xref: /openbmc/linux/fs/ocfs2/namei.c (revision f17f06a0)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* -*- mode: c; c-basic-offset: 8; -*-
3  * vim: noexpandtab sw=8 ts=8 sts=0:
4  *
5  * namei.c
6  *
7  * Create and rename file, directory, symlinks
8  *
9  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
10  *
11  *  Portions of this code from linux/fs/ext3/dir.c
12  *
13  *  Copyright (C) 1992, 1993, 1994, 1995
14  *  Remy Card (card@masi.ibp.fr)
15  *  Laboratoire MASI - Institut Blaise pascal
16  *  Universite Pierre et Marie Curie (Paris VI)
17  *
18  *   from
19  *
20  *   linux/fs/minix/dir.c
21  *
22  *   Copyright (C) 1991, 1992 Linux Torvalds
23  */
24 
25 #include <linux/fs.h>
26 #include <linux/types.h>
27 #include <linux/slab.h>
28 #include <linux/highmem.h>
29 #include <linux/quotaops.h>
30 #include <linux/iversion.h>
31 
32 #include <cluster/masklog.h>
33 
34 #include "ocfs2.h"
35 
36 #include "alloc.h"
37 #include "dcache.h"
38 #include "dir.h"
39 #include "dlmglue.h"
40 #include "extent_map.h"
41 #include "file.h"
42 #include "inode.h"
43 #include "journal.h"
44 #include "namei.h"
45 #include "suballoc.h"
46 #include "super.h"
47 #include "symlink.h"
48 #include "sysfile.h"
49 #include "uptodate.h"
50 #include "xattr.h"
51 #include "acl.h"
52 #include "ocfs2_trace.h"
53 
54 #include "buffer_head_io.h"
55 
56 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
57 			      struct inode *dir,
58 			      struct inode *inode,
59 			      dev_t dev,
60 			      struct buffer_head **new_fe_bh,
61 			      struct buffer_head *parent_fe_bh,
62 			      handle_t *handle,
63 			      struct ocfs2_alloc_context *inode_ac);
64 
65 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
66 				    struct inode **ret_orphan_dir,
67 				    u64 blkno,
68 				    char *name,
69 				    struct ocfs2_dir_lookup_result *lookup,
70 				    bool dio);
71 
72 static int ocfs2_orphan_add(struct ocfs2_super *osb,
73 			    handle_t *handle,
74 			    struct inode *inode,
75 			    struct buffer_head *fe_bh,
76 			    char *name,
77 			    struct ocfs2_dir_lookup_result *lookup,
78 			    struct inode *orphan_dir_inode,
79 			    bool dio);
80 
81 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
82 				     handle_t *handle,
83 				     struct inode *inode,
84 				     const char *symname);
85 
86 static int ocfs2_double_lock(struct ocfs2_super *osb,
87 			     struct buffer_head **bh1,
88 			     struct inode *inode1,
89 			     struct buffer_head **bh2,
90 			     struct inode *inode2,
91 			     int rename);
92 
93 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2);
94 /* An orphan dir name is an 8 byte value, printed as a hex string */
95 #define OCFS2_ORPHAN_NAMELEN ((int)(2 * sizeof(u64)))
96 
97 static struct dentry *ocfs2_lookup(struct inode *dir, struct dentry *dentry,
98 				   unsigned int flags)
99 {
100 	int status;
101 	u64 blkno;
102 	struct inode *inode = NULL;
103 	struct dentry *ret;
104 	struct ocfs2_inode_info *oi;
105 
106 	trace_ocfs2_lookup(dir, dentry, dentry->d_name.len,
107 			   dentry->d_name.name,
108 			   (unsigned long long)OCFS2_I(dir)->ip_blkno, 0);
109 
110 	if (dentry->d_name.len > OCFS2_MAX_FILENAME_LEN) {
111 		ret = ERR_PTR(-ENAMETOOLONG);
112 		goto bail;
113 	}
114 
115 	status = ocfs2_inode_lock_nested(dir, NULL, 0, OI_LS_PARENT);
116 	if (status < 0) {
117 		if (status != -ENOENT)
118 			mlog_errno(status);
119 		ret = ERR_PTR(status);
120 		goto bail;
121 	}
122 
123 	status = ocfs2_lookup_ino_from_name(dir, dentry->d_name.name,
124 					    dentry->d_name.len, &blkno);
125 	if (status < 0)
126 		goto bail_add;
127 
128 	inode = ocfs2_iget(OCFS2_SB(dir->i_sb), blkno, 0, 0);
129 	if (IS_ERR(inode)) {
130 		ret = ERR_PTR(-EACCES);
131 		goto bail_unlock;
132 	}
133 
134 	oi = OCFS2_I(inode);
135 	/* Clear any orphaned state... If we were able to look up the
136 	 * inode from a directory, it certainly can't be orphaned. We
137 	 * might have the bad state from a node which intended to
138 	 * orphan this inode but crashed before it could commit the
139 	 * unlink. */
140 	spin_lock(&oi->ip_lock);
141 	oi->ip_flags &= ~OCFS2_INODE_MAYBE_ORPHANED;
142 	spin_unlock(&oi->ip_lock);
143 
144 bail_add:
145 	ret = d_splice_alias(inode, dentry);
146 
147 	if (inode) {
148 		/*
149 		 * If d_splice_alias() finds a DCACHE_DISCONNECTED
150 		 * dentry, it will d_move() it on top of ourse. The
151 		 * return value will indicate this however, so in
152 		 * those cases, we switch them around for the locking
153 		 * code.
154 		 *
155 		 * NOTE: This dentry already has ->d_op set from
156 		 * ocfs2_get_parent() and ocfs2_get_dentry()
157 		 */
158 		if (!IS_ERR_OR_NULL(ret))
159 			dentry = ret;
160 
161 		status = ocfs2_dentry_attach_lock(dentry, inode,
162 						  OCFS2_I(dir)->ip_blkno);
163 		if (status) {
164 			mlog_errno(status);
165 			ret = ERR_PTR(status);
166 			goto bail_unlock;
167 		}
168 	} else
169 		ocfs2_dentry_attach_gen(dentry);
170 
171 bail_unlock:
172 	/* Don't drop the cluster lock until *after* the d_add --
173 	 * unlink on another node will message us to remove that
174 	 * dentry under this lock so otherwise we can race this with
175 	 * the downconvert thread and have a stale dentry. */
176 	ocfs2_inode_unlock(dir, 0);
177 
178 bail:
179 
180 	trace_ocfs2_lookup_ret(ret);
181 
182 	return ret;
183 }
184 
185 static struct inode *ocfs2_get_init_inode(struct inode *dir, umode_t mode)
186 {
187 	struct inode *inode;
188 	int status;
189 
190 	inode = new_inode(dir->i_sb);
191 	if (!inode) {
192 		mlog(ML_ERROR, "new_inode failed!\n");
193 		return ERR_PTR(-ENOMEM);
194 	}
195 
196 	/* populate as many fields early on as possible - many of
197 	 * these are used by the support functions here and in
198 	 * callers. */
199 	if (S_ISDIR(mode))
200 		set_nlink(inode, 2);
201 	inode_init_owner(inode, dir, mode);
202 	status = dquot_initialize(inode);
203 	if (status)
204 		return ERR_PTR(status);
205 
206 	return inode;
207 }
208 
209 static void ocfs2_cleanup_add_entry_failure(struct ocfs2_super *osb,
210 		struct dentry *dentry, struct inode *inode)
211 {
212 	struct ocfs2_dentry_lock *dl = dentry->d_fsdata;
213 
214 	ocfs2_simple_drop_lockres(osb, &dl->dl_lockres);
215 	ocfs2_lock_res_free(&dl->dl_lockres);
216 	BUG_ON(dl->dl_count != 1);
217 	spin_lock(&dentry_attach_lock);
218 	dentry->d_fsdata = NULL;
219 	spin_unlock(&dentry_attach_lock);
220 	kfree(dl);
221 	iput(inode);
222 }
223 
224 static int ocfs2_mknod(struct inode *dir,
225 		       struct dentry *dentry,
226 		       umode_t mode,
227 		       dev_t dev)
228 {
229 	int status = 0;
230 	struct buffer_head *parent_fe_bh = NULL;
231 	handle_t *handle = NULL;
232 	struct ocfs2_super *osb;
233 	struct ocfs2_dinode *dirfe;
234 	struct buffer_head *new_fe_bh = NULL;
235 	struct inode *inode = NULL;
236 	struct ocfs2_alloc_context *inode_ac = NULL;
237 	struct ocfs2_alloc_context *data_ac = NULL;
238 	struct ocfs2_alloc_context *meta_ac = NULL;
239 	int want_clusters = 0;
240 	int want_meta = 0;
241 	int xattr_credits = 0;
242 	struct ocfs2_security_xattr_info si = {
243 		.enable = 1,
244 	};
245 	int did_quota_inode = 0;
246 	struct ocfs2_dir_lookup_result lookup = { NULL, };
247 	sigset_t oldset;
248 	int did_block_signals = 0;
249 	struct ocfs2_dentry_lock *dl = NULL;
250 
251 	trace_ocfs2_mknod(dir, dentry, dentry->d_name.len, dentry->d_name.name,
252 			  (unsigned long long)OCFS2_I(dir)->ip_blkno,
253 			  (unsigned long)dev, mode);
254 
255 	status = dquot_initialize(dir);
256 	if (status) {
257 		mlog_errno(status);
258 		return status;
259 	}
260 
261 	/* get our super block */
262 	osb = OCFS2_SB(dir->i_sb);
263 
264 	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
265 	if (status < 0) {
266 		if (status != -ENOENT)
267 			mlog_errno(status);
268 		return status;
269 	}
270 
271 	if (S_ISDIR(mode) && (dir->i_nlink >= ocfs2_link_max(osb))) {
272 		status = -EMLINK;
273 		goto leave;
274 	}
275 
276 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
277 	if (!ocfs2_read_links_count(dirfe)) {
278 		/* can't make a file in a deleted directory. */
279 		status = -ENOENT;
280 		goto leave;
281 	}
282 
283 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
284 					   dentry->d_name.len);
285 	if (status)
286 		goto leave;
287 
288 	/* get a spot inside the dir. */
289 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
290 					      dentry->d_name.name,
291 					      dentry->d_name.len, &lookup);
292 	if (status < 0) {
293 		mlog_errno(status);
294 		goto leave;
295 	}
296 
297 	/* reserve an inode spot */
298 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
299 	if (status < 0) {
300 		if (status != -ENOSPC)
301 			mlog_errno(status);
302 		goto leave;
303 	}
304 
305 	inode = ocfs2_get_init_inode(dir, mode);
306 	if (IS_ERR(inode)) {
307 		status = PTR_ERR(inode);
308 		inode = NULL;
309 		mlog_errno(status);
310 		goto leave;
311 	}
312 
313 	/* get security xattr */
314 	status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
315 	if (status) {
316 		if (status == -EOPNOTSUPP)
317 			si.enable = 0;
318 		else {
319 			mlog_errno(status);
320 			goto leave;
321 		}
322 	}
323 
324 	/* calculate meta data/clusters for setting security and acl xattr */
325 	status = ocfs2_calc_xattr_init(dir, parent_fe_bh, mode,
326 				       &si, &want_clusters,
327 				       &xattr_credits, &want_meta);
328 	if (status < 0) {
329 		mlog_errno(status);
330 		goto leave;
331 	}
332 
333 	/* Reserve a cluster if creating an extent based directory. */
334 	if (S_ISDIR(mode) && !ocfs2_supports_inline_data(osb)) {
335 		want_clusters += 1;
336 
337 		/* Dir indexing requires extra space as well */
338 		if (ocfs2_supports_indexed_dirs(osb))
339 			want_meta++;
340 	}
341 
342 	status = ocfs2_reserve_new_metadata_blocks(osb, want_meta, &meta_ac);
343 	if (status < 0) {
344 		if (status != -ENOSPC)
345 			mlog_errno(status);
346 		goto leave;
347 	}
348 
349 	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
350 	if (status < 0) {
351 		if (status != -ENOSPC)
352 			mlog_errno(status);
353 		goto leave;
354 	}
355 
356 	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb,
357 							    S_ISDIR(mode),
358 							    xattr_credits));
359 	if (IS_ERR(handle)) {
360 		status = PTR_ERR(handle);
361 		handle = NULL;
362 		mlog_errno(status);
363 		goto leave;
364 	}
365 
366 	/* Starting to change things, restart is no longer possible. */
367 	ocfs2_block_signals(&oldset);
368 	did_block_signals = 1;
369 
370 	status = dquot_alloc_inode(inode);
371 	if (status)
372 		goto leave;
373 	did_quota_inode = 1;
374 
375 	/* do the real work now. */
376 	status = ocfs2_mknod_locked(osb, dir, inode, dev,
377 				    &new_fe_bh, parent_fe_bh, handle,
378 				    inode_ac);
379 	if (status < 0) {
380 		mlog_errno(status);
381 		goto leave;
382 	}
383 
384 	if (S_ISDIR(mode)) {
385 		status = ocfs2_fill_new_dir(osb, handle, dir, inode,
386 					    new_fe_bh, data_ac, meta_ac);
387 		if (status < 0) {
388 			mlog_errno(status);
389 			goto leave;
390 		}
391 
392 		status = ocfs2_journal_access_di(handle, INODE_CACHE(dir),
393 						 parent_fe_bh,
394 						 OCFS2_JOURNAL_ACCESS_WRITE);
395 		if (status < 0) {
396 			mlog_errno(status);
397 			goto leave;
398 		}
399 		ocfs2_add_links_count(dirfe, 1);
400 		ocfs2_journal_dirty(handle, parent_fe_bh);
401 		inc_nlink(dir);
402 	}
403 
404 	status = ocfs2_init_acl(handle, inode, dir, new_fe_bh, parent_fe_bh,
405 			 meta_ac, data_ac);
406 
407 	if (status < 0) {
408 		mlog_errno(status);
409 		goto leave;
410 	}
411 
412 	if (si.enable) {
413 		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
414 						 meta_ac, data_ac);
415 		if (status < 0) {
416 			mlog_errno(status);
417 			goto leave;
418 		}
419 	}
420 
421 	/*
422 	 * Do this before adding the entry to the directory. We add
423 	 * also set d_op after success so that ->d_iput() will cleanup
424 	 * the dentry lock even if ocfs2_add_entry() fails below.
425 	 */
426 	status = ocfs2_dentry_attach_lock(dentry, inode,
427 					  OCFS2_I(dir)->ip_blkno);
428 	if (status) {
429 		mlog_errno(status);
430 		goto leave;
431 	}
432 
433 	dl = dentry->d_fsdata;
434 
435 	status = ocfs2_add_entry(handle, dentry, inode,
436 				 OCFS2_I(inode)->ip_blkno, parent_fe_bh,
437 				 &lookup);
438 	if (status < 0) {
439 		mlog_errno(status);
440 		goto leave;
441 	}
442 
443 	insert_inode_hash(inode);
444 	d_instantiate(dentry, inode);
445 	status = 0;
446 leave:
447 	if (status < 0 && did_quota_inode)
448 		dquot_free_inode(inode);
449 	if (handle)
450 		ocfs2_commit_trans(osb, handle);
451 
452 	ocfs2_inode_unlock(dir, 1);
453 	if (did_block_signals)
454 		ocfs2_unblock_signals(&oldset);
455 
456 	brelse(new_fe_bh);
457 	brelse(parent_fe_bh);
458 	kfree(si.value);
459 
460 	ocfs2_free_dir_lookup_result(&lookup);
461 
462 	if (inode_ac)
463 		ocfs2_free_alloc_context(inode_ac);
464 
465 	if (data_ac)
466 		ocfs2_free_alloc_context(data_ac);
467 
468 	if (meta_ac)
469 		ocfs2_free_alloc_context(meta_ac);
470 
471 	/*
472 	 * We should call iput after the i_mutex of the bitmap been
473 	 * unlocked in ocfs2_free_alloc_context, or the
474 	 * ocfs2_delete_inode will mutex_lock again.
475 	 */
476 	if ((status < 0) && inode) {
477 		if (dl)
478 			ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
479 
480 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
481 		clear_nlink(inode);
482 		iput(inode);
483 	}
484 
485 	if (status)
486 		mlog_errno(status);
487 
488 	return status;
489 }
490 
491 static int __ocfs2_mknod_locked(struct inode *dir,
492 				struct inode *inode,
493 				dev_t dev,
494 				struct buffer_head **new_fe_bh,
495 				struct buffer_head *parent_fe_bh,
496 				handle_t *handle,
497 				struct ocfs2_alloc_context *inode_ac,
498 				u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
499 {
500 	int status = 0;
501 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
502 	struct ocfs2_dinode *fe = NULL;
503 	struct ocfs2_extent_list *fel;
504 	u16 feat;
505 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
506 	struct timespec64 ts;
507 
508 	*new_fe_bh = NULL;
509 
510 	/* populate as many fields early on as possible - many of
511 	 * these are used by the support functions here and in
512 	 * callers. */
513 	inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
514 	oi->ip_blkno = fe_blkno;
515 	spin_lock(&osb->osb_lock);
516 	inode->i_generation = osb->s_next_generation++;
517 	spin_unlock(&osb->osb_lock);
518 
519 	*new_fe_bh = sb_getblk(osb->sb, fe_blkno);
520 	if (!*new_fe_bh) {
521 		status = -ENOMEM;
522 		mlog_errno(status);
523 		goto leave;
524 	}
525 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
526 
527 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
528 					 *new_fe_bh,
529 					 OCFS2_JOURNAL_ACCESS_CREATE);
530 	if (status < 0) {
531 		mlog_errno(status);
532 		goto leave;
533 	}
534 
535 	fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
536 	memset(fe, 0, osb->sb->s_blocksize);
537 
538 	fe->i_generation = cpu_to_le32(inode->i_generation);
539 	fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
540 	fe->i_blkno = cpu_to_le64(fe_blkno);
541 	fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
542 	fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
543 	fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
544 	fe->i_uid = cpu_to_le32(i_uid_read(inode));
545 	fe->i_gid = cpu_to_le32(i_gid_read(inode));
546 	fe->i_mode = cpu_to_le16(inode->i_mode);
547 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
548 		fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
549 
550 	ocfs2_set_links_count(fe, inode->i_nlink);
551 
552 	fe->i_last_eb_blk = 0;
553 	strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
554 	fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
555 	ktime_get_real_ts64(&ts);
556 	fe->i_atime = fe->i_ctime = fe->i_mtime =
557 		cpu_to_le64(ts.tv_sec);
558 	fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
559 		cpu_to_le32(ts.tv_nsec);
560 	fe->i_dtime = 0;
561 
562 	/*
563 	 * If supported, directories start with inline data. If inline
564 	 * isn't supported, but indexing is, we start them as indexed.
565 	 */
566 	feat = le16_to_cpu(fe->i_dyn_features);
567 	if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
568 		fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
569 
570 		fe->id2.i_data.id_count = cpu_to_le16(
571 				ocfs2_max_inline_data_with_xattr(osb->sb, fe));
572 	} else {
573 		fel = &fe->id2.i_list;
574 		fel->l_tree_depth = 0;
575 		fel->l_next_free_rec = 0;
576 		fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
577 	}
578 
579 	ocfs2_journal_dirty(handle, *new_fe_bh);
580 
581 	ocfs2_populate_inode(inode, fe, 1);
582 	ocfs2_ci_set_new(osb, INODE_CACHE(inode));
583 	if (!ocfs2_mount_local(osb)) {
584 		status = ocfs2_create_new_inode_locks(inode);
585 		if (status < 0)
586 			mlog_errno(status);
587 	}
588 
589 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
590 
591 leave:
592 	if (status < 0) {
593 		if (*new_fe_bh) {
594 			brelse(*new_fe_bh);
595 			*new_fe_bh = NULL;
596 		}
597 	}
598 
599 	if (status)
600 		mlog_errno(status);
601 	return status;
602 }
603 
604 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
605 			      struct inode *dir,
606 			      struct inode *inode,
607 			      dev_t dev,
608 			      struct buffer_head **new_fe_bh,
609 			      struct buffer_head *parent_fe_bh,
610 			      handle_t *handle,
611 			      struct ocfs2_alloc_context *inode_ac)
612 {
613 	int status = 0;
614 	u64 suballoc_loc, fe_blkno = 0;
615 	u16 suballoc_bit;
616 
617 	*new_fe_bh = NULL;
618 
619 	status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
620 				       inode_ac, &suballoc_loc,
621 				       &suballoc_bit, &fe_blkno);
622 	if (status < 0) {
623 		mlog_errno(status);
624 		return status;
625 	}
626 
627 	status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
628 				    parent_fe_bh, handle, inode_ac,
629 				    fe_blkno, suballoc_loc, suballoc_bit);
630 	if (status < 0) {
631 		u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
632 		int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
633 				inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
634 		if (tmp)
635 			mlog_errno(tmp);
636 	}
637 
638 	return status;
639 }
640 
641 static int ocfs2_mkdir(struct inode *dir,
642 		       struct dentry *dentry,
643 		       umode_t mode)
644 {
645 	int ret;
646 
647 	trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
648 			  OCFS2_I(dir)->ip_blkno, mode);
649 	ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
650 	if (ret)
651 		mlog_errno(ret);
652 
653 	return ret;
654 }
655 
656 static int ocfs2_create(struct inode *dir,
657 			struct dentry *dentry,
658 			umode_t mode,
659 			bool excl)
660 {
661 	int ret;
662 
663 	trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
664 			   (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
665 	ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
666 	if (ret)
667 		mlog_errno(ret);
668 
669 	return ret;
670 }
671 
672 static int ocfs2_link(struct dentry *old_dentry,
673 		      struct inode *dir,
674 		      struct dentry *dentry)
675 {
676 	handle_t *handle;
677 	struct inode *inode = d_inode(old_dentry);
678 	struct inode *old_dir = d_inode(old_dentry->d_parent);
679 	int err;
680 	struct buffer_head *fe_bh = NULL;
681 	struct buffer_head *old_dir_bh = NULL;
682 	struct buffer_head *parent_fe_bh = NULL;
683 	struct ocfs2_dinode *fe = NULL;
684 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
685 	struct ocfs2_dir_lookup_result lookup = { NULL, };
686 	sigset_t oldset;
687 	u64 old_de_ino;
688 
689 	trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
690 			 old_dentry->d_name.len, old_dentry->d_name.name,
691 			 dentry->d_name.len, dentry->d_name.name);
692 
693 	if (S_ISDIR(inode->i_mode))
694 		return -EPERM;
695 
696 	err = dquot_initialize(dir);
697 	if (err) {
698 		mlog_errno(err);
699 		return err;
700 	}
701 
702 	err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
703 			&parent_fe_bh, dir, 0);
704 	if (err < 0) {
705 		if (err != -ENOENT)
706 			mlog_errno(err);
707 		return err;
708 	}
709 
710 	/* make sure both dirs have bhs
711 	 * get an extra ref on old_dir_bh if old==new */
712 	if (!parent_fe_bh) {
713 		if (old_dir_bh) {
714 			parent_fe_bh = old_dir_bh;
715 			get_bh(parent_fe_bh);
716 		} else {
717 			mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
718 			err = -EIO;
719 			goto out;
720 		}
721 	}
722 
723 	if (!dir->i_nlink) {
724 		err = -ENOENT;
725 		goto out;
726 	}
727 
728 	err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
729 			old_dentry->d_name.len, &old_de_ino);
730 	if (err) {
731 		err = -ENOENT;
732 		goto out;
733 	}
734 
735 	/*
736 	 * Check whether another node removed the source inode while we
737 	 * were in the vfs.
738 	 */
739 	if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
740 		err = -ENOENT;
741 		goto out;
742 	}
743 
744 	err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
745 					dentry->d_name.len);
746 	if (err)
747 		goto out;
748 
749 	err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
750 					   dentry->d_name.name,
751 					   dentry->d_name.len, &lookup);
752 	if (err < 0) {
753 		mlog_errno(err);
754 		goto out;
755 	}
756 
757 	err = ocfs2_inode_lock(inode, &fe_bh, 1);
758 	if (err < 0) {
759 		if (err != -ENOENT)
760 			mlog_errno(err);
761 		goto out;
762 	}
763 
764 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
765 	if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
766 		err = -EMLINK;
767 		goto out_unlock_inode;
768 	}
769 
770 	handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
771 	if (IS_ERR(handle)) {
772 		err = PTR_ERR(handle);
773 		handle = NULL;
774 		mlog_errno(err);
775 		goto out_unlock_inode;
776 	}
777 
778 	/* Starting to change things, restart is no longer possible. */
779 	ocfs2_block_signals(&oldset);
780 
781 	err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
782 				      OCFS2_JOURNAL_ACCESS_WRITE);
783 	if (err < 0) {
784 		mlog_errno(err);
785 		goto out_commit;
786 	}
787 
788 	inc_nlink(inode);
789 	inode->i_ctime = current_time(inode);
790 	ocfs2_set_links_count(fe, inode->i_nlink);
791 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
792 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
793 	ocfs2_journal_dirty(handle, fe_bh);
794 
795 	err = ocfs2_add_entry(handle, dentry, inode,
796 			      OCFS2_I(inode)->ip_blkno,
797 			      parent_fe_bh, &lookup);
798 	if (err) {
799 		ocfs2_add_links_count(fe, -1);
800 		drop_nlink(inode);
801 		mlog_errno(err);
802 		goto out_commit;
803 	}
804 
805 	err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
806 	if (err) {
807 		mlog_errno(err);
808 		goto out_commit;
809 	}
810 
811 	ihold(inode);
812 	d_instantiate(dentry, inode);
813 
814 out_commit:
815 	ocfs2_commit_trans(osb, handle);
816 	ocfs2_unblock_signals(&oldset);
817 out_unlock_inode:
818 	ocfs2_inode_unlock(inode, 1);
819 
820 out:
821 	ocfs2_double_unlock(old_dir, dir);
822 
823 	brelse(fe_bh);
824 	brelse(parent_fe_bh);
825 	brelse(old_dir_bh);
826 
827 	ocfs2_free_dir_lookup_result(&lookup);
828 
829 	if (err)
830 		mlog_errno(err);
831 
832 	return err;
833 }
834 
835 /*
836  * Takes and drops an exclusive lock on the given dentry. This will
837  * force other nodes to drop it.
838  */
839 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
840 {
841 	int ret;
842 
843 	ret = ocfs2_dentry_lock(dentry, 1);
844 	if (ret)
845 		mlog_errno(ret);
846 	else
847 		ocfs2_dentry_unlock(dentry, 1);
848 
849 	return ret;
850 }
851 
852 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
853 {
854 	if (S_ISDIR(inode->i_mode)) {
855 		if (inode->i_nlink == 2)
856 			return 1;
857 		return 0;
858 	}
859 
860 	if (inode->i_nlink == 1)
861 		return 1;
862 	return 0;
863 }
864 
865 static int ocfs2_unlink(struct inode *dir,
866 			struct dentry *dentry)
867 {
868 	int status;
869 	int child_locked = 0;
870 	bool is_unlinkable = false;
871 	struct inode *inode = d_inode(dentry);
872 	struct inode *orphan_dir = NULL;
873 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
874 	u64 blkno;
875 	struct ocfs2_dinode *fe = NULL;
876 	struct buffer_head *fe_bh = NULL;
877 	struct buffer_head *parent_node_bh = NULL;
878 	handle_t *handle = NULL;
879 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
880 	struct ocfs2_dir_lookup_result lookup = { NULL, };
881 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
882 
883 	trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
884 			   dentry->d_name.name,
885 			   (unsigned long long)OCFS2_I(dir)->ip_blkno,
886 			   (unsigned long long)OCFS2_I(inode)->ip_blkno);
887 
888 	status = dquot_initialize(dir);
889 	if (status) {
890 		mlog_errno(status);
891 		return status;
892 	}
893 
894 	BUG_ON(d_inode(dentry->d_parent) != dir);
895 
896 	if (inode == osb->root_inode)
897 		return -EPERM;
898 
899 	status = ocfs2_inode_lock_nested(dir, &parent_node_bh, 1,
900 					 OI_LS_PARENT);
901 	if (status < 0) {
902 		if (status != -ENOENT)
903 			mlog_errno(status);
904 		return status;
905 	}
906 
907 	status = ocfs2_find_files_on_disk(dentry->d_name.name,
908 					  dentry->d_name.len, &blkno, dir,
909 					  &lookup);
910 	if (status < 0) {
911 		if (status != -ENOENT)
912 			mlog_errno(status);
913 		goto leave;
914 	}
915 
916 	if (OCFS2_I(inode)->ip_blkno != blkno) {
917 		status = -ENOENT;
918 
919 		trace_ocfs2_unlink_noent(
920 				(unsigned long long)OCFS2_I(inode)->ip_blkno,
921 				(unsigned long long)blkno,
922 				OCFS2_I(inode)->ip_flags);
923 		goto leave;
924 	}
925 
926 	status = ocfs2_inode_lock(inode, &fe_bh, 1);
927 	if (status < 0) {
928 		if (status != -ENOENT)
929 			mlog_errno(status);
930 		goto leave;
931 	}
932 	child_locked = 1;
933 
934 	if (S_ISDIR(inode->i_mode)) {
935 		if (inode->i_nlink != 2 || !ocfs2_empty_dir(inode)) {
936 			status = -ENOTEMPTY;
937 			goto leave;
938 		}
939 	}
940 
941 	status = ocfs2_remote_dentry_delete(dentry);
942 	if (status < 0) {
943 		/* This remote delete should succeed under all normal
944 		 * circumstances. */
945 		mlog_errno(status);
946 		goto leave;
947 	}
948 
949 	if (ocfs2_inode_is_unlinkable(inode)) {
950 		status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
951 						  OCFS2_I(inode)->ip_blkno,
952 						  orphan_name, &orphan_insert,
953 						  false);
954 		if (status < 0) {
955 			mlog_errno(status);
956 			goto leave;
957 		}
958 		is_unlinkable = true;
959 	}
960 
961 	handle = ocfs2_start_trans(osb, ocfs2_unlink_credits(osb->sb));
962 	if (IS_ERR(handle)) {
963 		status = PTR_ERR(handle);
964 		handle = NULL;
965 		mlog_errno(status);
966 		goto leave;
967 	}
968 
969 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
970 					 OCFS2_JOURNAL_ACCESS_WRITE);
971 	if (status < 0) {
972 		mlog_errno(status);
973 		goto leave;
974 	}
975 
976 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
977 
978 	/* delete the name from the parent dir */
979 	status = ocfs2_delete_entry(handle, dir, &lookup);
980 	if (status < 0) {
981 		mlog_errno(status);
982 		goto leave;
983 	}
984 
985 	if (S_ISDIR(inode->i_mode))
986 		drop_nlink(inode);
987 	drop_nlink(inode);
988 	ocfs2_set_links_count(fe, inode->i_nlink);
989 	ocfs2_journal_dirty(handle, fe_bh);
990 
991 	dir->i_ctime = dir->i_mtime = current_time(dir);
992 	if (S_ISDIR(inode->i_mode))
993 		drop_nlink(dir);
994 
995 	status = ocfs2_mark_inode_dirty(handle, dir, parent_node_bh);
996 	if (status < 0) {
997 		mlog_errno(status);
998 		if (S_ISDIR(inode->i_mode))
999 			inc_nlink(dir);
1000 		goto leave;
1001 	}
1002 
1003 	if (is_unlinkable) {
1004 		status = ocfs2_orphan_add(osb, handle, inode, fe_bh,
1005 				orphan_name, &orphan_insert, orphan_dir, false);
1006 		if (status < 0)
1007 			mlog_errno(status);
1008 	}
1009 
1010 leave:
1011 	if (handle)
1012 		ocfs2_commit_trans(osb, handle);
1013 
1014 	if (orphan_dir) {
1015 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
1016 		ocfs2_inode_unlock(orphan_dir, 1);
1017 		inode_unlock(orphan_dir);
1018 		iput(orphan_dir);
1019 	}
1020 
1021 	if (child_locked)
1022 		ocfs2_inode_unlock(inode, 1);
1023 
1024 	ocfs2_inode_unlock(dir, 1);
1025 
1026 	brelse(fe_bh);
1027 	brelse(parent_node_bh);
1028 
1029 	ocfs2_free_dir_lookup_result(&orphan_insert);
1030 	ocfs2_free_dir_lookup_result(&lookup);
1031 
1032 	if (status && (status != -ENOTEMPTY) && (status != -ENOENT))
1033 		mlog_errno(status);
1034 
1035 	return status;
1036 }
1037 
1038 static int ocfs2_check_if_ancestor(struct ocfs2_super *osb,
1039 		u64 src_inode_no, u64 dest_inode_no)
1040 {
1041 	int ret = 0, i = 0;
1042 	u64 parent_inode_no = 0;
1043 	u64 child_inode_no = src_inode_no;
1044 	struct inode *child_inode;
1045 
1046 #define MAX_LOOKUP_TIMES 32
1047 	while (1) {
1048 		child_inode = ocfs2_iget(osb, child_inode_no, 0, 0);
1049 		if (IS_ERR(child_inode)) {
1050 			ret = PTR_ERR(child_inode);
1051 			break;
1052 		}
1053 
1054 		ret = ocfs2_inode_lock(child_inode, NULL, 0);
1055 		if (ret < 0) {
1056 			iput(child_inode);
1057 			if (ret != -ENOENT)
1058 				mlog_errno(ret);
1059 			break;
1060 		}
1061 
1062 		ret = ocfs2_lookup_ino_from_name(child_inode, "..", 2,
1063 				&parent_inode_no);
1064 		ocfs2_inode_unlock(child_inode, 0);
1065 		iput(child_inode);
1066 		if (ret < 0) {
1067 			ret = -ENOENT;
1068 			break;
1069 		}
1070 
1071 		if (parent_inode_no == dest_inode_no) {
1072 			ret = 1;
1073 			break;
1074 		}
1075 
1076 		if (parent_inode_no == osb->root_inode->i_ino) {
1077 			ret = 0;
1078 			break;
1079 		}
1080 
1081 		child_inode_no = parent_inode_no;
1082 
1083 		if (++i >= MAX_LOOKUP_TIMES) {
1084 			mlog(ML_NOTICE, "max lookup times reached, filesystem "
1085 					"may have nested directories, "
1086 					"src inode: %llu, dest inode: %llu.\n",
1087 					(unsigned long long)src_inode_no,
1088 					(unsigned long long)dest_inode_no);
1089 			ret = 0;
1090 			break;
1091 		}
1092 	}
1093 
1094 	return ret;
1095 }
1096 
1097 /*
1098  * The only place this should be used is rename and link!
1099  * if they have the same id, then the 1st one is the only one locked.
1100  */
1101 static int ocfs2_double_lock(struct ocfs2_super *osb,
1102 			     struct buffer_head **bh1,
1103 			     struct inode *inode1,
1104 			     struct buffer_head **bh2,
1105 			     struct inode *inode2,
1106 			     int rename)
1107 {
1108 	int status;
1109 	int inode1_is_ancestor, inode2_is_ancestor;
1110 	struct ocfs2_inode_info *oi1 = OCFS2_I(inode1);
1111 	struct ocfs2_inode_info *oi2 = OCFS2_I(inode2);
1112 
1113 	trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1114 				(unsigned long long)oi2->ip_blkno);
1115 
1116 	if (*bh1)
1117 		*bh1 = NULL;
1118 	if (*bh2)
1119 		*bh2 = NULL;
1120 
1121 	/* we always want to lock the one with the lower lockid first.
1122 	 * and if they are nested, we lock ancestor first */
1123 	if (oi1->ip_blkno != oi2->ip_blkno) {
1124 		inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1125 				oi1->ip_blkno);
1126 		if (inode1_is_ancestor < 0) {
1127 			status = inode1_is_ancestor;
1128 			goto bail;
1129 		}
1130 
1131 		inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1132 				oi2->ip_blkno);
1133 		if (inode2_is_ancestor < 0) {
1134 			status = inode2_is_ancestor;
1135 			goto bail;
1136 		}
1137 
1138 		if ((inode1_is_ancestor == 1) ||
1139 				(oi1->ip_blkno < oi2->ip_blkno &&
1140 				inode2_is_ancestor == 0)) {
1141 			/* switch id1 and id2 around */
1142 			swap(bh2, bh1);
1143 			swap(inode2, inode1);
1144 		}
1145 		/* lock id2 */
1146 		status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1147 				rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1148 		if (status < 0) {
1149 			if (status != -ENOENT)
1150 				mlog_errno(status);
1151 			goto bail;
1152 		}
1153 	}
1154 
1155 	/* lock id1 */
1156 	status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1157 			rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1158 	if (status < 0) {
1159 		/*
1160 		 * An error return must mean that no cluster locks
1161 		 * were held on function exit.
1162 		 */
1163 		if (oi1->ip_blkno != oi2->ip_blkno) {
1164 			ocfs2_inode_unlock(inode2, 1);
1165 			brelse(*bh2);
1166 			*bh2 = NULL;
1167 		}
1168 
1169 		if (status != -ENOENT)
1170 			mlog_errno(status);
1171 	}
1172 
1173 	trace_ocfs2_double_lock_end(
1174 			(unsigned long long)oi1->ip_blkno,
1175 			(unsigned long long)oi2->ip_blkno);
1176 
1177 bail:
1178 	if (status)
1179 		mlog_errno(status);
1180 	return status;
1181 }
1182 
1183 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1184 {
1185 	ocfs2_inode_unlock(inode1, 1);
1186 
1187 	if (inode1 != inode2)
1188 		ocfs2_inode_unlock(inode2, 1);
1189 }
1190 
1191 static int ocfs2_rename(struct inode *old_dir,
1192 			struct dentry *old_dentry,
1193 			struct inode *new_dir,
1194 			struct dentry *new_dentry,
1195 			unsigned int flags)
1196 {
1197 	int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1198 	int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1199 	struct inode *old_inode = d_inode(old_dentry);
1200 	struct inode *new_inode = d_inode(new_dentry);
1201 	struct inode *orphan_dir = NULL;
1202 	struct ocfs2_dinode *newfe = NULL;
1203 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1204 	struct buffer_head *newfe_bh = NULL;
1205 	struct buffer_head *old_inode_bh = NULL;
1206 	struct ocfs2_super *osb = NULL;
1207 	u64 newfe_blkno, old_de_ino;
1208 	handle_t *handle = NULL;
1209 	struct buffer_head *old_dir_bh = NULL;
1210 	struct buffer_head *new_dir_bh = NULL;
1211 	u32 old_dir_nlink = old_dir->i_nlink;
1212 	struct ocfs2_dinode *old_di;
1213 	struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1214 	struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1215 	struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1216 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1217 	struct ocfs2_dir_lookup_result target_insert = { NULL, };
1218 	bool should_add_orphan = false;
1219 
1220 	if (flags)
1221 		return -EINVAL;
1222 
1223 	/* At some point it might be nice to break this function up a
1224 	 * bit. */
1225 
1226 	trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1227 			   old_dentry->d_name.len, old_dentry->d_name.name,
1228 			   new_dentry->d_name.len, new_dentry->d_name.name);
1229 
1230 	status = dquot_initialize(old_dir);
1231 	if (status) {
1232 		mlog_errno(status);
1233 		goto bail;
1234 	}
1235 	status = dquot_initialize(new_dir);
1236 	if (status) {
1237 		mlog_errno(status);
1238 		goto bail;
1239 	}
1240 
1241 	osb = OCFS2_SB(old_dir->i_sb);
1242 
1243 	if (new_inode) {
1244 		if (!igrab(new_inode))
1245 			BUG();
1246 	}
1247 
1248 	/* Assume a directory hierarchy thusly:
1249 	 * a/b/c
1250 	 * a/d
1251 	 * a,b,c, and d are all directories.
1252 	 *
1253 	 * from cwd of 'a' on both nodes:
1254 	 * node1: mv b/c d
1255 	 * node2: mv d   b/c
1256 	 *
1257 	 * And that's why, just like the VFS, we need a file system
1258 	 * rename lock. */
1259 	if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1260 		status = ocfs2_rename_lock(osb);
1261 		if (status < 0) {
1262 			mlog_errno(status);
1263 			goto bail;
1264 		}
1265 		rename_lock = 1;
1266 
1267 		/* here we cannot guarantee the inodes haven't just been
1268 		 * changed, so check if they are nested again */
1269 		status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1270 				old_inode->i_ino);
1271 		if (status < 0) {
1272 			mlog_errno(status);
1273 			goto bail;
1274 		} else if (status == 1) {
1275 			status = -EPERM;
1276 			trace_ocfs2_rename_not_permitted(
1277 					(unsigned long long)old_inode->i_ino,
1278 					(unsigned long long)new_dir->i_ino);
1279 			goto bail;
1280 		}
1281 	}
1282 
1283 	/* if old and new are the same, this'll just do one lock. */
1284 	status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1285 				   &new_dir_bh, new_dir, 1);
1286 	if (status < 0) {
1287 		mlog_errno(status);
1288 		goto bail;
1289 	}
1290 	parents_locked = 1;
1291 
1292 	if (!new_dir->i_nlink) {
1293 		status = -EACCES;
1294 		goto bail;
1295 	}
1296 
1297 	/* make sure both dirs have bhs
1298 	 * get an extra ref on old_dir_bh if old==new */
1299 	if (!new_dir_bh) {
1300 		if (old_dir_bh) {
1301 			new_dir_bh = old_dir_bh;
1302 			get_bh(new_dir_bh);
1303 		} else {
1304 			mlog(ML_ERROR, "no old_dir_bh!\n");
1305 			status = -EIO;
1306 			goto bail;
1307 		}
1308 	}
1309 
1310 	/*
1311 	 * Aside from allowing a meta data update, the locking here
1312 	 * also ensures that the downconvert thread on other nodes
1313 	 * won't have to concurrently downconvert the inode and the
1314 	 * dentry locks.
1315 	 */
1316 	status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1317 					 OI_LS_PARENT);
1318 	if (status < 0) {
1319 		if (status != -ENOENT)
1320 			mlog_errno(status);
1321 		goto bail;
1322 	}
1323 	old_child_locked = 1;
1324 
1325 	status = ocfs2_remote_dentry_delete(old_dentry);
1326 	if (status < 0) {
1327 		mlog_errno(status);
1328 		goto bail;
1329 	}
1330 
1331 	if (S_ISDIR(old_inode->i_mode)) {
1332 		u64 old_inode_parent;
1333 
1334 		update_dot_dot = 1;
1335 		status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1336 						  old_inode,
1337 						  &old_inode_dot_dot_res);
1338 		if (status) {
1339 			status = -EIO;
1340 			goto bail;
1341 		}
1342 
1343 		if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1344 			status = -EIO;
1345 			goto bail;
1346 		}
1347 
1348 		if (!new_inode && new_dir != old_dir &&
1349 		    new_dir->i_nlink >= ocfs2_link_max(osb)) {
1350 			status = -EMLINK;
1351 			goto bail;
1352 		}
1353 	}
1354 
1355 	status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1356 					    old_dentry->d_name.len,
1357 					    &old_de_ino);
1358 	if (status) {
1359 		status = -ENOENT;
1360 		goto bail;
1361 	}
1362 
1363 	/*
1364 	 *  Check for inode number is _not_ due to possible IO errors.
1365 	 *  We might rmdir the source, keep it as pwd of some process
1366 	 *  and merrily kill the link to whatever was created under the
1367 	 *  same name. Goodbye sticky bit ;-<
1368 	 */
1369 	if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1370 		status = -ENOENT;
1371 		goto bail;
1372 	}
1373 
1374 	/* check if the target already exists (in which case we need
1375 	 * to delete it */
1376 	status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1377 					  new_dentry->d_name.len,
1378 					  &newfe_blkno, new_dir,
1379 					  &target_lookup_res);
1380 	/* The only error we allow here is -ENOENT because the new
1381 	 * file not existing is perfectly valid. */
1382 	if ((status < 0) && (status != -ENOENT)) {
1383 		/* If we cannot find the file specified we should just */
1384 		/* return the error... */
1385 		mlog_errno(status);
1386 		goto bail;
1387 	}
1388 	if (status == 0)
1389 		target_exists = 1;
1390 
1391 	if (!target_exists && new_inode) {
1392 		/*
1393 		 * Target was unlinked by another node while we were
1394 		 * waiting to get to ocfs2_rename(). There isn't
1395 		 * anything we can do here to help the situation, so
1396 		 * bubble up the appropriate error.
1397 		 */
1398 		status = -ENOENT;
1399 		goto bail;
1400 	}
1401 
1402 	/* In case we need to overwrite an existing file, we blow it
1403 	 * away first */
1404 	if (target_exists) {
1405 		/* VFS didn't think there existed an inode here, but
1406 		 * someone else in the cluster must have raced our
1407 		 * rename to create one. Today we error cleanly, in
1408 		 * the future we should consider calling iget to build
1409 		 * a new struct inode for this entry. */
1410 		if (!new_inode) {
1411 			status = -EACCES;
1412 
1413 			trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1414 						new_dentry->d_name.name);
1415 			goto bail;
1416 		}
1417 
1418 		if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1419 			status = -EACCES;
1420 
1421 			trace_ocfs2_rename_disagree(
1422 			     (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1423 			     (unsigned long long)newfe_blkno,
1424 			     OCFS2_I(new_inode)->ip_flags);
1425 			goto bail;
1426 		}
1427 
1428 		status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1429 		if (status < 0) {
1430 			if (status != -ENOENT)
1431 				mlog_errno(status);
1432 			goto bail;
1433 		}
1434 		new_child_locked = 1;
1435 
1436 		status = ocfs2_remote_dentry_delete(new_dentry);
1437 		if (status < 0) {
1438 			mlog_errno(status);
1439 			goto bail;
1440 		}
1441 
1442 		newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1443 
1444 		trace_ocfs2_rename_over_existing(
1445 		     (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1446 		     (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1447 
1448 		if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1449 			status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1450 						OCFS2_I(new_inode)->ip_blkno,
1451 						orphan_name, &orphan_insert,
1452 						false);
1453 			if (status < 0) {
1454 				mlog_errno(status);
1455 				goto bail;
1456 			}
1457 			should_add_orphan = true;
1458 		}
1459 	} else {
1460 		BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1461 
1462 		status = ocfs2_check_dir_for_entry(new_dir,
1463 						   new_dentry->d_name.name,
1464 						   new_dentry->d_name.len);
1465 		if (status)
1466 			goto bail;
1467 
1468 		status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1469 						      new_dentry->d_name.name,
1470 						      new_dentry->d_name.len,
1471 						      &target_insert);
1472 		if (status < 0) {
1473 			mlog_errno(status);
1474 			goto bail;
1475 		}
1476 	}
1477 
1478 	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1479 	if (IS_ERR(handle)) {
1480 		status = PTR_ERR(handle);
1481 		handle = NULL;
1482 		mlog_errno(status);
1483 		goto bail;
1484 	}
1485 
1486 	if (target_exists) {
1487 		if (S_ISDIR(new_inode->i_mode)) {
1488 			if (new_inode->i_nlink != 2 ||
1489 			    !ocfs2_empty_dir(new_inode)) {
1490 				status = -ENOTEMPTY;
1491 				goto bail;
1492 			}
1493 		}
1494 		status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1495 						 newfe_bh,
1496 						 OCFS2_JOURNAL_ACCESS_WRITE);
1497 		if (status < 0) {
1498 			mlog_errno(status);
1499 			goto bail;
1500 		}
1501 
1502 		/* change the dirent to point to the correct inode */
1503 		status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1504 					    old_inode);
1505 		if (status < 0) {
1506 			mlog_errno(status);
1507 			goto bail;
1508 		}
1509 		inode_inc_iversion(new_dir);
1510 
1511 		if (S_ISDIR(new_inode->i_mode))
1512 			ocfs2_set_links_count(newfe, 0);
1513 		else
1514 			ocfs2_add_links_count(newfe, -1);
1515 		ocfs2_journal_dirty(handle, newfe_bh);
1516 		if (should_add_orphan) {
1517 			status = ocfs2_orphan_add(osb, handle, new_inode,
1518 					newfe_bh, orphan_name,
1519 					&orphan_insert, orphan_dir, false);
1520 			if (status < 0) {
1521 				mlog_errno(status);
1522 				goto bail;
1523 			}
1524 		}
1525 	} else {
1526 		/* if the name was not found in new_dir, add it now */
1527 		status = ocfs2_add_entry(handle, new_dentry, old_inode,
1528 					 OCFS2_I(old_inode)->ip_blkno,
1529 					 new_dir_bh, &target_insert);
1530 	}
1531 
1532 	old_inode->i_ctime = current_time(old_inode);
1533 	mark_inode_dirty(old_inode);
1534 
1535 	status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1536 					 old_inode_bh,
1537 					 OCFS2_JOURNAL_ACCESS_WRITE);
1538 	if (status >= 0) {
1539 		old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1540 
1541 		old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1542 		old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1543 		ocfs2_journal_dirty(handle, old_inode_bh);
1544 	} else
1545 		mlog_errno(status);
1546 
1547 	/*
1548 	 * Now that the name has been added to new_dir, remove the old name.
1549 	 *
1550 	 * We don't keep any directory entry context around until now
1551 	 * because the insert might have changed the type of directory
1552 	 * we're dealing with.
1553 	 */
1554 	status = ocfs2_find_entry(old_dentry->d_name.name,
1555 				  old_dentry->d_name.len, old_dir,
1556 				  &old_entry_lookup);
1557 	if (status) {
1558 		if (!is_journal_aborted(osb->journal->j_journal)) {
1559 			ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1560 					"is not deleted.",
1561 					new_dentry->d_name.len, new_dentry->d_name.name,
1562 					old_dentry->d_name.len, old_dentry->d_name.name);
1563 		}
1564 		goto bail;
1565 	}
1566 
1567 	status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1568 	if (status < 0) {
1569 		mlog_errno(status);
1570 		if (!is_journal_aborted(osb->journal->j_journal)) {
1571 			ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1572 					"is not deleted.",
1573 					new_dentry->d_name.len, new_dentry->d_name.name,
1574 					old_dentry->d_name.len, old_dentry->d_name.name);
1575 		}
1576 		goto bail;
1577 	}
1578 
1579 	if (new_inode) {
1580 		drop_nlink(new_inode);
1581 		new_inode->i_ctime = current_time(new_inode);
1582 	}
1583 	old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1584 
1585 	if (update_dot_dot) {
1586 		status = ocfs2_update_entry(old_inode, handle,
1587 					    &old_inode_dot_dot_res, new_dir);
1588 		drop_nlink(old_dir);
1589 		if (new_inode) {
1590 			drop_nlink(new_inode);
1591 		} else {
1592 			inc_nlink(new_dir);
1593 			mark_inode_dirty(new_dir);
1594 		}
1595 	}
1596 	mark_inode_dirty(old_dir);
1597 	ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1598 	if (new_inode) {
1599 		mark_inode_dirty(new_inode);
1600 		ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1601 	}
1602 
1603 	if (old_dir != new_dir) {
1604 		/* Keep the same times on both directories.*/
1605 		new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1606 
1607 		/*
1608 		 * This will also pick up the i_nlink change from the
1609 		 * block above.
1610 		 */
1611 		ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1612 	}
1613 
1614 	if (old_dir_nlink != old_dir->i_nlink) {
1615 		if (!old_dir_bh) {
1616 			mlog(ML_ERROR, "need to change nlink for old dir "
1617 			     "%llu from %d to %d but bh is NULL!\n",
1618 			     (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1619 			     (int)old_dir_nlink, old_dir->i_nlink);
1620 		} else {
1621 			struct ocfs2_dinode *fe;
1622 			status = ocfs2_journal_access_di(handle,
1623 							 INODE_CACHE(old_dir),
1624 							 old_dir_bh,
1625 							 OCFS2_JOURNAL_ACCESS_WRITE);
1626 			fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1627 			ocfs2_set_links_count(fe, old_dir->i_nlink);
1628 			ocfs2_journal_dirty(handle, old_dir_bh);
1629 		}
1630 	}
1631 	ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1632 	status = 0;
1633 bail:
1634 	if (handle)
1635 		ocfs2_commit_trans(osb, handle);
1636 
1637 	if (orphan_dir) {
1638 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
1639 		ocfs2_inode_unlock(orphan_dir, 1);
1640 		inode_unlock(orphan_dir);
1641 		iput(orphan_dir);
1642 	}
1643 
1644 	if (new_child_locked)
1645 		ocfs2_inode_unlock(new_inode, 1);
1646 
1647 	if (old_child_locked)
1648 		ocfs2_inode_unlock(old_inode, 1);
1649 
1650 	if (parents_locked)
1651 		ocfs2_double_unlock(old_dir, new_dir);
1652 
1653 	if (rename_lock)
1654 		ocfs2_rename_unlock(osb);
1655 
1656 	if (new_inode)
1657 		sync_mapping_buffers(old_inode->i_mapping);
1658 
1659 	iput(new_inode);
1660 
1661 	ocfs2_free_dir_lookup_result(&target_lookup_res);
1662 	ocfs2_free_dir_lookup_result(&old_entry_lookup);
1663 	ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1664 	ocfs2_free_dir_lookup_result(&orphan_insert);
1665 	ocfs2_free_dir_lookup_result(&target_insert);
1666 
1667 	brelse(newfe_bh);
1668 	brelse(old_inode_bh);
1669 	brelse(old_dir_bh);
1670 	brelse(new_dir_bh);
1671 
1672 	if (status)
1673 		mlog_errno(status);
1674 
1675 	return status;
1676 }
1677 
1678 /*
1679  * we expect i_size = strlen(symname). Copy symname into the file
1680  * data, including the null terminator.
1681  */
1682 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1683 				     handle_t *handle,
1684 				     struct inode *inode,
1685 				     const char *symname)
1686 {
1687 	struct buffer_head **bhs = NULL;
1688 	const char *c;
1689 	struct super_block *sb = osb->sb;
1690 	u64 p_blkno, p_blocks;
1691 	int virtual, blocks, status, i, bytes_left;
1692 
1693 	bytes_left = i_size_read(inode) + 1;
1694 	/* we can't trust i_blocks because we're actually going to
1695 	 * write i_size + 1 bytes. */
1696 	blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1697 
1698 	trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1699 					i_size_read(inode), blocks);
1700 
1701 	/* Sanity check -- make sure we're going to fit. */
1702 	if (bytes_left >
1703 	    ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1704 		status = -EIO;
1705 		mlog_errno(status);
1706 		goto bail;
1707 	}
1708 
1709 	bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1710 	if (!bhs) {
1711 		status = -ENOMEM;
1712 		mlog_errno(status);
1713 		goto bail;
1714 	}
1715 
1716 	status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1717 					     NULL);
1718 	if (status < 0) {
1719 		mlog_errno(status);
1720 		goto bail;
1721 	}
1722 
1723 	/* links can never be larger than one cluster so we know this
1724 	 * is all going to be contiguous, but do a sanity check
1725 	 * anyway. */
1726 	if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1727 		status = -EIO;
1728 		mlog_errno(status);
1729 		goto bail;
1730 	}
1731 
1732 	virtual = 0;
1733 	while(bytes_left > 0) {
1734 		c = &symname[virtual * sb->s_blocksize];
1735 
1736 		bhs[virtual] = sb_getblk(sb, p_blkno);
1737 		if (!bhs[virtual]) {
1738 			status = -ENOMEM;
1739 			mlog_errno(status);
1740 			goto bail;
1741 		}
1742 		ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1743 					      bhs[virtual]);
1744 
1745 		status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1746 					      bhs[virtual],
1747 					      OCFS2_JOURNAL_ACCESS_CREATE);
1748 		if (status < 0) {
1749 			mlog_errno(status);
1750 			goto bail;
1751 		}
1752 
1753 		memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1754 
1755 		memcpy(bhs[virtual]->b_data, c,
1756 		       (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1757 		       bytes_left);
1758 
1759 		ocfs2_journal_dirty(handle, bhs[virtual]);
1760 
1761 		virtual++;
1762 		p_blkno++;
1763 		bytes_left -= sb->s_blocksize;
1764 	}
1765 
1766 	status = 0;
1767 bail:
1768 
1769 	if (bhs) {
1770 		for(i = 0; i < blocks; i++)
1771 			brelse(bhs[i]);
1772 		kfree(bhs);
1773 	}
1774 
1775 	if (status)
1776 		mlog_errno(status);
1777 	return status;
1778 }
1779 
1780 static int ocfs2_symlink(struct inode *dir,
1781 			 struct dentry *dentry,
1782 			 const char *symname)
1783 {
1784 	int status, l, credits;
1785 	u64 newsize;
1786 	struct ocfs2_super *osb = NULL;
1787 	struct inode *inode = NULL;
1788 	struct super_block *sb;
1789 	struct buffer_head *new_fe_bh = NULL;
1790 	struct buffer_head *parent_fe_bh = NULL;
1791 	struct ocfs2_dinode *fe = NULL;
1792 	struct ocfs2_dinode *dirfe;
1793 	handle_t *handle = NULL;
1794 	struct ocfs2_alloc_context *inode_ac = NULL;
1795 	struct ocfs2_alloc_context *data_ac = NULL;
1796 	struct ocfs2_alloc_context *xattr_ac = NULL;
1797 	int want_clusters = 0;
1798 	int xattr_credits = 0;
1799 	struct ocfs2_security_xattr_info si = {
1800 		.enable = 1,
1801 	};
1802 	int did_quota = 0, did_quota_inode = 0;
1803 	struct ocfs2_dir_lookup_result lookup = { NULL, };
1804 	sigset_t oldset;
1805 	int did_block_signals = 0;
1806 	struct ocfs2_dentry_lock *dl = NULL;
1807 
1808 	trace_ocfs2_symlink_begin(dir, dentry, symname,
1809 				  dentry->d_name.len, dentry->d_name.name);
1810 
1811 	status = dquot_initialize(dir);
1812 	if (status) {
1813 		mlog_errno(status);
1814 		goto bail;
1815 	}
1816 
1817 	sb = dir->i_sb;
1818 	osb = OCFS2_SB(sb);
1819 
1820 	l = strlen(symname) + 1;
1821 
1822 	credits = ocfs2_calc_symlink_credits(sb);
1823 
1824 	/* lock the parent directory */
1825 	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1826 	if (status < 0) {
1827 		if (status != -ENOENT)
1828 			mlog_errno(status);
1829 		return status;
1830 	}
1831 
1832 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1833 	if (!ocfs2_read_links_count(dirfe)) {
1834 		/* can't make a file in a deleted directory. */
1835 		status = -ENOENT;
1836 		goto bail;
1837 	}
1838 
1839 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1840 					   dentry->d_name.len);
1841 	if (status)
1842 		goto bail;
1843 
1844 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1845 					      dentry->d_name.name,
1846 					      dentry->d_name.len, &lookup);
1847 	if (status < 0) {
1848 		mlog_errno(status);
1849 		goto bail;
1850 	}
1851 
1852 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
1853 	if (status < 0) {
1854 		if (status != -ENOSPC)
1855 			mlog_errno(status);
1856 		goto bail;
1857 	}
1858 
1859 	inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1860 	if (IS_ERR(inode)) {
1861 		status = PTR_ERR(inode);
1862 		inode = NULL;
1863 		mlog_errno(status);
1864 		goto bail;
1865 	}
1866 
1867 	/* get security xattr */
1868 	status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1869 	if (status) {
1870 		if (status == -EOPNOTSUPP)
1871 			si.enable = 0;
1872 		else {
1873 			mlog_errno(status);
1874 			goto bail;
1875 		}
1876 	}
1877 
1878 	/* calculate meta data/clusters for setting security xattr */
1879 	if (si.enable) {
1880 		status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1881 						  &xattr_credits, &xattr_ac);
1882 		if (status < 0) {
1883 			mlog_errno(status);
1884 			goto bail;
1885 		}
1886 	}
1887 
1888 	/* don't reserve bitmap space for fast symlinks. */
1889 	if (l > ocfs2_fast_symlink_chars(sb))
1890 		want_clusters += 1;
1891 
1892 	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1893 	if (status < 0) {
1894 		if (status != -ENOSPC)
1895 			mlog_errno(status);
1896 		goto bail;
1897 	}
1898 
1899 	handle = ocfs2_start_trans(osb, credits + xattr_credits);
1900 	if (IS_ERR(handle)) {
1901 		status = PTR_ERR(handle);
1902 		handle = NULL;
1903 		mlog_errno(status);
1904 		goto bail;
1905 	}
1906 
1907 	/* Starting to change things, restart is no longer possible. */
1908 	ocfs2_block_signals(&oldset);
1909 	did_block_signals = 1;
1910 
1911 	status = dquot_alloc_inode(inode);
1912 	if (status)
1913 		goto bail;
1914 	did_quota_inode = 1;
1915 
1916 	trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1917 				   dentry->d_name.name,
1918 				   (unsigned long long)OCFS2_I(dir)->ip_blkno,
1919 				   inode->i_mode);
1920 
1921 	status = ocfs2_mknod_locked(osb, dir, inode,
1922 				    0, &new_fe_bh, parent_fe_bh, handle,
1923 				    inode_ac);
1924 	if (status < 0) {
1925 		mlog_errno(status);
1926 		goto bail;
1927 	}
1928 
1929 	fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1930 	inode->i_rdev = 0;
1931 	newsize = l - 1;
1932 	inode->i_op = &ocfs2_symlink_inode_operations;
1933 	inode_nohighmem(inode);
1934 	if (l > ocfs2_fast_symlink_chars(sb)) {
1935 		u32 offset = 0;
1936 
1937 		status = dquot_alloc_space_nodirty(inode,
1938 		    ocfs2_clusters_to_bytes(osb->sb, 1));
1939 		if (status)
1940 			goto bail;
1941 		did_quota = 1;
1942 		inode->i_mapping->a_ops = &ocfs2_aops;
1943 		status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1944 					      new_fe_bh,
1945 					      handle, data_ac, NULL,
1946 					      NULL);
1947 		if (status < 0) {
1948 			if (status != -ENOSPC && status != -EINTR) {
1949 				mlog(ML_ERROR,
1950 				     "Failed to extend file to %llu\n",
1951 				     (unsigned long long)newsize);
1952 				mlog_errno(status);
1953 				status = -ENOSPC;
1954 			}
1955 			goto bail;
1956 		}
1957 		i_size_write(inode, newsize);
1958 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1959 	} else {
1960 		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1961 		memcpy((char *) fe->id2.i_symlink, symname, l);
1962 		i_size_write(inode, newsize);
1963 		inode->i_blocks = 0;
1964 	}
1965 
1966 	status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1967 	if (status < 0) {
1968 		mlog_errno(status);
1969 		goto bail;
1970 	}
1971 
1972 	if (!ocfs2_inode_is_fast_symlink(inode)) {
1973 		status = ocfs2_create_symlink_data(osb, handle, inode,
1974 						   symname);
1975 		if (status < 0) {
1976 			mlog_errno(status);
1977 			goto bail;
1978 		}
1979 	}
1980 
1981 	if (si.enable) {
1982 		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1983 						 xattr_ac, data_ac);
1984 		if (status < 0) {
1985 			mlog_errno(status);
1986 			goto bail;
1987 		}
1988 	}
1989 
1990 	/*
1991 	 * Do this before adding the entry to the directory. We add
1992 	 * also set d_op after success so that ->d_iput() will cleanup
1993 	 * the dentry lock even if ocfs2_add_entry() fails below.
1994 	 */
1995 	status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
1996 	if (status) {
1997 		mlog_errno(status);
1998 		goto bail;
1999 	}
2000 
2001 	dl = dentry->d_fsdata;
2002 
2003 	status = ocfs2_add_entry(handle, dentry, inode,
2004 				 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2005 				 &lookup);
2006 	if (status < 0) {
2007 		mlog_errno(status);
2008 		goto bail;
2009 	}
2010 
2011 	insert_inode_hash(inode);
2012 	d_instantiate(dentry, inode);
2013 bail:
2014 	if (status < 0 && did_quota)
2015 		dquot_free_space_nodirty(inode,
2016 					ocfs2_clusters_to_bytes(osb->sb, 1));
2017 	if (status < 0 && did_quota_inode)
2018 		dquot_free_inode(inode);
2019 	if (handle)
2020 		ocfs2_commit_trans(osb, handle);
2021 
2022 	ocfs2_inode_unlock(dir, 1);
2023 	if (did_block_signals)
2024 		ocfs2_unblock_signals(&oldset);
2025 
2026 	brelse(new_fe_bh);
2027 	brelse(parent_fe_bh);
2028 	kfree(si.value);
2029 	ocfs2_free_dir_lookup_result(&lookup);
2030 	if (inode_ac)
2031 		ocfs2_free_alloc_context(inode_ac);
2032 	if (data_ac)
2033 		ocfs2_free_alloc_context(data_ac);
2034 	if (xattr_ac)
2035 		ocfs2_free_alloc_context(xattr_ac);
2036 	if ((status < 0) && inode) {
2037 		if (dl)
2038 			ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2039 
2040 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2041 		clear_nlink(inode);
2042 		iput(inode);
2043 	}
2044 
2045 	if (status)
2046 		mlog_errno(status);
2047 
2048 	return status;
2049 }
2050 
2051 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2052 {
2053 	int status, namelen;
2054 
2055 	namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2056 			   (long long)blkno);
2057 	if (namelen <= 0) {
2058 		if (namelen)
2059 			status = namelen;
2060 		else
2061 			status = -EINVAL;
2062 		mlog_errno(status);
2063 		goto bail;
2064 	}
2065 	if (namelen != OCFS2_ORPHAN_NAMELEN) {
2066 		status = -EINVAL;
2067 		mlog_errno(status);
2068 		goto bail;
2069 	}
2070 
2071 	trace_ocfs2_blkno_stringify(blkno, name, namelen);
2072 
2073 	status = 0;
2074 bail:
2075 	if (status < 0)
2076 		mlog_errno(status);
2077 	return status;
2078 }
2079 
2080 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2081 					struct inode **ret_orphan_dir,
2082 					struct buffer_head **ret_orphan_dir_bh)
2083 {
2084 	struct inode *orphan_dir_inode;
2085 	struct buffer_head *orphan_dir_bh = NULL;
2086 	int ret = 0;
2087 
2088 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2089 						       ORPHAN_DIR_SYSTEM_INODE,
2090 						       osb->slot_num);
2091 	if (!orphan_dir_inode) {
2092 		ret = -ENOENT;
2093 		mlog_errno(ret);
2094 		return ret;
2095 	}
2096 
2097 	inode_lock(orphan_dir_inode);
2098 
2099 	ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2100 	if (ret < 0) {
2101 		inode_unlock(orphan_dir_inode);
2102 		iput(orphan_dir_inode);
2103 
2104 		mlog_errno(ret);
2105 		return ret;
2106 	}
2107 
2108 	*ret_orphan_dir = orphan_dir_inode;
2109 	*ret_orphan_dir_bh = orphan_dir_bh;
2110 
2111 	return 0;
2112 }
2113 
2114 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2115 				      struct buffer_head *orphan_dir_bh,
2116 				      u64 blkno,
2117 				      char *name,
2118 				      struct ocfs2_dir_lookup_result *lookup,
2119 				      bool dio)
2120 {
2121 	int ret;
2122 	struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2123 	int namelen = dio ?
2124 			(OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2125 			OCFS2_ORPHAN_NAMELEN;
2126 
2127 	if (dio) {
2128 		ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2129 				OCFS2_DIO_ORPHAN_PREFIX);
2130 		if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2131 			ret = -EINVAL;
2132 			mlog_errno(ret);
2133 			return ret;
2134 		}
2135 
2136 		ret = ocfs2_blkno_stringify(blkno,
2137 				name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2138 	} else
2139 		ret = ocfs2_blkno_stringify(blkno, name);
2140 	if (ret < 0) {
2141 		mlog_errno(ret);
2142 		return ret;
2143 	}
2144 
2145 	ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2146 					   orphan_dir_bh, name,
2147 					   namelen, lookup);
2148 	if (ret < 0) {
2149 		mlog_errno(ret);
2150 		return ret;
2151 	}
2152 
2153 	return 0;
2154 }
2155 
2156 /**
2157  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2158  * insertion of an orphan.
2159  * @osb: ocfs2 file system
2160  * @ret_orphan_dir: Orphan dir inode - returned locked!
2161  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2162  * @lookup: dir lookup result, to be passed back into functions like
2163  *          ocfs2_orphan_add
2164  *
2165  * Returns zero on success and the ret_orphan_dir, name and lookup
2166  * fields will be populated.
2167  *
2168  * Returns non-zero on failure.
2169  */
2170 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2171 				    struct inode **ret_orphan_dir,
2172 				    u64 blkno,
2173 				    char *name,
2174 				    struct ocfs2_dir_lookup_result *lookup,
2175 				    bool dio)
2176 {
2177 	struct inode *orphan_dir_inode = NULL;
2178 	struct buffer_head *orphan_dir_bh = NULL;
2179 	int ret = 0;
2180 
2181 	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2182 					   &orphan_dir_bh);
2183 	if (ret < 0) {
2184 		mlog_errno(ret);
2185 		return ret;
2186 	}
2187 
2188 	ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2189 					 blkno, name, lookup, dio);
2190 	if (ret < 0) {
2191 		mlog_errno(ret);
2192 		goto out;
2193 	}
2194 
2195 	*ret_orphan_dir = orphan_dir_inode;
2196 
2197 out:
2198 	brelse(orphan_dir_bh);
2199 
2200 	if (ret) {
2201 		ocfs2_inode_unlock(orphan_dir_inode, 1);
2202 		inode_unlock(orphan_dir_inode);
2203 		iput(orphan_dir_inode);
2204 	}
2205 
2206 	if (ret)
2207 		mlog_errno(ret);
2208 	return ret;
2209 }
2210 
2211 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2212 			    handle_t *handle,
2213 			    struct inode *inode,
2214 			    struct buffer_head *fe_bh,
2215 			    char *name,
2216 			    struct ocfs2_dir_lookup_result *lookup,
2217 			    struct inode *orphan_dir_inode,
2218 			    bool dio)
2219 {
2220 	struct buffer_head *orphan_dir_bh = NULL;
2221 	int status = 0;
2222 	struct ocfs2_dinode *orphan_fe;
2223 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2224 	int namelen = dio ?
2225 			(OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2226 			OCFS2_ORPHAN_NAMELEN;
2227 
2228 	trace_ocfs2_orphan_add_begin(
2229 				(unsigned long long)OCFS2_I(inode)->ip_blkno);
2230 
2231 	status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2232 	if (status < 0) {
2233 		mlog_errno(status);
2234 		goto leave;
2235 	}
2236 
2237 	status = ocfs2_journal_access_di(handle,
2238 					 INODE_CACHE(orphan_dir_inode),
2239 					 orphan_dir_bh,
2240 					 OCFS2_JOURNAL_ACCESS_WRITE);
2241 	if (status < 0) {
2242 		mlog_errno(status);
2243 		goto leave;
2244 	}
2245 
2246 	/*
2247 	 * We're going to journal the change of i_flags and i_orphaned_slot.
2248 	 * It's safe anyway, though some callers may duplicate the journaling.
2249 	 * Journaling within the func just make the logic look more
2250 	 * straightforward.
2251 	 */
2252 	status = ocfs2_journal_access_di(handle,
2253 					 INODE_CACHE(inode),
2254 					 fe_bh,
2255 					 OCFS2_JOURNAL_ACCESS_WRITE);
2256 	if (status < 0) {
2257 		mlog_errno(status);
2258 		goto leave;
2259 	}
2260 
2261 	/* we're a cluster, and nlink can change on disk from
2262 	 * underneath us... */
2263 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2264 	if (S_ISDIR(inode->i_mode))
2265 		ocfs2_add_links_count(orphan_fe, 1);
2266 	set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2267 	ocfs2_journal_dirty(handle, orphan_dir_bh);
2268 
2269 	status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2270 				   namelen, inode,
2271 				   OCFS2_I(inode)->ip_blkno,
2272 				   orphan_dir_bh, lookup);
2273 	if (status < 0) {
2274 		mlog_errno(status);
2275 		goto rollback;
2276 	}
2277 
2278 	if (dio) {
2279 		/* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2280 		 * slot.
2281 		 */
2282 		fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2283 		fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2284 	} else {
2285 		fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2286 		OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2287 
2288 		/* Record which orphan dir our inode now resides
2289 		 * in. delete_inode will use this to determine which orphan
2290 		 * dir to lock. */
2291 		fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2292 	}
2293 
2294 	ocfs2_journal_dirty(handle, fe_bh);
2295 
2296 	trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2297 				   osb->slot_num);
2298 
2299 rollback:
2300 	if (status < 0) {
2301 		if (S_ISDIR(inode->i_mode))
2302 			ocfs2_add_links_count(orphan_fe, -1);
2303 		set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2304 	}
2305 
2306 leave:
2307 	brelse(orphan_dir_bh);
2308 
2309 	return status;
2310 }
2311 
2312 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2313 int ocfs2_orphan_del(struct ocfs2_super *osb,
2314 		     handle_t *handle,
2315 		     struct inode *orphan_dir_inode,
2316 		     struct inode *inode,
2317 		     struct buffer_head *orphan_dir_bh,
2318 		     bool dio)
2319 {
2320 	char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2321 	struct ocfs2_dinode *orphan_fe;
2322 	int status = 0;
2323 	struct ocfs2_dir_lookup_result lookup = { NULL, };
2324 
2325 	if (dio) {
2326 		status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2327 				OCFS2_DIO_ORPHAN_PREFIX);
2328 		if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2329 			status = -EINVAL;
2330 			mlog_errno(status);
2331 			return status;
2332 		}
2333 
2334 		status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2335 				name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2336 	} else
2337 		status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2338 	if (status < 0) {
2339 		mlog_errno(status);
2340 		goto leave;
2341 	}
2342 
2343 	trace_ocfs2_orphan_del(
2344 	     (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2345 	     name, strlen(name));
2346 
2347 	status = ocfs2_journal_access_di(handle,
2348 					 INODE_CACHE(orphan_dir_inode),
2349 					 orphan_dir_bh,
2350 					 OCFS2_JOURNAL_ACCESS_WRITE);
2351 	if (status < 0) {
2352 		mlog_errno(status);
2353 		goto leave;
2354 	}
2355 
2356 	/* find it's spot in the orphan directory */
2357 	status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2358 				  &lookup);
2359 	if (status) {
2360 		mlog_errno(status);
2361 		goto leave;
2362 	}
2363 
2364 	/* remove it from the orphan directory */
2365 	status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2366 	if (status < 0) {
2367 		mlog_errno(status);
2368 		goto leave;
2369 	}
2370 
2371 	/* do the i_nlink dance! :) */
2372 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2373 	if (S_ISDIR(inode->i_mode))
2374 		ocfs2_add_links_count(orphan_fe, -1);
2375 	set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2376 	ocfs2_journal_dirty(handle, orphan_dir_bh);
2377 
2378 leave:
2379 	ocfs2_free_dir_lookup_result(&lookup);
2380 
2381 	if (status)
2382 		mlog_errno(status);
2383 	return status;
2384 }
2385 
2386 /**
2387  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2388  * allocated file. This is different from the typical 'add to orphan dir'
2389  * operation in that the inode does not yet exist. This is a problem because
2390  * the orphan dir stringifies the inode block number to come up with it's
2391  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2392  * problem. This function works around it by calling deeper into the orphan
2393  * and suballoc code than other callers. Use this only by necessity.
2394  * @dir: The directory which this inode will ultimately wind up under - not the
2395  * orphan dir!
2396  * @dir_bh: buffer_head the @dir inode block
2397  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2398  * with the string to be used for orphan dirent. Pass back to the orphan dir
2399  * code.
2400  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2401  * dir code.
2402  * @ret_di_blkno: block number where the new inode will be allocated.
2403  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2404  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2405  *
2406  * Returns zero on success and the ret_orphan_dir, name and lookup
2407  * fields will be populated.
2408  *
2409  * Returns non-zero on failure.
2410  */
2411 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2412 					struct buffer_head *dir_bh,
2413 					char *orphan_name,
2414 					struct inode **ret_orphan_dir,
2415 					u64 *ret_di_blkno,
2416 					struct ocfs2_dir_lookup_result *orphan_insert,
2417 					struct ocfs2_alloc_context **ret_inode_ac)
2418 {
2419 	int ret;
2420 	u64 di_blkno;
2421 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2422 	struct inode *orphan_dir = NULL;
2423 	struct buffer_head *orphan_dir_bh = NULL;
2424 	struct ocfs2_alloc_context *inode_ac = NULL;
2425 
2426 	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2427 	if (ret < 0) {
2428 		mlog_errno(ret);
2429 		return ret;
2430 	}
2431 
2432 	/* reserve an inode spot */
2433 	ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2434 	if (ret < 0) {
2435 		if (ret != -ENOSPC)
2436 			mlog_errno(ret);
2437 		goto out;
2438 	}
2439 
2440 	ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2441 				       &di_blkno);
2442 	if (ret) {
2443 		mlog_errno(ret);
2444 		goto out;
2445 	}
2446 
2447 	ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2448 					 di_blkno, orphan_name, orphan_insert,
2449 					 false);
2450 	if (ret < 0) {
2451 		mlog_errno(ret);
2452 		goto out;
2453 	}
2454 
2455 out:
2456 	if (ret == 0) {
2457 		*ret_orphan_dir = orphan_dir;
2458 		*ret_di_blkno = di_blkno;
2459 		*ret_inode_ac = inode_ac;
2460 		/*
2461 		 * orphan_name and orphan_insert are already up to
2462 		 * date via prepare_orphan_dir
2463 		 */
2464 	} else {
2465 		/* Unroll reserve_new_inode* */
2466 		if (inode_ac)
2467 			ocfs2_free_alloc_context(inode_ac);
2468 
2469 		/* Unroll orphan dir locking */
2470 		inode_unlock(orphan_dir);
2471 		ocfs2_inode_unlock(orphan_dir, 1);
2472 		iput(orphan_dir);
2473 	}
2474 
2475 	brelse(orphan_dir_bh);
2476 
2477 	return ret;
2478 }
2479 
2480 int ocfs2_create_inode_in_orphan(struct inode *dir,
2481 				 int mode,
2482 				 struct inode **new_inode)
2483 {
2484 	int status, did_quota_inode = 0;
2485 	struct inode *inode = NULL;
2486 	struct inode *orphan_dir = NULL;
2487 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2488 	handle_t *handle = NULL;
2489 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2490 	struct buffer_head *parent_di_bh = NULL;
2491 	struct buffer_head *new_di_bh = NULL;
2492 	struct ocfs2_alloc_context *inode_ac = NULL;
2493 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2494 	u64 uninitialized_var(di_blkno), suballoc_loc;
2495 	u16 suballoc_bit;
2496 
2497 	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2498 	if (status < 0) {
2499 		if (status != -ENOENT)
2500 			mlog_errno(status);
2501 		return status;
2502 	}
2503 
2504 	status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2505 					      orphan_name, &orphan_dir,
2506 					      &di_blkno, &orphan_insert, &inode_ac);
2507 	if (status < 0) {
2508 		if (status != -ENOSPC)
2509 			mlog_errno(status);
2510 		goto leave;
2511 	}
2512 
2513 	inode = ocfs2_get_init_inode(dir, mode);
2514 	if (IS_ERR(inode)) {
2515 		status = PTR_ERR(inode);
2516 		inode = NULL;
2517 		mlog_errno(status);
2518 		goto leave;
2519 	}
2520 
2521 	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2522 	if (IS_ERR(handle)) {
2523 		status = PTR_ERR(handle);
2524 		handle = NULL;
2525 		mlog_errno(status);
2526 		goto leave;
2527 	}
2528 
2529 	status = dquot_alloc_inode(inode);
2530 	if (status)
2531 		goto leave;
2532 	did_quota_inode = 1;
2533 
2534 	status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2535 					      &suballoc_loc,
2536 					      &suballoc_bit, di_blkno);
2537 	if (status < 0) {
2538 		mlog_errno(status);
2539 		goto leave;
2540 	}
2541 
2542 	clear_nlink(inode);
2543 	/* do the real work now. */
2544 	status = __ocfs2_mknod_locked(dir, inode,
2545 				      0, &new_di_bh, parent_di_bh, handle,
2546 				      inode_ac, di_blkno, suballoc_loc,
2547 				      suballoc_bit);
2548 	if (status < 0) {
2549 		mlog_errno(status);
2550 		goto leave;
2551 	}
2552 
2553 	status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2554 				  &orphan_insert, orphan_dir, false);
2555 	if (status < 0) {
2556 		mlog_errno(status);
2557 		goto leave;
2558 	}
2559 
2560 	/* get open lock so that only nodes can't remove it from orphan dir. */
2561 	status = ocfs2_open_lock(inode);
2562 	if (status < 0)
2563 		mlog_errno(status);
2564 
2565 	insert_inode_hash(inode);
2566 leave:
2567 	if (status < 0 && did_quota_inode)
2568 		dquot_free_inode(inode);
2569 	if (handle)
2570 		ocfs2_commit_trans(osb, handle);
2571 
2572 	if (orphan_dir) {
2573 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
2574 		ocfs2_inode_unlock(orphan_dir, 1);
2575 		inode_unlock(orphan_dir);
2576 		iput(orphan_dir);
2577 	}
2578 
2579 	if ((status < 0) && inode) {
2580 		clear_nlink(inode);
2581 		iput(inode);
2582 	}
2583 
2584 	if (inode_ac)
2585 		ocfs2_free_alloc_context(inode_ac);
2586 
2587 	brelse(new_di_bh);
2588 
2589 	if (!status)
2590 		*new_inode = inode;
2591 
2592 	ocfs2_free_dir_lookup_result(&orphan_insert);
2593 
2594 	ocfs2_inode_unlock(dir, 1);
2595 	brelse(parent_di_bh);
2596 	return status;
2597 }
2598 
2599 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2600 	struct inode *inode)
2601 {
2602 	char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2603 	struct inode *orphan_dir_inode = NULL;
2604 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2605 	struct buffer_head *di_bh = NULL;
2606 	int status = 0;
2607 	handle_t *handle = NULL;
2608 	struct ocfs2_dinode *di = NULL;
2609 
2610 	status = ocfs2_inode_lock(inode, &di_bh, 1);
2611 	if (status < 0) {
2612 		mlog_errno(status);
2613 		goto bail;
2614 	}
2615 
2616 	di = (struct ocfs2_dinode *) di_bh->b_data;
2617 	/*
2618 	 * Another append dio crashed?
2619 	 * If so, manually recover it first.
2620 	 */
2621 	if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2622 		status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2623 		if (status < 0) {
2624 			if (status != -ENOSPC)
2625 				mlog_errno(status);
2626 			goto bail_unlock_inode;
2627 		}
2628 
2629 		status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2630 		if (status < 0) {
2631 			mlog_errno(status);
2632 			goto bail_unlock_inode;
2633 		}
2634 	}
2635 
2636 	status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2637 			OCFS2_I(inode)->ip_blkno,
2638 			orphan_name,
2639 			&orphan_insert,
2640 			true);
2641 	if (status < 0) {
2642 		mlog_errno(status);
2643 		goto bail_unlock_inode;
2644 	}
2645 
2646 	handle = ocfs2_start_trans(osb,
2647 			OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2648 	if (IS_ERR(handle)) {
2649 		status = PTR_ERR(handle);
2650 		goto bail_unlock_orphan;
2651 	}
2652 
2653 	status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2654 			&orphan_insert, orphan_dir_inode, true);
2655 	if (status)
2656 		mlog_errno(status);
2657 
2658 	ocfs2_commit_trans(osb, handle);
2659 
2660 bail_unlock_orphan:
2661 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2662 	inode_unlock(orphan_dir_inode);
2663 	iput(orphan_dir_inode);
2664 
2665 	ocfs2_free_dir_lookup_result(&orphan_insert);
2666 
2667 bail_unlock_inode:
2668 	ocfs2_inode_unlock(inode, 1);
2669 	brelse(di_bh);
2670 
2671 bail:
2672 	return status;
2673 }
2674 
2675 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2676 		struct inode *inode, struct buffer_head *di_bh,
2677 		int update_isize, loff_t end)
2678 {
2679 	struct inode *orphan_dir_inode = NULL;
2680 	struct buffer_head *orphan_dir_bh = NULL;
2681 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2682 	handle_t *handle = NULL;
2683 	int status = 0;
2684 
2685 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2686 			ORPHAN_DIR_SYSTEM_INODE,
2687 			le16_to_cpu(di->i_dio_orphaned_slot));
2688 	if (!orphan_dir_inode) {
2689 		status = -ENOENT;
2690 		mlog_errno(status);
2691 		goto bail;
2692 	}
2693 
2694 	inode_lock(orphan_dir_inode);
2695 	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2696 	if (status < 0) {
2697 		inode_unlock(orphan_dir_inode);
2698 		iput(orphan_dir_inode);
2699 		mlog_errno(status);
2700 		goto bail;
2701 	}
2702 
2703 	handle = ocfs2_start_trans(osb,
2704 			OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2705 	if (IS_ERR(handle)) {
2706 		status = PTR_ERR(handle);
2707 		goto bail_unlock_orphan;
2708 	}
2709 
2710 	BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2711 
2712 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2713 				inode, orphan_dir_bh, true);
2714 	if (status < 0) {
2715 		mlog_errno(status);
2716 		goto bail_commit;
2717 	}
2718 
2719 	status = ocfs2_journal_access_di(handle,
2720 			INODE_CACHE(inode),
2721 			di_bh,
2722 			OCFS2_JOURNAL_ACCESS_WRITE);
2723 	if (status < 0) {
2724 		mlog_errno(status);
2725 		goto bail_commit;
2726 	}
2727 
2728 	di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2729 	di->i_dio_orphaned_slot = 0;
2730 
2731 	if (update_isize) {
2732 		status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2733 		if (status)
2734 			mlog_errno(status);
2735 	} else
2736 		ocfs2_journal_dirty(handle, di_bh);
2737 
2738 bail_commit:
2739 	ocfs2_commit_trans(osb, handle);
2740 
2741 bail_unlock_orphan:
2742 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2743 	inode_unlock(orphan_dir_inode);
2744 	brelse(orphan_dir_bh);
2745 	iput(orphan_dir_inode);
2746 
2747 bail:
2748 	return status;
2749 }
2750 
2751 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2752 				   struct inode *inode,
2753 				   struct dentry *dentry)
2754 {
2755 	int status = 0;
2756 	struct buffer_head *parent_di_bh = NULL;
2757 	handle_t *handle = NULL;
2758 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2759 	struct ocfs2_dinode *dir_di, *di;
2760 	struct inode *orphan_dir_inode = NULL;
2761 	struct buffer_head *orphan_dir_bh = NULL;
2762 	struct buffer_head *di_bh = NULL;
2763 	struct ocfs2_dir_lookup_result lookup = { NULL, };
2764 
2765 	trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2766 				dentry->d_name.len, dentry->d_name.name,
2767 				(unsigned long long)OCFS2_I(dir)->ip_blkno,
2768 				(unsigned long long)OCFS2_I(inode)->ip_blkno);
2769 
2770 	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2771 	if (status < 0) {
2772 		if (status != -ENOENT)
2773 			mlog_errno(status);
2774 		return status;
2775 	}
2776 
2777 	dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2778 	if (!dir_di->i_links_count) {
2779 		/* can't make a file in a deleted directory. */
2780 		status = -ENOENT;
2781 		goto leave;
2782 	}
2783 
2784 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2785 					   dentry->d_name.len);
2786 	if (status)
2787 		goto leave;
2788 
2789 	/* get a spot inside the dir. */
2790 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2791 					      dentry->d_name.name,
2792 					      dentry->d_name.len, &lookup);
2793 	if (status < 0) {
2794 		mlog_errno(status);
2795 		goto leave;
2796 	}
2797 
2798 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2799 						       ORPHAN_DIR_SYSTEM_INODE,
2800 						       osb->slot_num);
2801 	if (!orphan_dir_inode) {
2802 		status = -ENOENT;
2803 		mlog_errno(status);
2804 		goto leave;
2805 	}
2806 
2807 	inode_lock(orphan_dir_inode);
2808 
2809 	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2810 	if (status < 0) {
2811 		mlog_errno(status);
2812 		inode_unlock(orphan_dir_inode);
2813 		iput(orphan_dir_inode);
2814 		goto leave;
2815 	}
2816 
2817 	status = ocfs2_read_inode_block(inode, &di_bh);
2818 	if (status < 0) {
2819 		mlog_errno(status);
2820 		goto orphan_unlock;
2821 	}
2822 
2823 	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2824 	if (IS_ERR(handle)) {
2825 		status = PTR_ERR(handle);
2826 		handle = NULL;
2827 		mlog_errno(status);
2828 		goto orphan_unlock;
2829 	}
2830 
2831 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2832 					 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2833 	if (status < 0) {
2834 		mlog_errno(status);
2835 		goto out_commit;
2836 	}
2837 
2838 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2839 				  orphan_dir_bh, false);
2840 	if (status < 0) {
2841 		mlog_errno(status);
2842 		goto out_commit;
2843 	}
2844 
2845 	di = (struct ocfs2_dinode *)di_bh->b_data;
2846 	di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2847 	di->i_orphaned_slot = 0;
2848 	set_nlink(inode, 1);
2849 	ocfs2_set_links_count(di, inode->i_nlink);
2850 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
2851 	ocfs2_journal_dirty(handle, di_bh);
2852 
2853 	status = ocfs2_add_entry(handle, dentry, inode,
2854 				 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2855 				 &lookup);
2856 	if (status < 0) {
2857 		mlog_errno(status);
2858 		goto out_commit;
2859 	}
2860 
2861 	status = ocfs2_dentry_attach_lock(dentry, inode,
2862 					  OCFS2_I(dir)->ip_blkno);
2863 	if (status) {
2864 		mlog_errno(status);
2865 		goto out_commit;
2866 	}
2867 
2868 	d_instantiate(dentry, inode);
2869 	status = 0;
2870 out_commit:
2871 	ocfs2_commit_trans(osb, handle);
2872 orphan_unlock:
2873 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2874 	inode_unlock(orphan_dir_inode);
2875 	iput(orphan_dir_inode);
2876 leave:
2877 
2878 	ocfs2_inode_unlock(dir, 1);
2879 
2880 	brelse(di_bh);
2881 	brelse(parent_di_bh);
2882 	brelse(orphan_dir_bh);
2883 
2884 	ocfs2_free_dir_lookup_result(&lookup);
2885 
2886 	if (status)
2887 		mlog_errno(status);
2888 
2889 	return status;
2890 }
2891 
2892 const struct inode_operations ocfs2_dir_iops = {
2893 	.create		= ocfs2_create,
2894 	.lookup		= ocfs2_lookup,
2895 	.link		= ocfs2_link,
2896 	.unlink		= ocfs2_unlink,
2897 	.rmdir		= ocfs2_unlink,
2898 	.symlink	= ocfs2_symlink,
2899 	.mkdir		= ocfs2_mkdir,
2900 	.mknod		= ocfs2_mknod,
2901 	.rename		= ocfs2_rename,
2902 	.setattr	= ocfs2_setattr,
2903 	.getattr	= ocfs2_getattr,
2904 	.permission	= ocfs2_permission,
2905 	.listxattr	= ocfs2_listxattr,
2906 	.fiemap         = ocfs2_fiemap,
2907 	.get_acl	= ocfs2_iop_get_acl,
2908 	.set_acl	= ocfs2_iop_set_acl,
2909 };
2910