xref: /openbmc/linux/fs/ocfs2/namei.c (revision 4bce6fce)
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 #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 "xattr.h"
64 #include "acl.h"
65 #include "ocfs2_trace.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 				    bool dio);
84 
85 static int ocfs2_orphan_add(struct ocfs2_super *osb,
86 			    handle_t *handle,
87 			    struct inode *inode,
88 			    struct buffer_head *fe_bh,
89 			    char *name,
90 			    struct ocfs2_dir_lookup_result *lookup,
91 			    struct inode *orphan_dir_inode,
92 			    bool dio);
93 
94 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
95 				     handle_t *handle,
96 				     struct inode *inode,
97 				     const char *symname);
98 
99 static int ocfs2_double_lock(struct ocfs2_super *osb,
100 			     struct buffer_head **bh1,
101 			     struct inode *inode1,
102 			     struct buffer_head **bh2,
103 			     struct inode *inode2,
104 			     int rename);
105 
106 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
107 /* An orphan dir name is an 8 byte value, printed as a hex string */
108 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
109 #define OCFS2_DIO_ORPHAN_PREFIX "dio-"
110 #define OCFS2_DIO_ORPHAN_PREFIX_LEN 4
111 
112 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
113 				   unsigned int flags)
114 {
115 	int status;
116 	u64 blkno;
117 	struct inode *inode = NULL;
118 	struct dentry *ret;
119 	struct ocfs2_inode_info *oi;
120 
121 	trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
122 			   dentry->d_name.name,
123 			   (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
124 
125 	if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
126 		ret = ERR_PTR(-ENAMETOOLONG);
127 		goto bail;
128 	}
129 
130 	status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
131 	if (status < 0) {
132 		if (status != -ENOENT)
133 			mlog_errno(status);
134 		ret = ERR_PTR(status);
135 		goto bail;
136 	}
137 
138 	status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
139 					    dentry->d_name.len, &blkno);
140 	if (status < 0)
141 		goto bail_add;
142 
143 	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
144 	if (IS_ERR(inode)) {
145 		ret = ERR_PTR(-EACCES);
146 		goto bail_unlock;
147 	}
148 
149 	oi = OCFS2_I(inode);
150 	/* Clear any orphaned state... If we were able to look up the
151 	 * inode from a directory, it certainly can't be orphaned. We
152 	 * might have the bad state from a node which intended to
153 	 * orphan this inode but crashed before it could commit the
154 	 * unlink. */
155 	spin_lock(&oi->ip_lock);
156 	oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
157 	spin_unlock(&oi->ip_lock);
158 
159 bail_add:
160 	ret = d_splice_alias(inode, dentry);
161 
162 	if (inode) {
163 		/*
164 		 * If d_splice_alias() finds a DCACHE_DISCONNECTED
165 		 * dentry, it will d_move() it on top of ourse. The
166 		 * return value will indicate this however, so in
167 		 * those cases, we switch them around for the locking
168 		 * code.
169 		 *
170 		 * NOTE: This dentry already has ->d_op set from
171 		 * ocfs2_get_parent() and ocfs2_get_dentry()
172 		 */
173 		if (!IS_ERR_OR_NULL(ret))
174 			dentry = ret;
175 
176 		status = ocfs2_dentry_attach_lock(dentry, inode,
177 						  OCFS2_I(dir)->ip_blkno);
178 		if (status) {
179 			mlog_errno(status);
180 			ret = ERR_PTR(status);
181 			goto bail_unlock;
182 		}
183 	} else
184 		ocfs2_dentry_attach_gen(dentry);
185 
186 bail_unlock:
187 	/* Don't drop the cluster lock until *after* the d_add --
188 	 * unlink on another node will message us to remove that
189 	 * dentry under this lock so otherwise we can race this with
190 	 * the downconvert thread and have a stale dentry. */
191 	ocfs2_inode_unlock(dir, 0);
192 
193 bail:
194 
195 	trace_ocfs2_lookup_ret(ret);
196 
197 	return ret;
198 }
199 
200 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
201 {
202 	struct inode *inode;
203 
204 	inode = new_inode(dir->i_sb);
205 	if (!inode) {
206 		mlog(ML_ERROR, "new_inode failed!\n");
207 		return NULL;
208 	}
209 
210 	/* populate as many fields early on as possible - many of
211 	 * these are used by the support functions here and in
212 	 * callers. */
213 	if (S_ISDIR(mode))
214 		set_nlink(inode, 2);
215 	inode_init_owner(inode, dir, mode);
216 	dquot_initialize(inode);
217 	return inode;
218 }
219 
220 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
221 		struct dentry *dentry, struct inode *inode)
222 {
223 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
224 
225 	ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
226 	ocfs2_lock_res_free(&dl->dl_lockres);
227 	BUG_ON(dl->dl_count != 1);
228 	spin_lock(&dentry_attach_lock);
229 	dentry->d_fsdata = NULL;
230 	spin_unlock(&dentry_attach_lock);
231 	kfree(dl);
232 	iput(inode);
233 }
234 
235 static int ocfs2_mknod(struct inode *dir,
236 		       struct dentry *dentry,
237 		       umode_t mode,
238 		       dev_t dev)
239 {
240 	int status = 0;
241 	struct buffer_head *parent_fe_bh = NULL;
242 	handle_t *handle = NULL;
243 	struct ocfs2_super *osb;
244 	struct ocfs2_dinode *dirfe;
245 	struct buffer_head *new_fe_bh = NULL;
246 	struct inode *inode = NULL;
247 	struct ocfs2_alloc_context *inode_ac = NULL;
248 	struct ocfs2_alloc_context *data_ac = NULL;
249 	struct ocfs2_alloc_context *meta_ac = NULL;
250 	int want_clusters = 0;
251 	int want_meta = 0;
252 	int xattr_credits = 0;
253 	struct ocfs2_security_xattr_info si = {
254 		.enable = 1,
255 	};
256 	int did_quota_inode = 0;
257 	struct ocfs2_dir_lookup_result lookup = { NULL, };
258 	sigset_t oldset;
259 	int did_block_signals = 0;
260 	struct posix_acl *default_acl = NULL, *acl = NULL;
261 	struct ocfs2_dentry_lock *dl = NULL;
262 
263 	trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
264 			  (unsigned long long)OCFS2_I(dir)->ip_blkno,
265 			  (unsigned long)dev, mode);
266 
267 	dquot_initialize(dir);
268 
269 	/* get our super block */
270 	osb = OCFS2_SB(dir->i_sb);
271 
272 	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
273 	if (status < 0) {
274 		if (status != -ENOENT)
275 			mlog_errno(status);
276 		return status;
277 	}
278 
279 	if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
280 		status = -EMLINK;
281 		goto leave;
282 	}
283 
284 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
285 	if (!ocfs2_read_links_count(dirfe)) {
286 		/* can't make a file in a deleted directory. */
287 		status = -ENOENT;
288 		goto leave;
289 	}
290 
291 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
292 					   dentry->d_name.len);
293 	if (status)
294 		goto leave;
295 
296 	/* get a spot inside the dir. */
297 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
298 					      dentry->d_name.name,
299 					      dentry->d_name.len, &lookup);
300 	if (status < 0) {
301 		mlog_errno(status);
302 		goto leave;
303 	}
304 
305 	/* reserve an inode spot */
306 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
307 	if (status < 0) {
308 		if (status != -ENOSPC)
309 			mlog_errno(status);
310 		goto leave;
311 	}
312 
313 	inode = ocfs2_get_init_inode(dir, mode);
314 	if (!inode) {
315 		status = -ENOMEM;
316 		mlog_errno(status);
317 		goto leave;
318 	}
319 
320 	/* get security xattr */
321 	status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
322 	if (status) {
323 		if (status == -EOPNOTSUPP)
324 			si.enable = 0;
325 		else {
326 			mlog_errno(status);
327 			goto leave;
328 		}
329 	}
330 
331 	/* calculate meta data/clusters for setting security and acl xattr */
332 	status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
333 				       &si, &want_clusters,
334 				       &xattr_credits, &want_meta);
335 	if (status < 0) {
336 		mlog_errno(status);
337 		goto leave;
338 	}
339 
340 	/* Reserve a cluster if creating an extent based directory. */
341 	if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
342 		want_clusters += 1;
343 
344 		/* Dir indexing requires extra space as well */
345 		if (ocfs2_supports_indexed_dirs(osb))
346 			want_meta++;
347 	}
348 
349 	status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
350 	if (status < 0) {
351 		if (status != -ENOSPC)
352 			mlog_errno(status);
353 		goto leave;
354 	}
355 
356 	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
357 	if (status < 0) {
358 		if (status != -ENOSPC)
359 			mlog_errno(status);
360 		goto leave;
361 	}
362 
363 	status = posix_acl_create(dir, &mode, &default_acl, &acl);
364 	if (status) {
365 		mlog_errno(status);
366 		goto leave;
367 	}
368 
369 	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
370 							    S_ISDIR(mode),
371 							    xattr_credits));
372 	if (IS_ERR(handle)) {
373 		status = PTR_ERR(handle);
374 		handle = NULL;
375 		mlog_errno(status);
376 		goto leave;
377 	}
378 
379 	/* Starting to change things, restart is no longer possible. */
380 	ocfs2_block_signals(&oldset);
381 	did_block_signals = 1;
382 
383 	status = dquot_alloc_inode(inode);
384 	if (status)
385 		goto leave;
386 	did_quota_inode = 1;
387 
388 	/* do the real work now. */
389 	status = ocfs2_mknod_locked(osb, dir, inode, dev,
390 				    &new_fe_bh, parent_fe_bh, handle,
391 				    inode_ac);
392 	if (status < 0) {
393 		mlog_errno(status);
394 		goto leave;
395 	}
396 
397 	if (S_ISDIR(mode)) {
398 		status = ocfs2_fill_new_dir(osb, handle, dir, inode,
399 					    new_fe_bh, data_ac, meta_ac);
400 		if (status < 0) {
401 			mlog_errno(status);
402 			goto leave;
403 		}
404 
405 		status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
406 						 parent_fe_bh,
407 						 OCFS2_JOURNAL_ACCESS_WRITE);
408 		if (status < 0) {
409 			mlog_errno(status);
410 			goto leave;
411 		}
412 		ocfs2_add_links_count(dirfe, 1);
413 		ocfs2_journal_dirty(handle, parent_fe_bh);
414 		inc_nlink(dir);
415 	}
416 
417 	if (default_acl) {
418 		status = ocfs2_set_acl(handle, inode, new_fe_bh,
419 				       ACL_TYPE_DEFAULT, default_acl,
420 				       meta_ac, data_ac);
421 	}
422 	if (!status && acl) {
423 		status = ocfs2_set_acl(handle, inode, new_fe_bh,
424 				       ACL_TYPE_ACCESS, acl,
425 				       meta_ac, data_ac);
426 	}
427 
428 	if (status < 0) {
429 		mlog_errno(status);
430 		goto leave;
431 	}
432 
433 	if (si.enable) {
434 		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
435 						 meta_ac, data_ac);
436 		if (status < 0) {
437 			mlog_errno(status);
438 			goto leave;
439 		}
440 	}
441 
442 	/*
443 	 * Do this before adding the entry to the directory. We add
444 	 * also set d_op after success so that ->d_iput() will cleanup
445 	 * the dentry lock even if ocfs2_add_entry() fails below.
446 	 */
447 	status = ocfs2_dentry_attach_lock(dentry, inode,
448 					  OCFS2_I(dir)->ip_blkno);
449 	if (status) {
450 		mlog_errno(status);
451 		goto leave;
452 	}
453 
454 	dl = dentry->d_fsdata;
455 
456 	status = ocfs2_add_entry(handle, dentry, inode,
457 				 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
458 				 &lookup);
459 	if (status < 0) {
460 		mlog_errno(status);
461 		goto leave;
462 	}
463 
464 	insert_inode_hash(inode);
465 	d_instantiate(dentry, inode);
466 	status = 0;
467 leave:
468 	if (default_acl)
469 		posix_acl_release(default_acl);
470 	if (acl)
471 		posix_acl_release(acl);
472 	if (status < 0 && did_quota_inode)
473 		dquot_free_inode(inode);
474 	if (handle)
475 		ocfs2_commit_trans(osb, handle);
476 
477 	ocfs2_inode_unlock(dir, 1);
478 	if (did_block_signals)
479 		ocfs2_unblock_signals(&oldset);
480 
481 	brelse(new_fe_bh);
482 	brelse(parent_fe_bh);
483 	kfree(si.value);
484 
485 	ocfs2_free_dir_lookup_result(&lookup);
486 
487 	if (inode_ac)
488 		ocfs2_free_alloc_context(inode_ac);
489 
490 	if (data_ac)
491 		ocfs2_free_alloc_context(data_ac);
492 
493 	if (meta_ac)
494 		ocfs2_free_alloc_context(meta_ac);
495 
496 	/*
497 	 * We should call iput after the i_mutex of the bitmap been
498 	 * unlocked in ocfs2_free_alloc_context, or the
499 	 * ocfs2_delete_inode will mutex_lock again.
500 	 */
501 	if ((status < 0) && inode) {
502 		if (dl)
503 			ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
504 
505 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
506 		clear_nlink(inode);
507 		iput(inode);
508 	}
509 
510 	if (status)
511 		mlog_errno(status);
512 
513 	return status;
514 }
515 
516 static int __ocfs2_mknod_locked(struct inode *dir,
517 				struct inode *inode,
518 				dev_t dev,
519 				struct buffer_head **new_fe_bh,
520 				struct buffer_head *parent_fe_bh,
521 				handle_t *handle,
522 				struct ocfs2_alloc_context *inode_ac,
523 				u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
524 {
525 	int status = 0;
526 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
527 	struct ocfs2_dinode *fe = NULL;
528 	struct ocfs2_extent_list *fel;
529 	u16 feat;
530 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
531 
532 	*new_fe_bh = NULL;
533 
534 	/* populate as many fields early on as possible - many of
535 	 * these are used by the support functions here and in
536 	 * callers. */
537 	inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
538 	OCFS2_I(inode)->ip_blkno = fe_blkno;
539 	spin_lock(&osb->osb_lock);
540 	inode->i_generation = osb->s_next_generation++;
541 	spin_unlock(&osb->osb_lock);
542 
543 	*new_fe_bh = sb_getblk(osb->sb, fe_blkno);
544 	if (!*new_fe_bh) {
545 		status = -ENOMEM;
546 		mlog_errno(status);
547 		goto leave;
548 	}
549 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
550 
551 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
552 					 *new_fe_bh,
553 					 OCFS2_JOURNAL_ACCESS_CREATE);
554 	if (status < 0) {
555 		mlog_errno(status);
556 		goto leave;
557 	}
558 
559 	fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
560 	memset(fe, 0, osb->sb->s_blocksize);
561 
562 	fe->i_generation = cpu_to_le32(inode->i_generation);
563 	fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
564 	fe->i_blkno = cpu_to_le64(fe_blkno);
565 	fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
566 	fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
567 	fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
568 	fe->i_uid = cpu_to_le32(i_uid_read(inode));
569 	fe->i_gid = cpu_to_le32(i_gid_read(inode));
570 	fe->i_mode = cpu_to_le16(inode->i_mode);
571 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
572 		fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
573 
574 	ocfs2_set_links_count(fe, inode->i_nlink);
575 
576 	fe->i_last_eb_blk = 0;
577 	strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
578 	fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
579 	fe->i_atime = fe->i_ctime = fe->i_mtime =
580 		cpu_to_le64(CURRENT_TIME.tv_sec);
581 	fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
582 		cpu_to_le32(CURRENT_TIME.tv_nsec);
583 	fe->i_dtime = 0;
584 
585 	/*
586 	 * If supported, directories start with inline data. If inline
587 	 * isn't supported, but indexing is, we start them as indexed.
588 	 */
589 	feat = le16_to_cpu(fe->i_dyn_features);
590 	if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
591 		fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
592 
593 		fe->id2.i_data.id_count = cpu_to_le16(
594 				ocfs2_max_inline_data_with_xattr(osb->sb, fe));
595 	} else {
596 		fel = &fe->id2.i_list;
597 		fel->l_tree_depth = 0;
598 		fel->l_next_free_rec = 0;
599 		fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
600 	}
601 
602 	ocfs2_journal_dirty(handle, *new_fe_bh);
603 
604 	ocfs2_populate_inode(inode, fe, 1);
605 	ocfs2_ci_set_new(osb, INODE_CACHE(inode));
606 	if (!ocfs2_mount_local(osb)) {
607 		status = ocfs2_create_new_inode_locks(inode);
608 		if (status < 0)
609 			mlog_errno(status);
610 	}
611 
612 	oi->i_sync_tid = handle->h_transaction->t_tid;
613 	oi->i_datasync_tid = handle->h_transaction->t_tid;
614 
615 leave:
616 	if (status < 0) {
617 		if (*new_fe_bh) {
618 			brelse(*new_fe_bh);
619 			*new_fe_bh = NULL;
620 		}
621 	}
622 
623 	if (status)
624 		mlog_errno(status);
625 	return status;
626 }
627 
628 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
629 			      struct inode *dir,
630 			      struct inode *inode,
631 			      dev_t dev,
632 			      struct buffer_head **new_fe_bh,
633 			      struct buffer_head *parent_fe_bh,
634 			      handle_t *handle,
635 			      struct ocfs2_alloc_context *inode_ac)
636 {
637 	int status = 0;
638 	u64 suballoc_loc, fe_blkno = 0;
639 	u16 suballoc_bit;
640 
641 	*new_fe_bh = NULL;
642 
643 	status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
644 				       inode_ac, &suballoc_loc,
645 				       &suballoc_bit, &fe_blkno);
646 	if (status < 0) {
647 		mlog_errno(status);
648 		return status;
649 	}
650 
651 	return __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
652 				    parent_fe_bh, handle, inode_ac,
653 				    fe_blkno, suballoc_loc, suballoc_bit);
654 }
655 
656 static int ocfs2_mkdir(struct inode *dir,
657 		       struct dentry *dentry,
658 		       umode_t mode)
659 {
660 	int ret;
661 
662 	trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
663 			  OCFS2_I(dir)->ip_blkno, mode);
664 	ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
665 	if (ret)
666 		mlog_errno(ret);
667 
668 	return ret;
669 }
670 
671 static int ocfs2_create(struct inode *dir,
672 			struct dentry *dentry,
673 			umode_t mode,
674 			bool excl)
675 {
676 	int ret;
677 
678 	trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
679 			   (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
680 	ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
681 	if (ret)
682 		mlog_errno(ret);
683 
684 	return ret;
685 }
686 
687 static int ocfs2_link(struct dentry *old_dentry,
688 		      struct inode *dir,
689 		      struct dentry *dentry)
690 {
691 	handle_t *handle;
692 	struct inode *inode = d_inode(old_dentry);
693 	struct inode *old_dir = d_inode(old_dentry->d_parent);
694 	int err;
695 	struct buffer_head *fe_bh = NULL;
696 	struct buffer_head *old_dir_bh = NULL;
697 	struct buffer_head *parent_fe_bh = NULL;
698 	struct ocfs2_dinode *fe = NULL;
699 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
700 	struct ocfs2_dir_lookup_result lookup = { NULL, };
701 	sigset_t oldset;
702 	u64 old_de_ino;
703 
704 	trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
705 			 old_dentry->d_name.len, old_dentry->d_name.name,
706 			 dentry->d_name.len, dentry->d_name.name);
707 
708 	if (S_ISDIR(inode->i_mode))
709 		return -EPERM;
710 
711 	dquot_initialize(dir);
712 
713 	err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
714 			&parent_fe_bh, dir, 0);
715 	if (err < 0) {
716 		if (err != -ENOENT)
717 			mlog_errno(err);
718 		return err;
719 	}
720 
721 	/* make sure both dirs have bhs
722 	 * get an extra ref on old_dir_bh if old==new */
723 	if (!parent_fe_bh) {
724 		if (old_dir_bh) {
725 			parent_fe_bh = old_dir_bh;
726 			get_bh(parent_fe_bh);
727 		} else {
728 			mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
729 			err = -EIO;
730 			goto out;
731 		}
732 	}
733 
734 	if (!dir->i_nlink) {
735 		err = -ENOENT;
736 		goto out;
737 	}
738 
739 	err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
740 			old_dentry->d_name.len, &old_de_ino);
741 	if (err) {
742 		err = -ENOENT;
743 		goto out;
744 	}
745 
746 	/*
747 	 * Check whether another node removed the source inode while we
748 	 * were in the vfs.
749 	 */
750 	if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
751 		err = -ENOENT;
752 		goto out;
753 	}
754 
755 	err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
756 					dentry->d_name.len);
757 	if (err)
758 		goto out;
759 
760 	err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
761 					   dentry->d_name.name,
762 					   dentry->d_name.len, &lookup);
763 	if (err < 0) {
764 		mlog_errno(err);
765 		goto out;
766 	}
767 
768 	err = ocfs2_inode_lock(inode, &fe_bh, 1);
769 	if (err < 0) {
770 		if (err != -ENOENT)
771 			mlog_errno(err);
772 		goto out;
773 	}
774 
775 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
776 	if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
777 		err = -EMLINK;
778 		goto out_unlock_inode;
779 	}
780 
781 	handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
782 	if (IS_ERR(handle)) {
783 		err = PTR_ERR(handle);
784 		handle = NULL;
785 		mlog_errno(err);
786 		goto out_unlock_inode;
787 	}
788 
789 	/* Starting to change things, restart is no longer possible. */
790 	ocfs2_block_signals(&oldset);
791 
792 	err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
793 				      OCFS2_JOURNAL_ACCESS_WRITE);
794 	if (err < 0) {
795 		mlog_errno(err);
796 		goto out_commit;
797 	}
798 
799 	inc_nlink(inode);
800 	inode->i_ctime = CURRENT_TIME;
801 	ocfs2_set_links_count(fe, inode->i_nlink);
802 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
803 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
804 	ocfs2_journal_dirty(handle, fe_bh);
805 
806 	err = ocfs2_add_entry(handle, dentry, inode,
807 			      OCFS2_I(inode)->ip_blkno,
808 			      parent_fe_bh, &lookup);
809 	if (err) {
810 		ocfs2_add_links_count(fe, -1);
811 		drop_nlink(inode);
812 		mlog_errno(err);
813 		goto out_commit;
814 	}
815 
816 	err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
817 	if (err) {
818 		mlog_errno(err);
819 		goto out_commit;
820 	}
821 
822 	ihold(inode);
823 	d_instantiate(dentry, inode);
824 
825 out_commit:
826 	ocfs2_commit_trans(osb, handle);
827 	ocfs2_unblock_signals(&oldset);
828 out_unlock_inode:
829 	ocfs2_inode_unlock(inode, 1);
830 
831 out:
832 	ocfs2_double_unlock(old_dir, dir);
833 
834 	brelse(fe_bh);
835 	brelse(parent_fe_bh);
836 	brelse(old_dir_bh);
837 
838 	ocfs2_free_dir_lookup_result(&lookup);
839 
840 	if (err)
841 		mlog_errno(err);
842 
843 	return err;
844 }
845 
846 /*
847  * Takes and drops an exclusive lock on the given dentry. This will
848  * force other nodes to drop it.
849  */
850 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
851 {
852 	int ret;
853 
854 	ret = ocfs2_dentry_lock(dentry, 1);
855 	if (ret)
856 		mlog_errno(ret);
857 	else
858 		ocfs2_dentry_unlock(dentry, 1);
859 
860 	return ret;
861 }
862 
863 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
864 {
865 	if (S_ISDIR(inode->i_mode)) {
866 		if (inode->i_nlink == 2)
867 			return 1;
868 		return 0;
869 	}
870 
871 	if (inode->i_nlink == 1)
872 		return 1;
873 	return 0;
874 }
875 
876 static int ocfs2_unlink(struct inode *dir,
877 			struct dentry *dentry)
878 {
879 	int status;
880 	int child_locked = 0;
881 	bool is_unlinkable = false;
882 	struct inode *inode = d_inode(dentry);
883 	struct inode *orphan_dir = NULL;
884 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
885 	u64 blkno;
886 	struct ocfs2_dinode *fe = NULL;
887 	struct buffer_head *fe_bh = NULL;
888 	struct buffer_head *parent_node_bh = NULL;
889 	handle_t *handle = NULL;
890 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
891 	struct ocfs2_dir_lookup_result lookup = { NULL, };
892 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
893 
894 	trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
895 			   dentry->d_name.name,
896 			   (unsigned long long)OCFS2_I(dir)->ip_blkno,
897 			   (unsigned long long)OCFS2_I(inode)->ip_blkno);
898 
899 	dquot_initialize(dir);
900 
901 	BUG_ON(d_inode(dentry->d_parent) != dir);
902 
903 	if (inode == osb->root_inode)
904 		return -EPERM;
905 
906 	status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
907 					 OI_LS_PARENT);
908 	if (status < 0) {
909 		if (status != -ENOENT)
910 			mlog_errno(status);
911 		return status;
912 	}
913 
914 	status = ocfs2_find_files_on_disk(dentry->d_name.name,
915 					  dentry->d_name.len, &blkno, dir,
916 					  &lookup);
917 	if (status < 0) {
918 		if (status != -ENOENT)
919 			mlog_errno(status);
920 		goto leave;
921 	}
922 
923 	if (OCFS2_I(inode)->ip_blkno != blkno) {
924 		status = -ENOENT;
925 
926 		trace_ocfs2_unlink_noent(
927 				(unsigned long long)OCFS2_I(inode)->ip_blkno,
928 				(unsigned long long)blkno,
929 				OCFS2_I(inode)->ip_flags);
930 		goto leave;
931 	}
932 
933 	status = ocfs2_inode_lock(inode, &fe_bh, 1);
934 	if (status < 0) {
935 		if (status != -ENOENT)
936 			mlog_errno(status);
937 		goto leave;
938 	}
939 	child_locked = 1;
940 
941 	if (S_ISDIR(inode->i_mode)) {
942 		if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
943 			status = -ENOTEMPTY;
944 			goto leave;
945 		}
946 	}
947 
948 	status = ocfs2_remote_dentry_delete(dentry);
949 	if (status < 0) {
950 		/* This remote delete should succeed under all normal
951 		 * circumstances. */
952 		mlog_errno(status);
953 		goto leave;
954 	}
955 
956 	if (ocfs2_inode_is_unlinkable(inode)) {
957 		status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
958 						  OCFS2_I(inode)->ip_blkno,
959 						  orphan_name, &orphan_insert,
960 						  false);
961 		if (status < 0) {
962 			mlog_errno(status);
963 			goto leave;
964 		}
965 		is_unlinkable = true;
966 	}
967 
968 	handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
969 	if (IS_ERR(handle)) {
970 		status = PTR_ERR(handle);
971 		handle = NULL;
972 		mlog_errno(status);
973 		goto leave;
974 	}
975 
976 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
977 					 OCFS2_JOURNAL_ACCESS_WRITE);
978 	if (status < 0) {
979 		mlog_errno(status);
980 		goto leave;
981 	}
982 
983 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
984 
985 	/* delete the name from the parent dir */
986 	status = ocfs2_delete_entry(handle, dir, &lookup);
987 	if (status < 0) {
988 		mlog_errno(status);
989 		goto leave;
990 	}
991 
992 	if (S_ISDIR(inode->i_mode))
993 		drop_nlink(inode);
994 	drop_nlink(inode);
995 	ocfs2_set_links_count(fe, inode->i_nlink);
996 	ocfs2_journal_dirty(handle, fe_bh);
997 
998 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
999 	if (S_ISDIR(inode->i_mode))
1000 		drop_nlink(dir);
1001 
1002 	status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
1003 	if (status < 0) {
1004 		mlog_errno(status);
1005 		if (S_ISDIR(inode->i_mode))
1006 			inc_nlink(dir);
1007 		goto leave;
1008 	}
1009 
1010 	if (is_unlinkable) {
1011 		status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1012 				orphan_name, &orphan_insert, orphan_dir, false);
1013 		if (status < 0)
1014 			mlog_errno(status);
1015 	}
1016 
1017 leave:
1018 	if (handle)
1019 		ocfs2_commit_trans(osb, handle);
1020 
1021 	if (child_locked)
1022 		ocfs2_inode_unlock(inode, 1);
1023 
1024 	ocfs2_inode_unlock(dir, 1);
1025 
1026 	if (orphan_dir) {
1027 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
1028 		ocfs2_inode_unlock(orphan_dir, 1);
1029 		mutex_unlock(&orphan_dir->i_mutex);
1030 		iput(orphan_dir);
1031 	}
1032 
1033 	brelse(fe_bh);
1034 	brelse(parent_node_bh);
1035 
1036 	ocfs2_free_dir_lookup_result(&orphan_insert);
1037 	ocfs2_free_dir_lookup_result(&lookup);
1038 
1039 	if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1040 		mlog_errno(status);
1041 
1042 	return status;
1043 }
1044 
1045 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1046 		u64 src_inode_no, u64 dest_inode_no)
1047 {
1048 	int ret = 0, i = 0;
1049 	u64 parent_inode_no = 0;
1050 	u64 child_inode_no = src_inode_no;
1051 	struct inode *child_inode;
1052 
1053 #define MAX_LOOKUP_TIMES 32
1054 	while (1) {
1055 		child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1056 		if (IS_ERR(child_inode)) {
1057 			ret = PTR_ERR(child_inode);
1058 			break;
1059 		}
1060 
1061 		ret = ocfs2_inode_lock(child_inode, NULL, 0);
1062 		if (ret < 0) {
1063 			iput(child_inode);
1064 			if (ret != -ENOENT)
1065 				mlog_errno(ret);
1066 			break;
1067 		}
1068 
1069 		ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1070 				&parent_inode_no);
1071 		ocfs2_inode_unlock(child_inode, 0);
1072 		iput(child_inode);
1073 		if (ret < 0) {
1074 			ret = -ENOENT;
1075 			break;
1076 		}
1077 
1078 		if (parent_inode_no == dest_inode_no) {
1079 			ret = 1;
1080 			break;
1081 		}
1082 
1083 		if (parent_inode_no == osb->root_inode->i_ino) {
1084 			ret = 0;
1085 			break;
1086 		}
1087 
1088 		child_inode_no = parent_inode_no;
1089 
1090 		if (++i >= MAX_LOOKUP_TIMES) {
1091 			mlog(ML_NOTICE, "max lookup times reached, filesystem "
1092 					"may have nested directories, "
1093 					"src inode: %llu, dest inode: %llu.\n",
1094 					(unsigned long long)src_inode_no,
1095 					(unsigned long long)dest_inode_no);
1096 			ret = 0;
1097 			break;
1098 		}
1099 	}
1100 
1101 	return ret;
1102 }
1103 
1104 /*
1105  * The only place this should be used is rename and link!
1106  * if they have the same id, then the 1st one is the only one locked.
1107  */
1108 static int ocfs2_double_lock(struct ocfs2_super *osb,
1109 			     struct buffer_head **bh1,
1110 			     struct inode *inode1,
1111 			     struct buffer_head **bh2,
1112 			     struct inode *inode2,
1113 			     int rename)
1114 {
1115 	int status;
1116 	int inode1_is_ancestor, inode2_is_ancestor;
1117 	struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1118 	struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1119 	struct buffer_head **tmpbh;
1120 	struct inode *tmpinode;
1121 
1122 	trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1123 				(unsigned long long)oi2->ip_blkno);
1124 
1125 	if (*bh1)
1126 		*bh1 = NULL;
1127 	if (*bh2)
1128 		*bh2 = NULL;
1129 
1130 	/* we always want to lock the one with the lower lockid first.
1131 	 * and if they are nested, we lock ancestor first */
1132 	if (oi1->ip_blkno != oi2->ip_blkno) {
1133 		inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1134 				oi1->ip_blkno);
1135 		if (inode1_is_ancestor < 0) {
1136 			status = inode1_is_ancestor;
1137 			goto bail;
1138 		}
1139 
1140 		inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1141 				oi2->ip_blkno);
1142 		if (inode2_is_ancestor < 0) {
1143 			status = inode2_is_ancestor;
1144 			goto bail;
1145 		}
1146 
1147 		if ((inode1_is_ancestor == 1) ||
1148 				(oi1->ip_blkno < oi2->ip_blkno &&
1149 				inode2_is_ancestor == 0)) {
1150 			/* switch id1 and id2 around */
1151 			tmpbh = bh2;
1152 			bh2 = bh1;
1153 			bh1 = tmpbh;
1154 
1155 			tmpinode = inode2;
1156 			inode2 = inode1;
1157 			inode1 = tmpinode;
1158 		}
1159 		/* lock id2 */
1160 		status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1161 				rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1162 		if (status < 0) {
1163 			if (status != -ENOENT)
1164 				mlog_errno(status);
1165 			goto bail;
1166 		}
1167 	}
1168 
1169 	/* lock id1 */
1170 	status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1171 			rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1172 	if (status < 0) {
1173 		/*
1174 		 * An error return must mean that no cluster locks
1175 		 * were held on function exit.
1176 		 */
1177 		if (oi1->ip_blkno != oi2->ip_blkno) {
1178 			ocfs2_inode_unlock(inode2, 1);
1179 			brelse(*bh2);
1180 			*bh2 = NULL;
1181 		}
1182 
1183 		if (status != -ENOENT)
1184 			mlog_errno(status);
1185 	}
1186 
1187 	trace_ocfs2_double_lock_end(
1188 			(unsigned long long)OCFS2_I(inode1)->ip_blkno,
1189 			(unsigned long long)OCFS2_I(inode2)->ip_blkno);
1190 
1191 bail:
1192 	if (status)
1193 		mlog_errno(status);
1194 	return status;
1195 }
1196 
1197 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1198 {
1199 	ocfs2_inode_unlock(inode1, 1);
1200 
1201 	if (inode1 != inode2)
1202 		ocfs2_inode_unlock(inode2, 1);
1203 }
1204 
1205 static int ocfs2_rename(struct inode *old_dir,
1206 			struct dentry *old_dentry,
1207 			struct inode *new_dir,
1208 			struct dentry *new_dentry)
1209 {
1210 	int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1211 	int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1212 	struct inode *old_inode = d_inode(old_dentry);
1213 	struct inode *new_inode = d_inode(new_dentry);
1214 	struct inode *orphan_dir = NULL;
1215 	struct ocfs2_dinode *newfe = NULL;
1216 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1217 	struct buffer_head *newfe_bh = NULL;
1218 	struct buffer_head *old_inode_bh = NULL;
1219 	struct ocfs2_super *osb = NULL;
1220 	u64 newfe_blkno, old_de_ino;
1221 	handle_t *handle = NULL;
1222 	struct buffer_head *old_dir_bh = NULL;
1223 	struct buffer_head *new_dir_bh = NULL;
1224 	u32 old_dir_nlink = old_dir->i_nlink;
1225 	struct ocfs2_dinode *old_di;
1226 	struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1227 	struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1228 	struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1229 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1230 	struct ocfs2_dir_lookup_result target_insert = { NULL, };
1231 	bool should_add_orphan = false;
1232 
1233 	/* At some point it might be nice to break this function up a
1234 	 * bit. */
1235 
1236 	trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1237 			   old_dentry->d_name.len, old_dentry->d_name.name,
1238 			   new_dentry->d_name.len, new_dentry->d_name.name);
1239 
1240 	dquot_initialize(old_dir);
1241 	dquot_initialize(new_dir);
1242 
1243 	osb = OCFS2_SB(old_dir->i_sb);
1244 
1245 	if (new_inode) {
1246 		if (!igrab(new_inode))
1247 			BUG();
1248 	}
1249 
1250 	/* Assume a directory hierarchy thusly:
1251 	 * a/b/c
1252 	 * a/d
1253 	 * a,b,c, and d are all directories.
1254 	 *
1255 	 * from cwd of 'a' on both nodes:
1256 	 * node1: mv b/c d
1257 	 * node2: mv d   b/c
1258 	 *
1259 	 * And that's why, just like the VFS, we need a file system
1260 	 * rename lock. */
1261 	if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1262 		status = ocfs2_rename_lock(osb);
1263 		if (status < 0) {
1264 			mlog_errno(status);
1265 			goto bail;
1266 		}
1267 		rename_lock = 1;
1268 
1269 		/* here we cannot guarantee the inodes haven't just been
1270 		 * changed, so check if they are nested again */
1271 		status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1272 				old_inode->i_ino);
1273 		if (status < 0) {
1274 			mlog_errno(status);
1275 			goto bail;
1276 		} else if (status == 1) {
1277 			status = -EPERM;
1278 			trace_ocfs2_rename_not_permitted(
1279 					(unsigned long long)old_inode->i_ino,
1280 					(unsigned long long)new_dir->i_ino);
1281 			goto bail;
1282 		}
1283 	}
1284 
1285 	/* if old and new are the same, this'll just do one lock. */
1286 	status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1287 				   &new_dir_bh, new_dir, 1);
1288 	if (status < 0) {
1289 		mlog_errno(status);
1290 		goto bail;
1291 	}
1292 	parents_locked = 1;
1293 
1294 	/* make sure both dirs have bhs
1295 	 * get an extra ref on old_dir_bh if old==new */
1296 	if (!new_dir_bh) {
1297 		if (old_dir_bh) {
1298 			new_dir_bh = old_dir_bh;
1299 			get_bh(new_dir_bh);
1300 		} else {
1301 			mlog(ML_ERROR, "no old_dir_bh!\n");
1302 			status = -EIO;
1303 			goto bail;
1304 		}
1305 	}
1306 
1307 	/*
1308 	 * Aside from allowing a meta data update, the locking here
1309 	 * also ensures that the downconvert thread on other nodes
1310 	 * won't have to concurrently downconvert the inode and the
1311 	 * dentry locks.
1312 	 */
1313 	status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1314 					 OI_LS_PARENT);
1315 	if (status < 0) {
1316 		if (status != -ENOENT)
1317 			mlog_errno(status);
1318 		goto bail;
1319 	}
1320 	old_child_locked = 1;
1321 
1322 	status = ocfs2_remote_dentry_delete(old_dentry);
1323 	if (status < 0) {
1324 		mlog_errno(status);
1325 		goto bail;
1326 	}
1327 
1328 	if (S_ISDIR(old_inode->i_mode)) {
1329 		u64 old_inode_parent;
1330 
1331 		update_dot_dot = 1;
1332 		status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1333 						  old_inode,
1334 						  &old_inode_dot_dot_res);
1335 		if (status) {
1336 			status = -EIO;
1337 			goto bail;
1338 		}
1339 
1340 		if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1341 			status = -EIO;
1342 			goto bail;
1343 		}
1344 
1345 		if (!new_inode && new_dir != old_dir &&
1346 		    new_dir->i_nlink >= ocfs2_link_max(osb)) {
1347 			status = -EMLINK;
1348 			goto bail;
1349 		}
1350 	}
1351 
1352 	status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1353 					    old_dentry->d_name.len,
1354 					    &old_de_ino);
1355 	if (status) {
1356 		status = -ENOENT;
1357 		goto bail;
1358 	}
1359 
1360 	/*
1361 	 *  Check for inode number is _not_ due to possible IO errors.
1362 	 *  We might rmdir the source, keep it as pwd of some process
1363 	 *  and merrily kill the link to whatever was created under the
1364 	 *  same name. Goodbye sticky bit ;-<
1365 	 */
1366 	if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1367 		status = -ENOENT;
1368 		goto bail;
1369 	}
1370 
1371 	/* check if the target already exists (in which case we need
1372 	 * to delete it */
1373 	status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1374 					  new_dentry->d_name.len,
1375 					  &newfe_blkno, new_dir,
1376 					  &target_lookup_res);
1377 	/* The only error we allow here is -ENOENT because the new
1378 	 * file not existing is perfectly valid. */
1379 	if ((status < 0) && (status != -ENOENT)) {
1380 		/* If we cannot find the file specified we should just */
1381 		/* return the error... */
1382 		mlog_errno(status);
1383 		goto bail;
1384 	}
1385 	if (status == 0)
1386 		target_exists = 1;
1387 
1388 	if (!target_exists && new_inode) {
1389 		/*
1390 		 * Target was unlinked by another node while we were
1391 		 * waiting to get to ocfs2_rename(). There isn't
1392 		 * anything we can do here to help the situation, so
1393 		 * bubble up the appropriate error.
1394 		 */
1395 		status = -ENOENT;
1396 		goto bail;
1397 	}
1398 
1399 	/* In case we need to overwrite an existing file, we blow it
1400 	 * away first */
1401 	if (target_exists) {
1402 		/* VFS didn't think there existed an inode here, but
1403 		 * someone else in the cluster must have raced our
1404 		 * rename to create one. Today we error cleanly, in
1405 		 * the future we should consider calling iget to build
1406 		 * a new struct inode for this entry. */
1407 		if (!new_inode) {
1408 			status = -EACCES;
1409 
1410 			trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1411 						new_dentry->d_name.name);
1412 			goto bail;
1413 		}
1414 
1415 		if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1416 			status = -EACCES;
1417 
1418 			trace_ocfs2_rename_disagree(
1419 			     (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1420 			     (unsigned long long)newfe_blkno,
1421 			     OCFS2_I(new_inode)->ip_flags);
1422 			goto bail;
1423 		}
1424 
1425 		status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1426 		if (status < 0) {
1427 			if (status != -ENOENT)
1428 				mlog_errno(status);
1429 			goto bail;
1430 		}
1431 		new_child_locked = 1;
1432 
1433 		status = ocfs2_remote_dentry_delete(new_dentry);
1434 		if (status < 0) {
1435 			mlog_errno(status);
1436 			goto bail;
1437 		}
1438 
1439 		newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1440 
1441 		trace_ocfs2_rename_over_existing(
1442 		     (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1443 		     (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1444 
1445 		if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1446 			status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1447 						OCFS2_I(new_inode)->ip_blkno,
1448 						orphan_name, &orphan_insert,
1449 						false);
1450 			if (status < 0) {
1451 				mlog_errno(status);
1452 				goto bail;
1453 			}
1454 			should_add_orphan = true;
1455 		}
1456 	} else {
1457 		BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1458 
1459 		status = ocfs2_check_dir_for_entry(new_dir,
1460 						   new_dentry->d_name.name,
1461 						   new_dentry->d_name.len);
1462 		if (status)
1463 			goto bail;
1464 
1465 		status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1466 						      new_dentry->d_name.name,
1467 						      new_dentry->d_name.len,
1468 						      &target_insert);
1469 		if (status < 0) {
1470 			mlog_errno(status);
1471 			goto bail;
1472 		}
1473 	}
1474 
1475 	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1476 	if (IS_ERR(handle)) {
1477 		status = PTR_ERR(handle);
1478 		handle = NULL;
1479 		mlog_errno(status);
1480 		goto bail;
1481 	}
1482 
1483 	if (target_exists) {
1484 		if (S_ISDIR(new_inode->i_mode)) {
1485 			if (new_inode->i_nlink != 2 ||
1486 			    !ocfs2_empty_dir(new_inode)) {
1487 				status = -ENOTEMPTY;
1488 				goto bail;
1489 			}
1490 		}
1491 		status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1492 						 newfe_bh,
1493 						 OCFS2_JOURNAL_ACCESS_WRITE);
1494 		if (status < 0) {
1495 			mlog_errno(status);
1496 			goto bail;
1497 		}
1498 
1499 		/* change the dirent to point to the correct inode */
1500 		status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1501 					    old_inode);
1502 		if (status < 0) {
1503 			mlog_errno(status);
1504 			goto bail;
1505 		}
1506 		new_dir->i_version++;
1507 
1508 		if (S_ISDIR(new_inode->i_mode))
1509 			ocfs2_set_links_count(newfe, 0);
1510 		else
1511 			ocfs2_add_links_count(newfe, -1);
1512 		ocfs2_journal_dirty(handle, newfe_bh);
1513 		if (should_add_orphan) {
1514 			status = ocfs2_orphan_add(osb, handle, new_inode,
1515 					newfe_bh, orphan_name,
1516 					&orphan_insert, orphan_dir, false);
1517 			if (status < 0) {
1518 				mlog_errno(status);
1519 				goto bail;
1520 			}
1521 		}
1522 	} else {
1523 		/* if the name was not found in new_dir, add it now */
1524 		status = ocfs2_add_entry(handle, new_dentry, old_inode,
1525 					 OCFS2_I(old_inode)->ip_blkno,
1526 					 new_dir_bh, &target_insert);
1527 	}
1528 
1529 	old_inode->i_ctime = CURRENT_TIME;
1530 	mark_inode_dirty(old_inode);
1531 
1532 	status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1533 					 old_inode_bh,
1534 					 OCFS2_JOURNAL_ACCESS_WRITE);
1535 	if (status >= 0) {
1536 		old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1537 
1538 		old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1539 		old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1540 		ocfs2_journal_dirty(handle, old_inode_bh);
1541 	} else
1542 		mlog_errno(status);
1543 
1544 	/*
1545 	 * Now that the name has been added to new_dir, remove the old name.
1546 	 *
1547 	 * We don't keep any directory entry context around until now
1548 	 * because the insert might have changed the type of directory
1549 	 * we're dealing with.
1550 	 */
1551 	status = ocfs2_find_entry(old_dentry->d_name.name,
1552 				  old_dentry->d_name.len, old_dir,
1553 				  &old_entry_lookup);
1554 	if (status)
1555 		goto bail;
1556 
1557 	status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1558 	if (status < 0) {
1559 		mlog_errno(status);
1560 		goto bail;
1561 	}
1562 
1563 	if (new_inode) {
1564 		drop_nlink(new_inode);
1565 		new_inode->i_ctime = CURRENT_TIME;
1566 	}
1567 	old_dir->i_ctime = old_dir->i_mtime = CURRENT_TIME;
1568 
1569 	if (update_dot_dot) {
1570 		status = ocfs2_update_entry(old_inode, handle,
1571 					    &old_inode_dot_dot_res, new_dir);
1572 		drop_nlink(old_dir);
1573 		if (new_inode) {
1574 			drop_nlink(new_inode);
1575 		} else {
1576 			inc_nlink(new_dir);
1577 			mark_inode_dirty(new_dir);
1578 		}
1579 	}
1580 	mark_inode_dirty(old_dir);
1581 	ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1582 	if (new_inode) {
1583 		mark_inode_dirty(new_inode);
1584 		ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1585 	}
1586 
1587 	if (old_dir != new_dir) {
1588 		/* Keep the same times on both directories.*/
1589 		new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1590 
1591 		/*
1592 		 * This will also pick up the i_nlink change from the
1593 		 * block above.
1594 		 */
1595 		ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1596 	}
1597 
1598 	if (old_dir_nlink != old_dir->i_nlink) {
1599 		if (!old_dir_bh) {
1600 			mlog(ML_ERROR, "need to change nlink for old dir "
1601 			     "%llu from %d to %d but bh is NULL!\n",
1602 			     (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1603 			     (int)old_dir_nlink, old_dir->i_nlink);
1604 		} else {
1605 			struct ocfs2_dinode *fe;
1606 			status = ocfs2_journal_access_di(handle,
1607 							 INODE_CACHE(old_dir),
1608 							 old_dir_bh,
1609 							 OCFS2_JOURNAL_ACCESS_WRITE);
1610 			fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1611 			ocfs2_set_links_count(fe, old_dir->i_nlink);
1612 			ocfs2_journal_dirty(handle, old_dir_bh);
1613 		}
1614 	}
1615 	ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1616 	status = 0;
1617 bail:
1618 	if (rename_lock)
1619 		ocfs2_rename_unlock(osb);
1620 
1621 	if (handle)
1622 		ocfs2_commit_trans(osb, handle);
1623 
1624 	if (parents_locked)
1625 		ocfs2_double_unlock(old_dir, new_dir);
1626 
1627 	if (old_child_locked)
1628 		ocfs2_inode_unlock(old_inode, 1);
1629 
1630 	if (new_child_locked)
1631 		ocfs2_inode_unlock(new_inode, 1);
1632 
1633 	if (orphan_dir) {
1634 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
1635 		ocfs2_inode_unlock(orphan_dir, 1);
1636 		mutex_unlock(&orphan_dir->i_mutex);
1637 		iput(orphan_dir);
1638 	}
1639 
1640 	if (new_inode)
1641 		sync_mapping_buffers(old_inode->i_mapping);
1642 
1643 	if (new_inode)
1644 		iput(new_inode);
1645 
1646 	ocfs2_free_dir_lookup_result(&target_lookup_res);
1647 	ocfs2_free_dir_lookup_result(&old_entry_lookup);
1648 	ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1649 	ocfs2_free_dir_lookup_result(&orphan_insert);
1650 	ocfs2_free_dir_lookup_result(&target_insert);
1651 
1652 	brelse(newfe_bh);
1653 	brelse(old_inode_bh);
1654 	brelse(old_dir_bh);
1655 	brelse(new_dir_bh);
1656 
1657 	if (status)
1658 		mlog_errno(status);
1659 
1660 	return status;
1661 }
1662 
1663 /*
1664  * we expect i_size = strlen(symname). Copy symname into the file
1665  * data, including the null terminator.
1666  */
1667 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1668 				     handle_t *handle,
1669 				     struct inode *inode,
1670 				     const char *symname)
1671 {
1672 	struct buffer_head **bhs = NULL;
1673 	const char *c;
1674 	struct super_block *sb = osb->sb;
1675 	u64 p_blkno, p_blocks;
1676 	int virtual, blocks, status, i, bytes_left;
1677 
1678 	bytes_left = i_size_read(inode) + 1;
1679 	/* we can't trust i_blocks because we're actually going to
1680 	 * write i_size + 1 bytes. */
1681 	blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1682 
1683 	trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1684 					i_size_read(inode), blocks);
1685 
1686 	/* Sanity check -- make sure we're going to fit. */
1687 	if (bytes_left >
1688 	    ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1689 		status = -EIO;
1690 		mlog_errno(status);
1691 		goto bail;
1692 	}
1693 
1694 	bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1695 	if (!bhs) {
1696 		status = -ENOMEM;
1697 		mlog_errno(status);
1698 		goto bail;
1699 	}
1700 
1701 	status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1702 					     NULL);
1703 	if (status < 0) {
1704 		mlog_errno(status);
1705 		goto bail;
1706 	}
1707 
1708 	/* links can never be larger than one cluster so we know this
1709 	 * is all going to be contiguous, but do a sanity check
1710 	 * anyway. */
1711 	if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1712 		status = -EIO;
1713 		mlog_errno(status);
1714 		goto bail;
1715 	}
1716 
1717 	virtual = 0;
1718 	while(bytes_left > 0) {
1719 		c = &symname[virtual * sb->s_blocksize];
1720 
1721 		bhs[virtual] = sb_getblk(sb, p_blkno);
1722 		if (!bhs[virtual]) {
1723 			status = -ENOMEM;
1724 			mlog_errno(status);
1725 			goto bail;
1726 		}
1727 		ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1728 					      bhs[virtual]);
1729 
1730 		status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1731 					      bhs[virtual],
1732 					      OCFS2_JOURNAL_ACCESS_CREATE);
1733 		if (status < 0) {
1734 			mlog_errno(status);
1735 			goto bail;
1736 		}
1737 
1738 		memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1739 
1740 		memcpy(bhs[virtual]->b_data, c,
1741 		       (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1742 		       bytes_left);
1743 
1744 		ocfs2_journal_dirty(handle, bhs[virtual]);
1745 
1746 		virtual++;
1747 		p_blkno++;
1748 		bytes_left -= sb->s_blocksize;
1749 	}
1750 
1751 	status = 0;
1752 bail:
1753 
1754 	if (bhs) {
1755 		for(i = 0; i < blocks; i++)
1756 			brelse(bhs[i]);
1757 		kfree(bhs);
1758 	}
1759 
1760 	if (status)
1761 		mlog_errno(status);
1762 	return status;
1763 }
1764 
1765 static int ocfs2_symlink(struct inode *dir,
1766 			 struct dentry *dentry,
1767 			 const char *symname)
1768 {
1769 	int status, l, credits;
1770 	u64 newsize;
1771 	struct ocfs2_super *osb = NULL;
1772 	struct inode *inode = NULL;
1773 	struct super_block *sb;
1774 	struct buffer_head *new_fe_bh = NULL;
1775 	struct buffer_head *parent_fe_bh = NULL;
1776 	struct ocfs2_dinode *fe = NULL;
1777 	struct ocfs2_dinode *dirfe;
1778 	handle_t *handle = NULL;
1779 	struct ocfs2_alloc_context *inode_ac = NULL;
1780 	struct ocfs2_alloc_context *data_ac = NULL;
1781 	struct ocfs2_alloc_context *xattr_ac = NULL;
1782 	int want_clusters = 0;
1783 	int xattr_credits = 0;
1784 	struct ocfs2_security_xattr_info si = {
1785 		.enable = 1,
1786 	};
1787 	int did_quota = 0, did_quota_inode = 0;
1788 	struct ocfs2_dir_lookup_result lookup = { NULL, };
1789 	sigset_t oldset;
1790 	int did_block_signals = 0;
1791 	struct ocfs2_dentry_lock *dl = NULL;
1792 
1793 	trace_ocfs2_symlink_begin(dir, dentry, symname,
1794 				  dentry->d_name.len, dentry->d_name.name);
1795 
1796 	dquot_initialize(dir);
1797 
1798 	sb = dir->i_sb;
1799 	osb = OCFS2_SB(sb);
1800 
1801 	l = strlen(symname) + 1;
1802 
1803 	credits = ocfs2_calc_symlink_credits(sb);
1804 
1805 	/* lock the parent directory */
1806 	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1807 	if (status < 0) {
1808 		if (status != -ENOENT)
1809 			mlog_errno(status);
1810 		return status;
1811 	}
1812 
1813 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1814 	if (!ocfs2_read_links_count(dirfe)) {
1815 		/* can't make a file in a deleted directory. */
1816 		status = -ENOENT;
1817 		goto bail;
1818 	}
1819 
1820 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1821 					   dentry->d_name.len);
1822 	if (status)
1823 		goto bail;
1824 
1825 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1826 					      dentry->d_name.name,
1827 					      dentry->d_name.len, &lookup);
1828 	if (status < 0) {
1829 		mlog_errno(status);
1830 		goto bail;
1831 	}
1832 
1833 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
1834 	if (status < 0) {
1835 		if (status != -ENOSPC)
1836 			mlog_errno(status);
1837 		goto bail;
1838 	}
1839 
1840 	inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1841 	if (!inode) {
1842 		status = -ENOMEM;
1843 		mlog_errno(status);
1844 		goto bail;
1845 	}
1846 
1847 	/* get security xattr */
1848 	status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1849 	if (status) {
1850 		if (status == -EOPNOTSUPP)
1851 			si.enable = 0;
1852 		else {
1853 			mlog_errno(status);
1854 			goto bail;
1855 		}
1856 	}
1857 
1858 	/* calculate meta data/clusters for setting security xattr */
1859 	if (si.enable) {
1860 		status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1861 						  &xattr_credits, &xattr_ac);
1862 		if (status < 0) {
1863 			mlog_errno(status);
1864 			goto bail;
1865 		}
1866 	}
1867 
1868 	/* don't reserve bitmap space for fast symlinks. */
1869 	if (l > ocfs2_fast_symlink_chars(sb))
1870 		want_clusters += 1;
1871 
1872 	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1873 	if (status < 0) {
1874 		if (status != -ENOSPC)
1875 			mlog_errno(status);
1876 		goto bail;
1877 	}
1878 
1879 	handle = ocfs2_start_trans(osb, credits + xattr_credits);
1880 	if (IS_ERR(handle)) {
1881 		status = PTR_ERR(handle);
1882 		handle = NULL;
1883 		mlog_errno(status);
1884 		goto bail;
1885 	}
1886 
1887 	/* Starting to change things, restart is no longer possible. */
1888 	ocfs2_block_signals(&oldset);
1889 	did_block_signals = 1;
1890 
1891 	status = dquot_alloc_inode(inode);
1892 	if (status)
1893 		goto bail;
1894 	did_quota_inode = 1;
1895 
1896 	trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1897 				   dentry->d_name.name,
1898 				   (unsigned long long)OCFS2_I(dir)->ip_blkno,
1899 				   inode->i_mode);
1900 
1901 	status = ocfs2_mknod_locked(osb, dir, inode,
1902 				    0, &new_fe_bh, parent_fe_bh, handle,
1903 				    inode_ac);
1904 	if (status < 0) {
1905 		mlog_errno(status);
1906 		goto bail;
1907 	}
1908 
1909 	fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1910 	inode->i_rdev = 0;
1911 	newsize = l - 1;
1912 	inode->i_op = &ocfs2_symlink_inode_operations;
1913 	if (l > ocfs2_fast_symlink_chars(sb)) {
1914 		u32 offset = 0;
1915 
1916 		status = dquot_alloc_space_nodirty(inode,
1917 		    ocfs2_clusters_to_bytes(osb->sb, 1));
1918 		if (status)
1919 			goto bail;
1920 		did_quota = 1;
1921 		inode->i_mapping->a_ops = &ocfs2_aops;
1922 		status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1923 					      new_fe_bh,
1924 					      handle, data_ac, NULL,
1925 					      NULL);
1926 		if (status < 0) {
1927 			if (status != -ENOSPC && status != -EINTR) {
1928 				mlog(ML_ERROR,
1929 				     "Failed to extend file to %llu\n",
1930 				     (unsigned long long)newsize);
1931 				mlog_errno(status);
1932 				status = -ENOSPC;
1933 			}
1934 			goto bail;
1935 		}
1936 		i_size_write(inode, newsize);
1937 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1938 	} else {
1939 		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1940 		memcpy((char *) fe->id2.i_symlink, symname, l);
1941 		i_size_write(inode, newsize);
1942 		inode->i_blocks = 0;
1943 	}
1944 
1945 	status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1946 	if (status < 0) {
1947 		mlog_errno(status);
1948 		goto bail;
1949 	}
1950 
1951 	if (!ocfs2_inode_is_fast_symlink(inode)) {
1952 		status = ocfs2_create_symlink_data(osb, handle, inode,
1953 						   symname);
1954 		if (status < 0) {
1955 			mlog_errno(status);
1956 			goto bail;
1957 		}
1958 	}
1959 
1960 	if (si.enable) {
1961 		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1962 						 xattr_ac, data_ac);
1963 		if (status < 0) {
1964 			mlog_errno(status);
1965 			goto bail;
1966 		}
1967 	}
1968 
1969 	/*
1970 	 * Do this before adding the entry to the directory. We add
1971 	 * also set d_op after success so that ->d_iput() will cleanup
1972 	 * the dentry lock even if ocfs2_add_entry() fails below.
1973 	 */
1974 	status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1975 	if (status) {
1976 		mlog_errno(status);
1977 		goto bail;
1978 	}
1979 
1980 	dl = dentry->d_fsdata;
1981 
1982 	status = ocfs2_add_entry(handle, dentry, inode,
1983 				 le64_to_cpu(fe->i_blkno), parent_fe_bh,
1984 				 &lookup);
1985 	if (status < 0) {
1986 		mlog_errno(status);
1987 		goto bail;
1988 	}
1989 
1990 	insert_inode_hash(inode);
1991 	d_instantiate(dentry, inode);
1992 bail:
1993 	if (status < 0 && did_quota)
1994 		dquot_free_space_nodirty(inode,
1995 					ocfs2_clusters_to_bytes(osb->sb, 1));
1996 	if (status < 0 && did_quota_inode)
1997 		dquot_free_inode(inode);
1998 	if (handle)
1999 		ocfs2_commit_trans(osb, handle);
2000 
2001 	ocfs2_inode_unlock(dir, 1);
2002 	if (did_block_signals)
2003 		ocfs2_unblock_signals(&oldset);
2004 
2005 	brelse(new_fe_bh);
2006 	brelse(parent_fe_bh);
2007 	kfree(si.value);
2008 	ocfs2_free_dir_lookup_result(&lookup);
2009 	if (inode_ac)
2010 		ocfs2_free_alloc_context(inode_ac);
2011 	if (data_ac)
2012 		ocfs2_free_alloc_context(data_ac);
2013 	if (xattr_ac)
2014 		ocfs2_free_alloc_context(xattr_ac);
2015 	if ((status < 0) && inode) {
2016 		if (dl)
2017 			ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2018 
2019 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2020 		clear_nlink(inode);
2021 		iput(inode);
2022 	}
2023 
2024 	if (status)
2025 		mlog_errno(status);
2026 
2027 	return status;
2028 }
2029 
2030 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2031 {
2032 	int status, namelen;
2033 
2034 	namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2035 			   (long long)blkno);
2036 	if (namelen <= 0) {
2037 		if (namelen)
2038 			status = namelen;
2039 		else
2040 			status = -EINVAL;
2041 		mlog_errno(status);
2042 		goto bail;
2043 	}
2044 	if (namelen != OCFS2_ORPHAN_NAMELEN) {
2045 		status = -EINVAL;
2046 		mlog_errno(status);
2047 		goto bail;
2048 	}
2049 
2050 	trace_ocfs2_blkno_stringify(blkno, name, namelen);
2051 
2052 	status = 0;
2053 bail:
2054 	if (status < 0)
2055 		mlog_errno(status);
2056 	return status;
2057 }
2058 
2059 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2060 					struct inode **ret_orphan_dir,
2061 					struct buffer_head **ret_orphan_dir_bh)
2062 {
2063 	struct inode *orphan_dir_inode;
2064 	struct buffer_head *orphan_dir_bh = NULL;
2065 	int ret = 0;
2066 
2067 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2068 						       ORPHAN_DIR_SYSTEM_INODE,
2069 						       osb->slot_num);
2070 	if (!orphan_dir_inode) {
2071 		ret = -ENOENT;
2072 		mlog_errno(ret);
2073 		return ret;
2074 	}
2075 
2076 	mutex_lock(&orphan_dir_inode->i_mutex);
2077 
2078 	ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2079 	if (ret < 0) {
2080 		mutex_unlock(&orphan_dir_inode->i_mutex);
2081 		iput(orphan_dir_inode);
2082 
2083 		mlog_errno(ret);
2084 		return ret;
2085 	}
2086 
2087 	*ret_orphan_dir = orphan_dir_inode;
2088 	*ret_orphan_dir_bh = orphan_dir_bh;
2089 
2090 	return 0;
2091 }
2092 
2093 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2094 				      struct buffer_head *orphan_dir_bh,
2095 				      u64 blkno,
2096 				      char *name,
2097 				      struct ocfs2_dir_lookup_result *lookup,
2098 				      bool dio)
2099 {
2100 	int ret;
2101 	struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2102 	int namelen = dio ?
2103 			(OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2104 			OCFS2_ORPHAN_NAMELEN;
2105 
2106 	if (dio) {
2107 		ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2108 				OCFS2_DIO_ORPHAN_PREFIX);
2109 		if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2110 			ret = -EINVAL;
2111 			mlog_errno(ret);
2112 			return ret;
2113 		}
2114 
2115 		ret = ocfs2_blkno_stringify(blkno,
2116 				name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2117 	} else
2118 		ret = ocfs2_blkno_stringify(blkno, name);
2119 	if (ret < 0) {
2120 		mlog_errno(ret);
2121 		return ret;
2122 	}
2123 
2124 	ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2125 					   orphan_dir_bh, name,
2126 					   namelen, lookup);
2127 	if (ret < 0) {
2128 		mlog_errno(ret);
2129 		return ret;
2130 	}
2131 
2132 	return 0;
2133 }
2134 
2135 /**
2136  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2137  * insertion of an orphan.
2138  * @osb: ocfs2 file system
2139  * @ret_orphan_dir: Orphan dir inode - returned locked!
2140  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2141  * @lookup: dir lookup result, to be passed back into functions like
2142  *          ocfs2_orphan_add
2143  *
2144  * Returns zero on success and the ret_orphan_dir, name and lookup
2145  * fields will be populated.
2146  *
2147  * Returns non-zero on failure.
2148  */
2149 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2150 				    struct inode **ret_orphan_dir,
2151 				    u64 blkno,
2152 				    char *name,
2153 				    struct ocfs2_dir_lookup_result *lookup,
2154 				    bool dio)
2155 {
2156 	struct inode *orphan_dir_inode = NULL;
2157 	struct buffer_head *orphan_dir_bh = NULL;
2158 	int ret = 0;
2159 
2160 	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2161 					   &orphan_dir_bh);
2162 	if (ret < 0) {
2163 		mlog_errno(ret);
2164 		return ret;
2165 	}
2166 
2167 	ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2168 					 blkno, name, lookup, dio);
2169 	if (ret < 0) {
2170 		mlog_errno(ret);
2171 		goto out;
2172 	}
2173 
2174 	*ret_orphan_dir = orphan_dir_inode;
2175 
2176 out:
2177 	brelse(orphan_dir_bh);
2178 
2179 	if (ret) {
2180 		ocfs2_inode_unlock(orphan_dir_inode, 1);
2181 		mutex_unlock(&orphan_dir_inode->i_mutex);
2182 		iput(orphan_dir_inode);
2183 	}
2184 
2185 	if (ret)
2186 		mlog_errno(ret);
2187 	return ret;
2188 }
2189 
2190 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2191 			    handle_t *handle,
2192 			    struct inode *inode,
2193 			    struct buffer_head *fe_bh,
2194 			    char *name,
2195 			    struct ocfs2_dir_lookup_result *lookup,
2196 			    struct inode *orphan_dir_inode,
2197 			    bool dio)
2198 {
2199 	struct buffer_head *orphan_dir_bh = NULL;
2200 	int status = 0;
2201 	struct ocfs2_dinode *orphan_fe;
2202 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2203 	int namelen = dio ?
2204 			(OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2205 			OCFS2_ORPHAN_NAMELEN;
2206 
2207 	trace_ocfs2_orphan_add_begin(
2208 				(unsigned long long)OCFS2_I(inode)->ip_blkno);
2209 
2210 	status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2211 	if (status < 0) {
2212 		mlog_errno(status);
2213 		goto leave;
2214 	}
2215 
2216 	status = ocfs2_journal_access_di(handle,
2217 					 INODE_CACHE(orphan_dir_inode),
2218 					 orphan_dir_bh,
2219 					 OCFS2_JOURNAL_ACCESS_WRITE);
2220 	if (status < 0) {
2221 		mlog_errno(status);
2222 		goto leave;
2223 	}
2224 
2225 	/*
2226 	 * We're going to journal the change of i_flags and i_orphaned_slot.
2227 	 * It's safe anyway, though some callers may duplicate the journaling.
2228 	 * Journaling within the func just make the logic look more
2229 	 * straightforward.
2230 	 */
2231 	status = ocfs2_journal_access_di(handle,
2232 					 INODE_CACHE(inode),
2233 					 fe_bh,
2234 					 OCFS2_JOURNAL_ACCESS_WRITE);
2235 	if (status < 0) {
2236 		mlog_errno(status);
2237 		goto leave;
2238 	}
2239 
2240 	/* we're a cluster, and nlink can change on disk from
2241 	 * underneath us... */
2242 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2243 	if (S_ISDIR(inode->i_mode))
2244 		ocfs2_add_links_count(orphan_fe, 1);
2245 	set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2246 	ocfs2_journal_dirty(handle, orphan_dir_bh);
2247 
2248 	status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2249 				   namelen, inode,
2250 				   OCFS2_I(inode)->ip_blkno,
2251 				   orphan_dir_bh, lookup);
2252 	if (status < 0) {
2253 		mlog_errno(status);
2254 		goto rollback;
2255 	}
2256 
2257 	if (dio) {
2258 		/* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2259 		 * slot.
2260 		 */
2261 		fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2262 		fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2263 	} else {
2264 		fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2265 		OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2266 
2267 		/* Record which orphan dir our inode now resides
2268 		 * in. delete_inode will use this to determine which orphan
2269 		 * dir to lock. */
2270 		fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2271 	}
2272 
2273 	ocfs2_journal_dirty(handle, fe_bh);
2274 
2275 	trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2276 				   osb->slot_num);
2277 
2278 rollback:
2279 	if (status < 0) {
2280 		if (S_ISDIR(inode->i_mode))
2281 			ocfs2_add_links_count(orphan_fe, -1);
2282 		set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2283 	}
2284 
2285 leave:
2286 	brelse(orphan_dir_bh);
2287 
2288 	return status;
2289 }
2290 
2291 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2292 int ocfs2_orphan_del(struct ocfs2_super *osb,
2293 		     handle_t *handle,
2294 		     struct inode *orphan_dir_inode,
2295 		     struct inode *inode,
2296 		     struct buffer_head *orphan_dir_bh,
2297 		     bool dio)
2298 {
2299 	const int namelen = OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN;
2300 	char name[namelen + 1];
2301 	struct ocfs2_dinode *orphan_fe;
2302 	int status = 0;
2303 	struct ocfs2_dir_lookup_result lookup = { NULL, };
2304 
2305 	if (dio) {
2306 		status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2307 				OCFS2_DIO_ORPHAN_PREFIX);
2308 		if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2309 			status = -EINVAL;
2310 			mlog_errno(status);
2311 			return status;
2312 		}
2313 
2314 		status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2315 				name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2316 	} else
2317 		status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2318 	if (status < 0) {
2319 		mlog_errno(status);
2320 		goto leave;
2321 	}
2322 
2323 	trace_ocfs2_orphan_del(
2324 	     (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2325 	     name, strlen(name));
2326 
2327 	/* find it's spot in the orphan directory */
2328 	status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2329 				  &lookup);
2330 	if (status) {
2331 		mlog_errno(status);
2332 		goto leave;
2333 	}
2334 
2335 	/* remove it from the orphan directory */
2336 	status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2337 	if (status < 0) {
2338 		mlog_errno(status);
2339 		goto leave;
2340 	}
2341 
2342 	status = ocfs2_journal_access_di(handle,
2343 					 INODE_CACHE(orphan_dir_inode),
2344 					 orphan_dir_bh,
2345 					 OCFS2_JOURNAL_ACCESS_WRITE);
2346 	if (status < 0) {
2347 		mlog_errno(status);
2348 		goto leave;
2349 	}
2350 
2351 	/* do the i_nlink dance! :) */
2352 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2353 	if (S_ISDIR(inode->i_mode))
2354 		ocfs2_add_links_count(orphan_fe, -1);
2355 	set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2356 	ocfs2_journal_dirty(handle, orphan_dir_bh);
2357 
2358 leave:
2359 	ocfs2_free_dir_lookup_result(&lookup);
2360 
2361 	if (status)
2362 		mlog_errno(status);
2363 	return status;
2364 }
2365 
2366 /**
2367  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2368  * allocated file. This is different from the typical 'add to orphan dir'
2369  * operation in that the inode does not yet exist. This is a problem because
2370  * the orphan dir stringifies the inode block number to come up with it's
2371  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2372  * problem. This function works around it by calling deeper into the orphan
2373  * and suballoc code than other callers. Use this only by necessity.
2374  * @dir: The directory which this inode will ultimately wind up under - not the
2375  * orphan dir!
2376  * @dir_bh: buffer_head the @dir inode block
2377  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2378  * with the string to be used for orphan dirent. Pass back to the orphan dir
2379  * code.
2380  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2381  * dir code.
2382  * @ret_di_blkno: block number where the new inode will be allocated.
2383  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2384  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2385  *
2386  * Returns zero on success and the ret_orphan_dir, name and lookup
2387  * fields will be populated.
2388  *
2389  * Returns non-zero on failure.
2390  */
2391 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2392 					struct buffer_head *dir_bh,
2393 					char *orphan_name,
2394 					struct inode **ret_orphan_dir,
2395 					u64 *ret_di_blkno,
2396 					struct ocfs2_dir_lookup_result *orphan_insert,
2397 					struct ocfs2_alloc_context **ret_inode_ac)
2398 {
2399 	int ret;
2400 	u64 di_blkno;
2401 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2402 	struct inode *orphan_dir = NULL;
2403 	struct buffer_head *orphan_dir_bh = NULL;
2404 	struct ocfs2_alloc_context *inode_ac = NULL;
2405 
2406 	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2407 	if (ret < 0) {
2408 		mlog_errno(ret);
2409 		return ret;
2410 	}
2411 
2412 	/* reserve an inode spot */
2413 	ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2414 	if (ret < 0) {
2415 		if (ret != -ENOSPC)
2416 			mlog_errno(ret);
2417 		goto out;
2418 	}
2419 
2420 	ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2421 				       &di_blkno);
2422 	if (ret) {
2423 		mlog_errno(ret);
2424 		goto out;
2425 	}
2426 
2427 	ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2428 					 di_blkno, orphan_name, orphan_insert,
2429 					 false);
2430 	if (ret < 0) {
2431 		mlog_errno(ret);
2432 		goto out;
2433 	}
2434 
2435 out:
2436 	if (ret == 0) {
2437 		*ret_orphan_dir = orphan_dir;
2438 		*ret_di_blkno = di_blkno;
2439 		*ret_inode_ac = inode_ac;
2440 		/*
2441 		 * orphan_name and orphan_insert are already up to
2442 		 * date via prepare_orphan_dir
2443 		 */
2444 	} else {
2445 		/* Unroll reserve_new_inode* */
2446 		if (inode_ac)
2447 			ocfs2_free_alloc_context(inode_ac);
2448 
2449 		/* Unroll orphan dir locking */
2450 		mutex_unlock(&orphan_dir->i_mutex);
2451 		ocfs2_inode_unlock(orphan_dir, 1);
2452 		iput(orphan_dir);
2453 	}
2454 
2455 	brelse(orphan_dir_bh);
2456 
2457 	return ret;
2458 }
2459 
2460 int ocfs2_create_inode_in_orphan(struct inode *dir,
2461 				 int mode,
2462 				 struct inode **new_inode)
2463 {
2464 	int status, did_quota_inode = 0;
2465 	struct inode *inode = NULL;
2466 	struct inode *orphan_dir = NULL;
2467 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2468 	struct ocfs2_dinode *di = NULL;
2469 	handle_t *handle = NULL;
2470 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2471 	struct buffer_head *parent_di_bh = NULL;
2472 	struct buffer_head *new_di_bh = NULL;
2473 	struct ocfs2_alloc_context *inode_ac = NULL;
2474 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2475 	u64 uninitialized_var(di_blkno), suballoc_loc;
2476 	u16 suballoc_bit;
2477 
2478 	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2479 	if (status < 0) {
2480 		if (status != -ENOENT)
2481 			mlog_errno(status);
2482 		return status;
2483 	}
2484 
2485 	status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2486 					      orphan_name, &orphan_dir,
2487 					      &di_blkno, &orphan_insert, &inode_ac);
2488 	if (status < 0) {
2489 		if (status != -ENOSPC)
2490 			mlog_errno(status);
2491 		goto leave;
2492 	}
2493 
2494 	inode = ocfs2_get_init_inode(dir, mode);
2495 	if (!inode) {
2496 		status = -ENOMEM;
2497 		mlog_errno(status);
2498 		goto leave;
2499 	}
2500 
2501 	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2502 	if (IS_ERR(handle)) {
2503 		status = PTR_ERR(handle);
2504 		handle = NULL;
2505 		mlog_errno(status);
2506 		goto leave;
2507 	}
2508 
2509 	status = dquot_alloc_inode(inode);
2510 	if (status)
2511 		goto leave;
2512 	did_quota_inode = 1;
2513 
2514 	status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2515 					      &suballoc_loc,
2516 					      &suballoc_bit, di_blkno);
2517 	if (status < 0) {
2518 		mlog_errno(status);
2519 		goto leave;
2520 	}
2521 
2522 	clear_nlink(inode);
2523 	/* do the real work now. */
2524 	status = __ocfs2_mknod_locked(dir, inode,
2525 				      0, &new_di_bh, parent_di_bh, handle,
2526 				      inode_ac, di_blkno, suballoc_loc,
2527 				      suballoc_bit);
2528 	if (status < 0) {
2529 		mlog_errno(status);
2530 		goto leave;
2531 	}
2532 
2533 	di = (struct ocfs2_dinode *)new_di_bh->b_data;
2534 	status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2535 				  &orphan_insert, orphan_dir, false);
2536 	if (status < 0) {
2537 		mlog_errno(status);
2538 		goto leave;
2539 	}
2540 
2541 	/* get open lock so that only nodes can't remove it from orphan dir. */
2542 	status = ocfs2_open_lock(inode);
2543 	if (status < 0)
2544 		mlog_errno(status);
2545 
2546 	insert_inode_hash(inode);
2547 leave:
2548 	if (status < 0 && did_quota_inode)
2549 		dquot_free_inode(inode);
2550 	if (handle)
2551 		ocfs2_commit_trans(osb, handle);
2552 
2553 	if (orphan_dir) {
2554 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
2555 		ocfs2_inode_unlock(orphan_dir, 1);
2556 		mutex_unlock(&orphan_dir->i_mutex);
2557 		iput(orphan_dir);
2558 	}
2559 
2560 	if ((status < 0) && inode) {
2561 		clear_nlink(inode);
2562 		iput(inode);
2563 	}
2564 
2565 	if (inode_ac)
2566 		ocfs2_free_alloc_context(inode_ac);
2567 
2568 	brelse(new_di_bh);
2569 
2570 	if (!status)
2571 		*new_inode = inode;
2572 
2573 	ocfs2_free_dir_lookup_result(&orphan_insert);
2574 
2575 	ocfs2_inode_unlock(dir, 1);
2576 	brelse(parent_di_bh);
2577 	return status;
2578 }
2579 
2580 static int ocfs2_dio_orphan_recovered(struct inode *inode)
2581 {
2582 	int ret;
2583 	struct buffer_head *di_bh = NULL;
2584 	struct ocfs2_dinode *di = NULL;
2585 
2586 	ret = ocfs2_inode_lock(inode, &di_bh, 1);
2587 	if (ret < 0) {
2588 		mlog_errno(ret);
2589 		return 0;
2590 	}
2591 
2592 	di = (struct ocfs2_dinode *) di_bh->b_data;
2593 	ret = !(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL));
2594 	ocfs2_inode_unlock(inode, 1);
2595 	brelse(di_bh);
2596 
2597 	return ret;
2598 }
2599 
2600 #define OCFS2_DIO_ORPHANED_FL_CHECK_INTERVAL 10000
2601 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2602 	struct inode *inode)
2603 {
2604 	char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2605 	struct inode *orphan_dir_inode = NULL;
2606 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2607 	struct buffer_head *di_bh = NULL;
2608 	int status = 0;
2609 	handle_t *handle = NULL;
2610 	struct ocfs2_dinode *di = NULL;
2611 
2612 restart:
2613 	status = ocfs2_inode_lock(inode, &di_bh, 1);
2614 	if (status < 0) {
2615 		mlog_errno(status);
2616 		goto bail;
2617 	}
2618 
2619 	di = (struct ocfs2_dinode *) di_bh->b_data;
2620 	/*
2621 	 * Another append dio crashed?
2622 	 * If so, wait for recovery first.
2623 	 */
2624 	if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2625 		ocfs2_inode_unlock(inode, 1);
2626 		brelse(di_bh);
2627 		wait_event_interruptible_timeout(OCFS2_I(inode)->append_dio_wq,
2628 				ocfs2_dio_orphan_recovered(inode),
2629 				msecs_to_jiffies(OCFS2_DIO_ORPHANED_FL_CHECK_INTERVAL));
2630 		goto restart;
2631 	}
2632 
2633 	status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2634 			OCFS2_I(inode)->ip_blkno,
2635 			orphan_name,
2636 			&orphan_insert,
2637 			true);
2638 	if (status < 0) {
2639 		mlog_errno(status);
2640 		goto bail_unlock_inode;
2641 	}
2642 
2643 	handle = ocfs2_start_trans(osb,
2644 			OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2645 	if (IS_ERR(handle)) {
2646 		status = PTR_ERR(handle);
2647 		goto bail_unlock_orphan;
2648 	}
2649 
2650 	status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2651 			&orphan_insert, orphan_dir_inode, true);
2652 	if (status)
2653 		mlog_errno(status);
2654 
2655 	ocfs2_commit_trans(osb, handle);
2656 
2657 bail_unlock_orphan:
2658 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2659 	mutex_unlock(&orphan_dir_inode->i_mutex);
2660 	iput(orphan_dir_inode);
2661 
2662 	ocfs2_free_dir_lookup_result(&orphan_insert);
2663 
2664 bail_unlock_inode:
2665 	ocfs2_inode_unlock(inode, 1);
2666 	brelse(di_bh);
2667 
2668 bail:
2669 	return status;
2670 }
2671 
2672 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2673 		struct inode *inode, int update_isize,
2674 		loff_t end)
2675 {
2676 	struct inode *orphan_dir_inode = NULL;
2677 	struct buffer_head *orphan_dir_bh = NULL;
2678 	struct buffer_head *di_bh = NULL;
2679 	struct ocfs2_dinode *di = NULL;
2680 	handle_t *handle = NULL;
2681 	int status = 0;
2682 
2683 	status = ocfs2_inode_lock(inode, &di_bh, 1);
2684 	if (status < 0) {
2685 		mlog_errno(status);
2686 		goto bail;
2687 	}
2688 	di = (struct ocfs2_dinode *) di_bh->b_data;
2689 
2690 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2691 			ORPHAN_DIR_SYSTEM_INODE,
2692 			le16_to_cpu(di->i_dio_orphaned_slot));
2693 	if (!orphan_dir_inode) {
2694 		status = -ENOENT;
2695 		mlog_errno(status);
2696 		goto bail_unlock_inode;
2697 	}
2698 
2699 	mutex_lock(&orphan_dir_inode->i_mutex);
2700 	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2701 	if (status < 0) {
2702 		mutex_unlock(&orphan_dir_inode->i_mutex);
2703 		iput(orphan_dir_inode);
2704 		mlog_errno(status);
2705 		goto bail_unlock_inode;
2706 	}
2707 
2708 	handle = ocfs2_start_trans(osb,
2709 			OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2710 	if (IS_ERR(handle)) {
2711 		status = PTR_ERR(handle);
2712 		goto bail_unlock_orphan;
2713 	}
2714 
2715 	BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2716 
2717 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2718 				inode, orphan_dir_bh, true);
2719 	if (status < 0) {
2720 		mlog_errno(status);
2721 		goto bail_commit;
2722 	}
2723 
2724 	status = ocfs2_journal_access_di(handle,
2725 			INODE_CACHE(inode),
2726 			di_bh,
2727 			OCFS2_JOURNAL_ACCESS_WRITE);
2728 	if (status < 0) {
2729 		mlog_errno(status);
2730 		goto bail_commit;
2731 	}
2732 
2733 	di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2734 	di->i_dio_orphaned_slot = 0;
2735 
2736 	if (update_isize) {
2737 		status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2738 		if (status)
2739 			mlog_errno(status);
2740 	} else
2741 		ocfs2_journal_dirty(handle, di_bh);
2742 
2743 bail_commit:
2744 	ocfs2_commit_trans(osb, handle);
2745 
2746 bail_unlock_orphan:
2747 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2748 	mutex_unlock(&orphan_dir_inode->i_mutex);
2749 	brelse(orphan_dir_bh);
2750 	iput(orphan_dir_inode);
2751 
2752 bail_unlock_inode:
2753 	ocfs2_inode_unlock(inode, 1);
2754 	brelse(di_bh);
2755 
2756 bail:
2757 	return status;
2758 }
2759 
2760 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2761 				   struct inode *inode,
2762 				   struct dentry *dentry)
2763 {
2764 	int status = 0;
2765 	struct buffer_head *parent_di_bh = NULL;
2766 	handle_t *handle = NULL;
2767 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2768 	struct ocfs2_dinode *dir_di, *di;
2769 	struct inode *orphan_dir_inode = NULL;
2770 	struct buffer_head *orphan_dir_bh = NULL;
2771 	struct buffer_head *di_bh = NULL;
2772 	struct ocfs2_dir_lookup_result lookup = { NULL, };
2773 
2774 	trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2775 				dentry->d_name.len, dentry->d_name.name,
2776 				(unsigned long long)OCFS2_I(dir)->ip_blkno,
2777 				(unsigned long long)OCFS2_I(inode)->ip_blkno);
2778 
2779 	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2780 	if (status < 0) {
2781 		if (status != -ENOENT)
2782 			mlog_errno(status);
2783 		return status;
2784 	}
2785 
2786 	dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2787 	if (!dir_di->i_links_count) {
2788 		/* can't make a file in a deleted directory. */
2789 		status = -ENOENT;
2790 		goto leave;
2791 	}
2792 
2793 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2794 					   dentry->d_name.len);
2795 	if (status)
2796 		goto leave;
2797 
2798 	/* get a spot inside the dir. */
2799 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2800 					      dentry->d_name.name,
2801 					      dentry->d_name.len, &lookup);
2802 	if (status < 0) {
2803 		mlog_errno(status);
2804 		goto leave;
2805 	}
2806 
2807 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2808 						       ORPHAN_DIR_SYSTEM_INODE,
2809 						       osb->slot_num);
2810 	if (!orphan_dir_inode) {
2811 		status = -ENOENT;
2812 		mlog_errno(status);
2813 		goto leave;
2814 	}
2815 
2816 	mutex_lock(&orphan_dir_inode->i_mutex);
2817 
2818 	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2819 	if (status < 0) {
2820 		mlog_errno(status);
2821 		mutex_unlock(&orphan_dir_inode->i_mutex);
2822 		iput(orphan_dir_inode);
2823 		goto leave;
2824 	}
2825 
2826 	status = ocfs2_read_inode_block(inode, &di_bh);
2827 	if (status < 0) {
2828 		mlog_errno(status);
2829 		goto orphan_unlock;
2830 	}
2831 
2832 	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2833 	if (IS_ERR(handle)) {
2834 		status = PTR_ERR(handle);
2835 		handle = NULL;
2836 		mlog_errno(status);
2837 		goto orphan_unlock;
2838 	}
2839 
2840 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2841 					 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2842 	if (status < 0) {
2843 		mlog_errno(status);
2844 		goto out_commit;
2845 	}
2846 
2847 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2848 				  orphan_dir_bh, false);
2849 	if (status < 0) {
2850 		mlog_errno(status);
2851 		goto out_commit;
2852 	}
2853 
2854 	di = (struct ocfs2_dinode *)di_bh->b_data;
2855 	di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2856 	di->i_orphaned_slot = 0;
2857 	set_nlink(inode, 1);
2858 	ocfs2_set_links_count(di, inode->i_nlink);
2859 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
2860 	ocfs2_journal_dirty(handle, di_bh);
2861 
2862 	status = ocfs2_add_entry(handle, dentry, inode,
2863 				 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2864 				 &lookup);
2865 	if (status < 0) {
2866 		mlog_errno(status);
2867 		goto out_commit;
2868 	}
2869 
2870 	status = ocfs2_dentry_attach_lock(dentry, inode,
2871 					  OCFS2_I(dir)->ip_blkno);
2872 	if (status) {
2873 		mlog_errno(status);
2874 		goto out_commit;
2875 	}
2876 
2877 	d_instantiate(dentry, inode);
2878 	status = 0;
2879 out_commit:
2880 	ocfs2_commit_trans(osb, handle);
2881 orphan_unlock:
2882 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2883 	mutex_unlock(&orphan_dir_inode->i_mutex);
2884 	iput(orphan_dir_inode);
2885 leave:
2886 
2887 	ocfs2_inode_unlock(dir, 1);
2888 
2889 	brelse(di_bh);
2890 	brelse(parent_di_bh);
2891 	brelse(orphan_dir_bh);
2892 
2893 	ocfs2_free_dir_lookup_result(&lookup);
2894 
2895 	if (status)
2896 		mlog_errno(status);
2897 
2898 	return status;
2899 }
2900 
2901 const struct inode_operations ocfs2_dir_iops = {
2902 	.create		= ocfs2_create,
2903 	.lookup		= ocfs2_lookup,
2904 	.link		= ocfs2_link,
2905 	.unlink		= ocfs2_unlink,
2906 	.rmdir		= ocfs2_unlink,
2907 	.symlink	= ocfs2_symlink,
2908 	.mkdir		= ocfs2_mkdir,
2909 	.mknod		= ocfs2_mknod,
2910 	.rename		= ocfs2_rename,
2911 	.setattr	= ocfs2_setattr,
2912 	.getattr	= ocfs2_getattr,
2913 	.permission	= ocfs2_permission,
2914 	.setxattr	= generic_setxattr,
2915 	.getxattr	= generic_getxattr,
2916 	.listxattr	= ocfs2_listxattr,
2917 	.removexattr	= generic_removexattr,
2918 	.fiemap         = ocfs2_fiemap,
2919 	.get_acl	= ocfs2_iop_get_acl,
2920 	.set_acl	= ocfs2_iop_set_acl,
2921 };
2922