_bh, handle,
393 				    &inode, inode_ac);
394 	if (status < 0) {
395 		mlog_errno(status);
396 		goto leave;
397 	}
398 
399 	if (S_ISDIR(mode)) {
400 		status = ocfs2_fill_new_dir(osb, handle, dir, inode,
401 					    new_fe_bh, data_ac);
402 		if (status < 0) {
403 			mlog_errno(status);
404 			goto leave;
405 		}
406 
407 		status = ocfs2_journal_access(handle, dir, parent_fe_bh,
408 					      OCFS2_JOURNAL_ACCESS_WRITE);
409 		if (status < 0) {
410 			mlog_errno(status);
411 			goto leave;
412 		}
413 		le16_add_cpu(&dirfe->i_links_count, 1);
414 		status = ocfs2_journal_dirty(handle, parent_fe_bh);
415 		if (status < 0) {
416 			mlog_errno(status);
417 			goto leave;
418 		}
419 		inc_nlink(dir);
420 	}
421 
422 	status = ocfs2_add_entry(handle, dentry, inode,
423 				 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
424 				 de_bh);
425 	if (status < 0) {
426 		mlog_errno(status);
427 		goto leave;
428 	}
429 
430 	status = ocfs2_dentry_attach_lock(dentry, inode,
431 					  OCFS2_I(dir)->ip_blkno);
432 	if (status) {
433 		mlog_errno(status);
434 		goto leave;
435 	}
436 
437 	insert_inode_hash(inode);
438 	dentry->d_op = &ocfs2_dentry_ops;
439 	d_instantiate(dentry, inode);
440 	status = 0;
441 leave:
442 	if (handle)
443 		ocfs2_commit_trans(osb, handle);
444 
445 	ocfs2_meta_unlock(dir, 1);
446 
447 	if (status == -ENOSPC)
448 		mlog(0, "Disk is full\n");
449 
450 	if (new_fe_bh)
451 		brelse(new_fe_bh);
452 
453 	if (de_bh)
454 		brelse(de_bh);
455 
456 	if (parent_fe_bh)
457 		brelse(parent_fe_bh);
458 
459 	if ((status < 0) && inode)
460 		iput(inode);
461 
462 	if (inode_ac)
463 		ocfs2_free_alloc_context(inode_ac);
464 
465 	if (data_ac)
466 		ocfs2_free_alloc_context(data_ac);
467 
468 	mlog_exit(status);
469 
470 	return status;
471 }
472 
473 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
474 			      struct inode *dir,
475 			      struct dentry *dentry, int mode,
476 			      dev_t dev,
477 			      struct buffer_head **new_fe_bh,
478 			      struct buffer_head *parent_fe_bh,
479 			      handle_t *handle,
480 			      struct inode **ret_inode,
481 			      struct ocfs2_alloc_context *inode_ac)
482 {
483 	int status = 0;
484 	struct ocfs2_dinode *fe = NULL;
485 	struct ocfs2_extent_list *fel;
486 	u64 fe_blkno = 0;
487 	u16 suballoc_bit;
488 	struct inode *inode = NULL;
489 
490 	mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
491 		   (unsigned long)dev, dentry->d_name.len,
492 		   dentry->d_name.name);
493 
494 	*new_fe_bh = NULL;
495 	*ret_inode = NULL;
496 
497 	status = ocfs2_claim_new_inode(osb, handle, inode_ac, &suballoc_bit,
498 				       &fe_blkno);
499 	if (status < 0) {
500 		mlog_errno(status);
501 		goto leave;
502 	}
503 
504 	inode = new_inode(dir->i_sb);
505 	if (IS_ERR(inode)) {
506 		status = PTR_ERR(inode);
507 		mlog(ML_ERROR, "new_inode failed!\n");
508 		goto leave;
509 	}
510 
511 	/* populate as many fields early on as possible - many of
512 	 * these are used by the support functions here and in
513 	 * callers. */
514 	inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
515 	OCFS2_I(inode)->ip_blkno = fe_blkno;
516 	if (S_ISDIR(mode))
517 		inode->i_nlink = 2;
518 	else
519 		inode->i_nlink = 1;
520 	inode->i_mode = mode;
521 	spin_lock(&osb->osb_lock);
522 	inode->i_generation = osb->s_next_generation++;
523 	spin_unlock(&osb->osb_lock);
524 
525 	*new_fe_bh = sb_getblk(osb->sb, fe_blkno);
526 	if (!*new_fe_bh) {
527 		status = -EIO;
528 		mlog_errno(status);
529 		goto leave;
530 	}
531 	ocfs2_set_new_buffer_uptodate(inode, *new_fe_bh);
532 
533 	status = ocfs2_journal_access(handle, inode, *new_fe_bh,
534 				      OCFS2_JOURNAL_ACCESS_CREATE);
535 	if (status < 0) {
536 		mlog_errno(status);
537 		goto leave;
538 	}
539 
540 	fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
541 	memset(fe, 0, osb->sb->s_blocksize);
542 
543 	fe->i_generation = cpu_to_le32(inode->i_generation);
544 	fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
545 	fe->i_blkno = cpu_to_le64(fe_blkno);
546 	fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
547 	fe->i_suballoc_slot = cpu_to_le16(osb->slot_num);
548 	fe->i_uid = cpu_to_le32(current->fsuid);
549 	if (dir->i_mode & S_ISGID) {
550 		fe->i_gid = cpu_to_le32(dir->i_gid);
551 		if (S_ISDIR(mode))
552 			mode |= S_ISGID;
553 	} else
554 		fe->i_gid = cpu_to_le32(current->fsgid);
555 	fe->i_mode = cpu_to_le16(mode);
556 	if (S_ISCHR(mode) || S_ISBLK(mode))
557 		fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
558 
559 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
560 
561 	fe->i_last_eb_blk = 0;
562 	strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
563 	le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
564 	fe->i_atime = fe->i_ctime = fe->i_mtime =
565 		cpu_to_le64(CURRENT_TIME.tv_sec);
566 	fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
567 		cpu_to_le32(CURRENT_TIME.tv_nsec);
568 	fe->i_dtime = 0;
569 
570 	fel = &fe->id2.i_list;
571 	fel->l_tree_depth = 0;
572 	fel->l_next_free_rec = 0;
573 	fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
574 
575 	status = ocfs2_journal_dirty(handle, *new_fe_bh);
576 	if (status < 0) {
577 		mlog_errno(status);
578 		goto leave;
579 	}
580 
581 	if (ocfs2_populate_inode(inode, fe, 1) < 0) {
582 		mlog(ML_ERROR, "populate inode failed! bh->b_blocknr=%llu, "
583 		     "i_blkno=%llu, i_ino=%lu\n",
584 		     (unsigned long long) (*new_fe_bh)->b_blocknr,
585 		     (unsigned long long)fe->i_blkno, inode->i_ino);
586 		BUG();
587 	}
588 
589 	ocfs2_inode_set_new(osb, inode);
590 	if (!ocfs2_mount_local(osb)) {
591 		status = ocfs2_create_new_inode_locks(inode);
592 		if (status < 0)
593 			mlog_errno(status);
594 	}
595 
596 	status = 0; /* error in ocfs2_create_new_inode_locks is not
597 		     * critical */
598 
599 	*ret_inode = inode;
600 leave:
601 	if (status < 0) {
602 		if (*new_fe_bh) {
603 			brelse(*new_fe_bh);
604 			*new_fe_bh = NULL;
605 		}
606 		if (inode)
607 			iput(inode);
608 	}
609 
610 	mlog_exit(status);
611 	return status;
612 }
613 
614 static int ocfs2_mkdir(struct inode *dir,
615 		       struct dentry *dentry,
616 		       int mode)
617 {
618 	int ret;
619 
620 	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
621 		   dentry->d_name.len, dentry->d_name.name);
622 	ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
623 	mlog_exit(ret);
624 
625 	return ret;
626 }
627 
628 static int ocfs2_create(struct inode *dir,
629 			struct dentry *dentry,
630 			int mode,
631 			struct nameidata *nd)
632 {
633 	int ret;
634 
635 	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
636 		   dentry->d_name.len, dentry->d_name.name);
637 	ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
638 	mlog_exit(ret);
639 
640 	return ret;
641 }
642 
643 static int ocfs2_link(struct dentry *old_dentry,
644 		      struct inode *dir,
645 		      struct dentry *dentry)
646 {
647 	handle_t *handle;
648 	struct inode *inode = old_dentry->d_inode;
649 	int err;
650 	struct buffer_head *fe_bh = NULL;
651 	struct buffer_head *parent_fe_bh = NULL;
652 	struct buffer_head *de_bh = NULL;
653 	struct ocfs2_dinode *fe = NULL;
654 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
655 
656 	mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
657 		   old_dentry->d_name.len, old_dentry->d_name.name,
658 		   dentry->d_name.len, dentry->d_name.name);
659 
660 	if (S_ISDIR(inode->i_mode))
661 		return -EPERM;
662 
663 	err = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
664 	if (err < 0) {
665 		if (err != -ENOENT)
666 			mlog_errno(err);
667 		return err;
668 	}
669 
670 	if (!dir->i_nlink) {
671 		err = -ENOENT;
672 		goto out;
673 	}
674 
675 	err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
676 					dentry->d_name.len);
677 	if (err)
678 		goto out;
679 
680 	err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
681 					   dentry->d_name.name,
682 					   dentry->d_name.len, &de_bh);
683 	if (err < 0) {
684 		mlog_errno(err);
685 		goto out;
686 	}
687 
688 	err = ocfs2_meta_lock(inode, &fe_bh, 1);
689 	if (err < 0) {
690 		if (err != -ENOENT)
691 			mlog_errno(err);
692 		goto out;
693 	}
694 
695 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
696 	if (le16_to_cpu(fe->i_links_count) >= OCFS2_LINK_MAX) {
697 		err = -EMLINK;
698 		goto out_unlock_inode;
699 	}
700 
701 	handle = ocfs2_start_trans(osb, OCFS2_LINK_CREDITS);
702 	if (IS_ERR(handle)) {
703 		err = PTR_ERR(handle);
704 		handle = NULL;
705 		mlog_errno(err);
706 		goto out_unlock_inode;
707 	}
708 
709 	err = ocfs2_journal_access(handle, inode, fe_bh,
710 				   OCFS2_JOURNAL_ACCESS_WRITE);
711 	if (err < 0) {
712 		mlog_errno(err);
713 		goto out_commit;
714 	}
715 
716 	inc_nlink(inode);
717 	inode->i_ctime = CURRENT_TIME;
718 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
719 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
720 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
721 
722 	err = ocfs2_journal_dirty(handle, fe_bh);
723 	if (err < 0) {
724 		le16_add_cpu(&fe->i_links_count, -1);
725 		drop_nlink(inode);
726 		mlog_errno(err);
727 		goto out_commit;
728 	}
729 
730 	err = ocfs2_add_entry(handle, dentry, inode,
731 			      OCFS2_I(inode)->ip_blkno,
732 			      parent_fe_bh, de_bh);
733 	if (err) {
734 		le16_add_cpu(&fe->i_links_count, -1);
735 		drop_nlink(inode);
736 		mlog_errno(err);
737 		goto out_commit;
738 	}
739 
740 	err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
741 	if (err) {
742 		mlog_errno(err);
743 		goto out_commit;
744 	}
745 
746 	atomic_inc(&inode->i_count);
747 	dentry->d_op = &ocfs2_dentry_ops;
748 	d_instantiate(dentry, inode);
749 
750 out_commit:
751 	ocfs2_commit_trans(osb, handle);
752 out_unlock_inode:
753 	ocfs2_meta_unlock(inode, 1);
754 
755 out:
756 	ocfs2_meta_unlock(dir, 1);
757 
758 	if (de_bh)
759 		brelse(de_bh);
760 	if (fe_bh)
761 		brelse(fe_bh);
762 	if (parent_fe_bh)
763 		brelse(parent_fe_bh);
764 
765 	mlog_exit(err);
766 
767 	return err;
768 }
769 
770 /*
771  * Takes and drops an exclusive lock on the given dentry. This will
772  * force other nodes to drop it.
773  */
774 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
775 {
776 	int ret;
777 
778 	ret = ocfs2_dentry_lock(dentry, 1);
779 	if (ret)
780 		mlog_errno(ret);
781 	else
782 		ocfs2_dentry_unlock(dentry, 1);
783 
784 	return ret;
785 }
786 
787 static inline int inode_is_unlinkable(struct inode *inode)
788 {
789 	if (S_ISDIR(inode->i_mode)) {
790 		if (inode->i_nlink == 2)
791 			return 1;
792 		return 0;
793 	}
794 
795 	if (inode->i_nlink == 1)
796 		return 1;
797 	return 0;
798 }
799 
800 static int ocfs2_unlink(struct inode *dir,
801 			struct dentry *dentry)
802 {
803 	int status;
804 	int child_locked = 0;
805 	struct inode *inode = dentry->d_inode;
806 	struct inode *orphan_dir = NULL;
807 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
808 	u64 blkno;
809 	struct ocfs2_dinode *fe = NULL;
810 	struct buffer_head *fe_bh = NULL;
811 	struct buffer_head *parent_node_bh = NULL;
812 	handle_t *handle = NULL;
813 	struct ocfs2_dir_entry *dirent = NULL;
814 	struct buffer_head *dirent_bh = NULL;
815 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
816 	struct buffer_head *orphan_entry_bh = NULL;
817 
818 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
819 		   dentry->d_name.len, dentry->d_name.name);
820 
821 	BUG_ON(dentry->d_parent->d_inode != dir);
822 
823 	mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
824 
825 	if (inode == osb->root_inode) {
826 		mlog(0, "Cannot delete the root directory\n");
827 		return -EPERM;
828 	}
829 
830 	status = ocfs2_meta_lock(dir, &parent_node_bh, 1);
831 	if (status < 0) {
832 		if (status != -ENOENT)
833 			mlog_errno(status);
834 		return status;
835 	}
836 
837 	status = ocfs2_find_files_on_disk(dentry->d_name.name,
838 					  dentry->d_name.len, &blkno,
839 					  dir, &dirent_bh, &dirent);
840 	if (status < 0) {
841 		if (status != -ENOENT)
842 			mlog_errno(status);
843 		goto leave;
844 	}
845 
846 	if (OCFS2_I(inode)->ip_blkno != blkno) {
847 		status = -ENOENT;
848 
849 		mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
850 		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
851 		     (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
852 		goto leave;
853 	}
854 
855 	status = ocfs2_meta_lock(inode, &fe_bh, 1);
856 	if (status < 0) {
857 		if (status != -ENOENT)
858 			mlog_errno(status);
859 		goto leave;
860 	}
861 	child_locked = 1;
862 
863 	if (S_ISDIR(inode->i_mode)) {
864 	       	if (!ocfs2_empty_dir(inode)) {
865 			status = -ENOTEMPTY;
866 			goto leave;
867 		} else if (inode->i_nlink != 2) {
868 			status = -ENOTEMPTY;
869 			goto leave;
870 		}
871 	}
872 
873 	status = ocfs2_remote_dentry_delete(dentry);
874 	if (status < 0) {
875 		/* This vote should succeed under all normal
876 		 * circumstances. */
877 		mlog_errno(status);
878 		goto leave;
879 	}
880 
881 	if (inode_is_unlinkable(inode)) {
882 		status = ocfs2_prepare_orphan_dir(osb, &orphan_dir, inode,
883 						  orphan_name,
884 						  &orphan_entry_bh);
885 		if (status < 0) {
886 			mlog_errno(status);
887 			goto leave;
888 		}
889 	}
890 
891 	handle = ocfs2_start_trans(osb, OCFS2_UNLINK_CREDITS);
892 	if (IS_ERR(handle)) {
893 		status = PTR_ERR(handle);
894 		handle = NULL;
895 		mlog_errno(status);
896 		goto leave;
897 	}
898 
899 	status = ocfs2_journal_access(handle, inode, fe_bh,
900 				      OCFS2_JOURNAL_ACCESS_WRITE);
901 	if (status < 0) {
902 		mlog_errno(status);
903 		goto leave;
904 	}
905 
906 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
907 
908 	if (inode_is_unlinkable(inode)) {
909 		status = ocfs2_orphan_add(osb, handle, inode, fe, orphan_name,
910 					  orphan_entry_bh, orphan_dir);
911 		if (status < 0) {
912 			mlog_errno(status);
913 			goto leave;
914 		}
915 	}
916 
917 	/* delete the name from the parent dir */
918 	status = ocfs2_delete_entry(handle, dir, dirent, dirent_bh);
919 	if (status < 0) {
920 		mlog_errno(status);
921 		goto leave;
922 	}
923 
924 	if (S_ISDIR(inode->i_mode))
925 		drop_nlink(inode);
926 	drop_nlink(inode);
927 	fe->i_links_count = cpu_to_le16(inode->i_nlink);
928 
929 	status = ocfs2_journal_dirty(handle, fe_bh);
930 	if (status < 0) {
931 		mlog_errno(status);
932 		goto leave;
933 	}
934 
935 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
936 	if (S_ISDIR(inode->i_mode))
937 		drop_nlink(dir);
938 
939 	status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
940 	if (status < 0) {
941 		mlog_errno(status);
942 		if (S_ISDIR(inode->i_mode))
943 			inc_nlink(dir);
944 	}
945 
946 leave:
947 	if (handle)
948 		ocfs2_commit_trans(osb, handle);
949 
950 	if (child_locked)
951 		ocfs2_meta_unlock(inode, 1);
952 
953 	ocfs2_meta_unlock(dir, 1);
954 
955 	if (orphan_dir) {
956 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
957 		ocfs2_meta_unlock(orphan_dir, 1);
958 		mutex_unlock(&orphan_dir->i_mutex);
959 		iput(orphan_dir);
960 	}
961 
962 	if (fe_bh)
963 		brelse(fe_bh);
964 
965 	if (dirent_bh)
966 		brelse(dirent_bh);
967 
968 	if (parent_node_bh)
969 		brelse(parent_node_bh);
970 
971 	if (orphan_entry_bh)
972 		brelse(orphan_entry_bh);
973 
974 	mlog_exit(status);
975 
976 	return status;
977 }
978 
979 /*
980  * The only place this should be used is rename!
981  * if they have the same id, then the 1st one is the only one locked.
982  */
983 static int ocfs2_double_lock(struct ocfs2_super *osb,
984 			     struct buffer_head **bh1,
985 			     struct inode *inode1,
986 			     struct buffer_head **bh2,
987 			     struct inode *inode2)
988 {
989 	int status;
990 	struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
991 	struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
992 	struct buffer_head **tmpbh;
993 	struct inode *tmpinode;
994 
995 	mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
996 		   (unsigned long long)oi1->ip_blkno,
997 		   (unsigned long long)oi2->ip_blkno);
998 
999 	if (*bh1)
1000 		*bh1 = NULL;
1001 	if (*bh2)
1002 		*bh2 = NULL;
1003 
1004 	/* we always want to lock the one with the lower lockid first. */
1005 	if (oi1->ip_blkno != oi2->ip_blkno) {
1006 		if (oi1->ip_blkno < oi2->ip_blkno) {
1007 			/* switch id1 and id2 around */
1008 			mlog(0, "switching them around...\n");
1009 			tmpbh = bh2;
1010 			bh2 = bh1;
1011 			bh1 = tmpbh;
1012 
1013 			tmpinode = inode2;
1014 			inode2 = inode1;
1015 			inode1 = tmpinode;
1016 		}
1017 		/* lock id2 */
1018 		status = ocfs2_meta_lock(inode2, bh2, 1);
1019 		if (status < 0) {
1020 			if (status != -ENOENT)
1021 				mlog_errno(status);
1022 			goto bail;
1023 		}
1024 	}
1025 
1026 	/* lock id1 */
1027 	status = ocfs2_meta_lock(inode1, bh1, 1);
1028 	if (status < 0) {
1029 		/*
1030 		 * An error return must mean that no cluster locks
1031 		 * were held on function exit.
1032 		 */
1033 		if (oi1->ip_blkno != oi2->ip_blkno)
1034 			ocfs2_meta_unlock(inode2, 1);
1035 
1036 		if (status != -ENOENT)
1037 			mlog_errno(status);
1038 	}
1039 
1040 bail:
1041 	mlog_exit(status);
1042 	return status;
1043 }
1044 
1045 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1046 {
1047 	ocfs2_meta_unlock(inode1, 1);
1048 
1049 	if (inode1 != inode2)
1050 		ocfs2_meta_unlock(inode2, 1);
1051 }
1052 
1053 #define PARENT_INO(buffer) \
1054 	((struct ocfs2_dir_entry *) \
1055 	 ((char *)buffer + \
1056 	  le16_to_cpu(((struct ocfs2_dir_entry *)buffer)->rec_len)))->inode
1057 
1058 static int ocfs2_rename(struct inode *old_dir,
1059 			struct dentry *old_dentry,
1060 			struct inode *new_dir,
1061 			struct dentry *new_dentry)
1062 {
1063 	int status = 0, rename_lock = 0, parents_locked = 0;
1064 	int old_child_locked = 0, new_child_locked = 0;
1065 	struct inode *old_inode = old_dentry->d_inode;
1066 	struct inode *new_inode = new_dentry->d_inode;
1067 	struct inode *orphan_dir = NULL;
1068 	struct ocfs2_dinode *newfe = NULL;
1069 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1070 	struct buffer_head *orphan_entry_bh = NULL;
1071 	struct buffer_head *newfe_bh = NULL;
1072 	struct buffer_head *old_inode_bh = NULL;
1073 	struct buffer_head *insert_entry_bh = NULL;
1074 	struct ocfs2_super *osb = NULL;
1075 	u64 newfe_blkno;
1076 	handle_t *handle = NULL;
1077 	struct buffer_head *old_dir_bh = NULL;
1078 	struct buffer_head *new_dir_bh = NULL;
1079 	struct ocfs2_dir_entry *old_de = NULL, *new_de = NULL; // dirent for old_dentry
1080 							       // and new_dentry
1081 	struct buffer_head *new_de_bh = NULL, *old_de_bh = NULL; // bhs for above
1082 	struct buffer_head *old_inode_de_bh = NULL; // if old_dentry is a dir,
1083 						    // this is the 1st dirent bh
1084 	nlink_t old_dir_nlink = old_dir->i_nlink;
1085 
1086 	/* At some point it might be nice to break this function up a
1087 	 * bit. */
1088 
1089 	mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1090 		   old_dir, old_dentry, new_dir, new_dentry,
1091 		   old_dentry->d_name.len, old_dentry->d_name.name,
1092 		   new_dentry->d_name.len, new_dentry->d_name.name);
1093 
1094 	osb = OCFS2_SB(old_dir->i_sb);
1095 
1096 	if (new_inode) {
1097 		if (!igrab(new_inode))
1098 			BUG();
1099 	}
1100 
1101 	/* Assume a directory hierarchy thusly:
1102 	 * a/b/c
1103 	 * a/d
1104 	 * a,b,c, and d are all directories.
1105 	 *
1106 	 * from cwd of 'a' on both nodes:
1107 	 * node1: mv b/c d
1108 	 * node2: mv d   b/c
1109 	 *
1110 	 * And that's why, just like the VFS, we need a file system
1111 	 * rename lock. */
1112 	if (old_dentry != new_dentry) {
1113 		status = ocfs2_rename_lock(osb);
1114 		if (status < 0) {
1115 			mlog_errno(status);
1116 			goto bail;
1117 		}
1118 		rename_lock = 1;
1119 	}
1120 
1121 	/* if old and new are the same, this'll just do one lock. */
1122 	status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1123 				   &new_dir_bh, new_dir);
1124 	if (status < 0) {
1125 		mlog_errno(status);
1126 		goto bail;
1127 	}
1128 	parents_locked = 1;
1129 
1130 	/* make sure both dirs have bhs
1131 	 * get an extra ref on old_dir_bh if old==new */
1132 	if (!new_dir_bh) {
1133 		if (old_dir_bh) {
1134 			new_dir_bh = old_dir_bh;
1135 			get_bh(new_dir_bh);
1136 		} else {
1137 			mlog(ML_ERROR, "no old_dir_bh!\n");
1138 			status = -EIO;
1139 			goto bail;
1140 		}
1141 	}
1142 
1143 	/*
1144 	 * Aside from allowing a meta data update, the locking here
1145 	 * also ensures that the vote thread on other nodes won't have
1146 	 * to concurrently downconvert the inode and the dentry locks.
1147 	 */
1148 	status = ocfs2_meta_lock(old_inode, &old_inode_bh, 1);
1149 	if (status < 0) {
1150 		if (status != -ENOENT)
1151 			mlog_errno(status);
1152 		goto bail;
1153 	}
1154 	old_child_locked = 1;
1155 
1156 	status = ocfs2_remote_dentry_delete(old_dentry);
1157 	if (status < 0) {
1158 		mlog_errno(status);
1159 		goto bail;
1160 	}
1161 
1162 	if (S_ISDIR(old_inode->i_mode)) {
1163 		status = -EIO;
1164 		old_inode_de_bh = ocfs2_bread(old_inode, 0, &status, 0);
1165 		if (!old_inode_de_bh)
1166 			goto bail;
1167 
1168 		status = -EIO;
1169 		if (le64_to_cpu(PARENT_INO(old_inode_de_bh->b_data)) !=
1170 		    OCFS2_I(old_dir)->ip_blkno)
1171 			goto bail;
1172 		status = -EMLINK;
1173 		if (!new_inode && new_dir!=old_dir &&
1174 		    new_dir->i_nlink >= OCFS2_LINK_MAX)
1175 			goto bail;
1176 	}
1177 
1178 	status = -ENOENT;
1179 	old_de_bh = ocfs2_find_entry(old_dentry->d_name.name,
1180 				     old_dentry->d_name.len,
1181 				     old_dir, &old_de);
1182 	if (!old_de_bh)
1183 		goto bail;
1184 
1185 	/*
1186 	 *  Check for inode number is _not_ due to possible IO errors.
1187 	 *  We might rmdir the source, keep it as pwd of some process
1188 	 *  and merrily kill the link to whatever was created under the
1189 	 *  same name. Goodbye sticky bit ;-<
1190 	 */
1191 	if (le64_to_cpu(old_de->inode) != OCFS2_I(old_inode)->ip_blkno)
1192 		goto bail;
1193 
1194 	/* check if the target already exists (in which case we need
1195 	 * to delete it */
1196 	status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1197 					  new_dentry->d_name.len,
1198 					  &newfe_blkno, new_dir, &new_de_bh,
1199 					  &new_de);
1200 	/* The only error we allow here is -ENOENT because the new
1201 	 * file not existing is perfectly valid. */
1202 	if ((status < 0) && (status != -ENOENT)) {
1203 		/* If we cannot find the file specified we should just */
1204 		/* return the error... */
1205 		mlog_errno(status);
1206 		goto bail;
1207 	}
1208 
1209 	if (!new_de && new_inode)
1210 		mlog(ML_ERROR, "inode %lu does not exist in it's parent "
1211 		     "directory!", new_inode->i_ino);
1212 
1213 	/* In case we need to overwrite an existing file, we blow it
1214 	 * away first */
1215 	if (new_de) {
1216 		/* VFS didn't think there existed an inode here, but
1217 		 * someone else in the cluster must have raced our
1218 		 * rename to create one. Today we error cleanly, in
1219 		 * the future we should consider calling iget to build
1220 		 * a new struct inode for this entry. */
1221 		if (!new_inode) {
1222 			status = -EACCES;
1223 
1224 			mlog(0, "We found an inode for name %.*s but VFS "
1225 			     "didn't give us one.\n", new_dentry->d_name.len,
1226 			     new_dentry->d_name.name);
1227 			goto bail;
1228 		}
1229 
1230 		if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1231 			status = -EACCES;
1232 
1233 			mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1234 			     (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1235 			     (unsigned long long)newfe_blkno,
1236 			     OCFS2_I(new_inode)->ip_flags);
1237 			goto bail;
1238 		}
1239 
1240 		status = ocfs2_meta_lock(new_inode, &newfe_bh, 1);
1241 		if (status < 0) {
1242 			if (status != -ENOENT)
1243 				mlog_errno(status);
1244 			goto bail;
1245 		}
1246 		new_child_locked = 1;
1247 
1248 		status = ocfs2_remote_dentry_delete(new_dentry);
1249 		if (status < 0) {
1250 			mlog_errno(status);
1251 			goto bail;
1252 		}
1253 
1254 		newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1255 
1256 		mlog(0, "aha rename over existing... new_de=%p new_blkno=%llu "
1257 		     "newfebh=%p bhblocknr=%llu\n", new_de,
1258 		     (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1259 		     (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1260 
1261 		if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1262 			status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1263 							  new_inode,
1264 							  orphan_name,
1265 							  &orphan_entry_bh);
1266 			if (status < 0) {
1267 				mlog_errno(status);
1268 				goto bail;
1269 			}
1270 		}
1271 	} else {
1272 		BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1273 
1274 		status = ocfs2_check_dir_for_entry(new_dir,
1275 						   new_dentry->d_name.name,
1276 						   new_dentry->d_name.len);
1277 		if (status)
1278 			goto bail;
1279 
1280 		status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1281 						      new_dentry->d_name.name,
1282 						      new_dentry->d_name.len,
1283 						      &insert_entry_bh);
1284 		if (status < 0) {
1285 			mlog_errno(status);
1286 			goto bail;
1287 		}
1288 	}
1289 
1290 	handle = ocfs2_start_trans(osb, OCFS2_RENAME_CREDITS);
1291 	if (IS_ERR(handle)) {
1292 		status = PTR_ERR(handle);
1293 		handle = NULL;
1294 		mlog_errno(status);
1295 		goto bail;
1296 	}
1297 
1298 	if (new_de) {
1299 		if (S_ISDIR(new_inode->i_mode)) {
1300 			if (!ocfs2_empty_dir(new_inode) ||
1301 			    new_inode->i_nlink != 2) {
1302 				status = -ENOTEMPTY;
1303 				goto bail;
1304 			}
1305 		}
1306 		status = ocfs2_journal_access(handle, new_inode, newfe_bh,
1307 					      OCFS2_JOURNAL_ACCESS_WRITE);
1308 		if (status < 0) {
1309 			mlog_errno(status);
1310 			goto bail;
1311 		}
1312 
1313 		if (S_ISDIR(new_inode->i_mode) ||
1314 		    (newfe->i_links_count == cpu_to_le16(1))){
1315 			status = ocfs2_orphan_add(osb, handle, new_inode,
1316 						  newfe, orphan_name,
1317 						  orphan_entry_bh, orphan_dir);
1318 			if (status < 0) {
1319 				mlog_errno(status);
1320 				goto bail;
1321 			}
1322 		}
1323 
1324 		/* change the dirent to point to the correct inode */
1325 		status = ocfs2_journal_access(handle, new_dir, new_de_bh,
1326 					      OCFS2_JOURNAL_ACCESS_WRITE);
1327 		if (status < 0) {
1328 			mlog_errno(status);
1329 			goto bail;
1330 		}
1331 		new_de->inode = cpu_to_le64(OCFS2_I(old_inode)->ip_blkno);
1332 		new_de->file_type = old_de->file_type;
1333 		new_dir->i_version++;
1334 		status = ocfs2_journal_dirty(handle, new_de_bh);
1335 		if (status < 0) {
1336 			mlog_errno(status);
1337 			goto bail;
1338 		}
1339 
1340 		if (S_ISDIR(new_inode->i_mode))
1341 			newfe->i_links_count = 0;
1342 		else
1343 			le16_add_cpu(&newfe->i_links_count, -1);
1344 
1345 		status = ocfs2_journal_dirty(handle, newfe_bh);
1346 		if (status < 0) {
1347 			mlog_errno(status);
1348 			goto bail;
1349 		}
1350 	} else {
1351 		/* if the name was not found in new_dir, add it now */
1352 		status = ocfs2_add_entry(handle, new_dentry, old_inode,
1353 					 OCFS2_I(old_inode)->ip_blkno,
1354 					 new_dir_bh, insert_entry_bh);
1355 	}
1356 
1357 	old_inode->i_ctime = CURRENT_TIME;
1358 	mark_inode_dirty(old_inode);
1359 	ocfs2_mark_inode_dirty(handle, old_inode, old_inode_bh);
1360 
1361 	/* now that the name has been added to new_dir, remove the old name */
1362 	status = ocfs2_delete_entry(handle, old_dir, old_de, old_de_bh);
1363 	if (status < 0) {
1364 		mlog_errno(status);
1365 		goto bail;
1366 	}
1367 
1368 	if (new_inode) {
1369 		new_inode->i_nlink--;
1370 		new_inode->i_ctime = CURRENT_TIME;
1371 	}
1372 	old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1373 	if (old_inode_de_bh) {
1374 		status = ocfs2_journal_access(handle, old_inode,
1375 					     old_inode_de_bh,
1376 					     OCFS2_JOURNAL_ACCESS_WRITE);
1377 		PARENT_INO(old_inode_de_bh->b_data) =
1378 			cpu_to_le64(OCFS2_I(new_dir)->ip_blkno);
1379 		status = ocfs2_journal_dirty(handle, old_inode_de_bh);
1380 		old_dir->i_nlink--;
1381 		if (new_inode) {
1382 			new_inode->i_nlink--;
1383 		} else {
1384 			inc_nlink(new_dir);
1385 			mark_inode_dirty(new_dir);
1386 		}
1387 	}
1388 	mark_inode_dirty(old_dir);
1389 	ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1390 	if (new_inode) {
1391 		mark_inode_dirty(new_inode);
1392 		ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1393 	}
1394 
1395 	if (old_dir != new_dir) {
1396 		/* Keep the same times on both directories.*/
1397 		new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1398 
1399 		/*
1400 		 * This will also pick up the i_nlink change from the
1401 		 * block above.
1402 		 */
1403 		ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1404 	}
1405 
1406 	if (old_dir_nlink != old_dir->i_nlink) {
1407 		if (!old_dir_bh) {
1408 			mlog(ML_ERROR, "need to change nlink for old dir "
1409 			     "%llu from %d to %d but bh is NULL!\n",
1410 			     (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1411 			     (int)old_dir_nlink, old_dir->i_nlink);
1412 		} else {
1413 			struct ocfs2_dinode *fe;
1414 			status = ocfs2_journal_access(handle, old_dir,
1415 						      old_dir_bh,
1416 						      OCFS2_JOURNAL_ACCESS_WRITE);
1417 			fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1418 			fe->i_links_count = cpu_to_le16(old_dir->i_nlink);
1419 			status = ocfs2_journal_dirty(handle, old_dir_bh);
1420 		}
1421 	}
1422 
1423 	ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1424 	status = 0;
1425 bail:
1426 	if (rename_lock)
1427 		ocfs2_rename_unlock(osb);
1428 
1429 	if (handle)
1430 		ocfs2_commit_trans(osb, handle);
1431 
1432 	if (parents_locked)
1433 		ocfs2_double_unlock(old_dir, new_dir);
1434 
1435 	if (old_child_locked)
1436 		ocfs2_meta_unlock(old_inode, 1);
1437 
1438 	if (new_child_locked)
1439 		ocfs2_meta_unlock(new_inode, 1);
1440 
1441 	if (orphan_dir) {
1442 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
1443 		ocfs2_meta_unlock(orphan_dir, 1);
1444 		mutex_unlock(&orphan_dir->i_mutex);
1445 		iput(orphan_dir);
1446 	}
1447 
1448 	if (new_inode)
1449 		sync_mapping_buffers(old_inode->i_mapping);
1450 
1451 	if (new_inode)
1452 		iput(new_inode);
1453 	if (newfe_bh)
1454 		brelse(newfe_bh);
1455 	if (old_inode_bh)
1456 		brelse(old_inode_bh);
1457 	if (old_dir_bh)
1458 		brelse(old_dir_bh);
1459 	if (new_dir_bh)
1460 		brelse(new_dir_bh);
1461 	if (new_de_bh)
1462 		brelse(new_de_bh);
1463 	if (old_de_bh)
1464 		brelse(old_de_bh);
1465 	if (old_inode_de_bh)
1466 		brelse(old_inode_de_bh);
1467 	if (orphan_entry_bh)
1468 		brelse(orphan_entry_bh);
1469 	if (insert_entry_bh)
1470 		brelse(insert_entry_bh);
1471 
1472 	mlog_exit(status);
1473 
1474 	return status;
1475 }
1476 
1477 /*
1478  * we expect i_size = strlen(symname). Copy symname into the file
1479  * data, including the null terminator.
1480  */
1481 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1482 				     handle_t *handle,
1483 				     struct inode *inode,
1484 				     const char *symname)
1485 {
1486 	struct buffer_head **bhs = NULL;
1487 	const char *c;
1488 	struct super_block *sb = osb->sb;
1489 	u64 p_blkno;
1490 	int p_blocks;
1491 	int virtual, blocks, status, i, bytes_left;
1492 
1493 	bytes_left = i_size_read(inode) + 1;
1494 	/* we can't trust i_blocks because we're actually going to
1495 	 * write i_size + 1 bytes. */
1496 	blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1497 
1498 	mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1499 			(unsigned long long)inode->i_blocks,
1500 			i_size_read(inode), blocks);
1501 
1502 	/* Sanity check -- make sure we're going to fit. */
1503 	if (bytes_left >
1504 	    ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1505 		status = -EIO;
1506 		mlog_errno(status);
1507 		goto bail;
1508 	}
1509 
1510 	bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1511 	if (!bhs) {
1512 		status = -ENOMEM;
1513 		mlog_errno(status);
1514 		goto bail;
1515 	}
1516 
1517 	status = ocfs2_extent_map_get_blocks(inode, 0, 1, &p_blkno,
1518 					     &p_blocks);
1519 	if (status < 0) {
1520 		mlog_errno(status);
1521 		goto bail;
1522 	}
1523 
1524 	/* links can never be larger than one cluster so we know this
1525 	 * is all going to be contiguous, but do a sanity check
1526 	 * anyway. */
1527 	if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1528 		status = -EIO;
1529 		mlog_errno(status);
1530 		goto bail;
1531 	}
1532 
1533 	virtual = 0;
1534 	while(bytes_left > 0) {
1535 		c = &symname[virtual * sb->s_blocksize];
1536 
1537 		bhs[virtual] = sb_getblk(sb, p_blkno);
1538 		if (!bhs[virtual]) {
1539 			status = -ENOMEM;
1540 			mlog_errno(status);
1541 			goto bail;
1542 		}
1543 		ocfs2_set_new_buffer_uptodate(inode, bhs[virtual]);
1544 
1545 		status = ocfs2_journal_access(handle, inode, bhs[virtual],
1546 					      OCFS2_JOURNAL_ACCESS_CREATE);
1547 		if (status < 0) {
1548 			mlog_errno(status);
1549 			goto bail;
1550 		}
1551 
1552 		memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1553 
1554 		memcpy(bhs[virtual]->b_data, c,
1555 		       (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1556 		       bytes_left);
1557 
1558 		status = ocfs2_journal_dirty(handle, bhs[virtual]);
1559 		if (status < 0) {
1560 			mlog_errno(status);
1561 			goto bail;
1562 		}
1563 
1564 		virtual++;
1565 		p_blkno++;
1566 		bytes_left -= sb->s_blocksize;
1567 	}
1568 
1569 	status = 0;
1570 bail:
1571 
1572 	if (bhs) {
1573 		for(i = 0; i < blocks; i++)
1574 			if (bhs[i])
1575 				brelse(bhs[i]);
1576 		kfree(bhs);
1577 	}
1578 
1579 	mlog_exit(status);
1580 	return status;
1581 }
1582 
1583 static int ocfs2_symlink(struct inode *dir,
1584 			 struct dentry *dentry,
1585 			 const char *symname)
1586 {
1587 	int status, l, credits;
1588 	u64 newsize;
1589 	struct ocfs2_super *osb = NULL;
1590 	struct inode *inode = NULL;
1591 	struct super_block *sb;
1592 	struct buffer_head *new_fe_bh = NULL;
1593 	struct buffer_head *de_bh = NULL;
1594 	struct buffer_head *parent_fe_bh = NULL;
1595 	struct ocfs2_dinode *fe = NULL;
1596 	struct ocfs2_dinode *dirfe;
1597 	handle_t *handle = NULL;
1598 	struct ocfs2_alloc_context *inode_ac = NULL;
1599 	struct ocfs2_alloc_context *data_ac = NULL;
1600 
1601 	mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1602 		   dentry, symname, dentry->d_name.len, dentry->d_name.name);
1603 
1604 	sb = dir->i_sb;
1605 	osb = OCFS2_SB(sb);
1606 
1607 	l = strlen(symname) + 1;
1608 
1609 	credits = ocfs2_calc_symlink_credits(sb);
1610 
1611 	/* lock the parent directory */
1612 	status = ocfs2_meta_lock(dir, &parent_fe_bh, 1);
1613 	if (status < 0) {
1614 		if (status != -ENOENT)
1615 			mlog_errno(status);
1616 		return status;
1617 	}
1618 
1619 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1620 	if (!dirfe->i_links_count) {
1621 		/* can't make a file in a deleted directory. */
1622 		status = -ENOENT;
1623 		goto bail;
1624 	}
1625 
1626 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1627 					   dentry->d_name.len);
1628 	if (status)
1629 		goto bail;
1630 
1631 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1632 					      dentry->d_name.name,
1633 					      dentry->d_name.len, &de_bh);
1634 	if (status < 0) {
1635 		mlog_errno(status);
1636 		goto bail;
1637 	}
1638 
1639 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
1640 	if (status < 0) {
1641 		if (status != -ENOSPC)
1642 			mlog_errno(status);
1643 		goto bail;
1644 	}
1645 
1646 	/* don't reserve bitmap space for fast symlinks. */
1647 	if (l > ocfs2_fast_symlink_chars(sb)) {
1648 		status = ocfs2_reserve_clusters(osb, 1, &data_ac);
1649 		if (status < 0) {
1650 			if (status != -ENOSPC)
1651 				mlog_errno(status);
1652 			goto bail;
1653 		}
1654 	}
1655 
1656 	handle = ocfs2_start_trans(osb, credits);
1657 	if (IS_ERR(handle)) {
1658 		status = PTR_ERR(handle);
1659 		handle = NULL;
1660 		mlog_errno(status);
1661 		goto bail;
1662 	}
1663 
1664 	status = ocfs2_mknod_locked(osb, dir, dentry,
1665 				    S_IFLNK | S_IRWXUGO, 0,
1666 				    &new_fe_bh, parent_fe_bh, handle,
1667 				    &inode, inode_ac);
1668 	if (status < 0) {
1669 		mlog_errno(status);
1670 		goto bail;
1671 	}
1672 
1673 	fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1674 	inode->i_rdev = 0;
1675 	newsize = l - 1;
1676 	if (l > ocfs2_fast_symlink_chars(sb)) {
1677 		inode->i_op = &ocfs2_symlink_inode_operations;
1678 		status = ocfs2_do_extend_allocation(osb, inode, 1, new_fe_bh,
1679 						    handle, data_ac, NULL,
1680 						    NULL);
1681 		if (status < 0) {
1682 			if (status != -ENOSPC && status != -EINTR) {
1683 				mlog(ML_ERROR,
1684 				     "Failed to extend file to %llu\n",
1685 				     (unsigned long long)newsize);
1686 				mlog_errno(status);
1687 				status = -ENOSPC;
1688 			}
1689 			goto bail;
1690 		}
1691 		i_size_write(inode, newsize);
1692 		inode->i_blocks = ocfs2_align_bytes_to_sectors(newsize);
1693 	} else {
1694 		inode->i_op = &ocfs2_fast_symlink_inode_operations;
1695 		memcpy((char *) fe->id2.i_symlink, symname, l);
1696 		i_size_write(inode, newsize);
1697 		inode->i_blocks = 0;
1698 	}
1699 
1700 	status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1701 	if (status < 0) {
1702 		mlog_errno(status);
1703 		goto bail;
1704 	}
1705 
1706 	if (!ocfs2_inode_is_fast_symlink(inode)) {
1707 		status = ocfs2_create_symlink_data(osb, handle, inode,
1708 						   symname);
1709 		if (status < 0) {
1710 			mlog_errno(status);
1711 			goto bail;
1712 		}
1713 	}
1714 
1715 	status = ocfs2_add_entry(handle, dentry, inode,
1716 				 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1717 				 de_bh);
1718 	if (status < 0) {
1719 		mlog_errno(status);
1720 		goto bail;
1721 	}
1722 
1723 	status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1724 	if (status) {
1725 		mlog_errno(status);
1726 		goto bail;
1727 	}
1728 
1729 	insert_inode_hash(inode);
1730 	dentry->d_op = &ocfs2_dentry_ops;
1731 	d_instantiate(dentry, inode);
1732 bail:
1733 	if (handle)
1734 		ocfs2_commit_trans(osb, handle);
1735 
1736 	ocfs2_meta_unlock(dir, 1);
1737 
1738 	if (new_fe_bh)
1739 		brelse(new_fe_bh);
1740 	if (parent_fe_bh)
1741 		brelse(parent_fe_bh);
1742 	if (de_bh)
1743 		brelse(de_bh);
1744 	if (inode_ac)
1745 		ocfs2_free_alloc_context(inode_ac);
1746 	if (data_ac)
1747 		ocfs2_free_alloc_context(data_ac);
1748 	if ((status < 0) && inode)
1749 		iput(inode);
1750 
1751 	mlog_exit(status);
1752 
1753 	return status;
1754 }
1755 
1756 int ocfs2_check_dir_entry(struct inode * dir,
1757 			  struct ocfs2_dir_entry * de,
1758 			  struct buffer_head * bh,
1759 			  unsigned long offset)
1760 {
1761 	const char *error_msg = NULL;
1762 	const int rlen = le16_to_cpu(de->rec_len);
1763 
1764 	if (rlen < OCFS2_DIR_REC_LEN(1))
1765 		error_msg = "rec_len is smaller than minimal";
1766 	else if (rlen % 4 != 0)
1767 		error_msg = "rec_len % 4 != 0";
1768 	else if (rlen < OCFS2_DIR_REC_LEN(de->name_len))
1769 		error_msg = "rec_len is too small for name_len";
1770 	else if (((char *) de - bh->b_data) + rlen > dir->i_sb->s_blocksize)
1771 		error_msg = "directory entry across blocks";
1772 
1773 	if (error_msg != NULL)
1774 		mlog(ML_ERROR, "bad entry in directory #%llu: %s - "
1775 		     "offset=%lu, inode=%llu, rec_len=%d, name_len=%d\n",
1776 		     (unsigned long long)OCFS2_I(dir)->ip_blkno, error_msg,
1777 		     offset, (unsigned long long)le64_to_cpu(de->inode), rlen,
1778 		     de->name_len);
1779 	return error_msg == NULL ? 1 : 0;
1780 }
1781 
1782 /* we don't always have a dentry for what we want to add, so people
1783  * like orphan dir can call this instead.
1784  *
1785  * If you pass me insert_bh, I'll skip the search of the other dir
1786  * blocks and put the record in there.
1787  */
1788 static int __ocfs2_add_entry(handle_t *handle,
1789 			     struct inode *dir,
1790 			     const char *name, int namelen,
1791 			     struct inode *inode, u64 blkno,
1792 			     struct buffer_head *parent_fe_bh,
1793 			     struct buffer_head *insert_bh)
1794 {
1795 	unsigned long offset;
1796 	unsigned short rec_len;
1797 	struct ocfs2_dir_entry *de, *de1;
1798 	struct super_block *sb;
1799 	int retval, status;
1800 
1801 	mlog_entry_void();
1802 
1803 	sb = dir->i_sb;
1804 
1805 	if (!namelen)
1806 		return -EINVAL;
1807 
1808 	rec_len = OCFS2_DIR_REC_LEN(namelen);
1809 	offset = 0;
1810 	de = (struct ocfs2_dir_entry *) insert_bh->b_data;
1811 	while (1) {
1812 		BUG_ON((char *)de >= sb->s_blocksize + insert_bh->b_data);
1813 		/* These checks should've already been passed by the
1814 		 * prepare function, but I guess we can leave them
1815 		 * here anyway. */
1816 		if (!ocfs2_check_dir_entry(dir, de, insert_bh, offset)) {
1817 			retval = -ENOENT;
1818 			goto bail;
1819 		}
1820 		if (ocfs2_match(namelen, name, de)) {
1821 			retval = -EEXIST;
1822 			goto bail;
1823 		}
1824 		if (((le64_to_cpu(de->inode) == 0) &&
1825 		     (le16_to_cpu(de->rec_len) >= rec_len)) ||
1826 		    (le16_to_cpu(de->rec_len) >=
1827 		     (OCFS2_DIR_REC_LEN(de->name_len) + rec_len))) {
1828 			dir->i_mtime = dir->i_ctime = CURRENT_TIME;
1829 			retval = ocfs2_mark_inode_dirty(handle, dir, parent_fe_bh);
1830 			if (retval < 0) {
1831 				mlog_errno(retval);
1832 				goto bail;
1833 			}
1834 
1835 			status = ocfs2_journal_access(handle, dir, insert_bh,
1836 						      OCFS2_JOURNAL_ACCESS_WRITE);
1837 			/* By now the buffer is marked for journaling */
1838 			offset += le16_to_cpu(de->rec_len);
1839 			if (le64_to_cpu(de->inode)) {
1840 				de1 = (struct ocfs2_dir_entry *)((char *) de +
1841 					OCFS2_DIR_REC_LEN(de->name_len));
1842 				de1->rec_len =
1843 					cpu_to_le16(le16_to_cpu(de->rec_len) -
1844 					OCFS2_DIR_REC_LEN(de->name_len));
1845 				de->rec_len = cpu_to_le16(OCFS2_DIR_REC_LEN(de->name_len));
1846 				de = de1;
1847 			}
1848 			de->file_type = OCFS2_FT_UNKNOWN;
1849 			if (blkno) {
1850 				de->inode = cpu_to_le64(blkno);
1851 				ocfs2_set_de_type(de, inode->i_mode);
1852 			} else
1853 				de->inode = 0;
1854 			de->name_len = namelen;
1855 			memcpy(de->name, name, namelen);
1856 
1857 			dir->i_version++;
1858 			status = ocfs2_journal_dirty(handle, insert_bh);
1859 			retval = 0;
1860 			goto bail;
1861 		}
1862 		offset += le16_to_cpu(de->rec_len);
1863 		de = (struct ocfs2_dir_entry *) ((char *) de + le16_to_cpu(de->rec_len));
1864 	}
1865 
1866 	/* when you think about it, the assert above should prevent us
1867 	 * from ever getting here. */
1868 	retval = -ENOSPC;
1869 bail:
1870 
1871 	mlog_exit(retval);
1872 	return retval;
1873 }
1874 
1875 
1876 /*
1877  * ocfs2_delete_entry deletes a directory entry by merging it with the
1878  * previous entry
1879  */
1880 static int ocfs2_delete_entry(handle_t *handle,
1881 			      struct inode *dir,
1882 			      struct ocfs2_dir_entry *de_del,
1883 			      struct buffer_head *bh)
1884 {
1885 	struct ocfs2_dir_entry *de, *pde;
1886 	int i, status = -ENOENT;
1887 
1888 	mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p)\n", handle, dir, de_del, bh);
1889 
1890 	i = 0;
1891 	pde = NULL;
1892 	de = (struct ocfs2_dir_entry *) bh->b_data;
1893 	while (i < bh->b_size) {
1894 		if (!ocfs2_check_dir_entry(dir, de, bh, i)) {
1895 			status = -EIO;
1896 			mlog_errno(status);
1897 			goto bail;
1898 		}
1899 		if (de == de_del)  {
1900 			status = ocfs2_journal_access(handle, dir, bh,
1901 						      OCFS2_JOURNAL_ACCESS_WRITE);
1902 			if (status < 0) {
1903 				status = -EIO;
1904 				mlog_errno(status);
1905 				goto bail;
1906 			}
1907 			if (pde)
1908 				pde->rec_len =
1909 					cpu_to_le16(le16_to_cpu(pde->rec_len) +
1910 						    le16_to_cpu(de->rec_len));
1911 			else
1912 				de->inode = 0;
1913 			dir->i_version++;
1914 			status = ocfs2_journal_dirty(handle, bh);
1915 			goto bail;
1916 		}
1917 		i += le16_to_cpu(de->rec_len);
1918 		pde = de;
1919 		de = (struct ocfs2_dir_entry *)((char *)de + le16_to_cpu(de->rec_len));
1920 	}
1921 bail:
1922 	mlog_exit(status);
1923 	return status;
1924 }
1925 
1926 /*
1927  * Returns 0 if not found, -1 on failure, and 1 on success
1928  */
1929 static int inline ocfs2_search_dirblock(struct buffer_head *bh,
1930 					struct inode *dir,
1931 					const char *name, int namelen,
1932 					unsigned long offset,
1933 					struct ocfs2_dir_entry **res_dir)
1934 {
1935 	struct ocfs2_dir_entry *de;
1936 	char *dlimit, *de_buf;
1937 	int de_len;
1938 	int ret = 0;
1939 
1940 	mlog_entry_void();
1941 
1942 	de_buf = bh->b_data;
1943 	dlimit = de_buf + dir->i_sb->s_blocksize;
1944 
1945 	while (de_buf < dlimit) {
1946 		/* this code is executed quadratically often */
1947 		/* do minimal checking `by hand' */
1948 
1949 		de = (struct ocfs2_dir_entry *) de_buf;
1950 
1951 		if (de_buf + namelen <= dlimit &&
1952 		    ocfs2_match(namelen, name, de)) {
1953 			/* found a match - just to be sure, do a full check */
1954 			if (!ocfs2_check_dir_entry(dir, de, bh, offset)) {
1955 				ret = -1;
1956 				goto bail;
1957 			}
1958 			*res_dir = de;
1959 			ret = 1;
1960 			goto bail;
1961 		}
1962 
1963 		/* prevent looping on a bad block */
1964 		de_len = le16_to_cpu(de->rec_len);
1965 		if (de_len <= 0) {
1966 			ret = -1;
1967 			goto bail;
1968 		}
1969 
1970 		de_buf += de_len;
1971 		offset += de_len;
1972 	}
1973 
1974 bail:
1975 	mlog_exit(ret);
1976 	return ret;
1977 }
1978 
1979 struct buffer_head *ocfs2_find_entry(const char *name, int namelen,
1980 				     struct inode *dir,
1981 				     struct ocfs2_dir_entry **res_dir)
1982 {
1983 	struct super_block *sb;
1984 	struct buffer_head *bh_use[NAMEI_RA_SIZE];
1985 	struct buffer_head *bh, *ret = NULL;
1986 	unsigned long start, block, b;
1987 	int ra_max = 0;		/* Number of bh's in the readahead
1988 				   buffer, bh_use[] */
1989 	int ra_ptr = 0;		/* Current index into readahead
1990 				   buffer */
1991 	int num = 0;
1992 	int nblocks, i, err;
1993 
1994 	mlog_entry_void();
1995 
1996 	*res_dir = NULL;
1997 	sb = dir->i_sb;
1998 
1999 	nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2000 	start = OCFS2_I(dir)->ip_dir_start_lookup;
2001 	if (start >= nblocks)
2002 		start = 0;
2003 	block = start;
2004 
2005 restart:
2006 	do {
2007 		/*
2008 		 * We deal with the read-ahead logic here.
2009 		 */
2010 		if (ra_ptr >= ra_max) {
2011 			/* Refill the readahead buffer */
2012 			ra_ptr = 0;
2013 			b = block;
2014 			for (ra_max = 0; ra_max < NAMEI_RA_SIZE; ra_max++) {
2015 				/*
2016 				 * Terminate if we reach the end of the
2017 				 * directory and must wrap, or if our
2018 				 * search has finished at this block.
2019 				 */
2020 				if (b >= nblocks || (num && block == start)) {
2021 					bh_use[ra_max] = NULL;
2022 					break;
2023 				}
2024 				num++;
2025 
2026 				bh = ocfs2_bread(dir, b++, &err, 1);
2027 				bh_use[ra_max] = bh;
2028 			}
2029 		}
2030 		if ((bh = bh_use[ra_ptr++]) == NULL)
2031 			goto next;
2032 		wait_on_buffer(bh);
2033 		if (!buffer_uptodate(bh)) {
2034 			/* read error, skip block & hope for the best */
2035 			ocfs2_error(dir->i_sb, "reading directory %llu, "
2036 				    "offset %lu\n",
2037 				    (unsigned long long)OCFS2_I(dir)->ip_blkno,
2038 				    block);
2039 			brelse(bh);
2040 			goto next;
2041 		}
2042 		i = ocfs2_search_dirblock(bh, dir, name, namelen,
2043 					  block << sb->s_blocksize_bits,
2044 					  res_dir);
2045 		if (i == 1) {
2046 			OCFS2_I(dir)->ip_dir_start_lookup = block;
2047 			ret = bh;
2048 			goto cleanup_and_exit;
2049 		} else {
2050 			brelse(bh);
2051 			if (i < 0)
2052 				goto cleanup_and_exit;
2053 		}
2054 	next:
2055 		if (++block >= nblocks)
2056 			block = 0;
2057 	} while (block != start);
2058 
2059 	/*
2060 	 * If the directory has grown while we were searching, then
2061 	 * search the last part of the directory before giving up.
2062 	 */
2063 	block = nblocks;
2064 	nblocks = i_size_read(dir) >> sb->s_blocksize_bits;
2065 	if (block < nblocks) {
2066 		start = 0;
2067 		goto restart;
2068 	}
2069 
2070 cleanup_and_exit:
2071 	/* Clean up the read-ahead blocks */
2072 	for (; ra_ptr < ra_max; ra_ptr++)
2073 		brelse(bh_use[ra_ptr]);
2074 
2075 	mlog_exit_ptr(ret);
2076 	return ret;
2077 }
2078 
2079 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2080 {
2081 	int status, namelen;
2082 
2083 	mlog_entry_void();
2084 
2085 	namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2086 			   (long long)blkno);
2087 	if (namelen <= 0) {
2088 		if (namelen)
2089 			status = namelen;
2090 		else
2091 			status = -EINVAL;
2092 		mlog_errno(status);
2093 		goto bail;
2094 	}
2095 	if (namelen != OCFS2_ORPHAN_NAMELEN) {
2096 		status = -EINVAL;
2097 		mlog_errno(status);
2098 		goto bail;
2099 	}
2100 
2101 	mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
2102 	     namelen);
2103 
2104 	status = 0;
2105 bail:
2106 	mlog_exit(status);
2107 	return status;
2108 }
2109 
2110 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2111 				    struct inode **ret_orphan_dir,
2112 				    struct inode *inode,
2113 				    char *name,
2114 				    struct buffer_head **de_bh)
2115 {
2116 	struct inode *orphan_dir_inode;
2117 	struct buffer_head *orphan_dir_bh = NULL;
2118 	int status = 0;
2119 
2120 	status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2121 	if (status < 0) {
2122 		mlog_errno(status);
2123 		return status;
2124 	}
2125 
2126 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2127 						       ORPHAN_DIR_SYSTEM_INODE,
2128 						       osb->slot_num);
2129 	if (!orphan_dir_inode) {
2130 		status = -ENOENT;
2131 		mlog_errno(status);
2132 		return status;
2133 	}
2134 
2135 	mutex_lock(&orphan_dir_inode->i_mutex);
2136 
2137 	status = ocfs2_meta_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2138 	if (status < 0) {
2139 		mlog_errno(status);
2140 		goto leave;
2141 	}
2142 
2143 	status = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2144 					      orphan_dir_bh, name,
2145 					      OCFS2_ORPHAN_NAMELEN, de_bh);
2146 	if (status < 0) {
2147 		ocfs2_meta_unlock(orphan_dir_inode, 1);
2148 
2149 		mlog_errno(status);
2150 		goto leave;
2151 	}
2152 
2153 	*ret_orphan_dir = orphan_dir_inode;
2154 
2155 leave:
2156 	if (status) {
2157 		mutex_unlock(&orphan_dir_inode->i_mutex);
2158 		iput(orphan_dir_inode);
2159 	}
2160 
2161 	if (orphan_dir_bh)
2162 		brelse(orphan_dir_bh);
2163 
2164 	mlog_exit(status);
2165 	return status;
2166 }
2167 
2168 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2169 			    handle_t *handle,
2170 			    struct inode *inode,
2171 			    struct ocfs2_dinode *fe,
2172 			    char *name,
2173 			    struct buffer_head *de_bh,
2174 			    struct inode *orphan_dir_inode)
2175 {
2176 	struct buffer_head *orphan_dir_bh = NULL;
2177 	int status = 0;
2178 	struct ocfs2_dinode *orphan_fe;
2179 
2180 	mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2181 
2182 	status = ocfs2_read_block(osb,
2183 				  OCFS2_I(orphan_dir_inode)->ip_blkno,
2184 				  &orphan_dir_bh, OCFS2_BH_CACHED,
2185 				  orphan_dir_inode);
2186 	if (status < 0) {
2187 		mlog_errno(status);
2188 		goto leave;
2189 	}
2190 
2191 	status = ocfs2_journal_access(handle, orphan_dir_inode, orphan_dir_bh,
2192 				      OCFS2_JOURNAL_ACCESS_WRITE);
2193 	if (status < 0) {
2194 		mlog_errno(status);
2195 		goto leave;
2196 	}
2197 
2198 	/* we're a cluster, and nlink can change on disk from
2199 	 * underneath us... */
2200 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2201 	if (S_ISDIR(inode->i_mode))
2202 		le16_add_cpu(&orphan_fe->i_links_count, 1);
2203 	orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2204 
2205 	status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2206 	if (status < 0) {
2207 		mlog_errno(status);
2208 		goto leave;
2209 	}
2210 
2211 	status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2212 				   OCFS2_ORPHAN_NAMELEN, inode,
2213 				   OCFS2_I(inode)->ip_blkno,
2214 				   orphan_dir_bh, de_bh);
2215 	if (status < 0) {
2216 		mlog_errno(status);
2217 		goto leave;
2218 	}
2219 
2220 	le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2221 
2222 	/* Record which orphan dir our inode now resides
2223 	 * in. delete_inode will use this to determine which orphan
2224 	 * dir to lock. */
2225 	spin_lock(&OCFS2_I(inode)->ip_lock);
2226 	OCFS2_I(inode)->ip_orphaned_slot = osb->slot_num;
2227 	spin_unlock(&OCFS2_I(inode)->ip_lock);
2228 
2229 	mlog(0, "Inode %llu orphaned in slot %d\n",
2230 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2231 
2232 leave:
2233 	if (orphan_dir_bh)
2234 		brelse(orphan_dir_bh);
2235 
2236 	mlog_exit(status);
2237 	return status;
2238 }
2239 
2240 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2241 int ocfs2_orphan_del(struct ocfs2_super *osb,
2242 		     handle_t *handle,
2243 		     struct inode *orphan_dir_inode,
2244 		     struct inode *inode,
2245 		     struct buffer_head *orphan_dir_bh)
2246 {
2247 	char name[OCFS2_ORPHAN_NAMELEN + 1];
2248 	struct ocfs2_dinode *orphan_fe;
2249 	int status = 0;
2250 	struct buffer_head *target_de_bh = NULL;
2251 	struct ocfs2_dir_entry *target_de = NULL;
2252 
2253 	mlog_entry_void();
2254 
2255 	status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2256 	if (status < 0) {
2257 		mlog_errno(status);
2258 		goto leave;
2259 	}
2260 
2261 	mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2262 	     name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2263 	     OCFS2_ORPHAN_NAMELEN);
2264 
2265 	/* find it's spot in the orphan directory */
2266 	target_de_bh = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN,
2267 					orphan_dir_inode, &target_de);
2268 	if (!target_de_bh) {
2269 		status = -ENOENT;
2270 		mlog_errno(status);
2271 		goto leave;
2272 	}
2273 
2274 	/* remove it from the orphan directory */
2275 	status = ocfs2_delete_entry(handle, orphan_dir_inode, target_de,
2276 				    target_de_bh);
2277 	if (status < 0) {
2278 		mlog_errno(status);
2279 		goto leave;
2280 	}
2281 
2282 	status = ocfs2_journal_access(handle,orphan_dir_inode,  orphan_dir_bh,
2283 				      OCFS2_JOURNAL_ACCESS_WRITE);
2284 	if (status < 0) {
2285 		mlog_errno(status);
2286 		goto leave;
2287 	}
2288 
2289 	/* do the i_nlink dance! :) */
2290 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2291 	if (S_ISDIR(inode->i_mode))
2292 		le16_add_cpu(&orphan_fe->i_links_count, -1);
2293 	orphan_dir_inode->i_nlink = le16_to_cpu(orphan_fe->i_links_count);
2294 
2295 	status = ocfs2_journal_dirty(handle, orphan_dir_bh);
2296 	if (status < 0) {
2297 		mlog_errno(status);
2298 		goto leave;
2299 	}
2300 
2301 leave:
2302 	if (target_de_bh)
2303 		brelse(target_de_bh);
2304 
2305 	mlog_exit(status);
2306 	return status;
2307 }
2308 
2309 const struct inode_operations ocfs2_dir_iops = {
2310 	.create		= ocfs2_create,
2311 	.lookup		= ocfs2_lookup,
2312 	.link		= ocfs2_link,
2313 	.unlink		= ocfs2_unlink,
2314 	.rmdir		= ocfs2_unlink,
2315 	.symlink	= ocfs2_symlink,
2316 	.mkdir		= ocfs2_mkdir,
2317 	.mknod		= ocfs2_mknod,
2318 	.rename		= ocfs2_rename,
2319 	.setattr	= ocfs2_setattr,
2320 	.getattr	= ocfs2_getattr,
2321 	.permission	= ocfs2_permission,
2322 };
2323