xref: /linux/fs/ocfs2/namei.c (revision 48dea9a700c8728cc31a1dd44588b97578de86ee)
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 roll_back;
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 roll_back;
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 roll_back;
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 roll_back;
441 	}
442 
443 	insert_inode_hash(inode);
444 	d_instantiate(dentry, inode);
445 	status = 0;
446 
447 roll_back:
448 	if (status < 0 && S_ISDIR(mode)) {
449 		ocfs2_add_links_count(dirfe, -1);
450 		drop_nlink(dir);
451 	}
452 
453 leave:
454 	if (status < 0 && did_quota_inode)
455 		dquot_free_inode(inode);
456 	if (handle)
457 		ocfs2_commit_trans(osb, handle);
458 
459 	ocfs2_inode_unlock(dir, 1);
460 	if (did_block_signals)
461 		ocfs2_unblock_signals(&oldset);
462 
463 	brelse(new_fe_bh);
464 	brelse(parent_fe_bh);
465 	kfree(si.value);
466 
467 	ocfs2_free_dir_lookup_result(&lookup);
468 
469 	if (inode_ac)
470 		ocfs2_free_alloc_context(inode_ac);
471 
472 	if (data_ac)
473 		ocfs2_free_alloc_context(data_ac);
474 
475 	if (meta_ac)
476 		ocfs2_free_alloc_context(meta_ac);
477 
478 	/*
479 	 * We should call iput after the i_mutex of the bitmap been
480 	 * unlocked in ocfs2_free_alloc_context, or the
481 	 * ocfs2_delete_inode will mutex_lock again.
482 	 */
483 	if ((status < 0) && inode) {
484 		if (dl)
485 			ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
486 
487 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
488 		clear_nlink(inode);
489 		iput(inode);
490 	}
491 
492 	if (status)
493 		mlog_errno(status);
494 
495 	return status;
496 }
497 
498 static int __ocfs2_mknod_locked(struct inode *dir,
499 				struct inode *inode,
500 				dev_t dev,
501 				struct buffer_head **new_fe_bh,
502 				struct buffer_head *parent_fe_bh,
503 				handle_t *handle,
504 				struct ocfs2_alloc_context *inode_ac,
505 				u64 fe_blkno, u64 suballoc_loc, u16 suballoc_bit)
506 {
507 	int status = 0;
508 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
509 	struct ocfs2_dinode *fe = NULL;
510 	struct ocfs2_extent_list *fel;
511 	u16 feat;
512 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
513 	struct timespec64 ts;
514 
515 	*new_fe_bh = NULL;
516 
517 	/* populate as many fields early on as possible - many of
518 	 * these are used by the support functions here and in
519 	 * callers. */
520 	inode->i_ino = ino_from_blkno(osb->sb, fe_blkno);
521 	oi->ip_blkno = fe_blkno;
522 	spin_lock(&osb->osb_lock);
523 	inode->i_generation = osb->s_next_generation++;
524 	spin_unlock(&osb->osb_lock);
525 
526 	*new_fe_bh = sb_getblk(osb->sb, fe_blkno);
527 	if (!*new_fe_bh) {
528 		status = -ENOMEM;
529 		mlog_errno(status);
530 		goto leave;
531 	}
532 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), *new_fe_bh);
533 
534 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
535 					 *new_fe_bh,
536 					 OCFS2_JOURNAL_ACCESS_CREATE);
537 	if (status < 0) {
538 		mlog_errno(status);
539 		goto leave;
540 	}
541 
542 	fe = (struct ocfs2_dinode *) (*new_fe_bh)->b_data;
543 	memset(fe, 0, osb->sb->s_blocksize);
544 
545 	fe->i_generation = cpu_to_le32(inode->i_generation);
546 	fe->i_fs_generation = cpu_to_le32(osb->fs_generation);
547 	fe->i_blkno = cpu_to_le64(fe_blkno);
548 	fe->i_suballoc_loc = cpu_to_le64(suballoc_loc);
549 	fe->i_suballoc_bit = cpu_to_le16(suballoc_bit);
550 	fe->i_suballoc_slot = cpu_to_le16(inode_ac->ac_alloc_slot);
551 	fe->i_uid = cpu_to_le32(i_uid_read(inode));
552 	fe->i_gid = cpu_to_le32(i_gid_read(inode));
553 	fe->i_mode = cpu_to_le16(inode->i_mode);
554 	if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode))
555 		fe->id1.dev1.i_rdev = cpu_to_le64(huge_encode_dev(dev));
556 
557 	ocfs2_set_links_count(fe, inode->i_nlink);
558 
559 	fe->i_last_eb_blk = 0;
560 	strcpy(fe->i_signature, OCFS2_INODE_SIGNATURE);
561 	fe->i_flags |= cpu_to_le32(OCFS2_VALID_FL);
562 	ktime_get_real_ts64(&ts);
563 	fe->i_atime = fe->i_ctime = fe->i_mtime =
564 		cpu_to_le64(ts.tv_sec);
565 	fe->i_mtime_nsec = fe->i_ctime_nsec = fe->i_atime_nsec =
566 		cpu_to_le32(ts.tv_nsec);
567 	fe->i_dtime = 0;
568 
569 	/*
570 	 * If supported, directories start with inline data. If inline
571 	 * isn't supported, but indexing is, we start them as indexed.
572 	 */
573 	feat = le16_to_cpu(fe->i_dyn_features);
574 	if (S_ISDIR(inode->i_mode) && ocfs2_supports_inline_data(osb)) {
575 		fe->i_dyn_features = cpu_to_le16(feat | OCFS2_INLINE_DATA_FL);
576 
577 		fe->id2.i_data.id_count = cpu_to_le16(
578 				ocfs2_max_inline_data_with_xattr(osb->sb, fe));
579 	} else {
580 		fel = &fe->id2.i_list;
581 		fel->l_tree_depth = 0;
582 		fel->l_next_free_rec = 0;
583 		fel->l_count = cpu_to_le16(ocfs2_extent_recs_per_inode(osb->sb));
584 	}
585 
586 	ocfs2_journal_dirty(handle, *new_fe_bh);
587 
588 	ocfs2_populate_inode(inode, fe, 1);
589 	ocfs2_ci_set_new(osb, INODE_CACHE(inode));
590 	if (!ocfs2_mount_local(osb)) {
591 		status = ocfs2_create_new_inode_locks(inode);
592 		if (status < 0)
593 			mlog_errno(status);
594 	}
595 
596 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
597 
598 leave:
599 	if (status < 0) {
600 		if (*new_fe_bh) {
601 			brelse(*new_fe_bh);
602 			*new_fe_bh = NULL;
603 		}
604 	}
605 
606 	if (status)
607 		mlog_errno(status);
608 	return status;
609 }
610 
611 static int ocfs2_mknod_locked(struct ocfs2_super *osb,
612 			      struct inode *dir,
613 			      struct inode *inode,
614 			      dev_t dev,
615 			      struct buffer_head **new_fe_bh,
616 			      struct buffer_head *parent_fe_bh,
617 			      handle_t *handle,
618 			      struct ocfs2_alloc_context *inode_ac)
619 {
620 	int status = 0;
621 	u64 suballoc_loc, fe_blkno = 0;
622 	u16 suballoc_bit;
623 
624 	*new_fe_bh = NULL;
625 
626 	status = ocfs2_claim_new_inode(handle, dir, parent_fe_bh,
627 				       inode_ac, &suballoc_loc,
628 				       &suballoc_bit, &fe_blkno);
629 	if (status < 0) {
630 		mlog_errno(status);
631 		return status;
632 	}
633 
634 	status = __ocfs2_mknod_locked(dir, inode, dev, new_fe_bh,
635 				    parent_fe_bh, handle, inode_ac,
636 				    fe_blkno, suballoc_loc, suballoc_bit);
637 	if (status < 0) {
638 		u64 bg_blkno = ocfs2_which_suballoc_group(fe_blkno, suballoc_bit);
639 		int tmp = ocfs2_free_suballoc_bits(handle, inode_ac->ac_inode,
640 				inode_ac->ac_bh, suballoc_bit, bg_blkno, 1);
641 		if (tmp)
642 			mlog_errno(tmp);
643 	}
644 
645 	return status;
646 }
647 
648 static int ocfs2_mkdir(struct inode *dir,
649 		       struct dentry *dentry,
650 		       umode_t mode)
651 {
652 	int ret;
653 
654 	trace_ocfs2_mkdir(dir, dentry, dentry->d_name.len, dentry->d_name.name,
655 			  OCFS2_I(dir)->ip_blkno, mode);
656 	ret = ocfs2_mknod(dir, dentry, mode | S_IFDIR, 0);
657 	if (ret)
658 		mlog_errno(ret);
659 
660 	return ret;
661 }
662 
663 static int ocfs2_create(struct inode *dir,
664 			struct dentry *dentry,
665 			umode_t mode,
666 			bool excl)
667 {
668 	int ret;
669 
670 	trace_ocfs2_create(dir, dentry, dentry->d_name.len, dentry->d_name.name,
671 			   (unsigned long long)OCFS2_I(dir)->ip_blkno, mode);
672 	ret = ocfs2_mknod(dir, dentry, mode | S_IFREG, 0);
673 	if (ret)
674 		mlog_errno(ret);
675 
676 	return ret;
677 }
678 
679 static int ocfs2_link(struct dentry *old_dentry,
680 		      struct inode *dir,
681 		      struct dentry *dentry)
682 {
683 	handle_t *handle;
684 	struct inode *inode = d_inode(old_dentry);
685 	struct inode *old_dir = d_inode(old_dentry->d_parent);
686 	int err;
687 	struct buffer_head *fe_bh = NULL;
688 	struct buffer_head *old_dir_bh = NULL;
689 	struct buffer_head *parent_fe_bh = NULL;
690 	struct ocfs2_dinode *fe = NULL;
691 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
692 	struct ocfs2_dir_lookup_result lookup = { NULL, };
693 	sigset_t oldset;
694 	u64 old_de_ino;
695 
696 	trace_ocfs2_link((unsigned long long)OCFS2_I(inode)->ip_blkno,
697 			 old_dentry->d_name.len, old_dentry->d_name.name,
698 			 dentry->d_name.len, dentry->d_name.name);
699 
700 	if (S_ISDIR(inode->i_mode))
701 		return -EPERM;
702 
703 	err = dquot_initialize(dir);
704 	if (err) {
705 		mlog_errno(err);
706 		return err;
707 	}
708 
709 	err = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
710 			&parent_fe_bh, dir, 0);
711 	if (err < 0) {
712 		if (err != -ENOENT)
713 			mlog_errno(err);
714 		return err;
715 	}
716 
717 	/* make sure both dirs have bhs
718 	 * get an extra ref on old_dir_bh if old==new */
719 	if (!parent_fe_bh) {
720 		if (old_dir_bh) {
721 			parent_fe_bh = old_dir_bh;
722 			get_bh(parent_fe_bh);
723 		} else {
724 			mlog(ML_ERROR, "%s: no old_dir_bh!\n", osb->uuid_str);
725 			err = -EIO;
726 			goto out;
727 		}
728 	}
729 
730 	if (!dir->i_nlink) {
731 		err = -ENOENT;
732 		goto out;
733 	}
734 
735 	err = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
736 			old_dentry->d_name.len, &old_de_ino);
737 	if (err) {
738 		err = -ENOENT;
739 		goto out;
740 	}
741 
742 	/*
743 	 * Check whether another node removed the source inode while we
744 	 * were in the vfs.
745 	 */
746 	if (old_de_ino != OCFS2_I(inode)->ip_blkno) {
747 		err = -ENOENT;
748 		goto out;
749 	}
750 
751 	err = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
752 					dentry->d_name.len);
753 	if (err)
754 		goto out;
755 
756 	err = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
757 					   dentry->d_name.name,
758 					   dentry->d_name.len, &lookup);
759 	if (err < 0) {
760 		mlog_errno(err);
761 		goto out;
762 	}
763 
764 	err = ocfs2_inode_lock(inode, &fe_bh, 1);
765 	if (err < 0) {
766 		if (err != -ENOENT)
767 			mlog_errno(err);
768 		goto out;
769 	}
770 
771 	fe = (struct ocfs2_dinode *) fe_bh->b_data;
772 	if (ocfs2_read_links_count(fe) >= ocfs2_link_max(osb)) {
773 		err = -EMLINK;
774 		goto out_unlock_inode;
775 	}
776 
777 	handle = ocfs2_start_trans(osb, ocfs2_link_credits(osb->sb));
778 	if (IS_ERR(handle)) {
779 		err = PTR_ERR(handle);
780 		handle = NULL;
781 		mlog_errno(err);
782 		goto out_unlock_inode;
783 	}
784 
785 	/* Starting to change things, restart is no longer possible. */
786 	ocfs2_block_signals(&oldset);
787 
788 	err = ocfs2_journal_access_di(handle, INODE_CACHE(inode), fe_bh,
789 				      OCFS2_JOURNAL_ACCESS_WRITE);
790 	if (err < 0) {
791 		mlog_errno(err);
792 		goto out_commit;
793 	}
794 
795 	inc_nlink(inode);
796 	inode->i_ctime = current_time(inode);
797 	ocfs2_set_links_count(fe, inode->i_nlink);
798 	fe->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
799 	fe->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
800 	ocfs2_journal_dirty(handle, fe_bh);
801 
802 	err = ocfs2_add_entry(handle, dentry, inode,
803 			      OCFS2_I(inode)->ip_blkno,
804 			      parent_fe_bh, &lookup);
805 	if (err) {
806 		ocfs2_add_links_count(fe, -1);
807 		drop_nlink(inode);
808 		mlog_errno(err);
809 		goto out_commit;
810 	}
811 
812 	err = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
813 	if (err) {
814 		mlog_errno(err);
815 		goto out_commit;
816 	}
817 
818 	ihold(inode);
819 	d_instantiate(dentry, inode);
820 
821 out_commit:
822 	ocfs2_commit_trans(osb, handle);
823 	ocfs2_unblock_signals(&oldset);
824 out_unlock_inode:
825 	ocfs2_inode_unlock(inode, 1);
826 
827 out:
828 	ocfs2_double_unlock(old_dir, dir);
829 
830 	brelse(fe_bh);
831 	brelse(parent_fe_bh);
832 	brelse(old_dir_bh);
833 
834 	ocfs2_free_dir_lookup_result(&lookup);
835 
836 	if (err)
837 		mlog_errno(err);
838 
839 	return err;
840 }
841 
842 /*
843  * Takes and drops an exclusive lock on the given dentry. This will
844  * force other nodes to drop it.
845  */
846 static int ocfs2_remote_dentry_delete(struct dentry *dentry)
847 {
848 	int ret;
849 
850 	ret = ocfs2_dentry_lock(dentry, 1);
851 	if (ret)
852 		mlog_errno(ret);
853 	else
854 		ocfs2_dentry_unlock(dentry, 1);
855 
856 	return ret;
857 }
858 
859 static inline int ocfs2_inode_is_unlinkable(struct inode *inode)
860 {
861 	if (S_ISDIR(inode->i_mode)) {
862 		if (inode->i_nlink == 2)
863 			return 1;
864 		return 0;
865 	}
866 
867 	if (inode->i_nlink == 1)
868 		return 1;
869 	return 0;
870 }
871 
872 static int ocfs2_unlink(struct inode *dir,
873 			struct dentry *dentry)
874 {
875 	int status;
876 	int child_locked = 0;
877 	bool is_unlinkable = false;
878 	struct inode *inode = d_inode(dentry);
879 	struct inode *orphan_dir = NULL;
880 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
881 	u64 blkno;
882 	struct ocfs2_dinode *fe = NULL;
883 	struct buffer_head *fe_bh = NULL;
884 	struct buffer_head *parent_node_bh = NULL;
885 	handle_t *handle = NULL;
886 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
887 	struct ocfs2_dir_lookup_result lookup = { NULL, };
888 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
889 
890 	trace_ocfs2_unlink(dir, dentry, dentry->d_name.len,
891 			   dentry->d_name.name,
892 			   (unsigned long long)OCFS2_I(dir)->ip_blkno,
893 			   (unsigned long long)OCFS2_I(inode)->ip_blkno);
894 
895 	status = dquot_initialize(dir);
896 	if (status) {
897 		mlog_errno(status);
898 		return status;
899 	}
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(dir);
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 (orphan_dir) {
1022 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
1023 		ocfs2_inode_unlock(orphan_dir, 1);
1024 		inode_unlock(orphan_dir);
1025 		iput(orphan_dir);
1026 	}
1027 
1028 	if (child_locked)
1029 		ocfs2_inode_unlock(inode, 1);
1030 
1031 	ocfs2_inode_unlock(dir, 1);
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 
1120 	trace_ocfs2_double_lock((unsigned long long)oi1->ip_blkno,
1121 				(unsigned long long)oi2->ip_blkno);
1122 
1123 	if (*bh1)
1124 		*bh1 = NULL;
1125 	if (*bh2)
1126 		*bh2 = NULL;
1127 
1128 	/* we always want to lock the one with the lower lockid first.
1129 	 * and if they are nested, we lock ancestor first */
1130 	if (oi1->ip_blkno != oi2->ip_blkno) {
1131 		inode1_is_ancestor = ocfs2_check_if_ancestor(osb, oi2->ip_blkno,
1132 				oi1->ip_blkno);
1133 		if (inode1_is_ancestor < 0) {
1134 			status = inode1_is_ancestor;
1135 			goto bail;
1136 		}
1137 
1138 		inode2_is_ancestor = ocfs2_check_if_ancestor(osb, oi1->ip_blkno,
1139 				oi2->ip_blkno);
1140 		if (inode2_is_ancestor < 0) {
1141 			status = inode2_is_ancestor;
1142 			goto bail;
1143 		}
1144 
1145 		if ((inode1_is_ancestor == 1) ||
1146 				(oi1->ip_blkno < oi2->ip_blkno &&
1147 				inode2_is_ancestor == 0)) {
1148 			/* switch id1 and id2 around */
1149 			swap(bh2, bh1);
1150 			swap(inode2, inode1);
1151 		}
1152 		/* lock id2 */
1153 		status = ocfs2_inode_lock_nested(inode2, bh2, 1,
1154 				rename == 1 ? OI_LS_RENAME1 : OI_LS_PARENT);
1155 		if (status < 0) {
1156 			if (status != -ENOENT)
1157 				mlog_errno(status);
1158 			goto bail;
1159 		}
1160 	}
1161 
1162 	/* lock id1 */
1163 	status = ocfs2_inode_lock_nested(inode1, bh1, 1,
1164 			rename == 1 ?  OI_LS_RENAME2 : OI_LS_PARENT);
1165 	if (status < 0) {
1166 		/*
1167 		 * An error return must mean that no cluster locks
1168 		 * were held on function exit.
1169 		 */
1170 		if (oi1->ip_blkno != oi2->ip_blkno) {
1171 			ocfs2_inode_unlock(inode2, 1);
1172 			brelse(*bh2);
1173 			*bh2 = NULL;
1174 		}
1175 
1176 		if (status != -ENOENT)
1177 			mlog_errno(status);
1178 	}
1179 
1180 	trace_ocfs2_double_lock_end(
1181 			(unsigned long long)oi1->ip_blkno,
1182 			(unsigned long long)oi2->ip_blkno);
1183 
1184 bail:
1185 	if (status)
1186 		mlog_errno(status);
1187 	return status;
1188 }
1189 
1190 static void ocfs2_double_unlock(struct inode *inode1, struct inode *inode2)
1191 {
1192 	ocfs2_inode_unlock(inode1, 1);
1193 
1194 	if (inode1 != inode2)
1195 		ocfs2_inode_unlock(inode2, 1);
1196 }
1197 
1198 static int ocfs2_rename(struct inode *old_dir,
1199 			struct dentry *old_dentry,
1200 			struct inode *new_dir,
1201 			struct dentry *new_dentry,
1202 			unsigned int flags)
1203 {
1204 	int status = 0, rename_lock = 0, parents_locked = 0, target_exists = 0;
1205 	int old_child_locked = 0, new_child_locked = 0, update_dot_dot = 0;
1206 	struct inode *old_inode = d_inode(old_dentry);
1207 	struct inode *new_inode = d_inode(new_dentry);
1208 	struct inode *orphan_dir = NULL;
1209 	struct ocfs2_dinode *newfe = NULL;
1210 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
1211 	struct buffer_head *newfe_bh = NULL;
1212 	struct buffer_head *old_inode_bh = NULL;
1213 	struct ocfs2_super *osb = NULL;
1214 	u64 newfe_blkno, old_de_ino;
1215 	handle_t *handle = NULL;
1216 	struct buffer_head *old_dir_bh = NULL;
1217 	struct buffer_head *new_dir_bh = NULL;
1218 	u32 old_dir_nlink = old_dir->i_nlink;
1219 	struct ocfs2_dinode *old_di;
1220 	struct ocfs2_dir_lookup_result old_inode_dot_dot_res = { NULL, };
1221 	struct ocfs2_dir_lookup_result target_lookup_res = { NULL, };
1222 	struct ocfs2_dir_lookup_result old_entry_lookup = { NULL, };
1223 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
1224 	struct ocfs2_dir_lookup_result target_insert = { NULL, };
1225 	bool should_add_orphan = false;
1226 
1227 	if (flags)
1228 		return -EINVAL;
1229 
1230 	/* At some point it might be nice to break this function up a
1231 	 * bit. */
1232 
1233 	trace_ocfs2_rename(old_dir, old_dentry, new_dir, new_dentry,
1234 			   old_dentry->d_name.len, old_dentry->d_name.name,
1235 			   new_dentry->d_name.len, new_dentry->d_name.name);
1236 
1237 	status = dquot_initialize(old_dir);
1238 	if (status) {
1239 		mlog_errno(status);
1240 		goto bail;
1241 	}
1242 	status = dquot_initialize(new_dir);
1243 	if (status) {
1244 		mlog_errno(status);
1245 		goto bail;
1246 	}
1247 
1248 	osb = OCFS2_SB(old_dir->i_sb);
1249 
1250 	if (new_inode) {
1251 		if (!igrab(new_inode))
1252 			BUG();
1253 	}
1254 
1255 	/* Assume a directory hierarchy thusly:
1256 	 * a/b/c
1257 	 * a/d
1258 	 * a,b,c, and d are all directories.
1259 	 *
1260 	 * from cwd of 'a' on both nodes:
1261 	 * node1: mv b/c d
1262 	 * node2: mv d   b/c
1263 	 *
1264 	 * And that's why, just like the VFS, we need a file system
1265 	 * rename lock. */
1266 	if (old_dir != new_dir && S_ISDIR(old_inode->i_mode)) {
1267 		status = ocfs2_rename_lock(osb);
1268 		if (status < 0) {
1269 			mlog_errno(status);
1270 			goto bail;
1271 		}
1272 		rename_lock = 1;
1273 
1274 		/* here we cannot guarantee the inodes haven't just been
1275 		 * changed, so check if they are nested again */
1276 		status = ocfs2_check_if_ancestor(osb, new_dir->i_ino,
1277 				old_inode->i_ino);
1278 		if (status < 0) {
1279 			mlog_errno(status);
1280 			goto bail;
1281 		} else if (status == 1) {
1282 			status = -EPERM;
1283 			trace_ocfs2_rename_not_permitted(
1284 					(unsigned long long)old_inode->i_ino,
1285 					(unsigned long long)new_dir->i_ino);
1286 			goto bail;
1287 		}
1288 	}
1289 
1290 	/* if old and new are the same, this'll just do one lock. */
1291 	status = ocfs2_double_lock(osb, &old_dir_bh, old_dir,
1292 				   &new_dir_bh, new_dir, 1);
1293 	if (status < 0) {
1294 		mlog_errno(status);
1295 		goto bail;
1296 	}
1297 	parents_locked = 1;
1298 
1299 	if (!new_dir->i_nlink) {
1300 		status = -EACCES;
1301 		goto bail;
1302 	}
1303 
1304 	/* make sure both dirs have bhs
1305 	 * get an extra ref on old_dir_bh if old==new */
1306 	if (!new_dir_bh) {
1307 		if (old_dir_bh) {
1308 			new_dir_bh = old_dir_bh;
1309 			get_bh(new_dir_bh);
1310 		} else {
1311 			mlog(ML_ERROR, "no old_dir_bh!\n");
1312 			status = -EIO;
1313 			goto bail;
1314 		}
1315 	}
1316 
1317 	/*
1318 	 * Aside from allowing a meta data update, the locking here
1319 	 * also ensures that the downconvert thread on other nodes
1320 	 * won't have to concurrently downconvert the inode and the
1321 	 * dentry locks.
1322 	 */
1323 	status = ocfs2_inode_lock_nested(old_inode, &old_inode_bh, 1,
1324 					 OI_LS_PARENT);
1325 	if (status < 0) {
1326 		if (status != -ENOENT)
1327 			mlog_errno(status);
1328 		goto bail;
1329 	}
1330 	old_child_locked = 1;
1331 
1332 	status = ocfs2_remote_dentry_delete(old_dentry);
1333 	if (status < 0) {
1334 		mlog_errno(status);
1335 		goto bail;
1336 	}
1337 
1338 	if (S_ISDIR(old_inode->i_mode)) {
1339 		u64 old_inode_parent;
1340 
1341 		update_dot_dot = 1;
1342 		status = ocfs2_find_files_on_disk("..", 2, &old_inode_parent,
1343 						  old_inode,
1344 						  &old_inode_dot_dot_res);
1345 		if (status) {
1346 			status = -EIO;
1347 			goto bail;
1348 		}
1349 
1350 		if (old_inode_parent != OCFS2_I(old_dir)->ip_blkno) {
1351 			status = -EIO;
1352 			goto bail;
1353 		}
1354 
1355 		if (!new_inode && new_dir != old_dir &&
1356 		    new_dir->i_nlink >= ocfs2_link_max(osb)) {
1357 			status = -EMLINK;
1358 			goto bail;
1359 		}
1360 	}
1361 
1362 	status = ocfs2_lookup_ino_from_name(old_dir, old_dentry->d_name.name,
1363 					    old_dentry->d_name.len,
1364 					    &old_de_ino);
1365 	if (status) {
1366 		status = -ENOENT;
1367 		goto bail;
1368 	}
1369 
1370 	/*
1371 	 *  Check for inode number is _not_ due to possible IO errors.
1372 	 *  We might rmdir the source, keep it as pwd of some process
1373 	 *  and merrily kill the link to whatever was created under the
1374 	 *  same name. Goodbye sticky bit ;-<
1375 	 */
1376 	if (old_de_ino != OCFS2_I(old_inode)->ip_blkno) {
1377 		status = -ENOENT;
1378 		goto bail;
1379 	}
1380 
1381 	/* check if the target already exists (in which case we need
1382 	 * to delete it */
1383 	status = ocfs2_find_files_on_disk(new_dentry->d_name.name,
1384 					  new_dentry->d_name.len,
1385 					  &newfe_blkno, new_dir,
1386 					  &target_lookup_res);
1387 	/* The only error we allow here is -ENOENT because the new
1388 	 * file not existing is perfectly valid. */
1389 	if ((status < 0) && (status != -ENOENT)) {
1390 		/* If we cannot find the file specified we should just */
1391 		/* return the error... */
1392 		mlog_errno(status);
1393 		goto bail;
1394 	}
1395 	if (status == 0)
1396 		target_exists = 1;
1397 
1398 	if (!target_exists && new_inode) {
1399 		/*
1400 		 * Target was unlinked by another node while we were
1401 		 * waiting to get to ocfs2_rename(). There isn't
1402 		 * anything we can do here to help the situation, so
1403 		 * bubble up the appropriate error.
1404 		 */
1405 		status = -ENOENT;
1406 		goto bail;
1407 	}
1408 
1409 	/* In case we need to overwrite an existing file, we blow it
1410 	 * away first */
1411 	if (target_exists) {
1412 		/* VFS didn't think there existed an inode here, but
1413 		 * someone else in the cluster must have raced our
1414 		 * rename to create one. Today we error cleanly, in
1415 		 * the future we should consider calling iget to build
1416 		 * a new struct inode for this entry. */
1417 		if (!new_inode) {
1418 			status = -EACCES;
1419 
1420 			trace_ocfs2_rename_target_exists(new_dentry->d_name.len,
1421 						new_dentry->d_name.name);
1422 			goto bail;
1423 		}
1424 
1425 		if (OCFS2_I(new_inode)->ip_blkno != newfe_blkno) {
1426 			status = -EACCES;
1427 
1428 			trace_ocfs2_rename_disagree(
1429 			     (unsigned long long)OCFS2_I(new_inode)->ip_blkno,
1430 			     (unsigned long long)newfe_blkno,
1431 			     OCFS2_I(new_inode)->ip_flags);
1432 			goto bail;
1433 		}
1434 
1435 		status = ocfs2_inode_lock(new_inode, &newfe_bh, 1);
1436 		if (status < 0) {
1437 			if (status != -ENOENT)
1438 				mlog_errno(status);
1439 			goto bail;
1440 		}
1441 		new_child_locked = 1;
1442 
1443 		status = ocfs2_remote_dentry_delete(new_dentry);
1444 		if (status < 0) {
1445 			mlog_errno(status);
1446 			goto bail;
1447 		}
1448 
1449 		newfe = (struct ocfs2_dinode *) newfe_bh->b_data;
1450 
1451 		trace_ocfs2_rename_over_existing(
1452 		     (unsigned long long)newfe_blkno, newfe_bh, newfe_bh ?
1453 		     (unsigned long long)newfe_bh->b_blocknr : 0ULL);
1454 
1455 		if (S_ISDIR(new_inode->i_mode) || (new_inode->i_nlink == 1)) {
1456 			status = ocfs2_prepare_orphan_dir(osb, &orphan_dir,
1457 						OCFS2_I(new_inode)->ip_blkno,
1458 						orphan_name, &orphan_insert,
1459 						false);
1460 			if (status < 0) {
1461 				mlog_errno(status);
1462 				goto bail;
1463 			}
1464 			should_add_orphan = true;
1465 		}
1466 	} else {
1467 		BUG_ON(d_inode(new_dentry->d_parent) != new_dir);
1468 
1469 		status = ocfs2_check_dir_for_entry(new_dir,
1470 						   new_dentry->d_name.name,
1471 						   new_dentry->d_name.len);
1472 		if (status)
1473 			goto bail;
1474 
1475 		status = ocfs2_prepare_dir_for_insert(osb, new_dir, new_dir_bh,
1476 						      new_dentry->d_name.name,
1477 						      new_dentry->d_name.len,
1478 						      &target_insert);
1479 		if (status < 0) {
1480 			mlog_errno(status);
1481 			goto bail;
1482 		}
1483 	}
1484 
1485 	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
1486 	if (IS_ERR(handle)) {
1487 		status = PTR_ERR(handle);
1488 		handle = NULL;
1489 		mlog_errno(status);
1490 		goto bail;
1491 	}
1492 
1493 	if (target_exists) {
1494 		if (S_ISDIR(new_inode->i_mode)) {
1495 			if (new_inode->i_nlink != 2 ||
1496 			    !ocfs2_empty_dir(new_inode)) {
1497 				status = -ENOTEMPTY;
1498 				goto bail;
1499 			}
1500 		}
1501 		status = ocfs2_journal_access_di(handle, INODE_CACHE(new_inode),
1502 						 newfe_bh,
1503 						 OCFS2_JOURNAL_ACCESS_WRITE);
1504 		if (status < 0) {
1505 			mlog_errno(status);
1506 			goto bail;
1507 		}
1508 
1509 		/* change the dirent to point to the correct inode */
1510 		status = ocfs2_update_entry(new_dir, handle, &target_lookup_res,
1511 					    old_inode);
1512 		if (status < 0) {
1513 			mlog_errno(status);
1514 			goto bail;
1515 		}
1516 		inode_inc_iversion(new_dir);
1517 
1518 		if (S_ISDIR(new_inode->i_mode))
1519 			ocfs2_set_links_count(newfe, 0);
1520 		else
1521 			ocfs2_add_links_count(newfe, -1);
1522 		ocfs2_journal_dirty(handle, newfe_bh);
1523 		if (should_add_orphan) {
1524 			status = ocfs2_orphan_add(osb, handle, new_inode,
1525 					newfe_bh, orphan_name,
1526 					&orphan_insert, orphan_dir, false);
1527 			if (status < 0) {
1528 				mlog_errno(status);
1529 				goto bail;
1530 			}
1531 		}
1532 	} else {
1533 		/* if the name was not found in new_dir, add it now */
1534 		status = ocfs2_add_entry(handle, new_dentry, old_inode,
1535 					 OCFS2_I(old_inode)->ip_blkno,
1536 					 new_dir_bh, &target_insert);
1537 	}
1538 
1539 	old_inode->i_ctime = current_time(old_inode);
1540 	mark_inode_dirty(old_inode);
1541 
1542 	status = ocfs2_journal_access_di(handle, INODE_CACHE(old_inode),
1543 					 old_inode_bh,
1544 					 OCFS2_JOURNAL_ACCESS_WRITE);
1545 	if (status >= 0) {
1546 		old_di = (struct ocfs2_dinode *) old_inode_bh->b_data;
1547 
1548 		old_di->i_ctime = cpu_to_le64(old_inode->i_ctime.tv_sec);
1549 		old_di->i_ctime_nsec = cpu_to_le32(old_inode->i_ctime.tv_nsec);
1550 		ocfs2_journal_dirty(handle, old_inode_bh);
1551 	} else
1552 		mlog_errno(status);
1553 
1554 	/*
1555 	 * Now that the name has been added to new_dir, remove the old name.
1556 	 *
1557 	 * We don't keep any directory entry context around until now
1558 	 * because the insert might have changed the type of directory
1559 	 * we're dealing with.
1560 	 */
1561 	status = ocfs2_find_entry(old_dentry->d_name.name,
1562 				  old_dentry->d_name.len, old_dir,
1563 				  &old_entry_lookup);
1564 	if (status) {
1565 		if (!is_journal_aborted(osb->journal->j_journal)) {
1566 			ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1567 					"is not deleted.",
1568 					new_dentry->d_name.len, new_dentry->d_name.name,
1569 					old_dentry->d_name.len, old_dentry->d_name.name);
1570 		}
1571 		goto bail;
1572 	}
1573 
1574 	status = ocfs2_delete_entry(handle, old_dir, &old_entry_lookup);
1575 	if (status < 0) {
1576 		mlog_errno(status);
1577 		if (!is_journal_aborted(osb->journal->j_journal)) {
1578 			ocfs2_error(osb->sb, "new entry %.*s is added, but old entry %.*s "
1579 					"is not deleted.",
1580 					new_dentry->d_name.len, new_dentry->d_name.name,
1581 					old_dentry->d_name.len, old_dentry->d_name.name);
1582 		}
1583 		goto bail;
1584 	}
1585 
1586 	if (new_inode) {
1587 		drop_nlink(new_inode);
1588 		new_inode->i_ctime = current_time(new_inode);
1589 	}
1590 	old_dir->i_ctime = old_dir->i_mtime = current_time(old_dir);
1591 
1592 	if (update_dot_dot) {
1593 		status = ocfs2_update_entry(old_inode, handle,
1594 					    &old_inode_dot_dot_res, new_dir);
1595 		drop_nlink(old_dir);
1596 		if (new_inode) {
1597 			drop_nlink(new_inode);
1598 		} else {
1599 			inc_nlink(new_dir);
1600 			mark_inode_dirty(new_dir);
1601 		}
1602 	}
1603 	mark_inode_dirty(old_dir);
1604 	ocfs2_mark_inode_dirty(handle, old_dir, old_dir_bh);
1605 	if (new_inode) {
1606 		mark_inode_dirty(new_inode);
1607 		ocfs2_mark_inode_dirty(handle, new_inode, newfe_bh);
1608 	}
1609 
1610 	if (old_dir != new_dir) {
1611 		/* Keep the same times on both directories.*/
1612 		new_dir->i_ctime = new_dir->i_mtime = old_dir->i_ctime;
1613 
1614 		/*
1615 		 * This will also pick up the i_nlink change from the
1616 		 * block above.
1617 		 */
1618 		ocfs2_mark_inode_dirty(handle, new_dir, new_dir_bh);
1619 	}
1620 
1621 	if (old_dir_nlink != old_dir->i_nlink) {
1622 		if (!old_dir_bh) {
1623 			mlog(ML_ERROR, "need to change nlink for old dir "
1624 			     "%llu from %d to %d but bh is NULL!\n",
1625 			     (unsigned long long)OCFS2_I(old_dir)->ip_blkno,
1626 			     (int)old_dir_nlink, old_dir->i_nlink);
1627 		} else {
1628 			struct ocfs2_dinode *fe;
1629 			status = ocfs2_journal_access_di(handle,
1630 							 INODE_CACHE(old_dir),
1631 							 old_dir_bh,
1632 							 OCFS2_JOURNAL_ACCESS_WRITE);
1633 			fe = (struct ocfs2_dinode *) old_dir_bh->b_data;
1634 			ocfs2_set_links_count(fe, old_dir->i_nlink);
1635 			ocfs2_journal_dirty(handle, old_dir_bh);
1636 		}
1637 	}
1638 	ocfs2_dentry_move(old_dentry, new_dentry, old_dir, new_dir);
1639 	status = 0;
1640 bail:
1641 	if (handle)
1642 		ocfs2_commit_trans(osb, handle);
1643 
1644 	if (orphan_dir) {
1645 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
1646 		ocfs2_inode_unlock(orphan_dir, 1);
1647 		inode_unlock(orphan_dir);
1648 		iput(orphan_dir);
1649 	}
1650 
1651 	if (new_child_locked)
1652 		ocfs2_inode_unlock(new_inode, 1);
1653 
1654 	if (old_child_locked)
1655 		ocfs2_inode_unlock(old_inode, 1);
1656 
1657 	if (parents_locked)
1658 		ocfs2_double_unlock(old_dir, new_dir);
1659 
1660 	if (rename_lock)
1661 		ocfs2_rename_unlock(osb);
1662 
1663 	if (new_inode)
1664 		sync_mapping_buffers(old_inode->i_mapping);
1665 
1666 	iput(new_inode);
1667 
1668 	ocfs2_free_dir_lookup_result(&target_lookup_res);
1669 	ocfs2_free_dir_lookup_result(&old_entry_lookup);
1670 	ocfs2_free_dir_lookup_result(&old_inode_dot_dot_res);
1671 	ocfs2_free_dir_lookup_result(&orphan_insert);
1672 	ocfs2_free_dir_lookup_result(&target_insert);
1673 
1674 	brelse(newfe_bh);
1675 	brelse(old_inode_bh);
1676 	brelse(old_dir_bh);
1677 	brelse(new_dir_bh);
1678 
1679 	if (status)
1680 		mlog_errno(status);
1681 
1682 	return status;
1683 }
1684 
1685 /*
1686  * we expect i_size = strlen(symname). Copy symname into the file
1687  * data, including the null terminator.
1688  */
1689 static int ocfs2_create_symlink_data(struct ocfs2_super *osb,
1690 				     handle_t *handle,
1691 				     struct inode *inode,
1692 				     const char *symname)
1693 {
1694 	struct buffer_head **bhs = NULL;
1695 	const char *c;
1696 	struct super_block *sb = osb->sb;
1697 	u64 p_blkno, p_blocks;
1698 	int virtual, blocks, status, i, bytes_left;
1699 
1700 	bytes_left = i_size_read(inode) + 1;
1701 	/* we can't trust i_blocks because we're actually going to
1702 	 * write i_size + 1 bytes. */
1703 	blocks = (bytes_left + sb->s_blocksize - 1) >> sb->s_blocksize_bits;
1704 
1705 	trace_ocfs2_create_symlink_data((unsigned long long)inode->i_blocks,
1706 					i_size_read(inode), blocks);
1707 
1708 	/* Sanity check -- make sure we're going to fit. */
1709 	if (bytes_left >
1710 	    ocfs2_clusters_to_bytes(sb, OCFS2_I(inode)->ip_clusters)) {
1711 		status = -EIO;
1712 		mlog_errno(status);
1713 		goto bail;
1714 	}
1715 
1716 	bhs = kcalloc(blocks, sizeof(struct buffer_head *), GFP_KERNEL);
1717 	if (!bhs) {
1718 		status = -ENOMEM;
1719 		mlog_errno(status);
1720 		goto bail;
1721 	}
1722 
1723 	status = ocfs2_extent_map_get_blocks(inode, 0, &p_blkno, &p_blocks,
1724 					     NULL);
1725 	if (status < 0) {
1726 		mlog_errno(status);
1727 		goto bail;
1728 	}
1729 
1730 	/* links can never be larger than one cluster so we know this
1731 	 * is all going to be contiguous, but do a sanity check
1732 	 * anyway. */
1733 	if ((p_blocks << sb->s_blocksize_bits) < bytes_left) {
1734 		status = -EIO;
1735 		mlog_errno(status);
1736 		goto bail;
1737 	}
1738 
1739 	virtual = 0;
1740 	while(bytes_left > 0) {
1741 		c = &symname[virtual * sb->s_blocksize];
1742 
1743 		bhs[virtual] = sb_getblk(sb, p_blkno);
1744 		if (!bhs[virtual]) {
1745 			status = -ENOMEM;
1746 			mlog_errno(status);
1747 			goto bail;
1748 		}
1749 		ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode),
1750 					      bhs[virtual]);
1751 
1752 		status = ocfs2_journal_access(handle, INODE_CACHE(inode),
1753 					      bhs[virtual],
1754 					      OCFS2_JOURNAL_ACCESS_CREATE);
1755 		if (status < 0) {
1756 			mlog_errno(status);
1757 			goto bail;
1758 		}
1759 
1760 		memset(bhs[virtual]->b_data, 0, sb->s_blocksize);
1761 
1762 		memcpy(bhs[virtual]->b_data, c,
1763 		       (bytes_left > sb->s_blocksize) ? sb->s_blocksize :
1764 		       bytes_left);
1765 
1766 		ocfs2_journal_dirty(handle, bhs[virtual]);
1767 
1768 		virtual++;
1769 		p_blkno++;
1770 		bytes_left -= sb->s_blocksize;
1771 	}
1772 
1773 	status = 0;
1774 bail:
1775 
1776 	if (bhs) {
1777 		for(i = 0; i < blocks; i++)
1778 			brelse(bhs[i]);
1779 		kfree(bhs);
1780 	}
1781 
1782 	if (status)
1783 		mlog_errno(status);
1784 	return status;
1785 }
1786 
1787 static int ocfs2_symlink(struct inode *dir,
1788 			 struct dentry *dentry,
1789 			 const char *symname)
1790 {
1791 	int status, l, credits;
1792 	u64 newsize;
1793 	struct ocfs2_super *osb = NULL;
1794 	struct inode *inode = NULL;
1795 	struct super_block *sb;
1796 	struct buffer_head *new_fe_bh = NULL;
1797 	struct buffer_head *parent_fe_bh = NULL;
1798 	struct ocfs2_dinode *fe = NULL;
1799 	struct ocfs2_dinode *dirfe;
1800 	handle_t *handle = NULL;
1801 	struct ocfs2_alloc_context *inode_ac = NULL;
1802 	struct ocfs2_alloc_context *data_ac = NULL;
1803 	struct ocfs2_alloc_context *xattr_ac = NULL;
1804 	int want_clusters = 0;
1805 	int xattr_credits = 0;
1806 	struct ocfs2_security_xattr_info si = {
1807 		.enable = 1,
1808 	};
1809 	int did_quota = 0, did_quota_inode = 0;
1810 	struct ocfs2_dir_lookup_result lookup = { NULL, };
1811 	sigset_t oldset;
1812 	int did_block_signals = 0;
1813 	struct ocfs2_dentry_lock *dl = NULL;
1814 
1815 	trace_ocfs2_symlink_begin(dir, dentry, symname,
1816 				  dentry->d_name.len, dentry->d_name.name);
1817 
1818 	status = dquot_initialize(dir);
1819 	if (status) {
1820 		mlog_errno(status);
1821 		goto bail;
1822 	}
1823 
1824 	sb = dir->i_sb;
1825 	osb = OCFS2_SB(sb);
1826 
1827 	l = strlen(symname) + 1;
1828 
1829 	credits = ocfs2_calc_symlink_credits(sb);
1830 
1831 	/* lock the parent directory */
1832 	status = ocfs2_inode_lock(dir, &parent_fe_bh, 1);
1833 	if (status < 0) {
1834 		if (status != -ENOENT)
1835 			mlog_errno(status);
1836 		return status;
1837 	}
1838 
1839 	dirfe = (struct ocfs2_dinode *) parent_fe_bh->b_data;
1840 	if (!ocfs2_read_links_count(dirfe)) {
1841 		/* can't make a file in a deleted directory. */
1842 		status = -ENOENT;
1843 		goto bail;
1844 	}
1845 
1846 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
1847 					   dentry->d_name.len);
1848 	if (status)
1849 		goto bail;
1850 
1851 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_fe_bh,
1852 					      dentry->d_name.name,
1853 					      dentry->d_name.len, &lookup);
1854 	if (status < 0) {
1855 		mlog_errno(status);
1856 		goto bail;
1857 	}
1858 
1859 	status = ocfs2_reserve_new_inode(osb, &inode_ac);
1860 	if (status < 0) {
1861 		if (status != -ENOSPC)
1862 			mlog_errno(status);
1863 		goto bail;
1864 	}
1865 
1866 	inode = ocfs2_get_init_inode(dir, S_IFLNK | S_IRWXUGO);
1867 	if (IS_ERR(inode)) {
1868 		status = PTR_ERR(inode);
1869 		inode = NULL;
1870 		mlog_errno(status);
1871 		goto bail;
1872 	}
1873 
1874 	/* get security xattr */
1875 	status = ocfs2_init_security_get(inode, dir, &dentry->d_name, &si);
1876 	if (status) {
1877 		if (status == -EOPNOTSUPP)
1878 			si.enable = 0;
1879 		else {
1880 			mlog_errno(status);
1881 			goto bail;
1882 		}
1883 	}
1884 
1885 	/* calculate meta data/clusters for setting security xattr */
1886 	if (si.enable) {
1887 		status = ocfs2_calc_security_init(dir, &si, &want_clusters,
1888 						  &xattr_credits, &xattr_ac);
1889 		if (status < 0) {
1890 			mlog_errno(status);
1891 			goto bail;
1892 		}
1893 	}
1894 
1895 	/* don't reserve bitmap space for fast symlinks. */
1896 	if (l > ocfs2_fast_symlink_chars(sb))
1897 		want_clusters += 1;
1898 
1899 	status = ocfs2_reserve_clusters(osb, want_clusters, &data_ac);
1900 	if (status < 0) {
1901 		if (status != -ENOSPC)
1902 			mlog_errno(status);
1903 		goto bail;
1904 	}
1905 
1906 	handle = ocfs2_start_trans(osb, credits + xattr_credits);
1907 	if (IS_ERR(handle)) {
1908 		status = PTR_ERR(handle);
1909 		handle = NULL;
1910 		mlog_errno(status);
1911 		goto bail;
1912 	}
1913 
1914 	/* Starting to change things, restart is no longer possible. */
1915 	ocfs2_block_signals(&oldset);
1916 	did_block_signals = 1;
1917 
1918 	status = dquot_alloc_inode(inode);
1919 	if (status)
1920 		goto bail;
1921 	did_quota_inode = 1;
1922 
1923 	trace_ocfs2_symlink_create(dir, dentry, dentry->d_name.len,
1924 				   dentry->d_name.name,
1925 				   (unsigned long long)OCFS2_I(dir)->ip_blkno,
1926 				   inode->i_mode);
1927 
1928 	status = ocfs2_mknod_locked(osb, dir, inode,
1929 				    0, &new_fe_bh, parent_fe_bh, handle,
1930 				    inode_ac);
1931 	if (status < 0) {
1932 		mlog_errno(status);
1933 		goto bail;
1934 	}
1935 
1936 	fe = (struct ocfs2_dinode *) new_fe_bh->b_data;
1937 	inode->i_rdev = 0;
1938 	newsize = l - 1;
1939 	inode->i_op = &ocfs2_symlink_inode_operations;
1940 	inode_nohighmem(inode);
1941 	if (l > ocfs2_fast_symlink_chars(sb)) {
1942 		u32 offset = 0;
1943 
1944 		status = dquot_alloc_space_nodirty(inode,
1945 		    ocfs2_clusters_to_bytes(osb->sb, 1));
1946 		if (status)
1947 			goto bail;
1948 		did_quota = 1;
1949 		inode->i_mapping->a_ops = &ocfs2_aops;
1950 		status = ocfs2_add_inode_data(osb, inode, &offset, 1, 0,
1951 					      new_fe_bh,
1952 					      handle, data_ac, NULL,
1953 					      NULL);
1954 		if (status < 0) {
1955 			if (status != -ENOSPC && status != -EINTR) {
1956 				mlog(ML_ERROR,
1957 				     "Failed to extend file to %llu\n",
1958 				     (unsigned long long)newsize);
1959 				mlog_errno(status);
1960 				status = -ENOSPC;
1961 			}
1962 			goto bail;
1963 		}
1964 		i_size_write(inode, newsize);
1965 		inode->i_blocks = ocfs2_inode_sector_count(inode);
1966 	} else {
1967 		inode->i_mapping->a_ops = &ocfs2_fast_symlink_aops;
1968 		memcpy((char *) fe->id2.i_symlink, symname, l);
1969 		i_size_write(inode, newsize);
1970 		inode->i_blocks = 0;
1971 	}
1972 
1973 	status = ocfs2_mark_inode_dirty(handle, inode, new_fe_bh);
1974 	if (status < 0) {
1975 		mlog_errno(status);
1976 		goto bail;
1977 	}
1978 
1979 	if (!ocfs2_inode_is_fast_symlink(inode)) {
1980 		status = ocfs2_create_symlink_data(osb, handle, inode,
1981 						   symname);
1982 		if (status < 0) {
1983 			mlog_errno(status);
1984 			goto bail;
1985 		}
1986 	}
1987 
1988 	if (si.enable) {
1989 		status = ocfs2_init_security_set(handle, inode, new_fe_bh, &si,
1990 						 xattr_ac, data_ac);
1991 		if (status < 0) {
1992 			mlog_errno(status);
1993 			goto bail;
1994 		}
1995 	}
1996 
1997 	/*
1998 	 * Do this before adding the entry to the directory. We add
1999 	 * also set d_op after success so that ->d_iput() will cleanup
2000 	 * the dentry lock even if ocfs2_add_entry() fails below.
2001 	 */
2002 	status = ocfs2_dentry_attach_lock(dentry, inode, OCFS2_I(dir)->ip_blkno);
2003 	if (status) {
2004 		mlog_errno(status);
2005 		goto bail;
2006 	}
2007 
2008 	dl = dentry->d_fsdata;
2009 
2010 	status = ocfs2_add_entry(handle, dentry, inode,
2011 				 le64_to_cpu(fe->i_blkno), parent_fe_bh,
2012 				 &lookup);
2013 	if (status < 0) {
2014 		mlog_errno(status);
2015 		goto bail;
2016 	}
2017 
2018 	insert_inode_hash(inode);
2019 	d_instantiate(dentry, inode);
2020 bail:
2021 	if (status < 0 && did_quota)
2022 		dquot_free_space_nodirty(inode,
2023 					ocfs2_clusters_to_bytes(osb->sb, 1));
2024 	if (status < 0 && did_quota_inode)
2025 		dquot_free_inode(inode);
2026 	if (handle)
2027 		ocfs2_commit_trans(osb, handle);
2028 
2029 	ocfs2_inode_unlock(dir, 1);
2030 	if (did_block_signals)
2031 		ocfs2_unblock_signals(&oldset);
2032 
2033 	brelse(new_fe_bh);
2034 	brelse(parent_fe_bh);
2035 	kfree(si.value);
2036 	ocfs2_free_dir_lookup_result(&lookup);
2037 	if (inode_ac)
2038 		ocfs2_free_alloc_context(inode_ac);
2039 	if (data_ac)
2040 		ocfs2_free_alloc_context(data_ac);
2041 	if (xattr_ac)
2042 		ocfs2_free_alloc_context(xattr_ac);
2043 	if ((status < 0) && inode) {
2044 		if (dl)
2045 			ocfs2_cleanup_add_entry_failure(osb, dentry, inode);
2046 
2047 		OCFS2_I(inode)->ip_flags |= OCFS2_INODE_SKIP_ORPHAN_DIR;
2048 		clear_nlink(inode);
2049 		iput(inode);
2050 	}
2051 
2052 	if (status)
2053 		mlog_errno(status);
2054 
2055 	return status;
2056 }
2057 
2058 static int ocfs2_blkno_stringify(u64 blkno, char *name)
2059 {
2060 	int status, namelen;
2061 
2062 	namelen = snprintf(name, OCFS2_ORPHAN_NAMELEN + 1, "%016llx",
2063 			   (long long)blkno);
2064 	if (namelen <= 0) {
2065 		if (namelen)
2066 			status = namelen;
2067 		else
2068 			status = -EINVAL;
2069 		mlog_errno(status);
2070 		goto bail;
2071 	}
2072 	if (namelen != OCFS2_ORPHAN_NAMELEN) {
2073 		status = -EINVAL;
2074 		mlog_errno(status);
2075 		goto bail;
2076 	}
2077 
2078 	trace_ocfs2_blkno_stringify(blkno, name, namelen);
2079 
2080 	status = 0;
2081 bail:
2082 	if (status < 0)
2083 		mlog_errno(status);
2084 	return status;
2085 }
2086 
2087 static int ocfs2_lookup_lock_orphan_dir(struct ocfs2_super *osb,
2088 					struct inode **ret_orphan_dir,
2089 					struct buffer_head **ret_orphan_dir_bh)
2090 {
2091 	struct inode *orphan_dir_inode;
2092 	struct buffer_head *orphan_dir_bh = NULL;
2093 	int ret = 0;
2094 
2095 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2096 						       ORPHAN_DIR_SYSTEM_INODE,
2097 						       osb->slot_num);
2098 	if (!orphan_dir_inode) {
2099 		ret = -ENOENT;
2100 		mlog_errno(ret);
2101 		return ret;
2102 	}
2103 
2104 	inode_lock(orphan_dir_inode);
2105 
2106 	ret = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2107 	if (ret < 0) {
2108 		inode_unlock(orphan_dir_inode);
2109 		iput(orphan_dir_inode);
2110 
2111 		mlog_errno(ret);
2112 		return ret;
2113 	}
2114 
2115 	*ret_orphan_dir = orphan_dir_inode;
2116 	*ret_orphan_dir_bh = orphan_dir_bh;
2117 
2118 	return 0;
2119 }
2120 
2121 static int __ocfs2_prepare_orphan_dir(struct inode *orphan_dir_inode,
2122 				      struct buffer_head *orphan_dir_bh,
2123 				      u64 blkno,
2124 				      char *name,
2125 				      struct ocfs2_dir_lookup_result *lookup,
2126 				      bool dio)
2127 {
2128 	int ret;
2129 	struct ocfs2_super *osb = OCFS2_SB(orphan_dir_inode->i_sb);
2130 	int namelen = dio ?
2131 			(OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2132 			OCFS2_ORPHAN_NAMELEN;
2133 
2134 	if (dio) {
2135 		ret = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2136 				OCFS2_DIO_ORPHAN_PREFIX);
2137 		if (ret != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2138 			ret = -EINVAL;
2139 			mlog_errno(ret);
2140 			return ret;
2141 		}
2142 
2143 		ret = ocfs2_blkno_stringify(blkno,
2144 				name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2145 	} else
2146 		ret = ocfs2_blkno_stringify(blkno, name);
2147 	if (ret < 0) {
2148 		mlog_errno(ret);
2149 		return ret;
2150 	}
2151 
2152 	ret = ocfs2_prepare_dir_for_insert(osb, orphan_dir_inode,
2153 					   orphan_dir_bh, name,
2154 					   namelen, lookup);
2155 	if (ret < 0) {
2156 		mlog_errno(ret);
2157 		return ret;
2158 	}
2159 
2160 	return 0;
2161 }
2162 
2163 /**
2164  * ocfs2_prepare_orphan_dir() - Prepare an orphan directory for
2165  * insertion of an orphan.
2166  * @osb: ocfs2 file system
2167  * @ret_orphan_dir: Orphan dir inode - returned locked!
2168  * @blkno: Actual block number of the inode to be inserted into orphan dir.
2169  * @lookup: dir lookup result, to be passed back into functions like
2170  *          ocfs2_orphan_add
2171  *
2172  * Returns zero on success and the ret_orphan_dir, name and lookup
2173  * fields will be populated.
2174  *
2175  * Returns non-zero on failure.
2176  */
2177 static int ocfs2_prepare_orphan_dir(struct ocfs2_super *osb,
2178 				    struct inode **ret_orphan_dir,
2179 				    u64 blkno,
2180 				    char *name,
2181 				    struct ocfs2_dir_lookup_result *lookup,
2182 				    bool dio)
2183 {
2184 	struct inode *orphan_dir_inode = NULL;
2185 	struct buffer_head *orphan_dir_bh = NULL;
2186 	int ret = 0;
2187 
2188 	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir_inode,
2189 					   &orphan_dir_bh);
2190 	if (ret < 0) {
2191 		mlog_errno(ret);
2192 		return ret;
2193 	}
2194 
2195 	ret = __ocfs2_prepare_orphan_dir(orphan_dir_inode, orphan_dir_bh,
2196 					 blkno, name, lookup, dio);
2197 	if (ret < 0) {
2198 		mlog_errno(ret);
2199 		goto out;
2200 	}
2201 
2202 	*ret_orphan_dir = orphan_dir_inode;
2203 
2204 out:
2205 	brelse(orphan_dir_bh);
2206 
2207 	if (ret) {
2208 		ocfs2_inode_unlock(orphan_dir_inode, 1);
2209 		inode_unlock(orphan_dir_inode);
2210 		iput(orphan_dir_inode);
2211 	}
2212 
2213 	if (ret)
2214 		mlog_errno(ret);
2215 	return ret;
2216 }
2217 
2218 static int ocfs2_orphan_add(struct ocfs2_super *osb,
2219 			    handle_t *handle,
2220 			    struct inode *inode,
2221 			    struct buffer_head *fe_bh,
2222 			    char *name,
2223 			    struct ocfs2_dir_lookup_result *lookup,
2224 			    struct inode *orphan_dir_inode,
2225 			    bool dio)
2226 {
2227 	struct buffer_head *orphan_dir_bh = NULL;
2228 	int status = 0;
2229 	struct ocfs2_dinode *orphan_fe;
2230 	struct ocfs2_dinode *fe = (struct ocfs2_dinode *) fe_bh->b_data;
2231 	int namelen = dio ?
2232 			(OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN) :
2233 			OCFS2_ORPHAN_NAMELEN;
2234 
2235 	trace_ocfs2_orphan_add_begin(
2236 				(unsigned long long)OCFS2_I(inode)->ip_blkno);
2237 
2238 	status = ocfs2_read_inode_block(orphan_dir_inode, &orphan_dir_bh);
2239 	if (status < 0) {
2240 		mlog_errno(status);
2241 		goto leave;
2242 	}
2243 
2244 	status = ocfs2_journal_access_di(handle,
2245 					 INODE_CACHE(orphan_dir_inode),
2246 					 orphan_dir_bh,
2247 					 OCFS2_JOURNAL_ACCESS_WRITE);
2248 	if (status < 0) {
2249 		mlog_errno(status);
2250 		goto leave;
2251 	}
2252 
2253 	/*
2254 	 * We're going to journal the change of i_flags and i_orphaned_slot.
2255 	 * It's safe anyway, though some callers may duplicate the journaling.
2256 	 * Journaling within the func just make the logic look more
2257 	 * straightforward.
2258 	 */
2259 	status = ocfs2_journal_access_di(handle,
2260 					 INODE_CACHE(inode),
2261 					 fe_bh,
2262 					 OCFS2_JOURNAL_ACCESS_WRITE);
2263 	if (status < 0) {
2264 		mlog_errno(status);
2265 		goto leave;
2266 	}
2267 
2268 	/* we're a cluster, and nlink can change on disk from
2269 	 * underneath us... */
2270 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2271 	if (S_ISDIR(inode->i_mode))
2272 		ocfs2_add_links_count(orphan_fe, 1);
2273 	set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2274 	ocfs2_journal_dirty(handle, orphan_dir_bh);
2275 
2276 	status = __ocfs2_add_entry(handle, orphan_dir_inode, name,
2277 				   namelen, inode,
2278 				   OCFS2_I(inode)->ip_blkno,
2279 				   orphan_dir_bh, lookup);
2280 	if (status < 0) {
2281 		mlog_errno(status);
2282 		goto rollback;
2283 	}
2284 
2285 	if (dio) {
2286 		/* Update flag OCFS2_DIO_ORPHANED_FL and record the orphan
2287 		 * slot.
2288 		 */
2289 		fe->i_flags |= cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2290 		fe->i_dio_orphaned_slot = cpu_to_le16(osb->slot_num);
2291 	} else {
2292 		fe->i_flags |= cpu_to_le32(OCFS2_ORPHANED_FL);
2293 		OCFS2_I(inode)->ip_flags &= ~OCFS2_INODE_SKIP_ORPHAN_DIR;
2294 
2295 		/* Record which orphan dir our inode now resides
2296 		 * in. delete_inode will use this to determine which orphan
2297 		 * dir to lock. */
2298 		fe->i_orphaned_slot = cpu_to_le16(osb->slot_num);
2299 	}
2300 
2301 	ocfs2_journal_dirty(handle, fe_bh);
2302 
2303 	trace_ocfs2_orphan_add_end((unsigned long long)OCFS2_I(inode)->ip_blkno,
2304 				   osb->slot_num);
2305 
2306 rollback:
2307 	if (status < 0) {
2308 		if (S_ISDIR(inode->i_mode))
2309 			ocfs2_add_links_count(orphan_fe, -1);
2310 		set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2311 	}
2312 
2313 leave:
2314 	brelse(orphan_dir_bh);
2315 
2316 	return status;
2317 }
2318 
2319 /* unlike orphan_add, we expect the orphan dir to already be locked here. */
2320 int ocfs2_orphan_del(struct ocfs2_super *osb,
2321 		     handle_t *handle,
2322 		     struct inode *orphan_dir_inode,
2323 		     struct inode *inode,
2324 		     struct buffer_head *orphan_dir_bh,
2325 		     bool dio)
2326 {
2327 	char name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2328 	struct ocfs2_dinode *orphan_fe;
2329 	int status = 0;
2330 	struct ocfs2_dir_lookup_result lookup = { NULL, };
2331 
2332 	if (dio) {
2333 		status = snprintf(name, OCFS2_DIO_ORPHAN_PREFIX_LEN + 1, "%s",
2334 				OCFS2_DIO_ORPHAN_PREFIX);
2335 		if (status != OCFS2_DIO_ORPHAN_PREFIX_LEN) {
2336 			status = -EINVAL;
2337 			mlog_errno(status);
2338 			return status;
2339 		}
2340 
2341 		status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno,
2342 				name + OCFS2_DIO_ORPHAN_PREFIX_LEN);
2343 	} else
2344 		status = ocfs2_blkno_stringify(OCFS2_I(inode)->ip_blkno, name);
2345 	if (status < 0) {
2346 		mlog_errno(status);
2347 		goto leave;
2348 	}
2349 
2350 	trace_ocfs2_orphan_del(
2351 	     (unsigned long long)OCFS2_I(orphan_dir_inode)->ip_blkno,
2352 	     name, strlen(name));
2353 
2354 	status = ocfs2_journal_access_di(handle,
2355 					 INODE_CACHE(orphan_dir_inode),
2356 					 orphan_dir_bh,
2357 					 OCFS2_JOURNAL_ACCESS_WRITE);
2358 	if (status < 0) {
2359 		mlog_errno(status);
2360 		goto leave;
2361 	}
2362 
2363 	/* find it's spot in the orphan directory */
2364 	status = ocfs2_find_entry(name, strlen(name), orphan_dir_inode,
2365 				  &lookup);
2366 	if (status) {
2367 		mlog_errno(status);
2368 		goto leave;
2369 	}
2370 
2371 	/* remove it from the orphan directory */
2372 	status = ocfs2_delete_entry(handle, orphan_dir_inode, &lookup);
2373 	if (status < 0) {
2374 		mlog_errno(status);
2375 		goto leave;
2376 	}
2377 
2378 	/* do the i_nlink dance! :) */
2379 	orphan_fe = (struct ocfs2_dinode *) orphan_dir_bh->b_data;
2380 	if (S_ISDIR(inode->i_mode))
2381 		ocfs2_add_links_count(orphan_fe, -1);
2382 	set_nlink(orphan_dir_inode, ocfs2_read_links_count(orphan_fe));
2383 	ocfs2_journal_dirty(handle, orphan_dir_bh);
2384 
2385 leave:
2386 	ocfs2_free_dir_lookup_result(&lookup);
2387 
2388 	if (status)
2389 		mlog_errno(status);
2390 	return status;
2391 }
2392 
2393 /**
2394  * ocfs2_prep_new_orphaned_file() - Prepare the orphan dir to receive a newly
2395  * allocated file. This is different from the typical 'add to orphan dir'
2396  * operation in that the inode does not yet exist. This is a problem because
2397  * the orphan dir stringifies the inode block number to come up with it's
2398  * dirent. Obviously if the inode does not yet exist we have a chicken and egg
2399  * problem. This function works around it by calling deeper into the orphan
2400  * and suballoc code than other callers. Use this only by necessity.
2401  * @dir: The directory which this inode will ultimately wind up under - not the
2402  * orphan dir!
2403  * @dir_bh: buffer_head the @dir inode block
2404  * @orphan_name: string of length (CFS2_ORPHAN_NAMELEN + 1). Will be filled
2405  * with the string to be used for orphan dirent. Pass back to the orphan dir
2406  * code.
2407  * @ret_orphan_dir: orphan dir inode returned to be passed back into orphan
2408  * dir code.
2409  * @ret_di_blkno: block number where the new inode will be allocated.
2410  * @orphan_insert: Dir insert context to be passed back into orphan dir code.
2411  * @ret_inode_ac: Inode alloc context to be passed back to the allocator.
2412  *
2413  * Returns zero on success and the ret_orphan_dir, name and lookup
2414  * fields will be populated.
2415  *
2416  * Returns non-zero on failure.
2417  */
2418 static int ocfs2_prep_new_orphaned_file(struct inode *dir,
2419 					struct buffer_head *dir_bh,
2420 					char *orphan_name,
2421 					struct inode **ret_orphan_dir,
2422 					u64 *ret_di_blkno,
2423 					struct ocfs2_dir_lookup_result *orphan_insert,
2424 					struct ocfs2_alloc_context **ret_inode_ac)
2425 {
2426 	int ret;
2427 	u64 di_blkno;
2428 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2429 	struct inode *orphan_dir = NULL;
2430 	struct buffer_head *orphan_dir_bh = NULL;
2431 	struct ocfs2_alloc_context *inode_ac = NULL;
2432 
2433 	ret = ocfs2_lookup_lock_orphan_dir(osb, &orphan_dir, &orphan_dir_bh);
2434 	if (ret < 0) {
2435 		mlog_errno(ret);
2436 		return ret;
2437 	}
2438 
2439 	/* reserve an inode spot */
2440 	ret = ocfs2_reserve_new_inode(osb, &inode_ac);
2441 	if (ret < 0) {
2442 		if (ret != -ENOSPC)
2443 			mlog_errno(ret);
2444 		goto out;
2445 	}
2446 
2447 	ret = ocfs2_find_new_inode_loc(dir, dir_bh, inode_ac,
2448 				       &di_blkno);
2449 	if (ret) {
2450 		mlog_errno(ret);
2451 		goto out;
2452 	}
2453 
2454 	ret = __ocfs2_prepare_orphan_dir(orphan_dir, orphan_dir_bh,
2455 					 di_blkno, orphan_name, orphan_insert,
2456 					 false);
2457 	if (ret < 0) {
2458 		mlog_errno(ret);
2459 		goto out;
2460 	}
2461 
2462 out:
2463 	if (ret == 0) {
2464 		*ret_orphan_dir = orphan_dir;
2465 		*ret_di_blkno = di_blkno;
2466 		*ret_inode_ac = inode_ac;
2467 		/*
2468 		 * orphan_name and orphan_insert are already up to
2469 		 * date via prepare_orphan_dir
2470 		 */
2471 	} else {
2472 		/* Unroll reserve_new_inode* */
2473 		if (inode_ac)
2474 			ocfs2_free_alloc_context(inode_ac);
2475 
2476 		/* Unroll orphan dir locking */
2477 		inode_unlock(orphan_dir);
2478 		ocfs2_inode_unlock(orphan_dir, 1);
2479 		iput(orphan_dir);
2480 	}
2481 
2482 	brelse(orphan_dir_bh);
2483 
2484 	return ret;
2485 }
2486 
2487 int ocfs2_create_inode_in_orphan(struct inode *dir,
2488 				 int mode,
2489 				 struct inode **new_inode)
2490 {
2491 	int status, did_quota_inode = 0;
2492 	struct inode *inode = NULL;
2493 	struct inode *orphan_dir = NULL;
2494 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2495 	handle_t *handle = NULL;
2496 	char orphan_name[OCFS2_ORPHAN_NAMELEN + 1];
2497 	struct buffer_head *parent_di_bh = NULL;
2498 	struct buffer_head *new_di_bh = NULL;
2499 	struct ocfs2_alloc_context *inode_ac = NULL;
2500 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2501 	u64 di_blkno, suballoc_loc;
2502 	u16 suballoc_bit;
2503 
2504 	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2505 	if (status < 0) {
2506 		if (status != -ENOENT)
2507 			mlog_errno(status);
2508 		return status;
2509 	}
2510 
2511 	status = ocfs2_prep_new_orphaned_file(dir, parent_di_bh,
2512 					      orphan_name, &orphan_dir,
2513 					      &di_blkno, &orphan_insert, &inode_ac);
2514 	if (status < 0) {
2515 		if (status != -ENOSPC)
2516 			mlog_errno(status);
2517 		goto leave;
2518 	}
2519 
2520 	inode = ocfs2_get_init_inode(dir, mode);
2521 	if (IS_ERR(inode)) {
2522 		status = PTR_ERR(inode);
2523 		inode = NULL;
2524 		mlog_errno(status);
2525 		goto leave;
2526 	}
2527 
2528 	handle = ocfs2_start_trans(osb, ocfs2_mknod_credits(osb->sb, 0, 0));
2529 	if (IS_ERR(handle)) {
2530 		status = PTR_ERR(handle);
2531 		handle = NULL;
2532 		mlog_errno(status);
2533 		goto leave;
2534 	}
2535 
2536 	status = dquot_alloc_inode(inode);
2537 	if (status)
2538 		goto leave;
2539 	did_quota_inode = 1;
2540 
2541 	status = ocfs2_claim_new_inode_at_loc(handle, dir, inode_ac,
2542 					      &suballoc_loc,
2543 					      &suballoc_bit, di_blkno);
2544 	if (status < 0) {
2545 		mlog_errno(status);
2546 		goto leave;
2547 	}
2548 
2549 	clear_nlink(inode);
2550 	/* do the real work now. */
2551 	status = __ocfs2_mknod_locked(dir, inode,
2552 				      0, &new_di_bh, parent_di_bh, handle,
2553 				      inode_ac, di_blkno, suballoc_loc,
2554 				      suballoc_bit);
2555 	if (status < 0) {
2556 		mlog_errno(status);
2557 		goto leave;
2558 	}
2559 
2560 	status = ocfs2_orphan_add(osb, handle, inode, new_di_bh, orphan_name,
2561 				  &orphan_insert, orphan_dir, false);
2562 	if (status < 0) {
2563 		mlog_errno(status);
2564 		goto leave;
2565 	}
2566 
2567 	/* get open lock so that only nodes can't remove it from orphan dir. */
2568 	status = ocfs2_open_lock(inode);
2569 	if (status < 0)
2570 		mlog_errno(status);
2571 
2572 	insert_inode_hash(inode);
2573 leave:
2574 	if (status < 0 && did_quota_inode)
2575 		dquot_free_inode(inode);
2576 	if (handle)
2577 		ocfs2_commit_trans(osb, handle);
2578 
2579 	if (orphan_dir) {
2580 		/* This was locked for us in ocfs2_prepare_orphan_dir() */
2581 		ocfs2_inode_unlock(orphan_dir, 1);
2582 		inode_unlock(orphan_dir);
2583 		iput(orphan_dir);
2584 	}
2585 
2586 	if ((status < 0) && inode) {
2587 		clear_nlink(inode);
2588 		iput(inode);
2589 	}
2590 
2591 	if (inode_ac)
2592 		ocfs2_free_alloc_context(inode_ac);
2593 
2594 	brelse(new_di_bh);
2595 
2596 	if (!status)
2597 		*new_inode = inode;
2598 
2599 	ocfs2_free_dir_lookup_result(&orphan_insert);
2600 
2601 	ocfs2_inode_unlock(dir, 1);
2602 	brelse(parent_di_bh);
2603 	return status;
2604 }
2605 
2606 int ocfs2_add_inode_to_orphan(struct ocfs2_super *osb,
2607 	struct inode *inode)
2608 {
2609 	char orphan_name[OCFS2_DIO_ORPHAN_PREFIX_LEN + OCFS2_ORPHAN_NAMELEN + 1];
2610 	struct inode *orphan_dir_inode = NULL;
2611 	struct ocfs2_dir_lookup_result orphan_insert = { NULL, };
2612 	struct buffer_head *di_bh = NULL;
2613 	int status = 0;
2614 	handle_t *handle = NULL;
2615 	struct ocfs2_dinode *di = NULL;
2616 
2617 	status = ocfs2_inode_lock(inode, &di_bh, 1);
2618 	if (status < 0) {
2619 		mlog_errno(status);
2620 		goto bail;
2621 	}
2622 
2623 	di = (struct ocfs2_dinode *) di_bh->b_data;
2624 	/*
2625 	 * Another append dio crashed?
2626 	 * If so, manually recover it first.
2627 	 */
2628 	if (unlikely(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL))) {
2629 		status = ocfs2_truncate_file(inode, di_bh, i_size_read(inode));
2630 		if (status < 0) {
2631 			if (status != -ENOSPC)
2632 				mlog_errno(status);
2633 			goto bail_unlock_inode;
2634 		}
2635 
2636 		status = ocfs2_del_inode_from_orphan(osb, inode, di_bh, 0, 0);
2637 		if (status < 0) {
2638 			mlog_errno(status);
2639 			goto bail_unlock_inode;
2640 		}
2641 	}
2642 
2643 	status = ocfs2_prepare_orphan_dir(osb, &orphan_dir_inode,
2644 			OCFS2_I(inode)->ip_blkno,
2645 			orphan_name,
2646 			&orphan_insert,
2647 			true);
2648 	if (status < 0) {
2649 		mlog_errno(status);
2650 		goto bail_unlock_inode;
2651 	}
2652 
2653 	handle = ocfs2_start_trans(osb,
2654 			OCFS2_INODE_ADD_TO_ORPHAN_CREDITS);
2655 	if (IS_ERR(handle)) {
2656 		status = PTR_ERR(handle);
2657 		goto bail_unlock_orphan;
2658 	}
2659 
2660 	status = ocfs2_orphan_add(osb, handle, inode, di_bh, orphan_name,
2661 			&orphan_insert, orphan_dir_inode, true);
2662 	if (status)
2663 		mlog_errno(status);
2664 
2665 	ocfs2_commit_trans(osb, handle);
2666 
2667 bail_unlock_orphan:
2668 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2669 	inode_unlock(orphan_dir_inode);
2670 	iput(orphan_dir_inode);
2671 
2672 	ocfs2_free_dir_lookup_result(&orphan_insert);
2673 
2674 bail_unlock_inode:
2675 	ocfs2_inode_unlock(inode, 1);
2676 	brelse(di_bh);
2677 
2678 bail:
2679 	return status;
2680 }
2681 
2682 int ocfs2_del_inode_from_orphan(struct ocfs2_super *osb,
2683 		struct inode *inode, struct buffer_head *di_bh,
2684 		int update_isize, loff_t end)
2685 {
2686 	struct inode *orphan_dir_inode = NULL;
2687 	struct buffer_head *orphan_dir_bh = NULL;
2688 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2689 	handle_t *handle = NULL;
2690 	int status = 0;
2691 
2692 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2693 			ORPHAN_DIR_SYSTEM_INODE,
2694 			le16_to_cpu(di->i_dio_orphaned_slot));
2695 	if (!orphan_dir_inode) {
2696 		status = -ENOENT;
2697 		mlog_errno(status);
2698 		goto bail;
2699 	}
2700 
2701 	inode_lock(orphan_dir_inode);
2702 	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2703 	if (status < 0) {
2704 		inode_unlock(orphan_dir_inode);
2705 		iput(orphan_dir_inode);
2706 		mlog_errno(status);
2707 		goto bail;
2708 	}
2709 
2710 	handle = ocfs2_start_trans(osb,
2711 			OCFS2_INODE_DEL_FROM_ORPHAN_CREDITS);
2712 	if (IS_ERR(handle)) {
2713 		status = PTR_ERR(handle);
2714 		goto bail_unlock_orphan;
2715 	}
2716 
2717 	BUG_ON(!(di->i_flags & cpu_to_le32(OCFS2_DIO_ORPHANED_FL)));
2718 
2719 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode,
2720 				inode, orphan_dir_bh, true);
2721 	if (status < 0) {
2722 		mlog_errno(status);
2723 		goto bail_commit;
2724 	}
2725 
2726 	status = ocfs2_journal_access_di(handle,
2727 			INODE_CACHE(inode),
2728 			di_bh,
2729 			OCFS2_JOURNAL_ACCESS_WRITE);
2730 	if (status < 0) {
2731 		mlog_errno(status);
2732 		goto bail_commit;
2733 	}
2734 
2735 	di->i_flags &= ~cpu_to_le32(OCFS2_DIO_ORPHANED_FL);
2736 	di->i_dio_orphaned_slot = 0;
2737 
2738 	if (update_isize) {
2739 		status = ocfs2_set_inode_size(handle, inode, di_bh, end);
2740 		if (status)
2741 			mlog_errno(status);
2742 	} else
2743 		ocfs2_journal_dirty(handle, di_bh);
2744 
2745 bail_commit:
2746 	ocfs2_commit_trans(osb, handle);
2747 
2748 bail_unlock_orphan:
2749 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2750 	inode_unlock(orphan_dir_inode);
2751 	brelse(orphan_dir_bh);
2752 	iput(orphan_dir_inode);
2753 
2754 bail:
2755 	return status;
2756 }
2757 
2758 int ocfs2_mv_orphaned_inode_to_new(struct inode *dir,
2759 				   struct inode *inode,
2760 				   struct dentry *dentry)
2761 {
2762 	int status = 0;
2763 	struct buffer_head *parent_di_bh = NULL;
2764 	handle_t *handle = NULL;
2765 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
2766 	struct ocfs2_dinode *dir_di, *di;
2767 	struct inode *orphan_dir_inode = NULL;
2768 	struct buffer_head *orphan_dir_bh = NULL;
2769 	struct buffer_head *di_bh = NULL;
2770 	struct ocfs2_dir_lookup_result lookup = { NULL, };
2771 
2772 	trace_ocfs2_mv_orphaned_inode_to_new(dir, dentry,
2773 				dentry->d_name.len, dentry->d_name.name,
2774 				(unsigned long long)OCFS2_I(dir)->ip_blkno,
2775 				(unsigned long long)OCFS2_I(inode)->ip_blkno);
2776 
2777 	status = ocfs2_inode_lock(dir, &parent_di_bh, 1);
2778 	if (status < 0) {
2779 		if (status != -ENOENT)
2780 			mlog_errno(status);
2781 		return status;
2782 	}
2783 
2784 	dir_di = (struct ocfs2_dinode *) parent_di_bh->b_data;
2785 	if (!dir_di->i_links_count) {
2786 		/* can't make a file in a deleted directory. */
2787 		status = -ENOENT;
2788 		goto leave;
2789 	}
2790 
2791 	status = ocfs2_check_dir_for_entry(dir, dentry->d_name.name,
2792 					   dentry->d_name.len);
2793 	if (status)
2794 		goto leave;
2795 
2796 	/* get a spot inside the dir. */
2797 	status = ocfs2_prepare_dir_for_insert(osb, dir, parent_di_bh,
2798 					      dentry->d_name.name,
2799 					      dentry->d_name.len, &lookup);
2800 	if (status < 0) {
2801 		mlog_errno(status);
2802 		goto leave;
2803 	}
2804 
2805 	orphan_dir_inode = ocfs2_get_system_file_inode(osb,
2806 						       ORPHAN_DIR_SYSTEM_INODE,
2807 						       osb->slot_num);
2808 	if (!orphan_dir_inode) {
2809 		status = -ENOENT;
2810 		mlog_errno(status);
2811 		goto leave;
2812 	}
2813 
2814 	inode_lock(orphan_dir_inode);
2815 
2816 	status = ocfs2_inode_lock(orphan_dir_inode, &orphan_dir_bh, 1);
2817 	if (status < 0) {
2818 		mlog_errno(status);
2819 		inode_unlock(orphan_dir_inode);
2820 		iput(orphan_dir_inode);
2821 		goto leave;
2822 	}
2823 
2824 	status = ocfs2_read_inode_block(inode, &di_bh);
2825 	if (status < 0) {
2826 		mlog_errno(status);
2827 		goto orphan_unlock;
2828 	}
2829 
2830 	handle = ocfs2_start_trans(osb, ocfs2_rename_credits(osb->sb));
2831 	if (IS_ERR(handle)) {
2832 		status = PTR_ERR(handle);
2833 		handle = NULL;
2834 		mlog_errno(status);
2835 		goto orphan_unlock;
2836 	}
2837 
2838 	status = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2839 					 di_bh, OCFS2_JOURNAL_ACCESS_WRITE);
2840 	if (status < 0) {
2841 		mlog_errno(status);
2842 		goto out_commit;
2843 	}
2844 
2845 	status = ocfs2_orphan_del(osb, handle, orphan_dir_inode, inode,
2846 				  orphan_dir_bh, false);
2847 	if (status < 0) {
2848 		mlog_errno(status);
2849 		goto out_commit;
2850 	}
2851 
2852 	di = (struct ocfs2_dinode *)di_bh->b_data;
2853 	di->i_flags &= ~cpu_to_le32(OCFS2_ORPHANED_FL);
2854 	di->i_orphaned_slot = 0;
2855 	set_nlink(inode, 1);
2856 	ocfs2_set_links_count(di, inode->i_nlink);
2857 	ocfs2_update_inode_fsync_trans(handle, inode, 1);
2858 	ocfs2_journal_dirty(handle, di_bh);
2859 
2860 	status = ocfs2_add_entry(handle, dentry, inode,
2861 				 OCFS2_I(inode)->ip_blkno, parent_di_bh,
2862 				 &lookup);
2863 	if (status < 0) {
2864 		mlog_errno(status);
2865 		goto out_commit;
2866 	}
2867 
2868 	status = ocfs2_dentry_attach_lock(dentry, inode,
2869 					  OCFS2_I(dir)->ip_blkno);
2870 	if (status) {
2871 		mlog_errno(status);
2872 		goto out_commit;
2873 	}
2874 
2875 	d_instantiate(dentry, inode);
2876 	status = 0;
2877 out_commit:
2878 	ocfs2_commit_trans(osb, handle);
2879 orphan_unlock:
2880 	ocfs2_inode_unlock(orphan_dir_inode, 1);
2881 	inode_unlock(orphan_dir_inode);
2882 	iput(orphan_dir_inode);
2883 leave:
2884 
2885 	ocfs2_inode_unlock(dir, 1);
2886 
2887 	brelse(di_bh);
2888 	brelse(parent_di_bh);
2889 	brelse(orphan_dir_bh);
2890 
2891 	ocfs2_free_dir_lookup_result(&lookup);
2892 
2893 	if (status)
2894 		mlog_errno(status);
2895 
2896 	return status;
2897 }
2898 
2899 const struct inode_operations ocfs2_dir_iops = {
2900 	.create		= ocfs2_create,
2901 	.lookup		= ocfs2_lookup,
2902 	.link		= ocfs2_link,
2903 	.unlink		= ocfs2_unlink,
2904 	.rmdir		= ocfs2_unlink,
2905 	.symlink	= ocfs2_symlink,
2906 	.mkdir		= ocfs2_mkdir,
2907 	.mknod		= ocfs2_mknod,
2908 	.rename		= ocfs2_rename,
2909 	.setattr	= ocfs2_setattr,
2910 	.getattr	= ocfs2_getattr,
2911 	.permission	= ocfs2_permission,
2912 	.listxattr	= ocfs2_listxattr,
2913 	.fiemap         = ocfs2_fiemap,
2914 	.get_acl	= ocfs2_iop_get_acl,
2915 	.set_acl	= ocfs2_iop_set_acl,
2916 };
2917