xref: /openbmc/linux/fs/ocfs2/alloc.c (revision d62e74be)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * alloc.c
5  *
6  * Extent allocs and frees
7  *
8  * Copyright (C) 2002, 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  */
25 
26 #include <linux/fs.h>
27 #include <linux/types.h>
28 #include <linux/slab.h>
29 #include <linux/highmem.h>
30 #include <linux/swap.h>
31 #include <linux/quotaops.h>
32 #include <linux/blkdev.h>
33 
34 #include <cluster/masklog.h>
35 
36 #include "ocfs2.h"
37 
38 #include "alloc.h"
39 #include "aops.h"
40 #include "blockcheck.h"
41 #include "dlmglue.h"
42 #include "extent_map.h"
43 #include "inode.h"
44 #include "journal.h"
45 #include "localalloc.h"
46 #include "suballoc.h"
47 #include "sysfile.h"
48 #include "file.h"
49 #include "super.h"
50 #include "uptodate.h"
51 #include "xattr.h"
52 #include "refcounttree.h"
53 #include "ocfs2_trace.h"
54 
55 #include "buffer_head_io.h"
56 
57 enum ocfs2_contig_type {
58 	CONTIG_NONE = 0,
59 	CONTIG_LEFT,
60 	CONTIG_RIGHT,
61 	CONTIG_LEFTRIGHT,
62 };
63 
64 static enum ocfs2_contig_type
65 	ocfs2_extent_rec_contig(struct super_block *sb,
66 				struct ocfs2_extent_rec *ext,
67 				struct ocfs2_extent_rec *insert_rec);
68 /*
69  * Operations for a specific extent tree type.
70  *
71  * To implement an on-disk btree (extent tree) type in ocfs2, add
72  * an ocfs2_extent_tree_operations structure and the matching
73  * ocfs2_init_<thingy>_extent_tree() function.  That's pretty much it
74  * for the allocation portion of the extent tree.
75  */
76 struct ocfs2_extent_tree_operations {
77 	/*
78 	 * last_eb_blk is the block number of the right most leaf extent
79 	 * block.  Most on-disk structures containing an extent tree store
80 	 * this value for fast access.  The ->eo_set_last_eb_blk() and
81 	 * ->eo_get_last_eb_blk() operations access this value.  They are
82 	 *  both required.
83 	 */
84 	void (*eo_set_last_eb_blk)(struct ocfs2_extent_tree *et,
85 				   u64 blkno);
86 	u64 (*eo_get_last_eb_blk)(struct ocfs2_extent_tree *et);
87 
88 	/*
89 	 * The on-disk structure usually keeps track of how many total
90 	 * clusters are stored in this extent tree.  This function updates
91 	 * that value.  new_clusters is the delta, and must be
92 	 * added to the total.  Required.
93 	 */
94 	void (*eo_update_clusters)(struct ocfs2_extent_tree *et,
95 				   u32 new_clusters);
96 
97 	/*
98 	 * If this extent tree is supported by an extent map, insert
99 	 * a record into the map.
100 	 */
101 	void (*eo_extent_map_insert)(struct ocfs2_extent_tree *et,
102 				     struct ocfs2_extent_rec *rec);
103 
104 	/*
105 	 * If this extent tree is supported by an extent map, truncate the
106 	 * map to clusters,
107 	 */
108 	void (*eo_extent_map_truncate)(struct ocfs2_extent_tree *et,
109 				       u32 clusters);
110 
111 	/*
112 	 * If ->eo_insert_check() exists, it is called before rec is
113 	 * inserted into the extent tree.  It is optional.
114 	 */
115 	int (*eo_insert_check)(struct ocfs2_extent_tree *et,
116 			       struct ocfs2_extent_rec *rec);
117 	int (*eo_sanity_check)(struct ocfs2_extent_tree *et);
118 
119 	/*
120 	 * --------------------------------------------------------------
121 	 * The remaining are internal to ocfs2_extent_tree and don't have
122 	 * accessor functions
123 	 */
124 
125 	/*
126 	 * ->eo_fill_root_el() takes et->et_object and sets et->et_root_el.
127 	 * It is required.
128 	 */
129 	void (*eo_fill_root_el)(struct ocfs2_extent_tree *et);
130 
131 	/*
132 	 * ->eo_fill_max_leaf_clusters sets et->et_max_leaf_clusters if
133 	 * it exists.  If it does not, et->et_max_leaf_clusters is set
134 	 * to 0 (unlimited).  Optional.
135 	 */
136 	void (*eo_fill_max_leaf_clusters)(struct ocfs2_extent_tree *et);
137 
138 	/*
139 	 * ->eo_extent_contig test whether the 2 ocfs2_extent_rec
140 	 * are contiguous or not. Optional. Don't need to set it if use
141 	 * ocfs2_extent_rec as the tree leaf.
142 	 */
143 	enum ocfs2_contig_type
144 		(*eo_extent_contig)(struct ocfs2_extent_tree *et,
145 				    struct ocfs2_extent_rec *ext,
146 				    struct ocfs2_extent_rec *insert_rec);
147 };
148 
149 
150 /*
151  * Pre-declare ocfs2_dinode_et_ops so we can use it as a sanity check
152  * in the methods.
153  */
154 static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et);
155 static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
156 					 u64 blkno);
157 static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
158 					 u32 clusters);
159 static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
160 					   struct ocfs2_extent_rec *rec);
161 static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
162 					     u32 clusters);
163 static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
164 				     struct ocfs2_extent_rec *rec);
165 static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et);
166 static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et);
167 static struct ocfs2_extent_tree_operations ocfs2_dinode_et_ops = {
168 	.eo_set_last_eb_blk	= ocfs2_dinode_set_last_eb_blk,
169 	.eo_get_last_eb_blk	= ocfs2_dinode_get_last_eb_blk,
170 	.eo_update_clusters	= ocfs2_dinode_update_clusters,
171 	.eo_extent_map_insert	= ocfs2_dinode_extent_map_insert,
172 	.eo_extent_map_truncate	= ocfs2_dinode_extent_map_truncate,
173 	.eo_insert_check	= ocfs2_dinode_insert_check,
174 	.eo_sanity_check	= ocfs2_dinode_sanity_check,
175 	.eo_fill_root_el	= ocfs2_dinode_fill_root_el,
176 };
177 
178 static void ocfs2_dinode_set_last_eb_blk(struct ocfs2_extent_tree *et,
179 					 u64 blkno)
180 {
181 	struct ocfs2_dinode *di = et->et_object;
182 
183 	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
184 	di->i_last_eb_blk = cpu_to_le64(blkno);
185 }
186 
187 static u64 ocfs2_dinode_get_last_eb_blk(struct ocfs2_extent_tree *et)
188 {
189 	struct ocfs2_dinode *di = et->et_object;
190 
191 	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
192 	return le64_to_cpu(di->i_last_eb_blk);
193 }
194 
195 static void ocfs2_dinode_update_clusters(struct ocfs2_extent_tree *et,
196 					 u32 clusters)
197 {
198 	struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
199 	struct ocfs2_dinode *di = et->et_object;
200 
201 	le32_add_cpu(&di->i_clusters, clusters);
202 	spin_lock(&oi->ip_lock);
203 	oi->ip_clusters = le32_to_cpu(di->i_clusters);
204 	spin_unlock(&oi->ip_lock);
205 }
206 
207 static void ocfs2_dinode_extent_map_insert(struct ocfs2_extent_tree *et,
208 					   struct ocfs2_extent_rec *rec)
209 {
210 	struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
211 
212 	ocfs2_extent_map_insert_rec(inode, rec);
213 }
214 
215 static void ocfs2_dinode_extent_map_truncate(struct ocfs2_extent_tree *et,
216 					     u32 clusters)
217 {
218 	struct inode *inode = &cache_info_to_inode(et->et_ci)->vfs_inode;
219 
220 	ocfs2_extent_map_trunc(inode, clusters);
221 }
222 
223 static int ocfs2_dinode_insert_check(struct ocfs2_extent_tree *et,
224 				     struct ocfs2_extent_rec *rec)
225 {
226 	struct ocfs2_inode_info *oi = cache_info_to_inode(et->et_ci);
227 	struct ocfs2_super *osb = OCFS2_SB(oi->vfs_inode.i_sb);
228 
229 	BUG_ON(oi->ip_dyn_features & OCFS2_INLINE_DATA_FL);
230 	mlog_bug_on_msg(!ocfs2_sparse_alloc(osb) &&
231 			(oi->ip_clusters != le32_to_cpu(rec->e_cpos)),
232 			"Device %s, asking for sparse allocation: inode %llu, "
233 			"cpos %u, clusters %u\n",
234 			osb->dev_str,
235 			(unsigned long long)oi->ip_blkno,
236 			rec->e_cpos, oi->ip_clusters);
237 
238 	return 0;
239 }
240 
241 static int ocfs2_dinode_sanity_check(struct ocfs2_extent_tree *et)
242 {
243 	struct ocfs2_dinode *di = et->et_object;
244 
245 	BUG_ON(et->et_ops != &ocfs2_dinode_et_ops);
246 	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
247 
248 	return 0;
249 }
250 
251 static void ocfs2_dinode_fill_root_el(struct ocfs2_extent_tree *et)
252 {
253 	struct ocfs2_dinode *di = et->et_object;
254 
255 	et->et_root_el = &di->id2.i_list;
256 }
257 
258 
259 static void ocfs2_xattr_value_fill_root_el(struct ocfs2_extent_tree *et)
260 {
261 	struct ocfs2_xattr_value_buf *vb = et->et_object;
262 
263 	et->et_root_el = &vb->vb_xv->xr_list;
264 }
265 
266 static void ocfs2_xattr_value_set_last_eb_blk(struct ocfs2_extent_tree *et,
267 					      u64 blkno)
268 {
269 	struct ocfs2_xattr_value_buf *vb = et->et_object;
270 
271 	vb->vb_xv->xr_last_eb_blk = cpu_to_le64(blkno);
272 }
273 
274 static u64 ocfs2_xattr_value_get_last_eb_blk(struct ocfs2_extent_tree *et)
275 {
276 	struct ocfs2_xattr_value_buf *vb = et->et_object;
277 
278 	return le64_to_cpu(vb->vb_xv->xr_last_eb_blk);
279 }
280 
281 static void ocfs2_xattr_value_update_clusters(struct ocfs2_extent_tree *et,
282 					      u32 clusters)
283 {
284 	struct ocfs2_xattr_value_buf *vb = et->et_object;
285 
286 	le32_add_cpu(&vb->vb_xv->xr_clusters, clusters);
287 }
288 
289 static struct ocfs2_extent_tree_operations ocfs2_xattr_value_et_ops = {
290 	.eo_set_last_eb_blk	= ocfs2_xattr_value_set_last_eb_blk,
291 	.eo_get_last_eb_blk	= ocfs2_xattr_value_get_last_eb_blk,
292 	.eo_update_clusters	= ocfs2_xattr_value_update_clusters,
293 	.eo_fill_root_el	= ocfs2_xattr_value_fill_root_el,
294 };
295 
296 static void ocfs2_xattr_tree_fill_root_el(struct ocfs2_extent_tree *et)
297 {
298 	struct ocfs2_xattr_block *xb = et->et_object;
299 
300 	et->et_root_el = &xb->xb_attrs.xb_root.xt_list;
301 }
302 
303 static void ocfs2_xattr_tree_fill_max_leaf_clusters(struct ocfs2_extent_tree *et)
304 {
305 	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
306 	et->et_max_leaf_clusters =
307 		ocfs2_clusters_for_bytes(sb, OCFS2_MAX_XATTR_TREE_LEAF_SIZE);
308 }
309 
310 static void ocfs2_xattr_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
311 					     u64 blkno)
312 {
313 	struct ocfs2_xattr_block *xb = et->et_object;
314 	struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
315 
316 	xt->xt_last_eb_blk = cpu_to_le64(blkno);
317 }
318 
319 static u64 ocfs2_xattr_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
320 {
321 	struct ocfs2_xattr_block *xb = et->et_object;
322 	struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
323 
324 	return le64_to_cpu(xt->xt_last_eb_blk);
325 }
326 
327 static void ocfs2_xattr_tree_update_clusters(struct ocfs2_extent_tree *et,
328 					     u32 clusters)
329 {
330 	struct ocfs2_xattr_block *xb = et->et_object;
331 
332 	le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, clusters);
333 }
334 
335 static struct ocfs2_extent_tree_operations ocfs2_xattr_tree_et_ops = {
336 	.eo_set_last_eb_blk	= ocfs2_xattr_tree_set_last_eb_blk,
337 	.eo_get_last_eb_blk	= ocfs2_xattr_tree_get_last_eb_blk,
338 	.eo_update_clusters	= ocfs2_xattr_tree_update_clusters,
339 	.eo_fill_root_el	= ocfs2_xattr_tree_fill_root_el,
340 	.eo_fill_max_leaf_clusters = ocfs2_xattr_tree_fill_max_leaf_clusters,
341 };
342 
343 static void ocfs2_dx_root_set_last_eb_blk(struct ocfs2_extent_tree *et,
344 					  u64 blkno)
345 {
346 	struct ocfs2_dx_root_block *dx_root = et->et_object;
347 
348 	dx_root->dr_last_eb_blk = cpu_to_le64(blkno);
349 }
350 
351 static u64 ocfs2_dx_root_get_last_eb_blk(struct ocfs2_extent_tree *et)
352 {
353 	struct ocfs2_dx_root_block *dx_root = et->et_object;
354 
355 	return le64_to_cpu(dx_root->dr_last_eb_blk);
356 }
357 
358 static void ocfs2_dx_root_update_clusters(struct ocfs2_extent_tree *et,
359 					  u32 clusters)
360 {
361 	struct ocfs2_dx_root_block *dx_root = et->et_object;
362 
363 	le32_add_cpu(&dx_root->dr_clusters, clusters);
364 }
365 
366 static int ocfs2_dx_root_sanity_check(struct ocfs2_extent_tree *et)
367 {
368 	struct ocfs2_dx_root_block *dx_root = et->et_object;
369 
370 	BUG_ON(!OCFS2_IS_VALID_DX_ROOT(dx_root));
371 
372 	return 0;
373 }
374 
375 static void ocfs2_dx_root_fill_root_el(struct ocfs2_extent_tree *et)
376 {
377 	struct ocfs2_dx_root_block *dx_root = et->et_object;
378 
379 	et->et_root_el = &dx_root->dr_list;
380 }
381 
382 static struct ocfs2_extent_tree_operations ocfs2_dx_root_et_ops = {
383 	.eo_set_last_eb_blk	= ocfs2_dx_root_set_last_eb_blk,
384 	.eo_get_last_eb_blk	= ocfs2_dx_root_get_last_eb_blk,
385 	.eo_update_clusters	= ocfs2_dx_root_update_clusters,
386 	.eo_sanity_check	= ocfs2_dx_root_sanity_check,
387 	.eo_fill_root_el	= ocfs2_dx_root_fill_root_el,
388 };
389 
390 static void ocfs2_refcount_tree_fill_root_el(struct ocfs2_extent_tree *et)
391 {
392 	struct ocfs2_refcount_block *rb = et->et_object;
393 
394 	et->et_root_el = &rb->rf_list;
395 }
396 
397 static void ocfs2_refcount_tree_set_last_eb_blk(struct ocfs2_extent_tree *et,
398 						u64 blkno)
399 {
400 	struct ocfs2_refcount_block *rb = et->et_object;
401 
402 	rb->rf_last_eb_blk = cpu_to_le64(blkno);
403 }
404 
405 static u64 ocfs2_refcount_tree_get_last_eb_blk(struct ocfs2_extent_tree *et)
406 {
407 	struct ocfs2_refcount_block *rb = et->et_object;
408 
409 	return le64_to_cpu(rb->rf_last_eb_blk);
410 }
411 
412 static void ocfs2_refcount_tree_update_clusters(struct ocfs2_extent_tree *et,
413 						u32 clusters)
414 {
415 	struct ocfs2_refcount_block *rb = et->et_object;
416 
417 	le32_add_cpu(&rb->rf_clusters, clusters);
418 }
419 
420 static enum ocfs2_contig_type
421 ocfs2_refcount_tree_extent_contig(struct ocfs2_extent_tree *et,
422 				  struct ocfs2_extent_rec *ext,
423 				  struct ocfs2_extent_rec *insert_rec)
424 {
425 	return CONTIG_NONE;
426 }
427 
428 static struct ocfs2_extent_tree_operations ocfs2_refcount_tree_et_ops = {
429 	.eo_set_last_eb_blk	= ocfs2_refcount_tree_set_last_eb_blk,
430 	.eo_get_last_eb_blk	= ocfs2_refcount_tree_get_last_eb_blk,
431 	.eo_update_clusters	= ocfs2_refcount_tree_update_clusters,
432 	.eo_fill_root_el	= ocfs2_refcount_tree_fill_root_el,
433 	.eo_extent_contig	= ocfs2_refcount_tree_extent_contig,
434 };
435 
436 static void __ocfs2_init_extent_tree(struct ocfs2_extent_tree *et,
437 				     struct ocfs2_caching_info *ci,
438 				     struct buffer_head *bh,
439 				     ocfs2_journal_access_func access,
440 				     void *obj,
441 				     struct ocfs2_extent_tree_operations *ops)
442 {
443 	et->et_ops = ops;
444 	et->et_root_bh = bh;
445 	et->et_ci = ci;
446 	et->et_root_journal_access = access;
447 	if (!obj)
448 		obj = (void *)bh->b_data;
449 	et->et_object = obj;
450 
451 	et->et_ops->eo_fill_root_el(et);
452 	if (!et->et_ops->eo_fill_max_leaf_clusters)
453 		et->et_max_leaf_clusters = 0;
454 	else
455 		et->et_ops->eo_fill_max_leaf_clusters(et);
456 }
457 
458 void ocfs2_init_dinode_extent_tree(struct ocfs2_extent_tree *et,
459 				   struct ocfs2_caching_info *ci,
460 				   struct buffer_head *bh)
461 {
462 	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_di,
463 				 NULL, &ocfs2_dinode_et_ops);
464 }
465 
466 void ocfs2_init_xattr_tree_extent_tree(struct ocfs2_extent_tree *et,
467 				       struct ocfs2_caching_info *ci,
468 				       struct buffer_head *bh)
469 {
470 	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_xb,
471 				 NULL, &ocfs2_xattr_tree_et_ops);
472 }
473 
474 void ocfs2_init_xattr_value_extent_tree(struct ocfs2_extent_tree *et,
475 					struct ocfs2_caching_info *ci,
476 					struct ocfs2_xattr_value_buf *vb)
477 {
478 	__ocfs2_init_extent_tree(et, ci, vb->vb_bh, vb->vb_access, vb,
479 				 &ocfs2_xattr_value_et_ops);
480 }
481 
482 void ocfs2_init_dx_root_extent_tree(struct ocfs2_extent_tree *et,
483 				    struct ocfs2_caching_info *ci,
484 				    struct buffer_head *bh)
485 {
486 	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_dr,
487 				 NULL, &ocfs2_dx_root_et_ops);
488 }
489 
490 void ocfs2_init_refcount_extent_tree(struct ocfs2_extent_tree *et,
491 				     struct ocfs2_caching_info *ci,
492 				     struct buffer_head *bh)
493 {
494 	__ocfs2_init_extent_tree(et, ci, bh, ocfs2_journal_access_rb,
495 				 NULL, &ocfs2_refcount_tree_et_ops);
496 }
497 
498 static inline void ocfs2_et_set_last_eb_blk(struct ocfs2_extent_tree *et,
499 					    u64 new_last_eb_blk)
500 {
501 	et->et_ops->eo_set_last_eb_blk(et, new_last_eb_blk);
502 }
503 
504 static inline u64 ocfs2_et_get_last_eb_blk(struct ocfs2_extent_tree *et)
505 {
506 	return et->et_ops->eo_get_last_eb_blk(et);
507 }
508 
509 static inline void ocfs2_et_update_clusters(struct ocfs2_extent_tree *et,
510 					    u32 clusters)
511 {
512 	et->et_ops->eo_update_clusters(et, clusters);
513 }
514 
515 static inline void ocfs2_et_extent_map_insert(struct ocfs2_extent_tree *et,
516 					      struct ocfs2_extent_rec *rec)
517 {
518 	if (et->et_ops->eo_extent_map_insert)
519 		et->et_ops->eo_extent_map_insert(et, rec);
520 }
521 
522 static inline void ocfs2_et_extent_map_truncate(struct ocfs2_extent_tree *et,
523 						u32 clusters)
524 {
525 	if (et->et_ops->eo_extent_map_truncate)
526 		et->et_ops->eo_extent_map_truncate(et, clusters);
527 }
528 
529 static inline int ocfs2_et_root_journal_access(handle_t *handle,
530 					       struct ocfs2_extent_tree *et,
531 					       int type)
532 {
533 	return et->et_root_journal_access(handle, et->et_ci, et->et_root_bh,
534 					  type);
535 }
536 
537 static inline enum ocfs2_contig_type
538 	ocfs2_et_extent_contig(struct ocfs2_extent_tree *et,
539 			       struct ocfs2_extent_rec *rec,
540 			       struct ocfs2_extent_rec *insert_rec)
541 {
542 	if (et->et_ops->eo_extent_contig)
543 		return et->et_ops->eo_extent_contig(et, rec, insert_rec);
544 
545 	return ocfs2_extent_rec_contig(
546 				ocfs2_metadata_cache_get_super(et->et_ci),
547 				rec, insert_rec);
548 }
549 
550 static inline int ocfs2_et_insert_check(struct ocfs2_extent_tree *et,
551 					struct ocfs2_extent_rec *rec)
552 {
553 	int ret = 0;
554 
555 	if (et->et_ops->eo_insert_check)
556 		ret = et->et_ops->eo_insert_check(et, rec);
557 	return ret;
558 }
559 
560 static inline int ocfs2_et_sanity_check(struct ocfs2_extent_tree *et)
561 {
562 	int ret = 0;
563 
564 	if (et->et_ops->eo_sanity_check)
565 		ret = et->et_ops->eo_sanity_check(et);
566 	return ret;
567 }
568 
569 static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
570 					 struct ocfs2_extent_block *eb);
571 static void ocfs2_adjust_rightmost_records(handle_t *handle,
572 					   struct ocfs2_extent_tree *et,
573 					   struct ocfs2_path *path,
574 					   struct ocfs2_extent_rec *insert_rec);
575 /*
576  * Reset the actual path elements so that we can re-use the structure
577  * to build another path. Generally, this involves freeing the buffer
578  * heads.
579  */
580 void ocfs2_reinit_path(struct ocfs2_path *path, int keep_root)
581 {
582 	int i, start = 0, depth = 0;
583 	struct ocfs2_path_item *node;
584 
585 	if (keep_root)
586 		start = 1;
587 
588 	for(i = start; i < path_num_items(path); i++) {
589 		node = &path->p_node[i];
590 
591 		brelse(node->bh);
592 		node->bh = NULL;
593 		node->el = NULL;
594 	}
595 
596 	/*
597 	 * Tree depth may change during truncate, or insert. If we're
598 	 * keeping the root extent list, then make sure that our path
599 	 * structure reflects the proper depth.
600 	 */
601 	if (keep_root)
602 		depth = le16_to_cpu(path_root_el(path)->l_tree_depth);
603 	else
604 		path_root_access(path) = NULL;
605 
606 	path->p_tree_depth = depth;
607 }
608 
609 void ocfs2_free_path(struct ocfs2_path *path)
610 {
611 	if (path) {
612 		ocfs2_reinit_path(path, 0);
613 		kfree(path);
614 	}
615 }
616 
617 /*
618  * All the elements of src into dest. After this call, src could be freed
619  * without affecting dest.
620  *
621  * Both paths should have the same root. Any non-root elements of dest
622  * will be freed.
623  */
624 static void ocfs2_cp_path(struct ocfs2_path *dest, struct ocfs2_path *src)
625 {
626 	int i;
627 
628 	BUG_ON(path_root_bh(dest) != path_root_bh(src));
629 	BUG_ON(path_root_el(dest) != path_root_el(src));
630 	BUG_ON(path_root_access(dest) != path_root_access(src));
631 
632 	ocfs2_reinit_path(dest, 1);
633 
634 	for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
635 		dest->p_node[i].bh = src->p_node[i].bh;
636 		dest->p_node[i].el = src->p_node[i].el;
637 
638 		if (dest->p_node[i].bh)
639 			get_bh(dest->p_node[i].bh);
640 	}
641 }
642 
643 /*
644  * Make the *dest path the same as src and re-initialize src path to
645  * have a root only.
646  */
647 static void ocfs2_mv_path(struct ocfs2_path *dest, struct ocfs2_path *src)
648 {
649 	int i;
650 
651 	BUG_ON(path_root_bh(dest) != path_root_bh(src));
652 	BUG_ON(path_root_access(dest) != path_root_access(src));
653 
654 	for(i = 1; i < OCFS2_MAX_PATH_DEPTH; i++) {
655 		brelse(dest->p_node[i].bh);
656 
657 		dest->p_node[i].bh = src->p_node[i].bh;
658 		dest->p_node[i].el = src->p_node[i].el;
659 
660 		src->p_node[i].bh = NULL;
661 		src->p_node[i].el = NULL;
662 	}
663 }
664 
665 /*
666  * Insert an extent block at given index.
667  *
668  * This will not take an additional reference on eb_bh.
669  */
670 static inline void ocfs2_path_insert_eb(struct ocfs2_path *path, int index,
671 					struct buffer_head *eb_bh)
672 {
673 	struct ocfs2_extent_block *eb = (struct ocfs2_extent_block *)eb_bh->b_data;
674 
675 	/*
676 	 * Right now, no root bh is an extent block, so this helps
677 	 * catch code errors with dinode trees. The assertion can be
678 	 * safely removed if we ever need to insert extent block
679 	 * structures at the root.
680 	 */
681 	BUG_ON(index == 0);
682 
683 	path->p_node[index].bh = eb_bh;
684 	path->p_node[index].el = &eb->h_list;
685 }
686 
687 static struct ocfs2_path *ocfs2_new_path(struct buffer_head *root_bh,
688 					 struct ocfs2_extent_list *root_el,
689 					 ocfs2_journal_access_func access)
690 {
691 	struct ocfs2_path *path;
692 
693 	BUG_ON(le16_to_cpu(root_el->l_tree_depth) >= OCFS2_MAX_PATH_DEPTH);
694 
695 	path = kzalloc(sizeof(*path), GFP_NOFS);
696 	if (path) {
697 		path->p_tree_depth = le16_to_cpu(root_el->l_tree_depth);
698 		get_bh(root_bh);
699 		path_root_bh(path) = root_bh;
700 		path_root_el(path) = root_el;
701 		path_root_access(path) = access;
702 	}
703 
704 	return path;
705 }
706 
707 struct ocfs2_path *ocfs2_new_path_from_path(struct ocfs2_path *path)
708 {
709 	return ocfs2_new_path(path_root_bh(path), path_root_el(path),
710 			      path_root_access(path));
711 }
712 
713 struct ocfs2_path *ocfs2_new_path_from_et(struct ocfs2_extent_tree *et)
714 {
715 	return ocfs2_new_path(et->et_root_bh, et->et_root_el,
716 			      et->et_root_journal_access);
717 }
718 
719 /*
720  * Journal the buffer at depth idx.  All idx>0 are extent_blocks,
721  * otherwise it's the root_access function.
722  *
723  * I don't like the way this function's name looks next to
724  * ocfs2_journal_access_path(), but I don't have a better one.
725  */
726 int ocfs2_path_bh_journal_access(handle_t *handle,
727 				 struct ocfs2_caching_info *ci,
728 				 struct ocfs2_path *path,
729 				 int idx)
730 {
731 	ocfs2_journal_access_func access = path_root_access(path);
732 
733 	if (!access)
734 		access = ocfs2_journal_access;
735 
736 	if (idx)
737 		access = ocfs2_journal_access_eb;
738 
739 	return access(handle, ci, path->p_node[idx].bh,
740 		      OCFS2_JOURNAL_ACCESS_WRITE);
741 }
742 
743 /*
744  * Convenience function to journal all components in a path.
745  */
746 int ocfs2_journal_access_path(struct ocfs2_caching_info *ci,
747 			      handle_t *handle,
748 			      struct ocfs2_path *path)
749 {
750 	int i, ret = 0;
751 
752 	if (!path)
753 		goto out;
754 
755 	for(i = 0; i < path_num_items(path); i++) {
756 		ret = ocfs2_path_bh_journal_access(handle, ci, path, i);
757 		if (ret < 0) {
758 			mlog_errno(ret);
759 			goto out;
760 		}
761 	}
762 
763 out:
764 	return ret;
765 }
766 
767 /*
768  * Return the index of the extent record which contains cluster #v_cluster.
769  * -1 is returned if it was not found.
770  *
771  * Should work fine on interior and exterior nodes.
772  */
773 int ocfs2_search_extent_list(struct ocfs2_extent_list *el, u32 v_cluster)
774 {
775 	int ret = -1;
776 	int i;
777 	struct ocfs2_extent_rec *rec;
778 	u32 rec_end, rec_start, clusters;
779 
780 	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
781 		rec = &el->l_recs[i];
782 
783 		rec_start = le32_to_cpu(rec->e_cpos);
784 		clusters = ocfs2_rec_clusters(el, rec);
785 
786 		rec_end = rec_start + clusters;
787 
788 		if (v_cluster >= rec_start && v_cluster < rec_end) {
789 			ret = i;
790 			break;
791 		}
792 	}
793 
794 	return ret;
795 }
796 
797 /*
798  * NOTE: ocfs2_block_extent_contig(), ocfs2_extents_adjacent() and
799  * ocfs2_extent_rec_contig only work properly against leaf nodes!
800  */
801 static int ocfs2_block_extent_contig(struct super_block *sb,
802 				     struct ocfs2_extent_rec *ext,
803 				     u64 blkno)
804 {
805 	u64 blk_end = le64_to_cpu(ext->e_blkno);
806 
807 	blk_end += ocfs2_clusters_to_blocks(sb,
808 				    le16_to_cpu(ext->e_leaf_clusters));
809 
810 	return blkno == blk_end;
811 }
812 
813 static int ocfs2_extents_adjacent(struct ocfs2_extent_rec *left,
814 				  struct ocfs2_extent_rec *right)
815 {
816 	u32 left_range;
817 
818 	left_range = le32_to_cpu(left->e_cpos) +
819 		le16_to_cpu(left->e_leaf_clusters);
820 
821 	return (left_range == le32_to_cpu(right->e_cpos));
822 }
823 
824 static enum ocfs2_contig_type
825 	ocfs2_extent_rec_contig(struct super_block *sb,
826 				struct ocfs2_extent_rec *ext,
827 				struct ocfs2_extent_rec *insert_rec)
828 {
829 	u64 blkno = le64_to_cpu(insert_rec->e_blkno);
830 
831 	/*
832 	 * Refuse to coalesce extent records with different flag
833 	 * fields - we don't want to mix unwritten extents with user
834 	 * data.
835 	 */
836 	if (ext->e_flags != insert_rec->e_flags)
837 		return CONTIG_NONE;
838 
839 	if (ocfs2_extents_adjacent(ext, insert_rec) &&
840 	    ocfs2_block_extent_contig(sb, ext, blkno))
841 			return CONTIG_RIGHT;
842 
843 	blkno = le64_to_cpu(ext->e_blkno);
844 	if (ocfs2_extents_adjacent(insert_rec, ext) &&
845 	    ocfs2_block_extent_contig(sb, insert_rec, blkno))
846 		return CONTIG_LEFT;
847 
848 	return CONTIG_NONE;
849 }
850 
851 /*
852  * NOTE: We can have pretty much any combination of contiguousness and
853  * appending.
854  *
855  * The usefulness of APPEND_TAIL is more in that it lets us know that
856  * we'll have to update the path to that leaf.
857  */
858 enum ocfs2_append_type {
859 	APPEND_NONE = 0,
860 	APPEND_TAIL,
861 };
862 
863 enum ocfs2_split_type {
864 	SPLIT_NONE = 0,
865 	SPLIT_LEFT,
866 	SPLIT_RIGHT,
867 };
868 
869 struct ocfs2_insert_type {
870 	enum ocfs2_split_type	ins_split;
871 	enum ocfs2_append_type	ins_appending;
872 	enum ocfs2_contig_type	ins_contig;
873 	int			ins_contig_index;
874 	int			ins_tree_depth;
875 };
876 
877 struct ocfs2_merge_ctxt {
878 	enum ocfs2_contig_type	c_contig_type;
879 	int			c_has_empty_extent;
880 	int			c_split_covers_rec;
881 };
882 
883 static int ocfs2_validate_extent_block(struct super_block *sb,
884 				       struct buffer_head *bh)
885 {
886 	int rc;
887 	struct ocfs2_extent_block *eb =
888 		(struct ocfs2_extent_block *)bh->b_data;
889 
890 	trace_ocfs2_validate_extent_block((unsigned long long)bh->b_blocknr);
891 
892 	BUG_ON(!buffer_uptodate(bh));
893 
894 	/*
895 	 * If the ecc fails, we return the error but otherwise
896 	 * leave the filesystem running.  We know any error is
897 	 * local to this block.
898 	 */
899 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &eb->h_check);
900 	if (rc) {
901 		mlog(ML_ERROR, "Checksum failed for extent block %llu\n",
902 		     (unsigned long long)bh->b_blocknr);
903 		return rc;
904 	}
905 
906 	/*
907 	 * Errors after here are fatal.
908 	 */
909 
910 	if (!OCFS2_IS_VALID_EXTENT_BLOCK(eb)) {
911 		ocfs2_error(sb,
912 			    "Extent block #%llu has bad signature %.*s",
913 			    (unsigned long long)bh->b_blocknr, 7,
914 			    eb->h_signature);
915 		return -EINVAL;
916 	}
917 
918 	if (le64_to_cpu(eb->h_blkno) != bh->b_blocknr) {
919 		ocfs2_error(sb,
920 			    "Extent block #%llu has an invalid h_blkno "
921 			    "of %llu",
922 			    (unsigned long long)bh->b_blocknr,
923 			    (unsigned long long)le64_to_cpu(eb->h_blkno));
924 		return -EINVAL;
925 	}
926 
927 	if (le32_to_cpu(eb->h_fs_generation) != OCFS2_SB(sb)->fs_generation) {
928 		ocfs2_error(sb,
929 			    "Extent block #%llu has an invalid "
930 			    "h_fs_generation of #%u",
931 			    (unsigned long long)bh->b_blocknr,
932 			    le32_to_cpu(eb->h_fs_generation));
933 		return -EINVAL;
934 	}
935 
936 	return 0;
937 }
938 
939 int ocfs2_read_extent_block(struct ocfs2_caching_info *ci, u64 eb_blkno,
940 			    struct buffer_head **bh)
941 {
942 	int rc;
943 	struct buffer_head *tmp = *bh;
944 
945 	rc = ocfs2_read_block(ci, eb_blkno, &tmp,
946 			      ocfs2_validate_extent_block);
947 
948 	/* If ocfs2_read_block() got us a new bh, pass it up. */
949 	if (!rc && !*bh)
950 		*bh = tmp;
951 
952 	return rc;
953 }
954 
955 
956 /*
957  * How many free extents have we got before we need more meta data?
958  */
959 int ocfs2_num_free_extents(struct ocfs2_super *osb,
960 			   struct ocfs2_extent_tree *et)
961 {
962 	int retval;
963 	struct ocfs2_extent_list *el = NULL;
964 	struct ocfs2_extent_block *eb;
965 	struct buffer_head *eb_bh = NULL;
966 	u64 last_eb_blk = 0;
967 
968 	el = et->et_root_el;
969 	last_eb_blk = ocfs2_et_get_last_eb_blk(et);
970 
971 	if (last_eb_blk) {
972 		retval = ocfs2_read_extent_block(et->et_ci, last_eb_blk,
973 						 &eb_bh);
974 		if (retval < 0) {
975 			mlog_errno(retval);
976 			goto bail;
977 		}
978 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
979 		el = &eb->h_list;
980 	}
981 
982 	BUG_ON(el->l_tree_depth != 0);
983 
984 	retval = le16_to_cpu(el->l_count) - le16_to_cpu(el->l_next_free_rec);
985 bail:
986 	brelse(eb_bh);
987 
988 	trace_ocfs2_num_free_extents(retval);
989 	return retval;
990 }
991 
992 /* expects array to already be allocated
993  *
994  * sets h_signature, h_blkno, h_suballoc_bit, h_suballoc_slot, and
995  * l_count for you
996  */
997 static int ocfs2_create_new_meta_bhs(handle_t *handle,
998 				     struct ocfs2_extent_tree *et,
999 				     int wanted,
1000 				     struct ocfs2_alloc_context *meta_ac,
1001 				     struct buffer_head *bhs[])
1002 {
1003 	int count, status, i;
1004 	u16 suballoc_bit_start;
1005 	u32 num_got;
1006 	u64 suballoc_loc, first_blkno;
1007 	struct ocfs2_super *osb =
1008 		OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
1009 	struct ocfs2_extent_block *eb;
1010 
1011 	count = 0;
1012 	while (count < wanted) {
1013 		status = ocfs2_claim_metadata(handle,
1014 					      meta_ac,
1015 					      wanted - count,
1016 					      &suballoc_loc,
1017 					      &suballoc_bit_start,
1018 					      &num_got,
1019 					      &first_blkno);
1020 		if (status < 0) {
1021 			mlog_errno(status);
1022 			goto bail;
1023 		}
1024 
1025 		for(i = count;  i < (num_got + count); i++) {
1026 			bhs[i] = sb_getblk(osb->sb, first_blkno);
1027 			if (bhs[i] == NULL) {
1028 				status = -ENOMEM;
1029 				mlog_errno(status);
1030 				goto bail;
1031 			}
1032 			ocfs2_set_new_buffer_uptodate(et->et_ci, bhs[i]);
1033 
1034 			status = ocfs2_journal_access_eb(handle, et->et_ci,
1035 							 bhs[i],
1036 							 OCFS2_JOURNAL_ACCESS_CREATE);
1037 			if (status < 0) {
1038 				mlog_errno(status);
1039 				goto bail;
1040 			}
1041 
1042 			memset(bhs[i]->b_data, 0, osb->sb->s_blocksize);
1043 			eb = (struct ocfs2_extent_block *) bhs[i]->b_data;
1044 			/* Ok, setup the minimal stuff here. */
1045 			strcpy(eb->h_signature, OCFS2_EXTENT_BLOCK_SIGNATURE);
1046 			eb->h_blkno = cpu_to_le64(first_blkno);
1047 			eb->h_fs_generation = cpu_to_le32(osb->fs_generation);
1048 			eb->h_suballoc_slot =
1049 				cpu_to_le16(meta_ac->ac_alloc_slot);
1050 			eb->h_suballoc_loc = cpu_to_le64(suballoc_loc);
1051 			eb->h_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1052 			eb->h_list.l_count =
1053 				cpu_to_le16(ocfs2_extent_recs_per_eb(osb->sb));
1054 
1055 			suballoc_bit_start++;
1056 			first_blkno++;
1057 
1058 			/* We'll also be dirtied by the caller, so
1059 			 * this isn't absolutely necessary. */
1060 			ocfs2_journal_dirty(handle, bhs[i]);
1061 		}
1062 
1063 		count += num_got;
1064 	}
1065 
1066 	status = 0;
1067 bail:
1068 	if (status < 0) {
1069 		for(i = 0; i < wanted; i++) {
1070 			brelse(bhs[i]);
1071 			bhs[i] = NULL;
1072 		}
1073 		mlog_errno(status);
1074 	}
1075 	return status;
1076 }
1077 
1078 /*
1079  * Helper function for ocfs2_add_branch() and ocfs2_shift_tree_depth().
1080  *
1081  * Returns the sum of the rightmost extent rec logical offset and
1082  * cluster count.
1083  *
1084  * ocfs2_add_branch() uses this to determine what logical cluster
1085  * value should be populated into the leftmost new branch records.
1086  *
1087  * ocfs2_shift_tree_depth() uses this to determine the # clusters
1088  * value for the new topmost tree record.
1089  */
1090 static inline u32 ocfs2_sum_rightmost_rec(struct ocfs2_extent_list  *el)
1091 {
1092 	int i;
1093 
1094 	i = le16_to_cpu(el->l_next_free_rec) - 1;
1095 
1096 	return le32_to_cpu(el->l_recs[i].e_cpos) +
1097 		ocfs2_rec_clusters(el, &el->l_recs[i]);
1098 }
1099 
1100 /*
1101  * Change range of the branches in the right most path according to the leaf
1102  * extent block's rightmost record.
1103  */
1104 static int ocfs2_adjust_rightmost_branch(handle_t *handle,
1105 					 struct ocfs2_extent_tree *et)
1106 {
1107 	int status;
1108 	struct ocfs2_path *path = NULL;
1109 	struct ocfs2_extent_list *el;
1110 	struct ocfs2_extent_rec *rec;
1111 
1112 	path = ocfs2_new_path_from_et(et);
1113 	if (!path) {
1114 		status = -ENOMEM;
1115 		return status;
1116 	}
1117 
1118 	status = ocfs2_find_path(et->et_ci, path, UINT_MAX);
1119 	if (status < 0) {
1120 		mlog_errno(status);
1121 		goto out;
1122 	}
1123 
1124 	status = ocfs2_extend_trans(handle, path_num_items(path));
1125 	if (status < 0) {
1126 		mlog_errno(status);
1127 		goto out;
1128 	}
1129 
1130 	status = ocfs2_journal_access_path(et->et_ci, handle, path);
1131 	if (status < 0) {
1132 		mlog_errno(status);
1133 		goto out;
1134 	}
1135 
1136 	el = path_leaf_el(path);
1137 	rec = &el->l_recs[le16_to_cpu(el->l_next_free_rec) - 1];
1138 
1139 	ocfs2_adjust_rightmost_records(handle, et, path, rec);
1140 
1141 out:
1142 	ocfs2_free_path(path);
1143 	return status;
1144 }
1145 
1146 /*
1147  * Add an entire tree branch to our inode. eb_bh is the extent block
1148  * to start at, if we don't want to start the branch at the root
1149  * structure.
1150  *
1151  * last_eb_bh is required as we have to update it's next_leaf pointer
1152  * for the new last extent block.
1153  *
1154  * the new branch will be 'empty' in the sense that every block will
1155  * contain a single record with cluster count == 0.
1156  */
1157 static int ocfs2_add_branch(handle_t *handle,
1158 			    struct ocfs2_extent_tree *et,
1159 			    struct buffer_head *eb_bh,
1160 			    struct buffer_head **last_eb_bh,
1161 			    struct ocfs2_alloc_context *meta_ac)
1162 {
1163 	int status, new_blocks, i;
1164 	u64 next_blkno, new_last_eb_blk;
1165 	struct buffer_head *bh;
1166 	struct buffer_head **new_eb_bhs = NULL;
1167 	struct ocfs2_extent_block *eb;
1168 	struct ocfs2_extent_list  *eb_el;
1169 	struct ocfs2_extent_list  *el;
1170 	u32 new_cpos, root_end;
1171 
1172 	BUG_ON(!last_eb_bh || !*last_eb_bh);
1173 
1174 	if (eb_bh) {
1175 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
1176 		el = &eb->h_list;
1177 	} else
1178 		el = et->et_root_el;
1179 
1180 	/* we never add a branch to a leaf. */
1181 	BUG_ON(!el->l_tree_depth);
1182 
1183 	new_blocks = le16_to_cpu(el->l_tree_depth);
1184 
1185 	eb = (struct ocfs2_extent_block *)(*last_eb_bh)->b_data;
1186 	new_cpos = ocfs2_sum_rightmost_rec(&eb->h_list);
1187 	root_end = ocfs2_sum_rightmost_rec(et->et_root_el);
1188 
1189 	/*
1190 	 * If there is a gap before the root end and the real end
1191 	 * of the righmost leaf block, we need to remove the gap
1192 	 * between new_cpos and root_end first so that the tree
1193 	 * is consistent after we add a new branch(it will start
1194 	 * from new_cpos).
1195 	 */
1196 	if (root_end > new_cpos) {
1197 		trace_ocfs2_adjust_rightmost_branch(
1198 			(unsigned long long)
1199 			ocfs2_metadata_cache_owner(et->et_ci),
1200 			root_end, new_cpos);
1201 
1202 		status = ocfs2_adjust_rightmost_branch(handle, et);
1203 		if (status) {
1204 			mlog_errno(status);
1205 			goto bail;
1206 		}
1207 	}
1208 
1209 	/* allocate the number of new eb blocks we need */
1210 	new_eb_bhs = kcalloc(new_blocks, sizeof(struct buffer_head *),
1211 			     GFP_KERNEL);
1212 	if (!new_eb_bhs) {
1213 		status = -ENOMEM;
1214 		mlog_errno(status);
1215 		goto bail;
1216 	}
1217 
1218 	status = ocfs2_create_new_meta_bhs(handle, et, new_blocks,
1219 					   meta_ac, new_eb_bhs);
1220 	if (status < 0) {
1221 		mlog_errno(status);
1222 		goto bail;
1223 	}
1224 
1225 	/* Note: new_eb_bhs[new_blocks - 1] is the guy which will be
1226 	 * linked with the rest of the tree.
1227 	 * conversly, new_eb_bhs[0] is the new bottommost leaf.
1228 	 *
1229 	 * when we leave the loop, new_last_eb_blk will point to the
1230 	 * newest leaf, and next_blkno will point to the topmost extent
1231 	 * block. */
1232 	next_blkno = new_last_eb_blk = 0;
1233 	for(i = 0; i < new_blocks; i++) {
1234 		bh = new_eb_bhs[i];
1235 		eb = (struct ocfs2_extent_block *) bh->b_data;
1236 		/* ocfs2_create_new_meta_bhs() should create it right! */
1237 		BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
1238 		eb_el = &eb->h_list;
1239 
1240 		status = ocfs2_journal_access_eb(handle, et->et_ci, bh,
1241 						 OCFS2_JOURNAL_ACCESS_CREATE);
1242 		if (status < 0) {
1243 			mlog_errno(status);
1244 			goto bail;
1245 		}
1246 
1247 		eb->h_next_leaf_blk = 0;
1248 		eb_el->l_tree_depth = cpu_to_le16(i);
1249 		eb_el->l_next_free_rec = cpu_to_le16(1);
1250 		/*
1251 		 * This actually counts as an empty extent as
1252 		 * c_clusters == 0
1253 		 */
1254 		eb_el->l_recs[0].e_cpos = cpu_to_le32(new_cpos);
1255 		eb_el->l_recs[0].e_blkno = cpu_to_le64(next_blkno);
1256 		/*
1257 		 * eb_el isn't always an interior node, but even leaf
1258 		 * nodes want a zero'd flags and reserved field so
1259 		 * this gets the whole 32 bits regardless of use.
1260 		 */
1261 		eb_el->l_recs[0].e_int_clusters = cpu_to_le32(0);
1262 		if (!eb_el->l_tree_depth)
1263 			new_last_eb_blk = le64_to_cpu(eb->h_blkno);
1264 
1265 		ocfs2_journal_dirty(handle, bh);
1266 		next_blkno = le64_to_cpu(eb->h_blkno);
1267 	}
1268 
1269 	/* This is a bit hairy. We want to update up to three blocks
1270 	 * here without leaving any of them in an inconsistent state
1271 	 * in case of error. We don't have to worry about
1272 	 * journal_dirty erroring as it won't unless we've aborted the
1273 	 * handle (in which case we would never be here) so reserving
1274 	 * the write with journal_access is all we need to do. */
1275 	status = ocfs2_journal_access_eb(handle, et->et_ci, *last_eb_bh,
1276 					 OCFS2_JOURNAL_ACCESS_WRITE);
1277 	if (status < 0) {
1278 		mlog_errno(status);
1279 		goto bail;
1280 	}
1281 	status = ocfs2_et_root_journal_access(handle, et,
1282 					      OCFS2_JOURNAL_ACCESS_WRITE);
1283 	if (status < 0) {
1284 		mlog_errno(status);
1285 		goto bail;
1286 	}
1287 	if (eb_bh) {
1288 		status = ocfs2_journal_access_eb(handle, et->et_ci, eb_bh,
1289 						 OCFS2_JOURNAL_ACCESS_WRITE);
1290 		if (status < 0) {
1291 			mlog_errno(status);
1292 			goto bail;
1293 		}
1294 	}
1295 
1296 	/* Link the new branch into the rest of the tree (el will
1297 	 * either be on the root_bh, or the extent block passed in. */
1298 	i = le16_to_cpu(el->l_next_free_rec);
1299 	el->l_recs[i].e_blkno = cpu_to_le64(next_blkno);
1300 	el->l_recs[i].e_cpos = cpu_to_le32(new_cpos);
1301 	el->l_recs[i].e_int_clusters = 0;
1302 	le16_add_cpu(&el->l_next_free_rec, 1);
1303 
1304 	/* fe needs a new last extent block pointer, as does the
1305 	 * next_leaf on the previously last-extent-block. */
1306 	ocfs2_et_set_last_eb_blk(et, new_last_eb_blk);
1307 
1308 	eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
1309 	eb->h_next_leaf_blk = cpu_to_le64(new_last_eb_blk);
1310 
1311 	ocfs2_journal_dirty(handle, *last_eb_bh);
1312 	ocfs2_journal_dirty(handle, et->et_root_bh);
1313 	if (eb_bh)
1314 		ocfs2_journal_dirty(handle, eb_bh);
1315 
1316 	/*
1317 	 * Some callers want to track the rightmost leaf so pass it
1318 	 * back here.
1319 	 */
1320 	brelse(*last_eb_bh);
1321 	get_bh(new_eb_bhs[0]);
1322 	*last_eb_bh = new_eb_bhs[0];
1323 
1324 	status = 0;
1325 bail:
1326 	if (new_eb_bhs) {
1327 		for (i = 0; i < new_blocks; i++)
1328 			brelse(new_eb_bhs[i]);
1329 		kfree(new_eb_bhs);
1330 	}
1331 
1332 	return status;
1333 }
1334 
1335 /*
1336  * adds another level to the allocation tree.
1337  * returns back the new extent block so you can add a branch to it
1338  * after this call.
1339  */
1340 static int ocfs2_shift_tree_depth(handle_t *handle,
1341 				  struct ocfs2_extent_tree *et,
1342 				  struct ocfs2_alloc_context *meta_ac,
1343 				  struct buffer_head **ret_new_eb_bh)
1344 {
1345 	int status, i;
1346 	u32 new_clusters;
1347 	struct buffer_head *new_eb_bh = NULL;
1348 	struct ocfs2_extent_block *eb;
1349 	struct ocfs2_extent_list  *root_el;
1350 	struct ocfs2_extent_list  *eb_el;
1351 
1352 	status = ocfs2_create_new_meta_bhs(handle, et, 1, meta_ac,
1353 					   &new_eb_bh);
1354 	if (status < 0) {
1355 		mlog_errno(status);
1356 		goto bail;
1357 	}
1358 
1359 	eb = (struct ocfs2_extent_block *) new_eb_bh->b_data;
1360 	/* ocfs2_create_new_meta_bhs() should create it right! */
1361 	BUG_ON(!OCFS2_IS_VALID_EXTENT_BLOCK(eb));
1362 
1363 	eb_el = &eb->h_list;
1364 	root_el = et->et_root_el;
1365 
1366 	status = ocfs2_journal_access_eb(handle, et->et_ci, new_eb_bh,
1367 					 OCFS2_JOURNAL_ACCESS_CREATE);
1368 	if (status < 0) {
1369 		mlog_errno(status);
1370 		goto bail;
1371 	}
1372 
1373 	/* copy the root extent list data into the new extent block */
1374 	eb_el->l_tree_depth = root_el->l_tree_depth;
1375 	eb_el->l_next_free_rec = root_el->l_next_free_rec;
1376 	for (i = 0; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1377 		eb_el->l_recs[i] = root_el->l_recs[i];
1378 
1379 	ocfs2_journal_dirty(handle, new_eb_bh);
1380 
1381 	status = ocfs2_et_root_journal_access(handle, et,
1382 					      OCFS2_JOURNAL_ACCESS_WRITE);
1383 	if (status < 0) {
1384 		mlog_errno(status);
1385 		goto bail;
1386 	}
1387 
1388 	new_clusters = ocfs2_sum_rightmost_rec(eb_el);
1389 
1390 	/* update root_bh now */
1391 	le16_add_cpu(&root_el->l_tree_depth, 1);
1392 	root_el->l_recs[0].e_cpos = 0;
1393 	root_el->l_recs[0].e_blkno = eb->h_blkno;
1394 	root_el->l_recs[0].e_int_clusters = cpu_to_le32(new_clusters);
1395 	for (i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
1396 		memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
1397 	root_el->l_next_free_rec = cpu_to_le16(1);
1398 
1399 	/* If this is our 1st tree depth shift, then last_eb_blk
1400 	 * becomes the allocated extent block */
1401 	if (root_el->l_tree_depth == cpu_to_le16(1))
1402 		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
1403 
1404 	ocfs2_journal_dirty(handle, et->et_root_bh);
1405 
1406 	*ret_new_eb_bh = new_eb_bh;
1407 	new_eb_bh = NULL;
1408 	status = 0;
1409 bail:
1410 	brelse(new_eb_bh);
1411 
1412 	return status;
1413 }
1414 
1415 /*
1416  * Should only be called when there is no space left in any of the
1417  * leaf nodes. What we want to do is find the lowest tree depth
1418  * non-leaf extent block with room for new records. There are three
1419  * valid results of this search:
1420  *
1421  * 1) a lowest extent block is found, then we pass it back in
1422  *    *lowest_eb_bh and return '0'
1423  *
1424  * 2) the search fails to find anything, but the root_el has room. We
1425  *    pass NULL back in *lowest_eb_bh, but still return '0'
1426  *
1427  * 3) the search fails to find anything AND the root_el is full, in
1428  *    which case we return > 0
1429  *
1430  * return status < 0 indicates an error.
1431  */
1432 static int ocfs2_find_branch_target(struct ocfs2_extent_tree *et,
1433 				    struct buffer_head **target_bh)
1434 {
1435 	int status = 0, i;
1436 	u64 blkno;
1437 	struct ocfs2_extent_block *eb;
1438 	struct ocfs2_extent_list  *el;
1439 	struct buffer_head *bh = NULL;
1440 	struct buffer_head *lowest_bh = NULL;
1441 
1442 	*target_bh = NULL;
1443 
1444 	el = et->et_root_el;
1445 
1446 	while(le16_to_cpu(el->l_tree_depth) > 1) {
1447 		if (le16_to_cpu(el->l_next_free_rec) == 0) {
1448 			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
1449 				    "Owner %llu has empty "
1450 				    "extent list (next_free_rec == 0)",
1451 				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
1452 			status = -EIO;
1453 			goto bail;
1454 		}
1455 		i = le16_to_cpu(el->l_next_free_rec) - 1;
1456 		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1457 		if (!blkno) {
1458 			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
1459 				    "Owner %llu has extent "
1460 				    "list where extent # %d has no physical "
1461 				    "block start",
1462 				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci), i);
1463 			status = -EIO;
1464 			goto bail;
1465 		}
1466 
1467 		brelse(bh);
1468 		bh = NULL;
1469 
1470 		status = ocfs2_read_extent_block(et->et_ci, blkno, &bh);
1471 		if (status < 0) {
1472 			mlog_errno(status);
1473 			goto bail;
1474 		}
1475 
1476 		eb = (struct ocfs2_extent_block *) bh->b_data;
1477 		el = &eb->h_list;
1478 
1479 		if (le16_to_cpu(el->l_next_free_rec) <
1480 		    le16_to_cpu(el->l_count)) {
1481 			brelse(lowest_bh);
1482 			lowest_bh = bh;
1483 			get_bh(lowest_bh);
1484 		}
1485 	}
1486 
1487 	/* If we didn't find one and the fe doesn't have any room,
1488 	 * then return '1' */
1489 	el = et->et_root_el;
1490 	if (!lowest_bh && (el->l_next_free_rec == el->l_count))
1491 		status = 1;
1492 
1493 	*target_bh = lowest_bh;
1494 bail:
1495 	brelse(bh);
1496 
1497 	return status;
1498 }
1499 
1500 /*
1501  * Grow a b-tree so that it has more records.
1502  *
1503  * We might shift the tree depth in which case existing paths should
1504  * be considered invalid.
1505  *
1506  * Tree depth after the grow is returned via *final_depth.
1507  *
1508  * *last_eb_bh will be updated by ocfs2_add_branch().
1509  */
1510 static int ocfs2_grow_tree(handle_t *handle, struct ocfs2_extent_tree *et,
1511 			   int *final_depth, struct buffer_head **last_eb_bh,
1512 			   struct ocfs2_alloc_context *meta_ac)
1513 {
1514 	int ret, shift;
1515 	struct ocfs2_extent_list *el = et->et_root_el;
1516 	int depth = le16_to_cpu(el->l_tree_depth);
1517 	struct buffer_head *bh = NULL;
1518 
1519 	BUG_ON(meta_ac == NULL);
1520 
1521 	shift = ocfs2_find_branch_target(et, &bh);
1522 	if (shift < 0) {
1523 		ret = shift;
1524 		mlog_errno(ret);
1525 		goto out;
1526 	}
1527 
1528 	/* We traveled all the way to the bottom of the allocation tree
1529 	 * and didn't find room for any more extents - we need to add
1530 	 * another tree level */
1531 	if (shift) {
1532 		BUG_ON(bh);
1533 		trace_ocfs2_grow_tree(
1534 			(unsigned long long)
1535 			ocfs2_metadata_cache_owner(et->et_ci),
1536 			depth);
1537 
1538 		/* ocfs2_shift_tree_depth will return us a buffer with
1539 		 * the new extent block (so we can pass that to
1540 		 * ocfs2_add_branch). */
1541 		ret = ocfs2_shift_tree_depth(handle, et, meta_ac, &bh);
1542 		if (ret < 0) {
1543 			mlog_errno(ret);
1544 			goto out;
1545 		}
1546 		depth++;
1547 		if (depth == 1) {
1548 			/*
1549 			 * Special case: we have room now if we shifted from
1550 			 * tree_depth 0, so no more work needs to be done.
1551 			 *
1552 			 * We won't be calling add_branch, so pass
1553 			 * back *last_eb_bh as the new leaf. At depth
1554 			 * zero, it should always be null so there's
1555 			 * no reason to brelse.
1556 			 */
1557 			BUG_ON(*last_eb_bh);
1558 			get_bh(bh);
1559 			*last_eb_bh = bh;
1560 			goto out;
1561 		}
1562 	}
1563 
1564 	/* call ocfs2_add_branch to add the final part of the tree with
1565 	 * the new data. */
1566 	ret = ocfs2_add_branch(handle, et, bh, last_eb_bh,
1567 			       meta_ac);
1568 	if (ret < 0) {
1569 		mlog_errno(ret);
1570 		goto out;
1571 	}
1572 
1573 out:
1574 	if (final_depth)
1575 		*final_depth = depth;
1576 	brelse(bh);
1577 	return ret;
1578 }
1579 
1580 /*
1581  * This function will discard the rightmost extent record.
1582  */
1583 static void ocfs2_shift_records_right(struct ocfs2_extent_list *el)
1584 {
1585 	int next_free = le16_to_cpu(el->l_next_free_rec);
1586 	int count = le16_to_cpu(el->l_count);
1587 	unsigned int num_bytes;
1588 
1589 	BUG_ON(!next_free);
1590 	/* This will cause us to go off the end of our extent list. */
1591 	BUG_ON(next_free >= count);
1592 
1593 	num_bytes = sizeof(struct ocfs2_extent_rec) * next_free;
1594 
1595 	memmove(&el->l_recs[1], &el->l_recs[0], num_bytes);
1596 }
1597 
1598 static void ocfs2_rotate_leaf(struct ocfs2_extent_list *el,
1599 			      struct ocfs2_extent_rec *insert_rec)
1600 {
1601 	int i, insert_index, next_free, has_empty, num_bytes;
1602 	u32 insert_cpos = le32_to_cpu(insert_rec->e_cpos);
1603 	struct ocfs2_extent_rec *rec;
1604 
1605 	next_free = le16_to_cpu(el->l_next_free_rec);
1606 	has_empty = ocfs2_is_empty_extent(&el->l_recs[0]);
1607 
1608 	BUG_ON(!next_free);
1609 
1610 	/* The tree code before us didn't allow enough room in the leaf. */
1611 	BUG_ON(el->l_next_free_rec == el->l_count && !has_empty);
1612 
1613 	/*
1614 	 * The easiest way to approach this is to just remove the
1615 	 * empty extent and temporarily decrement next_free.
1616 	 */
1617 	if (has_empty) {
1618 		/*
1619 		 * If next_free was 1 (only an empty extent), this
1620 		 * loop won't execute, which is fine. We still want
1621 		 * the decrement above to happen.
1622 		 */
1623 		for(i = 0; i < (next_free - 1); i++)
1624 			el->l_recs[i] = el->l_recs[i+1];
1625 
1626 		next_free--;
1627 	}
1628 
1629 	/*
1630 	 * Figure out what the new record index should be.
1631 	 */
1632 	for(i = 0; i < next_free; i++) {
1633 		rec = &el->l_recs[i];
1634 
1635 		if (insert_cpos < le32_to_cpu(rec->e_cpos))
1636 			break;
1637 	}
1638 	insert_index = i;
1639 
1640 	trace_ocfs2_rotate_leaf(insert_cpos, insert_index,
1641 				has_empty, next_free,
1642 				le16_to_cpu(el->l_count));
1643 
1644 	BUG_ON(insert_index < 0);
1645 	BUG_ON(insert_index >= le16_to_cpu(el->l_count));
1646 	BUG_ON(insert_index > next_free);
1647 
1648 	/*
1649 	 * No need to memmove if we're just adding to the tail.
1650 	 */
1651 	if (insert_index != next_free) {
1652 		BUG_ON(next_free >= le16_to_cpu(el->l_count));
1653 
1654 		num_bytes = next_free - insert_index;
1655 		num_bytes *= sizeof(struct ocfs2_extent_rec);
1656 		memmove(&el->l_recs[insert_index + 1],
1657 			&el->l_recs[insert_index],
1658 			num_bytes);
1659 	}
1660 
1661 	/*
1662 	 * Either we had an empty extent, and need to re-increment or
1663 	 * there was no empty extent on a non full rightmost leaf node,
1664 	 * in which case we still need to increment.
1665 	 */
1666 	next_free++;
1667 	el->l_next_free_rec = cpu_to_le16(next_free);
1668 	/*
1669 	 * Make sure none of the math above just messed up our tree.
1670 	 */
1671 	BUG_ON(le16_to_cpu(el->l_next_free_rec) > le16_to_cpu(el->l_count));
1672 
1673 	el->l_recs[insert_index] = *insert_rec;
1674 
1675 }
1676 
1677 static void ocfs2_remove_empty_extent(struct ocfs2_extent_list *el)
1678 {
1679 	int size, num_recs = le16_to_cpu(el->l_next_free_rec);
1680 
1681 	BUG_ON(num_recs == 0);
1682 
1683 	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
1684 		num_recs--;
1685 		size = num_recs * sizeof(struct ocfs2_extent_rec);
1686 		memmove(&el->l_recs[0], &el->l_recs[1], size);
1687 		memset(&el->l_recs[num_recs], 0,
1688 		       sizeof(struct ocfs2_extent_rec));
1689 		el->l_next_free_rec = cpu_to_le16(num_recs);
1690 	}
1691 }
1692 
1693 /*
1694  * Create an empty extent record .
1695  *
1696  * l_next_free_rec may be updated.
1697  *
1698  * If an empty extent already exists do nothing.
1699  */
1700 static void ocfs2_create_empty_extent(struct ocfs2_extent_list *el)
1701 {
1702 	int next_free = le16_to_cpu(el->l_next_free_rec);
1703 
1704 	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
1705 
1706 	if (next_free == 0)
1707 		goto set_and_inc;
1708 
1709 	if (ocfs2_is_empty_extent(&el->l_recs[0]))
1710 		return;
1711 
1712 	mlog_bug_on_msg(el->l_count == el->l_next_free_rec,
1713 			"Asked to create an empty extent in a full list:\n"
1714 			"count = %u, tree depth = %u",
1715 			le16_to_cpu(el->l_count),
1716 			le16_to_cpu(el->l_tree_depth));
1717 
1718 	ocfs2_shift_records_right(el);
1719 
1720 set_and_inc:
1721 	le16_add_cpu(&el->l_next_free_rec, 1);
1722 	memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
1723 }
1724 
1725 /*
1726  * For a rotation which involves two leaf nodes, the "root node" is
1727  * the lowest level tree node which contains a path to both leafs. This
1728  * resulting set of information can be used to form a complete "subtree"
1729  *
1730  * This function is passed two full paths from the dinode down to a
1731  * pair of adjacent leaves. It's task is to figure out which path
1732  * index contains the subtree root - this can be the root index itself
1733  * in a worst-case rotation.
1734  *
1735  * The array index of the subtree root is passed back.
1736  */
1737 int ocfs2_find_subtree_root(struct ocfs2_extent_tree *et,
1738 			    struct ocfs2_path *left,
1739 			    struct ocfs2_path *right)
1740 {
1741 	int i = 0;
1742 
1743 	/*
1744 	 * Check that the caller passed in two paths from the same tree.
1745 	 */
1746 	BUG_ON(path_root_bh(left) != path_root_bh(right));
1747 
1748 	do {
1749 		i++;
1750 
1751 		/*
1752 		 * The caller didn't pass two adjacent paths.
1753 		 */
1754 		mlog_bug_on_msg(i > left->p_tree_depth,
1755 				"Owner %llu, left depth %u, right depth %u\n"
1756 				"left leaf blk %llu, right leaf blk %llu\n",
1757 				(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
1758 				left->p_tree_depth, right->p_tree_depth,
1759 				(unsigned long long)path_leaf_bh(left)->b_blocknr,
1760 				(unsigned long long)path_leaf_bh(right)->b_blocknr);
1761 	} while (left->p_node[i].bh->b_blocknr ==
1762 		 right->p_node[i].bh->b_blocknr);
1763 
1764 	return i - 1;
1765 }
1766 
1767 typedef void (path_insert_t)(void *, struct buffer_head *);
1768 
1769 /*
1770  * Traverse a btree path in search of cpos, starting at root_el.
1771  *
1772  * This code can be called with a cpos larger than the tree, in which
1773  * case it will return the rightmost path.
1774  */
1775 static int __ocfs2_find_path(struct ocfs2_caching_info *ci,
1776 			     struct ocfs2_extent_list *root_el, u32 cpos,
1777 			     path_insert_t *func, void *data)
1778 {
1779 	int i, ret = 0;
1780 	u32 range;
1781 	u64 blkno;
1782 	struct buffer_head *bh = NULL;
1783 	struct ocfs2_extent_block *eb;
1784 	struct ocfs2_extent_list *el;
1785 	struct ocfs2_extent_rec *rec;
1786 
1787 	el = root_el;
1788 	while (el->l_tree_depth) {
1789 		if (le16_to_cpu(el->l_next_free_rec) == 0) {
1790 			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1791 				    "Owner %llu has empty extent list at "
1792 				    "depth %u\n",
1793 				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
1794 				    le16_to_cpu(el->l_tree_depth));
1795 			ret = -EROFS;
1796 			goto out;
1797 
1798 		}
1799 
1800 		for(i = 0; i < le16_to_cpu(el->l_next_free_rec) - 1; i++) {
1801 			rec = &el->l_recs[i];
1802 
1803 			/*
1804 			 * In the case that cpos is off the allocation
1805 			 * tree, this should just wind up returning the
1806 			 * rightmost record.
1807 			 */
1808 			range = le32_to_cpu(rec->e_cpos) +
1809 				ocfs2_rec_clusters(el, rec);
1810 			if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
1811 			    break;
1812 		}
1813 
1814 		blkno = le64_to_cpu(el->l_recs[i].e_blkno);
1815 		if (blkno == 0) {
1816 			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1817 				    "Owner %llu has bad blkno in extent list "
1818 				    "at depth %u (index %d)\n",
1819 				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
1820 				    le16_to_cpu(el->l_tree_depth), i);
1821 			ret = -EROFS;
1822 			goto out;
1823 		}
1824 
1825 		brelse(bh);
1826 		bh = NULL;
1827 		ret = ocfs2_read_extent_block(ci, blkno, &bh);
1828 		if (ret) {
1829 			mlog_errno(ret);
1830 			goto out;
1831 		}
1832 
1833 		eb = (struct ocfs2_extent_block *) bh->b_data;
1834 		el = &eb->h_list;
1835 
1836 		if (le16_to_cpu(el->l_next_free_rec) >
1837 		    le16_to_cpu(el->l_count)) {
1838 			ocfs2_error(ocfs2_metadata_cache_get_super(ci),
1839 				    "Owner %llu has bad count in extent list "
1840 				    "at block %llu (next free=%u, count=%u)\n",
1841 				    (unsigned long long)ocfs2_metadata_cache_owner(ci),
1842 				    (unsigned long long)bh->b_blocknr,
1843 				    le16_to_cpu(el->l_next_free_rec),
1844 				    le16_to_cpu(el->l_count));
1845 			ret = -EROFS;
1846 			goto out;
1847 		}
1848 
1849 		if (func)
1850 			func(data, bh);
1851 	}
1852 
1853 out:
1854 	/*
1855 	 * Catch any trailing bh that the loop didn't handle.
1856 	 */
1857 	brelse(bh);
1858 
1859 	return ret;
1860 }
1861 
1862 /*
1863  * Given an initialized path (that is, it has a valid root extent
1864  * list), this function will traverse the btree in search of the path
1865  * which would contain cpos.
1866  *
1867  * The path traveled is recorded in the path structure.
1868  *
1869  * Note that this will not do any comparisons on leaf node extent
1870  * records, so it will work fine in the case that we just added a tree
1871  * branch.
1872  */
1873 struct find_path_data {
1874 	int index;
1875 	struct ocfs2_path *path;
1876 };
1877 static void find_path_ins(void *data, struct buffer_head *bh)
1878 {
1879 	struct find_path_data *fp = data;
1880 
1881 	get_bh(bh);
1882 	ocfs2_path_insert_eb(fp->path, fp->index, bh);
1883 	fp->index++;
1884 }
1885 int ocfs2_find_path(struct ocfs2_caching_info *ci,
1886 		    struct ocfs2_path *path, u32 cpos)
1887 {
1888 	struct find_path_data data;
1889 
1890 	data.index = 1;
1891 	data.path = path;
1892 	return __ocfs2_find_path(ci, path_root_el(path), cpos,
1893 				 find_path_ins, &data);
1894 }
1895 
1896 static void find_leaf_ins(void *data, struct buffer_head *bh)
1897 {
1898 	struct ocfs2_extent_block *eb =(struct ocfs2_extent_block *)bh->b_data;
1899 	struct ocfs2_extent_list *el = &eb->h_list;
1900 	struct buffer_head **ret = data;
1901 
1902 	/* We want to retain only the leaf block. */
1903 	if (le16_to_cpu(el->l_tree_depth) == 0) {
1904 		get_bh(bh);
1905 		*ret = bh;
1906 	}
1907 }
1908 /*
1909  * Find the leaf block in the tree which would contain cpos. No
1910  * checking of the actual leaf is done.
1911  *
1912  * Some paths want to call this instead of allocating a path structure
1913  * and calling ocfs2_find_path().
1914  *
1915  * This function doesn't handle non btree extent lists.
1916  */
1917 int ocfs2_find_leaf(struct ocfs2_caching_info *ci,
1918 		    struct ocfs2_extent_list *root_el, u32 cpos,
1919 		    struct buffer_head **leaf_bh)
1920 {
1921 	int ret;
1922 	struct buffer_head *bh = NULL;
1923 
1924 	ret = __ocfs2_find_path(ci, root_el, cpos, find_leaf_ins, &bh);
1925 	if (ret) {
1926 		mlog_errno(ret);
1927 		goto out;
1928 	}
1929 
1930 	*leaf_bh = bh;
1931 out:
1932 	return ret;
1933 }
1934 
1935 /*
1936  * Adjust the adjacent records (left_rec, right_rec) involved in a rotation.
1937  *
1938  * Basically, we've moved stuff around at the bottom of the tree and
1939  * we need to fix up the extent records above the changes to reflect
1940  * the new changes.
1941  *
1942  * left_rec: the record on the left.
1943  * left_child_el: is the child list pointed to by left_rec
1944  * right_rec: the record to the right of left_rec
1945  * right_child_el: is the child list pointed to by right_rec
1946  *
1947  * By definition, this only works on interior nodes.
1948  */
1949 static void ocfs2_adjust_adjacent_records(struct ocfs2_extent_rec *left_rec,
1950 				  struct ocfs2_extent_list *left_child_el,
1951 				  struct ocfs2_extent_rec *right_rec,
1952 				  struct ocfs2_extent_list *right_child_el)
1953 {
1954 	u32 left_clusters, right_end;
1955 
1956 	/*
1957 	 * Interior nodes never have holes. Their cpos is the cpos of
1958 	 * the leftmost record in their child list. Their cluster
1959 	 * count covers the full theoretical range of their child list
1960 	 * - the range between their cpos and the cpos of the record
1961 	 * immediately to their right.
1962 	 */
1963 	left_clusters = le32_to_cpu(right_child_el->l_recs[0].e_cpos);
1964 	if (!ocfs2_rec_clusters(right_child_el, &right_child_el->l_recs[0])) {
1965 		BUG_ON(right_child_el->l_tree_depth);
1966 		BUG_ON(le16_to_cpu(right_child_el->l_next_free_rec) <= 1);
1967 		left_clusters = le32_to_cpu(right_child_el->l_recs[1].e_cpos);
1968 	}
1969 	left_clusters -= le32_to_cpu(left_rec->e_cpos);
1970 	left_rec->e_int_clusters = cpu_to_le32(left_clusters);
1971 
1972 	/*
1973 	 * Calculate the rightmost cluster count boundary before
1974 	 * moving cpos - we will need to adjust clusters after
1975 	 * updating e_cpos to keep the same highest cluster count.
1976 	 */
1977 	right_end = le32_to_cpu(right_rec->e_cpos);
1978 	right_end += le32_to_cpu(right_rec->e_int_clusters);
1979 
1980 	right_rec->e_cpos = left_rec->e_cpos;
1981 	le32_add_cpu(&right_rec->e_cpos, left_clusters);
1982 
1983 	right_end -= le32_to_cpu(right_rec->e_cpos);
1984 	right_rec->e_int_clusters = cpu_to_le32(right_end);
1985 }
1986 
1987 /*
1988  * Adjust the adjacent root node records involved in a
1989  * rotation. left_el_blkno is passed in as a key so that we can easily
1990  * find it's index in the root list.
1991  */
1992 static void ocfs2_adjust_root_records(struct ocfs2_extent_list *root_el,
1993 				      struct ocfs2_extent_list *left_el,
1994 				      struct ocfs2_extent_list *right_el,
1995 				      u64 left_el_blkno)
1996 {
1997 	int i;
1998 
1999 	BUG_ON(le16_to_cpu(root_el->l_tree_depth) <=
2000 	       le16_to_cpu(left_el->l_tree_depth));
2001 
2002 	for(i = 0; i < le16_to_cpu(root_el->l_next_free_rec) - 1; i++) {
2003 		if (le64_to_cpu(root_el->l_recs[i].e_blkno) == left_el_blkno)
2004 			break;
2005 	}
2006 
2007 	/*
2008 	 * The path walking code should have never returned a root and
2009 	 * two paths which are not adjacent.
2010 	 */
2011 	BUG_ON(i >= (le16_to_cpu(root_el->l_next_free_rec) - 1));
2012 
2013 	ocfs2_adjust_adjacent_records(&root_el->l_recs[i], left_el,
2014 				      &root_el->l_recs[i + 1], right_el);
2015 }
2016 
2017 /*
2018  * We've changed a leaf block (in right_path) and need to reflect that
2019  * change back up the subtree.
2020  *
2021  * This happens in multiple places:
2022  *   - When we've moved an extent record from the left path leaf to the right
2023  *     path leaf to make room for an empty extent in the left path leaf.
2024  *   - When our insert into the right path leaf is at the leftmost edge
2025  *     and requires an update of the path immediately to it's left. This
2026  *     can occur at the end of some types of rotation and appending inserts.
2027  *   - When we've adjusted the last extent record in the left path leaf and the
2028  *     1st extent record in the right path leaf during cross extent block merge.
2029  */
2030 static void ocfs2_complete_edge_insert(handle_t *handle,
2031 				       struct ocfs2_path *left_path,
2032 				       struct ocfs2_path *right_path,
2033 				       int subtree_index)
2034 {
2035 	int i, idx;
2036 	struct ocfs2_extent_list *el, *left_el, *right_el;
2037 	struct ocfs2_extent_rec *left_rec, *right_rec;
2038 	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2039 
2040 	/*
2041 	 * Update the counts and position values within all the
2042 	 * interior nodes to reflect the leaf rotation we just did.
2043 	 *
2044 	 * The root node is handled below the loop.
2045 	 *
2046 	 * We begin the loop with right_el and left_el pointing to the
2047 	 * leaf lists and work our way up.
2048 	 *
2049 	 * NOTE: within this loop, left_el and right_el always refer
2050 	 * to the *child* lists.
2051 	 */
2052 	left_el = path_leaf_el(left_path);
2053 	right_el = path_leaf_el(right_path);
2054 	for(i = left_path->p_tree_depth - 1; i > subtree_index; i--) {
2055 		trace_ocfs2_complete_edge_insert(i);
2056 
2057 		/*
2058 		 * One nice property of knowing that all of these
2059 		 * nodes are below the root is that we only deal with
2060 		 * the leftmost right node record and the rightmost
2061 		 * left node record.
2062 		 */
2063 		el = left_path->p_node[i].el;
2064 		idx = le16_to_cpu(left_el->l_next_free_rec) - 1;
2065 		left_rec = &el->l_recs[idx];
2066 
2067 		el = right_path->p_node[i].el;
2068 		right_rec = &el->l_recs[0];
2069 
2070 		ocfs2_adjust_adjacent_records(left_rec, left_el, right_rec,
2071 					      right_el);
2072 
2073 		ocfs2_journal_dirty(handle, left_path->p_node[i].bh);
2074 		ocfs2_journal_dirty(handle, right_path->p_node[i].bh);
2075 
2076 		/*
2077 		 * Setup our list pointers now so that the current
2078 		 * parents become children in the next iteration.
2079 		 */
2080 		left_el = left_path->p_node[i].el;
2081 		right_el = right_path->p_node[i].el;
2082 	}
2083 
2084 	/*
2085 	 * At the root node, adjust the two adjacent records which
2086 	 * begin our path to the leaves.
2087 	 */
2088 
2089 	el = left_path->p_node[subtree_index].el;
2090 	left_el = left_path->p_node[subtree_index + 1].el;
2091 	right_el = right_path->p_node[subtree_index + 1].el;
2092 
2093 	ocfs2_adjust_root_records(el, left_el, right_el,
2094 				  left_path->p_node[subtree_index + 1].bh->b_blocknr);
2095 
2096 	root_bh = left_path->p_node[subtree_index].bh;
2097 
2098 	ocfs2_journal_dirty(handle, root_bh);
2099 }
2100 
2101 static int ocfs2_rotate_subtree_right(handle_t *handle,
2102 				      struct ocfs2_extent_tree *et,
2103 				      struct ocfs2_path *left_path,
2104 				      struct ocfs2_path *right_path,
2105 				      int subtree_index)
2106 {
2107 	int ret, i;
2108 	struct buffer_head *right_leaf_bh;
2109 	struct buffer_head *left_leaf_bh = NULL;
2110 	struct buffer_head *root_bh;
2111 	struct ocfs2_extent_list *right_el, *left_el;
2112 	struct ocfs2_extent_rec move_rec;
2113 
2114 	left_leaf_bh = path_leaf_bh(left_path);
2115 	left_el = path_leaf_el(left_path);
2116 
2117 	if (left_el->l_next_free_rec != left_el->l_count) {
2118 		ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
2119 			    "Inode %llu has non-full interior leaf node %llu"
2120 			    "(next free = %u)",
2121 			    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2122 			    (unsigned long long)left_leaf_bh->b_blocknr,
2123 			    le16_to_cpu(left_el->l_next_free_rec));
2124 		return -EROFS;
2125 	}
2126 
2127 	/*
2128 	 * This extent block may already have an empty record, so we
2129 	 * return early if so.
2130 	 */
2131 	if (ocfs2_is_empty_extent(&left_el->l_recs[0]))
2132 		return 0;
2133 
2134 	root_bh = left_path->p_node[subtree_index].bh;
2135 	BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2136 
2137 	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
2138 					   subtree_index);
2139 	if (ret) {
2140 		mlog_errno(ret);
2141 		goto out;
2142 	}
2143 
2144 	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
2145 		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2146 						   right_path, i);
2147 		if (ret) {
2148 			mlog_errno(ret);
2149 			goto out;
2150 		}
2151 
2152 		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2153 						   left_path, i);
2154 		if (ret) {
2155 			mlog_errno(ret);
2156 			goto out;
2157 		}
2158 	}
2159 
2160 	right_leaf_bh = path_leaf_bh(right_path);
2161 	right_el = path_leaf_el(right_path);
2162 
2163 	/* This is a code error, not a disk corruption. */
2164 	mlog_bug_on_msg(!right_el->l_next_free_rec, "Inode %llu: Rotate fails "
2165 			"because rightmost leaf block %llu is empty\n",
2166 			(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2167 			(unsigned long long)right_leaf_bh->b_blocknr);
2168 
2169 	ocfs2_create_empty_extent(right_el);
2170 
2171 	ocfs2_journal_dirty(handle, right_leaf_bh);
2172 
2173 	/* Do the copy now. */
2174 	i = le16_to_cpu(left_el->l_next_free_rec) - 1;
2175 	move_rec = left_el->l_recs[i];
2176 	right_el->l_recs[0] = move_rec;
2177 
2178 	/*
2179 	 * Clear out the record we just copied and shift everything
2180 	 * over, leaving an empty extent in the left leaf.
2181 	 *
2182 	 * We temporarily subtract from next_free_rec so that the
2183 	 * shift will lose the tail record (which is now defunct).
2184 	 */
2185 	le16_add_cpu(&left_el->l_next_free_rec, -1);
2186 	ocfs2_shift_records_right(left_el);
2187 	memset(&left_el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2188 	le16_add_cpu(&left_el->l_next_free_rec, 1);
2189 
2190 	ocfs2_journal_dirty(handle, left_leaf_bh);
2191 
2192 	ocfs2_complete_edge_insert(handle, left_path, right_path,
2193 				   subtree_index);
2194 
2195 out:
2196 	return ret;
2197 }
2198 
2199 /*
2200  * Given a full path, determine what cpos value would return us a path
2201  * containing the leaf immediately to the left of the current one.
2202  *
2203  * Will return zero if the path passed in is already the leftmost path.
2204  */
2205 int ocfs2_find_cpos_for_left_leaf(struct super_block *sb,
2206 				  struct ocfs2_path *path, u32 *cpos)
2207 {
2208 	int i, j, ret = 0;
2209 	u64 blkno;
2210 	struct ocfs2_extent_list *el;
2211 
2212 	BUG_ON(path->p_tree_depth == 0);
2213 
2214 	*cpos = 0;
2215 
2216 	blkno = path_leaf_bh(path)->b_blocknr;
2217 
2218 	/* Start at the tree node just above the leaf and work our way up. */
2219 	i = path->p_tree_depth - 1;
2220 	while (i >= 0) {
2221 		el = path->p_node[i].el;
2222 
2223 		/*
2224 		 * Find the extent record just before the one in our
2225 		 * path.
2226 		 */
2227 		for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2228 			if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2229 				if (j == 0) {
2230 					if (i == 0) {
2231 						/*
2232 						 * We've determined that the
2233 						 * path specified is already
2234 						 * the leftmost one - return a
2235 						 * cpos of zero.
2236 						 */
2237 						goto out;
2238 					}
2239 					/*
2240 					 * The leftmost record points to our
2241 					 * leaf - we need to travel up the
2242 					 * tree one level.
2243 					 */
2244 					goto next_node;
2245 				}
2246 
2247 				*cpos = le32_to_cpu(el->l_recs[j - 1].e_cpos);
2248 				*cpos = *cpos + ocfs2_rec_clusters(el,
2249 							   &el->l_recs[j - 1]);
2250 				*cpos = *cpos - 1;
2251 				goto out;
2252 			}
2253 		}
2254 
2255 		/*
2256 		 * If we got here, we never found a valid node where
2257 		 * the tree indicated one should be.
2258 		 */
2259 		ocfs2_error(sb,
2260 			    "Invalid extent tree at extent block %llu\n",
2261 			    (unsigned long long)blkno);
2262 		ret = -EROFS;
2263 		goto out;
2264 
2265 next_node:
2266 		blkno = path->p_node[i].bh->b_blocknr;
2267 		i--;
2268 	}
2269 
2270 out:
2271 	return ret;
2272 }
2273 
2274 /*
2275  * Extend the transaction by enough credits to complete the rotation,
2276  * and still leave at least the original number of credits allocated
2277  * to this transaction.
2278  */
2279 static int ocfs2_extend_rotate_transaction(handle_t *handle, int subtree_depth,
2280 					   int op_credits,
2281 					   struct ocfs2_path *path)
2282 {
2283 	int ret = 0;
2284 	int credits = (path->p_tree_depth - subtree_depth) * 2 + 1 + op_credits;
2285 
2286 	if (handle->h_buffer_credits < credits)
2287 		ret = ocfs2_extend_trans(handle,
2288 					 credits - handle->h_buffer_credits);
2289 
2290 	return ret;
2291 }
2292 
2293 /*
2294  * Trap the case where we're inserting into the theoretical range past
2295  * the _actual_ left leaf range. Otherwise, we'll rotate a record
2296  * whose cpos is less than ours into the right leaf.
2297  *
2298  * It's only necessary to look at the rightmost record of the left
2299  * leaf because the logic that calls us should ensure that the
2300  * theoretical ranges in the path components above the leaves are
2301  * correct.
2302  */
2303 static int ocfs2_rotate_requires_path_adjustment(struct ocfs2_path *left_path,
2304 						 u32 insert_cpos)
2305 {
2306 	struct ocfs2_extent_list *left_el;
2307 	struct ocfs2_extent_rec *rec;
2308 	int next_free;
2309 
2310 	left_el = path_leaf_el(left_path);
2311 	next_free = le16_to_cpu(left_el->l_next_free_rec);
2312 	rec = &left_el->l_recs[next_free - 1];
2313 
2314 	if (insert_cpos > le32_to_cpu(rec->e_cpos))
2315 		return 1;
2316 	return 0;
2317 }
2318 
2319 static int ocfs2_leftmost_rec_contains(struct ocfs2_extent_list *el, u32 cpos)
2320 {
2321 	int next_free = le16_to_cpu(el->l_next_free_rec);
2322 	unsigned int range;
2323 	struct ocfs2_extent_rec *rec;
2324 
2325 	if (next_free == 0)
2326 		return 0;
2327 
2328 	rec = &el->l_recs[0];
2329 	if (ocfs2_is_empty_extent(rec)) {
2330 		/* Empty list. */
2331 		if (next_free == 1)
2332 			return 0;
2333 		rec = &el->l_recs[1];
2334 	}
2335 
2336 	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2337 	if (cpos >= le32_to_cpu(rec->e_cpos) && cpos < range)
2338 		return 1;
2339 	return 0;
2340 }
2341 
2342 /*
2343  * Rotate all the records in a btree right one record, starting at insert_cpos.
2344  *
2345  * The path to the rightmost leaf should be passed in.
2346  *
2347  * The array is assumed to be large enough to hold an entire path (tree depth).
2348  *
2349  * Upon successful return from this function:
2350  *
2351  * - The 'right_path' array will contain a path to the leaf block
2352  *   whose range contains e_cpos.
2353  * - That leaf block will have a single empty extent in list index 0.
2354  * - In the case that the rotation requires a post-insert update,
2355  *   *ret_left_path will contain a valid path which can be passed to
2356  *   ocfs2_insert_path().
2357  */
2358 static int ocfs2_rotate_tree_right(handle_t *handle,
2359 				   struct ocfs2_extent_tree *et,
2360 				   enum ocfs2_split_type split,
2361 				   u32 insert_cpos,
2362 				   struct ocfs2_path *right_path,
2363 				   struct ocfs2_path **ret_left_path)
2364 {
2365 	int ret, start, orig_credits = handle->h_buffer_credits;
2366 	u32 cpos;
2367 	struct ocfs2_path *left_path = NULL;
2368 	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
2369 
2370 	*ret_left_path = NULL;
2371 
2372 	left_path = ocfs2_new_path_from_path(right_path);
2373 	if (!left_path) {
2374 		ret = -ENOMEM;
2375 		mlog_errno(ret);
2376 		goto out;
2377 	}
2378 
2379 	ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
2380 	if (ret) {
2381 		mlog_errno(ret);
2382 		goto out;
2383 	}
2384 
2385 	trace_ocfs2_rotate_tree_right(
2386 		(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2387 		insert_cpos, cpos);
2388 
2389 	/*
2390 	 * What we want to do here is:
2391 	 *
2392 	 * 1) Start with the rightmost path.
2393 	 *
2394 	 * 2) Determine a path to the leaf block directly to the left
2395 	 *    of that leaf.
2396 	 *
2397 	 * 3) Determine the 'subtree root' - the lowest level tree node
2398 	 *    which contains a path to both leaves.
2399 	 *
2400 	 * 4) Rotate the subtree.
2401 	 *
2402 	 * 5) Find the next subtree by considering the left path to be
2403 	 *    the new right path.
2404 	 *
2405 	 * The check at the top of this while loop also accepts
2406 	 * insert_cpos == cpos because cpos is only a _theoretical_
2407 	 * value to get us the left path - insert_cpos might very well
2408 	 * be filling that hole.
2409 	 *
2410 	 * Stop at a cpos of '0' because we either started at the
2411 	 * leftmost branch (i.e., a tree with one branch and a
2412 	 * rotation inside of it), or we've gone as far as we can in
2413 	 * rotating subtrees.
2414 	 */
2415 	while (cpos && insert_cpos <= cpos) {
2416 		trace_ocfs2_rotate_tree_right(
2417 			(unsigned long long)
2418 			ocfs2_metadata_cache_owner(et->et_ci),
2419 			insert_cpos, cpos);
2420 
2421 		ret = ocfs2_find_path(et->et_ci, left_path, cpos);
2422 		if (ret) {
2423 			mlog_errno(ret);
2424 			goto out;
2425 		}
2426 
2427 		mlog_bug_on_msg(path_leaf_bh(left_path) ==
2428 				path_leaf_bh(right_path),
2429 				"Owner %llu: error during insert of %u "
2430 				"(left path cpos %u) results in two identical "
2431 				"paths ending at %llu\n",
2432 				(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2433 				insert_cpos, cpos,
2434 				(unsigned long long)
2435 				path_leaf_bh(left_path)->b_blocknr);
2436 
2437 		if (split == SPLIT_NONE &&
2438 		    ocfs2_rotate_requires_path_adjustment(left_path,
2439 							  insert_cpos)) {
2440 
2441 			/*
2442 			 * We've rotated the tree as much as we
2443 			 * should. The rest is up to
2444 			 * ocfs2_insert_path() to complete, after the
2445 			 * record insertion. We indicate this
2446 			 * situation by returning the left path.
2447 			 *
2448 			 * The reason we don't adjust the records here
2449 			 * before the record insert is that an error
2450 			 * later might break the rule where a parent
2451 			 * record e_cpos will reflect the actual
2452 			 * e_cpos of the 1st nonempty record of the
2453 			 * child list.
2454 			 */
2455 			*ret_left_path = left_path;
2456 			goto out_ret_path;
2457 		}
2458 
2459 		start = ocfs2_find_subtree_root(et, left_path, right_path);
2460 
2461 		trace_ocfs2_rotate_subtree(start,
2462 			(unsigned long long)
2463 			right_path->p_node[start].bh->b_blocknr,
2464 			right_path->p_tree_depth);
2465 
2466 		ret = ocfs2_extend_rotate_transaction(handle, start,
2467 						      orig_credits, right_path);
2468 		if (ret) {
2469 			mlog_errno(ret);
2470 			goto out;
2471 		}
2472 
2473 		ret = ocfs2_rotate_subtree_right(handle, et, left_path,
2474 						 right_path, start);
2475 		if (ret) {
2476 			mlog_errno(ret);
2477 			goto out;
2478 		}
2479 
2480 		if (split != SPLIT_NONE &&
2481 		    ocfs2_leftmost_rec_contains(path_leaf_el(right_path),
2482 						insert_cpos)) {
2483 			/*
2484 			 * A rotate moves the rightmost left leaf
2485 			 * record over to the leftmost right leaf
2486 			 * slot. If we're doing an extent split
2487 			 * instead of a real insert, then we have to
2488 			 * check that the extent to be split wasn't
2489 			 * just moved over. If it was, then we can
2490 			 * exit here, passing left_path back -
2491 			 * ocfs2_split_extent() is smart enough to
2492 			 * search both leaves.
2493 			 */
2494 			*ret_left_path = left_path;
2495 			goto out_ret_path;
2496 		}
2497 
2498 		/*
2499 		 * There is no need to re-read the next right path
2500 		 * as we know that it'll be our current left
2501 		 * path. Optimize by copying values instead.
2502 		 */
2503 		ocfs2_mv_path(right_path, left_path);
2504 
2505 		ret = ocfs2_find_cpos_for_left_leaf(sb, right_path, &cpos);
2506 		if (ret) {
2507 			mlog_errno(ret);
2508 			goto out;
2509 		}
2510 	}
2511 
2512 out:
2513 	ocfs2_free_path(left_path);
2514 
2515 out_ret_path:
2516 	return ret;
2517 }
2518 
2519 static int ocfs2_update_edge_lengths(handle_t *handle,
2520 				     struct ocfs2_extent_tree *et,
2521 				     int subtree_index, struct ocfs2_path *path)
2522 {
2523 	int i, idx, ret;
2524 	struct ocfs2_extent_rec *rec;
2525 	struct ocfs2_extent_list *el;
2526 	struct ocfs2_extent_block *eb;
2527 	u32 range;
2528 
2529 	/*
2530 	 * In normal tree rotation process, we will never touch the
2531 	 * tree branch above subtree_index and ocfs2_extend_rotate_transaction
2532 	 * doesn't reserve the credits for them either.
2533 	 *
2534 	 * But we do have a special case here which will update the rightmost
2535 	 * records for all the bh in the path.
2536 	 * So we have to allocate extra credits and access them.
2537 	 */
2538 	ret = ocfs2_extend_trans(handle, subtree_index);
2539 	if (ret) {
2540 		mlog_errno(ret);
2541 		goto out;
2542 	}
2543 
2544 	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
2545 	if (ret) {
2546 		mlog_errno(ret);
2547 		goto out;
2548 	}
2549 
2550 	/* Path should always be rightmost. */
2551 	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
2552 	BUG_ON(eb->h_next_leaf_blk != 0ULL);
2553 
2554 	el = &eb->h_list;
2555 	BUG_ON(le16_to_cpu(el->l_next_free_rec) == 0);
2556 	idx = le16_to_cpu(el->l_next_free_rec) - 1;
2557 	rec = &el->l_recs[idx];
2558 	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
2559 
2560 	for (i = 0; i < path->p_tree_depth; i++) {
2561 		el = path->p_node[i].el;
2562 		idx = le16_to_cpu(el->l_next_free_rec) - 1;
2563 		rec = &el->l_recs[idx];
2564 
2565 		rec->e_int_clusters = cpu_to_le32(range);
2566 		le32_add_cpu(&rec->e_int_clusters, -le32_to_cpu(rec->e_cpos));
2567 
2568 		ocfs2_journal_dirty(handle, path->p_node[i].bh);
2569 	}
2570 out:
2571 	return ret;
2572 }
2573 
2574 static void ocfs2_unlink_path(handle_t *handle,
2575 			      struct ocfs2_extent_tree *et,
2576 			      struct ocfs2_cached_dealloc_ctxt *dealloc,
2577 			      struct ocfs2_path *path, int unlink_start)
2578 {
2579 	int ret, i;
2580 	struct ocfs2_extent_block *eb;
2581 	struct ocfs2_extent_list *el;
2582 	struct buffer_head *bh;
2583 
2584 	for(i = unlink_start; i < path_num_items(path); i++) {
2585 		bh = path->p_node[i].bh;
2586 
2587 		eb = (struct ocfs2_extent_block *)bh->b_data;
2588 		/*
2589 		 * Not all nodes might have had their final count
2590 		 * decremented by the caller - handle this here.
2591 		 */
2592 		el = &eb->h_list;
2593 		if (le16_to_cpu(el->l_next_free_rec) > 1) {
2594 			mlog(ML_ERROR,
2595 			     "Inode %llu, attempted to remove extent block "
2596 			     "%llu with %u records\n",
2597 			     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
2598 			     (unsigned long long)le64_to_cpu(eb->h_blkno),
2599 			     le16_to_cpu(el->l_next_free_rec));
2600 
2601 			ocfs2_journal_dirty(handle, bh);
2602 			ocfs2_remove_from_cache(et->et_ci, bh);
2603 			continue;
2604 		}
2605 
2606 		el->l_next_free_rec = 0;
2607 		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
2608 
2609 		ocfs2_journal_dirty(handle, bh);
2610 
2611 		ret = ocfs2_cache_extent_block_free(dealloc, eb);
2612 		if (ret)
2613 			mlog_errno(ret);
2614 
2615 		ocfs2_remove_from_cache(et->et_ci, bh);
2616 	}
2617 }
2618 
2619 static void ocfs2_unlink_subtree(handle_t *handle,
2620 				 struct ocfs2_extent_tree *et,
2621 				 struct ocfs2_path *left_path,
2622 				 struct ocfs2_path *right_path,
2623 				 int subtree_index,
2624 				 struct ocfs2_cached_dealloc_ctxt *dealloc)
2625 {
2626 	int i;
2627 	struct buffer_head *root_bh = left_path->p_node[subtree_index].bh;
2628 	struct ocfs2_extent_list *root_el = left_path->p_node[subtree_index].el;
2629 	struct ocfs2_extent_list *el;
2630 	struct ocfs2_extent_block *eb;
2631 
2632 	el = path_leaf_el(left_path);
2633 
2634 	eb = (struct ocfs2_extent_block *)right_path->p_node[subtree_index + 1].bh->b_data;
2635 
2636 	for(i = 1; i < le16_to_cpu(root_el->l_next_free_rec); i++)
2637 		if (root_el->l_recs[i].e_blkno == eb->h_blkno)
2638 			break;
2639 
2640 	BUG_ON(i >= le16_to_cpu(root_el->l_next_free_rec));
2641 
2642 	memset(&root_el->l_recs[i], 0, sizeof(struct ocfs2_extent_rec));
2643 	le16_add_cpu(&root_el->l_next_free_rec, -1);
2644 
2645 	eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2646 	eb->h_next_leaf_blk = 0;
2647 
2648 	ocfs2_journal_dirty(handle, root_bh);
2649 	ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2650 
2651 	ocfs2_unlink_path(handle, et, dealloc, right_path,
2652 			  subtree_index + 1);
2653 }
2654 
2655 static int ocfs2_rotate_subtree_left(handle_t *handle,
2656 				     struct ocfs2_extent_tree *et,
2657 				     struct ocfs2_path *left_path,
2658 				     struct ocfs2_path *right_path,
2659 				     int subtree_index,
2660 				     struct ocfs2_cached_dealloc_ctxt *dealloc,
2661 				     int *deleted)
2662 {
2663 	int ret, i, del_right_subtree = 0, right_has_empty = 0;
2664 	struct buffer_head *root_bh, *et_root_bh = path_root_bh(right_path);
2665 	struct ocfs2_extent_list *right_leaf_el, *left_leaf_el;
2666 	struct ocfs2_extent_block *eb;
2667 
2668 	*deleted = 0;
2669 
2670 	right_leaf_el = path_leaf_el(right_path);
2671 	left_leaf_el = path_leaf_el(left_path);
2672 	root_bh = left_path->p_node[subtree_index].bh;
2673 	BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
2674 
2675 	if (!ocfs2_is_empty_extent(&left_leaf_el->l_recs[0]))
2676 		return 0;
2677 
2678 	eb = (struct ocfs2_extent_block *)path_leaf_bh(right_path)->b_data;
2679 	if (ocfs2_is_empty_extent(&right_leaf_el->l_recs[0])) {
2680 		/*
2681 		 * It's legal for us to proceed if the right leaf is
2682 		 * the rightmost one and it has an empty extent. There
2683 		 * are two cases to handle - whether the leaf will be
2684 		 * empty after removal or not. If the leaf isn't empty
2685 		 * then just remove the empty extent up front. The
2686 		 * next block will handle empty leaves by flagging
2687 		 * them for unlink.
2688 		 *
2689 		 * Non rightmost leaves will throw -EAGAIN and the
2690 		 * caller can manually move the subtree and retry.
2691 		 */
2692 
2693 		if (eb->h_next_leaf_blk != 0ULL)
2694 			return -EAGAIN;
2695 
2696 		if (le16_to_cpu(right_leaf_el->l_next_free_rec) > 1) {
2697 			ret = ocfs2_journal_access_eb(handle, et->et_ci,
2698 						      path_leaf_bh(right_path),
2699 						      OCFS2_JOURNAL_ACCESS_WRITE);
2700 			if (ret) {
2701 				mlog_errno(ret);
2702 				goto out;
2703 			}
2704 
2705 			ocfs2_remove_empty_extent(right_leaf_el);
2706 		} else
2707 			right_has_empty = 1;
2708 	}
2709 
2710 	if (eb->h_next_leaf_blk == 0ULL &&
2711 	    le16_to_cpu(right_leaf_el->l_next_free_rec) == 1) {
2712 		/*
2713 		 * We have to update i_last_eb_blk during the meta
2714 		 * data delete.
2715 		 */
2716 		ret = ocfs2_et_root_journal_access(handle, et,
2717 						   OCFS2_JOURNAL_ACCESS_WRITE);
2718 		if (ret) {
2719 			mlog_errno(ret);
2720 			goto out;
2721 		}
2722 
2723 		del_right_subtree = 1;
2724 	}
2725 
2726 	/*
2727 	 * Getting here with an empty extent in the right path implies
2728 	 * that it's the rightmost path and will be deleted.
2729 	 */
2730 	BUG_ON(right_has_empty && !del_right_subtree);
2731 
2732 	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
2733 					   subtree_index);
2734 	if (ret) {
2735 		mlog_errno(ret);
2736 		goto out;
2737 	}
2738 
2739 	for(i = subtree_index + 1; i < path_num_items(right_path); i++) {
2740 		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2741 						   right_path, i);
2742 		if (ret) {
2743 			mlog_errno(ret);
2744 			goto out;
2745 		}
2746 
2747 		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2748 						   left_path, i);
2749 		if (ret) {
2750 			mlog_errno(ret);
2751 			goto out;
2752 		}
2753 	}
2754 
2755 	if (!right_has_empty) {
2756 		/*
2757 		 * Only do this if we're moving a real
2758 		 * record. Otherwise, the action is delayed until
2759 		 * after removal of the right path in which case we
2760 		 * can do a simple shift to remove the empty extent.
2761 		 */
2762 		ocfs2_rotate_leaf(left_leaf_el, &right_leaf_el->l_recs[0]);
2763 		memset(&right_leaf_el->l_recs[0], 0,
2764 		       sizeof(struct ocfs2_extent_rec));
2765 	}
2766 	if (eb->h_next_leaf_blk == 0ULL) {
2767 		/*
2768 		 * Move recs over to get rid of empty extent, decrease
2769 		 * next_free. This is allowed to remove the last
2770 		 * extent in our leaf (setting l_next_free_rec to
2771 		 * zero) - the delete code below won't care.
2772 		 */
2773 		ocfs2_remove_empty_extent(right_leaf_el);
2774 	}
2775 
2776 	ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
2777 	ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
2778 
2779 	if (del_right_subtree) {
2780 		ocfs2_unlink_subtree(handle, et, left_path, right_path,
2781 				     subtree_index, dealloc);
2782 		ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
2783 						left_path);
2784 		if (ret) {
2785 			mlog_errno(ret);
2786 			goto out;
2787 		}
2788 
2789 		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
2790 		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
2791 
2792 		/*
2793 		 * Removal of the extent in the left leaf was skipped
2794 		 * above so we could delete the right path
2795 		 * 1st.
2796 		 */
2797 		if (right_has_empty)
2798 			ocfs2_remove_empty_extent(left_leaf_el);
2799 
2800 		ocfs2_journal_dirty(handle, et_root_bh);
2801 
2802 		*deleted = 1;
2803 	} else
2804 		ocfs2_complete_edge_insert(handle, left_path, right_path,
2805 					   subtree_index);
2806 
2807 out:
2808 	return ret;
2809 }
2810 
2811 /*
2812  * Given a full path, determine what cpos value would return us a path
2813  * containing the leaf immediately to the right of the current one.
2814  *
2815  * Will return zero if the path passed in is already the rightmost path.
2816  *
2817  * This looks similar, but is subtly different to
2818  * ocfs2_find_cpos_for_left_leaf().
2819  */
2820 int ocfs2_find_cpos_for_right_leaf(struct super_block *sb,
2821 				   struct ocfs2_path *path, u32 *cpos)
2822 {
2823 	int i, j, ret = 0;
2824 	u64 blkno;
2825 	struct ocfs2_extent_list *el;
2826 
2827 	*cpos = 0;
2828 
2829 	if (path->p_tree_depth == 0)
2830 		return 0;
2831 
2832 	blkno = path_leaf_bh(path)->b_blocknr;
2833 
2834 	/* Start at the tree node just above the leaf and work our way up. */
2835 	i = path->p_tree_depth - 1;
2836 	while (i >= 0) {
2837 		int next_free;
2838 
2839 		el = path->p_node[i].el;
2840 
2841 		/*
2842 		 * Find the extent record just after the one in our
2843 		 * path.
2844 		 */
2845 		next_free = le16_to_cpu(el->l_next_free_rec);
2846 		for(j = 0; j < le16_to_cpu(el->l_next_free_rec); j++) {
2847 			if (le64_to_cpu(el->l_recs[j].e_blkno) == blkno) {
2848 				if (j == (next_free - 1)) {
2849 					if (i == 0) {
2850 						/*
2851 						 * We've determined that the
2852 						 * path specified is already
2853 						 * the rightmost one - return a
2854 						 * cpos of zero.
2855 						 */
2856 						goto out;
2857 					}
2858 					/*
2859 					 * The rightmost record points to our
2860 					 * leaf - we need to travel up the
2861 					 * tree one level.
2862 					 */
2863 					goto next_node;
2864 				}
2865 
2866 				*cpos = le32_to_cpu(el->l_recs[j + 1].e_cpos);
2867 				goto out;
2868 			}
2869 		}
2870 
2871 		/*
2872 		 * If we got here, we never found a valid node where
2873 		 * the tree indicated one should be.
2874 		 */
2875 		ocfs2_error(sb,
2876 			    "Invalid extent tree at extent block %llu\n",
2877 			    (unsigned long long)blkno);
2878 		ret = -EROFS;
2879 		goto out;
2880 
2881 next_node:
2882 		blkno = path->p_node[i].bh->b_blocknr;
2883 		i--;
2884 	}
2885 
2886 out:
2887 	return ret;
2888 }
2889 
2890 static int ocfs2_rotate_rightmost_leaf_left(handle_t *handle,
2891 					    struct ocfs2_extent_tree *et,
2892 					    struct ocfs2_path *path)
2893 {
2894 	int ret;
2895 	struct buffer_head *bh = path_leaf_bh(path);
2896 	struct ocfs2_extent_list *el = path_leaf_el(path);
2897 
2898 	if (!ocfs2_is_empty_extent(&el->l_recs[0]))
2899 		return 0;
2900 
2901 	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
2902 					   path_num_items(path) - 1);
2903 	if (ret) {
2904 		mlog_errno(ret);
2905 		goto out;
2906 	}
2907 
2908 	ocfs2_remove_empty_extent(el);
2909 	ocfs2_journal_dirty(handle, bh);
2910 
2911 out:
2912 	return ret;
2913 }
2914 
2915 static int __ocfs2_rotate_tree_left(handle_t *handle,
2916 				    struct ocfs2_extent_tree *et,
2917 				    int orig_credits,
2918 				    struct ocfs2_path *path,
2919 				    struct ocfs2_cached_dealloc_ctxt *dealloc,
2920 				    struct ocfs2_path **empty_extent_path)
2921 {
2922 	int ret, subtree_root, deleted;
2923 	u32 right_cpos;
2924 	struct ocfs2_path *left_path = NULL;
2925 	struct ocfs2_path *right_path = NULL;
2926 	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
2927 
2928 	BUG_ON(!ocfs2_is_empty_extent(&(path_leaf_el(path)->l_recs[0])));
2929 
2930 	*empty_extent_path = NULL;
2931 
2932 	ret = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
2933 	if (ret) {
2934 		mlog_errno(ret);
2935 		goto out;
2936 	}
2937 
2938 	left_path = ocfs2_new_path_from_path(path);
2939 	if (!left_path) {
2940 		ret = -ENOMEM;
2941 		mlog_errno(ret);
2942 		goto out;
2943 	}
2944 
2945 	ocfs2_cp_path(left_path, path);
2946 
2947 	right_path = ocfs2_new_path_from_path(path);
2948 	if (!right_path) {
2949 		ret = -ENOMEM;
2950 		mlog_errno(ret);
2951 		goto out;
2952 	}
2953 
2954 	while (right_cpos) {
2955 		ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
2956 		if (ret) {
2957 			mlog_errno(ret);
2958 			goto out;
2959 		}
2960 
2961 		subtree_root = ocfs2_find_subtree_root(et, left_path,
2962 						       right_path);
2963 
2964 		trace_ocfs2_rotate_subtree(subtree_root,
2965 		     (unsigned long long)
2966 		     right_path->p_node[subtree_root].bh->b_blocknr,
2967 		     right_path->p_tree_depth);
2968 
2969 		ret = ocfs2_extend_rotate_transaction(handle, subtree_root,
2970 						      orig_credits, left_path);
2971 		if (ret) {
2972 			mlog_errno(ret);
2973 			goto out;
2974 		}
2975 
2976 		/*
2977 		 * Caller might still want to make changes to the
2978 		 * tree root, so re-add it to the journal here.
2979 		 */
2980 		ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
2981 						   left_path, 0);
2982 		if (ret) {
2983 			mlog_errno(ret);
2984 			goto out;
2985 		}
2986 
2987 		ret = ocfs2_rotate_subtree_left(handle, et, left_path,
2988 						right_path, subtree_root,
2989 						dealloc, &deleted);
2990 		if (ret == -EAGAIN) {
2991 			/*
2992 			 * The rotation has to temporarily stop due to
2993 			 * the right subtree having an empty
2994 			 * extent. Pass it back to the caller for a
2995 			 * fixup.
2996 			 */
2997 			*empty_extent_path = right_path;
2998 			right_path = NULL;
2999 			goto out;
3000 		}
3001 		if (ret) {
3002 			mlog_errno(ret);
3003 			goto out;
3004 		}
3005 
3006 		/*
3007 		 * The subtree rotate might have removed records on
3008 		 * the rightmost edge. If so, then rotation is
3009 		 * complete.
3010 		 */
3011 		if (deleted)
3012 			break;
3013 
3014 		ocfs2_mv_path(left_path, right_path);
3015 
3016 		ret = ocfs2_find_cpos_for_right_leaf(sb, left_path,
3017 						     &right_cpos);
3018 		if (ret) {
3019 			mlog_errno(ret);
3020 			goto out;
3021 		}
3022 	}
3023 
3024 out:
3025 	ocfs2_free_path(right_path);
3026 	ocfs2_free_path(left_path);
3027 
3028 	return ret;
3029 }
3030 
3031 static int ocfs2_remove_rightmost_path(handle_t *handle,
3032 				struct ocfs2_extent_tree *et,
3033 				struct ocfs2_path *path,
3034 				struct ocfs2_cached_dealloc_ctxt *dealloc)
3035 {
3036 	int ret, subtree_index;
3037 	u32 cpos;
3038 	struct ocfs2_path *left_path = NULL;
3039 	struct ocfs2_extent_block *eb;
3040 	struct ocfs2_extent_list *el;
3041 
3042 
3043 	ret = ocfs2_et_sanity_check(et);
3044 	if (ret)
3045 		goto out;
3046 	/*
3047 	 * There's two ways we handle this depending on
3048 	 * whether path is the only existing one.
3049 	 */
3050 	ret = ocfs2_extend_rotate_transaction(handle, 0,
3051 					      handle->h_buffer_credits,
3052 					      path);
3053 	if (ret) {
3054 		mlog_errno(ret);
3055 		goto out;
3056 	}
3057 
3058 	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
3059 	if (ret) {
3060 		mlog_errno(ret);
3061 		goto out;
3062 	}
3063 
3064 	ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3065 					    path, &cpos);
3066 	if (ret) {
3067 		mlog_errno(ret);
3068 		goto out;
3069 	}
3070 
3071 	if (cpos) {
3072 		/*
3073 		 * We have a path to the left of this one - it needs
3074 		 * an update too.
3075 		 */
3076 		left_path = ocfs2_new_path_from_path(path);
3077 		if (!left_path) {
3078 			ret = -ENOMEM;
3079 			mlog_errno(ret);
3080 			goto out;
3081 		}
3082 
3083 		ret = ocfs2_find_path(et->et_ci, left_path, cpos);
3084 		if (ret) {
3085 			mlog_errno(ret);
3086 			goto out;
3087 		}
3088 
3089 		ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
3090 		if (ret) {
3091 			mlog_errno(ret);
3092 			goto out;
3093 		}
3094 
3095 		subtree_index = ocfs2_find_subtree_root(et, left_path, path);
3096 
3097 		ocfs2_unlink_subtree(handle, et, left_path, path,
3098 				     subtree_index, dealloc);
3099 		ret = ocfs2_update_edge_lengths(handle, et, subtree_index,
3100 						left_path);
3101 		if (ret) {
3102 			mlog_errno(ret);
3103 			goto out;
3104 		}
3105 
3106 		eb = (struct ocfs2_extent_block *)path_leaf_bh(left_path)->b_data;
3107 		ocfs2_et_set_last_eb_blk(et, le64_to_cpu(eb->h_blkno));
3108 	} else {
3109 		/*
3110 		 * 'path' is also the leftmost path which
3111 		 * means it must be the only one. This gets
3112 		 * handled differently because we want to
3113 		 * revert the root back to having extents
3114 		 * in-line.
3115 		 */
3116 		ocfs2_unlink_path(handle, et, dealloc, path, 1);
3117 
3118 		el = et->et_root_el;
3119 		el->l_tree_depth = 0;
3120 		el->l_next_free_rec = 0;
3121 		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3122 
3123 		ocfs2_et_set_last_eb_blk(et, 0);
3124 	}
3125 
3126 	ocfs2_journal_dirty(handle, path_root_bh(path));
3127 
3128 out:
3129 	ocfs2_free_path(left_path);
3130 	return ret;
3131 }
3132 
3133 /*
3134  * Left rotation of btree records.
3135  *
3136  * In many ways, this is (unsurprisingly) the opposite of right
3137  * rotation. We start at some non-rightmost path containing an empty
3138  * extent in the leaf block. The code works its way to the rightmost
3139  * path by rotating records to the left in every subtree.
3140  *
3141  * This is used by any code which reduces the number of extent records
3142  * in a leaf. After removal, an empty record should be placed in the
3143  * leftmost list position.
3144  *
3145  * This won't handle a length update of the rightmost path records if
3146  * the rightmost tree leaf record is removed so the caller is
3147  * responsible for detecting and correcting that.
3148  */
3149 static int ocfs2_rotate_tree_left(handle_t *handle,
3150 				  struct ocfs2_extent_tree *et,
3151 				  struct ocfs2_path *path,
3152 				  struct ocfs2_cached_dealloc_ctxt *dealloc)
3153 {
3154 	int ret, orig_credits = handle->h_buffer_credits;
3155 	struct ocfs2_path *tmp_path = NULL, *restart_path = NULL;
3156 	struct ocfs2_extent_block *eb;
3157 	struct ocfs2_extent_list *el;
3158 
3159 	el = path_leaf_el(path);
3160 	if (!ocfs2_is_empty_extent(&el->l_recs[0]))
3161 		return 0;
3162 
3163 	if (path->p_tree_depth == 0) {
3164 rightmost_no_delete:
3165 		/*
3166 		 * Inline extents. This is trivially handled, so do
3167 		 * it up front.
3168 		 */
3169 		ret = ocfs2_rotate_rightmost_leaf_left(handle, et, path);
3170 		if (ret)
3171 			mlog_errno(ret);
3172 		goto out;
3173 	}
3174 
3175 	/*
3176 	 * Handle rightmost branch now. There's several cases:
3177 	 *  1) simple rotation leaving records in there. That's trivial.
3178 	 *  2) rotation requiring a branch delete - there's no more
3179 	 *     records left. Two cases of this:
3180 	 *     a) There are branches to the left.
3181 	 *     b) This is also the leftmost (the only) branch.
3182 	 *
3183 	 *  1) is handled via ocfs2_rotate_rightmost_leaf_left()
3184 	 *  2a) we need the left branch so that we can update it with the unlink
3185 	 *  2b) we need to bring the root back to inline extents.
3186 	 */
3187 
3188 	eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
3189 	el = &eb->h_list;
3190 	if (eb->h_next_leaf_blk == 0) {
3191 		/*
3192 		 * This gets a bit tricky if we're going to delete the
3193 		 * rightmost path. Get the other cases out of the way
3194 		 * 1st.
3195 		 */
3196 		if (le16_to_cpu(el->l_next_free_rec) > 1)
3197 			goto rightmost_no_delete;
3198 
3199 		if (le16_to_cpu(el->l_next_free_rec) == 0) {
3200 			ret = -EIO;
3201 			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
3202 				    "Owner %llu has empty extent block at %llu",
3203 				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
3204 				    (unsigned long long)le64_to_cpu(eb->h_blkno));
3205 			goto out;
3206 		}
3207 
3208 		/*
3209 		 * XXX: The caller can not trust "path" any more after
3210 		 * this as it will have been deleted. What do we do?
3211 		 *
3212 		 * In theory the rotate-for-merge code will never get
3213 		 * here because it'll always ask for a rotate in a
3214 		 * nonempty list.
3215 		 */
3216 
3217 		ret = ocfs2_remove_rightmost_path(handle, et, path,
3218 						  dealloc);
3219 		if (ret)
3220 			mlog_errno(ret);
3221 		goto out;
3222 	}
3223 
3224 	/*
3225 	 * Now we can loop, remembering the path we get from -EAGAIN
3226 	 * and restarting from there.
3227 	 */
3228 try_rotate:
3229 	ret = __ocfs2_rotate_tree_left(handle, et, orig_credits, path,
3230 				       dealloc, &restart_path);
3231 	if (ret && ret != -EAGAIN) {
3232 		mlog_errno(ret);
3233 		goto out;
3234 	}
3235 
3236 	while (ret == -EAGAIN) {
3237 		tmp_path = restart_path;
3238 		restart_path = NULL;
3239 
3240 		ret = __ocfs2_rotate_tree_left(handle, et, orig_credits,
3241 					       tmp_path, dealloc,
3242 					       &restart_path);
3243 		if (ret && ret != -EAGAIN) {
3244 			mlog_errno(ret);
3245 			goto out;
3246 		}
3247 
3248 		ocfs2_free_path(tmp_path);
3249 		tmp_path = NULL;
3250 
3251 		if (ret == 0)
3252 			goto try_rotate;
3253 	}
3254 
3255 out:
3256 	ocfs2_free_path(tmp_path);
3257 	ocfs2_free_path(restart_path);
3258 	return ret;
3259 }
3260 
3261 static void ocfs2_cleanup_merge(struct ocfs2_extent_list *el,
3262 				int index)
3263 {
3264 	struct ocfs2_extent_rec *rec = &el->l_recs[index];
3265 	unsigned int size;
3266 
3267 	if (rec->e_leaf_clusters == 0) {
3268 		/*
3269 		 * We consumed all of the merged-from record. An empty
3270 		 * extent cannot exist anywhere but the 1st array
3271 		 * position, so move things over if the merged-from
3272 		 * record doesn't occupy that position.
3273 		 *
3274 		 * This creates a new empty extent so the caller
3275 		 * should be smart enough to have removed any existing
3276 		 * ones.
3277 		 */
3278 		if (index > 0) {
3279 			BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
3280 			size = index * sizeof(struct ocfs2_extent_rec);
3281 			memmove(&el->l_recs[1], &el->l_recs[0], size);
3282 		}
3283 
3284 		/*
3285 		 * Always memset - the caller doesn't check whether it
3286 		 * created an empty extent, so there could be junk in
3287 		 * the other fields.
3288 		 */
3289 		memset(&el->l_recs[0], 0, sizeof(struct ocfs2_extent_rec));
3290 	}
3291 }
3292 
3293 static int ocfs2_get_right_path(struct ocfs2_extent_tree *et,
3294 				struct ocfs2_path *left_path,
3295 				struct ocfs2_path **ret_right_path)
3296 {
3297 	int ret;
3298 	u32 right_cpos;
3299 	struct ocfs2_path *right_path = NULL;
3300 	struct ocfs2_extent_list *left_el;
3301 
3302 	*ret_right_path = NULL;
3303 
3304 	/* This function shouldn't be called for non-trees. */
3305 	BUG_ON(left_path->p_tree_depth == 0);
3306 
3307 	left_el = path_leaf_el(left_path);
3308 	BUG_ON(left_el->l_next_free_rec != left_el->l_count);
3309 
3310 	ret = ocfs2_find_cpos_for_right_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3311 					     left_path, &right_cpos);
3312 	if (ret) {
3313 		mlog_errno(ret);
3314 		goto out;
3315 	}
3316 
3317 	/* This function shouldn't be called for the rightmost leaf. */
3318 	BUG_ON(right_cpos == 0);
3319 
3320 	right_path = ocfs2_new_path_from_path(left_path);
3321 	if (!right_path) {
3322 		ret = -ENOMEM;
3323 		mlog_errno(ret);
3324 		goto out;
3325 	}
3326 
3327 	ret = ocfs2_find_path(et->et_ci, right_path, right_cpos);
3328 	if (ret) {
3329 		mlog_errno(ret);
3330 		goto out;
3331 	}
3332 
3333 	*ret_right_path = right_path;
3334 out:
3335 	if (ret)
3336 		ocfs2_free_path(right_path);
3337 	return ret;
3338 }
3339 
3340 /*
3341  * Remove split_rec clusters from the record at index and merge them
3342  * onto the beginning of the record "next" to it.
3343  * For index < l_count - 1, the next means the extent rec at index + 1.
3344  * For index == l_count - 1, the "next" means the 1st extent rec of the
3345  * next extent block.
3346  */
3347 static int ocfs2_merge_rec_right(struct ocfs2_path *left_path,
3348 				 handle_t *handle,
3349 				 struct ocfs2_extent_tree *et,
3350 				 struct ocfs2_extent_rec *split_rec,
3351 				 int index)
3352 {
3353 	int ret, next_free, i;
3354 	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3355 	struct ocfs2_extent_rec *left_rec;
3356 	struct ocfs2_extent_rec *right_rec;
3357 	struct ocfs2_extent_list *right_el;
3358 	struct ocfs2_path *right_path = NULL;
3359 	int subtree_index = 0;
3360 	struct ocfs2_extent_list *el = path_leaf_el(left_path);
3361 	struct buffer_head *bh = path_leaf_bh(left_path);
3362 	struct buffer_head *root_bh = NULL;
3363 
3364 	BUG_ON(index >= le16_to_cpu(el->l_next_free_rec));
3365 	left_rec = &el->l_recs[index];
3366 
3367 	if (index == le16_to_cpu(el->l_next_free_rec) - 1 &&
3368 	    le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count)) {
3369 		/* we meet with a cross extent block merge. */
3370 		ret = ocfs2_get_right_path(et, left_path, &right_path);
3371 		if (ret) {
3372 			mlog_errno(ret);
3373 			goto out;
3374 		}
3375 
3376 		right_el = path_leaf_el(right_path);
3377 		next_free = le16_to_cpu(right_el->l_next_free_rec);
3378 		BUG_ON(next_free <= 0);
3379 		right_rec = &right_el->l_recs[0];
3380 		if (ocfs2_is_empty_extent(right_rec)) {
3381 			BUG_ON(next_free <= 1);
3382 			right_rec = &right_el->l_recs[1];
3383 		}
3384 
3385 		BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3386 		       le16_to_cpu(left_rec->e_leaf_clusters) !=
3387 		       le32_to_cpu(right_rec->e_cpos));
3388 
3389 		subtree_index = ocfs2_find_subtree_root(et, left_path,
3390 							right_path);
3391 
3392 		ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3393 						      handle->h_buffer_credits,
3394 						      right_path);
3395 		if (ret) {
3396 			mlog_errno(ret);
3397 			goto out;
3398 		}
3399 
3400 		root_bh = left_path->p_node[subtree_index].bh;
3401 		BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3402 
3403 		ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3404 						   subtree_index);
3405 		if (ret) {
3406 			mlog_errno(ret);
3407 			goto out;
3408 		}
3409 
3410 		for (i = subtree_index + 1;
3411 		     i < path_num_items(right_path); i++) {
3412 			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3413 							   right_path, i);
3414 			if (ret) {
3415 				mlog_errno(ret);
3416 				goto out;
3417 			}
3418 
3419 			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3420 							   left_path, i);
3421 			if (ret) {
3422 				mlog_errno(ret);
3423 				goto out;
3424 			}
3425 		}
3426 
3427 	} else {
3428 		BUG_ON(index == le16_to_cpu(el->l_next_free_rec) - 1);
3429 		right_rec = &el->l_recs[index + 1];
3430 	}
3431 
3432 	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, left_path,
3433 					   path_num_items(left_path) - 1);
3434 	if (ret) {
3435 		mlog_errno(ret);
3436 		goto out;
3437 	}
3438 
3439 	le16_add_cpu(&left_rec->e_leaf_clusters, -split_clusters);
3440 
3441 	le32_add_cpu(&right_rec->e_cpos, -split_clusters);
3442 	le64_add_cpu(&right_rec->e_blkno,
3443 		     -ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3444 					       split_clusters));
3445 	le16_add_cpu(&right_rec->e_leaf_clusters, split_clusters);
3446 
3447 	ocfs2_cleanup_merge(el, index);
3448 
3449 	ocfs2_journal_dirty(handle, bh);
3450 	if (right_path) {
3451 		ocfs2_journal_dirty(handle, path_leaf_bh(right_path));
3452 		ocfs2_complete_edge_insert(handle, left_path, right_path,
3453 					   subtree_index);
3454 	}
3455 out:
3456 	if (right_path)
3457 		ocfs2_free_path(right_path);
3458 	return ret;
3459 }
3460 
3461 static int ocfs2_get_left_path(struct ocfs2_extent_tree *et,
3462 			       struct ocfs2_path *right_path,
3463 			       struct ocfs2_path **ret_left_path)
3464 {
3465 	int ret;
3466 	u32 left_cpos;
3467 	struct ocfs2_path *left_path = NULL;
3468 
3469 	*ret_left_path = NULL;
3470 
3471 	/* This function shouldn't be called for non-trees. */
3472 	BUG_ON(right_path->p_tree_depth == 0);
3473 
3474 	ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3475 					    right_path, &left_cpos);
3476 	if (ret) {
3477 		mlog_errno(ret);
3478 		goto out;
3479 	}
3480 
3481 	/* This function shouldn't be called for the leftmost leaf. */
3482 	BUG_ON(left_cpos == 0);
3483 
3484 	left_path = ocfs2_new_path_from_path(right_path);
3485 	if (!left_path) {
3486 		ret = -ENOMEM;
3487 		mlog_errno(ret);
3488 		goto out;
3489 	}
3490 
3491 	ret = ocfs2_find_path(et->et_ci, left_path, left_cpos);
3492 	if (ret) {
3493 		mlog_errno(ret);
3494 		goto out;
3495 	}
3496 
3497 	*ret_left_path = left_path;
3498 out:
3499 	if (ret)
3500 		ocfs2_free_path(left_path);
3501 	return ret;
3502 }
3503 
3504 /*
3505  * Remove split_rec clusters from the record at index and merge them
3506  * onto the tail of the record "before" it.
3507  * For index > 0, the "before" means the extent rec at index - 1.
3508  *
3509  * For index == 0, the "before" means the last record of the previous
3510  * extent block. And there is also a situation that we may need to
3511  * remove the rightmost leaf extent block in the right_path and change
3512  * the right path to indicate the new rightmost path.
3513  */
3514 static int ocfs2_merge_rec_left(struct ocfs2_path *right_path,
3515 				handle_t *handle,
3516 				struct ocfs2_extent_tree *et,
3517 				struct ocfs2_extent_rec *split_rec,
3518 				struct ocfs2_cached_dealloc_ctxt *dealloc,
3519 				int index)
3520 {
3521 	int ret, i, subtree_index = 0, has_empty_extent = 0;
3522 	unsigned int split_clusters = le16_to_cpu(split_rec->e_leaf_clusters);
3523 	struct ocfs2_extent_rec *left_rec;
3524 	struct ocfs2_extent_rec *right_rec;
3525 	struct ocfs2_extent_list *el = path_leaf_el(right_path);
3526 	struct buffer_head *bh = path_leaf_bh(right_path);
3527 	struct buffer_head *root_bh = NULL;
3528 	struct ocfs2_path *left_path = NULL;
3529 	struct ocfs2_extent_list *left_el;
3530 
3531 	BUG_ON(index < 0);
3532 
3533 	right_rec = &el->l_recs[index];
3534 	if (index == 0) {
3535 		/* we meet with a cross extent block merge. */
3536 		ret = ocfs2_get_left_path(et, right_path, &left_path);
3537 		if (ret) {
3538 			mlog_errno(ret);
3539 			goto out;
3540 		}
3541 
3542 		left_el = path_leaf_el(left_path);
3543 		BUG_ON(le16_to_cpu(left_el->l_next_free_rec) !=
3544 		       le16_to_cpu(left_el->l_count));
3545 
3546 		left_rec = &left_el->l_recs[
3547 				le16_to_cpu(left_el->l_next_free_rec) - 1];
3548 		BUG_ON(le32_to_cpu(left_rec->e_cpos) +
3549 		       le16_to_cpu(left_rec->e_leaf_clusters) !=
3550 		       le32_to_cpu(split_rec->e_cpos));
3551 
3552 		subtree_index = ocfs2_find_subtree_root(et, left_path,
3553 							right_path);
3554 
3555 		ret = ocfs2_extend_rotate_transaction(handle, subtree_index,
3556 						      handle->h_buffer_credits,
3557 						      left_path);
3558 		if (ret) {
3559 			mlog_errno(ret);
3560 			goto out;
3561 		}
3562 
3563 		root_bh = left_path->p_node[subtree_index].bh;
3564 		BUG_ON(root_bh != right_path->p_node[subtree_index].bh);
3565 
3566 		ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3567 						   subtree_index);
3568 		if (ret) {
3569 			mlog_errno(ret);
3570 			goto out;
3571 		}
3572 
3573 		for (i = subtree_index + 1;
3574 		     i < path_num_items(right_path); i++) {
3575 			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3576 							   right_path, i);
3577 			if (ret) {
3578 				mlog_errno(ret);
3579 				goto out;
3580 			}
3581 
3582 			ret = ocfs2_path_bh_journal_access(handle, et->et_ci,
3583 							   left_path, i);
3584 			if (ret) {
3585 				mlog_errno(ret);
3586 				goto out;
3587 			}
3588 		}
3589 	} else {
3590 		left_rec = &el->l_recs[index - 1];
3591 		if (ocfs2_is_empty_extent(&el->l_recs[0]))
3592 			has_empty_extent = 1;
3593 	}
3594 
3595 	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, right_path,
3596 					   path_num_items(right_path) - 1);
3597 	if (ret) {
3598 		mlog_errno(ret);
3599 		goto out;
3600 	}
3601 
3602 	if (has_empty_extent && index == 1) {
3603 		/*
3604 		 * The easy case - we can just plop the record right in.
3605 		 */
3606 		*left_rec = *split_rec;
3607 
3608 		has_empty_extent = 0;
3609 	} else
3610 		le16_add_cpu(&left_rec->e_leaf_clusters, split_clusters);
3611 
3612 	le32_add_cpu(&right_rec->e_cpos, split_clusters);
3613 	le64_add_cpu(&right_rec->e_blkno,
3614 		     ocfs2_clusters_to_blocks(ocfs2_metadata_cache_get_super(et->et_ci),
3615 					      split_clusters));
3616 	le16_add_cpu(&right_rec->e_leaf_clusters, -split_clusters);
3617 
3618 	ocfs2_cleanup_merge(el, index);
3619 
3620 	ocfs2_journal_dirty(handle, bh);
3621 	if (left_path) {
3622 		ocfs2_journal_dirty(handle, path_leaf_bh(left_path));
3623 
3624 		/*
3625 		 * In the situation that the right_rec is empty and the extent
3626 		 * block is empty also,  ocfs2_complete_edge_insert can't handle
3627 		 * it and we need to delete the right extent block.
3628 		 */
3629 		if (le16_to_cpu(right_rec->e_leaf_clusters) == 0 &&
3630 		    le16_to_cpu(el->l_next_free_rec) == 1) {
3631 
3632 			ret = ocfs2_remove_rightmost_path(handle, et,
3633 							  right_path,
3634 							  dealloc);
3635 			if (ret) {
3636 				mlog_errno(ret);
3637 				goto out;
3638 			}
3639 
3640 			/* Now the rightmost extent block has been deleted.
3641 			 * So we use the new rightmost path.
3642 			 */
3643 			ocfs2_mv_path(right_path, left_path);
3644 			left_path = NULL;
3645 		} else
3646 			ocfs2_complete_edge_insert(handle, left_path,
3647 						   right_path, subtree_index);
3648 	}
3649 out:
3650 	if (left_path)
3651 		ocfs2_free_path(left_path);
3652 	return ret;
3653 }
3654 
3655 static int ocfs2_try_to_merge_extent(handle_t *handle,
3656 				     struct ocfs2_extent_tree *et,
3657 				     struct ocfs2_path *path,
3658 				     int split_index,
3659 				     struct ocfs2_extent_rec *split_rec,
3660 				     struct ocfs2_cached_dealloc_ctxt *dealloc,
3661 				     struct ocfs2_merge_ctxt *ctxt)
3662 {
3663 	int ret = 0;
3664 	struct ocfs2_extent_list *el = path_leaf_el(path);
3665 	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
3666 
3667 	BUG_ON(ctxt->c_contig_type == CONTIG_NONE);
3668 
3669 	if (ctxt->c_split_covers_rec && ctxt->c_has_empty_extent) {
3670 		/*
3671 		 * The merge code will need to create an empty
3672 		 * extent to take the place of the newly
3673 		 * emptied slot. Remove any pre-existing empty
3674 		 * extents - having more than one in a leaf is
3675 		 * illegal.
3676 		 */
3677 		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
3678 		if (ret) {
3679 			mlog_errno(ret);
3680 			goto out;
3681 		}
3682 		split_index--;
3683 		rec = &el->l_recs[split_index];
3684 	}
3685 
3686 	if (ctxt->c_contig_type == CONTIG_LEFTRIGHT) {
3687 		/*
3688 		 * Left-right contig implies this.
3689 		 */
3690 		BUG_ON(!ctxt->c_split_covers_rec);
3691 
3692 		/*
3693 		 * Since the leftright insert always covers the entire
3694 		 * extent, this call will delete the insert record
3695 		 * entirely, resulting in an empty extent record added to
3696 		 * the extent block.
3697 		 *
3698 		 * Since the adding of an empty extent shifts
3699 		 * everything back to the right, there's no need to
3700 		 * update split_index here.
3701 		 *
3702 		 * When the split_index is zero, we need to merge it to the
3703 		 * prevoius extent block. It is more efficient and easier
3704 		 * if we do merge_right first and merge_left later.
3705 		 */
3706 		ret = ocfs2_merge_rec_right(path, handle, et, split_rec,
3707 					    split_index);
3708 		if (ret) {
3709 			mlog_errno(ret);
3710 			goto out;
3711 		}
3712 
3713 		/*
3714 		 * We can only get this from logic error above.
3715 		 */
3716 		BUG_ON(!ocfs2_is_empty_extent(&el->l_recs[0]));
3717 
3718 		/* The merge left us with an empty extent, remove it. */
3719 		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
3720 		if (ret) {
3721 			mlog_errno(ret);
3722 			goto out;
3723 		}
3724 
3725 		rec = &el->l_recs[split_index];
3726 
3727 		/*
3728 		 * Note that we don't pass split_rec here on purpose -
3729 		 * we've merged it into the rec already.
3730 		 */
3731 		ret = ocfs2_merge_rec_left(path, handle, et, rec,
3732 					   dealloc, split_index);
3733 
3734 		if (ret) {
3735 			mlog_errno(ret);
3736 			goto out;
3737 		}
3738 
3739 		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
3740 		/*
3741 		 * Error from this last rotate is not critical, so
3742 		 * print but don't bubble it up.
3743 		 */
3744 		if (ret)
3745 			mlog_errno(ret);
3746 		ret = 0;
3747 	} else {
3748 		/*
3749 		 * Merge a record to the left or right.
3750 		 *
3751 		 * 'contig_type' is relative to the existing record,
3752 		 * so for example, if we're "right contig", it's to
3753 		 * the record on the left (hence the left merge).
3754 		 */
3755 		if (ctxt->c_contig_type == CONTIG_RIGHT) {
3756 			ret = ocfs2_merge_rec_left(path, handle, et,
3757 						   split_rec, dealloc,
3758 						   split_index);
3759 			if (ret) {
3760 				mlog_errno(ret);
3761 				goto out;
3762 			}
3763 		} else {
3764 			ret = ocfs2_merge_rec_right(path, handle,
3765 						    et, split_rec,
3766 						    split_index);
3767 			if (ret) {
3768 				mlog_errno(ret);
3769 				goto out;
3770 			}
3771 		}
3772 
3773 		if (ctxt->c_split_covers_rec) {
3774 			/*
3775 			 * The merge may have left an empty extent in
3776 			 * our leaf. Try to rotate it away.
3777 			 */
3778 			ret = ocfs2_rotate_tree_left(handle, et, path,
3779 						     dealloc);
3780 			if (ret)
3781 				mlog_errno(ret);
3782 			ret = 0;
3783 		}
3784 	}
3785 
3786 out:
3787 	return ret;
3788 }
3789 
3790 static void ocfs2_subtract_from_rec(struct super_block *sb,
3791 				    enum ocfs2_split_type split,
3792 				    struct ocfs2_extent_rec *rec,
3793 				    struct ocfs2_extent_rec *split_rec)
3794 {
3795 	u64 len_blocks;
3796 
3797 	len_blocks = ocfs2_clusters_to_blocks(sb,
3798 				le16_to_cpu(split_rec->e_leaf_clusters));
3799 
3800 	if (split == SPLIT_LEFT) {
3801 		/*
3802 		 * Region is on the left edge of the existing
3803 		 * record.
3804 		 */
3805 		le32_add_cpu(&rec->e_cpos,
3806 			     le16_to_cpu(split_rec->e_leaf_clusters));
3807 		le64_add_cpu(&rec->e_blkno, len_blocks);
3808 		le16_add_cpu(&rec->e_leaf_clusters,
3809 			     -le16_to_cpu(split_rec->e_leaf_clusters));
3810 	} else {
3811 		/*
3812 		 * Region is on the right edge of the existing
3813 		 * record.
3814 		 */
3815 		le16_add_cpu(&rec->e_leaf_clusters,
3816 			     -le16_to_cpu(split_rec->e_leaf_clusters));
3817 	}
3818 }
3819 
3820 /*
3821  * Do the final bits of extent record insertion at the target leaf
3822  * list. If this leaf is part of an allocation tree, it is assumed
3823  * that the tree above has been prepared.
3824  */
3825 static void ocfs2_insert_at_leaf(struct ocfs2_extent_tree *et,
3826 				 struct ocfs2_extent_rec *insert_rec,
3827 				 struct ocfs2_extent_list *el,
3828 				 struct ocfs2_insert_type *insert)
3829 {
3830 	int i = insert->ins_contig_index;
3831 	unsigned int range;
3832 	struct ocfs2_extent_rec *rec;
3833 
3834 	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
3835 
3836 	if (insert->ins_split != SPLIT_NONE) {
3837 		i = ocfs2_search_extent_list(el, le32_to_cpu(insert_rec->e_cpos));
3838 		BUG_ON(i == -1);
3839 		rec = &el->l_recs[i];
3840 		ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
3841 					insert->ins_split, rec,
3842 					insert_rec);
3843 		goto rotate;
3844 	}
3845 
3846 	/*
3847 	 * Contiguous insert - either left or right.
3848 	 */
3849 	if (insert->ins_contig != CONTIG_NONE) {
3850 		rec = &el->l_recs[i];
3851 		if (insert->ins_contig == CONTIG_LEFT) {
3852 			rec->e_blkno = insert_rec->e_blkno;
3853 			rec->e_cpos = insert_rec->e_cpos;
3854 		}
3855 		le16_add_cpu(&rec->e_leaf_clusters,
3856 			     le16_to_cpu(insert_rec->e_leaf_clusters));
3857 		return;
3858 	}
3859 
3860 	/*
3861 	 * Handle insert into an empty leaf.
3862 	 */
3863 	if (le16_to_cpu(el->l_next_free_rec) == 0 ||
3864 	    ((le16_to_cpu(el->l_next_free_rec) == 1) &&
3865 	     ocfs2_is_empty_extent(&el->l_recs[0]))) {
3866 		el->l_recs[0] = *insert_rec;
3867 		el->l_next_free_rec = cpu_to_le16(1);
3868 		return;
3869 	}
3870 
3871 	/*
3872 	 * Appending insert.
3873 	 */
3874 	if (insert->ins_appending == APPEND_TAIL) {
3875 		i = le16_to_cpu(el->l_next_free_rec) - 1;
3876 		rec = &el->l_recs[i];
3877 		range = le32_to_cpu(rec->e_cpos)
3878 			+ le16_to_cpu(rec->e_leaf_clusters);
3879 		BUG_ON(le32_to_cpu(insert_rec->e_cpos) < range);
3880 
3881 		mlog_bug_on_msg(le16_to_cpu(el->l_next_free_rec) >=
3882 				le16_to_cpu(el->l_count),
3883 				"owner %llu, depth %u, count %u, next free %u, "
3884 				"rec.cpos %u, rec.clusters %u, "
3885 				"insert.cpos %u, insert.clusters %u\n",
3886 				ocfs2_metadata_cache_owner(et->et_ci),
3887 				le16_to_cpu(el->l_tree_depth),
3888 				le16_to_cpu(el->l_count),
3889 				le16_to_cpu(el->l_next_free_rec),
3890 				le32_to_cpu(el->l_recs[i].e_cpos),
3891 				le16_to_cpu(el->l_recs[i].e_leaf_clusters),
3892 				le32_to_cpu(insert_rec->e_cpos),
3893 				le16_to_cpu(insert_rec->e_leaf_clusters));
3894 		i++;
3895 		el->l_recs[i] = *insert_rec;
3896 		le16_add_cpu(&el->l_next_free_rec, 1);
3897 		return;
3898 	}
3899 
3900 rotate:
3901 	/*
3902 	 * Ok, we have to rotate.
3903 	 *
3904 	 * At this point, it is safe to assume that inserting into an
3905 	 * empty leaf and appending to a leaf have both been handled
3906 	 * above.
3907 	 *
3908 	 * This leaf needs to have space, either by the empty 1st
3909 	 * extent record, or by virtue of an l_next_rec < l_count.
3910 	 */
3911 	ocfs2_rotate_leaf(el, insert_rec);
3912 }
3913 
3914 static void ocfs2_adjust_rightmost_records(handle_t *handle,
3915 					   struct ocfs2_extent_tree *et,
3916 					   struct ocfs2_path *path,
3917 					   struct ocfs2_extent_rec *insert_rec)
3918 {
3919 	int ret, i, next_free;
3920 	struct buffer_head *bh;
3921 	struct ocfs2_extent_list *el;
3922 	struct ocfs2_extent_rec *rec;
3923 
3924 	/*
3925 	 * Update everything except the leaf block.
3926 	 */
3927 	for (i = 0; i < path->p_tree_depth; i++) {
3928 		bh = path->p_node[i].bh;
3929 		el = path->p_node[i].el;
3930 
3931 		next_free = le16_to_cpu(el->l_next_free_rec);
3932 		if (next_free == 0) {
3933 			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
3934 				    "Owner %llu has a bad extent list",
3935 				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci));
3936 			ret = -EIO;
3937 			return;
3938 		}
3939 
3940 		rec = &el->l_recs[next_free - 1];
3941 
3942 		rec->e_int_clusters = insert_rec->e_cpos;
3943 		le32_add_cpu(&rec->e_int_clusters,
3944 			     le16_to_cpu(insert_rec->e_leaf_clusters));
3945 		le32_add_cpu(&rec->e_int_clusters,
3946 			     -le32_to_cpu(rec->e_cpos));
3947 
3948 		ocfs2_journal_dirty(handle, bh);
3949 	}
3950 }
3951 
3952 static int ocfs2_append_rec_to_path(handle_t *handle,
3953 				    struct ocfs2_extent_tree *et,
3954 				    struct ocfs2_extent_rec *insert_rec,
3955 				    struct ocfs2_path *right_path,
3956 				    struct ocfs2_path **ret_left_path)
3957 {
3958 	int ret, next_free;
3959 	struct ocfs2_extent_list *el;
3960 	struct ocfs2_path *left_path = NULL;
3961 
3962 	*ret_left_path = NULL;
3963 
3964 	/*
3965 	 * This shouldn't happen for non-trees. The extent rec cluster
3966 	 * count manipulation below only works for interior nodes.
3967 	 */
3968 	BUG_ON(right_path->p_tree_depth == 0);
3969 
3970 	/*
3971 	 * If our appending insert is at the leftmost edge of a leaf,
3972 	 * then we might need to update the rightmost records of the
3973 	 * neighboring path.
3974 	 */
3975 	el = path_leaf_el(right_path);
3976 	next_free = le16_to_cpu(el->l_next_free_rec);
3977 	if (next_free == 0 ||
3978 	    (next_free == 1 && ocfs2_is_empty_extent(&el->l_recs[0]))) {
3979 		u32 left_cpos;
3980 
3981 		ret = ocfs2_find_cpos_for_left_leaf(ocfs2_metadata_cache_get_super(et->et_ci),
3982 						    right_path, &left_cpos);
3983 		if (ret) {
3984 			mlog_errno(ret);
3985 			goto out;
3986 		}
3987 
3988 		trace_ocfs2_append_rec_to_path(
3989 			(unsigned long long)
3990 			ocfs2_metadata_cache_owner(et->et_ci),
3991 			le32_to_cpu(insert_rec->e_cpos),
3992 			left_cpos);
3993 
3994 		/*
3995 		 * No need to worry if the append is already in the
3996 		 * leftmost leaf.
3997 		 */
3998 		if (left_cpos) {
3999 			left_path = ocfs2_new_path_from_path(right_path);
4000 			if (!left_path) {
4001 				ret = -ENOMEM;
4002 				mlog_errno(ret);
4003 				goto out;
4004 			}
4005 
4006 			ret = ocfs2_find_path(et->et_ci, left_path,
4007 					      left_cpos);
4008 			if (ret) {
4009 				mlog_errno(ret);
4010 				goto out;
4011 			}
4012 
4013 			/*
4014 			 * ocfs2_insert_path() will pass the left_path to the
4015 			 * journal for us.
4016 			 */
4017 		}
4018 	}
4019 
4020 	ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
4021 	if (ret) {
4022 		mlog_errno(ret);
4023 		goto out;
4024 	}
4025 
4026 	ocfs2_adjust_rightmost_records(handle, et, right_path, insert_rec);
4027 
4028 	*ret_left_path = left_path;
4029 	ret = 0;
4030 out:
4031 	if (ret != 0)
4032 		ocfs2_free_path(left_path);
4033 
4034 	return ret;
4035 }
4036 
4037 static void ocfs2_split_record(struct ocfs2_extent_tree *et,
4038 			       struct ocfs2_path *left_path,
4039 			       struct ocfs2_path *right_path,
4040 			       struct ocfs2_extent_rec *split_rec,
4041 			       enum ocfs2_split_type split)
4042 {
4043 	int index;
4044 	u32 cpos = le32_to_cpu(split_rec->e_cpos);
4045 	struct ocfs2_extent_list *left_el = NULL, *right_el, *insert_el, *el;
4046 	struct ocfs2_extent_rec *rec, *tmprec;
4047 
4048 	right_el = path_leaf_el(right_path);
4049 	if (left_path)
4050 		left_el = path_leaf_el(left_path);
4051 
4052 	el = right_el;
4053 	insert_el = right_el;
4054 	index = ocfs2_search_extent_list(el, cpos);
4055 	if (index != -1) {
4056 		if (index == 0 && left_path) {
4057 			BUG_ON(ocfs2_is_empty_extent(&el->l_recs[0]));
4058 
4059 			/*
4060 			 * This typically means that the record
4061 			 * started in the left path but moved to the
4062 			 * right as a result of rotation. We either
4063 			 * move the existing record to the left, or we
4064 			 * do the later insert there.
4065 			 *
4066 			 * In this case, the left path should always
4067 			 * exist as the rotate code will have passed
4068 			 * it back for a post-insert update.
4069 			 */
4070 
4071 			if (split == SPLIT_LEFT) {
4072 				/*
4073 				 * It's a left split. Since we know
4074 				 * that the rotate code gave us an
4075 				 * empty extent in the left path, we
4076 				 * can just do the insert there.
4077 				 */
4078 				insert_el = left_el;
4079 			} else {
4080 				/*
4081 				 * Right split - we have to move the
4082 				 * existing record over to the left
4083 				 * leaf. The insert will be into the
4084 				 * newly created empty extent in the
4085 				 * right leaf.
4086 				 */
4087 				tmprec = &right_el->l_recs[index];
4088 				ocfs2_rotate_leaf(left_el, tmprec);
4089 				el = left_el;
4090 
4091 				memset(tmprec, 0, sizeof(*tmprec));
4092 				index = ocfs2_search_extent_list(left_el, cpos);
4093 				BUG_ON(index == -1);
4094 			}
4095 		}
4096 	} else {
4097 		BUG_ON(!left_path);
4098 		BUG_ON(!ocfs2_is_empty_extent(&left_el->l_recs[0]));
4099 		/*
4100 		 * Left path is easy - we can just allow the insert to
4101 		 * happen.
4102 		 */
4103 		el = left_el;
4104 		insert_el = left_el;
4105 		index = ocfs2_search_extent_list(el, cpos);
4106 		BUG_ON(index == -1);
4107 	}
4108 
4109 	rec = &el->l_recs[index];
4110 	ocfs2_subtract_from_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4111 				split, rec, split_rec);
4112 	ocfs2_rotate_leaf(insert_el, split_rec);
4113 }
4114 
4115 /*
4116  * This function only does inserts on an allocation b-tree. For tree
4117  * depth = 0, ocfs2_insert_at_leaf() is called directly.
4118  *
4119  * right_path is the path we want to do the actual insert
4120  * in. left_path should only be passed in if we need to update that
4121  * portion of the tree after an edge insert.
4122  */
4123 static int ocfs2_insert_path(handle_t *handle,
4124 			     struct ocfs2_extent_tree *et,
4125 			     struct ocfs2_path *left_path,
4126 			     struct ocfs2_path *right_path,
4127 			     struct ocfs2_extent_rec *insert_rec,
4128 			     struct ocfs2_insert_type *insert)
4129 {
4130 	int ret, subtree_index;
4131 	struct buffer_head *leaf_bh = path_leaf_bh(right_path);
4132 
4133 	if (left_path) {
4134 		/*
4135 		 * There's a chance that left_path got passed back to
4136 		 * us without being accounted for in the
4137 		 * journal. Extend our transaction here to be sure we
4138 		 * can change those blocks.
4139 		 */
4140 		ret = ocfs2_extend_trans(handle, left_path->p_tree_depth);
4141 		if (ret < 0) {
4142 			mlog_errno(ret);
4143 			goto out;
4144 		}
4145 
4146 		ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
4147 		if (ret < 0) {
4148 			mlog_errno(ret);
4149 			goto out;
4150 		}
4151 	}
4152 
4153 	/*
4154 	 * Pass both paths to the journal. The majority of inserts
4155 	 * will be touching all components anyway.
4156 	 */
4157 	ret = ocfs2_journal_access_path(et->et_ci, handle, right_path);
4158 	if (ret < 0) {
4159 		mlog_errno(ret);
4160 		goto out;
4161 	}
4162 
4163 	if (insert->ins_split != SPLIT_NONE) {
4164 		/*
4165 		 * We could call ocfs2_insert_at_leaf() for some types
4166 		 * of splits, but it's easier to just let one separate
4167 		 * function sort it all out.
4168 		 */
4169 		ocfs2_split_record(et, left_path, right_path,
4170 				   insert_rec, insert->ins_split);
4171 
4172 		/*
4173 		 * Split might have modified either leaf and we don't
4174 		 * have a guarantee that the later edge insert will
4175 		 * dirty this for us.
4176 		 */
4177 		if (left_path)
4178 			ocfs2_journal_dirty(handle,
4179 					    path_leaf_bh(left_path));
4180 	} else
4181 		ocfs2_insert_at_leaf(et, insert_rec, path_leaf_el(right_path),
4182 				     insert);
4183 
4184 	ocfs2_journal_dirty(handle, leaf_bh);
4185 
4186 	if (left_path) {
4187 		/*
4188 		 * The rotate code has indicated that we need to fix
4189 		 * up portions of the tree after the insert.
4190 		 *
4191 		 * XXX: Should we extend the transaction here?
4192 		 */
4193 		subtree_index = ocfs2_find_subtree_root(et, left_path,
4194 							right_path);
4195 		ocfs2_complete_edge_insert(handle, left_path, right_path,
4196 					   subtree_index);
4197 	}
4198 
4199 	ret = 0;
4200 out:
4201 	return ret;
4202 }
4203 
4204 static int ocfs2_do_insert_extent(handle_t *handle,
4205 				  struct ocfs2_extent_tree *et,
4206 				  struct ocfs2_extent_rec *insert_rec,
4207 				  struct ocfs2_insert_type *type)
4208 {
4209 	int ret, rotate = 0;
4210 	u32 cpos;
4211 	struct ocfs2_path *right_path = NULL;
4212 	struct ocfs2_path *left_path = NULL;
4213 	struct ocfs2_extent_list *el;
4214 
4215 	el = et->et_root_el;
4216 
4217 	ret = ocfs2_et_root_journal_access(handle, et,
4218 					   OCFS2_JOURNAL_ACCESS_WRITE);
4219 	if (ret) {
4220 		mlog_errno(ret);
4221 		goto out;
4222 	}
4223 
4224 	if (le16_to_cpu(el->l_tree_depth) == 0) {
4225 		ocfs2_insert_at_leaf(et, insert_rec, el, type);
4226 		goto out_update_clusters;
4227 	}
4228 
4229 	right_path = ocfs2_new_path_from_et(et);
4230 	if (!right_path) {
4231 		ret = -ENOMEM;
4232 		mlog_errno(ret);
4233 		goto out;
4234 	}
4235 
4236 	/*
4237 	 * Determine the path to start with. Rotations need the
4238 	 * rightmost path, everything else can go directly to the
4239 	 * target leaf.
4240 	 */
4241 	cpos = le32_to_cpu(insert_rec->e_cpos);
4242 	if (type->ins_appending == APPEND_NONE &&
4243 	    type->ins_contig == CONTIG_NONE) {
4244 		rotate = 1;
4245 		cpos = UINT_MAX;
4246 	}
4247 
4248 	ret = ocfs2_find_path(et->et_ci, right_path, cpos);
4249 	if (ret) {
4250 		mlog_errno(ret);
4251 		goto out;
4252 	}
4253 
4254 	/*
4255 	 * Rotations and appends need special treatment - they modify
4256 	 * parts of the tree's above them.
4257 	 *
4258 	 * Both might pass back a path immediate to the left of the
4259 	 * one being inserted to. This will be cause
4260 	 * ocfs2_insert_path() to modify the rightmost records of
4261 	 * left_path to account for an edge insert.
4262 	 *
4263 	 * XXX: When modifying this code, keep in mind that an insert
4264 	 * can wind up skipping both of these two special cases...
4265 	 */
4266 	if (rotate) {
4267 		ret = ocfs2_rotate_tree_right(handle, et, type->ins_split,
4268 					      le32_to_cpu(insert_rec->e_cpos),
4269 					      right_path, &left_path);
4270 		if (ret) {
4271 			mlog_errno(ret);
4272 			goto out;
4273 		}
4274 
4275 		/*
4276 		 * ocfs2_rotate_tree_right() might have extended the
4277 		 * transaction without re-journaling our tree root.
4278 		 */
4279 		ret = ocfs2_et_root_journal_access(handle, et,
4280 						   OCFS2_JOURNAL_ACCESS_WRITE);
4281 		if (ret) {
4282 			mlog_errno(ret);
4283 			goto out;
4284 		}
4285 	} else if (type->ins_appending == APPEND_TAIL
4286 		   && type->ins_contig != CONTIG_LEFT) {
4287 		ret = ocfs2_append_rec_to_path(handle, et, insert_rec,
4288 					       right_path, &left_path);
4289 		if (ret) {
4290 			mlog_errno(ret);
4291 			goto out;
4292 		}
4293 	}
4294 
4295 	ret = ocfs2_insert_path(handle, et, left_path, right_path,
4296 				insert_rec, type);
4297 	if (ret) {
4298 		mlog_errno(ret);
4299 		goto out;
4300 	}
4301 
4302 out_update_clusters:
4303 	if (type->ins_split == SPLIT_NONE)
4304 		ocfs2_et_update_clusters(et,
4305 					 le16_to_cpu(insert_rec->e_leaf_clusters));
4306 
4307 	ocfs2_journal_dirty(handle, et->et_root_bh);
4308 
4309 out:
4310 	ocfs2_free_path(left_path);
4311 	ocfs2_free_path(right_path);
4312 
4313 	return ret;
4314 }
4315 
4316 static enum ocfs2_contig_type
4317 ocfs2_figure_merge_contig_type(struct ocfs2_extent_tree *et,
4318 			       struct ocfs2_path *path,
4319 			       struct ocfs2_extent_list *el, int index,
4320 			       struct ocfs2_extent_rec *split_rec)
4321 {
4322 	int status;
4323 	enum ocfs2_contig_type ret = CONTIG_NONE;
4324 	u32 left_cpos, right_cpos;
4325 	struct ocfs2_extent_rec *rec = NULL;
4326 	struct ocfs2_extent_list *new_el;
4327 	struct ocfs2_path *left_path = NULL, *right_path = NULL;
4328 	struct buffer_head *bh;
4329 	struct ocfs2_extent_block *eb;
4330 	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
4331 
4332 	if (index > 0) {
4333 		rec = &el->l_recs[index - 1];
4334 	} else if (path->p_tree_depth > 0) {
4335 		status = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
4336 		if (status)
4337 			goto out;
4338 
4339 		if (left_cpos != 0) {
4340 			left_path = ocfs2_new_path_from_path(path);
4341 			if (!left_path)
4342 				goto out;
4343 
4344 			status = ocfs2_find_path(et->et_ci, left_path,
4345 						 left_cpos);
4346 			if (status)
4347 				goto out;
4348 
4349 			new_el = path_leaf_el(left_path);
4350 
4351 			if (le16_to_cpu(new_el->l_next_free_rec) !=
4352 			    le16_to_cpu(new_el->l_count)) {
4353 				bh = path_leaf_bh(left_path);
4354 				eb = (struct ocfs2_extent_block *)bh->b_data;
4355 				ocfs2_error(sb,
4356 					    "Extent block #%llu has an "
4357 					    "invalid l_next_free_rec of "
4358 					    "%d.  It should have "
4359 					    "matched the l_count of %d",
4360 					    (unsigned long long)le64_to_cpu(eb->h_blkno),
4361 					    le16_to_cpu(new_el->l_next_free_rec),
4362 					    le16_to_cpu(new_el->l_count));
4363 				status = -EINVAL;
4364 				goto out;
4365 			}
4366 			rec = &new_el->l_recs[
4367 				le16_to_cpu(new_el->l_next_free_rec) - 1];
4368 		}
4369 	}
4370 
4371 	/*
4372 	 * We're careful to check for an empty extent record here -
4373 	 * the merge code will know what to do if it sees one.
4374 	 */
4375 	if (rec) {
4376 		if (index == 1 && ocfs2_is_empty_extent(rec)) {
4377 			if (split_rec->e_cpos == el->l_recs[index].e_cpos)
4378 				ret = CONTIG_RIGHT;
4379 		} else {
4380 			ret = ocfs2_et_extent_contig(et, rec, split_rec);
4381 		}
4382 	}
4383 
4384 	rec = NULL;
4385 	if (index < (le16_to_cpu(el->l_next_free_rec) - 1))
4386 		rec = &el->l_recs[index + 1];
4387 	else if (le16_to_cpu(el->l_next_free_rec) == le16_to_cpu(el->l_count) &&
4388 		 path->p_tree_depth > 0) {
4389 		status = ocfs2_find_cpos_for_right_leaf(sb, path, &right_cpos);
4390 		if (status)
4391 			goto out;
4392 
4393 		if (right_cpos == 0)
4394 			goto out;
4395 
4396 		right_path = ocfs2_new_path_from_path(path);
4397 		if (!right_path)
4398 			goto out;
4399 
4400 		status = ocfs2_find_path(et->et_ci, right_path, right_cpos);
4401 		if (status)
4402 			goto out;
4403 
4404 		new_el = path_leaf_el(right_path);
4405 		rec = &new_el->l_recs[0];
4406 		if (ocfs2_is_empty_extent(rec)) {
4407 			if (le16_to_cpu(new_el->l_next_free_rec) <= 1) {
4408 				bh = path_leaf_bh(right_path);
4409 				eb = (struct ocfs2_extent_block *)bh->b_data;
4410 				ocfs2_error(sb,
4411 					    "Extent block #%llu has an "
4412 					    "invalid l_next_free_rec of %d",
4413 					    (unsigned long long)le64_to_cpu(eb->h_blkno),
4414 					    le16_to_cpu(new_el->l_next_free_rec));
4415 				status = -EINVAL;
4416 				goto out;
4417 			}
4418 			rec = &new_el->l_recs[1];
4419 		}
4420 	}
4421 
4422 	if (rec) {
4423 		enum ocfs2_contig_type contig_type;
4424 
4425 		contig_type = ocfs2_et_extent_contig(et, rec, split_rec);
4426 
4427 		if (contig_type == CONTIG_LEFT && ret == CONTIG_RIGHT)
4428 			ret = CONTIG_LEFTRIGHT;
4429 		else if (ret == CONTIG_NONE)
4430 			ret = contig_type;
4431 	}
4432 
4433 out:
4434 	if (left_path)
4435 		ocfs2_free_path(left_path);
4436 	if (right_path)
4437 		ocfs2_free_path(right_path);
4438 
4439 	return ret;
4440 }
4441 
4442 static void ocfs2_figure_contig_type(struct ocfs2_extent_tree *et,
4443 				     struct ocfs2_insert_type *insert,
4444 				     struct ocfs2_extent_list *el,
4445 				     struct ocfs2_extent_rec *insert_rec)
4446 {
4447 	int i;
4448 	enum ocfs2_contig_type contig_type = CONTIG_NONE;
4449 
4450 	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
4451 
4452 	for(i = 0; i < le16_to_cpu(el->l_next_free_rec); i++) {
4453 		contig_type = ocfs2_et_extent_contig(et, &el->l_recs[i],
4454 						     insert_rec);
4455 		if (contig_type != CONTIG_NONE) {
4456 			insert->ins_contig_index = i;
4457 			break;
4458 		}
4459 	}
4460 	insert->ins_contig = contig_type;
4461 
4462 	if (insert->ins_contig != CONTIG_NONE) {
4463 		struct ocfs2_extent_rec *rec =
4464 				&el->l_recs[insert->ins_contig_index];
4465 		unsigned int len = le16_to_cpu(rec->e_leaf_clusters) +
4466 				   le16_to_cpu(insert_rec->e_leaf_clusters);
4467 
4468 		/*
4469 		 * Caller might want us to limit the size of extents, don't
4470 		 * calculate contiguousness if we might exceed that limit.
4471 		 */
4472 		if (et->et_max_leaf_clusters &&
4473 		    (len > et->et_max_leaf_clusters))
4474 			insert->ins_contig = CONTIG_NONE;
4475 	}
4476 }
4477 
4478 /*
4479  * This should only be called against the righmost leaf extent list.
4480  *
4481  * ocfs2_figure_appending_type() will figure out whether we'll have to
4482  * insert at the tail of the rightmost leaf.
4483  *
4484  * This should also work against the root extent list for tree's with 0
4485  * depth. If we consider the root extent list to be the rightmost leaf node
4486  * then the logic here makes sense.
4487  */
4488 static void ocfs2_figure_appending_type(struct ocfs2_insert_type *insert,
4489 					struct ocfs2_extent_list *el,
4490 					struct ocfs2_extent_rec *insert_rec)
4491 {
4492 	int i;
4493 	u32 cpos = le32_to_cpu(insert_rec->e_cpos);
4494 	struct ocfs2_extent_rec *rec;
4495 
4496 	insert->ins_appending = APPEND_NONE;
4497 
4498 	BUG_ON(le16_to_cpu(el->l_tree_depth) != 0);
4499 
4500 	if (!el->l_next_free_rec)
4501 		goto set_tail_append;
4502 
4503 	if (ocfs2_is_empty_extent(&el->l_recs[0])) {
4504 		/* Were all records empty? */
4505 		if (le16_to_cpu(el->l_next_free_rec) == 1)
4506 			goto set_tail_append;
4507 	}
4508 
4509 	i = le16_to_cpu(el->l_next_free_rec) - 1;
4510 	rec = &el->l_recs[i];
4511 
4512 	if (cpos >=
4513 	    (le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)))
4514 		goto set_tail_append;
4515 
4516 	return;
4517 
4518 set_tail_append:
4519 	insert->ins_appending = APPEND_TAIL;
4520 }
4521 
4522 /*
4523  * Helper function called at the beginning of an insert.
4524  *
4525  * This computes a few things that are commonly used in the process of
4526  * inserting into the btree:
4527  *   - Whether the new extent is contiguous with an existing one.
4528  *   - The current tree depth.
4529  *   - Whether the insert is an appending one.
4530  *   - The total # of free records in the tree.
4531  *
4532  * All of the information is stored on the ocfs2_insert_type
4533  * structure.
4534  */
4535 static int ocfs2_figure_insert_type(struct ocfs2_extent_tree *et,
4536 				    struct buffer_head **last_eb_bh,
4537 				    struct ocfs2_extent_rec *insert_rec,
4538 				    int *free_records,
4539 				    struct ocfs2_insert_type *insert)
4540 {
4541 	int ret;
4542 	struct ocfs2_extent_block *eb;
4543 	struct ocfs2_extent_list *el;
4544 	struct ocfs2_path *path = NULL;
4545 	struct buffer_head *bh = NULL;
4546 
4547 	insert->ins_split = SPLIT_NONE;
4548 
4549 	el = et->et_root_el;
4550 	insert->ins_tree_depth = le16_to_cpu(el->l_tree_depth);
4551 
4552 	if (el->l_tree_depth) {
4553 		/*
4554 		 * If we have tree depth, we read in the
4555 		 * rightmost extent block ahead of time as
4556 		 * ocfs2_figure_insert_type() and ocfs2_add_branch()
4557 		 * may want it later.
4558 		 */
4559 		ret = ocfs2_read_extent_block(et->et_ci,
4560 					      ocfs2_et_get_last_eb_blk(et),
4561 					      &bh);
4562 		if (ret) {
4563 			mlog_errno(ret);
4564 			goto out;
4565 		}
4566 		eb = (struct ocfs2_extent_block *) bh->b_data;
4567 		el = &eb->h_list;
4568 	}
4569 
4570 	/*
4571 	 * Unless we have a contiguous insert, we'll need to know if
4572 	 * there is room left in our allocation tree for another
4573 	 * extent record.
4574 	 *
4575 	 * XXX: This test is simplistic, we can search for empty
4576 	 * extent records too.
4577 	 */
4578 	*free_records = le16_to_cpu(el->l_count) -
4579 		le16_to_cpu(el->l_next_free_rec);
4580 
4581 	if (!insert->ins_tree_depth) {
4582 		ocfs2_figure_contig_type(et, insert, el, insert_rec);
4583 		ocfs2_figure_appending_type(insert, el, insert_rec);
4584 		return 0;
4585 	}
4586 
4587 	path = ocfs2_new_path_from_et(et);
4588 	if (!path) {
4589 		ret = -ENOMEM;
4590 		mlog_errno(ret);
4591 		goto out;
4592 	}
4593 
4594 	/*
4595 	 * In the case that we're inserting past what the tree
4596 	 * currently accounts for, ocfs2_find_path() will return for
4597 	 * us the rightmost tree path. This is accounted for below in
4598 	 * the appending code.
4599 	 */
4600 	ret = ocfs2_find_path(et->et_ci, path, le32_to_cpu(insert_rec->e_cpos));
4601 	if (ret) {
4602 		mlog_errno(ret);
4603 		goto out;
4604 	}
4605 
4606 	el = path_leaf_el(path);
4607 
4608 	/*
4609 	 * Now that we have the path, there's two things we want to determine:
4610 	 * 1) Contiguousness (also set contig_index if this is so)
4611 	 *
4612 	 * 2) Are we doing an append? We can trivially break this up
4613          *     into two types of appends: simple record append, or a
4614          *     rotate inside the tail leaf.
4615 	 */
4616 	ocfs2_figure_contig_type(et, insert, el, insert_rec);
4617 
4618 	/*
4619 	 * The insert code isn't quite ready to deal with all cases of
4620 	 * left contiguousness. Specifically, if it's an insert into
4621 	 * the 1st record in a leaf, it will require the adjustment of
4622 	 * cluster count on the last record of the path directly to it's
4623 	 * left. For now, just catch that case and fool the layers
4624 	 * above us. This works just fine for tree_depth == 0, which
4625 	 * is why we allow that above.
4626 	 */
4627 	if (insert->ins_contig == CONTIG_LEFT &&
4628 	    insert->ins_contig_index == 0)
4629 		insert->ins_contig = CONTIG_NONE;
4630 
4631 	/*
4632 	 * Ok, so we can simply compare against last_eb to figure out
4633 	 * whether the path doesn't exist. This will only happen in
4634 	 * the case that we're doing a tail append, so maybe we can
4635 	 * take advantage of that information somehow.
4636 	 */
4637 	if (ocfs2_et_get_last_eb_blk(et) ==
4638 	    path_leaf_bh(path)->b_blocknr) {
4639 		/*
4640 		 * Ok, ocfs2_find_path() returned us the rightmost
4641 		 * tree path. This might be an appending insert. There are
4642 		 * two cases:
4643 		 *    1) We're doing a true append at the tail:
4644 		 *	-This might even be off the end of the leaf
4645 		 *    2) We're "appending" by rotating in the tail
4646 		 */
4647 		ocfs2_figure_appending_type(insert, el, insert_rec);
4648 	}
4649 
4650 out:
4651 	ocfs2_free_path(path);
4652 
4653 	if (ret == 0)
4654 		*last_eb_bh = bh;
4655 	else
4656 		brelse(bh);
4657 	return ret;
4658 }
4659 
4660 /*
4661  * Insert an extent into a btree.
4662  *
4663  * The caller needs to update the owning btree's cluster count.
4664  */
4665 int ocfs2_insert_extent(handle_t *handle,
4666 			struct ocfs2_extent_tree *et,
4667 			u32 cpos,
4668 			u64 start_blk,
4669 			u32 new_clusters,
4670 			u8 flags,
4671 			struct ocfs2_alloc_context *meta_ac)
4672 {
4673 	int status;
4674 	int uninitialized_var(free_records);
4675 	struct buffer_head *last_eb_bh = NULL;
4676 	struct ocfs2_insert_type insert = {0, };
4677 	struct ocfs2_extent_rec rec;
4678 
4679 	trace_ocfs2_insert_extent_start(
4680 		(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4681 		cpos, new_clusters);
4682 
4683 	memset(&rec, 0, sizeof(rec));
4684 	rec.e_cpos = cpu_to_le32(cpos);
4685 	rec.e_blkno = cpu_to_le64(start_blk);
4686 	rec.e_leaf_clusters = cpu_to_le16(new_clusters);
4687 	rec.e_flags = flags;
4688 	status = ocfs2_et_insert_check(et, &rec);
4689 	if (status) {
4690 		mlog_errno(status);
4691 		goto bail;
4692 	}
4693 
4694 	status = ocfs2_figure_insert_type(et, &last_eb_bh, &rec,
4695 					  &free_records, &insert);
4696 	if (status < 0) {
4697 		mlog_errno(status);
4698 		goto bail;
4699 	}
4700 
4701 	trace_ocfs2_insert_extent(insert.ins_appending, insert.ins_contig,
4702 				  insert.ins_contig_index, free_records,
4703 				  insert.ins_tree_depth);
4704 
4705 	if (insert.ins_contig == CONTIG_NONE && free_records == 0) {
4706 		status = ocfs2_grow_tree(handle, et,
4707 					 &insert.ins_tree_depth, &last_eb_bh,
4708 					 meta_ac);
4709 		if (status) {
4710 			mlog_errno(status);
4711 			goto bail;
4712 		}
4713 	}
4714 
4715 	/* Finally, we can add clusters. This might rotate the tree for us. */
4716 	status = ocfs2_do_insert_extent(handle, et, &rec, &insert);
4717 	if (status < 0)
4718 		mlog_errno(status);
4719 	else
4720 		ocfs2_et_extent_map_insert(et, &rec);
4721 
4722 bail:
4723 	brelse(last_eb_bh);
4724 
4725 	return status;
4726 }
4727 
4728 /*
4729  * Allcate and add clusters into the extent b-tree.
4730  * The new clusters(clusters_to_add) will be inserted at logical_offset.
4731  * The extent b-tree's root is specified by et, and
4732  * it is not limited to the file storage. Any extent tree can use this
4733  * function if it implements the proper ocfs2_extent_tree.
4734  */
4735 int ocfs2_add_clusters_in_btree(handle_t *handle,
4736 				struct ocfs2_extent_tree *et,
4737 				u32 *logical_offset,
4738 				u32 clusters_to_add,
4739 				int mark_unwritten,
4740 				struct ocfs2_alloc_context *data_ac,
4741 				struct ocfs2_alloc_context *meta_ac,
4742 				enum ocfs2_alloc_restarted *reason_ret)
4743 {
4744 	int status = 0, err = 0;
4745 	int need_free = 0;
4746 	int free_extents;
4747 	enum ocfs2_alloc_restarted reason = RESTART_NONE;
4748 	u32 bit_off, num_bits;
4749 	u64 block;
4750 	u8 flags = 0;
4751 	struct ocfs2_super *osb =
4752 		OCFS2_SB(ocfs2_metadata_cache_get_super(et->et_ci));
4753 
4754 	BUG_ON(!clusters_to_add);
4755 
4756 	if (mark_unwritten)
4757 		flags = OCFS2_EXT_UNWRITTEN;
4758 
4759 	free_extents = ocfs2_num_free_extents(osb, et);
4760 	if (free_extents < 0) {
4761 		status = free_extents;
4762 		mlog_errno(status);
4763 		goto leave;
4764 	}
4765 
4766 	/* there are two cases which could cause us to EAGAIN in the
4767 	 * we-need-more-metadata case:
4768 	 * 1) we haven't reserved *any*
4769 	 * 2) we are so fragmented, we've needed to add metadata too
4770 	 *    many times. */
4771 	if (!free_extents && !meta_ac) {
4772 		err = -1;
4773 		status = -EAGAIN;
4774 		reason = RESTART_META;
4775 		goto leave;
4776 	} else if ((!free_extents)
4777 		   && (ocfs2_alloc_context_bits_left(meta_ac)
4778 		       < ocfs2_extend_meta_needed(et->et_root_el))) {
4779 		err = -2;
4780 		status = -EAGAIN;
4781 		reason = RESTART_META;
4782 		goto leave;
4783 	}
4784 
4785 	status = __ocfs2_claim_clusters(handle, data_ac, 1,
4786 					clusters_to_add, &bit_off, &num_bits);
4787 	if (status < 0) {
4788 		if (status != -ENOSPC)
4789 			mlog_errno(status);
4790 		goto leave;
4791 	}
4792 
4793 	BUG_ON(num_bits > clusters_to_add);
4794 
4795 	/* reserve our write early -- insert_extent may update the tree root */
4796 	status = ocfs2_et_root_journal_access(handle, et,
4797 					      OCFS2_JOURNAL_ACCESS_WRITE);
4798 	if (status < 0) {
4799 		mlog_errno(status);
4800 		need_free = 1;
4801 		goto bail;
4802 	}
4803 
4804 	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4805 	trace_ocfs2_add_clusters_in_btree(
4806 	     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
4807 	     bit_off, num_bits);
4808 	status = ocfs2_insert_extent(handle, et, *logical_offset, block,
4809 				     num_bits, flags, meta_ac);
4810 	if (status < 0) {
4811 		mlog_errno(status);
4812 		need_free = 1;
4813 		goto bail;
4814 	}
4815 
4816 	ocfs2_journal_dirty(handle, et->et_root_bh);
4817 
4818 	clusters_to_add -= num_bits;
4819 	*logical_offset += num_bits;
4820 
4821 	if (clusters_to_add) {
4822 		err = clusters_to_add;
4823 		status = -EAGAIN;
4824 		reason = RESTART_TRANS;
4825 	}
4826 
4827 bail:
4828 	if (need_free) {
4829 		if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
4830 			ocfs2_free_local_alloc_bits(osb, handle, data_ac,
4831 					bit_off, num_bits);
4832 		else
4833 			ocfs2_free_clusters(handle,
4834 					data_ac->ac_inode,
4835 					data_ac->ac_bh,
4836 					ocfs2_clusters_to_blocks(osb->sb, bit_off),
4837 					num_bits);
4838 	}
4839 
4840 leave:
4841 	if (reason_ret)
4842 		*reason_ret = reason;
4843 	trace_ocfs2_add_clusters_in_btree_ret(status, reason, err);
4844 	return status;
4845 }
4846 
4847 static void ocfs2_make_right_split_rec(struct super_block *sb,
4848 				       struct ocfs2_extent_rec *split_rec,
4849 				       u32 cpos,
4850 				       struct ocfs2_extent_rec *rec)
4851 {
4852 	u32 rec_cpos = le32_to_cpu(rec->e_cpos);
4853 	u32 rec_range = rec_cpos + le16_to_cpu(rec->e_leaf_clusters);
4854 
4855 	memset(split_rec, 0, sizeof(struct ocfs2_extent_rec));
4856 
4857 	split_rec->e_cpos = cpu_to_le32(cpos);
4858 	split_rec->e_leaf_clusters = cpu_to_le16(rec_range - cpos);
4859 
4860 	split_rec->e_blkno = rec->e_blkno;
4861 	le64_add_cpu(&split_rec->e_blkno,
4862 		     ocfs2_clusters_to_blocks(sb, cpos - rec_cpos));
4863 
4864 	split_rec->e_flags = rec->e_flags;
4865 }
4866 
4867 static int ocfs2_split_and_insert(handle_t *handle,
4868 				  struct ocfs2_extent_tree *et,
4869 				  struct ocfs2_path *path,
4870 				  struct buffer_head **last_eb_bh,
4871 				  int split_index,
4872 				  struct ocfs2_extent_rec *orig_split_rec,
4873 				  struct ocfs2_alloc_context *meta_ac)
4874 {
4875 	int ret = 0, depth;
4876 	unsigned int insert_range, rec_range, do_leftright = 0;
4877 	struct ocfs2_extent_rec tmprec;
4878 	struct ocfs2_extent_list *rightmost_el;
4879 	struct ocfs2_extent_rec rec;
4880 	struct ocfs2_extent_rec split_rec = *orig_split_rec;
4881 	struct ocfs2_insert_type insert;
4882 	struct ocfs2_extent_block *eb;
4883 
4884 leftright:
4885 	/*
4886 	 * Store a copy of the record on the stack - it might move
4887 	 * around as the tree is manipulated below.
4888 	 */
4889 	rec = path_leaf_el(path)->l_recs[split_index];
4890 
4891 	rightmost_el = et->et_root_el;
4892 
4893 	depth = le16_to_cpu(rightmost_el->l_tree_depth);
4894 	if (depth) {
4895 		BUG_ON(!(*last_eb_bh));
4896 		eb = (struct ocfs2_extent_block *) (*last_eb_bh)->b_data;
4897 		rightmost_el = &eb->h_list;
4898 	}
4899 
4900 	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
4901 	    le16_to_cpu(rightmost_el->l_count)) {
4902 		ret = ocfs2_grow_tree(handle, et,
4903 				      &depth, last_eb_bh, meta_ac);
4904 		if (ret) {
4905 			mlog_errno(ret);
4906 			goto out;
4907 		}
4908 	}
4909 
4910 	memset(&insert, 0, sizeof(struct ocfs2_insert_type));
4911 	insert.ins_appending = APPEND_NONE;
4912 	insert.ins_contig = CONTIG_NONE;
4913 	insert.ins_tree_depth = depth;
4914 
4915 	insert_range = le32_to_cpu(split_rec.e_cpos) +
4916 		le16_to_cpu(split_rec.e_leaf_clusters);
4917 	rec_range = le32_to_cpu(rec.e_cpos) +
4918 		le16_to_cpu(rec.e_leaf_clusters);
4919 
4920 	if (split_rec.e_cpos == rec.e_cpos) {
4921 		insert.ins_split = SPLIT_LEFT;
4922 	} else if (insert_range == rec_range) {
4923 		insert.ins_split = SPLIT_RIGHT;
4924 	} else {
4925 		/*
4926 		 * Left/right split. We fake this as a right split
4927 		 * first and then make a second pass as a left split.
4928 		 */
4929 		insert.ins_split = SPLIT_RIGHT;
4930 
4931 		ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
4932 					   &tmprec, insert_range, &rec);
4933 
4934 		split_rec = tmprec;
4935 
4936 		BUG_ON(do_leftright);
4937 		do_leftright = 1;
4938 	}
4939 
4940 	ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
4941 	if (ret) {
4942 		mlog_errno(ret);
4943 		goto out;
4944 	}
4945 
4946 	if (do_leftright == 1) {
4947 		u32 cpos;
4948 		struct ocfs2_extent_list *el;
4949 
4950 		do_leftright++;
4951 		split_rec = *orig_split_rec;
4952 
4953 		ocfs2_reinit_path(path, 1);
4954 
4955 		cpos = le32_to_cpu(split_rec.e_cpos);
4956 		ret = ocfs2_find_path(et->et_ci, path, cpos);
4957 		if (ret) {
4958 			mlog_errno(ret);
4959 			goto out;
4960 		}
4961 
4962 		el = path_leaf_el(path);
4963 		split_index = ocfs2_search_extent_list(el, cpos);
4964 		goto leftright;
4965 	}
4966 out:
4967 
4968 	return ret;
4969 }
4970 
4971 static int ocfs2_replace_extent_rec(handle_t *handle,
4972 				    struct ocfs2_extent_tree *et,
4973 				    struct ocfs2_path *path,
4974 				    struct ocfs2_extent_list *el,
4975 				    int split_index,
4976 				    struct ocfs2_extent_rec *split_rec)
4977 {
4978 	int ret;
4979 
4980 	ret = ocfs2_path_bh_journal_access(handle, et->et_ci, path,
4981 					   path_num_items(path) - 1);
4982 	if (ret) {
4983 		mlog_errno(ret);
4984 		goto out;
4985 	}
4986 
4987 	el->l_recs[split_index] = *split_rec;
4988 
4989 	ocfs2_journal_dirty(handle, path_leaf_bh(path));
4990 out:
4991 	return ret;
4992 }
4993 
4994 /*
4995  * Split part or all of the extent record at split_index in the leaf
4996  * pointed to by path. Merge with the contiguous extent record if needed.
4997  *
4998  * Care is taken to handle contiguousness so as to not grow the tree.
4999  *
5000  * meta_ac is not strictly necessary - we only truly need it if growth
5001  * of the tree is required. All other cases will degrade into a less
5002  * optimal tree layout.
5003  *
5004  * last_eb_bh should be the rightmost leaf block for any extent
5005  * btree. Since a split may grow the tree or a merge might shrink it,
5006  * the caller cannot trust the contents of that buffer after this call.
5007  *
5008  * This code is optimized for readability - several passes might be
5009  * made over certain portions of the tree. All of those blocks will
5010  * have been brought into cache (and pinned via the journal), so the
5011  * extra overhead is not expressed in terms of disk reads.
5012  */
5013 int ocfs2_split_extent(handle_t *handle,
5014 		       struct ocfs2_extent_tree *et,
5015 		       struct ocfs2_path *path,
5016 		       int split_index,
5017 		       struct ocfs2_extent_rec *split_rec,
5018 		       struct ocfs2_alloc_context *meta_ac,
5019 		       struct ocfs2_cached_dealloc_ctxt *dealloc)
5020 {
5021 	int ret = 0;
5022 	struct ocfs2_extent_list *el = path_leaf_el(path);
5023 	struct buffer_head *last_eb_bh = NULL;
5024 	struct ocfs2_extent_rec *rec = &el->l_recs[split_index];
5025 	struct ocfs2_merge_ctxt ctxt;
5026 	struct ocfs2_extent_list *rightmost_el;
5027 
5028 	if (le32_to_cpu(rec->e_cpos) > le32_to_cpu(split_rec->e_cpos) ||
5029 	    ((le32_to_cpu(rec->e_cpos) + le16_to_cpu(rec->e_leaf_clusters)) <
5030 	     (le32_to_cpu(split_rec->e_cpos) + le16_to_cpu(split_rec->e_leaf_clusters)))) {
5031 		ret = -EIO;
5032 		mlog_errno(ret);
5033 		goto out;
5034 	}
5035 
5036 	ctxt.c_contig_type = ocfs2_figure_merge_contig_type(et, path, el,
5037 							    split_index,
5038 							    split_rec);
5039 
5040 	/*
5041 	 * The core merge / split code wants to know how much room is
5042 	 * left in this allocation tree, so we pass the
5043 	 * rightmost extent list.
5044 	 */
5045 	if (path->p_tree_depth) {
5046 		struct ocfs2_extent_block *eb;
5047 
5048 		ret = ocfs2_read_extent_block(et->et_ci,
5049 					      ocfs2_et_get_last_eb_blk(et),
5050 					      &last_eb_bh);
5051 		if (ret) {
5052 			mlog_errno(ret);
5053 			goto out;
5054 		}
5055 
5056 		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
5057 		rightmost_el = &eb->h_list;
5058 	} else
5059 		rightmost_el = path_root_el(path);
5060 
5061 	if (rec->e_cpos == split_rec->e_cpos &&
5062 	    rec->e_leaf_clusters == split_rec->e_leaf_clusters)
5063 		ctxt.c_split_covers_rec = 1;
5064 	else
5065 		ctxt.c_split_covers_rec = 0;
5066 
5067 	ctxt.c_has_empty_extent = ocfs2_is_empty_extent(&el->l_recs[0]);
5068 
5069 	trace_ocfs2_split_extent(split_index, ctxt.c_contig_type,
5070 				 ctxt.c_has_empty_extent,
5071 				 ctxt.c_split_covers_rec);
5072 
5073 	if (ctxt.c_contig_type == CONTIG_NONE) {
5074 		if (ctxt.c_split_covers_rec)
5075 			ret = ocfs2_replace_extent_rec(handle, et, path, el,
5076 						       split_index, split_rec);
5077 		else
5078 			ret = ocfs2_split_and_insert(handle, et, path,
5079 						     &last_eb_bh, split_index,
5080 						     split_rec, meta_ac);
5081 		if (ret)
5082 			mlog_errno(ret);
5083 	} else {
5084 		ret = ocfs2_try_to_merge_extent(handle, et, path,
5085 						split_index, split_rec,
5086 						dealloc, &ctxt);
5087 		if (ret)
5088 			mlog_errno(ret);
5089 	}
5090 
5091 out:
5092 	brelse(last_eb_bh);
5093 	return ret;
5094 }
5095 
5096 /*
5097  * Change the flags of the already-existing extent at cpos for len clusters.
5098  *
5099  * new_flags: the flags we want to set.
5100  * clear_flags: the flags we want to clear.
5101  * phys: the new physical offset we want this new extent starts from.
5102  *
5103  * If the existing extent is larger than the request, initiate a
5104  * split. An attempt will be made at merging with adjacent extents.
5105  *
5106  * The caller is responsible for passing down meta_ac if we'll need it.
5107  */
5108 int ocfs2_change_extent_flag(handle_t *handle,
5109 			     struct ocfs2_extent_tree *et,
5110 			     u32 cpos, u32 len, u32 phys,
5111 			     struct ocfs2_alloc_context *meta_ac,
5112 			     struct ocfs2_cached_dealloc_ctxt *dealloc,
5113 			     int new_flags, int clear_flags)
5114 {
5115 	int ret, index;
5116 	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5117 	u64 start_blkno = ocfs2_clusters_to_blocks(sb, phys);
5118 	struct ocfs2_extent_rec split_rec;
5119 	struct ocfs2_path *left_path = NULL;
5120 	struct ocfs2_extent_list *el;
5121 	struct ocfs2_extent_rec *rec;
5122 
5123 	left_path = ocfs2_new_path_from_et(et);
5124 	if (!left_path) {
5125 		ret = -ENOMEM;
5126 		mlog_errno(ret);
5127 		goto out;
5128 	}
5129 
5130 	ret = ocfs2_find_path(et->et_ci, left_path, cpos);
5131 	if (ret) {
5132 		mlog_errno(ret);
5133 		goto out;
5134 	}
5135 	el = path_leaf_el(left_path);
5136 
5137 	index = ocfs2_search_extent_list(el, cpos);
5138 	if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
5139 		ocfs2_error(sb,
5140 			    "Owner %llu has an extent at cpos %u which can no "
5141 			    "longer be found.\n",
5142 			     (unsigned long long)
5143 			     ocfs2_metadata_cache_owner(et->et_ci), cpos);
5144 		ret = -EROFS;
5145 		goto out;
5146 	}
5147 
5148 	ret = -EIO;
5149 	rec = &el->l_recs[index];
5150 	if (new_flags && (rec->e_flags & new_flags)) {
5151 		mlog(ML_ERROR, "Owner %llu tried to set %d flags on an "
5152 		     "extent that already had them",
5153 		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5154 		     new_flags);
5155 		goto out;
5156 	}
5157 
5158 	if (clear_flags && !(rec->e_flags & clear_flags)) {
5159 		mlog(ML_ERROR, "Owner %llu tried to clear %d flags on an "
5160 		     "extent that didn't have them",
5161 		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5162 		     clear_flags);
5163 		goto out;
5164 	}
5165 
5166 	memset(&split_rec, 0, sizeof(struct ocfs2_extent_rec));
5167 	split_rec.e_cpos = cpu_to_le32(cpos);
5168 	split_rec.e_leaf_clusters = cpu_to_le16(len);
5169 	split_rec.e_blkno = cpu_to_le64(start_blkno);
5170 	split_rec.e_flags = rec->e_flags;
5171 	if (new_flags)
5172 		split_rec.e_flags |= new_flags;
5173 	if (clear_flags)
5174 		split_rec.e_flags &= ~clear_flags;
5175 
5176 	ret = ocfs2_split_extent(handle, et, left_path,
5177 				 index, &split_rec, meta_ac,
5178 				 dealloc);
5179 	if (ret)
5180 		mlog_errno(ret);
5181 
5182 out:
5183 	ocfs2_free_path(left_path);
5184 	return ret;
5185 
5186 }
5187 
5188 /*
5189  * Mark the already-existing extent at cpos as written for len clusters.
5190  * This removes the unwritten extent flag.
5191  *
5192  * If the existing extent is larger than the request, initiate a
5193  * split. An attempt will be made at merging with adjacent extents.
5194  *
5195  * The caller is responsible for passing down meta_ac if we'll need it.
5196  */
5197 int ocfs2_mark_extent_written(struct inode *inode,
5198 			      struct ocfs2_extent_tree *et,
5199 			      handle_t *handle, u32 cpos, u32 len, u32 phys,
5200 			      struct ocfs2_alloc_context *meta_ac,
5201 			      struct ocfs2_cached_dealloc_ctxt *dealloc)
5202 {
5203 	int ret;
5204 
5205 	trace_ocfs2_mark_extent_written(
5206 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
5207 		cpos, len, phys);
5208 
5209 	if (!ocfs2_writes_unwritten_extents(OCFS2_SB(inode->i_sb))) {
5210 		ocfs2_error(inode->i_sb, "Inode %llu has unwritten extents "
5211 			    "that are being written to, but the feature bit "
5212 			    "is not set in the super block.",
5213 			    (unsigned long long)OCFS2_I(inode)->ip_blkno);
5214 		ret = -EROFS;
5215 		goto out;
5216 	}
5217 
5218 	/*
5219 	 * XXX: This should be fixed up so that we just re-insert the
5220 	 * next extent records.
5221 	 */
5222 	ocfs2_et_extent_map_truncate(et, 0);
5223 
5224 	ret = ocfs2_change_extent_flag(handle, et, cpos,
5225 				       len, phys, meta_ac, dealloc,
5226 				       0, OCFS2_EXT_UNWRITTEN);
5227 	if (ret)
5228 		mlog_errno(ret);
5229 
5230 out:
5231 	return ret;
5232 }
5233 
5234 static int ocfs2_split_tree(handle_t *handle, struct ocfs2_extent_tree *et,
5235 			    struct ocfs2_path *path,
5236 			    int index, u32 new_range,
5237 			    struct ocfs2_alloc_context *meta_ac)
5238 {
5239 	int ret, depth, credits;
5240 	struct buffer_head *last_eb_bh = NULL;
5241 	struct ocfs2_extent_block *eb;
5242 	struct ocfs2_extent_list *rightmost_el, *el;
5243 	struct ocfs2_extent_rec split_rec;
5244 	struct ocfs2_extent_rec *rec;
5245 	struct ocfs2_insert_type insert;
5246 
5247 	/*
5248 	 * Setup the record to split before we grow the tree.
5249 	 */
5250 	el = path_leaf_el(path);
5251 	rec = &el->l_recs[index];
5252 	ocfs2_make_right_split_rec(ocfs2_metadata_cache_get_super(et->et_ci),
5253 				   &split_rec, new_range, rec);
5254 
5255 	depth = path->p_tree_depth;
5256 	if (depth > 0) {
5257 		ret = ocfs2_read_extent_block(et->et_ci,
5258 					      ocfs2_et_get_last_eb_blk(et),
5259 					      &last_eb_bh);
5260 		if (ret < 0) {
5261 			mlog_errno(ret);
5262 			goto out;
5263 		}
5264 
5265 		eb = (struct ocfs2_extent_block *) last_eb_bh->b_data;
5266 		rightmost_el = &eb->h_list;
5267 	} else
5268 		rightmost_el = path_leaf_el(path);
5269 
5270 	credits = path->p_tree_depth +
5271 		  ocfs2_extend_meta_needed(et->et_root_el);
5272 	ret = ocfs2_extend_trans(handle, credits);
5273 	if (ret) {
5274 		mlog_errno(ret);
5275 		goto out;
5276 	}
5277 
5278 	if (le16_to_cpu(rightmost_el->l_next_free_rec) ==
5279 	    le16_to_cpu(rightmost_el->l_count)) {
5280 		ret = ocfs2_grow_tree(handle, et, &depth, &last_eb_bh,
5281 				      meta_ac);
5282 		if (ret) {
5283 			mlog_errno(ret);
5284 			goto out;
5285 		}
5286 	}
5287 
5288 	memset(&insert, 0, sizeof(struct ocfs2_insert_type));
5289 	insert.ins_appending = APPEND_NONE;
5290 	insert.ins_contig = CONTIG_NONE;
5291 	insert.ins_split = SPLIT_RIGHT;
5292 	insert.ins_tree_depth = depth;
5293 
5294 	ret = ocfs2_do_insert_extent(handle, et, &split_rec, &insert);
5295 	if (ret)
5296 		mlog_errno(ret);
5297 
5298 out:
5299 	brelse(last_eb_bh);
5300 	return ret;
5301 }
5302 
5303 static int ocfs2_truncate_rec(handle_t *handle,
5304 			      struct ocfs2_extent_tree *et,
5305 			      struct ocfs2_path *path, int index,
5306 			      struct ocfs2_cached_dealloc_ctxt *dealloc,
5307 			      u32 cpos, u32 len)
5308 {
5309 	int ret;
5310 	u32 left_cpos, rec_range, trunc_range;
5311 	int wants_rotate = 0, is_rightmost_tree_rec = 0;
5312 	struct super_block *sb = ocfs2_metadata_cache_get_super(et->et_ci);
5313 	struct ocfs2_path *left_path = NULL;
5314 	struct ocfs2_extent_list *el = path_leaf_el(path);
5315 	struct ocfs2_extent_rec *rec;
5316 	struct ocfs2_extent_block *eb;
5317 
5318 	if (ocfs2_is_empty_extent(&el->l_recs[0]) && index > 0) {
5319 		ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
5320 		if (ret) {
5321 			mlog_errno(ret);
5322 			goto out;
5323 		}
5324 
5325 		index--;
5326 	}
5327 
5328 	if (index == (le16_to_cpu(el->l_next_free_rec) - 1) &&
5329 	    path->p_tree_depth) {
5330 		/*
5331 		 * Check whether this is the rightmost tree record. If
5332 		 * we remove all of this record or part of its right
5333 		 * edge then an update of the record lengths above it
5334 		 * will be required.
5335 		 */
5336 		eb = (struct ocfs2_extent_block *)path_leaf_bh(path)->b_data;
5337 		if (eb->h_next_leaf_blk == 0)
5338 			is_rightmost_tree_rec = 1;
5339 	}
5340 
5341 	rec = &el->l_recs[index];
5342 	if (index == 0 && path->p_tree_depth &&
5343 	    le32_to_cpu(rec->e_cpos) == cpos) {
5344 		/*
5345 		 * Changing the leftmost offset (via partial or whole
5346 		 * record truncate) of an interior (or rightmost) path
5347 		 * means we have to update the subtree that is formed
5348 		 * by this leaf and the one to it's left.
5349 		 *
5350 		 * There are two cases we can skip:
5351 		 *   1) Path is the leftmost one in our btree.
5352 		 *   2) The leaf is rightmost and will be empty after
5353 		 *      we remove the extent record - the rotate code
5354 		 *      knows how to update the newly formed edge.
5355 		 */
5356 
5357 		ret = ocfs2_find_cpos_for_left_leaf(sb, path, &left_cpos);
5358 		if (ret) {
5359 			mlog_errno(ret);
5360 			goto out;
5361 		}
5362 
5363 		if (left_cpos && le16_to_cpu(el->l_next_free_rec) > 1) {
5364 			left_path = ocfs2_new_path_from_path(path);
5365 			if (!left_path) {
5366 				ret = -ENOMEM;
5367 				mlog_errno(ret);
5368 				goto out;
5369 			}
5370 
5371 			ret = ocfs2_find_path(et->et_ci, left_path,
5372 					      left_cpos);
5373 			if (ret) {
5374 				mlog_errno(ret);
5375 				goto out;
5376 			}
5377 		}
5378 	}
5379 
5380 	ret = ocfs2_extend_rotate_transaction(handle, 0,
5381 					      handle->h_buffer_credits,
5382 					      path);
5383 	if (ret) {
5384 		mlog_errno(ret);
5385 		goto out;
5386 	}
5387 
5388 	ret = ocfs2_journal_access_path(et->et_ci, handle, path);
5389 	if (ret) {
5390 		mlog_errno(ret);
5391 		goto out;
5392 	}
5393 
5394 	ret = ocfs2_journal_access_path(et->et_ci, handle, left_path);
5395 	if (ret) {
5396 		mlog_errno(ret);
5397 		goto out;
5398 	}
5399 
5400 	rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5401 	trunc_range = cpos + len;
5402 
5403 	if (le32_to_cpu(rec->e_cpos) == cpos && rec_range == trunc_range) {
5404 		int next_free;
5405 
5406 		memset(rec, 0, sizeof(*rec));
5407 		ocfs2_cleanup_merge(el, index);
5408 		wants_rotate = 1;
5409 
5410 		next_free = le16_to_cpu(el->l_next_free_rec);
5411 		if (is_rightmost_tree_rec && next_free > 1) {
5412 			/*
5413 			 * We skip the edge update if this path will
5414 			 * be deleted by the rotate code.
5415 			 */
5416 			rec = &el->l_recs[next_free - 1];
5417 			ocfs2_adjust_rightmost_records(handle, et, path,
5418 						       rec);
5419 		}
5420 	} else if (le32_to_cpu(rec->e_cpos) == cpos) {
5421 		/* Remove leftmost portion of the record. */
5422 		le32_add_cpu(&rec->e_cpos, len);
5423 		le64_add_cpu(&rec->e_blkno, ocfs2_clusters_to_blocks(sb, len));
5424 		le16_add_cpu(&rec->e_leaf_clusters, -len);
5425 	} else if (rec_range == trunc_range) {
5426 		/* Remove rightmost portion of the record */
5427 		le16_add_cpu(&rec->e_leaf_clusters, -len);
5428 		if (is_rightmost_tree_rec)
5429 			ocfs2_adjust_rightmost_records(handle, et, path, rec);
5430 	} else {
5431 		/* Caller should have trapped this. */
5432 		mlog(ML_ERROR, "Owner %llu: Invalid record truncate: (%u, %u) "
5433 		     "(%u, %u)\n",
5434 		     (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5435 		     le32_to_cpu(rec->e_cpos),
5436 		     le16_to_cpu(rec->e_leaf_clusters), cpos, len);
5437 		BUG();
5438 	}
5439 
5440 	if (left_path) {
5441 		int subtree_index;
5442 
5443 		subtree_index = ocfs2_find_subtree_root(et, left_path, path);
5444 		ocfs2_complete_edge_insert(handle, left_path, path,
5445 					   subtree_index);
5446 	}
5447 
5448 	ocfs2_journal_dirty(handle, path_leaf_bh(path));
5449 
5450 	ret = ocfs2_rotate_tree_left(handle, et, path, dealloc);
5451 	if (ret) {
5452 		mlog_errno(ret);
5453 		goto out;
5454 	}
5455 
5456 out:
5457 	ocfs2_free_path(left_path);
5458 	return ret;
5459 }
5460 
5461 int ocfs2_remove_extent(handle_t *handle,
5462 			struct ocfs2_extent_tree *et,
5463 			u32 cpos, u32 len,
5464 			struct ocfs2_alloc_context *meta_ac,
5465 			struct ocfs2_cached_dealloc_ctxt *dealloc)
5466 {
5467 	int ret, index;
5468 	u32 rec_range, trunc_range;
5469 	struct ocfs2_extent_rec *rec;
5470 	struct ocfs2_extent_list *el;
5471 	struct ocfs2_path *path = NULL;
5472 
5473 	/*
5474 	 * XXX: Why are we truncating to 0 instead of wherever this
5475 	 * affects us?
5476 	 */
5477 	ocfs2_et_extent_map_truncate(et, 0);
5478 
5479 	path = ocfs2_new_path_from_et(et);
5480 	if (!path) {
5481 		ret = -ENOMEM;
5482 		mlog_errno(ret);
5483 		goto out;
5484 	}
5485 
5486 	ret = ocfs2_find_path(et->et_ci, path, cpos);
5487 	if (ret) {
5488 		mlog_errno(ret);
5489 		goto out;
5490 	}
5491 
5492 	el = path_leaf_el(path);
5493 	index = ocfs2_search_extent_list(el, cpos);
5494 	if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
5495 		ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5496 			    "Owner %llu has an extent at cpos %u which can no "
5497 			    "longer be found.\n",
5498 			    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5499 			    cpos);
5500 		ret = -EROFS;
5501 		goto out;
5502 	}
5503 
5504 	/*
5505 	 * We have 3 cases of extent removal:
5506 	 *   1) Range covers the entire extent rec
5507 	 *   2) Range begins or ends on one edge of the extent rec
5508 	 *   3) Range is in the middle of the extent rec (no shared edges)
5509 	 *
5510 	 * For case 1 we remove the extent rec and left rotate to
5511 	 * fill the hole.
5512 	 *
5513 	 * For case 2 we just shrink the existing extent rec, with a
5514 	 * tree update if the shrinking edge is also the edge of an
5515 	 * extent block.
5516 	 *
5517 	 * For case 3 we do a right split to turn the extent rec into
5518 	 * something case 2 can handle.
5519 	 */
5520 	rec = &el->l_recs[index];
5521 	rec_range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
5522 	trunc_range = cpos + len;
5523 
5524 	BUG_ON(cpos < le32_to_cpu(rec->e_cpos) || trunc_range > rec_range);
5525 
5526 	trace_ocfs2_remove_extent(
5527 		(unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5528 		cpos, len, index, le32_to_cpu(rec->e_cpos),
5529 		ocfs2_rec_clusters(el, rec));
5530 
5531 	if (le32_to_cpu(rec->e_cpos) == cpos || rec_range == trunc_range) {
5532 		ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5533 					 cpos, len);
5534 		if (ret) {
5535 			mlog_errno(ret);
5536 			goto out;
5537 		}
5538 	} else {
5539 		ret = ocfs2_split_tree(handle, et, path, index,
5540 				       trunc_range, meta_ac);
5541 		if (ret) {
5542 			mlog_errno(ret);
5543 			goto out;
5544 		}
5545 
5546 		/*
5547 		 * The split could have manipulated the tree enough to
5548 		 * move the record location, so we have to look for it again.
5549 		 */
5550 		ocfs2_reinit_path(path, 1);
5551 
5552 		ret = ocfs2_find_path(et->et_ci, path, cpos);
5553 		if (ret) {
5554 			mlog_errno(ret);
5555 			goto out;
5556 		}
5557 
5558 		el = path_leaf_el(path);
5559 		index = ocfs2_search_extent_list(el, cpos);
5560 		if (index == -1 || index >= le16_to_cpu(el->l_next_free_rec)) {
5561 			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5562 				    "Owner %llu: split at cpos %u lost record.",
5563 				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5564 				    cpos);
5565 			ret = -EROFS;
5566 			goto out;
5567 		}
5568 
5569 		/*
5570 		 * Double check our values here. If anything is fishy,
5571 		 * it's easier to catch it at the top level.
5572 		 */
5573 		rec = &el->l_recs[index];
5574 		rec_range = le32_to_cpu(rec->e_cpos) +
5575 			ocfs2_rec_clusters(el, rec);
5576 		if (rec_range != trunc_range) {
5577 			ocfs2_error(ocfs2_metadata_cache_get_super(et->et_ci),
5578 				    "Owner %llu: error after split at cpos %u"
5579 				    "trunc len %u, existing record is (%u,%u)",
5580 				    (unsigned long long)ocfs2_metadata_cache_owner(et->et_ci),
5581 				    cpos, len, le32_to_cpu(rec->e_cpos),
5582 				    ocfs2_rec_clusters(el, rec));
5583 			ret = -EROFS;
5584 			goto out;
5585 		}
5586 
5587 		ret = ocfs2_truncate_rec(handle, et, path, index, dealloc,
5588 					 cpos, len);
5589 		if (ret) {
5590 			mlog_errno(ret);
5591 			goto out;
5592 		}
5593 	}
5594 
5595 out:
5596 	ocfs2_free_path(path);
5597 	return ret;
5598 }
5599 
5600 /*
5601  * ocfs2_reserve_blocks_for_rec_trunc() would look basically the
5602  * same as ocfs2_lock_alloctors(), except for it accepts a blocks
5603  * number to reserve some extra blocks, and it only handles meta
5604  * data allocations.
5605  *
5606  * Currently, only ocfs2_remove_btree_range() uses it for truncating
5607  * and punching holes.
5608  */
5609 static int ocfs2_reserve_blocks_for_rec_trunc(struct inode *inode,
5610 					      struct ocfs2_extent_tree *et,
5611 					      u32 extents_to_split,
5612 					      struct ocfs2_alloc_context **ac,
5613 					      int extra_blocks)
5614 {
5615 	int ret = 0, num_free_extents;
5616 	unsigned int max_recs_needed = 2 * extents_to_split;
5617 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5618 
5619 	*ac = NULL;
5620 
5621 	num_free_extents = ocfs2_num_free_extents(osb, et);
5622 	if (num_free_extents < 0) {
5623 		ret = num_free_extents;
5624 		mlog_errno(ret);
5625 		goto out;
5626 	}
5627 
5628 	if (!num_free_extents ||
5629 	    (ocfs2_sparse_alloc(osb) && num_free_extents < max_recs_needed))
5630 		extra_blocks += ocfs2_extend_meta_needed(et->et_root_el);
5631 
5632 	if (extra_blocks) {
5633 		ret = ocfs2_reserve_new_metadata_blocks(osb, extra_blocks, ac);
5634 		if (ret < 0) {
5635 			if (ret != -ENOSPC)
5636 				mlog_errno(ret);
5637 			goto out;
5638 		}
5639 	}
5640 
5641 out:
5642 	if (ret) {
5643 		if (*ac) {
5644 			ocfs2_free_alloc_context(*ac);
5645 			*ac = NULL;
5646 		}
5647 	}
5648 
5649 	return ret;
5650 }
5651 
5652 int ocfs2_remove_btree_range(struct inode *inode,
5653 			     struct ocfs2_extent_tree *et,
5654 			     u32 cpos, u32 phys_cpos, u32 len, int flags,
5655 			     struct ocfs2_cached_dealloc_ctxt *dealloc,
5656 			     u64 refcount_loc)
5657 {
5658 	int ret, credits = 0, extra_blocks = 0;
5659 	u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
5660 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5661 	struct inode *tl_inode = osb->osb_tl_inode;
5662 	handle_t *handle;
5663 	struct ocfs2_alloc_context *meta_ac = NULL;
5664 	struct ocfs2_refcount_tree *ref_tree = NULL;
5665 
5666 	if ((flags & OCFS2_EXT_REFCOUNTED) && len) {
5667 		BUG_ON(!(OCFS2_I(inode)->ip_dyn_features &
5668 			 OCFS2_HAS_REFCOUNT_FL));
5669 
5670 		ret = ocfs2_lock_refcount_tree(osb, refcount_loc, 1,
5671 					       &ref_tree, NULL);
5672 		if (ret) {
5673 			mlog_errno(ret);
5674 			goto bail;
5675 		}
5676 
5677 		ret = ocfs2_prepare_refcount_change_for_del(inode,
5678 							    refcount_loc,
5679 							    phys_blkno,
5680 							    len,
5681 							    &credits,
5682 							    &extra_blocks);
5683 		if (ret < 0) {
5684 			mlog_errno(ret);
5685 			goto bail;
5686 		}
5687 	}
5688 
5689 	ret = ocfs2_reserve_blocks_for_rec_trunc(inode, et, 1, &meta_ac,
5690 						 extra_blocks);
5691 	if (ret) {
5692 		mlog_errno(ret);
5693 		goto bail;
5694 	}
5695 
5696 	mutex_lock(&tl_inode->i_mutex);
5697 
5698 	if (ocfs2_truncate_log_needs_flush(osb)) {
5699 		ret = __ocfs2_flush_truncate_log(osb);
5700 		if (ret < 0) {
5701 			mlog_errno(ret);
5702 			goto out;
5703 		}
5704 	}
5705 
5706 	handle = ocfs2_start_trans(osb,
5707 			ocfs2_remove_extent_credits(osb->sb) + credits);
5708 	if (IS_ERR(handle)) {
5709 		ret = PTR_ERR(handle);
5710 		mlog_errno(ret);
5711 		goto out;
5712 	}
5713 
5714 	ret = ocfs2_et_root_journal_access(handle, et,
5715 					   OCFS2_JOURNAL_ACCESS_WRITE);
5716 	if (ret) {
5717 		mlog_errno(ret);
5718 		goto out_commit;
5719 	}
5720 
5721 	dquot_free_space_nodirty(inode,
5722 				  ocfs2_clusters_to_bytes(inode->i_sb, len));
5723 
5724 	ret = ocfs2_remove_extent(handle, et, cpos, len, meta_ac, dealloc);
5725 	if (ret) {
5726 		mlog_errno(ret);
5727 		goto out_commit;
5728 	}
5729 
5730 	ocfs2_et_update_clusters(et, -len);
5731 
5732 	ocfs2_journal_dirty(handle, et->et_root_bh);
5733 
5734 	if (phys_blkno) {
5735 		if (flags & OCFS2_EXT_REFCOUNTED)
5736 			ret = ocfs2_decrease_refcount(inode, handle,
5737 					ocfs2_blocks_to_clusters(osb->sb,
5738 								 phys_blkno),
5739 					len, meta_ac,
5740 					dealloc, 1);
5741 		else
5742 			ret = ocfs2_truncate_log_append(osb, handle,
5743 							phys_blkno, len);
5744 		if (ret)
5745 			mlog_errno(ret);
5746 
5747 	}
5748 
5749 out_commit:
5750 	ocfs2_commit_trans(osb, handle);
5751 out:
5752 	mutex_unlock(&tl_inode->i_mutex);
5753 bail:
5754 	if (meta_ac)
5755 		ocfs2_free_alloc_context(meta_ac);
5756 
5757 	if (ref_tree)
5758 		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
5759 
5760 	return ret;
5761 }
5762 
5763 int ocfs2_truncate_log_needs_flush(struct ocfs2_super *osb)
5764 {
5765 	struct buffer_head *tl_bh = osb->osb_tl_bh;
5766 	struct ocfs2_dinode *di;
5767 	struct ocfs2_truncate_log *tl;
5768 
5769 	di = (struct ocfs2_dinode *) tl_bh->b_data;
5770 	tl = &di->id2.i_dealloc;
5771 
5772 	mlog_bug_on_msg(le16_to_cpu(tl->tl_used) > le16_to_cpu(tl->tl_count),
5773 			"slot %d, invalid truncate log parameters: used = "
5774 			"%u, count = %u\n", osb->slot_num,
5775 			le16_to_cpu(tl->tl_used), le16_to_cpu(tl->tl_count));
5776 	return le16_to_cpu(tl->tl_used) == le16_to_cpu(tl->tl_count);
5777 }
5778 
5779 static int ocfs2_truncate_log_can_coalesce(struct ocfs2_truncate_log *tl,
5780 					   unsigned int new_start)
5781 {
5782 	unsigned int tail_index;
5783 	unsigned int current_tail;
5784 
5785 	/* No records, nothing to coalesce */
5786 	if (!le16_to_cpu(tl->tl_used))
5787 		return 0;
5788 
5789 	tail_index = le16_to_cpu(tl->tl_used) - 1;
5790 	current_tail = le32_to_cpu(tl->tl_recs[tail_index].t_start);
5791 	current_tail += le32_to_cpu(tl->tl_recs[tail_index].t_clusters);
5792 
5793 	return current_tail == new_start;
5794 }
5795 
5796 int ocfs2_truncate_log_append(struct ocfs2_super *osb,
5797 			      handle_t *handle,
5798 			      u64 start_blk,
5799 			      unsigned int num_clusters)
5800 {
5801 	int status, index;
5802 	unsigned int start_cluster, tl_count;
5803 	struct inode *tl_inode = osb->osb_tl_inode;
5804 	struct buffer_head *tl_bh = osb->osb_tl_bh;
5805 	struct ocfs2_dinode *di;
5806 	struct ocfs2_truncate_log *tl;
5807 
5808 	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
5809 
5810 	start_cluster = ocfs2_blocks_to_clusters(osb->sb, start_blk);
5811 
5812 	di = (struct ocfs2_dinode *) tl_bh->b_data;
5813 
5814 	/* tl_bh is loaded from ocfs2_truncate_log_init().  It's validated
5815 	 * by the underlying call to ocfs2_read_inode_block(), so any
5816 	 * corruption is a code bug */
5817 	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5818 
5819 	tl = &di->id2.i_dealloc;
5820 	tl_count = le16_to_cpu(tl->tl_count);
5821 	mlog_bug_on_msg(tl_count > ocfs2_truncate_recs_per_inode(osb->sb) ||
5822 			tl_count == 0,
5823 			"Truncate record count on #%llu invalid "
5824 			"wanted %u, actual %u\n",
5825 			(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5826 			ocfs2_truncate_recs_per_inode(osb->sb),
5827 			le16_to_cpu(tl->tl_count));
5828 
5829 	/* Caller should have known to flush before calling us. */
5830 	index = le16_to_cpu(tl->tl_used);
5831 	if (index >= tl_count) {
5832 		status = -ENOSPC;
5833 		mlog_errno(status);
5834 		goto bail;
5835 	}
5836 
5837 	status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
5838 					 OCFS2_JOURNAL_ACCESS_WRITE);
5839 	if (status < 0) {
5840 		mlog_errno(status);
5841 		goto bail;
5842 	}
5843 
5844 	trace_ocfs2_truncate_log_append(
5845 		(unsigned long long)OCFS2_I(tl_inode)->ip_blkno, index,
5846 		start_cluster, num_clusters);
5847 	if (ocfs2_truncate_log_can_coalesce(tl, start_cluster)) {
5848 		/*
5849 		 * Move index back to the record we are coalescing with.
5850 		 * ocfs2_truncate_log_can_coalesce() guarantees nonzero
5851 		 */
5852 		index--;
5853 
5854 		num_clusters += le32_to_cpu(tl->tl_recs[index].t_clusters);
5855 		trace_ocfs2_truncate_log_append(
5856 			(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5857 			index, le32_to_cpu(tl->tl_recs[index].t_start),
5858 			num_clusters);
5859 	} else {
5860 		tl->tl_recs[index].t_start = cpu_to_le32(start_cluster);
5861 		tl->tl_used = cpu_to_le16(index + 1);
5862 	}
5863 	tl->tl_recs[index].t_clusters = cpu_to_le32(num_clusters);
5864 
5865 	ocfs2_journal_dirty(handle, tl_bh);
5866 
5867 	osb->truncated_clusters += num_clusters;
5868 bail:
5869 	return status;
5870 }
5871 
5872 static int ocfs2_replay_truncate_records(struct ocfs2_super *osb,
5873 					 handle_t *handle,
5874 					 struct inode *data_alloc_inode,
5875 					 struct buffer_head *data_alloc_bh)
5876 {
5877 	int status = 0;
5878 	int i;
5879 	unsigned int num_clusters;
5880 	u64 start_blk;
5881 	struct ocfs2_truncate_rec rec;
5882 	struct ocfs2_dinode *di;
5883 	struct ocfs2_truncate_log *tl;
5884 	struct inode *tl_inode = osb->osb_tl_inode;
5885 	struct buffer_head *tl_bh = osb->osb_tl_bh;
5886 
5887 	di = (struct ocfs2_dinode *) tl_bh->b_data;
5888 	tl = &di->id2.i_dealloc;
5889 	i = le16_to_cpu(tl->tl_used) - 1;
5890 	while (i >= 0) {
5891 		/* Caller has given us at least enough credits to
5892 		 * update the truncate log dinode */
5893 		status = ocfs2_journal_access_di(handle, INODE_CACHE(tl_inode), tl_bh,
5894 						 OCFS2_JOURNAL_ACCESS_WRITE);
5895 		if (status < 0) {
5896 			mlog_errno(status);
5897 			goto bail;
5898 		}
5899 
5900 		tl->tl_used = cpu_to_le16(i);
5901 
5902 		ocfs2_journal_dirty(handle, tl_bh);
5903 
5904 		/* TODO: Perhaps we can calculate the bulk of the
5905 		 * credits up front rather than extending like
5906 		 * this. */
5907 		status = ocfs2_extend_trans(handle,
5908 					    OCFS2_TRUNCATE_LOG_FLUSH_ONE_REC);
5909 		if (status < 0) {
5910 			mlog_errno(status);
5911 			goto bail;
5912 		}
5913 
5914 		rec = tl->tl_recs[i];
5915 		start_blk = ocfs2_clusters_to_blocks(data_alloc_inode->i_sb,
5916 						    le32_to_cpu(rec.t_start));
5917 		num_clusters = le32_to_cpu(rec.t_clusters);
5918 
5919 		/* if start_blk is not set, we ignore the record as
5920 		 * invalid. */
5921 		if (start_blk) {
5922 			trace_ocfs2_replay_truncate_records(
5923 				(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5924 				i, le32_to_cpu(rec.t_start), num_clusters);
5925 
5926 			status = ocfs2_free_clusters(handle, data_alloc_inode,
5927 						     data_alloc_bh, start_blk,
5928 						     num_clusters);
5929 			if (status < 0) {
5930 				mlog_errno(status);
5931 				goto bail;
5932 			}
5933 		}
5934 		i--;
5935 	}
5936 
5937 	osb->truncated_clusters = 0;
5938 
5939 bail:
5940 	return status;
5941 }
5942 
5943 /* Expects you to already be holding tl_inode->i_mutex */
5944 int __ocfs2_flush_truncate_log(struct ocfs2_super *osb)
5945 {
5946 	int status;
5947 	unsigned int num_to_flush;
5948 	handle_t *handle;
5949 	struct inode *tl_inode = osb->osb_tl_inode;
5950 	struct inode *data_alloc_inode = NULL;
5951 	struct buffer_head *tl_bh = osb->osb_tl_bh;
5952 	struct buffer_head *data_alloc_bh = NULL;
5953 	struct ocfs2_dinode *di;
5954 	struct ocfs2_truncate_log *tl;
5955 
5956 	BUG_ON(mutex_trylock(&tl_inode->i_mutex));
5957 
5958 	di = (struct ocfs2_dinode *) tl_bh->b_data;
5959 
5960 	/* tl_bh is loaded from ocfs2_truncate_log_init().  It's validated
5961 	 * by the underlying call to ocfs2_read_inode_block(), so any
5962 	 * corruption is a code bug */
5963 	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
5964 
5965 	tl = &di->id2.i_dealloc;
5966 	num_to_flush = le16_to_cpu(tl->tl_used);
5967 	trace_ocfs2_flush_truncate_log(
5968 		(unsigned long long)OCFS2_I(tl_inode)->ip_blkno,
5969 		num_to_flush);
5970 	if (!num_to_flush) {
5971 		status = 0;
5972 		goto out;
5973 	}
5974 
5975 	data_alloc_inode = ocfs2_get_system_file_inode(osb,
5976 						       GLOBAL_BITMAP_SYSTEM_INODE,
5977 						       OCFS2_INVALID_SLOT);
5978 	if (!data_alloc_inode) {
5979 		status = -EINVAL;
5980 		mlog(ML_ERROR, "Could not get bitmap inode!\n");
5981 		goto out;
5982 	}
5983 
5984 	mutex_lock(&data_alloc_inode->i_mutex);
5985 
5986 	status = ocfs2_inode_lock(data_alloc_inode, &data_alloc_bh, 1);
5987 	if (status < 0) {
5988 		mlog_errno(status);
5989 		goto out_mutex;
5990 	}
5991 
5992 	handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
5993 	if (IS_ERR(handle)) {
5994 		status = PTR_ERR(handle);
5995 		mlog_errno(status);
5996 		goto out_unlock;
5997 	}
5998 
5999 	status = ocfs2_replay_truncate_records(osb, handle, data_alloc_inode,
6000 					       data_alloc_bh);
6001 	if (status < 0)
6002 		mlog_errno(status);
6003 
6004 	ocfs2_commit_trans(osb, handle);
6005 
6006 out_unlock:
6007 	brelse(data_alloc_bh);
6008 	ocfs2_inode_unlock(data_alloc_inode, 1);
6009 
6010 out_mutex:
6011 	mutex_unlock(&data_alloc_inode->i_mutex);
6012 	iput(data_alloc_inode);
6013 
6014 out:
6015 	return status;
6016 }
6017 
6018 int ocfs2_flush_truncate_log(struct ocfs2_super *osb)
6019 {
6020 	int status;
6021 	struct inode *tl_inode = osb->osb_tl_inode;
6022 
6023 	mutex_lock(&tl_inode->i_mutex);
6024 	status = __ocfs2_flush_truncate_log(osb);
6025 	mutex_unlock(&tl_inode->i_mutex);
6026 
6027 	return status;
6028 }
6029 
6030 static void ocfs2_truncate_log_worker(struct work_struct *work)
6031 {
6032 	int status;
6033 	struct ocfs2_super *osb =
6034 		container_of(work, struct ocfs2_super,
6035 			     osb_truncate_log_wq.work);
6036 
6037 	status = ocfs2_flush_truncate_log(osb);
6038 	if (status < 0)
6039 		mlog_errno(status);
6040 	else
6041 		ocfs2_init_steal_slots(osb);
6042 }
6043 
6044 #define OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL (2 * HZ)
6045 void ocfs2_schedule_truncate_log_flush(struct ocfs2_super *osb,
6046 				       int cancel)
6047 {
6048 	if (osb->osb_tl_inode) {
6049 		/* We want to push off log flushes while truncates are
6050 		 * still running. */
6051 		if (cancel)
6052 			cancel_delayed_work(&osb->osb_truncate_log_wq);
6053 
6054 		queue_delayed_work(ocfs2_wq, &osb->osb_truncate_log_wq,
6055 				   OCFS2_TRUNCATE_LOG_FLUSH_INTERVAL);
6056 	}
6057 }
6058 
6059 static int ocfs2_get_truncate_log_info(struct ocfs2_super *osb,
6060 				       int slot_num,
6061 				       struct inode **tl_inode,
6062 				       struct buffer_head **tl_bh)
6063 {
6064 	int status;
6065 	struct inode *inode = NULL;
6066 	struct buffer_head *bh = NULL;
6067 
6068 	inode = ocfs2_get_system_file_inode(osb,
6069 					   TRUNCATE_LOG_SYSTEM_INODE,
6070 					   slot_num);
6071 	if (!inode) {
6072 		status = -EINVAL;
6073 		mlog(ML_ERROR, "Could not get load truncate log inode!\n");
6074 		goto bail;
6075 	}
6076 
6077 	status = ocfs2_read_inode_block(inode, &bh);
6078 	if (status < 0) {
6079 		iput(inode);
6080 		mlog_errno(status);
6081 		goto bail;
6082 	}
6083 
6084 	*tl_inode = inode;
6085 	*tl_bh    = bh;
6086 bail:
6087 	return status;
6088 }
6089 
6090 /* called during the 1st stage of node recovery. we stamp a clean
6091  * truncate log and pass back a copy for processing later. if the
6092  * truncate log does not require processing, a *tl_copy is set to
6093  * NULL. */
6094 int ocfs2_begin_truncate_log_recovery(struct ocfs2_super *osb,
6095 				      int slot_num,
6096 				      struct ocfs2_dinode **tl_copy)
6097 {
6098 	int status;
6099 	struct inode *tl_inode = NULL;
6100 	struct buffer_head *tl_bh = NULL;
6101 	struct ocfs2_dinode *di;
6102 	struct ocfs2_truncate_log *tl;
6103 
6104 	*tl_copy = NULL;
6105 
6106 	trace_ocfs2_begin_truncate_log_recovery(slot_num);
6107 
6108 	status = ocfs2_get_truncate_log_info(osb, slot_num, &tl_inode, &tl_bh);
6109 	if (status < 0) {
6110 		mlog_errno(status);
6111 		goto bail;
6112 	}
6113 
6114 	di = (struct ocfs2_dinode *) tl_bh->b_data;
6115 
6116 	/* tl_bh is loaded from ocfs2_get_truncate_log_info().  It's
6117 	 * validated by the underlying call to ocfs2_read_inode_block(),
6118 	 * so any corruption is a code bug */
6119 	BUG_ON(!OCFS2_IS_VALID_DINODE(di));
6120 
6121 	tl = &di->id2.i_dealloc;
6122 	if (le16_to_cpu(tl->tl_used)) {
6123 		trace_ocfs2_truncate_log_recovery_num(le16_to_cpu(tl->tl_used));
6124 
6125 		*tl_copy = kmalloc(tl_bh->b_size, GFP_KERNEL);
6126 		if (!(*tl_copy)) {
6127 			status = -ENOMEM;
6128 			mlog_errno(status);
6129 			goto bail;
6130 		}
6131 
6132 		/* Assuming the write-out below goes well, this copy
6133 		 * will be passed back to recovery for processing. */
6134 		memcpy(*tl_copy, tl_bh->b_data, tl_bh->b_size);
6135 
6136 		/* All we need to do to clear the truncate log is set
6137 		 * tl_used. */
6138 		tl->tl_used = 0;
6139 
6140 		ocfs2_compute_meta_ecc(osb->sb, tl_bh->b_data, &di->i_check);
6141 		status = ocfs2_write_block(osb, tl_bh, INODE_CACHE(tl_inode));
6142 		if (status < 0) {
6143 			mlog_errno(status);
6144 			goto bail;
6145 		}
6146 	}
6147 
6148 bail:
6149 	if (tl_inode)
6150 		iput(tl_inode);
6151 	brelse(tl_bh);
6152 
6153 	if (status < 0 && (*tl_copy)) {
6154 		kfree(*tl_copy);
6155 		*tl_copy = NULL;
6156 		mlog_errno(status);
6157 	}
6158 
6159 	return status;
6160 }
6161 
6162 int ocfs2_complete_truncate_log_recovery(struct ocfs2_super *osb,
6163 					 struct ocfs2_dinode *tl_copy)
6164 {
6165 	int status = 0;
6166 	int i;
6167 	unsigned int clusters, num_recs, start_cluster;
6168 	u64 start_blk;
6169 	handle_t *handle;
6170 	struct inode *tl_inode = osb->osb_tl_inode;
6171 	struct ocfs2_truncate_log *tl;
6172 
6173 	if (OCFS2_I(tl_inode)->ip_blkno == le64_to_cpu(tl_copy->i_blkno)) {
6174 		mlog(ML_ERROR, "Asked to recover my own truncate log!\n");
6175 		return -EINVAL;
6176 	}
6177 
6178 	tl = &tl_copy->id2.i_dealloc;
6179 	num_recs = le16_to_cpu(tl->tl_used);
6180 	trace_ocfs2_complete_truncate_log_recovery(
6181 		(unsigned long long)le64_to_cpu(tl_copy->i_blkno),
6182 		num_recs);
6183 
6184 	mutex_lock(&tl_inode->i_mutex);
6185 	for(i = 0; i < num_recs; i++) {
6186 		if (ocfs2_truncate_log_needs_flush(osb)) {
6187 			status = __ocfs2_flush_truncate_log(osb);
6188 			if (status < 0) {
6189 				mlog_errno(status);
6190 				goto bail_up;
6191 			}
6192 		}
6193 
6194 		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6195 		if (IS_ERR(handle)) {
6196 			status = PTR_ERR(handle);
6197 			mlog_errno(status);
6198 			goto bail_up;
6199 		}
6200 
6201 		clusters = le32_to_cpu(tl->tl_recs[i].t_clusters);
6202 		start_cluster = le32_to_cpu(tl->tl_recs[i].t_start);
6203 		start_blk = ocfs2_clusters_to_blocks(osb->sb, start_cluster);
6204 
6205 		status = ocfs2_truncate_log_append(osb, handle,
6206 						   start_blk, clusters);
6207 		ocfs2_commit_trans(osb, handle);
6208 		if (status < 0) {
6209 			mlog_errno(status);
6210 			goto bail_up;
6211 		}
6212 	}
6213 
6214 bail_up:
6215 	mutex_unlock(&tl_inode->i_mutex);
6216 
6217 	return status;
6218 }
6219 
6220 void ocfs2_truncate_log_shutdown(struct ocfs2_super *osb)
6221 {
6222 	int status;
6223 	struct inode *tl_inode = osb->osb_tl_inode;
6224 
6225 	if (tl_inode) {
6226 		cancel_delayed_work(&osb->osb_truncate_log_wq);
6227 		flush_workqueue(ocfs2_wq);
6228 
6229 		status = ocfs2_flush_truncate_log(osb);
6230 		if (status < 0)
6231 			mlog_errno(status);
6232 
6233 		brelse(osb->osb_tl_bh);
6234 		iput(osb->osb_tl_inode);
6235 	}
6236 }
6237 
6238 int ocfs2_truncate_log_init(struct ocfs2_super *osb)
6239 {
6240 	int status;
6241 	struct inode *tl_inode = NULL;
6242 	struct buffer_head *tl_bh = NULL;
6243 
6244 	status = ocfs2_get_truncate_log_info(osb,
6245 					     osb->slot_num,
6246 					     &tl_inode,
6247 					     &tl_bh);
6248 	if (status < 0)
6249 		mlog_errno(status);
6250 
6251 	/* ocfs2_truncate_log_shutdown keys on the existence of
6252 	 * osb->osb_tl_inode so we don't set any of the osb variables
6253 	 * until we're sure all is well. */
6254 	INIT_DELAYED_WORK(&osb->osb_truncate_log_wq,
6255 			  ocfs2_truncate_log_worker);
6256 	osb->osb_tl_bh    = tl_bh;
6257 	osb->osb_tl_inode = tl_inode;
6258 
6259 	return status;
6260 }
6261 
6262 /*
6263  * Delayed de-allocation of suballocator blocks.
6264  *
6265  * Some sets of block de-allocations might involve multiple suballocator inodes.
6266  *
6267  * The locking for this can get extremely complicated, especially when
6268  * the suballocator inodes to delete from aren't known until deep
6269  * within an unrelated codepath.
6270  *
6271  * ocfs2_extent_block structures are a good example of this - an inode
6272  * btree could have been grown by any number of nodes each allocating
6273  * out of their own suballoc inode.
6274  *
6275  * These structures allow the delay of block de-allocation until a
6276  * later time, when locking of multiple cluster inodes won't cause
6277  * deadlock.
6278  */
6279 
6280 /*
6281  * Describe a single bit freed from a suballocator.  For the block
6282  * suballocators, it represents one block.  For the global cluster
6283  * allocator, it represents some clusters and free_bit indicates
6284  * clusters number.
6285  */
6286 struct ocfs2_cached_block_free {
6287 	struct ocfs2_cached_block_free		*free_next;
6288 	u64					free_bg;
6289 	u64					free_blk;
6290 	unsigned int				free_bit;
6291 };
6292 
6293 struct ocfs2_per_slot_free_list {
6294 	struct ocfs2_per_slot_free_list		*f_next_suballocator;
6295 	int					f_inode_type;
6296 	int					f_slot;
6297 	struct ocfs2_cached_block_free		*f_first;
6298 };
6299 
6300 static int ocfs2_free_cached_blocks(struct ocfs2_super *osb,
6301 				    int sysfile_type,
6302 				    int slot,
6303 				    struct ocfs2_cached_block_free *head)
6304 {
6305 	int ret;
6306 	u64 bg_blkno;
6307 	handle_t *handle;
6308 	struct inode *inode;
6309 	struct buffer_head *di_bh = NULL;
6310 	struct ocfs2_cached_block_free *tmp;
6311 
6312 	inode = ocfs2_get_system_file_inode(osb, sysfile_type, slot);
6313 	if (!inode) {
6314 		ret = -EINVAL;
6315 		mlog_errno(ret);
6316 		goto out;
6317 	}
6318 
6319 	mutex_lock(&inode->i_mutex);
6320 
6321 	ret = ocfs2_inode_lock(inode, &di_bh, 1);
6322 	if (ret) {
6323 		mlog_errno(ret);
6324 		goto out_mutex;
6325 	}
6326 
6327 	handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
6328 	if (IS_ERR(handle)) {
6329 		ret = PTR_ERR(handle);
6330 		mlog_errno(ret);
6331 		goto out_unlock;
6332 	}
6333 
6334 	while (head) {
6335 		if (head->free_bg)
6336 			bg_blkno = head->free_bg;
6337 		else
6338 			bg_blkno = ocfs2_which_suballoc_group(head->free_blk,
6339 							      head->free_bit);
6340 		trace_ocfs2_free_cached_blocks(
6341 		     (unsigned long long)head->free_blk, head->free_bit);
6342 
6343 		ret = ocfs2_free_suballoc_bits(handle, inode, di_bh,
6344 					       head->free_bit, bg_blkno, 1);
6345 		if (ret) {
6346 			mlog_errno(ret);
6347 			goto out_journal;
6348 		}
6349 
6350 		ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
6351 		if (ret) {
6352 			mlog_errno(ret);
6353 			goto out_journal;
6354 		}
6355 
6356 		tmp = head;
6357 		head = head->free_next;
6358 		kfree(tmp);
6359 	}
6360 
6361 out_journal:
6362 	ocfs2_commit_trans(osb, handle);
6363 
6364 out_unlock:
6365 	ocfs2_inode_unlock(inode, 1);
6366 	brelse(di_bh);
6367 out_mutex:
6368 	mutex_unlock(&inode->i_mutex);
6369 	iput(inode);
6370 out:
6371 	while(head) {
6372 		/* Premature exit may have left some dangling items. */
6373 		tmp = head;
6374 		head = head->free_next;
6375 		kfree(tmp);
6376 	}
6377 
6378 	return ret;
6379 }
6380 
6381 int ocfs2_cache_cluster_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6382 				u64 blkno, unsigned int bit)
6383 {
6384 	int ret = 0;
6385 	struct ocfs2_cached_block_free *item;
6386 
6387 	item = kzalloc(sizeof(*item), GFP_NOFS);
6388 	if (item == NULL) {
6389 		ret = -ENOMEM;
6390 		mlog_errno(ret);
6391 		return ret;
6392 	}
6393 
6394 	trace_ocfs2_cache_cluster_dealloc((unsigned long long)blkno, bit);
6395 
6396 	item->free_blk = blkno;
6397 	item->free_bit = bit;
6398 	item->free_next = ctxt->c_global_allocator;
6399 
6400 	ctxt->c_global_allocator = item;
6401 	return ret;
6402 }
6403 
6404 static int ocfs2_free_cached_clusters(struct ocfs2_super *osb,
6405 				      struct ocfs2_cached_block_free *head)
6406 {
6407 	struct ocfs2_cached_block_free *tmp;
6408 	struct inode *tl_inode = osb->osb_tl_inode;
6409 	handle_t *handle;
6410 	int ret = 0;
6411 
6412 	mutex_lock(&tl_inode->i_mutex);
6413 
6414 	while (head) {
6415 		if (ocfs2_truncate_log_needs_flush(osb)) {
6416 			ret = __ocfs2_flush_truncate_log(osb);
6417 			if (ret < 0) {
6418 				mlog_errno(ret);
6419 				break;
6420 			}
6421 		}
6422 
6423 		handle = ocfs2_start_trans(osb, OCFS2_TRUNCATE_LOG_UPDATE);
6424 		if (IS_ERR(handle)) {
6425 			ret = PTR_ERR(handle);
6426 			mlog_errno(ret);
6427 			break;
6428 		}
6429 
6430 		ret = ocfs2_truncate_log_append(osb, handle, head->free_blk,
6431 						head->free_bit);
6432 
6433 		ocfs2_commit_trans(osb, handle);
6434 		tmp = head;
6435 		head = head->free_next;
6436 		kfree(tmp);
6437 
6438 		if (ret < 0) {
6439 			mlog_errno(ret);
6440 			break;
6441 		}
6442 	}
6443 
6444 	mutex_unlock(&tl_inode->i_mutex);
6445 
6446 	while (head) {
6447 		/* Premature exit may have left some dangling items. */
6448 		tmp = head;
6449 		head = head->free_next;
6450 		kfree(tmp);
6451 	}
6452 
6453 	return ret;
6454 }
6455 
6456 int ocfs2_run_deallocs(struct ocfs2_super *osb,
6457 		       struct ocfs2_cached_dealloc_ctxt *ctxt)
6458 {
6459 	int ret = 0, ret2;
6460 	struct ocfs2_per_slot_free_list *fl;
6461 
6462 	if (!ctxt)
6463 		return 0;
6464 
6465 	while (ctxt->c_first_suballocator) {
6466 		fl = ctxt->c_first_suballocator;
6467 
6468 		if (fl->f_first) {
6469 			trace_ocfs2_run_deallocs(fl->f_inode_type,
6470 						 fl->f_slot);
6471 			ret2 = ocfs2_free_cached_blocks(osb,
6472 							fl->f_inode_type,
6473 							fl->f_slot,
6474 							fl->f_first);
6475 			if (ret2)
6476 				mlog_errno(ret2);
6477 			if (!ret)
6478 				ret = ret2;
6479 		}
6480 
6481 		ctxt->c_first_suballocator = fl->f_next_suballocator;
6482 		kfree(fl);
6483 	}
6484 
6485 	if (ctxt->c_global_allocator) {
6486 		ret2 = ocfs2_free_cached_clusters(osb,
6487 						  ctxt->c_global_allocator);
6488 		if (ret2)
6489 			mlog_errno(ret2);
6490 		if (!ret)
6491 			ret = ret2;
6492 
6493 		ctxt->c_global_allocator = NULL;
6494 	}
6495 
6496 	return ret;
6497 }
6498 
6499 static struct ocfs2_per_slot_free_list *
6500 ocfs2_find_per_slot_free_list(int type,
6501 			      int slot,
6502 			      struct ocfs2_cached_dealloc_ctxt *ctxt)
6503 {
6504 	struct ocfs2_per_slot_free_list *fl = ctxt->c_first_suballocator;
6505 
6506 	while (fl) {
6507 		if (fl->f_inode_type == type && fl->f_slot == slot)
6508 			return fl;
6509 
6510 		fl = fl->f_next_suballocator;
6511 	}
6512 
6513 	fl = kmalloc(sizeof(*fl), GFP_NOFS);
6514 	if (fl) {
6515 		fl->f_inode_type = type;
6516 		fl->f_slot = slot;
6517 		fl->f_first = NULL;
6518 		fl->f_next_suballocator = ctxt->c_first_suballocator;
6519 
6520 		ctxt->c_first_suballocator = fl;
6521 	}
6522 	return fl;
6523 }
6524 
6525 int ocfs2_cache_block_dealloc(struct ocfs2_cached_dealloc_ctxt *ctxt,
6526 			      int type, int slot, u64 suballoc,
6527 			      u64 blkno, unsigned int bit)
6528 {
6529 	int ret;
6530 	struct ocfs2_per_slot_free_list *fl;
6531 	struct ocfs2_cached_block_free *item;
6532 
6533 	fl = ocfs2_find_per_slot_free_list(type, slot, ctxt);
6534 	if (fl == NULL) {
6535 		ret = -ENOMEM;
6536 		mlog_errno(ret);
6537 		goto out;
6538 	}
6539 
6540 	item = kzalloc(sizeof(*item), GFP_NOFS);
6541 	if (item == NULL) {
6542 		ret = -ENOMEM;
6543 		mlog_errno(ret);
6544 		goto out;
6545 	}
6546 
6547 	trace_ocfs2_cache_block_dealloc(type, slot,
6548 					(unsigned long long)suballoc,
6549 					(unsigned long long)blkno, bit);
6550 
6551 	item->free_bg = suballoc;
6552 	item->free_blk = blkno;
6553 	item->free_bit = bit;
6554 	item->free_next = fl->f_first;
6555 
6556 	fl->f_first = item;
6557 
6558 	ret = 0;
6559 out:
6560 	return ret;
6561 }
6562 
6563 static int ocfs2_cache_extent_block_free(struct ocfs2_cached_dealloc_ctxt *ctxt,
6564 					 struct ocfs2_extent_block *eb)
6565 {
6566 	return ocfs2_cache_block_dealloc(ctxt, EXTENT_ALLOC_SYSTEM_INODE,
6567 					 le16_to_cpu(eb->h_suballoc_slot),
6568 					 le64_to_cpu(eb->h_suballoc_loc),
6569 					 le64_to_cpu(eb->h_blkno),
6570 					 le16_to_cpu(eb->h_suballoc_bit));
6571 }
6572 
6573 static int ocfs2_zero_func(handle_t *handle, struct buffer_head *bh)
6574 {
6575 	set_buffer_uptodate(bh);
6576 	mark_buffer_dirty(bh);
6577 	return 0;
6578 }
6579 
6580 void ocfs2_map_and_dirty_page(struct inode *inode, handle_t *handle,
6581 			      unsigned int from, unsigned int to,
6582 			      struct page *page, int zero, u64 *phys)
6583 {
6584 	int ret, partial = 0;
6585 
6586 	ret = ocfs2_map_page_blocks(page, phys, inode, from, to, 0);
6587 	if (ret)
6588 		mlog_errno(ret);
6589 
6590 	if (zero)
6591 		zero_user_segment(page, from, to);
6592 
6593 	/*
6594 	 * Need to set the buffers we zero'd into uptodate
6595 	 * here if they aren't - ocfs2_map_page_blocks()
6596 	 * might've skipped some
6597 	 */
6598 	ret = walk_page_buffers(handle, page_buffers(page),
6599 				from, to, &partial,
6600 				ocfs2_zero_func);
6601 	if (ret < 0)
6602 		mlog_errno(ret);
6603 	else if (ocfs2_should_order_data(inode)) {
6604 		ret = ocfs2_jbd2_file_inode(handle, inode);
6605 		if (ret < 0)
6606 			mlog_errno(ret);
6607 	}
6608 
6609 	if (!partial)
6610 		SetPageUptodate(page);
6611 
6612 	flush_dcache_page(page);
6613 }
6614 
6615 static void ocfs2_zero_cluster_pages(struct inode *inode, loff_t start,
6616 				     loff_t end, struct page **pages,
6617 				     int numpages, u64 phys, handle_t *handle)
6618 {
6619 	int i;
6620 	struct page *page;
6621 	unsigned int from, to = PAGE_CACHE_SIZE;
6622 	struct super_block *sb = inode->i_sb;
6623 
6624 	BUG_ON(!ocfs2_sparse_alloc(OCFS2_SB(sb)));
6625 
6626 	if (numpages == 0)
6627 		goto out;
6628 
6629 	to = PAGE_CACHE_SIZE;
6630 	for(i = 0; i < numpages; i++) {
6631 		page = pages[i];
6632 
6633 		from = start & (PAGE_CACHE_SIZE - 1);
6634 		if ((end >> PAGE_CACHE_SHIFT) == page->index)
6635 			to = end & (PAGE_CACHE_SIZE - 1);
6636 
6637 		BUG_ON(from > PAGE_CACHE_SIZE);
6638 		BUG_ON(to > PAGE_CACHE_SIZE);
6639 
6640 		ocfs2_map_and_dirty_page(inode, handle, from, to, page, 1,
6641 					 &phys);
6642 
6643 		start = (page->index + 1) << PAGE_CACHE_SHIFT;
6644 	}
6645 out:
6646 	if (pages)
6647 		ocfs2_unlock_and_free_pages(pages, numpages);
6648 }
6649 
6650 int ocfs2_grab_pages(struct inode *inode, loff_t start, loff_t end,
6651 		     struct page **pages, int *num)
6652 {
6653 	int numpages, ret = 0;
6654 	struct address_space *mapping = inode->i_mapping;
6655 	unsigned long index;
6656 	loff_t last_page_bytes;
6657 
6658 	BUG_ON(start > end);
6659 
6660 	numpages = 0;
6661 	last_page_bytes = PAGE_ALIGN(end);
6662 	index = start >> PAGE_CACHE_SHIFT;
6663 	do {
6664 		pages[numpages] = find_or_create_page(mapping, index, GFP_NOFS);
6665 		if (!pages[numpages]) {
6666 			ret = -ENOMEM;
6667 			mlog_errno(ret);
6668 			goto out;
6669 		}
6670 
6671 		numpages++;
6672 		index++;
6673 	} while (index < (last_page_bytes >> PAGE_CACHE_SHIFT));
6674 
6675 out:
6676 	if (ret != 0) {
6677 		if (pages)
6678 			ocfs2_unlock_and_free_pages(pages, numpages);
6679 		numpages = 0;
6680 	}
6681 
6682 	*num = numpages;
6683 
6684 	return ret;
6685 }
6686 
6687 static int ocfs2_grab_eof_pages(struct inode *inode, loff_t start, loff_t end,
6688 				struct page **pages, int *num)
6689 {
6690 	struct super_block *sb = inode->i_sb;
6691 
6692 	BUG_ON(start >> OCFS2_SB(sb)->s_clustersize_bits !=
6693 	       (end - 1) >> OCFS2_SB(sb)->s_clustersize_bits);
6694 
6695 	return ocfs2_grab_pages(inode, start, end, pages, num);
6696 }
6697 
6698 /*
6699  * Zero the area past i_size but still within an allocated
6700  * cluster. This avoids exposing nonzero data on subsequent file
6701  * extends.
6702  *
6703  * We need to call this before i_size is updated on the inode because
6704  * otherwise block_write_full_page() will skip writeout of pages past
6705  * i_size. The new_i_size parameter is passed for this reason.
6706  */
6707 int ocfs2_zero_range_for_truncate(struct inode *inode, handle_t *handle,
6708 				  u64 range_start, u64 range_end)
6709 {
6710 	int ret = 0, numpages;
6711 	struct page **pages = NULL;
6712 	u64 phys;
6713 	unsigned int ext_flags;
6714 	struct super_block *sb = inode->i_sb;
6715 
6716 	/*
6717 	 * File systems which don't support sparse files zero on every
6718 	 * extend.
6719 	 */
6720 	if (!ocfs2_sparse_alloc(OCFS2_SB(sb)))
6721 		return 0;
6722 
6723 	pages = kcalloc(ocfs2_pages_per_cluster(sb),
6724 			sizeof(struct page *), GFP_NOFS);
6725 	if (pages == NULL) {
6726 		ret = -ENOMEM;
6727 		mlog_errno(ret);
6728 		goto out;
6729 	}
6730 
6731 	if (range_start == range_end)
6732 		goto out;
6733 
6734 	ret = ocfs2_extent_map_get_blocks(inode,
6735 					  range_start >> sb->s_blocksize_bits,
6736 					  &phys, NULL, &ext_flags);
6737 	if (ret) {
6738 		mlog_errno(ret);
6739 		goto out;
6740 	}
6741 
6742 	/*
6743 	 * Tail is a hole, or is marked unwritten. In either case, we
6744 	 * can count on read and write to return/push zero's.
6745 	 */
6746 	if (phys == 0 || ext_flags & OCFS2_EXT_UNWRITTEN)
6747 		goto out;
6748 
6749 	ret = ocfs2_grab_eof_pages(inode, range_start, range_end, pages,
6750 				   &numpages);
6751 	if (ret) {
6752 		mlog_errno(ret);
6753 		goto out;
6754 	}
6755 
6756 	ocfs2_zero_cluster_pages(inode, range_start, range_end, pages,
6757 				 numpages, phys, handle);
6758 
6759 	/*
6760 	 * Initiate writeout of the pages we zero'd here. We don't
6761 	 * wait on them - the truncate_inode_pages() call later will
6762 	 * do that for us.
6763 	 */
6764 	ret = filemap_fdatawrite_range(inode->i_mapping, range_start,
6765 				       range_end - 1);
6766 	if (ret)
6767 		mlog_errno(ret);
6768 
6769 out:
6770 	kfree(pages);
6771 
6772 	return ret;
6773 }
6774 
6775 static void ocfs2_zero_dinode_id2_with_xattr(struct inode *inode,
6776 					     struct ocfs2_dinode *di)
6777 {
6778 	unsigned int blocksize = 1 << inode->i_sb->s_blocksize_bits;
6779 	unsigned int xattrsize = le16_to_cpu(di->i_xattr_inline_size);
6780 
6781 	if (le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_XATTR_FL)
6782 		memset(&di->id2, 0, blocksize -
6783 				    offsetof(struct ocfs2_dinode, id2) -
6784 				    xattrsize);
6785 	else
6786 		memset(&di->id2, 0, blocksize -
6787 				    offsetof(struct ocfs2_dinode, id2));
6788 }
6789 
6790 void ocfs2_dinode_new_extent_list(struct inode *inode,
6791 				  struct ocfs2_dinode *di)
6792 {
6793 	ocfs2_zero_dinode_id2_with_xattr(inode, di);
6794 	di->id2.i_list.l_tree_depth = 0;
6795 	di->id2.i_list.l_next_free_rec = 0;
6796 	di->id2.i_list.l_count = cpu_to_le16(
6797 		ocfs2_extent_recs_per_inode_with_xattr(inode->i_sb, di));
6798 }
6799 
6800 void ocfs2_set_inode_data_inline(struct inode *inode, struct ocfs2_dinode *di)
6801 {
6802 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
6803 	struct ocfs2_inline_data *idata = &di->id2.i_data;
6804 
6805 	spin_lock(&oi->ip_lock);
6806 	oi->ip_dyn_features |= OCFS2_INLINE_DATA_FL;
6807 	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6808 	spin_unlock(&oi->ip_lock);
6809 
6810 	/*
6811 	 * We clear the entire i_data structure here so that all
6812 	 * fields can be properly initialized.
6813 	 */
6814 	ocfs2_zero_dinode_id2_with_xattr(inode, di);
6815 
6816 	idata->id_count = cpu_to_le16(
6817 			ocfs2_max_inline_data_with_xattr(inode->i_sb, di));
6818 }
6819 
6820 int ocfs2_convert_inline_data_to_extents(struct inode *inode,
6821 					 struct buffer_head *di_bh)
6822 {
6823 	int ret, i, has_data, num_pages = 0;
6824 	int need_free = 0;
6825 	u32 bit_off, num;
6826 	handle_t *handle;
6827 	u64 uninitialized_var(block);
6828 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
6829 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6830 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
6831 	struct ocfs2_alloc_context *data_ac = NULL;
6832 	struct page **pages = NULL;
6833 	loff_t end = osb->s_clustersize;
6834 	struct ocfs2_extent_tree et;
6835 	int did_quota = 0;
6836 
6837 	has_data = i_size_read(inode) ? 1 : 0;
6838 
6839 	if (has_data) {
6840 		pages = kcalloc(ocfs2_pages_per_cluster(osb->sb),
6841 				sizeof(struct page *), GFP_NOFS);
6842 		if (pages == NULL) {
6843 			ret = -ENOMEM;
6844 			mlog_errno(ret);
6845 			goto out;
6846 		}
6847 
6848 		ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
6849 		if (ret) {
6850 			mlog_errno(ret);
6851 			goto out;
6852 		}
6853 	}
6854 
6855 	handle = ocfs2_start_trans(osb,
6856 				   ocfs2_inline_to_extents_credits(osb->sb));
6857 	if (IS_ERR(handle)) {
6858 		ret = PTR_ERR(handle);
6859 		mlog_errno(ret);
6860 		goto out_unlock;
6861 	}
6862 
6863 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
6864 				      OCFS2_JOURNAL_ACCESS_WRITE);
6865 	if (ret) {
6866 		mlog_errno(ret);
6867 		goto out_commit;
6868 	}
6869 
6870 	if (has_data) {
6871 		unsigned int page_end;
6872 		u64 phys;
6873 
6874 		ret = dquot_alloc_space_nodirty(inode,
6875 				       ocfs2_clusters_to_bytes(osb->sb, 1));
6876 		if (ret)
6877 			goto out_commit;
6878 		did_quota = 1;
6879 
6880 		data_ac->ac_resv = &OCFS2_I(inode)->ip_la_data_resv;
6881 
6882 		ret = ocfs2_claim_clusters(handle, data_ac, 1, &bit_off,
6883 					   &num);
6884 		if (ret) {
6885 			mlog_errno(ret);
6886 			goto out_commit;
6887 		}
6888 
6889 		/*
6890 		 * Save two copies, one for insert, and one that can
6891 		 * be changed by ocfs2_map_and_dirty_page() below.
6892 		 */
6893 		block = phys = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
6894 
6895 		/*
6896 		 * Non sparse file systems zero on extend, so no need
6897 		 * to do that now.
6898 		 */
6899 		if (!ocfs2_sparse_alloc(osb) &&
6900 		    PAGE_CACHE_SIZE < osb->s_clustersize)
6901 			end = PAGE_CACHE_SIZE;
6902 
6903 		ret = ocfs2_grab_eof_pages(inode, 0, end, pages, &num_pages);
6904 		if (ret) {
6905 			mlog_errno(ret);
6906 			need_free = 1;
6907 			goto out_commit;
6908 		}
6909 
6910 		/*
6911 		 * This should populate the 1st page for us and mark
6912 		 * it up to date.
6913 		 */
6914 		ret = ocfs2_read_inline_data(inode, pages[0], di_bh);
6915 		if (ret) {
6916 			mlog_errno(ret);
6917 			need_free = 1;
6918 			goto out_commit;
6919 		}
6920 
6921 		page_end = PAGE_CACHE_SIZE;
6922 		if (PAGE_CACHE_SIZE > osb->s_clustersize)
6923 			page_end = osb->s_clustersize;
6924 
6925 		for (i = 0; i < num_pages; i++)
6926 			ocfs2_map_and_dirty_page(inode, handle, 0, page_end,
6927 						 pages[i], i > 0, &phys);
6928 	}
6929 
6930 	spin_lock(&oi->ip_lock);
6931 	oi->ip_dyn_features &= ~OCFS2_INLINE_DATA_FL;
6932 	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
6933 	spin_unlock(&oi->ip_lock);
6934 
6935 	ocfs2_dinode_new_extent_list(inode, di);
6936 
6937 	ocfs2_journal_dirty(handle, di_bh);
6938 
6939 	if (has_data) {
6940 		/*
6941 		 * An error at this point should be extremely rare. If
6942 		 * this proves to be false, we could always re-build
6943 		 * the in-inode data from our pages.
6944 		 */
6945 		ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
6946 		ret = ocfs2_insert_extent(handle, &et, 0, block, 1, 0, NULL);
6947 		if (ret) {
6948 			mlog_errno(ret);
6949 			need_free = 1;
6950 			goto out_commit;
6951 		}
6952 
6953 		inode->i_blocks = ocfs2_inode_sector_count(inode);
6954 	}
6955 
6956 out_commit:
6957 	if (ret < 0 && did_quota)
6958 		dquot_free_space_nodirty(inode,
6959 					  ocfs2_clusters_to_bytes(osb->sb, 1));
6960 
6961 	if (need_free) {
6962 		if (data_ac->ac_which == OCFS2_AC_USE_LOCAL)
6963 			ocfs2_free_local_alloc_bits(osb, handle, data_ac,
6964 					bit_off, num);
6965 		else
6966 			ocfs2_free_clusters(handle,
6967 					data_ac->ac_inode,
6968 					data_ac->ac_bh,
6969 					ocfs2_clusters_to_blocks(osb->sb, bit_off),
6970 					num);
6971 	}
6972 
6973 	ocfs2_commit_trans(osb, handle);
6974 
6975 out_unlock:
6976 	if (data_ac)
6977 		ocfs2_free_alloc_context(data_ac);
6978 
6979 out:
6980 	if (pages) {
6981 		ocfs2_unlock_and_free_pages(pages, num_pages);
6982 		kfree(pages);
6983 	}
6984 
6985 	return ret;
6986 }
6987 
6988 /*
6989  * It is expected, that by the time you call this function,
6990  * inode->i_size and fe->i_size have been adjusted.
6991  *
6992  * WARNING: This will kfree the truncate context
6993  */
6994 int ocfs2_commit_truncate(struct ocfs2_super *osb,
6995 			  struct inode *inode,
6996 			  struct buffer_head *di_bh)
6997 {
6998 	int status = 0, i, flags = 0;
6999 	u32 new_highest_cpos, range, trunc_cpos, trunc_len, phys_cpos, coff;
7000 	u64 blkno = 0;
7001 	struct ocfs2_extent_list *el;
7002 	struct ocfs2_extent_rec *rec;
7003 	struct ocfs2_path *path = NULL;
7004 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7005 	struct ocfs2_extent_list *root_el = &(di->id2.i_list);
7006 	u64 refcount_loc = le64_to_cpu(di->i_refcount_loc);
7007 	struct ocfs2_extent_tree et;
7008 	struct ocfs2_cached_dealloc_ctxt dealloc;
7009 
7010 	ocfs2_init_dinode_extent_tree(&et, INODE_CACHE(inode), di_bh);
7011 	ocfs2_init_dealloc_ctxt(&dealloc);
7012 
7013 	new_highest_cpos = ocfs2_clusters_for_bytes(osb->sb,
7014 						     i_size_read(inode));
7015 
7016 	path = ocfs2_new_path(di_bh, &di->id2.i_list,
7017 			      ocfs2_journal_access_di);
7018 	if (!path) {
7019 		status = -ENOMEM;
7020 		mlog_errno(status);
7021 		goto bail;
7022 	}
7023 
7024 	ocfs2_extent_map_trunc(inode, new_highest_cpos);
7025 
7026 start:
7027 	/*
7028 	 * Check that we still have allocation to delete.
7029 	 */
7030 	if (OCFS2_I(inode)->ip_clusters == 0) {
7031 		status = 0;
7032 		goto bail;
7033 	}
7034 
7035 	/*
7036 	 * Truncate always works against the rightmost tree branch.
7037 	 */
7038 	status = ocfs2_find_path(INODE_CACHE(inode), path, UINT_MAX);
7039 	if (status) {
7040 		mlog_errno(status);
7041 		goto bail;
7042 	}
7043 
7044 	trace_ocfs2_commit_truncate(
7045 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
7046 		new_highest_cpos,
7047 		OCFS2_I(inode)->ip_clusters,
7048 		path->p_tree_depth);
7049 
7050 	/*
7051 	 * By now, el will point to the extent list on the bottom most
7052 	 * portion of this tree. Only the tail record is considered in
7053 	 * each pass.
7054 	 *
7055 	 * We handle the following cases, in order:
7056 	 * - empty extent: delete the remaining branch
7057 	 * - remove the entire record
7058 	 * - remove a partial record
7059 	 * - no record needs to be removed (truncate has completed)
7060 	 */
7061 	el = path_leaf_el(path);
7062 	if (le16_to_cpu(el->l_next_free_rec) == 0) {
7063 		ocfs2_error(inode->i_sb,
7064 			    "Inode %llu has empty extent block at %llu\n",
7065 			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
7066 			    (unsigned long long)path_leaf_bh(path)->b_blocknr);
7067 		status = -EROFS;
7068 		goto bail;
7069 	}
7070 
7071 	i = le16_to_cpu(el->l_next_free_rec) - 1;
7072 	rec = &el->l_recs[i];
7073 	flags = rec->e_flags;
7074 	range = le32_to_cpu(rec->e_cpos) + ocfs2_rec_clusters(el, rec);
7075 
7076 	if (i == 0 && ocfs2_is_empty_extent(rec)) {
7077 		/*
7078 		 * Lower levels depend on this never happening, but it's best
7079 		 * to check it up here before changing the tree.
7080 		*/
7081 		if (root_el->l_tree_depth && rec->e_int_clusters == 0) {
7082 			ocfs2_error(inode->i_sb, "Inode %lu has an empty "
7083 				    "extent record, depth %u\n", inode->i_ino,
7084 				    le16_to_cpu(root_el->l_tree_depth));
7085 			status = -EROFS;
7086 			goto bail;
7087 		}
7088 		trunc_cpos = le32_to_cpu(rec->e_cpos);
7089 		trunc_len = 0;
7090 		blkno = 0;
7091 	} else if (le32_to_cpu(rec->e_cpos) >= new_highest_cpos) {
7092 		/*
7093 		 * Truncate entire record.
7094 		 */
7095 		trunc_cpos = le32_to_cpu(rec->e_cpos);
7096 		trunc_len = ocfs2_rec_clusters(el, rec);
7097 		blkno = le64_to_cpu(rec->e_blkno);
7098 	} else if (range > new_highest_cpos) {
7099 		/*
7100 		 * Partial truncate. it also should be
7101 		 * the last truncate we're doing.
7102 		 */
7103 		trunc_cpos = new_highest_cpos;
7104 		trunc_len = range - new_highest_cpos;
7105 		coff = new_highest_cpos - le32_to_cpu(rec->e_cpos);
7106 		blkno = le64_to_cpu(rec->e_blkno) +
7107 				ocfs2_clusters_to_blocks(inode->i_sb, coff);
7108 	} else {
7109 		/*
7110 		 * Truncate completed, leave happily.
7111 		 */
7112 		status = 0;
7113 		goto bail;
7114 	}
7115 
7116 	phys_cpos = ocfs2_blocks_to_clusters(inode->i_sb, blkno);
7117 
7118 	status = ocfs2_remove_btree_range(inode, &et, trunc_cpos,
7119 					  phys_cpos, trunc_len, flags, &dealloc,
7120 					  refcount_loc);
7121 	if (status < 0) {
7122 		mlog_errno(status);
7123 		goto bail;
7124 	}
7125 
7126 	ocfs2_reinit_path(path, 1);
7127 
7128 	/*
7129 	 * The check above will catch the case where we've truncated
7130 	 * away all allocation.
7131 	 */
7132 	goto start;
7133 
7134 bail:
7135 
7136 	ocfs2_schedule_truncate_log_flush(osb, 1);
7137 
7138 	ocfs2_run_deallocs(osb, &dealloc);
7139 
7140 	ocfs2_free_path(path);
7141 
7142 	return status;
7143 }
7144 
7145 /*
7146  * 'start' is inclusive, 'end' is not.
7147  */
7148 int ocfs2_truncate_inline(struct inode *inode, struct buffer_head *di_bh,
7149 			  unsigned int start, unsigned int end, int trunc)
7150 {
7151 	int ret;
7152 	unsigned int numbytes;
7153 	handle_t *handle;
7154 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
7155 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
7156 	struct ocfs2_inline_data *idata = &di->id2.i_data;
7157 
7158 	if (end > i_size_read(inode))
7159 		end = i_size_read(inode);
7160 
7161 	BUG_ON(start > end);
7162 
7163 	if (!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_DATA_FL) ||
7164 	    !(le16_to_cpu(di->i_dyn_features) & OCFS2_INLINE_DATA_FL) ||
7165 	    !ocfs2_supports_inline_data(osb)) {
7166 		ocfs2_error(inode->i_sb,
7167 			    "Inline data flags for inode %llu don't agree! "
7168 			    "Disk: 0x%x, Memory: 0x%x, Superblock: 0x%x\n",
7169 			    (unsigned long long)OCFS2_I(inode)->ip_blkno,
7170 			    le16_to_cpu(di->i_dyn_features),
7171 			    OCFS2_I(inode)->ip_dyn_features,
7172 			    osb->s_feature_incompat);
7173 		ret = -EROFS;
7174 		goto out;
7175 	}
7176 
7177 	handle = ocfs2_start_trans(osb, OCFS2_INODE_UPDATE_CREDITS);
7178 	if (IS_ERR(handle)) {
7179 		ret = PTR_ERR(handle);
7180 		mlog_errno(ret);
7181 		goto out;
7182 	}
7183 
7184 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
7185 				      OCFS2_JOURNAL_ACCESS_WRITE);
7186 	if (ret) {
7187 		mlog_errno(ret);
7188 		goto out_commit;
7189 	}
7190 
7191 	numbytes = end - start;
7192 	memset(idata->id_data + start, 0, numbytes);
7193 
7194 	/*
7195 	 * No need to worry about the data page here - it's been
7196 	 * truncated already and inline data doesn't need it for
7197 	 * pushing zero's to disk, so we'll let readpage pick it up
7198 	 * later.
7199 	 */
7200 	if (trunc) {
7201 		i_size_write(inode, start);
7202 		di->i_size = cpu_to_le64(start);
7203 	}
7204 
7205 	inode->i_blocks = ocfs2_inode_sector_count(inode);
7206 	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
7207 
7208 	di->i_ctime = di->i_mtime = cpu_to_le64(inode->i_ctime.tv_sec);
7209 	di->i_ctime_nsec = di->i_mtime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
7210 
7211 	ocfs2_journal_dirty(handle, di_bh);
7212 
7213 out_commit:
7214 	ocfs2_commit_trans(osb, handle);
7215 
7216 out:
7217 	return ret;
7218 }
7219 
7220 static int ocfs2_trim_extent(struct super_block *sb,
7221 			     struct ocfs2_group_desc *gd,
7222 			     u32 start, u32 count)
7223 {
7224 	u64 discard, bcount;
7225 
7226 	bcount = ocfs2_clusters_to_blocks(sb, count);
7227 	discard = le64_to_cpu(gd->bg_blkno) +
7228 			ocfs2_clusters_to_blocks(sb, start);
7229 
7230 	trace_ocfs2_trim_extent(sb, (unsigned long long)discard, bcount);
7231 
7232 	return sb_issue_discard(sb, discard, bcount, GFP_NOFS, 0);
7233 }
7234 
7235 static int ocfs2_trim_group(struct super_block *sb,
7236 			    struct ocfs2_group_desc *gd,
7237 			    u32 start, u32 max, u32 minbits)
7238 {
7239 	int ret = 0, count = 0, next;
7240 	void *bitmap = gd->bg_bitmap;
7241 
7242 	if (le16_to_cpu(gd->bg_free_bits_count) < minbits)
7243 		return 0;
7244 
7245 	trace_ocfs2_trim_group((unsigned long long)le64_to_cpu(gd->bg_blkno),
7246 			       start, max, minbits);
7247 
7248 	while (start < max) {
7249 		start = ocfs2_find_next_zero_bit(bitmap, max, start);
7250 		if (start >= max)
7251 			break;
7252 		next = ocfs2_find_next_bit(bitmap, max, start);
7253 
7254 		if ((next - start) >= minbits) {
7255 			ret = ocfs2_trim_extent(sb, gd,
7256 						start, next - start);
7257 			if (ret < 0) {
7258 				mlog_errno(ret);
7259 				break;
7260 			}
7261 			count += next - start;
7262 		}
7263 		start = next + 1;
7264 
7265 		if (fatal_signal_pending(current)) {
7266 			count = -ERESTARTSYS;
7267 			break;
7268 		}
7269 
7270 		if ((le16_to_cpu(gd->bg_free_bits_count) - count) < minbits)
7271 			break;
7272 	}
7273 
7274 	if (ret < 0)
7275 		count = ret;
7276 
7277 	return count;
7278 }
7279 
7280 int ocfs2_trim_fs(struct super_block *sb, struct fstrim_range *range)
7281 {
7282 	struct ocfs2_super *osb = OCFS2_SB(sb);
7283 	u64 start, len, trimmed, first_group, last_group, group;
7284 	int ret, cnt;
7285 	u32 first_bit, last_bit, minlen;
7286 	struct buffer_head *main_bm_bh = NULL;
7287 	struct inode *main_bm_inode = NULL;
7288 	struct buffer_head *gd_bh = NULL;
7289 	struct ocfs2_dinode *main_bm;
7290 	struct ocfs2_group_desc *gd = NULL;
7291 
7292 	start = range->start >> osb->s_clustersize_bits;
7293 	len = range->len >> osb->s_clustersize_bits;
7294 	minlen = range->minlen >> osb->s_clustersize_bits;
7295 
7296 	if (minlen >= osb->bitmap_cpg || range->len < sb->s_blocksize)
7297 		return -EINVAL;
7298 
7299 	main_bm_inode = ocfs2_get_system_file_inode(osb,
7300 						    GLOBAL_BITMAP_SYSTEM_INODE,
7301 						    OCFS2_INVALID_SLOT);
7302 	if (!main_bm_inode) {
7303 		ret = -EIO;
7304 		mlog_errno(ret);
7305 		goto out;
7306 	}
7307 
7308 	mutex_lock(&main_bm_inode->i_mutex);
7309 
7310 	ret = ocfs2_inode_lock(main_bm_inode, &main_bm_bh, 0);
7311 	if (ret < 0) {
7312 		mlog_errno(ret);
7313 		goto out_mutex;
7314 	}
7315 	main_bm = (struct ocfs2_dinode *)main_bm_bh->b_data;
7316 
7317 	if (start >= le32_to_cpu(main_bm->i_clusters)) {
7318 		ret = -EINVAL;
7319 		goto out_unlock;
7320 	}
7321 
7322 	len = range->len >> osb->s_clustersize_bits;
7323 	if (start + len > le32_to_cpu(main_bm->i_clusters))
7324 		len = le32_to_cpu(main_bm->i_clusters) - start;
7325 
7326 	trace_ocfs2_trim_fs(start, len, minlen);
7327 
7328 	/* Determine first and last group to examine based on start and len */
7329 	first_group = ocfs2_which_cluster_group(main_bm_inode, start);
7330 	if (first_group == osb->first_cluster_group_blkno)
7331 		first_bit = start;
7332 	else
7333 		first_bit = start - ocfs2_blocks_to_clusters(sb, first_group);
7334 	last_group = ocfs2_which_cluster_group(main_bm_inode, start + len - 1);
7335 	last_bit = osb->bitmap_cpg;
7336 
7337 	trimmed = 0;
7338 	for (group = first_group; group <= last_group;) {
7339 		if (first_bit + len >= osb->bitmap_cpg)
7340 			last_bit = osb->bitmap_cpg;
7341 		else
7342 			last_bit = first_bit + len;
7343 
7344 		ret = ocfs2_read_group_descriptor(main_bm_inode,
7345 						  main_bm, group,
7346 						  &gd_bh);
7347 		if (ret < 0) {
7348 			mlog_errno(ret);
7349 			break;
7350 		}
7351 
7352 		gd = (struct ocfs2_group_desc *)gd_bh->b_data;
7353 		cnt = ocfs2_trim_group(sb, gd, first_bit, last_bit, minlen);
7354 		brelse(gd_bh);
7355 		gd_bh = NULL;
7356 		if (cnt < 0) {
7357 			ret = cnt;
7358 			mlog_errno(ret);
7359 			break;
7360 		}
7361 
7362 		trimmed += cnt;
7363 		len -= osb->bitmap_cpg - first_bit;
7364 		first_bit = 0;
7365 		if (group == osb->first_cluster_group_blkno)
7366 			group = ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7367 		else
7368 			group += ocfs2_clusters_to_blocks(sb, osb->bitmap_cpg);
7369 	}
7370 	range->len = trimmed * sb->s_blocksize;
7371 out_unlock:
7372 	ocfs2_inode_unlock(main_bm_inode, 0);
7373 	brelse(main_bm_bh);
7374 out_mutex:
7375 	mutex_unlock(&main_bm_inode->i_mutex);
7376 	iput(main_bm_inode);
7377 out:
7378 	return ret;
7379 }
7380