xref: /openbmc/linux/fs/ocfs2/namei.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * namei.c
5  *
6  * Create and rename file, directory, symlinks
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  *  Portions of this code from linux/fs/ext3/dir.c
11  *
12  *  Copyright (C) 1992, 1993, 1994, 1995
13  *  Remy Card (card@masi.ibp.fr)
14  *  Laboratoire MASI - Institut Blaise pascal
15  *  Universite Pierre et Marie Curie (Paris VI)
16  *
17  *   from
18  *
19  *   linux/fs/minix/dir.c
20  *
21  *   Copyright (C) 1991, 1992 Linux Torvalds
22  *
23  * This program is free software; you can redistribute it and/or
24  * modify it under the terms of the GNU General Public
25  * License as published by the Free Software Foundation; either
26  * version 2 of the License, or (at your option) any later version.
27  *
28  * This program is distributed in the hope that it will be useful,
29  * but WITHOUT ANY WARRANTY; without even the implied warranty of
30  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
31  * General Public License for more details.
32  *
33  * You should have received a copy of the GNU General Public
34  * License along with this program; if not, write to the
35  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
36  * Boston, MA 021110-1307, USA.
37  */
38 
39 #include <linux/fs.h>
40 #include <linux/types.h>
41 #include <linux/slab.h>
42 #include <linux/highmem.h>
43 #include <linux/quotaops.h>
44 
45 #define MLOG_MASK_PREFIX ML_NAMEI
46 #include <cluster/masklog.h>
47 
48 #include "ocfs2.h"
49 
50 #include "alloc.h"
51 #include "dcache.h"
52 #include "dir.h"
53 #include "dlmglue.h"
54 #include "extent_map.h"
55 #include "file.h"
56 #include "inode.h"
57 #include "journal.h"
58 #include "namei.h"
59 #include "suballoc.h"
60 #include "super.h"
61 #include "symlink.h"
62 #include "sysfile.h"
63 #include "uptodate.h"
64 #include "xattr.h"
65 #include "acl.h"
66 
67 #include "buffer_head_io.h"
68 
69 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
70 			      struct inode *dir,
71 			      struct inode *inode,
72 			      dev_t dev,
73 			      struct buffer_head **new_fe_bh,
74 			      struct buffer_head *parent_fe_bh,
75 			      handle_t *handle,
76 			      struct ocfs2_alloc_context *inode_ac);
77 
78 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
79 				    struct inode **ret_orphan_dir,
80 				    u64 blkno,
81 				    char *name,
82 				    struct ocfs2_dir_lookup_result *lookup);
83 
84 static int ocfs2_orphan_add(struct ocfs2_super *osb,
85 			    handle_t *handle,
86 			    struct inode *inode,
87 			    struct buffer_head *fe_bh,
88 			    char *name,
89 			    struct ocfs2_dir_lookup_result *lookup,
90 			    struct inode *orphan_dir_inode);
91 
92 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
93 				     handle_t *handle,
94 				     struct inode *inode,
95 				     const char *symname);
96 
97 /* An orphan dir name is an 8 byte value, printed as a hex string */
98 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
99 
100 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
101 				   struct nameidata *nd)
102 {
103 	int status;
104 	u64 blkno;
105 	struct inode *inode = NULL;
106 	struct dentry *ret;
107 	struct ocfs2_inode_info *oi;
108 
109 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
110 		   dentry->d_name.len, dentry->d_name.name);
111 
112 	if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
113 		ret = ERR_PTR(-ENAMETOOLONG);
114 		goto bail;
115 	}
116 
117 	mlog(0, "find name %.*s in directory %llu\n", dentry->d_name.len,
118 	     dentry->d_name.name, (unsigned long long)OCFS2_I(dir)->ip_blkno);
119 
120 	status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
121 	if (status < 0) {
122 		if (status != -ENOENT)
123 			mlog_errno(status);
124 		ret = ERR_PTR(status);
125 		goto bail;
126 	}
127 
128 	status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
129 					    dentry->d_name.len, &blkno);
130 	if (status < 0)
131 		goto bail_add;
132 
133 	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
134 	if (IS_ERR(inode)) {
135 		ret = ERR_PTR(-EACCES);
136 		goto bail_unlock;
137 	}
138 
139 	oi = OCFS2_I(inode);
140 	/* Clear any orphaned state... If we were able to look up the
141 	 * inode from a directory, it certainly can't be orphaned. We
142 	 * might have the bad state from a node which intended to
143 	 * orphan this inode but crashed before it could commit the
144 	 * unlink. */
145 	spin_lock(&oi->ip_lock);
146 	oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
147 	spin_unlock(&oi->ip_lock);
148 
149 bail_add:
150 	ret = d_splice_alias(inode, dentry);
151 
152 	if (inode) {
153 		/*
154 		 * If d_splice_alias() finds a DCACHE_DISCONNECTED
155 		 * dentry, it will d_move() it on top of ourse. The
156 		 * return value will indicate this however, so in
157 		 * those cases, we switch them around for the locking
158 		 * code.
159 		 *
160 		 * NOTE: This dentry already has ->d_op set from
161 		 * ocfs2_get_parent() and ocfs2_get_dentry()
162 		 */
163 		if (ret)
164 			dentry = ret;
165 
166 		status = ocfs2_dentry_attach_lock(dentry, inode,
167 						  OCFS2_I(dir)->ip_blkno);
168 		if (status) {
169 			mlog_errno(status);
170 			ret = ERR_PTR(status);
171 			goto bail_unlock;
172 		}
173 	} else
174 		ocfs2_dentry_attach_gen(dentry);
175 
176 bail_unlock:
177 	/* Don't drop the cluster lock until *after* the d_add --
178 	 * unlink on another node will message us to remove that
179 	 * dentry under this lock so otherwise we can race this with
180 	 * the downconvert thread and have a stale dentry. */
181 	ocfs2_inode_unlock(dir, 0);
182 
183 bail:
184 
185 	mlog_exit_ptr(ret);
186 
187 	return ret;
188 }
189 
190 static struct inode *ocfs2_get_init_inode(struct inode *dir, int mode)
191 {
192 	struct inode *inode;
193 
194 	inode = new_inode(dir->i_sb);
195 	if (!inode) {
196 		mlog(ML_ERROR, "new_inode failed!\n");
197 		return NULL;
198 	}
199 
200 	/* populate as many fields early on as possible - many of
201 	 * these are used by the support functions here and in
202 	 * callers. */
203 	if (S_ISDIR(mode))
204 		inode->i_nlink = 2;
205 	else
206 		inode->i_nlink = 1;
207 	inode_init_owner(inode, dir, mode);
208 	dquot_initialize(inode);
209 	return inode;
210 }
211 
212 static int ocfs2_mknod(struct inode *dir,
213 		       struct dentry *dentry,
214 		       int mode,
215 		       dev_t dev)
216 {
217 	int status = 0;
218 	struct buffer_head *parent_fe_bh = NULL;
219 	handle_t *handle = NULL;
220 	struct ocfs2_super *osb;
221 	struct ocfs2_dinode *dirfe;
222 	struct buffer_head *new_fe_bh = NULL;
223 	struct inode *inode = NULL;
224 	struct ocfs2_alloc_context *inode_ac = NULL;
225 	struct ocfs2_alloc_context *data_ac = NULL;
226 	struct ocfs2_alloc_context *meta_ac = NULL;
227 	int want_clusters = 0;
228 	int want_meta = 0;
229 	int xattr_credits = 0;
230 	struct ocfs2_security_xattr_info si = {
231 		.enable = 1,
232 	};
233 	int did_quota_inode = 0;
234 	struct ocfs2_dir_lookup_result lookup = { NULL, };
235 	sigset_t oldset;
236 	int did_block_signals = 0;
237 
238 	mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry, mode,
239 		   (unsigned long)dev, dentry->d_name.len,
240 		   dentry->d_name.name);
241 
242 	dquot_initialize(dir);
243 
244 	/* get our super block */
245 	osb = OCFS2_SB(dir->i_sb);
246 
247 	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
248 	if (status < 0) {
249 		if (status != -ENOENT)
250 			mlog_errno(status);
251 		return status;
252 	}
253 
254 	if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
255 		status = -EMLINK;
256 		goto leave;
257 	}
258 
259 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
260 	if (!ocfs2_read_links_count(dirfe)) {
261 		/* can't make a file in a deleted directory. */
262 		status = -ENOENT;
263 		goto leave;
264 	}
265 
266 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
267 					   dentry->d_name.len);
268 	if (status)
269 		goto leave;
270 
271 	/* get a spot inside the dir. */
272 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
273 					      dentry->d_name.name,
274 					      dentry->d_name.len, &lookup);
275 	if (status < 0) {
276 		mlog_errno(status);
277 		goto leave;
278 	}
279 
280 	/* reserve an inode spot */
281 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
282 	if (status < 0) {
283 		if (status != -ENOSPC)
284 			mlog_errno(status);
285 		goto leave;
286 	}
287 
288 	inode = ocfs2_get_init_inode(dir, mode);
289 	if (!inode) {
290 		status = -ENOMEM;
291 		mlog_errno(status);
292 		goto leave;
293 	}
294 
295 	/* get security xattr */
296 	status = ocfs2_init_security_get(inode, dir, &si);
297 	if (status) {
298 		if (status == -EOPNOTSUPP)
299 			si.enable = 0;
300 		else {
301 			mlog_errno(status);
302 			goto leave;
303 		}
304 	}
305 
306 	/* calculate meta data/clusters for setting security and acl xattr */
307 	status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
308 				       &si, &want_clusters,
309 				       &xattr_credits, &want_meta);
310 	if (status < 0) {
311 		mlog_errno(status);
312 		goto leave;
313 	}
314 
315 	/* Reserve a cluster if creating an extent based directory. */
316 	if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
317 		want_clusters += 1;
318 
319 		/* Dir indexing requires extra space as well */
320 		if (ocfs2_supports_indexed_dirs(osb))
321 			want_meta++;
322 	}
323 
324 	status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
325 	if (status < 0) {
326 		if (status != -ENOSPC)
327 			mlog_errno(status);
328 		goto leave;
329 	}
330 
331 	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
332 	if (status < 0) {
333 		if (status != -ENOSPC)
334 			mlog_errno(status);
335 		goto leave;
336 	}
337 
338 	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
339 							    S_ISDIR(mode),
340 							    xattr_credits));
341 	if (IS_ERR(handle)) {
342 		status = PTR_ERR(handle);
343 		handle = NULL;
344 		mlog_errno(status);
345 		goto leave;
346 	}
347 
348 	/* Starting to change things, restart is no longer possible. */
349 	ocfs2_block_signals(&oldset);
350 	did_block_signals = 1;
351 
352 	status = dquot_alloc_inode(inode);
353 	if (status)
354 		goto leave;
355 	did_quota_inode = 1;
356 
357 	mlog_entry("(0x%p, 0x%p, %d, %lu, '%.*s')\n", dir, dentry,
358 		   inode->i_mode, (unsigned long)dev, dentry->d_name.len,
359 		   dentry->d_name.name);
360 
361 	/* do the real work now. */
362 	status = ocfs2_mknod_locked(osb, dir, inode, dev,
363 				    &new_fe_bh, parent_fe_bh, handle,
364 				    inode_ac);
365 	if (status < 0) {
366 		mlog_errno(status);
367 		goto leave;
368 	}
369 
370 	if (S_ISDIR(mode)) {
371 		status = ocfs2_fill_new_dir(osb, handle, dir, inode,
372 					    new_fe_bh, data_ac, meta_ac);
373 		if (status < 0) {
374 			mlog_errno(status);
375 			goto leave;
376 		}
377 
378 		status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
379 						 parent_fe_bh,
380 						 OCFS2_JOURNAL_ACCESS_WRITE);
381 		if (status < 0) {
382 			mlog_errno(status);
383 			goto leave;
384 		}
385 		ocfs2_add_links_count(dirfe, 1);
386 		ocfs2_journal_dirty(handle, parent_fe_bh);
387 		inc_nlink(dir);
388 	}
389 
390 	status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
391 				meta_ac, data_ac);
392 	if (status < 0) {
393 		mlog_errno(status);
394 		goto leave;
395 	}
396 
397 	if (si.enable) {
398 		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
399 						 meta_ac, data_ac);
400 		if (status < 0) {
401 			mlog_errno(status);
402 			goto leave;
403 		}
404 	}
405 
406 	/*
407 	 * Do this before adding the entry to the directory. We add
408 	 * also set d_op after success so that ->d_iput() will cleanup
409 	 * the dentry lock even if ocfs2_add_entry() fails below.
410 	 */
411 	status = ocfs2_dentry_attach_lock(dentry, inode,
412 					  OCFS2_I(dir)->ip_blkno);
413 	if (status) {
414 		mlog_errno(status);
415 		goto leave;
416 	}
417 
418 	status = ocfs2_add_entry(handle, dentry, inode,
419 				 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
420 				 &lookup);
421 	if (status < 0) {
422 		mlog_errno(status);
423 		goto leave;
424 	}
425 
426 	insert_inode_hash(inode);
427 	d_instantiate(dentry, inode);
428 	status = 0;
429 leave:
430 	if (status < 0 && did_quota_inode)
431 		dquot_free_inode(inode);
432 	if (handle)
433 		ocfs2_commit_trans(osb, handle);
434 
435 	ocfs2_inode_unlock(dir, 1);
436 	if (did_block_signals)
437 		ocfs2_unblock_signals(&oldset);
438 
439 	if (status == -ENOSPC)
440 		mlog(0, "Disk is full\n");
441 
442 	brelse(new_fe_bh);
443 	brelse(parent_fe_bh);
444 	kfree(si.name);
445 	kfree(si.value);
446 
447 	ocfs2_free_dir_lookup_result(&lookup);
448 
449 	if (inode_ac)
450 		ocfs2_free_alloc_context(inode_ac);
451 
452 	if (data_ac)
453 		ocfs2_free_alloc_context(data_ac);
454 
455 	if (meta_ac)
456 		ocfs2_free_alloc_context(meta_ac);
457 
458 	/*
459 	 * We should call iput after the i_mutex of the bitmap been
460 	 * unlocked in ocfs2_free_alloc_context, or the
461 	 * ocfs2_delete_inode will mutex_lock again.
462 	 */
463 	if ((status < 0) && inode) {
464 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
465 		clear_nlink(inode);
466 		iput(inode);
467 	}
468 
469 	mlog_exit(status);
470 
471 	return status;
472 }
473 
474 static int __ocfs2_mknod_locked(struct inode *dir,
475 				struct inode *inode,
476 				dev_t dev,
477 				struct buffer_head **new_fe_bh,
478 				struct buffer_head *parent_fe_bh,
479 				handle_t *handle,
480 				struct ocfs2_alloc_context *inode_ac,
481 				u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
482 {
483 	int status = 0;
484 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
485 	struct ocfs2_dinode *fe = NULL;
486 	struct ocfs2_extent_list *fel;
487 	u16 feat;
488 
489 	*new_fe_bh = NULL;
490 
491 	/* populate as many fields early on as possible - many of
492 	 * these are used by the support functions here and in
493 	 * callers. */
494 	inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
495 	OCFS2_I(inode)->ip_blkno = fe_blkno;
496 	spin_lock(&osb->osb_lock);
497 	inode->i_generation = osb->s_next_generation++;
498 	spin_unlock(&osb->osb_lock);
499 
500 	*new_fe_bh = sb_getblk(osb->sb, fe_blkno);
501 	if (!*new_fe_bh) {
502 		status = -EIO;
503 		mlog_errno(status);
504 		goto leave;
505 	}
506 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
507 
508 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
509 					 *new_fe_bh,
510 					 OCFS2_JOURNAL_ACCESS_CREATE);
511 	if (status < 0) {
512 		mlog_errno(status);
513 		goto leave;
514 	}
515 
516 	fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
517 	memset(fe, 0, osb->sb->s_blocksize);
518 
519 	fe->i_generation = cpu_to_le32(inode->i_generation);
520 	fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
521 	fe->i_blkno = cpu_to_le64(fe_blkno);
522 	fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
523 	fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
524 	fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
525 	fe->i_uid = cpu_to_le32(inode->i_uid);
526 	fe->i_gid = cpu_to_le32(inode->i_gid);
527 	fe->i_mode = cpu_to_le16(inode->i_mode);
528 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
529 		fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
530 
531 	ocfs2_set_links_count(fe, inode->i_nlink);
532 
533 	fe->i_last_eb_blk = 0;
534 	strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
535 	le32_add_cpu(&fe->i_flags, OCFS2_VALID_FL);
536 	fe->i_atime = fe->i_ctime = fe->i_mtime =
537 		cpu_to_le64(CURRENT_TIME.tv_sec);
538 	fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
539 		cpu_to_le32(CURRENT_TIME.tv_nsec);
540 	fe->i_dtime = 0;
541 
542 	/*
543 	 * If supported, directories start with inline data. If inline
544 	 * isn't supported, but indexing is, we start them as indexed.
545 	 */
546 	feat = le16_to_cpu(fe->i_dyn_features);
547 	if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
548 		fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
549 
550 		fe->id2.i_data.id_count = cpu_to_le16(
551 				ocfs2_max_inline_data_with_xattr(osb->sb, fe));
552 	} else {
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 
559 	ocfs2_journal_dirty(handle, *new_fe_bh);
560 
561 	ocfs2_populate_inode(inode, fe, 1);
562 	ocfs2_ci_set_new(osb, INODE_CACHE(inode));
563 	if (!ocfs2_mount_local(osb)) {
564 		status = ocfs2_create_new_inode_locks(inode);
565 		if (status < 0)
566 			mlog_errno(status);
567 	}
568 
569 	status = 0; /* error in ocfs2_create_new_inode_locks is not
570 		     * critical */
571 
572 leave:
573 	if (status < 0) {
574 		if (*new_fe_bh) {
575 			brelse(*new_fe_bh);
576 			*new_fe_bh = NULL;
577 		}
578 	}
579 
580 	mlog_exit(status);
581 	return status;
582 }
583 
584 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
585 			      struct inode *dir,
586 			      struct inode *inode,
587 			      dev_t dev,
588 			      struct buffer_head **new_fe_bh,
589 			      struct buffer_head *parent_fe_bh,
590 			      handle_t *handle,
591 			      struct ocfs2_alloc_context *inode_ac)
592 {
593 	int status = 0;
594 	u64 suballoc_loc, fe_blkno = 0;
595 	u16 suballoc_bit;
596 
597 	*new_fe_bh = NULL;
598 
599 	status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
600 				       inode_ac, &suballoc_loc,
601 				       &suballoc_bit, &fe_blkno);
602 	if (status < 0) {
603 		mlog_errno(status);
604 		return status;
605 	}
606 
607 	return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
608 				    parent_fe_bh, handle, inode_ac,
609 				    fe_blkno, suballoc_loc, suballoc_bit);
610 }
611 
612 static int ocfs2_mkdir(struct inode *dir,
613 		       struct dentry *dentry,
614 		       int mode)
615 {
616 	int ret;
617 
618 	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
619 		   dentry->d_name.len, dentry->d_name.name);
620 	ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
621 	mlog_exit(ret);
622 
623 	return ret;
624 }
625 
626 static int ocfs2_create(struct inode *dir,
627 			struct dentry *dentry,
628 			int mode,
629 			struct nameidata *nd)
630 {
631 	int ret;
632 
633 	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry, mode,
634 		   dentry->d_name.len, dentry->d_name.name);
635 	ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
636 	mlog_exit(ret);
637 
638 	return ret;
639 }
640 
641 static int ocfs2_link(struct dentry *old_dentry,
642 		      struct inode *dir,
643 		      struct dentry *dentry)
644 {
645 	handle_t *handle;
646 	struct inode *inode = old_dentry->d_inode;
647 	int err;
648 	struct buffer_head *fe_bh = NULL;
649 	struct buffer_head *parent_fe_bh = NULL;
650 	struct ocfs2_dinode *fe = NULL;
651 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
652 	struct ocfs2_dir_lookup_result lookup = { NULL, };
653 	sigset_t oldset;
654 
655 	mlog_entry("(inode=%lu, old='%.*s' new='%.*s')\n", inode->i_ino,
656 		   old_dentry->d_name.len, old_dentry->d_name.name,
657 		   dentry->d_name.len, dentry->d_name.name);
658 
659 	if (S_ISDIR(inode->i_mode))
660 		return -EPERM;
661 
662 	dquot_initialize(dir);
663 
664 	err = ocfs2_inode_lock_nested(dir, &parent_fe_bh, 1, OI_LS_PARENT);
665 	if (err < 0) {
666 		if (err != -ENOENT)
667 			mlog_errno(err);
668 		return err;
669 	}
670 
671 	if (!dir->i_nlink) {
672 		err = -ENOENT;
673 		goto out;
674 	}
675 
676 	err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
677 					dentry->d_name.len);
678 	if (err)
679 		goto out;
680 
681 	err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
682 					   dentry->d_name.name,
683 					   dentry->d_name.len, &lookup);
684 	if (err < 0) {
685 		mlog_errno(err);
686 		goto out;
687 	}
688 
689 	err = ocfs2_inode_lock(inode, &fe_bh, 1);
690 	if (err < 0) {
691 		if (err != -ENOENT)
692 			mlog_errno(err);
693 		goto out;
694 	}
695 
696 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
697 	if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
698 		err = -EMLINK;
699 		goto out_unlock_inode;
700 	}
701 
702 	handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
703 	if (IS_ERR(handle)) {
704 		err = PTR_ERR(handle);
705 		handle = NULL;
706 		mlog_errno(err);
707 		goto out_unlock_inode;
708 	}
709 
710 	/* Starting to change things, restart is no longer possible. */
711 	ocfs2_block_signals(&oldset);
712 
713 	err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
714 				      OCFS2_JOURNAL_ACCESS_WRITE);
715 	if (err < 0) {
716 		mlog_errno(err);
717 		goto out_commit;
718 	}
719 
720 	inc_nlink(inode);
721 	inode->i_ctime = CURRENT_TIME;
722 	ocfs2_set_links_count(fe, inode->i_nlink);
723 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
724 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
725 	ocfs2_journal_dirty(handle, fe_bh);
726 
727 	err = ocfs2_add_entry(handle, dentry, inode,
728 			      OCFS2_I(inode)->ip_blkno,
729 			      parent_fe_bh, &lookup);
730 	if (err) {
731 		ocfs2_add_links_count(fe, -1);
732 		drop_nlink(inode);
733 		mlog_errno(err);
734 		goto out_commit;
735 	}
736 
737 	err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
738 	if (err) {
739 		mlog_errno(err);
740 		goto out_commit;
741 	}
742 
743 	ihold(inode);
744 	d_instantiate(dentry, inode);
745 
746 out_commit:
747 	ocfs2_commit_trans(osb, handle);
748 	ocfs2_unblock_signals(&oldset);
749 out_unlock_inode:
750 	ocfs2_inode_unlock(inode, 1);
751 
752 out:
753 	ocfs2_inode_unlock(dir, 1);
754 
755 	brelse(fe_bh);
756 	brelse(parent_fe_bh);
757 
758 	ocfs2_free_dir_lookup_result(&lookup);
759 
760 	mlog_exit(err);
761 
762 	return err;
763 }
764 
765 /*
766  * Takes and drops an exclusive lock on the given dentry. This will
767  * force other nodes to drop it.
768  */
769 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
770 {
771 	int ret;
772 
773 	ret = ocfs2_dentry_lock(dentry, 1);
774 	if (ret)
775 		mlog_errno(ret);
776 	else
777 		ocfs2_dentry_unlock(dentry, 1);
778 
779 	return ret;
780 }
781 
782 static inline int inode_is_unlinkable(struct inode *inode)
783 {
784 	if (S_ISDIR(inode->i_mode)) {
785 		if (inode->i_nlink == 2)
786 			return 1;
787 		return 0;
788 	}
789 
790 	if (inode->i_nlink == 1)
791 		return 1;
792 	return 0;
793 }
794 
795 static int ocfs2_unlink(struct inode *dir,
796 			struct dentry *dentry)
797 {
798 	int status;
799 	int child_locked = 0;
800 	struct inode *inode = dentry->d_inode;
801 	struct inode *orphan_dir = NULL;
802 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
803 	u64 blkno;
804 	struct ocfs2_dinode *fe = NULL;
805 	struct buffer_head *fe_bh = NULL;
806 	struct buffer_head *parent_node_bh = NULL;
807 	handle_t *handle = NULL;
808 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
809 	struct ocfs2_dir_lookup_result lookup = { NULL, };
810 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
811 
812 	mlog_entry("(0x%p, 0x%p, '%.*s')\n", dir, dentry,
813 		   dentry->d_name.len, dentry->d_name.name);
814 
815 	dquot_initialize(dir);
816 
817 	BUG_ON(dentry->d_parent->d_inode != dir);
818 
819 	mlog(0, "ino = %llu\n", (unsigned long long)OCFS2_I(inode)->ip_blkno);
820 
821 	if (inode == osb->root_inode) {
822 		mlog(0, "Cannot delete the root directory\n");
823 		return -EPERM;
824 	}
825 
826 	status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
827 					 OI_LS_PARENT);
828 	if (status < 0) {
829 		if (status != -ENOENT)
830 			mlog_errno(status);
831 		return status;
832 	}
833 
834 	status = ocfs2_find_files_on_disk(dentry->d_name.name,
835 					  dentry->d_name.len, &blkno, dir,
836 					  &lookup);
837 	if (status < 0) {
838 		if (status != -ENOENT)
839 			mlog_errno(status);
840 		goto leave;
841 	}
842 
843 	if (OCFS2_I(inode)->ip_blkno != blkno) {
844 		status = -ENOENT;
845 
846 		mlog(0, "ip_blkno %llu != dirent blkno %llu ip_flags = %x\n",
847 		     (unsigned long long)OCFS2_I(inode)->ip_blkno,
848 		     (unsigned long long)blkno, OCFS2_I(inode)->ip_flags);
849 		goto leave;
850 	}
851 
852 	status = ocfs2_inode_lock(inode, &fe_bh, 1);
853 	if (status < 0) {
854 		if (status != -ENOENT)
855 			mlog_errno(status);
856 		goto leave;
857 	}
858 	child_locked = 1;
859 
860 	if (S_ISDIR(inode->i_mode)) {
861 		if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
862 			status = -ENOTEMPTY;
863 			goto leave;
864 		}
865 	}
866 
867 	status = ocfs2_remote_dentry_delete(dentry);
868 	if (status < 0) {
869 		/* This remote delete should succeed under all normal
870 		 * circumstances. */
871 		mlog_errno(status);
872 		goto leave;
873 	}
874 
875 	if (inode_is_unlinkable(inode)) {
876 		status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
877 						  OCFS2_I(inode)->ip_blkno,
878 						  orphan_name, &orphan_insert);
879 		if (status < 0) {
880 			mlog_errno(status);
881 			goto leave;
882 		}
883 	}
884 
885 	handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
886 	if (IS_ERR(handle)) {
887 		status = PTR_ERR(handle);
888 		handle = NULL;
889 		mlog_errno(status);
890 		goto leave;
891 	}
892 
893 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
894 					 OCFS2_JOURNAL_ACCESS_WRITE);
895 	if (status < 0) {
896 		mlog_errno(status);
897 		goto leave;
898 	}
899 
900 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
901 
902 	if (inode_is_unlinkable(inode)) {
903 		status = ocfs2_orphan_add(osb, handle, inode, fe_bh, orphan_name,
904 					  &orphan_insert, orphan_dir);
905 		if (status < 0) {
906 			mlog_errno(status);
907 			goto leave;
908 		}
909 	}
910 
911 	/* delete the name from the parent dir */
912 	status = ocfs2_delete_entry(handle, dir, &lookup);
913 	if (status < 0) {
914 		mlog_errno(status);
915 		goto leave;
916 	}
917 
918 	if (S_ISDIR(inode->i_mode))
919 		drop_nlink(inode);
920 	drop_nlink(inode);
921 	ocfs2_set_links_count(fe, inode->i_nlink);
922 	ocfs2_journal_dirty(handle, fe_bh);
923 
924 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
925 	if (S_ISDIR(inode->i_mode))
926 		drop_nlink(dir);
927 
928 	status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
929 	if (status < 0) {
930 		mlog_errno(status);
931 		if (S_ISDIR(inode->i_mode))
932 			inc_nlink(dir);
933 	}
934 
935 leave:
936 	if (handle)
937 		ocfs2_commit_trans(osb, handle);
938 
939 	if (child_locked)
940 		ocfs2_inode_unlock(inode, 1);
941 
942 	ocfs2_inode_unlock(dir, 1);
943 
944 	if (orphan_dir) {
945 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
946 		ocfs2_inode_unlock(orphan_dir, 1);
947 		mutex_unlock(&orphan_dir->i_mutex);
948 		iput(orphan_dir);
949 	}
950 
951 	brelse(fe_bh);
952 	brelse(parent_node_bh);
953 
954 	ocfs2_free_dir_lookup_result(&orphan_insert);
955 	ocfs2_free_dir_lookup_result(&lookup);
956 
957 	mlog_exit(status);
958 
959 	return status;
960 }
961 
962 /*
963  * The only place this should be used is rename!
964  * if they have the same id, then the 1st one is the only one locked.
965  */
966 static int ocfs2_double_lock(struct ocfs2_super *osb,
967 			     struct buffer_head **bh1,
968 			     struct inode *inode1,
969 			     struct buffer_head **bh2,
970 			     struct inode *inode2)
971 {
972 	int status;
973 	struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
974 	struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
975 	struct buffer_head **tmpbh;
976 	struct inode *tmpinode;
977 
978 	mlog_entry("(inode1 = %llu, inode2 = %llu)\n",
979 		   (unsigned long long)oi1->ip_blkno,
980 		   (unsigned long long)oi2->ip_blkno);
981 
982 	if (*bh1)
983 		*bh1 = NULL;
984 	if (*bh2)
985 		*bh2 = NULL;
986 
987 	/* we always want to lock the one with the lower lockid first. */
988 	if (oi1->ip_blkno != oi2->ip_blkno) {
989 		if (oi1->ip_blkno < oi2->ip_blkno) {
990 			/* switch id1 and id2 around */
991 			mlog(0, "switching them around...\n");
992 			tmpbh = bh2;
993 			bh2 = bh1;
994 			bh1 = tmpbh;
995 
996 			tmpinode = inode2;
997 			inode2 = inode1;
998 			inode1 = tmpinode;
999 		}
1000 		/* lock id2 */
1001 		status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1002 						 OI_LS_RENAME1);
1003 		if (status < 0) {
1004 			if (status != -ENOENT)
1005 				mlog_errno(status);
1006 			goto bail;
1007 		}
1008 	}
1009 
1010 	/* lock id1 */
1011 	status = ocfs2_inode_lock_nested(inode1, bh1, 1, OI_LS_RENAME2);
1012 	if (status < 0) {
1013 		/*
1014 		 * An error return must mean that no cluster locks
1015 		 * were held on function exit.
1016 		 */
1017 		if (oi1->ip_blkno != oi2->ip_blkno) {
1018 			ocfs2_inode_unlock(inode2, 1);
1019 			brelse(*bh2);
1020 			*bh2 = NULL;
1021 		}
1022 
1023 		if (status != -ENOENT)
1024 			mlog_errno(status);
1025 	}
1026 
1027 bail:
1028 	mlog_exit(status);
1029 	return status;
1030 }
1031 
1032 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1033 {
1034 	ocfs2_inode_unlock(inode1, 1);
1035 
1036 	if (inode1 != inode2)
1037 		ocfs2_inode_unlock(inode2, 1);
1038 }
1039 
1040 static int ocfs2_rename(struct inode *old_dir,
1041 			struct dentry *old_dentry,
1042 			struct inode *new_dir,
1043 			struct dentry *new_dentry)
1044 {
1045 	int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1046 	int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1047 	struct inode *old_inode = old_dentry->d_inode;
1048 	struct inode *new_inode = new_dentry->d_inode;
1049 	struct inode *orphan_dir = NULL;
1050 	struct ocfs2_dinode *newfe = NULL;
1051 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1052 	struct buffer_head *newfe_bh = NULL;
1053 	struct buffer_head *old_inode_bh = NULL;
1054 	struct ocfs2_super *osb = NULL;
1055 	u64 newfe_blkno, old_de_ino;
1056 	handle_t *handle = NULL;
1057 	struct buffer_head *old_dir_bh = NULL;
1058 	struct buffer_head *new_dir_bh = NULL;
1059 	nlink_t old_dir_nlink = old_dir->i_nlink;
1060 	struct ocfs2_dinode *old_di;
1061 	struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1062 	struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1063 	struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1064 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1065 	struct ocfs2_dir_lookup_result target_insert = { NULL, };
1066 
1067 	/* At some point it might be nice to break this function up a
1068 	 * bit. */
1069 
1070 	mlog_entry("(0x%p, 0x%p, 0x%p, 0x%p, from='%.*s' to='%.*s')\n",
1071 		   old_dir, old_dentry, new_dir, new_dentry,
1072 		   old_dentry->d_name.len, old_dentry->d_name.name,
1073 		   new_dentry->d_name.len, new_dentry->d_name.name);
1074 
1075 	dquot_initialize(old_dir);
1076 	dquot_initialize(new_dir);
1077 
1078 	osb = OCFS2_SB(old_dir->i_sb);
1079 
1080 	if (new_inode) {
1081 		if (!igrab(new_inode))
1082 			BUG();
1083 	}
1084 
1085 	/* Assume a directory hierarchy thusly:
1086 	 * a/b/c
1087 	 * a/d
1088 	 * a,b,c, and d are all directories.
1089 	 *
1090 	 * from cwd of 'a' on both nodes:
1091 	 * node1: mv b/c d
1092 	 * node2: mv d   b/c
1093 	 *
1094 	 * And that's why, just like the VFS, we need a file system
1095 	 * rename lock. */
1096 	if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1097 		status = ocfs2_rename_lock(osb);
1098 		if (status < 0) {
1099 			mlog_errno(status);
1100 			goto bail;
1101 		}
1102 		rename_lock = 1;
1103 	}
1104 
1105 	/* if old and new are the same, this'll just do one lock. */
1106 	status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1107 				   &new_dir_bh, new_dir);
1108 	if (status < 0) {
1109 		mlog_errno(status);
1110 		goto bail;
1111 	}
1112 	parents_locked = 1;
1113 
1114 	/* make sure both dirs have bhs
1115 	 * get an extra ref on old_dir_bh if old==new */
1116 	if (!new_dir_bh) {
1117 		if (old_dir_bh) {
1118 			new_dir_bh = old_dir_bh;
1119 			get_bh(new_dir_bh);
1120 		} else {
1121 			mlog(ML_ERROR, "no old_dir_bh!\n");
1122 			status = -EIO;
1123 			goto bail;
1124 		}
1125 	}
1126 
1127 	/*
1128 	 * Aside from allowing a meta data update, the locking here
1129 	 * also ensures that the downconvert thread on other nodes
1130 	 * won't have to concurrently downconvert the inode and the
1131 	 * dentry locks.
1132 	 */
1133 	status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1134 					 OI_LS_PARENT);
1135 	if (status < 0) {
1136 		if (status != -ENOENT)
1137 			mlog_errno(status);
1138 		goto bail;
1139 	}
1140 	old_child_locked = 1;
1141 
1142 	status = ocfs2_remote_dentry_delete(old_dentry);
1143 	if (status < 0) {
1144 		mlog_errno(status);
1145 		goto bail;
1146 	}
1147 
1148 	if (S_ISDIR(old_inode->i_mode)) {
1149 		u64 old_inode_parent;
1150 
1151 		update_dot_dot = 1;
1152 		status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1153 						  old_inode,
1154 						  &old_inode_dot_dot_res);
1155 		if (status) {
1156 			status = -EIO;
1157 			goto bail;
1158 		}
1159 
1160 		if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1161 			status = -EIO;
1162 			goto bail;
1163 		}
1164 
1165 		if (!new_inode && new_dir != old_dir &&
1166 		    new_dir->i_nlink >= ocfs2_link_max(osb)) {
1167 			status = -EMLINK;
1168 			goto bail;
1169 		}
1170 	}
1171 
1172 	status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1173 					    old_dentry->d_name.len,
1174 					    &old_de_ino);
1175 	if (status) {
1176 		status = -ENOENT;
1177 		goto bail;
1178 	}
1179 
1180 	/*
1181 	 *  Check for inode number is _not_ due to possible IO errors.
1182 	 *  We might rmdir the source, keep it as pwd of some process
1183 	 *  and merrily kill the link to whatever was created under the
1184 	 *  same name. Goodbye sticky bit ;-<
1185 	 */
1186 	if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1187 		status = -ENOENT;
1188 		goto bail;
1189 	}
1190 
1191 	/* check if the target already exists (in which case we need
1192 	 * to delete it */
1193 	status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1194 					  new_dentry->d_name.len,
1195 					  &newfe_blkno, new_dir,
1196 					  &target_lookup_res);
1197 	/* The only error we allow here is -ENOENT because the new
1198 	 * file not existing is perfectly valid. */
1199 	if ((status < 0) && (status != -ENOENT)) {
1200 		/* If we cannot find the file specified we should just */
1201 		/* return the error... */
1202 		mlog_errno(status);
1203 		goto bail;
1204 	}
1205 	if (status == 0)
1206 		target_exists = 1;
1207 
1208 	if (!target_exists && new_inode) {
1209 		/*
1210 		 * Target was unlinked by another node while we were
1211 		 * waiting to get to ocfs2_rename(). There isn't
1212 		 * anything we can do here to help the situation, so
1213 		 * bubble up the appropriate error.
1214 		 */
1215 		status = -ENOENT;
1216 		goto bail;
1217 	}
1218 
1219 	/* In case we need to overwrite an existing file, we blow it
1220 	 * away first */
1221 	if (target_exists) {
1222 		/* VFS didn't think there existed an inode here, but
1223 		 * someone else in the cluster must have raced our
1224 		 * rename to create one. Today we error cleanly, in
1225 		 * the future we should consider calling iget to build
1226 		 * a new struct inode for this entry. */
1227 		if (!new_inode) {
1228 			status = -EACCES;
1229 
1230 			mlog(0, "We found an inode for name %.*s but VFS "
1231 			     "didn't give us one.\n", new_dentry->d_name.len,
1232 			     new_dentry->d_name.name);
1233 			goto bail;
1234 		}
1235 
1236 		if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1237 			status = -EACCES;
1238 
1239 			mlog(0, "Inode %llu and dir %llu disagree. flags = %x\n",
1240 			     (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1241 			     (unsigned long long)newfe_blkno,
1242 			     OCFS2_I(new_inode)->ip_flags);
1243 			goto bail;
1244 		}
1245 
1246 		status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1247 		if (status < 0) {
1248 			if (status != -ENOENT)
1249 				mlog_errno(status);
1250 			goto bail;
1251 		}
1252 		new_child_locked = 1;
1253 
1254 		status = ocfs2_remote_dentry_delete(new_dentry);
1255 		if (status < 0) {
1256 			mlog_errno(status);
1257 			goto bail;
1258 		}
1259 
1260 		newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1261 
1262 		mlog(0, "aha rename over existing... new_blkno=%llu "
1263 		     "newfebh=%p bhblocknr=%llu\n",
1264 		     (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1265 		     (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1266 
1267 		if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1268 			status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1269 						OCFS2_I(new_inode)->ip_blkno,
1270 						orphan_name, &orphan_insert);
1271 			if (status < 0) {
1272 				mlog_errno(status);
1273 				goto bail;
1274 			}
1275 		}
1276 	} else {
1277 		BUG_ON(new_dentry->d_parent->d_inode != new_dir);
1278 
1279 		status = ocfs2_check_dir_for_entry(new_dir,
1280 						   new_dentry->d_name.name,
1281 						   new_dentry->d_name.len);
1282 		if (status)
1283 			goto bail;
1284 
1285 		status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1286 						      new_dentry->d_name.name,
1287 						      new_dentry->d_name.len,
1288 						      &target_insert);
1289 		if (status < 0) {
1290 			mlog_errno(status);
1291 			goto bail;
1292 		}
1293 	}
1294 
1295 	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1296 	if (IS_ERR(handle)) {
1297 		status = PTR_ERR(handle);
1298 		handle = NULL;
1299 		mlog_errno(status);
1300 		goto bail;
1301 	}
1302 
1303 	if (target_exists) {
1304 		if (S_ISDIR(new_inode->i_mode)) {
1305 			if (new_inode->i_nlink != 2 ||
1306 			    !ocfs2_empty_dir(new_inode)) {
1307 				status = -ENOTEMPTY;
1308 				goto bail;
1309 			}
1310 		}
1311 		status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1312 						 newfe_bh,
1313 						 OCFS2_JOURNAL_ACCESS_WRITE);
1314 		if (status < 0) {
1315 			mlog_errno(status);
1316 			goto bail;
1317 		}
1318 
1319 		if (S_ISDIR(new_inode->i_mode) ||
1320 		    (ocfs2_read_links_count(newfe) == 1)) {
1321 			status = ocfs2_orphan_add(osb, handle, new_inode,
1322 						  newfe_bh, orphan_name,
1323 						  &orphan_insert, orphan_dir);
1324 			if (status < 0) {
1325 				mlog_errno(status);
1326 				goto bail;
1327 			}
1328 		}
1329 
1330 		/* change the dirent to point to the correct inode */
1331 		status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1332 					    old_inode);
1333 		if (status < 0) {
1334 			mlog_errno(status);
1335 			goto bail;
1336 		}
1337 		new_dir->i_version++;
1338 
1339 		if (S_ISDIR(new_inode->i_mode))
1340 			ocfs2_set_links_count(newfe, 0);
1341 		else
1342 			ocfs2_add_links_count(newfe, -1);
1343 		ocfs2_journal_dirty(handle, newfe_bh);
1344 	} else {
1345 		/* if the name was not found in new_dir, add it now */
1346 		status = ocfs2_add_entry(handle, new_dentry, old_inode,
1347 					 OCFS2_I(old_inode)->ip_blkno,
1348 					 new_dir_bh, &target_insert);
1349 	}
1350 
1351 	old_inode->i_ctime = CURRENT_TIME;
1352 	mark_inode_dirty(old_inode);
1353 
1354 	status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1355 					 old_inode_bh,
1356 					 OCFS2_JOURNAL_ACCESS_WRITE);
1357 	if (status >= 0) {
1358 		old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1359 
1360 		old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1361 		old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1362 		ocfs2_journal_dirty(handle, old_inode_bh);
1363 	} else
1364 		mlog_errno(status);
1365 
1366 	/*
1367 	 * Now that the name has been added to new_dir, remove the old name.
1368 	 *
1369 	 * We don't keep any directory entry context around until now
1370 	 * because the insert might have changed the type of directory
1371 	 * we're dealing with.
1372 	 */
1373 	status = ocfs2_find_entry(old_dentry->d_name.name,
1374 				  old_dentry->d_name.len, old_dir,
1375 				  &old_entry_lookup);
1376 	if (status)
1377 		goto bail;
1378 
1379 	status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1380 	if (status < 0) {
1381 		mlog_errno(status);
1382 		goto bail;
1383 	}
1384 
1385 	if (new_inode) {
1386 		new_inode->i_nlink--;
1387 		new_inode->i_ctime = CURRENT_TIME;
1388 	}
1389 	old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1390 
1391 	if (update_dot_dot) {
1392 		status = ocfs2_update_entry(old_inode, handle,
1393 					    &old_inode_dot_dot_res, new_dir);
1394 		old_dir->i_nlink--;
1395 		if (new_inode) {
1396 			new_inode->i_nlink--;
1397 		} else {
1398 			inc_nlink(new_dir);
1399 			mark_inode_dirty(new_dir);
1400 		}
1401 	}
1402 	mark_inode_dirty(old_dir);
1403 	ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1404 	if (new_inode) {
1405 		mark_inode_dirty(new_inode);
1406 		ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1407 	}
1408 
1409 	if (old_dir != new_dir) {
1410 		/* Keep the same times on both directories.*/
1411 		new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1412 
1413 		/*
1414 		 * This will also pick up the i_nlink change from the
1415 		 * block above.
1416 		 */
1417 		ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1418 	}
1419 
1420 	if (old_dir_nlink != old_dir->i_nlink) {
1421 		if (!old_dir_bh) {
1422 			mlog(ML_ERROR, "need to change nlink for old dir "
1423 			     "%llu from %d to %d but bh is NULL!\n",
1424 			     (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1425 			     (int)old_dir_nlink, old_dir->i_nlink);
1426 		} else {
1427 			struct ocfs2_dinode *fe;
1428 			status = ocfs2_journal_access_di(handle,
1429 							 INODE_CACHE(old_dir),
1430 							 old_dir_bh,
1431 							 OCFS2_JOURNAL_ACCESS_WRITE);
1432 			fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1433 			ocfs2_set_links_count(fe, old_dir->i_nlink);
1434 			ocfs2_journal_dirty(handle, old_dir_bh);
1435 		}
1436 	}
1437 	ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1438 	status = 0;
1439 bail:
1440 	if (rename_lock)
1441 		ocfs2_rename_unlock(osb);
1442 
1443 	if (handle)
1444 		ocfs2_commit_trans(osb, handle);
1445 
1446 	if (parents_locked)
1447 		ocfs2_double_unlock(old_dir, new_dir);
1448 
1449 	if (old_child_locked)
1450 		ocfs2_inode_unlock(old_inode, 1);
1451 
1452 	if (new_child_locked)
1453 		ocfs2_inode_unlock(new_inode, 1);
1454 
1455 	if (orphan_dir) {
1456 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
1457 		ocfs2_inode_unlock(orphan_dir, 1);
1458 		mutex_unlock(&orphan_dir->i_mutex);
1459 		iput(orphan_dir);
1460 	}
1461 
1462 	if (new_inode)
1463 		sync_mapping_buffers(old_inode->i_mapping);
1464 
1465 	if (new_inode)
1466 		iput(new_inode);
1467 
1468 	ocfs2_free_dir_lookup_result(&target_lookup_res);
1469 	ocfs2_free_dir_lookup_result(&old_entry_lookup);
1470 	ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1471 	ocfs2_free_dir_lookup_result(&orphan_insert);
1472 	ocfs2_free_dir_lookup_result(&target_insert);
1473 
1474 	brelse(newfe_bh);
1475 	brelse(old_inode_bh);
1476 	brelse(old_dir_bh);
1477 	brelse(new_dir_bh);
1478 
1479 	mlog_exit(status);
1480 
1481 	return status;
1482 }
1483 
1484 /*
1485  * we expect i_size = strlen(symname). Copy symname into the file
1486  * data, including the null terminator.
1487  */
1488 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1489 				     handle_t *handle,
1490 				     struct inode *inode,
1491 				     const char *symname)
1492 {
1493 	struct buffer_head **bhs = NULL;
1494 	const char *c;
1495 	struct super_block *sb = osb->sb;
1496 	u64 p_blkno, p_blocks;
1497 	int virtual, blocks, status, i, bytes_left;
1498 
1499 	bytes_left = i_size_read(inode) + 1;
1500 	/* we can't trust i_blocks because we're actually going to
1501 	 * write i_size + 1 bytes. */
1502 	blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1503 
1504 	mlog_entry("i_blocks = %llu, i_size = %llu, blocks = %d\n",
1505 			(unsigned long long)inode->i_blocks,
1506 			i_size_read(inode), blocks);
1507 
1508 	/* Sanity check -- make sure we're going to fit. */
1509 	if (bytes_left >
1510 	    ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1511 		status = -EIO;
1512 		mlog_errno(status);
1513 		goto bail;
1514 	}
1515 
1516 	bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1517 	if (!bhs) {
1518 		status = -ENOMEM;
1519 		mlog_errno(status);
1520 		goto bail;
1521 	}
1522 
1523 	status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1524 					     NULL);
1525 	if (status < 0) {
1526 		mlog_errno(status);
1527 		goto bail;
1528 	}
1529 
1530 	/* links can never be larger than one cluster so we know this
1531 	 * is all going to be contiguous, but do a sanity check
1532 	 * anyway. */
1533 	if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1534 		status = -EIO;
1535 		mlog_errno(status);
1536 		goto bail;
1537 	}
1538 
1539 	virtual = 0;
1540 	while(bytes_left > 0) {
1541 		c = &symname[virtual * sb->s_blocksize];
1542 
1543 		bhs[virtual] = sb_getblk(sb, p_blkno);
1544 		if (!bhs[virtual]) {
1545 			status = -ENOMEM;
1546 			mlog_errno(status);
1547 			goto bail;
1548 		}
1549 		ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1550 					      bhs[virtual]);
1551 
1552 		status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1553 					      bhs[virtual],
1554 					      OCFS2_JOURNAL_ACCESS_CREATE);
1555 		if (status < 0) {
1556 			mlog_errno(status);
1557 			goto bail;
1558 		}
1559 
1560 		memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1561 
1562 		memcpy(bhs[virtual]->b_data, c,
1563 		       (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1564 		       bytes_left);
1565 
1566 		ocfs2_journal_dirty(handle, bhs[virtual]);
1567 
1568 		virtual++;
1569 		p_blkno++;
1570 		bytes_left -= sb->s_blocksize;
1571 	}
1572 
1573 	status = 0;
1574 bail:
1575 
1576 	if (bhs) {
1577 		for(i = 0; i < blocks; i++)
1578 			brelse(bhs[i]);
1579 		kfree(bhs);
1580 	}
1581 
1582 	mlog_exit(status);
1583 	return status;
1584 }
1585 
1586 static int ocfs2_symlink(struct inode *dir,
1587 			 struct dentry *dentry,
1588 			 const char *symname)
1589 {
1590 	int status, l, credits;
1591 	u64 newsize;
1592 	struct ocfs2_super *osb = NULL;
1593 	struct inode *inode = NULL;
1594 	struct super_block *sb;
1595 	struct buffer_head *new_fe_bh = NULL;
1596 	struct buffer_head *parent_fe_bh = NULL;
1597 	struct ocfs2_dinode *fe = NULL;
1598 	struct ocfs2_dinode *dirfe;
1599 	handle_t *handle = NULL;
1600 	struct ocfs2_alloc_context *inode_ac = NULL;
1601 	struct ocfs2_alloc_context *data_ac = NULL;
1602 	struct ocfs2_alloc_context *xattr_ac = NULL;
1603 	int want_clusters = 0;
1604 	int xattr_credits = 0;
1605 	struct ocfs2_security_xattr_info si = {
1606 		.enable = 1,
1607 	};
1608 	int did_quota = 0, did_quota_inode = 0;
1609 	struct ocfs2_dir_lookup_result lookup = { NULL, };
1610 	sigset_t oldset;
1611 	int did_block_signals = 0;
1612 
1613 	mlog_entry("(0x%p, 0x%p, symname='%s' actual='%.*s')\n", dir,
1614 		   dentry, symname, dentry->d_name.len, dentry->d_name.name);
1615 
1616 	dquot_initialize(dir);
1617 
1618 	sb = dir->i_sb;
1619 	osb = OCFS2_SB(sb);
1620 
1621 	l = strlen(symname) + 1;
1622 
1623 	credits = ocfs2_calc_symlink_credits(sb);
1624 
1625 	/* lock the parent directory */
1626 	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1627 	if (status < 0) {
1628 		if (status != -ENOENT)
1629 			mlog_errno(status);
1630 		return status;
1631 	}
1632 
1633 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1634 	if (!ocfs2_read_links_count(dirfe)) {
1635 		/* can't make a file in a deleted directory. */
1636 		status = -ENOENT;
1637 		goto bail;
1638 	}
1639 
1640 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1641 					   dentry->d_name.len);
1642 	if (status)
1643 		goto bail;
1644 
1645 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1646 					      dentry->d_name.name,
1647 					      dentry->d_name.len, &lookup);
1648 	if (status < 0) {
1649 		mlog_errno(status);
1650 		goto bail;
1651 	}
1652 
1653 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
1654 	if (status < 0) {
1655 		if (status != -ENOSPC)
1656 			mlog_errno(status);
1657 		goto bail;
1658 	}
1659 
1660 	inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1661 	if (!inode) {
1662 		status = -ENOMEM;
1663 		mlog_errno(status);
1664 		goto bail;
1665 	}
1666 
1667 	/* get security xattr */
1668 	status = ocfs2_init_security_get(inode, dir, &si);
1669 	if (status) {
1670 		if (status == -EOPNOTSUPP)
1671 			si.enable = 0;
1672 		else {
1673 			mlog_errno(status);
1674 			goto bail;
1675 		}
1676 	}
1677 
1678 	/* calculate meta data/clusters for setting security xattr */
1679 	if (si.enable) {
1680 		status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1681 						  &xattr_credits, &xattr_ac);
1682 		if (status < 0) {
1683 			mlog_errno(status);
1684 			goto bail;
1685 		}
1686 	}
1687 
1688 	/* don't reserve bitmap space for fast symlinks. */
1689 	if (l > ocfs2_fast_symlink_chars(sb))
1690 		want_clusters += 1;
1691 
1692 	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1693 	if (status < 0) {
1694 		if (status != -ENOSPC)
1695 			mlog_errno(status);
1696 		goto bail;
1697 	}
1698 
1699 	handle = ocfs2_start_trans(osb, credits + xattr_credits);
1700 	if (IS_ERR(handle)) {
1701 		status = PTR_ERR(handle);
1702 		handle = NULL;
1703 		mlog_errno(status);
1704 		goto bail;
1705 	}
1706 
1707 	/* Starting to change things, restart is no longer possible. */
1708 	ocfs2_block_signals(&oldset);
1709 	did_block_signals = 1;
1710 
1711 	status = dquot_alloc_inode(inode);
1712 	if (status)
1713 		goto bail;
1714 	did_quota_inode = 1;
1715 
1716 	mlog_entry("(0x%p, 0x%p, %d, '%.*s')\n", dir, dentry,
1717 		   inode->i_mode, dentry->d_name.len,
1718 		   dentry->d_name.name);
1719 
1720 	status = ocfs2_mknod_locked(osb, dir, inode,
1721 				    0, &new_fe_bh, parent_fe_bh, handle,
1722 				    inode_ac);
1723 	if (status < 0) {
1724 		mlog_errno(status);
1725 		goto bail;
1726 	}
1727 
1728 	fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1729 	inode->i_rdev = 0;
1730 	newsize = l - 1;
1731 	if (l > ocfs2_fast_symlink_chars(sb)) {
1732 		u32 offset = 0;
1733 
1734 		inode->i_op = &ocfs2_symlink_inode_operations;
1735 		status = dquot_alloc_space_nodirty(inode,
1736 		    ocfs2_clusters_to_bytes(osb->sb, 1));
1737 		if (status)
1738 			goto bail;
1739 		did_quota = 1;
1740 		status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1741 					      new_fe_bh,
1742 					      handle, data_ac, NULL,
1743 					      NULL);
1744 		if (status < 0) {
1745 			if (status != -ENOSPC && status != -EINTR) {
1746 				mlog(ML_ERROR,
1747 				     "Failed to extend file to %llu\n",
1748 				     (unsigned long long)newsize);
1749 				mlog_errno(status);
1750 				status = -ENOSPC;
1751 			}
1752 			goto bail;
1753 		}
1754 		i_size_write(inode, newsize);
1755 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1756 	} else {
1757 		inode->i_op = &ocfs2_fast_symlink_inode_operations;
1758 		memcpy((char *) fe->id2.i_symlink, symname, l);
1759 		i_size_write(inode, newsize);
1760 		inode->i_blocks = 0;
1761 	}
1762 
1763 	status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1764 	if (status < 0) {
1765 		mlog_errno(status);
1766 		goto bail;
1767 	}
1768 
1769 	if (!ocfs2_inode_is_fast_symlink(inode)) {
1770 		status = ocfs2_create_symlink_data(osb, handle, inode,
1771 						   symname);
1772 		if (status < 0) {
1773 			mlog_errno(status);
1774 			goto bail;
1775 		}
1776 	}
1777 
1778 	if (si.enable) {
1779 		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1780 						 xattr_ac, data_ac);
1781 		if (status < 0) {
1782 			mlog_errno(status);
1783 			goto bail;
1784 		}
1785 	}
1786 
1787 	/*
1788 	 * Do this before adding the entry to the directory. We add
1789 	 * also set d_op after success so that ->d_iput() will cleanup
1790 	 * the dentry lock even if ocfs2_add_entry() fails below.
1791 	 */
1792 	status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1793 	if (status) {
1794 		mlog_errno(status);
1795 		goto bail;
1796 	}
1797 
1798 	status = ocfs2_add_entry(handle, dentry, inode,
1799 				 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1800 				 &lookup);
1801 	if (status < 0) {
1802 		mlog_errno(status);
1803 		goto bail;
1804 	}
1805 
1806 	insert_inode_hash(inode);
1807 	d_instantiate(dentry, inode);
1808 bail:
1809 	if (status < 0 && did_quota)
1810 		dquot_free_space_nodirty(inode,
1811 					ocfs2_clusters_to_bytes(osb->sb, 1));
1812 	if (status < 0 && did_quota_inode)
1813 		dquot_free_inode(inode);
1814 	if (handle)
1815 		ocfs2_commit_trans(osb, handle);
1816 
1817 	ocfs2_inode_unlock(dir, 1);
1818 	if (did_block_signals)
1819 		ocfs2_unblock_signals(&oldset);
1820 
1821 	brelse(new_fe_bh);
1822 	brelse(parent_fe_bh);
1823 	kfree(si.name);
1824 	kfree(si.value);
1825 	ocfs2_free_dir_lookup_result(&lookup);
1826 	if (inode_ac)
1827 		ocfs2_free_alloc_context(inode_ac);
1828 	if (data_ac)
1829 		ocfs2_free_alloc_context(data_ac);
1830 	if (xattr_ac)
1831 		ocfs2_free_alloc_context(xattr_ac);
1832 	if ((status < 0) && inode) {
1833 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
1834 		clear_nlink(inode);
1835 		iput(inode);
1836 	}
1837 
1838 	mlog_exit(status);
1839 
1840 	return status;
1841 }
1842 
1843 static int ocfs2_blkno_stringify(u64 blkno, char *name)
1844 {
1845 	int status, namelen;
1846 
1847 	mlog_entry_void();
1848 
1849 	namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
1850 			   (long long)blkno);
1851 	if (namelen <= 0) {
1852 		if (namelen)
1853 			status = namelen;
1854 		else
1855 			status = -EINVAL;
1856 		mlog_errno(status);
1857 		goto bail;
1858 	}
1859 	if (namelen != OCFS2_ORPHAN_NAMELEN) {
1860 		status = -EINVAL;
1861 		mlog_errno(status);
1862 		goto bail;
1863 	}
1864 
1865 	mlog(0, "built filename '%s' for orphan dir (len=%d)\n", name,
1866 	     namelen);
1867 
1868 	status = 0;
1869 bail:
1870 	mlog_exit(status);
1871 	return status;
1872 }
1873 
1874 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
1875 					struct inode **ret_orphan_dir,
1876 					struct buffer_head **ret_orphan_dir_bh)
1877 {
1878 	struct inode *orphan_dir_inode;
1879 	struct buffer_head *orphan_dir_bh = NULL;
1880 	int ret = 0;
1881 
1882 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
1883 						       ORPHAN_DIR_SYSTEM_INODE,
1884 						       osb->slot_num);
1885 	if (!orphan_dir_inode) {
1886 		ret = -ENOENT;
1887 		mlog_errno(ret);
1888 		return ret;
1889 	}
1890 
1891 	mutex_lock(&orphan_dir_inode->i_mutex);
1892 
1893 	ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
1894 	if (ret < 0) {
1895 		mutex_unlock(&orphan_dir_inode->i_mutex);
1896 		iput(orphan_dir_inode);
1897 
1898 		mlog_errno(ret);
1899 		return ret;
1900 	}
1901 
1902 	*ret_orphan_dir = orphan_dir_inode;
1903 	*ret_orphan_dir_bh = orphan_dir_bh;
1904 
1905 	return 0;
1906 }
1907 
1908 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
1909 				      struct buffer_head *orphan_dir_bh,
1910 				      u64 blkno,
1911 				      char *name,
1912 				      struct ocfs2_dir_lookup_result *lookup)
1913 {
1914 	int ret;
1915 	struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
1916 
1917 	ret = ocfs2_blkno_stringify(blkno, name);
1918 	if (ret < 0) {
1919 		mlog_errno(ret);
1920 		return ret;
1921 	}
1922 
1923 	ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
1924 					   orphan_dir_bh, name,
1925 					   OCFS2_ORPHAN_NAMELEN, lookup);
1926 	if (ret < 0) {
1927 		mlog_errno(ret);
1928 		return ret;
1929 	}
1930 
1931 	return 0;
1932 }
1933 
1934 /**
1935  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
1936  * insertion of an orphan.
1937  * @osb: ocfs2 file system
1938  * @ret_orphan_dir: Orphan dir inode - returned locked!
1939  * @blkno: Actual block number of the inode to be inserted into orphan dir.
1940  * @lookup: dir lookup result, to be passed back into functions like
1941  *          ocfs2_orphan_add
1942  *
1943  * Returns zero on success and the ret_orphan_dir, name and lookup
1944  * fields will be populated.
1945  *
1946  * Returns non-zero on failure.
1947  */
1948 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
1949 				    struct inode **ret_orphan_dir,
1950 				    u64 blkno,
1951 				    char *name,
1952 				    struct ocfs2_dir_lookup_result *lookup)
1953 {
1954 	struct inode *orphan_dir_inode = NULL;
1955 	struct buffer_head *orphan_dir_bh = NULL;
1956 	int ret = 0;
1957 
1958 	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
1959 					   &orphan_dir_bh);
1960 	if (ret < 0) {
1961 		mlog_errno(ret);
1962 		return ret;
1963 	}
1964 
1965 	ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
1966 					 blkno, name, lookup);
1967 	if (ret < 0) {
1968 		mlog_errno(ret);
1969 		goto out;
1970 	}
1971 
1972 	*ret_orphan_dir = orphan_dir_inode;
1973 
1974 out:
1975 	brelse(orphan_dir_bh);
1976 
1977 	if (ret) {
1978 		ocfs2_inode_unlock(orphan_dir_inode, 1);
1979 		mutex_unlock(&orphan_dir_inode->i_mutex);
1980 		iput(orphan_dir_inode);
1981 	}
1982 
1983 	mlog_exit(ret);
1984 	return ret;
1985 }
1986 
1987 static int ocfs2_orphan_add(struct ocfs2_super *osb,
1988 			    handle_t *handle,
1989 			    struct inode *inode,
1990 			    struct buffer_head *fe_bh,
1991 			    char *name,
1992 			    struct ocfs2_dir_lookup_result *lookup,
1993 			    struct inode *orphan_dir_inode)
1994 {
1995 	struct buffer_head *orphan_dir_bh = NULL;
1996 	int status = 0;
1997 	struct ocfs2_dinode *orphan_fe;
1998 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
1999 
2000 	mlog_entry("(inode->i_ino = %lu)\n", inode->i_ino);
2001 
2002 	status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2003 	if (status < 0) {
2004 		mlog_errno(status);
2005 		goto leave;
2006 	}
2007 
2008 	status = ocfs2_journal_access_di(handle,
2009 					 INODE_CACHE(orphan_dir_inode),
2010 					 orphan_dir_bh,
2011 					 OCFS2_JOURNAL_ACCESS_WRITE);
2012 	if (status < 0) {
2013 		mlog_errno(status);
2014 		goto leave;
2015 	}
2016 
2017 	/* we're a cluster, and nlink can change on disk from
2018 	 * underneath us... */
2019 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2020 	if (S_ISDIR(inode->i_mode))
2021 		ocfs2_add_links_count(orphan_fe, 1);
2022 	orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2023 	ocfs2_journal_dirty(handle, orphan_dir_bh);
2024 
2025 	status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2026 				   OCFS2_ORPHAN_NAMELEN, inode,
2027 				   OCFS2_I(inode)->ip_blkno,
2028 				   orphan_dir_bh, lookup);
2029 	if (status < 0) {
2030 		mlog_errno(status);
2031 		goto leave;
2032 	}
2033 
2034 	/*
2035 	 * We're going to journal the change of i_flags and i_orphaned_slot.
2036 	 * It's safe anyway, though some callers may duplicate the journaling.
2037 	 * Journaling within the func just make the logic look more
2038 	 * straightforward.
2039 	 */
2040 	status = ocfs2_journal_access_di(handle,
2041 					 INODE_CACHE(inode),
2042 					 fe_bh,
2043 					 OCFS2_JOURNAL_ACCESS_WRITE);
2044 	if (status < 0) {
2045 		mlog_errno(status);
2046 		goto leave;
2047 	}
2048 
2049 	le32_add_cpu(&fe->i_flags, OCFS2_ORPHANED_FL);
2050 	OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2051 
2052 	/* Record which orphan dir our inode now resides
2053 	 * in. delete_inode will use this to determine which orphan
2054 	 * dir to lock. */
2055 	fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2056 
2057 	ocfs2_journal_dirty(handle, fe_bh);
2058 
2059 	mlog(0, "Inode %llu orphaned in slot %d\n",
2060 	     (unsigned long long)OCFS2_I(inode)->ip_blkno, osb->slot_num);
2061 
2062 leave:
2063 	brelse(orphan_dir_bh);
2064 
2065 	mlog_exit(status);
2066 	return status;
2067 }
2068 
2069 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2070 int ocfs2_orphan_del(struct ocfs2_super *osb,
2071 		     handle_t *handle,
2072 		     struct inode *orphan_dir_inode,
2073 		     struct inode *inode,
2074 		     struct buffer_head *orphan_dir_bh)
2075 {
2076 	char name[OCFS2_ORPHAN_NAMELEN + 1];
2077 	struct ocfs2_dinode *orphan_fe;
2078 	int status = 0;
2079 	struct ocfs2_dir_lookup_result lookup = { NULL, };
2080 
2081 	mlog_entry_void();
2082 
2083 	status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2084 	if (status < 0) {
2085 		mlog_errno(status);
2086 		goto leave;
2087 	}
2088 
2089 	mlog(0, "removing '%s' from orphan dir %llu (namelen=%d)\n",
2090 	     name, (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2091 	     OCFS2_ORPHAN_NAMELEN);
2092 
2093 	/* find it's spot in the orphan directory */
2094 	status = ocfs2_find_entry(name, OCFS2_ORPHAN_NAMELEN, orphan_dir_inode,
2095 				  &lookup);
2096 	if (status) {
2097 		mlog_errno(status);
2098 		goto leave;
2099 	}
2100 
2101 	/* remove it from the orphan directory */
2102 	status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2103 	if (status < 0) {
2104 		mlog_errno(status);
2105 		goto leave;
2106 	}
2107 
2108 	status = ocfs2_journal_access_di(handle,
2109 					 INODE_CACHE(orphan_dir_inode),
2110 					 orphan_dir_bh,
2111 					 OCFS2_JOURNAL_ACCESS_WRITE);
2112 	if (status < 0) {
2113 		mlog_errno(status);
2114 		goto leave;
2115 	}
2116 
2117 	/* do the i_nlink dance! :) */
2118 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2119 	if (S_ISDIR(inode->i_mode))
2120 		ocfs2_add_links_count(orphan_fe, -1);
2121 	orphan_dir_inode->i_nlink = ocfs2_read_links_count(orphan_fe);
2122 	ocfs2_journal_dirty(handle, orphan_dir_bh);
2123 
2124 leave:
2125 	ocfs2_free_dir_lookup_result(&lookup);
2126 
2127 	mlog_exit(status);
2128 	return status;
2129 }
2130 
2131 /**
2132  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to recieve a newly
2133  * allocated file. This is different from the typical 'add to orphan dir'
2134  * operation in that the inode does not yet exist. This is a problem because
2135  * the orphan dir stringifies the inode block number to come up with it's
2136  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2137  * problem. This function works around it by calling deeper into the orphan
2138  * and suballoc code than other callers. Use this only by necessity.
2139  * @dir: The directory which this inode will ultimately wind up under - not the
2140  * orphan dir!
2141  * @dir_bh: buffer_head the @dir inode block
2142  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2143  * with the string to be used for orphan dirent. Pass back to the orphan dir
2144  * code.
2145  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2146  * dir code.
2147  * @ret_di_blkno: block number where the new inode will be allocated.
2148  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2149  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2150  *
2151  * Returns zero on success and the ret_orphan_dir, name and lookup
2152  * fields will be populated.
2153  *
2154  * Returns non-zero on failure.
2155  */
2156 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2157 					struct buffer_head *dir_bh,
2158 					char *orphan_name,
2159 					struct inode **ret_orphan_dir,
2160 					u64 *ret_di_blkno,
2161 					struct ocfs2_dir_lookup_result *orphan_insert,
2162 					struct ocfs2_alloc_context **ret_inode_ac)
2163 {
2164 	int ret;
2165 	u64 di_blkno;
2166 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2167 	struct inode *orphan_dir = NULL;
2168 	struct buffer_head *orphan_dir_bh = NULL;
2169 	struct ocfs2_alloc_context *inode_ac = NULL;
2170 
2171 	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2172 	if (ret < 0) {
2173 		mlog_errno(ret);
2174 		return ret;
2175 	}
2176 
2177 	/* reserve an inode spot */
2178 	ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2179 	if (ret < 0) {
2180 		if (ret != -ENOSPC)
2181 			mlog_errno(ret);
2182 		goto out;
2183 	}
2184 
2185 	ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2186 				       &di_blkno);
2187 	if (ret) {
2188 		mlog_errno(ret);
2189 		goto out;
2190 	}
2191 
2192 	ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2193 					 di_blkno, orphan_name, orphan_insert);
2194 	if (ret < 0) {
2195 		mlog_errno(ret);
2196 		goto out;
2197 	}
2198 
2199 out:
2200 	if (ret == 0) {
2201 		*ret_orphan_dir = orphan_dir;
2202 		*ret_di_blkno = di_blkno;
2203 		*ret_inode_ac = inode_ac;
2204 		/*
2205 		 * orphan_name and orphan_insert are already up to
2206 		 * date via prepare_orphan_dir
2207 		 */
2208 	} else {
2209 		/* Unroll reserve_new_inode* */
2210 		if (inode_ac)
2211 			ocfs2_free_alloc_context(inode_ac);
2212 
2213 		/* Unroll orphan dir locking */
2214 		mutex_unlock(&orphan_dir->i_mutex);
2215 		ocfs2_inode_unlock(orphan_dir, 1);
2216 		iput(orphan_dir);
2217 	}
2218 
2219 	brelse(orphan_dir_bh);
2220 
2221 	return 0;
2222 }
2223 
2224 int ocfs2_create_inode_in_orphan(struct inode *dir,
2225 				 int mode,
2226 				 struct inode **new_inode)
2227 {
2228 	int status, did_quota_inode = 0;
2229 	struct inode *inode = NULL;
2230 	struct inode *orphan_dir = NULL;
2231 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2232 	struct ocfs2_dinode *di = NULL;
2233 	handle_t *handle = NULL;
2234 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2235 	struct buffer_head *parent_di_bh = NULL;
2236 	struct buffer_head *new_di_bh = NULL;
2237 	struct ocfs2_alloc_context *inode_ac = NULL;
2238 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2239 	u64 uninitialized_var(di_blkno), suballoc_loc;
2240 	u16 suballoc_bit;
2241 
2242 	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2243 	if (status < 0) {
2244 		if (status != -ENOENT)
2245 			mlog_errno(status);
2246 		return status;
2247 	}
2248 
2249 	status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2250 					      orphan_name, &orphan_dir,
2251 					      &di_blkno, &orphan_insert, &inode_ac);
2252 	if (status < 0) {
2253 		if (status != -ENOSPC)
2254 			mlog_errno(status);
2255 		goto leave;
2256 	}
2257 
2258 	inode = ocfs2_get_init_inode(dir, mode);
2259 	if (!inode) {
2260 		status = -ENOMEM;
2261 		mlog_errno(status);
2262 		goto leave;
2263 	}
2264 
2265 	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2266 	if (IS_ERR(handle)) {
2267 		status = PTR_ERR(handle);
2268 		handle = NULL;
2269 		mlog_errno(status);
2270 		goto leave;
2271 	}
2272 
2273 	status = dquot_alloc_inode(inode);
2274 	if (status)
2275 		goto leave;
2276 	did_quota_inode = 1;
2277 
2278 	status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2279 					      &suballoc_loc,
2280 					      &suballoc_bit, di_blkno);
2281 	if (status < 0) {
2282 		mlog_errno(status);
2283 		goto leave;
2284 	}
2285 
2286 	inode->i_nlink = 0;
2287 	/* do the real work now. */
2288 	status = __ocfs2_mknod_locked(dir, inode,
2289 				      0, &new_di_bh, parent_di_bh, handle,
2290 				      inode_ac, di_blkno, suballoc_loc,
2291 				      suballoc_bit);
2292 	if (status < 0) {
2293 		mlog_errno(status);
2294 		goto leave;
2295 	}
2296 
2297 	di = (struct ocfs2_dinode *)new_di_bh->b_data;
2298 	status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2299 				  &orphan_insert, orphan_dir);
2300 	if (status < 0) {
2301 		mlog_errno(status);
2302 		goto leave;
2303 	}
2304 
2305 	/* get open lock so that only nodes can't remove it from orphan dir. */
2306 	status = ocfs2_open_lock(inode);
2307 	if (status < 0)
2308 		mlog_errno(status);
2309 
2310 	insert_inode_hash(inode);
2311 leave:
2312 	if (status < 0 && did_quota_inode)
2313 		dquot_free_inode(inode);
2314 	if (handle)
2315 		ocfs2_commit_trans(osb, handle);
2316 
2317 	if (orphan_dir) {
2318 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
2319 		ocfs2_inode_unlock(orphan_dir, 1);
2320 		mutex_unlock(&orphan_dir->i_mutex);
2321 		iput(orphan_dir);
2322 	}
2323 
2324 	if (status == -ENOSPC)
2325 		mlog(0, "Disk is full\n");
2326 
2327 	if ((status < 0) && inode) {
2328 		clear_nlink(inode);
2329 		iput(inode);
2330 	}
2331 
2332 	if (inode_ac)
2333 		ocfs2_free_alloc_context(inode_ac);
2334 
2335 	brelse(new_di_bh);
2336 
2337 	if (!status)
2338 		*new_inode = inode;
2339 
2340 	ocfs2_free_dir_lookup_result(&orphan_insert);
2341 
2342 	ocfs2_inode_unlock(dir, 1);
2343 	brelse(parent_di_bh);
2344 	return status;
2345 }
2346 
2347 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2348 				   struct inode *inode,
2349 				   struct dentry *dentry)
2350 {
2351 	int status = 0;
2352 	struct buffer_head *parent_di_bh = NULL;
2353 	handle_t *handle = NULL;
2354 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2355 	struct ocfs2_dinode *dir_di, *di;
2356 	struct inode *orphan_dir_inode = NULL;
2357 	struct buffer_head *orphan_dir_bh = NULL;
2358 	struct buffer_head *di_bh = NULL;
2359 	struct ocfs2_dir_lookup_result lookup = { NULL, };
2360 
2361 	mlog_entry("(0x%p, 0x%p, %.*s')\n", dir, dentry,
2362 		   dentry->d_name.len, dentry->d_name.name);
2363 
2364 	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2365 	if (status < 0) {
2366 		if (status != -ENOENT)
2367 			mlog_errno(status);
2368 		return status;
2369 	}
2370 
2371 	dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2372 	if (!dir_di->i_links_count) {
2373 		/* can't make a file in a deleted directory. */
2374 		status = -ENOENT;
2375 		goto leave;
2376 	}
2377 
2378 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2379 					   dentry->d_name.len);
2380 	if (status)
2381 		goto leave;
2382 
2383 	/* get a spot inside the dir. */
2384 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2385 					      dentry->d_name.name,
2386 					      dentry->d_name.len, &lookup);
2387 	if (status < 0) {
2388 		mlog_errno(status);
2389 		goto leave;
2390 	}
2391 
2392 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2393 						       ORPHAN_DIR_SYSTEM_INODE,
2394 						       osb->slot_num);
2395 	if (!orphan_dir_inode) {
2396 		status = -EEXIST;
2397 		mlog_errno(status);
2398 		goto leave;
2399 	}
2400 
2401 	mutex_lock(&orphan_dir_inode->i_mutex);
2402 
2403 	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2404 	if (status < 0) {
2405 		mlog_errno(status);
2406 		mutex_unlock(&orphan_dir_inode->i_mutex);
2407 		iput(orphan_dir_inode);
2408 		goto leave;
2409 	}
2410 
2411 	status = ocfs2_read_inode_block(inode, &di_bh);
2412 	if (status < 0) {
2413 		mlog_errno(status);
2414 		goto orphan_unlock;
2415 	}
2416 
2417 	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2418 	if (IS_ERR(handle)) {
2419 		status = PTR_ERR(handle);
2420 		handle = NULL;
2421 		mlog_errno(status);
2422 		goto orphan_unlock;
2423 	}
2424 
2425 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2426 					 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2427 	if (status < 0) {
2428 		mlog_errno(status);
2429 		goto out_commit;
2430 	}
2431 
2432 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2433 				  orphan_dir_bh);
2434 	if (status < 0) {
2435 		mlog_errno(status);
2436 		goto out_commit;
2437 	}
2438 
2439 	di = (struct ocfs2_dinode *)di_bh->b_data;
2440 	le32_add_cpu(&di->i_flags, -OCFS2_ORPHANED_FL);
2441 	di->i_orphaned_slot = 0;
2442 	inode->i_nlink = 1;
2443 	ocfs2_set_links_count(di, inode->i_nlink);
2444 	ocfs2_journal_dirty(handle, di_bh);
2445 
2446 	status = ocfs2_add_entry(handle, dentry, inode,
2447 				 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2448 				 &lookup);
2449 	if (status < 0) {
2450 		mlog_errno(status);
2451 		goto out_commit;
2452 	}
2453 
2454 	status = ocfs2_dentry_attach_lock(dentry, inode,
2455 					  OCFS2_I(dir)->ip_blkno);
2456 	if (status) {
2457 		mlog_errno(status);
2458 		goto out_commit;
2459 	}
2460 
2461 	d_instantiate(dentry, inode);
2462 	status = 0;
2463 out_commit:
2464 	ocfs2_commit_trans(osb, handle);
2465 orphan_unlock:
2466 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2467 	mutex_unlock(&orphan_dir_inode->i_mutex);
2468 	iput(orphan_dir_inode);
2469 leave:
2470 
2471 	ocfs2_inode_unlock(dir, 1);
2472 
2473 	brelse(di_bh);
2474 	brelse(parent_di_bh);
2475 	brelse(orphan_dir_bh);
2476 
2477 	ocfs2_free_dir_lookup_result(&lookup);
2478 
2479 	mlog_exit(status);
2480 
2481 	return status;
2482 }
2483 
2484 const struct inode_operations ocfs2_dir_iops = {
2485 	.create		= ocfs2_create,
2486 	.lookup		= ocfs2_lookup,
2487 	.link		= ocfs2_link,
2488 	.unlink		= ocfs2_unlink,
2489 	.rmdir		= ocfs2_unlink,
2490 	.symlink	= ocfs2_symlink,
2491 	.mkdir		= ocfs2_mkdir,
2492 	.mknod		= ocfs2_mknod,
2493 	.rename		= ocfs2_rename,
2494 	.setattr	= ocfs2_setattr,
2495 	.getattr	= ocfs2_getattr,
2496 	.permission	= ocfs2_permission,
2497 	.setxattr	= generic_setxattr,
2498 	.getxattr	= generic_getxattr,
2499 	.listxattr	= ocfs2_listxattr,
2500 	.removexattr	= generic_removexattr,
2501 	.fiemap         = ocfs2_fiemap,
2502 };
2503