xref: /openbmc/linux/fs/ocfs2/refcounttree.c (revision d7a3d85e)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * refcounttree.c
5  *
6  * Copyright (C) 2009 Oracle.  All rights reserved.
7  *
8  * This program is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU General Public
10  * License version 2 as published by the Free Software Foundation.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
15  * General Public License for more details.
16  */
17 
18 #include <linux/sort.h>
19 #include <cluster/masklog.h>
20 #include "ocfs2.h"
21 #include "inode.h"
22 #include "alloc.h"
23 #include "suballoc.h"
24 #include "journal.h"
25 #include "uptodate.h"
26 #include "super.h"
27 #include "buffer_head_io.h"
28 #include "blockcheck.h"
29 #include "refcounttree.h"
30 #include "sysfile.h"
31 #include "dlmglue.h"
32 #include "extent_map.h"
33 #include "aops.h"
34 #include "xattr.h"
35 #include "namei.h"
36 #include "ocfs2_trace.h"
37 
38 #include <linux/bio.h>
39 #include <linux/blkdev.h>
40 #include <linux/slab.h>
41 #include <linux/writeback.h>
42 #include <linux/pagevec.h>
43 #include <linux/swap.h>
44 #include <linux/security.h>
45 #include <linux/fsnotify.h>
46 #include <linux/quotaops.h>
47 #include <linux/namei.h>
48 #include <linux/mount.h>
49 #include <linux/posix_acl.h>
50 
51 struct ocfs2_cow_context {
52 	struct inode *inode;
53 	u32 cow_start;
54 	u32 cow_len;
55 	struct ocfs2_extent_tree data_et;
56 	struct ocfs2_refcount_tree *ref_tree;
57 	struct buffer_head *ref_root_bh;
58 	struct ocfs2_alloc_context *meta_ac;
59 	struct ocfs2_alloc_context *data_ac;
60 	struct ocfs2_cached_dealloc_ctxt dealloc;
61 	void *cow_object;
62 	struct ocfs2_post_refcount *post_refcount;
63 	int extra_credits;
64 	int (*get_clusters)(struct ocfs2_cow_context *context,
65 			    u32 v_cluster, u32 *p_cluster,
66 			    u32 *num_clusters,
67 			    unsigned int *extent_flags);
68 	int (*cow_duplicate_clusters)(handle_t *handle,
69 				      struct inode *inode,
70 				      u32 cpos, u32 old_cluster,
71 				      u32 new_cluster, u32 new_len);
72 };
73 
74 static inline struct ocfs2_refcount_tree *
75 cache_info_to_refcount(struct ocfs2_caching_info *ci)
76 {
77 	return container_of(ci, struct ocfs2_refcount_tree, rf_ci);
78 }
79 
80 static int ocfs2_validate_refcount_block(struct super_block *sb,
81 					 struct buffer_head *bh)
82 {
83 	int rc;
84 	struct ocfs2_refcount_block *rb =
85 		(struct ocfs2_refcount_block *)bh->b_data;
86 
87 	trace_ocfs2_validate_refcount_block((unsigned long long)bh->b_blocknr);
88 
89 	BUG_ON(!buffer_uptodate(bh));
90 
91 	/*
92 	 * If the ecc fails, we return the error but otherwise
93 	 * leave the filesystem running.  We know any error is
94 	 * local to this block.
95 	 */
96 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &rb->rf_check);
97 	if (rc) {
98 		mlog(ML_ERROR, "Checksum failed for refcount block %llu\n",
99 		     (unsigned long long)bh->b_blocknr);
100 		return rc;
101 	}
102 
103 
104 	if (!OCFS2_IS_VALID_REFCOUNT_BLOCK(rb)) {
105 		ocfs2_error(sb,
106 			    "Refcount block #%llu has bad signature %.*s",
107 			    (unsigned long long)bh->b_blocknr, 7,
108 			    rb->rf_signature);
109 		return -EINVAL;
110 	}
111 
112 	if (le64_to_cpu(rb->rf_blkno) != bh->b_blocknr) {
113 		ocfs2_error(sb,
114 			    "Refcount block #%llu has an invalid rf_blkno "
115 			    "of %llu",
116 			    (unsigned long long)bh->b_blocknr,
117 			    (unsigned long long)le64_to_cpu(rb->rf_blkno));
118 		return -EINVAL;
119 	}
120 
121 	if (le32_to_cpu(rb->rf_fs_generation) != OCFS2_SB(sb)->fs_generation) {
122 		ocfs2_error(sb,
123 			    "Refcount block #%llu has an invalid "
124 			    "rf_fs_generation of #%u",
125 			    (unsigned long long)bh->b_blocknr,
126 			    le32_to_cpu(rb->rf_fs_generation));
127 		return -EINVAL;
128 	}
129 
130 	return 0;
131 }
132 
133 static int ocfs2_read_refcount_block(struct ocfs2_caching_info *ci,
134 				     u64 rb_blkno,
135 				     struct buffer_head **bh)
136 {
137 	int rc;
138 	struct buffer_head *tmp = *bh;
139 
140 	rc = ocfs2_read_block(ci, rb_blkno, &tmp,
141 			      ocfs2_validate_refcount_block);
142 
143 	/* If ocfs2_read_block() got us a new bh, pass it up. */
144 	if (!rc && !*bh)
145 		*bh = tmp;
146 
147 	return rc;
148 }
149 
150 static u64 ocfs2_refcount_cache_owner(struct ocfs2_caching_info *ci)
151 {
152 	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
153 
154 	return rf->rf_blkno;
155 }
156 
157 static struct super_block *
158 ocfs2_refcount_cache_get_super(struct ocfs2_caching_info *ci)
159 {
160 	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
161 
162 	return rf->rf_sb;
163 }
164 
165 static void ocfs2_refcount_cache_lock(struct ocfs2_caching_info *ci)
166 {
167 	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
168 
169 	spin_lock(&rf->rf_lock);
170 }
171 
172 static void ocfs2_refcount_cache_unlock(struct ocfs2_caching_info *ci)
173 {
174 	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
175 
176 	spin_unlock(&rf->rf_lock);
177 }
178 
179 static void ocfs2_refcount_cache_io_lock(struct ocfs2_caching_info *ci)
180 {
181 	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
182 
183 	mutex_lock(&rf->rf_io_mutex);
184 }
185 
186 static void ocfs2_refcount_cache_io_unlock(struct ocfs2_caching_info *ci)
187 {
188 	struct ocfs2_refcount_tree *rf = cache_info_to_refcount(ci);
189 
190 	mutex_unlock(&rf->rf_io_mutex);
191 }
192 
193 static const struct ocfs2_caching_operations ocfs2_refcount_caching_ops = {
194 	.co_owner		= ocfs2_refcount_cache_owner,
195 	.co_get_super		= ocfs2_refcount_cache_get_super,
196 	.co_cache_lock		= ocfs2_refcount_cache_lock,
197 	.co_cache_unlock	= ocfs2_refcount_cache_unlock,
198 	.co_io_lock		= ocfs2_refcount_cache_io_lock,
199 	.co_io_unlock		= ocfs2_refcount_cache_io_unlock,
200 };
201 
202 static struct ocfs2_refcount_tree *
203 ocfs2_find_refcount_tree(struct ocfs2_super *osb, u64 blkno)
204 {
205 	struct rb_node *n = osb->osb_rf_lock_tree.rb_node;
206 	struct ocfs2_refcount_tree *tree = NULL;
207 
208 	while (n) {
209 		tree = rb_entry(n, struct ocfs2_refcount_tree, rf_node);
210 
211 		if (blkno < tree->rf_blkno)
212 			n = n->rb_left;
213 		else if (blkno > tree->rf_blkno)
214 			n = n->rb_right;
215 		else
216 			return tree;
217 	}
218 
219 	return NULL;
220 }
221 
222 /* osb_lock is already locked. */
223 static void ocfs2_insert_refcount_tree(struct ocfs2_super *osb,
224 				       struct ocfs2_refcount_tree *new)
225 {
226 	u64 rf_blkno = new->rf_blkno;
227 	struct rb_node *parent = NULL;
228 	struct rb_node **p = &osb->osb_rf_lock_tree.rb_node;
229 	struct ocfs2_refcount_tree *tmp;
230 
231 	while (*p) {
232 		parent = *p;
233 
234 		tmp = rb_entry(parent, struct ocfs2_refcount_tree,
235 			       rf_node);
236 
237 		if (rf_blkno < tmp->rf_blkno)
238 			p = &(*p)->rb_left;
239 		else if (rf_blkno > tmp->rf_blkno)
240 			p = &(*p)->rb_right;
241 		else {
242 			/* This should never happen! */
243 			mlog(ML_ERROR, "Duplicate refcount block %llu found!\n",
244 			     (unsigned long long)rf_blkno);
245 			BUG();
246 		}
247 	}
248 
249 	rb_link_node(&new->rf_node, parent, p);
250 	rb_insert_color(&new->rf_node, &osb->osb_rf_lock_tree);
251 }
252 
253 static void ocfs2_free_refcount_tree(struct ocfs2_refcount_tree *tree)
254 {
255 	ocfs2_metadata_cache_exit(&tree->rf_ci);
256 	ocfs2_simple_drop_lockres(OCFS2_SB(tree->rf_sb), &tree->rf_lockres);
257 	ocfs2_lock_res_free(&tree->rf_lockres);
258 	kfree(tree);
259 }
260 
261 static inline void
262 ocfs2_erase_refcount_tree_from_list_no_lock(struct ocfs2_super *osb,
263 					struct ocfs2_refcount_tree *tree)
264 {
265 	rb_erase(&tree->rf_node, &osb->osb_rf_lock_tree);
266 	if (osb->osb_ref_tree_lru && osb->osb_ref_tree_lru == tree)
267 		osb->osb_ref_tree_lru = NULL;
268 }
269 
270 static void ocfs2_erase_refcount_tree_from_list(struct ocfs2_super *osb,
271 					struct ocfs2_refcount_tree *tree)
272 {
273 	spin_lock(&osb->osb_lock);
274 	ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
275 	spin_unlock(&osb->osb_lock);
276 }
277 
278 static void ocfs2_kref_remove_refcount_tree(struct kref *kref)
279 {
280 	struct ocfs2_refcount_tree *tree =
281 		container_of(kref, struct ocfs2_refcount_tree, rf_getcnt);
282 
283 	ocfs2_free_refcount_tree(tree);
284 }
285 
286 static inline void
287 ocfs2_refcount_tree_get(struct ocfs2_refcount_tree *tree)
288 {
289 	kref_get(&tree->rf_getcnt);
290 }
291 
292 static inline void
293 ocfs2_refcount_tree_put(struct ocfs2_refcount_tree *tree)
294 {
295 	kref_put(&tree->rf_getcnt, ocfs2_kref_remove_refcount_tree);
296 }
297 
298 static inline void ocfs2_init_refcount_tree_ci(struct ocfs2_refcount_tree *new,
299 					       struct super_block *sb)
300 {
301 	ocfs2_metadata_cache_init(&new->rf_ci, &ocfs2_refcount_caching_ops);
302 	mutex_init(&new->rf_io_mutex);
303 	new->rf_sb = sb;
304 	spin_lock_init(&new->rf_lock);
305 }
306 
307 static inline void ocfs2_init_refcount_tree_lock(struct ocfs2_super *osb,
308 					struct ocfs2_refcount_tree *new,
309 					u64 rf_blkno, u32 generation)
310 {
311 	init_rwsem(&new->rf_sem);
312 	ocfs2_refcount_lock_res_init(&new->rf_lockres, osb,
313 				     rf_blkno, generation);
314 }
315 
316 static struct ocfs2_refcount_tree*
317 ocfs2_allocate_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno)
318 {
319 	struct ocfs2_refcount_tree *new;
320 
321 	new = kzalloc(sizeof(struct ocfs2_refcount_tree), GFP_NOFS);
322 	if (!new)
323 		return NULL;
324 
325 	new->rf_blkno = rf_blkno;
326 	kref_init(&new->rf_getcnt);
327 	ocfs2_init_refcount_tree_ci(new, osb->sb);
328 
329 	return new;
330 }
331 
332 static int ocfs2_get_refcount_tree(struct ocfs2_super *osb, u64 rf_blkno,
333 				   struct ocfs2_refcount_tree **ret_tree)
334 {
335 	int ret = 0;
336 	struct ocfs2_refcount_tree *tree, *new = NULL;
337 	struct buffer_head *ref_root_bh = NULL;
338 	struct ocfs2_refcount_block *ref_rb;
339 
340 	spin_lock(&osb->osb_lock);
341 	if (osb->osb_ref_tree_lru &&
342 	    osb->osb_ref_tree_lru->rf_blkno == rf_blkno)
343 		tree = osb->osb_ref_tree_lru;
344 	else
345 		tree = ocfs2_find_refcount_tree(osb, rf_blkno);
346 	if (tree)
347 		goto out;
348 
349 	spin_unlock(&osb->osb_lock);
350 
351 	new = ocfs2_allocate_refcount_tree(osb, rf_blkno);
352 	if (!new) {
353 		ret = -ENOMEM;
354 		mlog_errno(ret);
355 		return ret;
356 	}
357 	/*
358 	 * We need the generation to create the refcount tree lock and since
359 	 * it isn't changed during the tree modification, we are safe here to
360 	 * read without protection.
361 	 * We also have to purge the cache after we create the lock since the
362 	 * refcount block may have the stale data. It can only be trusted when
363 	 * we hold the refcount lock.
364 	 */
365 	ret = ocfs2_read_refcount_block(&new->rf_ci, rf_blkno, &ref_root_bh);
366 	if (ret) {
367 		mlog_errno(ret);
368 		ocfs2_metadata_cache_exit(&new->rf_ci);
369 		kfree(new);
370 		return ret;
371 	}
372 
373 	ref_rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
374 	new->rf_generation = le32_to_cpu(ref_rb->rf_generation);
375 	ocfs2_init_refcount_tree_lock(osb, new, rf_blkno,
376 				      new->rf_generation);
377 	ocfs2_metadata_cache_purge(&new->rf_ci);
378 
379 	spin_lock(&osb->osb_lock);
380 	tree = ocfs2_find_refcount_tree(osb, rf_blkno);
381 	if (tree)
382 		goto out;
383 
384 	ocfs2_insert_refcount_tree(osb, new);
385 
386 	tree = new;
387 	new = NULL;
388 
389 out:
390 	*ret_tree = tree;
391 
392 	osb->osb_ref_tree_lru = tree;
393 
394 	spin_unlock(&osb->osb_lock);
395 
396 	if (new)
397 		ocfs2_free_refcount_tree(new);
398 
399 	brelse(ref_root_bh);
400 	return ret;
401 }
402 
403 static int ocfs2_get_refcount_block(struct inode *inode, u64 *ref_blkno)
404 {
405 	int ret;
406 	struct buffer_head *di_bh = NULL;
407 	struct ocfs2_dinode *di;
408 
409 	ret = ocfs2_read_inode_block(inode, &di_bh);
410 	if (ret) {
411 		mlog_errno(ret);
412 		goto out;
413 	}
414 
415 	BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
416 
417 	di = (struct ocfs2_dinode *)di_bh->b_data;
418 	*ref_blkno = le64_to_cpu(di->i_refcount_loc);
419 	brelse(di_bh);
420 out:
421 	return ret;
422 }
423 
424 static int __ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
425 				      struct ocfs2_refcount_tree *tree, int rw)
426 {
427 	int ret;
428 
429 	ret = ocfs2_refcount_lock(tree, rw);
430 	if (ret) {
431 		mlog_errno(ret);
432 		goto out;
433 	}
434 
435 	if (rw)
436 		down_write(&tree->rf_sem);
437 	else
438 		down_read(&tree->rf_sem);
439 
440 out:
441 	return ret;
442 }
443 
444 /*
445  * Lock the refcount tree pointed by ref_blkno and return the tree.
446  * In most case, we lock the tree and read the refcount block.
447  * So read it here if the caller really needs it.
448  *
449  * If the tree has been re-created by other node, it will free the
450  * old one and re-create it.
451  */
452 int ocfs2_lock_refcount_tree(struct ocfs2_super *osb,
453 			     u64 ref_blkno, int rw,
454 			     struct ocfs2_refcount_tree **ret_tree,
455 			     struct buffer_head **ref_bh)
456 {
457 	int ret, delete_tree = 0;
458 	struct ocfs2_refcount_tree *tree = NULL;
459 	struct buffer_head *ref_root_bh = NULL;
460 	struct ocfs2_refcount_block *rb;
461 
462 again:
463 	ret = ocfs2_get_refcount_tree(osb, ref_blkno, &tree);
464 	if (ret) {
465 		mlog_errno(ret);
466 		return ret;
467 	}
468 
469 	ocfs2_refcount_tree_get(tree);
470 
471 	ret = __ocfs2_lock_refcount_tree(osb, tree, rw);
472 	if (ret) {
473 		mlog_errno(ret);
474 		ocfs2_refcount_tree_put(tree);
475 		goto out;
476 	}
477 
478 	ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
479 					&ref_root_bh);
480 	if (ret) {
481 		mlog_errno(ret);
482 		ocfs2_unlock_refcount_tree(osb, tree, rw);
483 		ocfs2_refcount_tree_put(tree);
484 		goto out;
485 	}
486 
487 	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
488 	/*
489 	 * If the refcount block has been freed and re-created, we may need
490 	 * to recreate the refcount tree also.
491 	 *
492 	 * Here we just remove the tree from the rb-tree, and the last
493 	 * kref holder will unlock and delete this refcount_tree.
494 	 * Then we goto "again" and ocfs2_get_refcount_tree will create
495 	 * the new refcount tree for us.
496 	 */
497 	if (tree->rf_generation != le32_to_cpu(rb->rf_generation)) {
498 		if (!tree->rf_removed) {
499 			ocfs2_erase_refcount_tree_from_list(osb, tree);
500 			tree->rf_removed = 1;
501 			delete_tree = 1;
502 		}
503 
504 		ocfs2_unlock_refcount_tree(osb, tree, rw);
505 		/*
506 		 * We get an extra reference when we create the refcount
507 		 * tree, so another put will destroy it.
508 		 */
509 		if (delete_tree)
510 			ocfs2_refcount_tree_put(tree);
511 		brelse(ref_root_bh);
512 		ref_root_bh = NULL;
513 		goto again;
514 	}
515 
516 	*ret_tree = tree;
517 	if (ref_bh) {
518 		*ref_bh = ref_root_bh;
519 		ref_root_bh = NULL;
520 	}
521 out:
522 	brelse(ref_root_bh);
523 	return ret;
524 }
525 
526 void ocfs2_unlock_refcount_tree(struct ocfs2_super *osb,
527 				struct ocfs2_refcount_tree *tree, int rw)
528 {
529 	if (rw)
530 		up_write(&tree->rf_sem);
531 	else
532 		up_read(&tree->rf_sem);
533 
534 	ocfs2_refcount_unlock(tree, rw);
535 	ocfs2_refcount_tree_put(tree);
536 }
537 
538 void ocfs2_purge_refcount_trees(struct ocfs2_super *osb)
539 {
540 	struct rb_node *node;
541 	struct ocfs2_refcount_tree *tree;
542 	struct rb_root *root = &osb->osb_rf_lock_tree;
543 
544 	while ((node = rb_last(root)) != NULL) {
545 		tree = rb_entry(node, struct ocfs2_refcount_tree, rf_node);
546 
547 		trace_ocfs2_purge_refcount_trees(
548 				(unsigned long long) tree->rf_blkno);
549 
550 		rb_erase(&tree->rf_node, root);
551 		ocfs2_free_refcount_tree(tree);
552 	}
553 }
554 
555 /*
556  * Create a refcount tree for an inode.
557  * We take for granted that the inode is already locked.
558  */
559 static int ocfs2_create_refcount_tree(struct inode *inode,
560 				      struct buffer_head *di_bh)
561 {
562 	int ret;
563 	handle_t *handle = NULL;
564 	struct ocfs2_alloc_context *meta_ac = NULL;
565 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
566 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
567 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
568 	struct buffer_head *new_bh = NULL;
569 	struct ocfs2_refcount_block *rb;
570 	struct ocfs2_refcount_tree *new_tree = NULL, *tree = NULL;
571 	u16 suballoc_bit_start;
572 	u32 num_got;
573 	u64 suballoc_loc, first_blkno;
574 
575 	BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
576 
577 	trace_ocfs2_create_refcount_tree(
578 		(unsigned long long)OCFS2_I(inode)->ip_blkno);
579 
580 	ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
581 	if (ret) {
582 		mlog_errno(ret);
583 		goto out;
584 	}
585 
586 	handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_CREATE_CREDITS);
587 	if (IS_ERR(handle)) {
588 		ret = PTR_ERR(handle);
589 		mlog_errno(ret);
590 		goto out;
591 	}
592 
593 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
594 				      OCFS2_JOURNAL_ACCESS_WRITE);
595 	if (ret) {
596 		mlog_errno(ret);
597 		goto out_commit;
598 	}
599 
600 	ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
601 				   &suballoc_bit_start, &num_got,
602 				   &first_blkno);
603 	if (ret) {
604 		mlog_errno(ret);
605 		goto out_commit;
606 	}
607 
608 	new_tree = ocfs2_allocate_refcount_tree(osb, first_blkno);
609 	if (!new_tree) {
610 		ret = -ENOMEM;
611 		mlog_errno(ret);
612 		goto out_commit;
613 	}
614 
615 	new_bh = sb_getblk(inode->i_sb, first_blkno);
616 	if (!new_bh) {
617 		ret = -ENOMEM;
618 		mlog_errno(ret);
619 		goto out_commit;
620 	}
621 	ocfs2_set_new_buffer_uptodate(&new_tree->rf_ci, new_bh);
622 
623 	ret = ocfs2_journal_access_rb(handle, &new_tree->rf_ci, new_bh,
624 				      OCFS2_JOURNAL_ACCESS_CREATE);
625 	if (ret) {
626 		mlog_errno(ret);
627 		goto out_commit;
628 	}
629 
630 	/* Initialize ocfs2_refcount_block. */
631 	rb = (struct ocfs2_refcount_block *)new_bh->b_data;
632 	memset(rb, 0, inode->i_sb->s_blocksize);
633 	strcpy((void *)rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
634 	rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
635 	rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
636 	rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
637 	rb->rf_fs_generation = cpu_to_le32(osb->fs_generation);
638 	rb->rf_blkno = cpu_to_le64(first_blkno);
639 	rb->rf_count = cpu_to_le32(1);
640 	rb->rf_records.rl_count =
641 			cpu_to_le16(ocfs2_refcount_recs_per_rb(osb->sb));
642 	spin_lock(&osb->osb_lock);
643 	rb->rf_generation = osb->s_next_generation++;
644 	spin_unlock(&osb->osb_lock);
645 
646 	ocfs2_journal_dirty(handle, new_bh);
647 
648 	spin_lock(&oi->ip_lock);
649 	oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
650 	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
651 	di->i_refcount_loc = cpu_to_le64(first_blkno);
652 	spin_unlock(&oi->ip_lock);
653 
654 	trace_ocfs2_create_refcount_tree_blkno((unsigned long long)first_blkno);
655 
656 	ocfs2_journal_dirty(handle, di_bh);
657 
658 	/*
659 	 * We have to init the tree lock here since it will use
660 	 * the generation number to create it.
661 	 */
662 	new_tree->rf_generation = le32_to_cpu(rb->rf_generation);
663 	ocfs2_init_refcount_tree_lock(osb, new_tree, first_blkno,
664 				      new_tree->rf_generation);
665 
666 	spin_lock(&osb->osb_lock);
667 	tree = ocfs2_find_refcount_tree(osb, first_blkno);
668 
669 	/*
670 	 * We've just created a new refcount tree in this block.  If
671 	 * we found a refcount tree on the ocfs2_super, it must be
672 	 * one we just deleted.  We free the old tree before
673 	 * inserting the new tree.
674 	 */
675 	BUG_ON(tree && tree->rf_generation == new_tree->rf_generation);
676 	if (tree)
677 		ocfs2_erase_refcount_tree_from_list_no_lock(osb, tree);
678 	ocfs2_insert_refcount_tree(osb, new_tree);
679 	spin_unlock(&osb->osb_lock);
680 	new_tree = NULL;
681 	if (tree)
682 		ocfs2_refcount_tree_put(tree);
683 
684 out_commit:
685 	ocfs2_commit_trans(osb, handle);
686 
687 out:
688 	if (new_tree) {
689 		ocfs2_metadata_cache_exit(&new_tree->rf_ci);
690 		kfree(new_tree);
691 	}
692 
693 	brelse(new_bh);
694 	if (meta_ac)
695 		ocfs2_free_alloc_context(meta_ac);
696 
697 	return ret;
698 }
699 
700 static int ocfs2_set_refcount_tree(struct inode *inode,
701 				   struct buffer_head *di_bh,
702 				   u64 refcount_loc)
703 {
704 	int ret;
705 	handle_t *handle = NULL;
706 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
707 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
708 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
709 	struct buffer_head *ref_root_bh = NULL;
710 	struct ocfs2_refcount_block *rb;
711 	struct ocfs2_refcount_tree *ref_tree;
712 
713 	BUG_ON(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL);
714 
715 	ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
716 				       &ref_tree, &ref_root_bh);
717 	if (ret) {
718 		mlog_errno(ret);
719 		return ret;
720 	}
721 
722 	handle = ocfs2_start_trans(osb, OCFS2_REFCOUNT_TREE_SET_CREDITS);
723 	if (IS_ERR(handle)) {
724 		ret = PTR_ERR(handle);
725 		mlog_errno(ret);
726 		goto out;
727 	}
728 
729 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
730 				      OCFS2_JOURNAL_ACCESS_WRITE);
731 	if (ret) {
732 		mlog_errno(ret);
733 		goto out_commit;
734 	}
735 
736 	ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, ref_root_bh,
737 				      OCFS2_JOURNAL_ACCESS_WRITE);
738 	if (ret) {
739 		mlog_errno(ret);
740 		goto out_commit;
741 	}
742 
743 	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
744 	le32_add_cpu(&rb->rf_count, 1);
745 
746 	ocfs2_journal_dirty(handle, ref_root_bh);
747 
748 	spin_lock(&oi->ip_lock);
749 	oi->ip_dyn_features |= OCFS2_HAS_REFCOUNT_FL;
750 	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
751 	di->i_refcount_loc = cpu_to_le64(refcount_loc);
752 	spin_unlock(&oi->ip_lock);
753 	ocfs2_journal_dirty(handle, di_bh);
754 
755 out_commit:
756 	ocfs2_commit_trans(osb, handle);
757 out:
758 	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
759 	brelse(ref_root_bh);
760 
761 	return ret;
762 }
763 
764 int ocfs2_remove_refcount_tree(struct inode *inode, struct buffer_head *di_bh)
765 {
766 	int ret, delete_tree = 0;
767 	handle_t *handle = NULL;
768 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
769 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
770 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
771 	struct ocfs2_refcount_block *rb;
772 	struct inode *alloc_inode = NULL;
773 	struct buffer_head *alloc_bh = NULL;
774 	struct buffer_head *blk_bh = NULL;
775 	struct ocfs2_refcount_tree *ref_tree;
776 	int credits = OCFS2_REFCOUNT_TREE_REMOVE_CREDITS;
777 	u64 blk = 0, bg_blkno = 0, ref_blkno = le64_to_cpu(di->i_refcount_loc);
778 	u16 bit = 0;
779 
780 	if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL))
781 		return 0;
782 
783 	BUG_ON(!ref_blkno);
784 	ret = ocfs2_lock_refcount_tree(osb, ref_blkno, 1, &ref_tree, &blk_bh);
785 	if (ret) {
786 		mlog_errno(ret);
787 		return ret;
788 	}
789 
790 	rb = (struct ocfs2_refcount_block *)blk_bh->b_data;
791 
792 	/*
793 	 * If we are the last user, we need to free the block.
794 	 * So lock the allocator ahead.
795 	 */
796 	if (le32_to_cpu(rb->rf_count) == 1) {
797 		blk = le64_to_cpu(rb->rf_blkno);
798 		bit = le16_to_cpu(rb->rf_suballoc_bit);
799 		if (rb->rf_suballoc_loc)
800 			bg_blkno = le64_to_cpu(rb->rf_suballoc_loc);
801 		else
802 			bg_blkno = ocfs2_which_suballoc_group(blk, bit);
803 
804 		alloc_inode = ocfs2_get_system_file_inode(osb,
805 					EXTENT_ALLOC_SYSTEM_INODE,
806 					le16_to_cpu(rb->rf_suballoc_slot));
807 		if (!alloc_inode) {
808 			ret = -ENOMEM;
809 			mlog_errno(ret);
810 			goto out;
811 		}
812 		mutex_lock(&alloc_inode->i_mutex);
813 
814 		ret = ocfs2_inode_lock(alloc_inode, &alloc_bh, 1);
815 		if (ret) {
816 			mlog_errno(ret);
817 			goto out_mutex;
818 		}
819 
820 		credits += OCFS2_SUBALLOC_FREE;
821 	}
822 
823 	handle = ocfs2_start_trans(osb, credits);
824 	if (IS_ERR(handle)) {
825 		ret = PTR_ERR(handle);
826 		mlog_errno(ret);
827 		goto out_unlock;
828 	}
829 
830 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
831 				      OCFS2_JOURNAL_ACCESS_WRITE);
832 	if (ret) {
833 		mlog_errno(ret);
834 		goto out_commit;
835 	}
836 
837 	ret = ocfs2_journal_access_rb(handle, &ref_tree->rf_ci, blk_bh,
838 				      OCFS2_JOURNAL_ACCESS_WRITE);
839 	if (ret) {
840 		mlog_errno(ret);
841 		goto out_commit;
842 	}
843 
844 	spin_lock(&oi->ip_lock);
845 	oi->ip_dyn_features &= ~OCFS2_HAS_REFCOUNT_FL;
846 	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
847 	di->i_refcount_loc = 0;
848 	spin_unlock(&oi->ip_lock);
849 	ocfs2_journal_dirty(handle, di_bh);
850 
851 	le32_add_cpu(&rb->rf_count , -1);
852 	ocfs2_journal_dirty(handle, blk_bh);
853 
854 	if (!rb->rf_count) {
855 		delete_tree = 1;
856 		ocfs2_erase_refcount_tree_from_list(osb, ref_tree);
857 		ret = ocfs2_free_suballoc_bits(handle, alloc_inode,
858 					       alloc_bh, bit, bg_blkno, 1);
859 		if (ret)
860 			mlog_errno(ret);
861 	}
862 
863 out_commit:
864 	ocfs2_commit_trans(osb, handle);
865 out_unlock:
866 	if (alloc_inode) {
867 		ocfs2_inode_unlock(alloc_inode, 1);
868 		brelse(alloc_bh);
869 	}
870 out_mutex:
871 	if (alloc_inode) {
872 		mutex_unlock(&alloc_inode->i_mutex);
873 		iput(alloc_inode);
874 	}
875 out:
876 	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
877 	if (delete_tree)
878 		ocfs2_refcount_tree_put(ref_tree);
879 	brelse(blk_bh);
880 
881 	return ret;
882 }
883 
884 static void ocfs2_find_refcount_rec_in_rl(struct ocfs2_caching_info *ci,
885 					  struct buffer_head *ref_leaf_bh,
886 					  u64 cpos, unsigned int len,
887 					  struct ocfs2_refcount_rec *ret_rec,
888 					  int *index)
889 {
890 	int i = 0;
891 	struct ocfs2_refcount_block *rb =
892 		(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
893 	struct ocfs2_refcount_rec *rec = NULL;
894 
895 	for (; i < le16_to_cpu(rb->rf_records.rl_used); i++) {
896 		rec = &rb->rf_records.rl_recs[i];
897 
898 		if (le64_to_cpu(rec->r_cpos) +
899 		    le32_to_cpu(rec->r_clusters) <= cpos)
900 			continue;
901 		else if (le64_to_cpu(rec->r_cpos) > cpos)
902 			break;
903 
904 		/* ok, cpos fail in this rec. Just return. */
905 		if (ret_rec)
906 			*ret_rec = *rec;
907 		goto out;
908 	}
909 
910 	if (ret_rec) {
911 		/* We meet with a hole here, so fake the rec. */
912 		ret_rec->r_cpos = cpu_to_le64(cpos);
913 		ret_rec->r_refcount = 0;
914 		if (i < le16_to_cpu(rb->rf_records.rl_used) &&
915 		    le64_to_cpu(rec->r_cpos) < cpos + len)
916 			ret_rec->r_clusters =
917 				cpu_to_le32(le64_to_cpu(rec->r_cpos) - cpos);
918 		else
919 			ret_rec->r_clusters = cpu_to_le32(len);
920 	}
921 
922 out:
923 	*index = i;
924 }
925 
926 /*
927  * Try to remove refcount tree. The mechanism is:
928  * 1) Check whether i_clusters == 0, if no, exit.
929  * 2) check whether we have i_xattr_loc in dinode. if yes, exit.
930  * 3) Check whether we have inline xattr stored outside, if yes, exit.
931  * 4) Remove the tree.
932  */
933 int ocfs2_try_remove_refcount_tree(struct inode *inode,
934 				   struct buffer_head *di_bh)
935 {
936 	int ret;
937 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
938 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
939 
940 	down_write(&oi->ip_xattr_sem);
941 	down_write(&oi->ip_alloc_sem);
942 
943 	if (oi->ip_clusters)
944 		goto out;
945 
946 	if ((oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) && di->i_xattr_loc)
947 		goto out;
948 
949 	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL &&
950 	    ocfs2_has_inline_xattr_value_outside(inode, di))
951 		goto out;
952 
953 	ret = ocfs2_remove_refcount_tree(inode, di_bh);
954 	if (ret)
955 		mlog_errno(ret);
956 out:
957 	up_write(&oi->ip_alloc_sem);
958 	up_write(&oi->ip_xattr_sem);
959 	return 0;
960 }
961 
962 /*
963  * Find the end range for a leaf refcount block indicated by
964  * el->l_recs[index].e_blkno.
965  */
966 static int ocfs2_get_refcount_cpos_end(struct ocfs2_caching_info *ci,
967 				       struct buffer_head *ref_root_bh,
968 				       struct ocfs2_extent_block *eb,
969 				       struct ocfs2_extent_list *el,
970 				       int index,  u32 *cpos_end)
971 {
972 	int ret, i, subtree_root;
973 	u32 cpos;
974 	u64 blkno;
975 	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
976 	struct ocfs2_path *left_path = NULL, *right_path = NULL;
977 	struct ocfs2_extent_tree et;
978 	struct ocfs2_extent_list *tmp_el;
979 
980 	if (index < le16_to_cpu(el->l_next_free_rec) - 1) {
981 		/*
982 		 * We have a extent rec after index, so just use the e_cpos
983 		 * of the next extent rec.
984 		 */
985 		*cpos_end = le32_to_cpu(el->l_recs[index+1].e_cpos);
986 		return 0;
987 	}
988 
989 	if (!eb || (eb && !eb->h_next_leaf_blk)) {
990 		/*
991 		 * We are the last extent rec, so any high cpos should
992 		 * be stored in this leaf refcount block.
993 		 */
994 		*cpos_end = UINT_MAX;
995 		return 0;
996 	}
997 
998 	/*
999 	 * If the extent block isn't the last one, we have to find
1000 	 * the subtree root between this extent block and the next
1001 	 * leaf extent block and get the corresponding e_cpos from
1002 	 * the subroot. Otherwise we may corrupt the b-tree.
1003 	 */
1004 	ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1005 
1006 	left_path = ocfs2_new_path_from_et(&et);
1007 	if (!left_path) {
1008 		ret = -ENOMEM;
1009 		mlog_errno(ret);
1010 		goto out;
1011 	}
1012 
1013 	cpos = le32_to_cpu(eb->h_list.l_recs[index].e_cpos);
1014 	ret = ocfs2_find_path(ci, left_path, cpos);
1015 	if (ret) {
1016 		mlog_errno(ret);
1017 		goto out;
1018 	}
1019 
1020 	right_path = ocfs2_new_path_from_path(left_path);
1021 	if (!right_path) {
1022 		ret = -ENOMEM;
1023 		mlog_errno(ret);
1024 		goto out;
1025 	}
1026 
1027 	ret = ocfs2_find_cpos_for_right_leaf(sb, left_path, &cpos);
1028 	if (ret) {
1029 		mlog_errno(ret);
1030 		goto out;
1031 	}
1032 
1033 	ret = ocfs2_find_path(ci, right_path, cpos);
1034 	if (ret) {
1035 		mlog_errno(ret);
1036 		goto out;
1037 	}
1038 
1039 	subtree_root = ocfs2_find_subtree_root(&et, left_path,
1040 					       right_path);
1041 
1042 	tmp_el = left_path->p_node[subtree_root].el;
1043 	blkno = left_path->p_node[subtree_root+1].bh->b_blocknr;
1044 	for (i = 0; i < le16_to_cpu(tmp_el->l_next_free_rec); i++) {
1045 		if (le64_to_cpu(tmp_el->l_recs[i].e_blkno) == blkno) {
1046 			*cpos_end = le32_to_cpu(tmp_el->l_recs[i+1].e_cpos);
1047 			break;
1048 		}
1049 	}
1050 
1051 	BUG_ON(i == le16_to_cpu(tmp_el->l_next_free_rec));
1052 
1053 out:
1054 	ocfs2_free_path(left_path);
1055 	ocfs2_free_path(right_path);
1056 	return ret;
1057 }
1058 
1059 /*
1060  * Given a cpos and len, try to find the refcount record which contains cpos.
1061  * 1. If cpos can be found in one refcount record, return the record.
1062  * 2. If cpos can't be found, return a fake record which start from cpos
1063  *    and end at a small value between cpos+len and start of the next record.
1064  *    This fake record has r_refcount = 0.
1065  */
1066 static int ocfs2_get_refcount_rec(struct ocfs2_caching_info *ci,
1067 				  struct buffer_head *ref_root_bh,
1068 				  u64 cpos, unsigned int len,
1069 				  struct ocfs2_refcount_rec *ret_rec,
1070 				  int *index,
1071 				  struct buffer_head **ret_bh)
1072 {
1073 	int ret = 0, i, found;
1074 	u32 low_cpos, uninitialized_var(cpos_end);
1075 	struct ocfs2_extent_list *el;
1076 	struct ocfs2_extent_rec *rec = NULL;
1077 	struct ocfs2_extent_block *eb = NULL;
1078 	struct buffer_head *eb_bh = NULL, *ref_leaf_bh = NULL;
1079 	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1080 	struct ocfs2_refcount_block *rb =
1081 			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
1082 
1083 	if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)) {
1084 		ocfs2_find_refcount_rec_in_rl(ci, ref_root_bh, cpos, len,
1085 					      ret_rec, index);
1086 		*ret_bh = ref_root_bh;
1087 		get_bh(ref_root_bh);
1088 		return 0;
1089 	}
1090 
1091 	el = &rb->rf_list;
1092 	low_cpos = cpos & OCFS2_32BIT_POS_MASK;
1093 
1094 	if (el->l_tree_depth) {
1095 		ret = ocfs2_find_leaf(ci, el, low_cpos, &eb_bh);
1096 		if (ret) {
1097 			mlog_errno(ret);
1098 			goto out;
1099 		}
1100 
1101 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1102 		el = &eb->h_list;
1103 
1104 		if (el->l_tree_depth) {
1105 			ocfs2_error(sb,
1106 			"refcount tree %llu has non zero tree "
1107 			"depth in leaf btree tree block %llu\n",
1108 			(unsigned long long)ocfs2_metadata_cache_owner(ci),
1109 			(unsigned long long)eb_bh->b_blocknr);
1110 			ret = -EROFS;
1111 			goto out;
1112 		}
1113 	}
1114 
1115 	found = 0;
1116 	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
1117 		rec = &el->l_recs[i];
1118 
1119 		if (le32_to_cpu(rec->e_cpos) <= low_cpos) {
1120 			found = 1;
1121 			break;
1122 		}
1123 	}
1124 
1125 	if (found) {
1126 		ret = ocfs2_get_refcount_cpos_end(ci, ref_root_bh,
1127 						  eb, el, i, &cpos_end);
1128 		if (ret) {
1129 			mlog_errno(ret);
1130 			goto out;
1131 		}
1132 
1133 		if (cpos_end < low_cpos + len)
1134 			len = cpos_end - low_cpos;
1135 	}
1136 
1137 	ret = ocfs2_read_refcount_block(ci, le64_to_cpu(rec->e_blkno),
1138 					&ref_leaf_bh);
1139 	if (ret) {
1140 		mlog_errno(ret);
1141 		goto out;
1142 	}
1143 
1144 	ocfs2_find_refcount_rec_in_rl(ci, ref_leaf_bh, cpos, len,
1145 				      ret_rec, index);
1146 	*ret_bh = ref_leaf_bh;
1147 out:
1148 	brelse(eb_bh);
1149 	return ret;
1150 }
1151 
1152 enum ocfs2_ref_rec_contig {
1153 	REF_CONTIG_NONE = 0,
1154 	REF_CONTIG_LEFT,
1155 	REF_CONTIG_RIGHT,
1156 	REF_CONTIG_LEFTRIGHT,
1157 };
1158 
1159 static enum ocfs2_ref_rec_contig
1160 	ocfs2_refcount_rec_adjacent(struct ocfs2_refcount_block *rb,
1161 				    int index)
1162 {
1163 	if ((rb->rf_records.rl_recs[index].r_refcount ==
1164 	    rb->rf_records.rl_recs[index + 1].r_refcount) &&
1165 	    (le64_to_cpu(rb->rf_records.rl_recs[index].r_cpos) +
1166 	    le32_to_cpu(rb->rf_records.rl_recs[index].r_clusters) ==
1167 	    le64_to_cpu(rb->rf_records.rl_recs[index + 1].r_cpos)))
1168 		return REF_CONTIG_RIGHT;
1169 
1170 	return REF_CONTIG_NONE;
1171 }
1172 
1173 static enum ocfs2_ref_rec_contig
1174 	ocfs2_refcount_rec_contig(struct ocfs2_refcount_block *rb,
1175 				  int index)
1176 {
1177 	enum ocfs2_ref_rec_contig ret = REF_CONTIG_NONE;
1178 
1179 	if (index < le16_to_cpu(rb->rf_records.rl_used) - 1)
1180 		ret = ocfs2_refcount_rec_adjacent(rb, index);
1181 
1182 	if (index > 0) {
1183 		enum ocfs2_ref_rec_contig tmp;
1184 
1185 		tmp = ocfs2_refcount_rec_adjacent(rb, index - 1);
1186 
1187 		if (tmp == REF_CONTIG_RIGHT) {
1188 			if (ret == REF_CONTIG_RIGHT)
1189 				ret = REF_CONTIG_LEFTRIGHT;
1190 			else
1191 				ret = REF_CONTIG_LEFT;
1192 		}
1193 	}
1194 
1195 	return ret;
1196 }
1197 
1198 static void ocfs2_rotate_refcount_rec_left(struct ocfs2_refcount_block *rb,
1199 					   int index)
1200 {
1201 	BUG_ON(rb->rf_records.rl_recs[index].r_refcount !=
1202 	       rb->rf_records.rl_recs[index+1].r_refcount);
1203 
1204 	le32_add_cpu(&rb->rf_records.rl_recs[index].r_clusters,
1205 		     le32_to_cpu(rb->rf_records.rl_recs[index+1].r_clusters));
1206 
1207 	if (index < le16_to_cpu(rb->rf_records.rl_used) - 2)
1208 		memmove(&rb->rf_records.rl_recs[index + 1],
1209 			&rb->rf_records.rl_recs[index + 2],
1210 			sizeof(struct ocfs2_refcount_rec) *
1211 			(le16_to_cpu(rb->rf_records.rl_used) - index - 2));
1212 
1213 	memset(&rb->rf_records.rl_recs[le16_to_cpu(rb->rf_records.rl_used) - 1],
1214 	       0, sizeof(struct ocfs2_refcount_rec));
1215 	le16_add_cpu(&rb->rf_records.rl_used, -1);
1216 }
1217 
1218 /*
1219  * Merge the refcount rec if we are contiguous with the adjacent recs.
1220  */
1221 static void ocfs2_refcount_rec_merge(struct ocfs2_refcount_block *rb,
1222 				     int index)
1223 {
1224 	enum ocfs2_ref_rec_contig contig =
1225 				ocfs2_refcount_rec_contig(rb, index);
1226 
1227 	if (contig == REF_CONTIG_NONE)
1228 		return;
1229 
1230 	if (contig == REF_CONTIG_LEFT || contig == REF_CONTIG_LEFTRIGHT) {
1231 		BUG_ON(index == 0);
1232 		index--;
1233 	}
1234 
1235 	ocfs2_rotate_refcount_rec_left(rb, index);
1236 
1237 	if (contig == REF_CONTIG_LEFTRIGHT)
1238 		ocfs2_rotate_refcount_rec_left(rb, index);
1239 }
1240 
1241 /*
1242  * Change the refcount indexed by "index" in ref_bh.
1243  * If refcount reaches 0, remove it.
1244  */
1245 static int ocfs2_change_refcount_rec(handle_t *handle,
1246 				     struct ocfs2_caching_info *ci,
1247 				     struct buffer_head *ref_leaf_bh,
1248 				     int index, int merge, int change)
1249 {
1250 	int ret;
1251 	struct ocfs2_refcount_block *rb =
1252 			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1253 	struct ocfs2_refcount_list *rl = &rb->rf_records;
1254 	struct ocfs2_refcount_rec *rec = &rl->rl_recs[index];
1255 
1256 	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1257 				      OCFS2_JOURNAL_ACCESS_WRITE);
1258 	if (ret) {
1259 		mlog_errno(ret);
1260 		goto out;
1261 	}
1262 
1263 	trace_ocfs2_change_refcount_rec(
1264 		(unsigned long long)ocfs2_metadata_cache_owner(ci),
1265 		index, le32_to_cpu(rec->r_refcount), change);
1266 	le32_add_cpu(&rec->r_refcount, change);
1267 
1268 	if (!rec->r_refcount) {
1269 		if (index != le16_to_cpu(rl->rl_used) - 1) {
1270 			memmove(rec, rec + 1,
1271 				(le16_to_cpu(rl->rl_used) - index - 1) *
1272 				sizeof(struct ocfs2_refcount_rec));
1273 			memset(&rl->rl_recs[le16_to_cpu(rl->rl_used) - 1],
1274 			       0, sizeof(struct ocfs2_refcount_rec));
1275 		}
1276 
1277 		le16_add_cpu(&rl->rl_used, -1);
1278 	} else if (merge)
1279 		ocfs2_refcount_rec_merge(rb, index);
1280 
1281 	ocfs2_journal_dirty(handle, ref_leaf_bh);
1282 out:
1283 	return ret;
1284 }
1285 
1286 static int ocfs2_expand_inline_ref_root(handle_t *handle,
1287 					struct ocfs2_caching_info *ci,
1288 					struct buffer_head *ref_root_bh,
1289 					struct buffer_head **ref_leaf_bh,
1290 					struct ocfs2_alloc_context *meta_ac)
1291 {
1292 	int ret;
1293 	u16 suballoc_bit_start;
1294 	u32 num_got;
1295 	u64 suballoc_loc, blkno;
1296 	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1297 	struct buffer_head *new_bh = NULL;
1298 	struct ocfs2_refcount_block *new_rb;
1299 	struct ocfs2_refcount_block *root_rb =
1300 			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
1301 
1302 	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1303 				      OCFS2_JOURNAL_ACCESS_WRITE);
1304 	if (ret) {
1305 		mlog_errno(ret);
1306 		goto out;
1307 	}
1308 
1309 	ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1310 				   &suballoc_bit_start, &num_got,
1311 				   &blkno);
1312 	if (ret) {
1313 		mlog_errno(ret);
1314 		goto out;
1315 	}
1316 
1317 	new_bh = sb_getblk(sb, blkno);
1318 	if (new_bh == NULL) {
1319 		ret = -ENOMEM;
1320 		mlog_errno(ret);
1321 		goto out;
1322 	}
1323 	ocfs2_set_new_buffer_uptodate(ci, new_bh);
1324 
1325 	ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1326 				      OCFS2_JOURNAL_ACCESS_CREATE);
1327 	if (ret) {
1328 		mlog_errno(ret);
1329 		goto out;
1330 	}
1331 
1332 	/*
1333 	 * Initialize ocfs2_refcount_block.
1334 	 * It should contain the same information as the old root.
1335 	 * so just memcpy it and change the corresponding field.
1336 	 */
1337 	memcpy(new_bh->b_data, ref_root_bh->b_data, sb->s_blocksize);
1338 
1339 	new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1340 	new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1341 	new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1342 	new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1343 	new_rb->rf_blkno = cpu_to_le64(blkno);
1344 	new_rb->rf_cpos = cpu_to_le32(0);
1345 	new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1346 	new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1347 	ocfs2_journal_dirty(handle, new_bh);
1348 
1349 	/* Now change the root. */
1350 	memset(&root_rb->rf_list, 0, sb->s_blocksize -
1351 	       offsetof(struct ocfs2_refcount_block, rf_list));
1352 	root_rb->rf_list.l_count = cpu_to_le16(ocfs2_extent_recs_per_rb(sb));
1353 	root_rb->rf_clusters = cpu_to_le32(1);
1354 	root_rb->rf_list.l_next_free_rec = cpu_to_le16(1);
1355 	root_rb->rf_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
1356 	root_rb->rf_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
1357 	root_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_TREE_FL);
1358 
1359 	ocfs2_journal_dirty(handle, ref_root_bh);
1360 
1361 	trace_ocfs2_expand_inline_ref_root((unsigned long long)blkno,
1362 		le16_to_cpu(new_rb->rf_records.rl_used));
1363 
1364 	*ref_leaf_bh = new_bh;
1365 	new_bh = NULL;
1366 out:
1367 	brelse(new_bh);
1368 	return ret;
1369 }
1370 
1371 static int ocfs2_refcount_rec_no_intersect(struct ocfs2_refcount_rec *prev,
1372 					   struct ocfs2_refcount_rec *next)
1373 {
1374 	if (ocfs2_get_ref_rec_low_cpos(prev) + le32_to_cpu(prev->r_clusters) <=
1375 		ocfs2_get_ref_rec_low_cpos(next))
1376 		return 1;
1377 
1378 	return 0;
1379 }
1380 
1381 static int cmp_refcount_rec_by_low_cpos(const void *a, const void *b)
1382 {
1383 	const struct ocfs2_refcount_rec *l = a, *r = b;
1384 	u32 l_cpos = ocfs2_get_ref_rec_low_cpos(l);
1385 	u32 r_cpos = ocfs2_get_ref_rec_low_cpos(r);
1386 
1387 	if (l_cpos > r_cpos)
1388 		return 1;
1389 	if (l_cpos < r_cpos)
1390 		return -1;
1391 	return 0;
1392 }
1393 
1394 static int cmp_refcount_rec_by_cpos(const void *a, const void *b)
1395 {
1396 	const struct ocfs2_refcount_rec *l = a, *r = b;
1397 	u64 l_cpos = le64_to_cpu(l->r_cpos);
1398 	u64 r_cpos = le64_to_cpu(r->r_cpos);
1399 
1400 	if (l_cpos > r_cpos)
1401 		return 1;
1402 	if (l_cpos < r_cpos)
1403 		return -1;
1404 	return 0;
1405 }
1406 
1407 static void swap_refcount_rec(void *a, void *b, int size)
1408 {
1409 	struct ocfs2_refcount_rec *l = a, *r = b, tmp;
1410 
1411 	tmp = *l;
1412 	*l = *r;
1413 	*r = tmp;
1414 }
1415 
1416 /*
1417  * The refcount cpos are ordered by their 64bit cpos,
1418  * But we will use the low 32 bit to be the e_cpos in the b-tree.
1419  * So we need to make sure that this pos isn't intersected with others.
1420  *
1421  * Note: The refcount block is already sorted by their low 32 bit cpos,
1422  *       So just try the middle pos first, and we will exit when we find
1423  *       the good position.
1424  */
1425 static int ocfs2_find_refcount_split_pos(struct ocfs2_refcount_list *rl,
1426 					 u32 *split_pos, int *split_index)
1427 {
1428 	int num_used = le16_to_cpu(rl->rl_used);
1429 	int delta, middle = num_used / 2;
1430 
1431 	for (delta = 0; delta < middle; delta++) {
1432 		/* Let's check delta earlier than middle */
1433 		if (ocfs2_refcount_rec_no_intersect(
1434 					&rl->rl_recs[middle - delta - 1],
1435 					&rl->rl_recs[middle - delta])) {
1436 			*split_index = middle - delta;
1437 			break;
1438 		}
1439 
1440 		/* For even counts, don't walk off the end */
1441 		if ((middle + delta + 1) == num_used)
1442 			continue;
1443 
1444 		/* Now try delta past middle */
1445 		if (ocfs2_refcount_rec_no_intersect(
1446 					&rl->rl_recs[middle + delta],
1447 					&rl->rl_recs[middle + delta + 1])) {
1448 			*split_index = middle + delta + 1;
1449 			break;
1450 		}
1451 	}
1452 
1453 	if (delta >= middle)
1454 		return -ENOSPC;
1455 
1456 	*split_pos = ocfs2_get_ref_rec_low_cpos(&rl->rl_recs[*split_index]);
1457 	return 0;
1458 }
1459 
1460 static int ocfs2_divide_leaf_refcount_block(struct buffer_head *ref_leaf_bh,
1461 					    struct buffer_head *new_bh,
1462 					    u32 *split_cpos)
1463 {
1464 	int split_index = 0, num_moved, ret;
1465 	u32 cpos = 0;
1466 	struct ocfs2_refcount_block *rb =
1467 			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1468 	struct ocfs2_refcount_list *rl = &rb->rf_records;
1469 	struct ocfs2_refcount_block *new_rb =
1470 			(struct ocfs2_refcount_block *)new_bh->b_data;
1471 	struct ocfs2_refcount_list *new_rl = &new_rb->rf_records;
1472 
1473 	trace_ocfs2_divide_leaf_refcount_block(
1474 		(unsigned long long)ref_leaf_bh->b_blocknr,
1475 		le16_to_cpu(rl->rl_count), le16_to_cpu(rl->rl_used));
1476 
1477 	/*
1478 	 * XXX: Improvement later.
1479 	 * If we know all the high 32 bit cpos is the same, no need to sort.
1480 	 *
1481 	 * In order to make the whole process safe, we do:
1482 	 * 1. sort the entries by their low 32 bit cpos first so that we can
1483 	 *    find the split cpos easily.
1484 	 * 2. call ocfs2_insert_extent to insert the new refcount block.
1485 	 * 3. move the refcount rec to the new block.
1486 	 * 4. sort the entries by their 64 bit cpos.
1487 	 * 5. dirty the new_rb and rb.
1488 	 */
1489 	sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1490 	     sizeof(struct ocfs2_refcount_rec),
1491 	     cmp_refcount_rec_by_low_cpos, swap_refcount_rec);
1492 
1493 	ret = ocfs2_find_refcount_split_pos(rl, &cpos, &split_index);
1494 	if (ret) {
1495 		mlog_errno(ret);
1496 		return ret;
1497 	}
1498 
1499 	new_rb->rf_cpos = cpu_to_le32(cpos);
1500 
1501 	/* move refcount records starting from split_index to the new block. */
1502 	num_moved = le16_to_cpu(rl->rl_used) - split_index;
1503 	memcpy(new_rl->rl_recs, &rl->rl_recs[split_index],
1504 	       num_moved * sizeof(struct ocfs2_refcount_rec));
1505 
1506 	/*ok, remove the entries we just moved over to the other block. */
1507 	memset(&rl->rl_recs[split_index], 0,
1508 	       num_moved * sizeof(struct ocfs2_refcount_rec));
1509 
1510 	/* change old and new rl_used accordingly. */
1511 	le16_add_cpu(&rl->rl_used, -num_moved);
1512 	new_rl->rl_used = cpu_to_le16(num_moved);
1513 
1514 	sort(&rl->rl_recs, le16_to_cpu(rl->rl_used),
1515 	     sizeof(struct ocfs2_refcount_rec),
1516 	     cmp_refcount_rec_by_cpos, swap_refcount_rec);
1517 
1518 	sort(&new_rl->rl_recs, le16_to_cpu(new_rl->rl_used),
1519 	     sizeof(struct ocfs2_refcount_rec),
1520 	     cmp_refcount_rec_by_cpos, swap_refcount_rec);
1521 
1522 	*split_cpos = cpos;
1523 	return 0;
1524 }
1525 
1526 static int ocfs2_new_leaf_refcount_block(handle_t *handle,
1527 					 struct ocfs2_caching_info *ci,
1528 					 struct buffer_head *ref_root_bh,
1529 					 struct buffer_head *ref_leaf_bh,
1530 					 struct ocfs2_alloc_context *meta_ac)
1531 {
1532 	int ret;
1533 	u16 suballoc_bit_start;
1534 	u32 num_got, new_cpos;
1535 	u64 suballoc_loc, blkno;
1536 	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
1537 	struct ocfs2_refcount_block *root_rb =
1538 			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
1539 	struct buffer_head *new_bh = NULL;
1540 	struct ocfs2_refcount_block *new_rb;
1541 	struct ocfs2_extent_tree ref_et;
1542 
1543 	BUG_ON(!(le32_to_cpu(root_rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL));
1544 
1545 	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
1546 				      OCFS2_JOURNAL_ACCESS_WRITE);
1547 	if (ret) {
1548 		mlog_errno(ret);
1549 		goto out;
1550 	}
1551 
1552 	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1553 				      OCFS2_JOURNAL_ACCESS_WRITE);
1554 	if (ret) {
1555 		mlog_errno(ret);
1556 		goto out;
1557 	}
1558 
1559 	ret = ocfs2_claim_metadata(handle, meta_ac, 1, &suballoc_loc,
1560 				   &suballoc_bit_start, &num_got,
1561 				   &blkno);
1562 	if (ret) {
1563 		mlog_errno(ret);
1564 		goto out;
1565 	}
1566 
1567 	new_bh = sb_getblk(sb, blkno);
1568 	if (new_bh == NULL) {
1569 		ret = -ENOMEM;
1570 		mlog_errno(ret);
1571 		goto out;
1572 	}
1573 	ocfs2_set_new_buffer_uptodate(ci, new_bh);
1574 
1575 	ret = ocfs2_journal_access_rb(handle, ci, new_bh,
1576 				      OCFS2_JOURNAL_ACCESS_CREATE);
1577 	if (ret) {
1578 		mlog_errno(ret);
1579 		goto out;
1580 	}
1581 
1582 	/* Initialize ocfs2_refcount_block. */
1583 	new_rb = (struct ocfs2_refcount_block *)new_bh->b_data;
1584 	memset(new_rb, 0, sb->s_blocksize);
1585 	strcpy((void *)new_rb, OCFS2_REFCOUNT_BLOCK_SIGNATURE);
1586 	new_rb->rf_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
1587 	new_rb->rf_suballoc_loc = cpu_to_le64(suballoc_loc);
1588 	new_rb->rf_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1589 	new_rb->rf_fs_generation = cpu_to_le32(OCFS2_SB(sb)->fs_generation);
1590 	new_rb->rf_blkno = cpu_to_le64(blkno);
1591 	new_rb->rf_parent = cpu_to_le64(ref_root_bh->b_blocknr);
1592 	new_rb->rf_flags = cpu_to_le32(OCFS2_REFCOUNT_LEAF_FL);
1593 	new_rb->rf_records.rl_count =
1594 				cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
1595 	new_rb->rf_generation = root_rb->rf_generation;
1596 
1597 	ret = ocfs2_divide_leaf_refcount_block(ref_leaf_bh, new_bh, &new_cpos);
1598 	if (ret) {
1599 		mlog_errno(ret);
1600 		goto out;
1601 	}
1602 
1603 	ocfs2_journal_dirty(handle, ref_leaf_bh);
1604 	ocfs2_journal_dirty(handle, new_bh);
1605 
1606 	ocfs2_init_refcount_extent_tree(&ref_et, ci, ref_root_bh);
1607 
1608 	trace_ocfs2_new_leaf_refcount_block(
1609 			(unsigned long long)new_bh->b_blocknr, new_cpos);
1610 
1611 	/* Insert the new leaf block with the specific offset cpos. */
1612 	ret = ocfs2_insert_extent(handle, &ref_et, new_cpos, new_bh->b_blocknr,
1613 				  1, 0, meta_ac);
1614 	if (ret)
1615 		mlog_errno(ret);
1616 
1617 out:
1618 	brelse(new_bh);
1619 	return ret;
1620 }
1621 
1622 static int ocfs2_expand_refcount_tree(handle_t *handle,
1623 				      struct ocfs2_caching_info *ci,
1624 				      struct buffer_head *ref_root_bh,
1625 				      struct buffer_head *ref_leaf_bh,
1626 				      struct ocfs2_alloc_context *meta_ac)
1627 {
1628 	int ret;
1629 	struct buffer_head *expand_bh = NULL;
1630 
1631 	if (ref_root_bh == ref_leaf_bh) {
1632 		/*
1633 		 * the old root bh hasn't been expanded to a b-tree,
1634 		 * so expand it first.
1635 		 */
1636 		ret = ocfs2_expand_inline_ref_root(handle, ci, ref_root_bh,
1637 						   &expand_bh, meta_ac);
1638 		if (ret) {
1639 			mlog_errno(ret);
1640 			goto out;
1641 		}
1642 	} else {
1643 		expand_bh = ref_leaf_bh;
1644 		get_bh(expand_bh);
1645 	}
1646 
1647 
1648 	/* Now add a new refcount block into the tree.*/
1649 	ret = ocfs2_new_leaf_refcount_block(handle, ci, ref_root_bh,
1650 					    expand_bh, meta_ac);
1651 	if (ret)
1652 		mlog_errno(ret);
1653 out:
1654 	brelse(expand_bh);
1655 	return ret;
1656 }
1657 
1658 /*
1659  * Adjust the extent rec in b-tree representing ref_leaf_bh.
1660  *
1661  * Only called when we have inserted a new refcount rec at index 0
1662  * which means ocfs2_extent_rec.e_cpos may need some change.
1663  */
1664 static int ocfs2_adjust_refcount_rec(handle_t *handle,
1665 				     struct ocfs2_caching_info *ci,
1666 				     struct buffer_head *ref_root_bh,
1667 				     struct buffer_head *ref_leaf_bh,
1668 				     struct ocfs2_refcount_rec *rec)
1669 {
1670 	int ret = 0, i;
1671 	u32 new_cpos, old_cpos;
1672 	struct ocfs2_path *path = NULL;
1673 	struct ocfs2_extent_tree et;
1674 	struct ocfs2_refcount_block *rb =
1675 		(struct ocfs2_refcount_block *)ref_root_bh->b_data;
1676 	struct ocfs2_extent_list *el;
1677 
1678 	if (!(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL))
1679 		goto out;
1680 
1681 	rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1682 	old_cpos = le32_to_cpu(rb->rf_cpos);
1683 	new_cpos = le64_to_cpu(rec->r_cpos) & OCFS2_32BIT_POS_MASK;
1684 	if (old_cpos <= new_cpos)
1685 		goto out;
1686 
1687 	ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
1688 
1689 	path = ocfs2_new_path_from_et(&et);
1690 	if (!path) {
1691 		ret = -ENOMEM;
1692 		mlog_errno(ret);
1693 		goto out;
1694 	}
1695 
1696 	ret = ocfs2_find_path(ci, path, old_cpos);
1697 	if (ret) {
1698 		mlog_errno(ret);
1699 		goto out;
1700 	}
1701 
1702 	/*
1703 	 * 2 more credits, one for the leaf refcount block, one for
1704 	 * the extent block contains the extent rec.
1705 	 */
1706 	ret = ocfs2_extend_trans(handle, 2);
1707 	if (ret < 0) {
1708 		mlog_errno(ret);
1709 		goto out;
1710 	}
1711 
1712 	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1713 				      OCFS2_JOURNAL_ACCESS_WRITE);
1714 	if (ret < 0) {
1715 		mlog_errno(ret);
1716 		goto out;
1717 	}
1718 
1719 	ret = ocfs2_journal_access_eb(handle, ci, path_leaf_bh(path),
1720 				      OCFS2_JOURNAL_ACCESS_WRITE);
1721 	if (ret < 0) {
1722 		mlog_errno(ret);
1723 		goto out;
1724 	}
1725 
1726 	/* change the leaf extent block first. */
1727 	el = path_leaf_el(path);
1728 
1729 	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++)
1730 		if (le32_to_cpu(el->l_recs[i].e_cpos) == old_cpos)
1731 			break;
1732 
1733 	BUG_ON(i == le16_to_cpu(el->l_next_free_rec));
1734 
1735 	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1736 
1737 	/* change the r_cpos in the leaf block. */
1738 	rb->rf_cpos = cpu_to_le32(new_cpos);
1739 
1740 	ocfs2_journal_dirty(handle, path_leaf_bh(path));
1741 	ocfs2_journal_dirty(handle, ref_leaf_bh);
1742 
1743 out:
1744 	ocfs2_free_path(path);
1745 	return ret;
1746 }
1747 
1748 static int ocfs2_insert_refcount_rec(handle_t *handle,
1749 				     struct ocfs2_caching_info *ci,
1750 				     struct buffer_head *ref_root_bh,
1751 				     struct buffer_head *ref_leaf_bh,
1752 				     struct ocfs2_refcount_rec *rec,
1753 				     int index, int merge,
1754 				     struct ocfs2_alloc_context *meta_ac)
1755 {
1756 	int ret;
1757 	struct ocfs2_refcount_block *rb =
1758 			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1759 	struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1760 	struct buffer_head *new_bh = NULL;
1761 
1762 	BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1763 
1764 	if (rf_list->rl_used == rf_list->rl_count) {
1765 		u64 cpos = le64_to_cpu(rec->r_cpos);
1766 		u32 len = le32_to_cpu(rec->r_clusters);
1767 
1768 		ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1769 						 ref_leaf_bh, meta_ac);
1770 		if (ret) {
1771 			mlog_errno(ret);
1772 			goto out;
1773 		}
1774 
1775 		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1776 					     cpos, len, NULL, &index,
1777 					     &new_bh);
1778 		if (ret) {
1779 			mlog_errno(ret);
1780 			goto out;
1781 		}
1782 
1783 		ref_leaf_bh = new_bh;
1784 		rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1785 		rf_list = &rb->rf_records;
1786 	}
1787 
1788 	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1789 				      OCFS2_JOURNAL_ACCESS_WRITE);
1790 	if (ret) {
1791 		mlog_errno(ret);
1792 		goto out;
1793 	}
1794 
1795 	if (index < le16_to_cpu(rf_list->rl_used))
1796 		memmove(&rf_list->rl_recs[index + 1],
1797 			&rf_list->rl_recs[index],
1798 			(le16_to_cpu(rf_list->rl_used) - index) *
1799 			 sizeof(struct ocfs2_refcount_rec));
1800 
1801 	trace_ocfs2_insert_refcount_rec(
1802 		(unsigned long long)ref_leaf_bh->b_blocknr, index,
1803 		(unsigned long long)le64_to_cpu(rec->r_cpos),
1804 		le32_to_cpu(rec->r_clusters), le32_to_cpu(rec->r_refcount));
1805 
1806 	rf_list->rl_recs[index] = *rec;
1807 
1808 	le16_add_cpu(&rf_list->rl_used, 1);
1809 
1810 	if (merge)
1811 		ocfs2_refcount_rec_merge(rb, index);
1812 
1813 	ocfs2_journal_dirty(handle, ref_leaf_bh);
1814 
1815 	if (index == 0) {
1816 		ret = ocfs2_adjust_refcount_rec(handle, ci,
1817 						ref_root_bh,
1818 						ref_leaf_bh, rec);
1819 		if (ret)
1820 			mlog_errno(ret);
1821 	}
1822 out:
1823 	brelse(new_bh);
1824 	return ret;
1825 }
1826 
1827 /*
1828  * Split the refcount_rec indexed by "index" in ref_leaf_bh.
1829  * This is much simple than our b-tree code.
1830  * split_rec is the new refcount rec we want to insert.
1831  * If split_rec->r_refcount > 0, we are changing the refcount(in case we
1832  * increase refcount or decrease a refcount to non-zero).
1833  * If split_rec->r_refcount == 0, we are punching a hole in current refcount
1834  * rec( in case we decrease a refcount to zero).
1835  */
1836 static int ocfs2_split_refcount_rec(handle_t *handle,
1837 				    struct ocfs2_caching_info *ci,
1838 				    struct buffer_head *ref_root_bh,
1839 				    struct buffer_head *ref_leaf_bh,
1840 				    struct ocfs2_refcount_rec *split_rec,
1841 				    int index, int merge,
1842 				    struct ocfs2_alloc_context *meta_ac,
1843 				    struct ocfs2_cached_dealloc_ctxt *dealloc)
1844 {
1845 	int ret, recs_need;
1846 	u32 len;
1847 	struct ocfs2_refcount_block *rb =
1848 			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1849 	struct ocfs2_refcount_list *rf_list = &rb->rf_records;
1850 	struct ocfs2_refcount_rec *orig_rec = &rf_list->rl_recs[index];
1851 	struct ocfs2_refcount_rec *tail_rec = NULL;
1852 	struct buffer_head *new_bh = NULL;
1853 
1854 	BUG_ON(le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL);
1855 
1856 	trace_ocfs2_split_refcount_rec(le64_to_cpu(orig_rec->r_cpos),
1857 		le32_to_cpu(orig_rec->r_clusters),
1858 		le32_to_cpu(orig_rec->r_refcount),
1859 		le64_to_cpu(split_rec->r_cpos),
1860 		le32_to_cpu(split_rec->r_clusters),
1861 		le32_to_cpu(split_rec->r_refcount));
1862 
1863 	/*
1864 	 * If we just need to split the header or tail clusters,
1865 	 * no more recs are needed, just split is OK.
1866 	 * Otherwise we at least need one new recs.
1867 	 */
1868 	if (!split_rec->r_refcount &&
1869 	    (split_rec->r_cpos == orig_rec->r_cpos ||
1870 	     le64_to_cpu(split_rec->r_cpos) +
1871 	     le32_to_cpu(split_rec->r_clusters) ==
1872 	     le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1873 		recs_need = 0;
1874 	else
1875 		recs_need = 1;
1876 
1877 	/*
1878 	 * We need one more rec if we split in the middle and the new rec have
1879 	 * some refcount in it.
1880 	 */
1881 	if (split_rec->r_refcount &&
1882 	    (split_rec->r_cpos != orig_rec->r_cpos &&
1883 	     le64_to_cpu(split_rec->r_cpos) +
1884 	     le32_to_cpu(split_rec->r_clusters) !=
1885 	     le64_to_cpu(orig_rec->r_cpos) + le32_to_cpu(orig_rec->r_clusters)))
1886 		recs_need++;
1887 
1888 	/* If the leaf block don't have enough record, expand it. */
1889 	if (le16_to_cpu(rf_list->rl_used) + recs_need >
1890 					 le16_to_cpu(rf_list->rl_count)) {
1891 		struct ocfs2_refcount_rec tmp_rec;
1892 		u64 cpos = le64_to_cpu(orig_rec->r_cpos);
1893 		len = le32_to_cpu(orig_rec->r_clusters);
1894 		ret = ocfs2_expand_refcount_tree(handle, ci, ref_root_bh,
1895 						 ref_leaf_bh, meta_ac);
1896 		if (ret) {
1897 			mlog_errno(ret);
1898 			goto out;
1899 		}
1900 
1901 		/*
1902 		 * We have to re-get it since now cpos may be moved to
1903 		 * another leaf block.
1904 		 */
1905 		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
1906 					     cpos, len, &tmp_rec, &index,
1907 					     &new_bh);
1908 		if (ret) {
1909 			mlog_errno(ret);
1910 			goto out;
1911 		}
1912 
1913 		ref_leaf_bh = new_bh;
1914 		rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
1915 		rf_list = &rb->rf_records;
1916 		orig_rec = &rf_list->rl_recs[index];
1917 	}
1918 
1919 	ret = ocfs2_journal_access_rb(handle, ci, ref_leaf_bh,
1920 				      OCFS2_JOURNAL_ACCESS_WRITE);
1921 	if (ret) {
1922 		mlog_errno(ret);
1923 		goto out;
1924 	}
1925 
1926 	/*
1927 	 * We have calculated out how many new records we need and store
1928 	 * in recs_need, so spare enough space first by moving the records
1929 	 * after "index" to the end.
1930 	 */
1931 	if (index != le16_to_cpu(rf_list->rl_used) - 1)
1932 		memmove(&rf_list->rl_recs[index + 1 + recs_need],
1933 			&rf_list->rl_recs[index + 1],
1934 			(le16_to_cpu(rf_list->rl_used) - index - 1) *
1935 			 sizeof(struct ocfs2_refcount_rec));
1936 
1937 	len = (le64_to_cpu(orig_rec->r_cpos) +
1938 	      le32_to_cpu(orig_rec->r_clusters)) -
1939 	      (le64_to_cpu(split_rec->r_cpos) +
1940 	      le32_to_cpu(split_rec->r_clusters));
1941 
1942 	/*
1943 	 * If we have "len", the we will split in the tail and move it
1944 	 * to the end of the space we have just spared.
1945 	 */
1946 	if (len) {
1947 		tail_rec = &rf_list->rl_recs[index + recs_need];
1948 
1949 		memcpy(tail_rec, orig_rec, sizeof(struct ocfs2_refcount_rec));
1950 		le64_add_cpu(&tail_rec->r_cpos,
1951 			     le32_to_cpu(tail_rec->r_clusters) - len);
1952 		tail_rec->r_clusters = cpu_to_le32(len);
1953 	}
1954 
1955 	/*
1956 	 * If the split pos isn't the same as the original one, we need to
1957 	 * split in the head.
1958 	 *
1959 	 * Note: We have the chance that split_rec.r_refcount = 0,
1960 	 * recs_need = 0 and len > 0, which means we just cut the head from
1961 	 * the orig_rec and in that case we have done some modification in
1962 	 * orig_rec above, so the check for r_cpos is faked.
1963 	 */
1964 	if (split_rec->r_cpos != orig_rec->r_cpos && tail_rec != orig_rec) {
1965 		len = le64_to_cpu(split_rec->r_cpos) -
1966 		      le64_to_cpu(orig_rec->r_cpos);
1967 		orig_rec->r_clusters = cpu_to_le32(len);
1968 		index++;
1969 	}
1970 
1971 	le16_add_cpu(&rf_list->rl_used, recs_need);
1972 
1973 	if (split_rec->r_refcount) {
1974 		rf_list->rl_recs[index] = *split_rec;
1975 		trace_ocfs2_split_refcount_rec_insert(
1976 			(unsigned long long)ref_leaf_bh->b_blocknr, index,
1977 			(unsigned long long)le64_to_cpu(split_rec->r_cpos),
1978 			le32_to_cpu(split_rec->r_clusters),
1979 			le32_to_cpu(split_rec->r_refcount));
1980 
1981 		if (merge)
1982 			ocfs2_refcount_rec_merge(rb, index);
1983 	}
1984 
1985 	ocfs2_journal_dirty(handle, ref_leaf_bh);
1986 
1987 out:
1988 	brelse(new_bh);
1989 	return ret;
1990 }
1991 
1992 static int __ocfs2_increase_refcount(handle_t *handle,
1993 				     struct ocfs2_caching_info *ci,
1994 				     struct buffer_head *ref_root_bh,
1995 				     u64 cpos, u32 len, int merge,
1996 				     struct ocfs2_alloc_context *meta_ac,
1997 				     struct ocfs2_cached_dealloc_ctxt *dealloc)
1998 {
1999 	int ret = 0, index;
2000 	struct buffer_head *ref_leaf_bh = NULL;
2001 	struct ocfs2_refcount_rec rec;
2002 	unsigned int set_len = 0;
2003 
2004 	trace_ocfs2_increase_refcount_begin(
2005 	     (unsigned long long)ocfs2_metadata_cache_owner(ci),
2006 	     (unsigned long long)cpos, len);
2007 
2008 	while (len) {
2009 		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2010 					     cpos, len, &rec, &index,
2011 					     &ref_leaf_bh);
2012 		if (ret) {
2013 			mlog_errno(ret);
2014 			goto out;
2015 		}
2016 
2017 		set_len = le32_to_cpu(rec.r_clusters);
2018 
2019 		/*
2020 		 * Here we may meet with 3 situations:
2021 		 *
2022 		 * 1. If we find an already existing record, and the length
2023 		 *    is the same, cool, we just need to increase the r_refcount
2024 		 *    and it is OK.
2025 		 * 2. If we find a hole, just insert it with r_refcount = 1.
2026 		 * 3. If we are in the middle of one extent record, split
2027 		 *    it.
2028 		 */
2029 		if (rec.r_refcount && le64_to_cpu(rec.r_cpos) == cpos &&
2030 		    set_len <= len) {
2031 			trace_ocfs2_increase_refcount_change(
2032 				(unsigned long long)cpos, set_len,
2033 				le32_to_cpu(rec.r_refcount));
2034 			ret = ocfs2_change_refcount_rec(handle, ci,
2035 							ref_leaf_bh, index,
2036 							merge, 1);
2037 			if (ret) {
2038 				mlog_errno(ret);
2039 				goto out;
2040 			}
2041 		} else if (!rec.r_refcount) {
2042 			rec.r_refcount = cpu_to_le32(1);
2043 
2044 			trace_ocfs2_increase_refcount_insert(
2045 			     (unsigned long long)le64_to_cpu(rec.r_cpos),
2046 			     set_len);
2047 			ret = ocfs2_insert_refcount_rec(handle, ci, ref_root_bh,
2048 							ref_leaf_bh,
2049 							&rec, index,
2050 							merge, meta_ac);
2051 			if (ret) {
2052 				mlog_errno(ret);
2053 				goto out;
2054 			}
2055 		} else  {
2056 			set_len = min((u64)(cpos + len),
2057 				      le64_to_cpu(rec.r_cpos) + set_len) - cpos;
2058 			rec.r_cpos = cpu_to_le64(cpos);
2059 			rec.r_clusters = cpu_to_le32(set_len);
2060 			le32_add_cpu(&rec.r_refcount, 1);
2061 
2062 			trace_ocfs2_increase_refcount_split(
2063 			     (unsigned long long)le64_to_cpu(rec.r_cpos),
2064 			     set_len, le32_to_cpu(rec.r_refcount));
2065 			ret = ocfs2_split_refcount_rec(handle, ci,
2066 						       ref_root_bh, ref_leaf_bh,
2067 						       &rec, index, merge,
2068 						       meta_ac, dealloc);
2069 			if (ret) {
2070 				mlog_errno(ret);
2071 				goto out;
2072 			}
2073 		}
2074 
2075 		cpos += set_len;
2076 		len -= set_len;
2077 		brelse(ref_leaf_bh);
2078 		ref_leaf_bh = NULL;
2079 	}
2080 
2081 out:
2082 	brelse(ref_leaf_bh);
2083 	return ret;
2084 }
2085 
2086 static int ocfs2_remove_refcount_extent(handle_t *handle,
2087 				struct ocfs2_caching_info *ci,
2088 				struct buffer_head *ref_root_bh,
2089 				struct buffer_head *ref_leaf_bh,
2090 				struct ocfs2_alloc_context *meta_ac,
2091 				struct ocfs2_cached_dealloc_ctxt *dealloc)
2092 {
2093 	int ret;
2094 	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2095 	struct ocfs2_refcount_block *rb =
2096 			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2097 	struct ocfs2_extent_tree et;
2098 
2099 	BUG_ON(rb->rf_records.rl_used);
2100 
2101 	trace_ocfs2_remove_refcount_extent(
2102 		(unsigned long long)ocfs2_metadata_cache_owner(ci),
2103 		(unsigned long long)ref_leaf_bh->b_blocknr,
2104 		le32_to_cpu(rb->rf_cpos));
2105 
2106 	ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2107 	ret = ocfs2_remove_extent(handle, &et, le32_to_cpu(rb->rf_cpos),
2108 				  1, meta_ac, dealloc);
2109 	if (ret) {
2110 		mlog_errno(ret);
2111 		goto out;
2112 	}
2113 
2114 	ocfs2_remove_from_cache(ci, ref_leaf_bh);
2115 
2116 	/*
2117 	 * add the freed block to the dealloc so that it will be freed
2118 	 * when we run dealloc.
2119 	 */
2120 	ret = ocfs2_cache_block_dealloc(dealloc, EXTENT_ALLOC_SYSTEM_INODE,
2121 					le16_to_cpu(rb->rf_suballoc_slot),
2122 					le64_to_cpu(rb->rf_suballoc_loc),
2123 					le64_to_cpu(rb->rf_blkno),
2124 					le16_to_cpu(rb->rf_suballoc_bit));
2125 	if (ret) {
2126 		mlog_errno(ret);
2127 		goto out;
2128 	}
2129 
2130 	ret = ocfs2_journal_access_rb(handle, ci, ref_root_bh,
2131 				      OCFS2_JOURNAL_ACCESS_WRITE);
2132 	if (ret) {
2133 		mlog_errno(ret);
2134 		goto out;
2135 	}
2136 
2137 	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2138 
2139 	le32_add_cpu(&rb->rf_clusters, -1);
2140 
2141 	/*
2142 	 * check whether we need to restore the root refcount block if
2143 	 * there is no leaf extent block at atll.
2144 	 */
2145 	if (!rb->rf_list.l_next_free_rec) {
2146 		BUG_ON(rb->rf_clusters);
2147 
2148 		trace_ocfs2_restore_refcount_block(
2149 		     (unsigned long long)ref_root_bh->b_blocknr);
2150 
2151 		rb->rf_flags = 0;
2152 		rb->rf_parent = 0;
2153 		rb->rf_cpos = 0;
2154 		memset(&rb->rf_records, 0, sb->s_blocksize -
2155 		       offsetof(struct ocfs2_refcount_block, rf_records));
2156 		rb->rf_records.rl_count =
2157 				cpu_to_le16(ocfs2_refcount_recs_per_rb(sb));
2158 	}
2159 
2160 	ocfs2_journal_dirty(handle, ref_root_bh);
2161 
2162 out:
2163 	return ret;
2164 }
2165 
2166 int ocfs2_increase_refcount(handle_t *handle,
2167 			    struct ocfs2_caching_info *ci,
2168 			    struct buffer_head *ref_root_bh,
2169 			    u64 cpos, u32 len,
2170 			    struct ocfs2_alloc_context *meta_ac,
2171 			    struct ocfs2_cached_dealloc_ctxt *dealloc)
2172 {
2173 	return __ocfs2_increase_refcount(handle, ci, ref_root_bh,
2174 					 cpos, len, 1,
2175 					 meta_ac, dealloc);
2176 }
2177 
2178 static int ocfs2_decrease_refcount_rec(handle_t *handle,
2179 				struct ocfs2_caching_info *ci,
2180 				struct buffer_head *ref_root_bh,
2181 				struct buffer_head *ref_leaf_bh,
2182 				int index, u64 cpos, unsigned int len,
2183 				struct ocfs2_alloc_context *meta_ac,
2184 				struct ocfs2_cached_dealloc_ctxt *dealloc)
2185 {
2186 	int ret;
2187 	struct ocfs2_refcount_block *rb =
2188 			(struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
2189 	struct ocfs2_refcount_rec *rec = &rb->rf_records.rl_recs[index];
2190 
2191 	BUG_ON(cpos < le64_to_cpu(rec->r_cpos));
2192 	BUG_ON(cpos + len >
2193 	       le64_to_cpu(rec->r_cpos) + le32_to_cpu(rec->r_clusters));
2194 
2195 	trace_ocfs2_decrease_refcount_rec(
2196 		(unsigned long long)ocfs2_metadata_cache_owner(ci),
2197 		(unsigned long long)cpos, len);
2198 
2199 	if (cpos == le64_to_cpu(rec->r_cpos) &&
2200 	    len == le32_to_cpu(rec->r_clusters))
2201 		ret = ocfs2_change_refcount_rec(handle, ci,
2202 						ref_leaf_bh, index, 1, -1);
2203 	else {
2204 		struct ocfs2_refcount_rec split = *rec;
2205 		split.r_cpos = cpu_to_le64(cpos);
2206 		split.r_clusters = cpu_to_le32(len);
2207 
2208 		le32_add_cpu(&split.r_refcount, -1);
2209 
2210 		ret = ocfs2_split_refcount_rec(handle, ci,
2211 					       ref_root_bh, ref_leaf_bh,
2212 					       &split, index, 1,
2213 					       meta_ac, dealloc);
2214 	}
2215 
2216 	if (ret) {
2217 		mlog_errno(ret);
2218 		goto out;
2219 	}
2220 
2221 	/* Remove the leaf refcount block if it contains no refcount record. */
2222 	if (!rb->rf_records.rl_used && ref_leaf_bh != ref_root_bh) {
2223 		ret = ocfs2_remove_refcount_extent(handle, ci, ref_root_bh,
2224 						   ref_leaf_bh, meta_ac,
2225 						   dealloc);
2226 		if (ret)
2227 			mlog_errno(ret);
2228 	}
2229 
2230 out:
2231 	return ret;
2232 }
2233 
2234 static int __ocfs2_decrease_refcount(handle_t *handle,
2235 				     struct ocfs2_caching_info *ci,
2236 				     struct buffer_head *ref_root_bh,
2237 				     u64 cpos, u32 len,
2238 				     struct ocfs2_alloc_context *meta_ac,
2239 				     struct ocfs2_cached_dealloc_ctxt *dealloc,
2240 				     int delete)
2241 {
2242 	int ret = 0, index = 0;
2243 	struct ocfs2_refcount_rec rec;
2244 	unsigned int r_count = 0, r_len;
2245 	struct super_block *sb = ocfs2_metadata_cache_get_super(ci);
2246 	struct buffer_head *ref_leaf_bh = NULL;
2247 
2248 	trace_ocfs2_decrease_refcount(
2249 		(unsigned long long)ocfs2_metadata_cache_owner(ci),
2250 		(unsigned long long)cpos, len, delete);
2251 
2252 	while (len) {
2253 		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2254 					     cpos, len, &rec, &index,
2255 					     &ref_leaf_bh);
2256 		if (ret) {
2257 			mlog_errno(ret);
2258 			goto out;
2259 		}
2260 
2261 		r_count = le32_to_cpu(rec.r_refcount);
2262 		BUG_ON(r_count == 0);
2263 		if (!delete)
2264 			BUG_ON(r_count > 1);
2265 
2266 		r_len = min((u64)(cpos + len), le64_to_cpu(rec.r_cpos) +
2267 			      le32_to_cpu(rec.r_clusters)) - cpos;
2268 
2269 		ret = ocfs2_decrease_refcount_rec(handle, ci, ref_root_bh,
2270 						  ref_leaf_bh, index,
2271 						  cpos, r_len,
2272 						  meta_ac, dealloc);
2273 		if (ret) {
2274 			mlog_errno(ret);
2275 			goto out;
2276 		}
2277 
2278 		if (le32_to_cpu(rec.r_refcount) == 1 && delete) {
2279 			ret = ocfs2_cache_cluster_dealloc(dealloc,
2280 					  ocfs2_clusters_to_blocks(sb, cpos),
2281 							  r_len);
2282 			if (ret) {
2283 				mlog_errno(ret);
2284 				goto out;
2285 			}
2286 		}
2287 
2288 		cpos += r_len;
2289 		len -= r_len;
2290 		brelse(ref_leaf_bh);
2291 		ref_leaf_bh = NULL;
2292 	}
2293 
2294 out:
2295 	brelse(ref_leaf_bh);
2296 	return ret;
2297 }
2298 
2299 /* Caller must hold refcount tree lock. */
2300 int ocfs2_decrease_refcount(struct inode *inode,
2301 			    handle_t *handle, u32 cpos, u32 len,
2302 			    struct ocfs2_alloc_context *meta_ac,
2303 			    struct ocfs2_cached_dealloc_ctxt *dealloc,
2304 			    int delete)
2305 {
2306 	int ret;
2307 	u64 ref_blkno;
2308 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2309 	struct buffer_head *ref_root_bh = NULL;
2310 	struct ocfs2_refcount_tree *tree;
2311 
2312 	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2313 
2314 	ret = ocfs2_get_refcount_block(inode, &ref_blkno);
2315 	if (ret) {
2316 		mlog_errno(ret);
2317 		goto out;
2318 	}
2319 
2320 	ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb), ref_blkno, &tree);
2321 	if (ret) {
2322 		mlog_errno(ret);
2323 		goto out;
2324 	}
2325 
2326 	ret = ocfs2_read_refcount_block(&tree->rf_ci, tree->rf_blkno,
2327 					&ref_root_bh);
2328 	if (ret) {
2329 		mlog_errno(ret);
2330 		goto out;
2331 	}
2332 
2333 	ret = __ocfs2_decrease_refcount(handle, &tree->rf_ci, ref_root_bh,
2334 					cpos, len, meta_ac, dealloc, delete);
2335 	if (ret)
2336 		mlog_errno(ret);
2337 out:
2338 	brelse(ref_root_bh);
2339 	return ret;
2340 }
2341 
2342 /*
2343  * Mark the already-existing extent at cpos as refcounted for len clusters.
2344  * This adds the refcount extent flag.
2345  *
2346  * If the existing extent is larger than the request, initiate a
2347  * split. An attempt will be made at merging with adjacent extents.
2348  *
2349  * The caller is responsible for passing down meta_ac if we'll need it.
2350  */
2351 static int ocfs2_mark_extent_refcounted(struct inode *inode,
2352 				struct ocfs2_extent_tree *et,
2353 				handle_t *handle, u32 cpos,
2354 				u32 len, u32 phys,
2355 				struct ocfs2_alloc_context *meta_ac,
2356 				struct ocfs2_cached_dealloc_ctxt *dealloc)
2357 {
2358 	int ret;
2359 
2360 	trace_ocfs2_mark_extent_refcounted(OCFS2_I(inode)->ip_blkno,
2361 					   cpos, len, phys);
2362 
2363 	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2364 		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2365 			    "tree, but the feature bit is not set in the "
2366 			    "super block.", inode->i_ino);
2367 		ret = -EROFS;
2368 		goto out;
2369 	}
2370 
2371 	ret = ocfs2_change_extent_flag(handle, et, cpos,
2372 				       len, phys, meta_ac, dealloc,
2373 				       OCFS2_EXT_REFCOUNTED, 0);
2374 	if (ret)
2375 		mlog_errno(ret);
2376 
2377 out:
2378 	return ret;
2379 }
2380 
2381 /*
2382  * Given some contiguous physical clusters, calculate what we need
2383  * for modifying their refcount.
2384  */
2385 static int ocfs2_calc_refcount_meta_credits(struct super_block *sb,
2386 					    struct ocfs2_caching_info *ci,
2387 					    struct buffer_head *ref_root_bh,
2388 					    u64 start_cpos,
2389 					    u32 clusters,
2390 					    int *meta_add,
2391 					    int *credits)
2392 {
2393 	int ret = 0, index, ref_blocks = 0, recs_add = 0;
2394 	u64 cpos = start_cpos;
2395 	struct ocfs2_refcount_block *rb;
2396 	struct ocfs2_refcount_rec rec;
2397 	struct buffer_head *ref_leaf_bh = NULL, *prev_bh = NULL;
2398 	u32 len;
2399 
2400 	while (clusters) {
2401 		ret = ocfs2_get_refcount_rec(ci, ref_root_bh,
2402 					     cpos, clusters, &rec,
2403 					     &index, &ref_leaf_bh);
2404 		if (ret) {
2405 			mlog_errno(ret);
2406 			goto out;
2407 		}
2408 
2409 		if (ref_leaf_bh != prev_bh) {
2410 			/*
2411 			 * Now we encounter a new leaf block, so calculate
2412 			 * whether we need to extend the old leaf.
2413 			 */
2414 			if (prev_bh) {
2415 				rb = (struct ocfs2_refcount_block *)
2416 							prev_bh->b_data;
2417 
2418 				if (le16_to_cpu(rb->rf_records.rl_used) +
2419 				    recs_add >
2420 				    le16_to_cpu(rb->rf_records.rl_count))
2421 					ref_blocks++;
2422 			}
2423 
2424 			recs_add = 0;
2425 			*credits += 1;
2426 			brelse(prev_bh);
2427 			prev_bh = ref_leaf_bh;
2428 			get_bh(prev_bh);
2429 		}
2430 
2431 		trace_ocfs2_calc_refcount_meta_credits_iterate(
2432 				recs_add, (unsigned long long)cpos, clusters,
2433 				(unsigned long long)le64_to_cpu(rec.r_cpos),
2434 				le32_to_cpu(rec.r_clusters),
2435 				le32_to_cpu(rec.r_refcount), index);
2436 
2437 		len = min((u64)cpos + clusters, le64_to_cpu(rec.r_cpos) +
2438 			  le32_to_cpu(rec.r_clusters)) - cpos;
2439 		/*
2440 		 * We record all the records which will be inserted to the
2441 		 * same refcount block, so that we can tell exactly whether
2442 		 * we need a new refcount block or not.
2443 		 *
2444 		 * If we will insert a new one, this is easy and only happens
2445 		 * during adding refcounted flag to the extent, so we don't
2446 		 * have a chance of spliting. We just need one record.
2447 		 *
2448 		 * If the refcount rec already exists, that would be a little
2449 		 * complicated. we may have to:
2450 		 * 1) split at the beginning if the start pos isn't aligned.
2451 		 *    we need 1 more record in this case.
2452 		 * 2) split int the end if the end pos isn't aligned.
2453 		 *    we need 1 more record in this case.
2454 		 * 3) split in the middle because of file system fragmentation.
2455 		 *    we need 2 more records in this case(we can't detect this
2456 		 *    beforehand, so always think of the worst case).
2457 		 */
2458 		if (rec.r_refcount) {
2459 			recs_add += 2;
2460 			/* Check whether we need a split at the beginning. */
2461 			if (cpos == start_cpos &&
2462 			    cpos != le64_to_cpu(rec.r_cpos))
2463 				recs_add++;
2464 
2465 			/* Check whether we need a split in the end. */
2466 			if (cpos + clusters < le64_to_cpu(rec.r_cpos) +
2467 			    le32_to_cpu(rec.r_clusters))
2468 				recs_add++;
2469 		} else
2470 			recs_add++;
2471 
2472 		brelse(ref_leaf_bh);
2473 		ref_leaf_bh = NULL;
2474 		clusters -= len;
2475 		cpos += len;
2476 	}
2477 
2478 	if (prev_bh) {
2479 		rb = (struct ocfs2_refcount_block *)prev_bh->b_data;
2480 
2481 		if (le16_to_cpu(rb->rf_records.rl_used) + recs_add >
2482 		    le16_to_cpu(rb->rf_records.rl_count))
2483 			ref_blocks++;
2484 
2485 		*credits += 1;
2486 	}
2487 
2488 	if (!ref_blocks)
2489 		goto out;
2490 
2491 	*meta_add += ref_blocks;
2492 	*credits += ref_blocks;
2493 
2494 	/*
2495 	 * So we may need ref_blocks to insert into the tree.
2496 	 * That also means we need to change the b-tree and add that number
2497 	 * of records since we never merge them.
2498 	 * We need one more block for expansion since the new created leaf
2499 	 * block is also full and needs split.
2500 	 */
2501 	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
2502 	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL) {
2503 		struct ocfs2_extent_tree et;
2504 
2505 		ocfs2_init_refcount_extent_tree(&et, ci, ref_root_bh);
2506 		*meta_add += ocfs2_extend_meta_needed(et.et_root_el);
2507 		*credits += ocfs2_calc_extend_credits(sb,
2508 						      et.et_root_el);
2509 	} else {
2510 		*credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
2511 		*meta_add += 1;
2512 	}
2513 
2514 out:
2515 
2516 	trace_ocfs2_calc_refcount_meta_credits(
2517 		(unsigned long long)start_cpos, clusters,
2518 		*meta_add, *credits);
2519 	brelse(ref_leaf_bh);
2520 	brelse(prev_bh);
2521 	return ret;
2522 }
2523 
2524 /*
2525  * For refcount tree, we will decrease some contiguous clusters
2526  * refcount count, so just go through it to see how many blocks
2527  * we gonna touch and whether we need to create new blocks.
2528  *
2529  * Normally the refcount blocks store these refcount should be
2530  * contiguous also, so that we can get the number easily.
2531  * We will at most add split 2 refcount records and 2 more
2532  * refcount blocks, so just check it in a rough way.
2533  *
2534  * Caller must hold refcount tree lock.
2535  */
2536 int ocfs2_prepare_refcount_change_for_del(struct inode *inode,
2537 					  u64 refcount_loc,
2538 					  u64 phys_blkno,
2539 					  u32 clusters,
2540 					  int *credits,
2541 					  int *ref_blocks)
2542 {
2543 	int ret;
2544 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2545 	struct buffer_head *ref_root_bh = NULL;
2546 	struct ocfs2_refcount_tree *tree;
2547 	u64 start_cpos = ocfs2_blocks_to_clusters(inode->i_sb, phys_blkno);
2548 
2549 	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
2550 		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
2551 			    "tree, but the feature bit is not set in the "
2552 			    "super block.", inode->i_ino);
2553 		ret = -EROFS;
2554 		goto out;
2555 	}
2556 
2557 	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
2558 
2559 	ret = ocfs2_get_refcount_tree(OCFS2_SB(inode->i_sb),
2560 				      refcount_loc, &tree);
2561 	if (ret) {
2562 		mlog_errno(ret);
2563 		goto out;
2564 	}
2565 
2566 	ret = ocfs2_read_refcount_block(&tree->rf_ci, refcount_loc,
2567 					&ref_root_bh);
2568 	if (ret) {
2569 		mlog_errno(ret);
2570 		goto out;
2571 	}
2572 
2573 	ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
2574 					       &tree->rf_ci,
2575 					       ref_root_bh,
2576 					       start_cpos, clusters,
2577 					       ref_blocks, credits);
2578 	if (ret) {
2579 		mlog_errno(ret);
2580 		goto out;
2581 	}
2582 
2583 	trace_ocfs2_prepare_refcount_change_for_del(*ref_blocks, *credits);
2584 
2585 out:
2586 	brelse(ref_root_bh);
2587 	return ret;
2588 }
2589 
2590 #define	MAX_CONTIG_BYTES	1048576
2591 
2592 static inline unsigned int ocfs2_cow_contig_clusters(struct super_block *sb)
2593 {
2594 	return ocfs2_clusters_for_bytes(sb, MAX_CONTIG_BYTES);
2595 }
2596 
2597 static inline unsigned int ocfs2_cow_contig_mask(struct super_block *sb)
2598 {
2599 	return ~(ocfs2_cow_contig_clusters(sb) - 1);
2600 }
2601 
2602 /*
2603  * Given an extent that starts at 'start' and an I/O that starts at 'cpos',
2604  * find an offset (start + (n * contig_clusters)) that is closest to cpos
2605  * while still being less than or equal to it.
2606  *
2607  * The goal is to break the extent at a multiple of contig_clusters.
2608  */
2609 static inline unsigned int ocfs2_cow_align_start(struct super_block *sb,
2610 						 unsigned int start,
2611 						 unsigned int cpos)
2612 {
2613 	BUG_ON(start > cpos);
2614 
2615 	return start + ((cpos - start) & ocfs2_cow_contig_mask(sb));
2616 }
2617 
2618 /*
2619  * Given a cluster count of len, pad it out so that it is a multiple
2620  * of contig_clusters.
2621  */
2622 static inline unsigned int ocfs2_cow_align_length(struct super_block *sb,
2623 						  unsigned int len)
2624 {
2625 	unsigned int padded =
2626 		(len + (ocfs2_cow_contig_clusters(sb) - 1)) &
2627 		ocfs2_cow_contig_mask(sb);
2628 
2629 	/* Did we wrap? */
2630 	if (padded < len)
2631 		padded = UINT_MAX;
2632 
2633 	return padded;
2634 }
2635 
2636 /*
2637  * Calculate out the start and number of virtual clusters we need to to CoW.
2638  *
2639  * cpos is vitual start cluster position we want to do CoW in a
2640  * file and write_len is the cluster length.
2641  * max_cpos is the place where we want to stop CoW intentionally.
2642  *
2643  * Normal we will start CoW from the beginning of extent record cotaining cpos.
2644  * We try to break up extents on boundaries of MAX_CONTIG_BYTES so that we
2645  * get good I/O from the resulting extent tree.
2646  */
2647 static int ocfs2_refcount_cal_cow_clusters(struct inode *inode,
2648 					   struct ocfs2_extent_list *el,
2649 					   u32 cpos,
2650 					   u32 write_len,
2651 					   u32 max_cpos,
2652 					   u32 *cow_start,
2653 					   u32 *cow_len)
2654 {
2655 	int ret = 0;
2656 	int tree_height = le16_to_cpu(el->l_tree_depth), i;
2657 	struct buffer_head *eb_bh = NULL;
2658 	struct ocfs2_extent_block *eb = NULL;
2659 	struct ocfs2_extent_rec *rec;
2660 	unsigned int want_clusters, rec_end = 0;
2661 	int contig_clusters = ocfs2_cow_contig_clusters(inode->i_sb);
2662 	int leaf_clusters;
2663 
2664 	BUG_ON(cpos + write_len > max_cpos);
2665 
2666 	if (tree_height > 0) {
2667 		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, cpos, &eb_bh);
2668 		if (ret) {
2669 			mlog_errno(ret);
2670 			goto out;
2671 		}
2672 
2673 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2674 		el = &eb->h_list;
2675 
2676 		if (el->l_tree_depth) {
2677 			ocfs2_error(inode->i_sb,
2678 				    "Inode %lu has non zero tree depth in "
2679 				    "leaf block %llu\n", inode->i_ino,
2680 				    (unsigned long long)eb_bh->b_blocknr);
2681 			ret = -EROFS;
2682 			goto out;
2683 		}
2684 	}
2685 
2686 	*cow_len = 0;
2687 	for (i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
2688 		rec = &el->l_recs[i];
2689 
2690 		if (ocfs2_is_empty_extent(rec)) {
2691 			mlog_bug_on_msg(i != 0, "Inode %lu has empty record in "
2692 					"index %d\n", inode->i_ino, i);
2693 			continue;
2694 		}
2695 
2696 		if (le32_to_cpu(rec->e_cpos) +
2697 		    le16_to_cpu(rec->e_leaf_clusters) <= cpos)
2698 			continue;
2699 
2700 		if (*cow_len == 0) {
2701 			/*
2702 			 * We should find a refcounted record in the
2703 			 * first pass.
2704 			 */
2705 			BUG_ON(!(rec->e_flags & OCFS2_EXT_REFCOUNTED));
2706 			*cow_start = le32_to_cpu(rec->e_cpos);
2707 		}
2708 
2709 		/*
2710 		 * If we encounter a hole, a non-refcounted record or
2711 		 * pass the max_cpos, stop the search.
2712 		 */
2713 		if ((!(rec->e_flags & OCFS2_EXT_REFCOUNTED)) ||
2714 		    (*cow_len && rec_end != le32_to_cpu(rec->e_cpos)) ||
2715 		    (max_cpos <= le32_to_cpu(rec->e_cpos)))
2716 			break;
2717 
2718 		leaf_clusters = le16_to_cpu(rec->e_leaf_clusters);
2719 		rec_end = le32_to_cpu(rec->e_cpos) + leaf_clusters;
2720 		if (rec_end > max_cpos) {
2721 			rec_end = max_cpos;
2722 			leaf_clusters = rec_end - le32_to_cpu(rec->e_cpos);
2723 		}
2724 
2725 		/*
2726 		 * How many clusters do we actually need from
2727 		 * this extent?  First we see how many we actually
2728 		 * need to complete the write.  If that's smaller
2729 		 * than contig_clusters, we try for contig_clusters.
2730 		 */
2731 		if (!*cow_len)
2732 			want_clusters = write_len;
2733 		else
2734 			want_clusters = (cpos + write_len) -
2735 				(*cow_start + *cow_len);
2736 		if (want_clusters < contig_clusters)
2737 			want_clusters = contig_clusters;
2738 
2739 		/*
2740 		 * If the write does not cover the whole extent, we
2741 		 * need to calculate how we're going to split the extent.
2742 		 * We try to do it on contig_clusters boundaries.
2743 		 *
2744 		 * Any extent smaller than contig_clusters will be
2745 		 * CoWed in its entirety.
2746 		 */
2747 		if (leaf_clusters <= contig_clusters)
2748 			*cow_len += leaf_clusters;
2749 		else if (*cow_len || (*cow_start == cpos)) {
2750 			/*
2751 			 * This extent needs to be CoW'd from its
2752 			 * beginning, so all we have to do is compute
2753 			 * how many clusters to grab.  We align
2754 			 * want_clusters to the edge of contig_clusters
2755 			 * to get better I/O.
2756 			 */
2757 			want_clusters = ocfs2_cow_align_length(inode->i_sb,
2758 							       want_clusters);
2759 
2760 			if (leaf_clusters < want_clusters)
2761 				*cow_len += leaf_clusters;
2762 			else
2763 				*cow_len += want_clusters;
2764 		} else if ((*cow_start + contig_clusters) >=
2765 			   (cpos + write_len)) {
2766 			/*
2767 			 * Breaking off contig_clusters at the front
2768 			 * of the extent will cover our write.  That's
2769 			 * easy.
2770 			 */
2771 			*cow_len = contig_clusters;
2772 		} else if ((rec_end - cpos) <= contig_clusters) {
2773 			/*
2774 			 * Breaking off contig_clusters at the tail of
2775 			 * this extent will cover cpos.
2776 			 */
2777 			*cow_start = rec_end - contig_clusters;
2778 			*cow_len = contig_clusters;
2779 		} else if ((rec_end - cpos) <= want_clusters) {
2780 			/*
2781 			 * While we can't fit the entire write in this
2782 			 * extent, we know that the write goes from cpos
2783 			 * to the end of the extent.  Break that off.
2784 			 * We try to break it at some multiple of
2785 			 * contig_clusters from the front of the extent.
2786 			 * Failing that (ie, cpos is within
2787 			 * contig_clusters of the front), we'll CoW the
2788 			 * entire extent.
2789 			 */
2790 			*cow_start = ocfs2_cow_align_start(inode->i_sb,
2791 							   *cow_start, cpos);
2792 			*cow_len = rec_end - *cow_start;
2793 		} else {
2794 			/*
2795 			 * Ok, the entire write lives in the middle of
2796 			 * this extent.  Let's try to slice the extent up
2797 			 * nicely.  Optimally, our CoW region starts at
2798 			 * m*contig_clusters from the beginning of the
2799 			 * extent and goes for n*contig_clusters,
2800 			 * covering the entire write.
2801 			 */
2802 			*cow_start = ocfs2_cow_align_start(inode->i_sb,
2803 							   *cow_start, cpos);
2804 
2805 			want_clusters = (cpos + write_len) - *cow_start;
2806 			want_clusters = ocfs2_cow_align_length(inode->i_sb,
2807 							       want_clusters);
2808 			if (*cow_start + want_clusters <= rec_end)
2809 				*cow_len = want_clusters;
2810 			else
2811 				*cow_len = rec_end - *cow_start;
2812 		}
2813 
2814 		/* Have we covered our entire write yet? */
2815 		if ((*cow_start + *cow_len) >= (cpos + write_len))
2816 			break;
2817 
2818 		/*
2819 		 * If we reach the end of the extent block and don't get enough
2820 		 * clusters, continue with the next extent block if possible.
2821 		 */
2822 		if (i + 1 == le16_to_cpu(el->l_next_free_rec) &&
2823 		    eb && eb->h_next_leaf_blk) {
2824 			brelse(eb_bh);
2825 			eb_bh = NULL;
2826 
2827 			ret = ocfs2_read_extent_block(INODE_CACHE(inode),
2828 					       le64_to_cpu(eb->h_next_leaf_blk),
2829 					       &eb_bh);
2830 			if (ret) {
2831 				mlog_errno(ret);
2832 				goto out;
2833 			}
2834 
2835 			eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2836 			el = &eb->h_list;
2837 			i = -1;
2838 		}
2839 	}
2840 
2841 out:
2842 	brelse(eb_bh);
2843 	return ret;
2844 }
2845 
2846 /*
2847  * Prepare meta_ac, data_ac and calculate credits when we want to add some
2848  * num_clusters in data_tree "et" and change the refcount for the old
2849  * clusters(starting form p_cluster) in the refcount tree.
2850  *
2851  * Note:
2852  * 1. since we may split the old tree, so we at most will need num_clusters + 2
2853  *    more new leaf records.
2854  * 2. In some case, we may not need to reserve new clusters(e.g, reflink), so
2855  *    just give data_ac = NULL.
2856  */
2857 static int ocfs2_lock_refcount_allocators(struct super_block *sb,
2858 					u32 p_cluster, u32 num_clusters,
2859 					struct ocfs2_extent_tree *et,
2860 					struct ocfs2_caching_info *ref_ci,
2861 					struct buffer_head *ref_root_bh,
2862 					struct ocfs2_alloc_context **meta_ac,
2863 					struct ocfs2_alloc_context **data_ac,
2864 					int *credits)
2865 {
2866 	int ret = 0, meta_add = 0;
2867 	int num_free_extents = ocfs2_num_free_extents(OCFS2_SB(sb), et);
2868 
2869 	if (num_free_extents < 0) {
2870 		ret = num_free_extents;
2871 		mlog_errno(ret);
2872 		goto out;
2873 	}
2874 
2875 	if (num_free_extents < num_clusters + 2)
2876 		meta_add =
2877 			ocfs2_extend_meta_needed(et->et_root_el);
2878 
2879 	*credits += ocfs2_calc_extend_credits(sb, et->et_root_el);
2880 
2881 	ret = ocfs2_calc_refcount_meta_credits(sb, ref_ci, ref_root_bh,
2882 					       p_cluster, num_clusters,
2883 					       &meta_add, credits);
2884 	if (ret) {
2885 		mlog_errno(ret);
2886 		goto out;
2887 	}
2888 
2889 	trace_ocfs2_lock_refcount_allocators(meta_add, *credits);
2890 	ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(sb), meta_add,
2891 						meta_ac);
2892 	if (ret) {
2893 		mlog_errno(ret);
2894 		goto out;
2895 	}
2896 
2897 	if (data_ac) {
2898 		ret = ocfs2_reserve_clusters(OCFS2_SB(sb), num_clusters,
2899 					     data_ac);
2900 		if (ret)
2901 			mlog_errno(ret);
2902 	}
2903 
2904 out:
2905 	if (ret) {
2906 		if (*meta_ac) {
2907 			ocfs2_free_alloc_context(*meta_ac);
2908 			*meta_ac = NULL;
2909 		}
2910 	}
2911 
2912 	return ret;
2913 }
2914 
2915 static int ocfs2_clear_cow_buffer(handle_t *handle, struct buffer_head *bh)
2916 {
2917 	BUG_ON(buffer_dirty(bh));
2918 
2919 	clear_buffer_mapped(bh);
2920 
2921 	return 0;
2922 }
2923 
2924 int ocfs2_duplicate_clusters_by_page(handle_t *handle,
2925 				     struct inode *inode,
2926 				     u32 cpos, u32 old_cluster,
2927 				     u32 new_cluster, u32 new_len)
2928 {
2929 	int ret = 0, partial;
2930 	struct super_block *sb = inode->i_sb;
2931 	u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
2932 	struct page *page;
2933 	pgoff_t page_index;
2934 	unsigned int from, to, readahead_pages;
2935 	loff_t offset, end, map_end;
2936 	struct address_space *mapping = inode->i_mapping;
2937 
2938 	trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
2939 					       new_cluster, new_len);
2940 
2941 	readahead_pages =
2942 		(ocfs2_cow_contig_clusters(sb) <<
2943 		 OCFS2_SB(sb)->s_clustersize_bits) >> PAGE_CACHE_SHIFT;
2944 	offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
2945 	end = offset + (new_len << OCFS2_SB(sb)->s_clustersize_bits);
2946 	/*
2947 	 * We only duplicate pages until we reach the page contains i_size - 1.
2948 	 * So trim 'end' to i_size.
2949 	 */
2950 	if (end > i_size_read(inode))
2951 		end = i_size_read(inode);
2952 
2953 	while (offset < end) {
2954 		page_index = offset >> PAGE_CACHE_SHIFT;
2955 		map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
2956 		if (map_end > end)
2957 			map_end = end;
2958 
2959 		/* from, to is the offset within the page. */
2960 		from = offset & (PAGE_CACHE_SIZE - 1);
2961 		to = PAGE_CACHE_SIZE;
2962 		if (map_end & (PAGE_CACHE_SIZE - 1))
2963 			to = map_end & (PAGE_CACHE_SIZE - 1);
2964 
2965 		page = find_or_create_page(mapping, page_index, GFP_NOFS);
2966 		if (!page) {
2967 			ret = -ENOMEM;
2968 			mlog_errno(ret);
2969 			break;
2970 		}
2971 
2972 		/*
2973 		 * In case PAGE_CACHE_SIZE <= CLUSTER_SIZE, This page
2974 		 * can't be dirtied before we CoW it out.
2975 		 */
2976 		if (PAGE_CACHE_SIZE <= OCFS2_SB(sb)->s_clustersize)
2977 			BUG_ON(PageDirty(page));
2978 
2979 		if (!PageUptodate(page)) {
2980 			ret = block_read_full_page(page, ocfs2_get_block);
2981 			if (ret) {
2982 				mlog_errno(ret);
2983 				goto unlock;
2984 			}
2985 			lock_page(page);
2986 		}
2987 
2988 		if (page_has_buffers(page)) {
2989 			ret = walk_page_buffers(handle, page_buffers(page),
2990 						from, to, &partial,
2991 						ocfs2_clear_cow_buffer);
2992 			if (ret) {
2993 				mlog_errno(ret);
2994 				goto unlock;
2995 			}
2996 		}
2997 
2998 		ocfs2_map_and_dirty_page(inode,
2999 					 handle, from, to,
3000 					 page, 0, &new_block);
3001 		mark_page_accessed(page);
3002 unlock:
3003 		unlock_page(page);
3004 		page_cache_release(page);
3005 		page = NULL;
3006 		offset = map_end;
3007 		if (ret)
3008 			break;
3009 	}
3010 
3011 	return ret;
3012 }
3013 
3014 int ocfs2_duplicate_clusters_by_jbd(handle_t *handle,
3015 				    struct inode *inode,
3016 				    u32 cpos, u32 old_cluster,
3017 				    u32 new_cluster, u32 new_len)
3018 {
3019 	int ret = 0;
3020 	struct super_block *sb = inode->i_sb;
3021 	struct ocfs2_caching_info *ci = INODE_CACHE(inode);
3022 	int i, blocks = ocfs2_clusters_to_blocks(sb, new_len);
3023 	u64 old_block = ocfs2_clusters_to_blocks(sb, old_cluster);
3024 	u64 new_block = ocfs2_clusters_to_blocks(sb, new_cluster);
3025 	struct ocfs2_super *osb = OCFS2_SB(sb);
3026 	struct buffer_head *old_bh = NULL;
3027 	struct buffer_head *new_bh = NULL;
3028 
3029 	trace_ocfs2_duplicate_clusters_by_page(cpos, old_cluster,
3030 					       new_cluster, new_len);
3031 
3032 	for (i = 0; i < blocks; i++, old_block++, new_block++) {
3033 		new_bh = sb_getblk(osb->sb, new_block);
3034 		if (new_bh == NULL) {
3035 			ret = -ENOMEM;
3036 			mlog_errno(ret);
3037 			break;
3038 		}
3039 
3040 		ocfs2_set_new_buffer_uptodate(ci, new_bh);
3041 
3042 		ret = ocfs2_read_block(ci, old_block, &old_bh, NULL);
3043 		if (ret) {
3044 			mlog_errno(ret);
3045 			break;
3046 		}
3047 
3048 		ret = ocfs2_journal_access(handle, ci, new_bh,
3049 					   OCFS2_JOURNAL_ACCESS_CREATE);
3050 		if (ret) {
3051 			mlog_errno(ret);
3052 			break;
3053 		}
3054 
3055 		memcpy(new_bh->b_data, old_bh->b_data, sb->s_blocksize);
3056 		ocfs2_journal_dirty(handle, new_bh);
3057 
3058 		brelse(new_bh);
3059 		brelse(old_bh);
3060 		new_bh = NULL;
3061 		old_bh = NULL;
3062 	}
3063 
3064 	brelse(new_bh);
3065 	brelse(old_bh);
3066 	return ret;
3067 }
3068 
3069 static int ocfs2_clear_ext_refcount(handle_t *handle,
3070 				    struct ocfs2_extent_tree *et,
3071 				    u32 cpos, u32 p_cluster, u32 len,
3072 				    unsigned int ext_flags,
3073 				    struct ocfs2_alloc_context *meta_ac,
3074 				    struct ocfs2_cached_dealloc_ctxt *dealloc)
3075 {
3076 	int ret, index;
3077 	struct ocfs2_extent_rec replace_rec;
3078 	struct ocfs2_path *path = NULL;
3079 	struct ocfs2_extent_list *el;
3080 	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
3081 	u64 ino = ocfs2_metadata_cache_owner(et->et_ci);
3082 
3083 	trace_ocfs2_clear_ext_refcount((unsigned long long)ino,
3084 				       cpos, len, p_cluster, ext_flags);
3085 
3086 	memset(&replace_rec, 0, sizeof(replace_rec));
3087 	replace_rec.e_cpos = cpu_to_le32(cpos);
3088 	replace_rec.e_leaf_clusters = cpu_to_le16(len);
3089 	replace_rec.e_blkno = cpu_to_le64(ocfs2_clusters_to_blocks(sb,
3090 								   p_cluster));
3091 	replace_rec.e_flags = ext_flags;
3092 	replace_rec.e_flags &= ~OCFS2_EXT_REFCOUNTED;
3093 
3094 	path = ocfs2_new_path_from_et(et);
3095 	if (!path) {
3096 		ret = -ENOMEM;
3097 		mlog_errno(ret);
3098 		goto out;
3099 	}
3100 
3101 	ret = ocfs2_find_path(et->et_ci, path, cpos);
3102 	if (ret) {
3103 		mlog_errno(ret);
3104 		goto out;
3105 	}
3106 
3107 	el = path_leaf_el(path);
3108 
3109 	index = ocfs2_search_extent_list(el, cpos);
3110 	if (index == -1) {
3111 		ocfs2_error(sb,
3112 			    "Inode %llu has an extent at cpos %u which can no "
3113 			    "longer be found.\n",
3114 			    (unsigned long long)ino, cpos);
3115 		ret = -EROFS;
3116 		goto out;
3117 	}
3118 
3119 	ret = ocfs2_split_extent(handle, et, path, index,
3120 				 &replace_rec, meta_ac, dealloc);
3121 	if (ret)
3122 		mlog_errno(ret);
3123 
3124 out:
3125 	ocfs2_free_path(path);
3126 	return ret;
3127 }
3128 
3129 static int ocfs2_replace_clusters(handle_t *handle,
3130 				  struct ocfs2_cow_context *context,
3131 				  u32 cpos, u32 old,
3132 				  u32 new, u32 len,
3133 				  unsigned int ext_flags)
3134 {
3135 	int ret;
3136 	struct ocfs2_caching_info *ci = context->data_et.et_ci;
3137 	u64 ino = ocfs2_metadata_cache_owner(ci);
3138 
3139 	trace_ocfs2_replace_clusters((unsigned long long)ino,
3140 				     cpos, old, new, len, ext_flags);
3141 
3142 	/*If the old clusters is unwritten, no need to duplicate. */
3143 	if (!(ext_flags & OCFS2_EXT_UNWRITTEN)) {
3144 		ret = context->cow_duplicate_clusters(handle, context->inode,
3145 						      cpos, old, new, len);
3146 		if (ret) {
3147 			mlog_errno(ret);
3148 			goto out;
3149 		}
3150 	}
3151 
3152 	ret = ocfs2_clear_ext_refcount(handle, &context->data_et,
3153 				       cpos, new, len, ext_flags,
3154 				       context->meta_ac, &context->dealloc);
3155 	if (ret)
3156 		mlog_errno(ret);
3157 out:
3158 	return ret;
3159 }
3160 
3161 int ocfs2_cow_sync_writeback(struct super_block *sb,
3162 			     struct inode *inode,
3163 			     u32 cpos, u32 num_clusters)
3164 {
3165 	int ret = 0;
3166 	loff_t offset, end, map_end;
3167 	pgoff_t page_index;
3168 	struct page *page;
3169 
3170 	if (ocfs2_should_order_data(inode))
3171 		return 0;
3172 
3173 	offset = ((loff_t)cpos) << OCFS2_SB(sb)->s_clustersize_bits;
3174 	end = offset + (num_clusters << OCFS2_SB(sb)->s_clustersize_bits);
3175 
3176 	ret = filemap_fdatawrite_range(inode->i_mapping,
3177 				       offset, end - 1);
3178 	if (ret < 0) {
3179 		mlog_errno(ret);
3180 		return ret;
3181 	}
3182 
3183 	while (offset < end) {
3184 		page_index = offset >> PAGE_CACHE_SHIFT;
3185 		map_end = ((loff_t)page_index + 1) << PAGE_CACHE_SHIFT;
3186 		if (map_end > end)
3187 			map_end = end;
3188 
3189 		page = find_or_create_page(inode->i_mapping,
3190 					   page_index, GFP_NOFS);
3191 		BUG_ON(!page);
3192 
3193 		wait_on_page_writeback(page);
3194 		if (PageError(page)) {
3195 			ret = -EIO;
3196 			mlog_errno(ret);
3197 		} else
3198 			mark_page_accessed(page);
3199 
3200 		unlock_page(page);
3201 		page_cache_release(page);
3202 		page = NULL;
3203 		offset = map_end;
3204 		if (ret)
3205 			break;
3206 	}
3207 
3208 	return ret;
3209 }
3210 
3211 static int ocfs2_di_get_clusters(struct ocfs2_cow_context *context,
3212 				 u32 v_cluster, u32 *p_cluster,
3213 				 u32 *num_clusters,
3214 				 unsigned int *extent_flags)
3215 {
3216 	return ocfs2_get_clusters(context->inode, v_cluster, p_cluster,
3217 				  num_clusters, extent_flags);
3218 }
3219 
3220 static int ocfs2_make_clusters_writable(struct super_block *sb,
3221 					struct ocfs2_cow_context *context,
3222 					u32 cpos, u32 p_cluster,
3223 					u32 num_clusters, unsigned int e_flags)
3224 {
3225 	int ret, delete, index, credits =  0;
3226 	u32 new_bit, new_len, orig_num_clusters;
3227 	unsigned int set_len;
3228 	struct ocfs2_super *osb = OCFS2_SB(sb);
3229 	handle_t *handle;
3230 	struct buffer_head *ref_leaf_bh = NULL;
3231 	struct ocfs2_caching_info *ref_ci = &context->ref_tree->rf_ci;
3232 	struct ocfs2_refcount_rec rec;
3233 
3234 	trace_ocfs2_make_clusters_writable(cpos, p_cluster,
3235 					   num_clusters, e_flags);
3236 
3237 	ret = ocfs2_lock_refcount_allocators(sb, p_cluster, num_clusters,
3238 					     &context->data_et,
3239 					     ref_ci,
3240 					     context->ref_root_bh,
3241 					     &context->meta_ac,
3242 					     &context->data_ac, &credits);
3243 	if (ret) {
3244 		mlog_errno(ret);
3245 		return ret;
3246 	}
3247 
3248 	if (context->post_refcount)
3249 		credits += context->post_refcount->credits;
3250 
3251 	credits += context->extra_credits;
3252 	handle = ocfs2_start_trans(osb, credits);
3253 	if (IS_ERR(handle)) {
3254 		ret = PTR_ERR(handle);
3255 		mlog_errno(ret);
3256 		goto out;
3257 	}
3258 
3259 	orig_num_clusters = num_clusters;
3260 
3261 	while (num_clusters) {
3262 		ret = ocfs2_get_refcount_rec(ref_ci, context->ref_root_bh,
3263 					     p_cluster, num_clusters,
3264 					     &rec, &index, &ref_leaf_bh);
3265 		if (ret) {
3266 			mlog_errno(ret);
3267 			goto out_commit;
3268 		}
3269 
3270 		BUG_ON(!rec.r_refcount);
3271 		set_len = min((u64)p_cluster + num_clusters,
3272 			      le64_to_cpu(rec.r_cpos) +
3273 			      le32_to_cpu(rec.r_clusters)) - p_cluster;
3274 
3275 		/*
3276 		 * There are many different situation here.
3277 		 * 1. If refcount == 1, remove the flag and don't COW.
3278 		 * 2. If refcount > 1, allocate clusters.
3279 		 *    Here we may not allocate r_len once at a time, so continue
3280 		 *    until we reach num_clusters.
3281 		 */
3282 		if (le32_to_cpu(rec.r_refcount) == 1) {
3283 			delete = 0;
3284 			ret = ocfs2_clear_ext_refcount(handle,
3285 						       &context->data_et,
3286 						       cpos, p_cluster,
3287 						       set_len, e_flags,
3288 						       context->meta_ac,
3289 						       &context->dealloc);
3290 			if (ret) {
3291 				mlog_errno(ret);
3292 				goto out_commit;
3293 			}
3294 		} else {
3295 			delete = 1;
3296 
3297 			ret = __ocfs2_claim_clusters(handle,
3298 						     context->data_ac,
3299 						     1, set_len,
3300 						     &new_bit, &new_len);
3301 			if (ret) {
3302 				mlog_errno(ret);
3303 				goto out_commit;
3304 			}
3305 
3306 			ret = ocfs2_replace_clusters(handle, context,
3307 						     cpos, p_cluster, new_bit,
3308 						     new_len, e_flags);
3309 			if (ret) {
3310 				mlog_errno(ret);
3311 				goto out_commit;
3312 			}
3313 			set_len = new_len;
3314 		}
3315 
3316 		ret = __ocfs2_decrease_refcount(handle, ref_ci,
3317 						context->ref_root_bh,
3318 						p_cluster, set_len,
3319 						context->meta_ac,
3320 						&context->dealloc, delete);
3321 		if (ret) {
3322 			mlog_errno(ret);
3323 			goto out_commit;
3324 		}
3325 
3326 		cpos += set_len;
3327 		p_cluster += set_len;
3328 		num_clusters -= set_len;
3329 		brelse(ref_leaf_bh);
3330 		ref_leaf_bh = NULL;
3331 	}
3332 
3333 	/* handle any post_cow action. */
3334 	if (context->post_refcount && context->post_refcount->func) {
3335 		ret = context->post_refcount->func(context->inode, handle,
3336 						context->post_refcount->para);
3337 		if (ret) {
3338 			mlog_errno(ret);
3339 			goto out_commit;
3340 		}
3341 	}
3342 
3343 	/*
3344 	 * Here we should write the new page out first if we are
3345 	 * in write-back mode.
3346 	 */
3347 	if (context->get_clusters == ocfs2_di_get_clusters) {
3348 		ret = ocfs2_cow_sync_writeback(sb, context->inode, cpos,
3349 					       orig_num_clusters);
3350 		if (ret)
3351 			mlog_errno(ret);
3352 	}
3353 
3354 out_commit:
3355 	ocfs2_commit_trans(osb, handle);
3356 
3357 out:
3358 	if (context->data_ac) {
3359 		ocfs2_free_alloc_context(context->data_ac);
3360 		context->data_ac = NULL;
3361 	}
3362 	if (context->meta_ac) {
3363 		ocfs2_free_alloc_context(context->meta_ac);
3364 		context->meta_ac = NULL;
3365 	}
3366 	brelse(ref_leaf_bh);
3367 
3368 	return ret;
3369 }
3370 
3371 static int ocfs2_replace_cow(struct ocfs2_cow_context *context)
3372 {
3373 	int ret = 0;
3374 	struct inode *inode = context->inode;
3375 	u32 cow_start = context->cow_start, cow_len = context->cow_len;
3376 	u32 p_cluster, num_clusters;
3377 	unsigned int ext_flags;
3378 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3379 
3380 	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb))) {
3381 		ocfs2_error(inode->i_sb, "Inode %lu want to use refcount "
3382 			    "tree, but the feature bit is not set in the "
3383 			    "super block.", inode->i_ino);
3384 		return -EROFS;
3385 	}
3386 
3387 	ocfs2_init_dealloc_ctxt(&context->dealloc);
3388 
3389 	while (cow_len) {
3390 		ret = context->get_clusters(context, cow_start, &p_cluster,
3391 					    &num_clusters, &ext_flags);
3392 		if (ret) {
3393 			mlog_errno(ret);
3394 			break;
3395 		}
3396 
3397 		BUG_ON(!(ext_flags & OCFS2_EXT_REFCOUNTED));
3398 
3399 		if (cow_len < num_clusters)
3400 			num_clusters = cow_len;
3401 
3402 		ret = ocfs2_make_clusters_writable(inode->i_sb, context,
3403 						   cow_start, p_cluster,
3404 						   num_clusters, ext_flags);
3405 		if (ret) {
3406 			mlog_errno(ret);
3407 			break;
3408 		}
3409 
3410 		cow_len -= num_clusters;
3411 		cow_start += num_clusters;
3412 	}
3413 
3414 	if (ocfs2_dealloc_has_cluster(&context->dealloc)) {
3415 		ocfs2_schedule_truncate_log_flush(osb, 1);
3416 		ocfs2_run_deallocs(osb, &context->dealloc);
3417 	}
3418 
3419 	return ret;
3420 }
3421 
3422 /*
3423  * Starting at cpos, try to CoW write_len clusters.  Don't CoW
3424  * past max_cpos.  This will stop when it runs into a hole or an
3425  * unrefcounted extent.
3426  */
3427 static int ocfs2_refcount_cow_hunk(struct inode *inode,
3428 				   struct buffer_head *di_bh,
3429 				   u32 cpos, u32 write_len, u32 max_cpos)
3430 {
3431 	int ret;
3432 	u32 cow_start = 0, cow_len = 0;
3433 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
3434 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3435 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3436 	struct buffer_head *ref_root_bh = NULL;
3437 	struct ocfs2_refcount_tree *ref_tree;
3438 	struct ocfs2_cow_context *context = NULL;
3439 
3440 	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3441 
3442 	ret = ocfs2_refcount_cal_cow_clusters(inode, &di->id2.i_list,
3443 					      cpos, write_len, max_cpos,
3444 					      &cow_start, &cow_len);
3445 	if (ret) {
3446 		mlog_errno(ret);
3447 		goto out;
3448 	}
3449 
3450 	trace_ocfs2_refcount_cow_hunk(OCFS2_I(inode)->ip_blkno,
3451 				      cpos, write_len, max_cpos,
3452 				      cow_start, cow_len);
3453 
3454 	BUG_ON(cow_len == 0);
3455 
3456 	context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3457 	if (!context) {
3458 		ret = -ENOMEM;
3459 		mlog_errno(ret);
3460 		goto out;
3461 	}
3462 
3463 	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
3464 				       1, &ref_tree, &ref_root_bh);
3465 	if (ret) {
3466 		mlog_errno(ret);
3467 		goto out;
3468 	}
3469 
3470 	context->inode = inode;
3471 	context->cow_start = cow_start;
3472 	context->cow_len = cow_len;
3473 	context->ref_tree = ref_tree;
3474 	context->ref_root_bh = ref_root_bh;
3475 	context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_page;
3476 	context->get_clusters = ocfs2_di_get_clusters;
3477 
3478 	ocfs2_init_dinode_extent_tree(&context->data_et,
3479 				      INODE_CACHE(inode), di_bh);
3480 
3481 	ret = ocfs2_replace_cow(context);
3482 	if (ret)
3483 		mlog_errno(ret);
3484 
3485 	/*
3486 	 * truncate the extent map here since no matter whether we meet with
3487 	 * any error during the action, we shouldn't trust cached extent map
3488 	 * any more.
3489 	 */
3490 	ocfs2_extent_map_trunc(inode, cow_start);
3491 
3492 	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3493 	brelse(ref_root_bh);
3494 out:
3495 	kfree(context);
3496 	return ret;
3497 }
3498 
3499 /*
3500  * CoW any and all clusters between cpos and cpos+write_len.
3501  * Don't CoW past max_cpos.  If this returns successfully, all
3502  * clusters between cpos and cpos+write_len are safe to modify.
3503  */
3504 int ocfs2_refcount_cow(struct inode *inode,
3505 		       struct buffer_head *di_bh,
3506 		       u32 cpos, u32 write_len, u32 max_cpos)
3507 {
3508 	int ret = 0;
3509 	u32 p_cluster, num_clusters;
3510 	unsigned int ext_flags;
3511 
3512 	while (write_len) {
3513 		ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3514 					 &num_clusters, &ext_flags);
3515 		if (ret) {
3516 			mlog_errno(ret);
3517 			break;
3518 		}
3519 
3520 		if (write_len < num_clusters)
3521 			num_clusters = write_len;
3522 
3523 		if (ext_flags & OCFS2_EXT_REFCOUNTED) {
3524 			ret = ocfs2_refcount_cow_hunk(inode, di_bh, cpos,
3525 						      num_clusters, max_cpos);
3526 			if (ret) {
3527 				mlog_errno(ret);
3528 				break;
3529 			}
3530 		}
3531 
3532 		write_len -= num_clusters;
3533 		cpos += num_clusters;
3534 	}
3535 
3536 	return ret;
3537 }
3538 
3539 static int ocfs2_xattr_value_get_clusters(struct ocfs2_cow_context *context,
3540 					  u32 v_cluster, u32 *p_cluster,
3541 					  u32 *num_clusters,
3542 					  unsigned int *extent_flags)
3543 {
3544 	struct inode *inode = context->inode;
3545 	struct ocfs2_xattr_value_root *xv = context->cow_object;
3546 
3547 	return ocfs2_xattr_get_clusters(inode, v_cluster, p_cluster,
3548 					num_clusters, &xv->xr_list,
3549 					extent_flags);
3550 }
3551 
3552 /*
3553  * Given a xattr value root, calculate the most meta/credits we need for
3554  * refcount tree change if we truncate it to 0.
3555  */
3556 int ocfs2_refcounted_xattr_delete_need(struct inode *inode,
3557 				       struct ocfs2_caching_info *ref_ci,
3558 				       struct buffer_head *ref_root_bh,
3559 				       struct ocfs2_xattr_value_root *xv,
3560 				       int *meta_add, int *credits)
3561 {
3562 	int ret = 0, index, ref_blocks = 0;
3563 	u32 p_cluster, num_clusters;
3564 	u32 cpos = 0, clusters = le32_to_cpu(xv->xr_clusters);
3565 	struct ocfs2_refcount_block *rb;
3566 	struct ocfs2_refcount_rec rec;
3567 	struct buffer_head *ref_leaf_bh = NULL;
3568 
3569 	while (cpos < clusters) {
3570 		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
3571 					       &num_clusters, &xv->xr_list,
3572 					       NULL);
3573 		if (ret) {
3574 			mlog_errno(ret);
3575 			goto out;
3576 		}
3577 
3578 		cpos += num_clusters;
3579 
3580 		while (num_clusters) {
3581 			ret = ocfs2_get_refcount_rec(ref_ci, ref_root_bh,
3582 						     p_cluster, num_clusters,
3583 						     &rec, &index,
3584 						     &ref_leaf_bh);
3585 			if (ret) {
3586 				mlog_errno(ret);
3587 				goto out;
3588 			}
3589 
3590 			BUG_ON(!rec.r_refcount);
3591 
3592 			rb = (struct ocfs2_refcount_block *)ref_leaf_bh->b_data;
3593 
3594 			/*
3595 			 * We really don't know whether the other clusters is in
3596 			 * this refcount block or not, so just take the worst
3597 			 * case that all the clusters are in this block and each
3598 			 * one will split a refcount rec, so totally we need
3599 			 * clusters * 2 new refcount rec.
3600 			 */
3601 			if (le16_to_cpu(rb->rf_records.rl_used) + clusters * 2 >
3602 			    le16_to_cpu(rb->rf_records.rl_count))
3603 				ref_blocks++;
3604 
3605 			*credits += 1;
3606 			brelse(ref_leaf_bh);
3607 			ref_leaf_bh = NULL;
3608 
3609 			if (num_clusters <= le32_to_cpu(rec.r_clusters))
3610 				break;
3611 			else
3612 				num_clusters -= le32_to_cpu(rec.r_clusters);
3613 			p_cluster += num_clusters;
3614 		}
3615 	}
3616 
3617 	*meta_add += ref_blocks;
3618 	if (!ref_blocks)
3619 		goto out;
3620 
3621 	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
3622 	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
3623 		*credits += OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
3624 	else {
3625 		struct ocfs2_extent_tree et;
3626 
3627 		ocfs2_init_refcount_extent_tree(&et, ref_ci, ref_root_bh);
3628 		*credits += ocfs2_calc_extend_credits(inode->i_sb,
3629 						      et.et_root_el);
3630 	}
3631 
3632 out:
3633 	brelse(ref_leaf_bh);
3634 	return ret;
3635 }
3636 
3637 /*
3638  * Do CoW for xattr.
3639  */
3640 int ocfs2_refcount_cow_xattr(struct inode *inode,
3641 			     struct ocfs2_dinode *di,
3642 			     struct ocfs2_xattr_value_buf *vb,
3643 			     struct ocfs2_refcount_tree *ref_tree,
3644 			     struct buffer_head *ref_root_bh,
3645 			     u32 cpos, u32 write_len,
3646 			     struct ocfs2_post_refcount *post)
3647 {
3648 	int ret;
3649 	struct ocfs2_xattr_value_root *xv = vb->vb_xv;
3650 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
3651 	struct ocfs2_cow_context *context = NULL;
3652 	u32 cow_start, cow_len;
3653 
3654 	BUG_ON(!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL));
3655 
3656 	ret = ocfs2_refcount_cal_cow_clusters(inode, &xv->xr_list,
3657 					      cpos, write_len, UINT_MAX,
3658 					      &cow_start, &cow_len);
3659 	if (ret) {
3660 		mlog_errno(ret);
3661 		goto out;
3662 	}
3663 
3664 	BUG_ON(cow_len == 0);
3665 
3666 	context = kzalloc(sizeof(struct ocfs2_cow_context), GFP_NOFS);
3667 	if (!context) {
3668 		ret = -ENOMEM;
3669 		mlog_errno(ret);
3670 		goto out;
3671 	}
3672 
3673 	context->inode = inode;
3674 	context->cow_start = cow_start;
3675 	context->cow_len = cow_len;
3676 	context->ref_tree = ref_tree;
3677 	context->ref_root_bh = ref_root_bh;
3678 	context->cow_object = xv;
3679 
3680 	context->cow_duplicate_clusters = ocfs2_duplicate_clusters_by_jbd;
3681 	/* We need the extra credits for duplicate_clusters by jbd. */
3682 	context->extra_credits =
3683 		ocfs2_clusters_to_blocks(inode->i_sb, 1) * cow_len;
3684 	context->get_clusters = ocfs2_xattr_value_get_clusters;
3685 	context->post_refcount = post;
3686 
3687 	ocfs2_init_xattr_value_extent_tree(&context->data_et,
3688 					   INODE_CACHE(inode), vb);
3689 
3690 	ret = ocfs2_replace_cow(context);
3691 	if (ret)
3692 		mlog_errno(ret);
3693 
3694 out:
3695 	kfree(context);
3696 	return ret;
3697 }
3698 
3699 /*
3700  * Insert a new extent into refcount tree and mark a extent rec
3701  * as refcounted in the dinode tree.
3702  */
3703 int ocfs2_add_refcount_flag(struct inode *inode,
3704 			    struct ocfs2_extent_tree *data_et,
3705 			    struct ocfs2_caching_info *ref_ci,
3706 			    struct buffer_head *ref_root_bh,
3707 			    u32 cpos, u32 p_cluster, u32 num_clusters,
3708 			    struct ocfs2_cached_dealloc_ctxt *dealloc,
3709 			    struct ocfs2_post_refcount *post)
3710 {
3711 	int ret;
3712 	handle_t *handle;
3713 	int credits = 1, ref_blocks = 0;
3714 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3715 	struct ocfs2_alloc_context *meta_ac = NULL;
3716 
3717 	ret = ocfs2_calc_refcount_meta_credits(inode->i_sb,
3718 					       ref_ci, ref_root_bh,
3719 					       p_cluster, num_clusters,
3720 					       &ref_blocks, &credits);
3721 	if (ret) {
3722 		mlog_errno(ret);
3723 		goto out;
3724 	}
3725 
3726 	trace_ocfs2_add_refcount_flag(ref_blocks, credits);
3727 
3728 	if (ref_blocks) {
3729 		ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
3730 							ref_blocks, &meta_ac);
3731 		if (ret) {
3732 			mlog_errno(ret);
3733 			goto out;
3734 		}
3735 	}
3736 
3737 	if (post)
3738 		credits += post->credits;
3739 
3740 	handle = ocfs2_start_trans(osb, credits);
3741 	if (IS_ERR(handle)) {
3742 		ret = PTR_ERR(handle);
3743 		mlog_errno(ret);
3744 		goto out;
3745 	}
3746 
3747 	ret = ocfs2_mark_extent_refcounted(inode, data_et, handle,
3748 					   cpos, num_clusters, p_cluster,
3749 					   meta_ac, dealloc);
3750 	if (ret) {
3751 		mlog_errno(ret);
3752 		goto out_commit;
3753 	}
3754 
3755 	ret = __ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3756 					p_cluster, num_clusters, 0,
3757 					meta_ac, dealloc);
3758 	if (ret) {
3759 		mlog_errno(ret);
3760 		goto out_commit;
3761 	}
3762 
3763 	if (post && post->func) {
3764 		ret = post->func(inode, handle, post->para);
3765 		if (ret)
3766 			mlog_errno(ret);
3767 	}
3768 
3769 out_commit:
3770 	ocfs2_commit_trans(osb, handle);
3771 out:
3772 	if (meta_ac)
3773 		ocfs2_free_alloc_context(meta_ac);
3774 	return ret;
3775 }
3776 
3777 static int ocfs2_change_ctime(struct inode *inode,
3778 			      struct buffer_head *di_bh)
3779 {
3780 	int ret;
3781 	handle_t *handle;
3782 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3783 
3784 	handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb),
3785 				   OCFS2_INODE_UPDATE_CREDITS);
3786 	if (IS_ERR(handle)) {
3787 		ret = PTR_ERR(handle);
3788 		mlog_errno(ret);
3789 		goto out;
3790 	}
3791 
3792 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
3793 				      OCFS2_JOURNAL_ACCESS_WRITE);
3794 	if (ret) {
3795 		mlog_errno(ret);
3796 		goto out_commit;
3797 	}
3798 
3799 	inode->i_ctime = CURRENT_TIME;
3800 	di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3801 	di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3802 
3803 	ocfs2_journal_dirty(handle, di_bh);
3804 
3805 out_commit:
3806 	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3807 out:
3808 	return ret;
3809 }
3810 
3811 static int ocfs2_attach_refcount_tree(struct inode *inode,
3812 				      struct buffer_head *di_bh)
3813 {
3814 	int ret, data_changed = 0;
3815 	struct buffer_head *ref_root_bh = NULL;
3816 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
3817 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
3818 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3819 	struct ocfs2_refcount_tree *ref_tree;
3820 	unsigned int ext_flags;
3821 	loff_t size;
3822 	u32 cpos, num_clusters, clusters, p_cluster;
3823 	struct ocfs2_cached_dealloc_ctxt dealloc;
3824 	struct ocfs2_extent_tree di_et;
3825 
3826 	ocfs2_init_dealloc_ctxt(&dealloc);
3827 
3828 	if (!(oi->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL)) {
3829 		ret = ocfs2_create_refcount_tree(inode, di_bh);
3830 		if (ret) {
3831 			mlog_errno(ret);
3832 			goto out;
3833 		}
3834 	}
3835 
3836 	BUG_ON(!di->i_refcount_loc);
3837 	ret = ocfs2_lock_refcount_tree(osb,
3838 				       le64_to_cpu(di->i_refcount_loc), 1,
3839 				       &ref_tree, &ref_root_bh);
3840 	if (ret) {
3841 		mlog_errno(ret);
3842 		goto out;
3843 	}
3844 
3845 	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL)
3846 		goto attach_xattr;
3847 
3848 	ocfs2_init_dinode_extent_tree(&di_et, INODE_CACHE(inode), di_bh);
3849 
3850 	size = i_size_read(inode);
3851 	clusters = ocfs2_clusters_for_bytes(inode->i_sb, size);
3852 
3853 	cpos = 0;
3854 	while (cpos < clusters) {
3855 		ret = ocfs2_get_clusters(inode, cpos, &p_cluster,
3856 					 &num_clusters, &ext_flags);
3857 		if (ret) {
3858 			mlog_errno(ret);
3859 			goto unlock;
3860 		}
3861 		if (p_cluster && !(ext_flags & OCFS2_EXT_REFCOUNTED)) {
3862 			ret = ocfs2_add_refcount_flag(inode, &di_et,
3863 						      &ref_tree->rf_ci,
3864 						      ref_root_bh, cpos,
3865 						      p_cluster, num_clusters,
3866 						      &dealloc, NULL);
3867 			if (ret) {
3868 				mlog_errno(ret);
3869 				goto unlock;
3870 			}
3871 
3872 			data_changed = 1;
3873 		}
3874 		cpos += num_clusters;
3875 	}
3876 
3877 attach_xattr:
3878 	if (oi->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
3879 		ret = ocfs2_xattr_attach_refcount_tree(inode, di_bh,
3880 						       &ref_tree->rf_ci,
3881 						       ref_root_bh,
3882 						       &dealloc);
3883 		if (ret) {
3884 			mlog_errno(ret);
3885 			goto unlock;
3886 		}
3887 	}
3888 
3889 	if (data_changed) {
3890 		ret = ocfs2_change_ctime(inode, di_bh);
3891 		if (ret)
3892 			mlog_errno(ret);
3893 	}
3894 
3895 unlock:
3896 	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3897 	brelse(ref_root_bh);
3898 
3899 	if (!ret && ocfs2_dealloc_has_cluster(&dealloc)) {
3900 		ocfs2_schedule_truncate_log_flush(osb, 1);
3901 		ocfs2_run_deallocs(osb, &dealloc);
3902 	}
3903 out:
3904 	/*
3905 	 * Empty the extent map so that we may get the right extent
3906 	 * record from the disk.
3907 	 */
3908 	ocfs2_extent_map_trunc(inode, 0);
3909 
3910 	return ret;
3911 }
3912 
3913 static int ocfs2_add_refcounted_extent(struct inode *inode,
3914 				   struct ocfs2_extent_tree *et,
3915 				   struct ocfs2_caching_info *ref_ci,
3916 				   struct buffer_head *ref_root_bh,
3917 				   u32 cpos, u32 p_cluster, u32 num_clusters,
3918 				   unsigned int ext_flags,
3919 				   struct ocfs2_cached_dealloc_ctxt *dealloc)
3920 {
3921 	int ret;
3922 	handle_t *handle;
3923 	int credits = 0;
3924 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3925 	struct ocfs2_alloc_context *meta_ac = NULL;
3926 
3927 	ret = ocfs2_lock_refcount_allocators(inode->i_sb,
3928 					     p_cluster, num_clusters,
3929 					     et, ref_ci,
3930 					     ref_root_bh, &meta_ac,
3931 					     NULL, &credits);
3932 	if (ret) {
3933 		mlog_errno(ret);
3934 		goto out;
3935 	}
3936 
3937 	handle = ocfs2_start_trans(osb, credits);
3938 	if (IS_ERR(handle)) {
3939 		ret = PTR_ERR(handle);
3940 		mlog_errno(ret);
3941 		goto out;
3942 	}
3943 
3944 	ret = ocfs2_insert_extent(handle, et, cpos,
3945 			ocfs2_clusters_to_blocks(inode->i_sb, p_cluster),
3946 			num_clusters, ext_flags, meta_ac);
3947 	if (ret) {
3948 		mlog_errno(ret);
3949 		goto out_commit;
3950 	}
3951 
3952 	ret = ocfs2_increase_refcount(handle, ref_ci, ref_root_bh,
3953 				      p_cluster, num_clusters,
3954 				      meta_ac, dealloc);
3955 	if (ret)
3956 		mlog_errno(ret);
3957 
3958 out_commit:
3959 	ocfs2_commit_trans(osb, handle);
3960 out:
3961 	if (meta_ac)
3962 		ocfs2_free_alloc_context(meta_ac);
3963 	return ret;
3964 }
3965 
3966 static int ocfs2_duplicate_inline_data(struct inode *s_inode,
3967 				       struct buffer_head *s_bh,
3968 				       struct inode *t_inode,
3969 				       struct buffer_head *t_bh)
3970 {
3971 	int ret;
3972 	handle_t *handle;
3973 	struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
3974 	struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
3975 	struct ocfs2_dinode *t_di = (struct ocfs2_dinode *)t_bh->b_data;
3976 
3977 	BUG_ON(!(OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL));
3978 
3979 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
3980 	if (IS_ERR(handle)) {
3981 		ret = PTR_ERR(handle);
3982 		mlog_errno(ret);
3983 		goto out;
3984 	}
3985 
3986 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
3987 				      OCFS2_JOURNAL_ACCESS_WRITE);
3988 	if (ret) {
3989 		mlog_errno(ret);
3990 		goto out_commit;
3991 	}
3992 
3993 	t_di->id2.i_data.id_count = s_di->id2.i_data.id_count;
3994 	memcpy(t_di->id2.i_data.id_data, s_di->id2.i_data.id_data,
3995 	       le16_to_cpu(s_di->id2.i_data.id_count));
3996 	spin_lock(&OCFS2_I(t_inode)->ip_lock);
3997 	OCFS2_I(t_inode)->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
3998 	t_di->i_dyn_features = cpu_to_le16(OCFS2_I(t_inode)->ip_dyn_features);
3999 	spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4000 
4001 	ocfs2_journal_dirty(handle, t_bh);
4002 
4003 out_commit:
4004 	ocfs2_commit_trans(osb, handle);
4005 out:
4006 	return ret;
4007 }
4008 
4009 static int ocfs2_duplicate_extent_list(struct inode *s_inode,
4010 				struct inode *t_inode,
4011 				struct buffer_head *t_bh,
4012 				struct ocfs2_caching_info *ref_ci,
4013 				struct buffer_head *ref_root_bh,
4014 				struct ocfs2_cached_dealloc_ctxt *dealloc)
4015 {
4016 	int ret = 0;
4017 	u32 p_cluster, num_clusters, clusters, cpos;
4018 	loff_t size;
4019 	unsigned int ext_flags;
4020 	struct ocfs2_extent_tree et;
4021 
4022 	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(t_inode), t_bh);
4023 
4024 	size = i_size_read(s_inode);
4025 	clusters = ocfs2_clusters_for_bytes(s_inode->i_sb, size);
4026 
4027 	cpos = 0;
4028 	while (cpos < clusters) {
4029 		ret = ocfs2_get_clusters(s_inode, cpos, &p_cluster,
4030 					 &num_clusters, &ext_flags);
4031 		if (ret) {
4032 			mlog_errno(ret);
4033 			goto out;
4034 		}
4035 		if (p_cluster) {
4036 			ret = ocfs2_add_refcounted_extent(t_inode, &et,
4037 							  ref_ci, ref_root_bh,
4038 							  cpos, p_cluster,
4039 							  num_clusters,
4040 							  ext_flags,
4041 							  dealloc);
4042 			if (ret) {
4043 				mlog_errno(ret);
4044 				goto out;
4045 			}
4046 		}
4047 
4048 		cpos += num_clusters;
4049 	}
4050 
4051 out:
4052 	return ret;
4053 }
4054 
4055 /*
4056  * change the new file's attributes to the src.
4057  *
4058  * reflink creates a snapshot of a file, that means the attributes
4059  * must be identical except for three exceptions - nlink, ino, and ctime.
4060  */
4061 static int ocfs2_complete_reflink(struct inode *s_inode,
4062 				  struct buffer_head *s_bh,
4063 				  struct inode *t_inode,
4064 				  struct buffer_head *t_bh,
4065 				  bool preserve)
4066 {
4067 	int ret;
4068 	handle_t *handle;
4069 	struct ocfs2_dinode *s_di = (struct ocfs2_dinode *)s_bh->b_data;
4070 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)t_bh->b_data;
4071 	loff_t size = i_size_read(s_inode);
4072 
4073 	handle = ocfs2_start_trans(OCFS2_SB(t_inode->i_sb),
4074 				   OCFS2_INODE_UPDATE_CREDITS);
4075 	if (IS_ERR(handle)) {
4076 		ret = PTR_ERR(handle);
4077 		mlog_errno(ret);
4078 		return ret;
4079 	}
4080 
4081 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(t_inode), t_bh,
4082 				      OCFS2_JOURNAL_ACCESS_WRITE);
4083 	if (ret) {
4084 		mlog_errno(ret);
4085 		goto out_commit;
4086 	}
4087 
4088 	spin_lock(&OCFS2_I(t_inode)->ip_lock);
4089 	OCFS2_I(t_inode)->ip_clusters = OCFS2_I(s_inode)->ip_clusters;
4090 	OCFS2_I(t_inode)->ip_attr = OCFS2_I(s_inode)->ip_attr;
4091 	OCFS2_I(t_inode)->ip_dyn_features = OCFS2_I(s_inode)->ip_dyn_features;
4092 	spin_unlock(&OCFS2_I(t_inode)->ip_lock);
4093 	i_size_write(t_inode, size);
4094 	t_inode->i_blocks = s_inode->i_blocks;
4095 
4096 	di->i_xattr_inline_size = s_di->i_xattr_inline_size;
4097 	di->i_clusters = s_di->i_clusters;
4098 	di->i_size = s_di->i_size;
4099 	di->i_dyn_features = s_di->i_dyn_features;
4100 	di->i_attr = s_di->i_attr;
4101 
4102 	if (preserve) {
4103 		t_inode->i_uid = s_inode->i_uid;
4104 		t_inode->i_gid = s_inode->i_gid;
4105 		t_inode->i_mode = s_inode->i_mode;
4106 		di->i_uid = s_di->i_uid;
4107 		di->i_gid = s_di->i_gid;
4108 		di->i_mode = s_di->i_mode;
4109 
4110 		/*
4111 		 * update time.
4112 		 * we want mtime to appear identical to the source and
4113 		 * update ctime.
4114 		 */
4115 		t_inode->i_ctime = CURRENT_TIME;
4116 
4117 		di->i_ctime = cpu_to_le64(t_inode->i_ctime.tv_sec);
4118 		di->i_ctime_nsec = cpu_to_le32(t_inode->i_ctime.tv_nsec);
4119 
4120 		t_inode->i_mtime = s_inode->i_mtime;
4121 		di->i_mtime = s_di->i_mtime;
4122 		di->i_mtime_nsec = s_di->i_mtime_nsec;
4123 	}
4124 
4125 	ocfs2_journal_dirty(handle, t_bh);
4126 
4127 out_commit:
4128 	ocfs2_commit_trans(OCFS2_SB(t_inode->i_sb), handle);
4129 	return ret;
4130 }
4131 
4132 static int ocfs2_create_reflink_node(struct inode *s_inode,
4133 				     struct buffer_head *s_bh,
4134 				     struct inode *t_inode,
4135 				     struct buffer_head *t_bh,
4136 				     bool preserve)
4137 {
4138 	int ret;
4139 	struct buffer_head *ref_root_bh = NULL;
4140 	struct ocfs2_cached_dealloc_ctxt dealloc;
4141 	struct ocfs2_super *osb = OCFS2_SB(s_inode->i_sb);
4142 	struct ocfs2_refcount_block *rb;
4143 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)s_bh->b_data;
4144 	struct ocfs2_refcount_tree *ref_tree;
4145 
4146 	ocfs2_init_dealloc_ctxt(&dealloc);
4147 
4148 	ret = ocfs2_set_refcount_tree(t_inode, t_bh,
4149 				      le64_to_cpu(di->i_refcount_loc));
4150 	if (ret) {
4151 		mlog_errno(ret);
4152 		goto out;
4153 	}
4154 
4155 	if (OCFS2_I(s_inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
4156 		ret = ocfs2_duplicate_inline_data(s_inode, s_bh,
4157 						  t_inode, t_bh);
4158 		if (ret)
4159 			mlog_errno(ret);
4160 		goto out;
4161 	}
4162 
4163 	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
4164 				       1, &ref_tree, &ref_root_bh);
4165 	if (ret) {
4166 		mlog_errno(ret);
4167 		goto out;
4168 	}
4169 	rb = (struct ocfs2_refcount_block *)ref_root_bh->b_data;
4170 
4171 	ret = ocfs2_duplicate_extent_list(s_inode, t_inode, t_bh,
4172 					  &ref_tree->rf_ci, ref_root_bh,
4173 					  &dealloc);
4174 	if (ret) {
4175 		mlog_errno(ret);
4176 		goto out_unlock_refcount;
4177 	}
4178 
4179 out_unlock_refcount:
4180 	ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
4181 	brelse(ref_root_bh);
4182 out:
4183 	if (ocfs2_dealloc_has_cluster(&dealloc)) {
4184 		ocfs2_schedule_truncate_log_flush(osb, 1);
4185 		ocfs2_run_deallocs(osb, &dealloc);
4186 	}
4187 
4188 	return ret;
4189 }
4190 
4191 static int __ocfs2_reflink(struct dentry *old_dentry,
4192 			   struct buffer_head *old_bh,
4193 			   struct inode *new_inode,
4194 			   bool preserve)
4195 {
4196 	int ret;
4197 	struct inode *inode = d_inode(old_dentry);
4198 	struct buffer_head *new_bh = NULL;
4199 
4200 	if (OCFS2_I(inode)->ip_flags & OCFS2_INODE_SYSTEM_FILE) {
4201 		ret = -EINVAL;
4202 		mlog_errno(ret);
4203 		goto out;
4204 	}
4205 
4206 	ret = filemap_fdatawrite(inode->i_mapping);
4207 	if (ret) {
4208 		mlog_errno(ret);
4209 		goto out;
4210 	}
4211 
4212 	ret = ocfs2_attach_refcount_tree(inode, old_bh);
4213 	if (ret) {
4214 		mlog_errno(ret);
4215 		goto out;
4216 	}
4217 
4218 	mutex_lock_nested(&new_inode->i_mutex, I_MUTEX_CHILD);
4219 	ret = ocfs2_inode_lock_nested(new_inode, &new_bh, 1,
4220 				      OI_LS_REFLINK_TARGET);
4221 	if (ret) {
4222 		mlog_errno(ret);
4223 		goto out_unlock;
4224 	}
4225 
4226 	ret = ocfs2_create_reflink_node(inode, old_bh,
4227 					new_inode, new_bh, preserve);
4228 	if (ret) {
4229 		mlog_errno(ret);
4230 		goto inode_unlock;
4231 	}
4232 
4233 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_XATTR_FL) {
4234 		ret = ocfs2_reflink_xattrs(inode, old_bh,
4235 					   new_inode, new_bh,
4236 					   preserve);
4237 		if (ret) {
4238 			mlog_errno(ret);
4239 			goto inode_unlock;
4240 		}
4241 	}
4242 
4243 	ret = ocfs2_complete_reflink(inode, old_bh,
4244 				     new_inode, new_bh, preserve);
4245 	if (ret)
4246 		mlog_errno(ret);
4247 
4248 inode_unlock:
4249 	ocfs2_inode_unlock(new_inode, 1);
4250 	brelse(new_bh);
4251 out_unlock:
4252 	mutex_unlock(&new_inode->i_mutex);
4253 out:
4254 	if (!ret) {
4255 		ret = filemap_fdatawait(inode->i_mapping);
4256 		if (ret)
4257 			mlog_errno(ret);
4258 	}
4259 	return ret;
4260 }
4261 
4262 static int ocfs2_reflink(struct dentry *old_dentry, struct inode *dir,
4263 			 struct dentry *new_dentry, bool preserve)
4264 {
4265 	int error;
4266 	struct inode *inode = d_inode(old_dentry);
4267 	struct buffer_head *old_bh = NULL;
4268 	struct inode *new_orphan_inode = NULL;
4269 	struct posix_acl *default_acl, *acl;
4270 	umode_t mode;
4271 
4272 	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4273 		return -EOPNOTSUPP;
4274 
4275 	mode = inode->i_mode;
4276 	error = posix_acl_create(dir, &mode, &default_acl, &acl);
4277 	if (error) {
4278 		mlog_errno(error);
4279 		return error;
4280 	}
4281 
4282 	error = ocfs2_create_inode_in_orphan(dir, mode,
4283 					     &new_orphan_inode);
4284 	if (error) {
4285 		mlog_errno(error);
4286 		goto out;
4287 	}
4288 
4289 	error = ocfs2_rw_lock(inode, 1);
4290 	if (error) {
4291 		mlog_errno(error);
4292 		goto out;
4293 	}
4294 
4295 	error = ocfs2_inode_lock(inode, &old_bh, 1);
4296 	if (error) {
4297 		mlog_errno(error);
4298 		ocfs2_rw_unlock(inode, 1);
4299 		goto out;
4300 	}
4301 
4302 	down_write(&OCFS2_I(inode)->ip_xattr_sem);
4303 	down_write(&OCFS2_I(inode)->ip_alloc_sem);
4304 	error = __ocfs2_reflink(old_dentry, old_bh,
4305 				new_orphan_inode, preserve);
4306 	up_write(&OCFS2_I(inode)->ip_alloc_sem);
4307 	up_write(&OCFS2_I(inode)->ip_xattr_sem);
4308 
4309 	ocfs2_inode_unlock(inode, 1);
4310 	ocfs2_rw_unlock(inode, 1);
4311 	brelse(old_bh);
4312 
4313 	if (error) {
4314 		mlog_errno(error);
4315 		goto out;
4316 	}
4317 
4318 	/* If the security isn't preserved, we need to re-initialize them. */
4319 	if (!preserve) {
4320 		error = ocfs2_init_security_and_acl(dir, new_orphan_inode,
4321 						    &new_dentry->d_name,
4322 						    default_acl, acl);
4323 		if (error)
4324 			mlog_errno(error);
4325 	}
4326 out:
4327 	if (default_acl)
4328 		posix_acl_release(default_acl);
4329 	if (acl)
4330 		posix_acl_release(acl);
4331 	if (!error) {
4332 		error = ocfs2_mv_orphaned_inode_to_new(dir, new_orphan_inode,
4333 						       new_dentry);
4334 		if (error)
4335 			mlog_errno(error);
4336 	}
4337 
4338 	if (new_orphan_inode) {
4339 		/*
4340 		 * We need to open_unlock the inode no matter whether we
4341 		 * succeed or not, so that other nodes can delete it later.
4342 		 */
4343 		ocfs2_open_unlock(new_orphan_inode);
4344 		if (error)
4345 			iput(new_orphan_inode);
4346 	}
4347 
4348 	return error;
4349 }
4350 
4351 /*
4352  * Below here are the bits used by OCFS2_IOC_REFLINK() to fake
4353  * sys_reflink().  This will go away when vfs_reflink() exists in
4354  * fs/namei.c.
4355  */
4356 
4357 /* copied from may_create in VFS. */
4358 static inline int ocfs2_may_create(struct inode *dir, struct dentry *child)
4359 {
4360 	if (d_really_is_positive(child))
4361 		return -EEXIST;
4362 	if (IS_DEADDIR(dir))
4363 		return -ENOENT;
4364 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
4365 }
4366 
4367 /**
4368  * ocfs2_vfs_reflink - Create a reference-counted link
4369  *
4370  * @old_dentry:        source dentry + inode
4371  * @dir:       directory to create the target
4372  * @new_dentry:        target dentry
4373  * @preserve:  if true, preserve all file attributes
4374  */
4375 static int ocfs2_vfs_reflink(struct dentry *old_dentry, struct inode *dir,
4376 			     struct dentry *new_dentry, bool preserve)
4377 {
4378 	struct inode *inode = d_inode(old_dentry);
4379 	int error;
4380 
4381 	if (!inode)
4382 		return -ENOENT;
4383 
4384 	error = ocfs2_may_create(dir, new_dentry);
4385 	if (error)
4386 		return error;
4387 
4388 	if (dir->i_sb != inode->i_sb)
4389 		return -EXDEV;
4390 
4391 	/*
4392 	 * A reflink to an append-only or immutable file cannot be created.
4393 	 */
4394 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4395 		return -EPERM;
4396 
4397 	/* Only regular files can be reflinked. */
4398 	if (!S_ISREG(inode->i_mode))
4399 		return -EPERM;
4400 
4401 	/*
4402 	 * If the caller wants to preserve ownership, they require the
4403 	 * rights to do so.
4404 	 */
4405 	if (preserve) {
4406 		if (!uid_eq(current_fsuid(), inode->i_uid) && !capable(CAP_CHOWN))
4407 			return -EPERM;
4408 		if (!in_group_p(inode->i_gid) && !capable(CAP_CHOWN))
4409 			return -EPERM;
4410 	}
4411 
4412 	/*
4413 	 * If the caller is modifying any aspect of the attributes, they
4414 	 * are not creating a snapshot.  They need read permission on the
4415 	 * file.
4416 	 */
4417 	if (!preserve) {
4418 		error = inode_permission(inode, MAY_READ);
4419 		if (error)
4420 			return error;
4421 	}
4422 
4423 	mutex_lock(&inode->i_mutex);
4424 	dquot_initialize(dir);
4425 	error = ocfs2_reflink(old_dentry, dir, new_dentry, preserve);
4426 	mutex_unlock(&inode->i_mutex);
4427 	if (!error)
4428 		fsnotify_create(dir, new_dentry);
4429 	return error;
4430 }
4431 /*
4432  * Most codes are copied from sys_linkat.
4433  */
4434 int ocfs2_reflink_ioctl(struct inode *inode,
4435 			const char __user *oldname,
4436 			const char __user *newname,
4437 			bool preserve)
4438 {
4439 	struct dentry *new_dentry;
4440 	struct path old_path, new_path;
4441 	int error;
4442 
4443 	if (!ocfs2_refcount_tree(OCFS2_SB(inode->i_sb)))
4444 		return -EOPNOTSUPP;
4445 
4446 	error = user_path_at(AT_FDCWD, oldname, 0, &old_path);
4447 	if (error) {
4448 		mlog_errno(error);
4449 		return error;
4450 	}
4451 
4452 	new_dentry = user_path_create(AT_FDCWD, newname, &new_path, 0);
4453 	error = PTR_ERR(new_dentry);
4454 	if (IS_ERR(new_dentry)) {
4455 		mlog_errno(error);
4456 		goto out;
4457 	}
4458 
4459 	error = -EXDEV;
4460 	if (old_path.mnt != new_path.mnt) {
4461 		mlog_errno(error);
4462 		goto out_dput;
4463 	}
4464 
4465 	error = ocfs2_vfs_reflink(old_path.dentry,
4466 				  d_inode(new_path.dentry),
4467 				  new_dentry, preserve);
4468 out_dput:
4469 	done_path_create(&new_path, new_dentry);
4470 out:
4471 	path_put(&old_path);
4472 
4473 	return error;
4474 }
4475