xref: /linux/fs/ocfs2/xattr.c (revision e0bf6c5ca2d3281f231c5f0c9bf145e9513644de)
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * xattr.c
5  *
6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
7  *
8  * CREDITS:
9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public
14  * License version 2 as published by the Free Software Foundation.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  */
21 
22 #include <linux/capability.h>
23 #include <linux/fs.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/security.h>
39 
40 #include <cluster/masklog.h>
41 
42 #include "ocfs2.h"
43 #include "alloc.h"
44 #include "blockcheck.h"
45 #include "dlmglue.h"
46 #include "file.h"
47 #include "symlink.h"
48 #include "sysfile.h"
49 #include "inode.h"
50 #include "journal.h"
51 #include "ocfs2_fs.h"
52 #include "suballoc.h"
53 #include "uptodate.h"
54 #include "buffer_head_io.h"
55 #include "super.h"
56 #include "xattr.h"
57 #include "refcounttree.h"
58 #include "acl.h"
59 #include "ocfs2_trace.h"
60 
61 struct ocfs2_xattr_def_value_root {
62 	struct ocfs2_xattr_value_root	xv;
63 	struct ocfs2_extent_rec		er;
64 };
65 
66 struct ocfs2_xattr_bucket {
67 	/* The inode these xattrs are associated with */
68 	struct inode *bu_inode;
69 
70 	/* The actual buffers that make up the bucket */
71 	struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
72 
73 	/* How many blocks make up one bucket for this filesystem */
74 	int bu_blocks;
75 };
76 
77 struct ocfs2_xattr_set_ctxt {
78 	handle_t *handle;
79 	struct ocfs2_alloc_context *meta_ac;
80 	struct ocfs2_alloc_context *data_ac;
81 	struct ocfs2_cached_dealloc_ctxt dealloc;
82 	int set_abort;
83 };
84 
85 #define OCFS2_XATTR_ROOT_SIZE	(sizeof(struct ocfs2_xattr_def_value_root))
86 #define OCFS2_XATTR_INLINE_SIZE	80
87 #define OCFS2_XATTR_HEADER_GAP	4
88 #define OCFS2_XATTR_FREE_IN_IBODY	(OCFS2_MIN_XATTR_INLINE_SIZE \
89 					 - sizeof(struct ocfs2_xattr_header) \
90 					 - OCFS2_XATTR_HEADER_GAP)
91 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)	((ptr)->i_sb->s_blocksize \
92 					 - sizeof(struct ocfs2_xattr_block) \
93 					 - sizeof(struct ocfs2_xattr_header) \
94 					 - OCFS2_XATTR_HEADER_GAP)
95 
96 static struct ocfs2_xattr_def_value_root def_xv = {
97 	.xv.xr_list.l_count = cpu_to_le16(1),
98 };
99 
100 const struct xattr_handler *ocfs2_xattr_handlers[] = {
101 	&ocfs2_xattr_user_handler,
102 	&posix_acl_access_xattr_handler,
103 	&posix_acl_default_xattr_handler,
104 	&ocfs2_xattr_trusted_handler,
105 	&ocfs2_xattr_security_handler,
106 	NULL
107 };
108 
109 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
110 	[OCFS2_XATTR_INDEX_USER]	= &ocfs2_xattr_user_handler,
111 	[OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
112 					= &posix_acl_access_xattr_handler,
113 	[OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
114 					= &posix_acl_default_xattr_handler,
115 	[OCFS2_XATTR_INDEX_TRUSTED]	= &ocfs2_xattr_trusted_handler,
116 	[OCFS2_XATTR_INDEX_SECURITY]	= &ocfs2_xattr_security_handler,
117 };
118 
119 struct ocfs2_xattr_info {
120 	int		xi_name_index;
121 	const char	*xi_name;
122 	int		xi_name_len;
123 	const void	*xi_value;
124 	size_t		xi_value_len;
125 };
126 
127 struct ocfs2_xattr_search {
128 	struct buffer_head *inode_bh;
129 	/*
130 	 * xattr_bh point to the block buffer head which has extended attribute
131 	 * when extended attribute in inode, xattr_bh is equal to inode_bh.
132 	 */
133 	struct buffer_head *xattr_bh;
134 	struct ocfs2_xattr_header *header;
135 	struct ocfs2_xattr_bucket *bucket;
136 	void *base;
137 	void *end;
138 	struct ocfs2_xattr_entry *here;
139 	int not_found;
140 };
141 
142 /* Operations on struct ocfs2_xa_entry */
143 struct ocfs2_xa_loc;
144 struct ocfs2_xa_loc_operations {
145 	/*
146 	 * Journal functions
147 	 */
148 	int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
149 				  int type);
150 	void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
151 
152 	/*
153 	 * Return a pointer to the appropriate buffer in loc->xl_storage
154 	 * at the given offset from loc->xl_header.
155 	 */
156 	void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
157 
158 	/* Can we reuse the existing entry for the new value? */
159 	int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
160 			     struct ocfs2_xattr_info *xi);
161 
162 	/* How much space is needed for the new value? */
163 	int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
164 			       struct ocfs2_xattr_info *xi);
165 
166 	/*
167 	 * Return the offset of the first name+value pair.  This is
168 	 * the start of our downward-filling free space.
169 	 */
170 	int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
171 
172 	/*
173 	 * Remove the name+value at this location.  Do whatever is
174 	 * appropriate with the remaining name+value pairs.
175 	 */
176 	void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
177 
178 	/* Fill xl_entry with a new entry */
179 	void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
180 
181 	/* Add name+value storage to an entry */
182 	void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
183 
184 	/*
185 	 * Initialize the value buf's access and bh fields for this entry.
186 	 * ocfs2_xa_fill_value_buf() will handle the xv pointer.
187 	 */
188 	void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
189 				   struct ocfs2_xattr_value_buf *vb);
190 };
191 
192 /*
193  * Describes an xattr entry location.  This is a memory structure
194  * tracking the on-disk structure.
195  */
196 struct ocfs2_xa_loc {
197 	/* This xattr belongs to this inode */
198 	struct inode *xl_inode;
199 
200 	/* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
201 	struct ocfs2_xattr_header *xl_header;
202 
203 	/* Bytes from xl_header to the end of the storage */
204 	int xl_size;
205 
206 	/*
207 	 * The ocfs2_xattr_entry this location describes.  If this is
208 	 * NULL, this location describes the on-disk structure where it
209 	 * would have been.
210 	 */
211 	struct ocfs2_xattr_entry *xl_entry;
212 
213 	/*
214 	 * Internal housekeeping
215 	 */
216 
217 	/* Buffer(s) containing this entry */
218 	void *xl_storage;
219 
220 	/* Operations on the storage backing this location */
221 	const struct ocfs2_xa_loc_operations *xl_ops;
222 };
223 
224 /*
225  * Convenience functions to calculate how much space is needed for a
226  * given name+value pair
227  */
228 static int namevalue_size(int name_len, uint64_t value_len)
229 {
230 	if (value_len > OCFS2_XATTR_INLINE_SIZE)
231 		return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
232 	else
233 		return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
234 }
235 
236 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
237 {
238 	return namevalue_size(xi->xi_name_len, xi->xi_value_len);
239 }
240 
241 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
242 {
243 	u64 value_len = le64_to_cpu(xe->xe_value_size);
244 
245 	BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
246 	       ocfs2_xattr_is_local(xe));
247 	return namevalue_size(xe->xe_name_len, value_len);
248 }
249 
250 
251 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
252 					     struct ocfs2_xattr_header *xh,
253 					     int index,
254 					     int *block_off,
255 					     int *new_offset);
256 
257 static int ocfs2_xattr_block_find(struct inode *inode,
258 				  int name_index,
259 				  const char *name,
260 				  struct ocfs2_xattr_search *xs);
261 static int ocfs2_xattr_index_block_find(struct inode *inode,
262 					struct buffer_head *root_bh,
263 					int name_index,
264 					const char *name,
265 					struct ocfs2_xattr_search *xs);
266 
267 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
268 					struct buffer_head *blk_bh,
269 					char *buffer,
270 					size_t buffer_size);
271 
272 static int ocfs2_xattr_create_index_block(struct inode *inode,
273 					  struct ocfs2_xattr_search *xs,
274 					  struct ocfs2_xattr_set_ctxt *ctxt);
275 
276 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
277 					     struct ocfs2_xattr_info *xi,
278 					     struct ocfs2_xattr_search *xs,
279 					     struct ocfs2_xattr_set_ctxt *ctxt);
280 
281 typedef int (xattr_tree_rec_func)(struct inode *inode,
282 				  struct buffer_head *root_bh,
283 				  u64 blkno, u32 cpos, u32 len, void *para);
284 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
285 					   struct buffer_head *root_bh,
286 					   xattr_tree_rec_func *rec_func,
287 					   void *para);
288 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
289 					struct ocfs2_xattr_bucket *bucket,
290 					void *para);
291 static int ocfs2_rm_xattr_cluster(struct inode *inode,
292 				  struct buffer_head *root_bh,
293 				  u64 blkno,
294 				  u32 cpos,
295 				  u32 len,
296 				  void *para);
297 
298 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
299 				  u64 src_blk, u64 last_blk, u64 to_blk,
300 				  unsigned int start_bucket,
301 				  u32 *first_hash);
302 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
303 					struct ocfs2_dinode *di,
304 					struct ocfs2_xattr_info *xi,
305 					struct ocfs2_xattr_search *xis,
306 					struct ocfs2_xattr_search *xbs,
307 					struct ocfs2_refcount_tree **ref_tree,
308 					int *meta_need,
309 					int *credits);
310 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
311 					   struct ocfs2_xattr_bucket *bucket,
312 					   int offset,
313 					   struct ocfs2_xattr_value_root **xv,
314 					   struct buffer_head **bh);
315 
316 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
317 {
318 	return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
319 }
320 
321 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
322 {
323 	return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
324 }
325 
326 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
327 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
328 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
329 
330 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
331 {
332 	struct ocfs2_xattr_bucket *bucket;
333 	int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
334 
335 	BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
336 
337 	bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
338 	if (bucket) {
339 		bucket->bu_inode = inode;
340 		bucket->bu_blocks = blks;
341 	}
342 
343 	return bucket;
344 }
345 
346 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
347 {
348 	int i;
349 
350 	for (i = 0; i < bucket->bu_blocks; i++) {
351 		brelse(bucket->bu_bhs[i]);
352 		bucket->bu_bhs[i] = NULL;
353 	}
354 }
355 
356 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
357 {
358 	if (bucket) {
359 		ocfs2_xattr_bucket_relse(bucket);
360 		bucket->bu_inode = NULL;
361 		kfree(bucket);
362 	}
363 }
364 
365 /*
366  * A bucket that has never been written to disk doesn't need to be
367  * read.  We just need the buffer_heads.  Don't call this for
368  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
369  * them fully.
370  */
371 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
372 				   u64 xb_blkno, int new)
373 {
374 	int i, rc = 0;
375 
376 	for (i = 0; i < bucket->bu_blocks; i++) {
377 		bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
378 					      xb_blkno + i);
379 		if (!bucket->bu_bhs[i]) {
380 			rc = -ENOMEM;
381 			mlog_errno(rc);
382 			break;
383 		}
384 
385 		if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
386 					   bucket->bu_bhs[i])) {
387 			if (new)
388 				ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
389 							      bucket->bu_bhs[i]);
390 			else {
391 				set_buffer_uptodate(bucket->bu_bhs[i]);
392 				ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
393 							  bucket->bu_bhs[i]);
394 			}
395 		}
396 	}
397 
398 	if (rc)
399 		ocfs2_xattr_bucket_relse(bucket);
400 	return rc;
401 }
402 
403 /* Read the xattr bucket at xb_blkno */
404 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
405 				   u64 xb_blkno)
406 {
407 	int rc;
408 
409 	rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
410 			       bucket->bu_blocks, bucket->bu_bhs, 0,
411 			       NULL);
412 	if (!rc) {
413 		spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
414 		rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
415 						 bucket->bu_bhs,
416 						 bucket->bu_blocks,
417 						 &bucket_xh(bucket)->xh_check);
418 		spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
419 		if (rc)
420 			mlog_errno(rc);
421 	}
422 
423 	if (rc)
424 		ocfs2_xattr_bucket_relse(bucket);
425 	return rc;
426 }
427 
428 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
429 					     struct ocfs2_xattr_bucket *bucket,
430 					     int type)
431 {
432 	int i, rc = 0;
433 
434 	for (i = 0; i < bucket->bu_blocks; i++) {
435 		rc = ocfs2_journal_access(handle,
436 					  INODE_CACHE(bucket->bu_inode),
437 					  bucket->bu_bhs[i], type);
438 		if (rc) {
439 			mlog_errno(rc);
440 			break;
441 		}
442 	}
443 
444 	return rc;
445 }
446 
447 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
448 					     struct ocfs2_xattr_bucket *bucket)
449 {
450 	int i;
451 
452 	spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
453 	ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
454 				   bucket->bu_bhs, bucket->bu_blocks,
455 				   &bucket_xh(bucket)->xh_check);
456 	spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
457 
458 	for (i = 0; i < bucket->bu_blocks; i++)
459 		ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
460 }
461 
462 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
463 					 struct ocfs2_xattr_bucket *src)
464 {
465 	int i;
466 	int blocksize = src->bu_inode->i_sb->s_blocksize;
467 
468 	BUG_ON(dest->bu_blocks != src->bu_blocks);
469 	BUG_ON(dest->bu_inode != src->bu_inode);
470 
471 	for (i = 0; i < src->bu_blocks; i++) {
472 		memcpy(bucket_block(dest, i), bucket_block(src, i),
473 		       blocksize);
474 	}
475 }
476 
477 static int ocfs2_validate_xattr_block(struct super_block *sb,
478 				      struct buffer_head *bh)
479 {
480 	int rc;
481 	struct ocfs2_xattr_block *xb =
482 		(struct ocfs2_xattr_block *)bh->b_data;
483 
484 	trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
485 
486 	BUG_ON(!buffer_uptodate(bh));
487 
488 	/*
489 	 * If the ecc fails, we return the error but otherwise
490 	 * leave the filesystem running.  We know any error is
491 	 * local to this block.
492 	 */
493 	rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
494 	if (rc)
495 		return rc;
496 
497 	/*
498 	 * Errors after here are fatal
499 	 */
500 
501 	if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
502 		ocfs2_error(sb,
503 			    "Extended attribute block #%llu has bad "
504 			    "signature %.*s",
505 			    (unsigned long long)bh->b_blocknr, 7,
506 			    xb->xb_signature);
507 		return -EINVAL;
508 	}
509 
510 	if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
511 		ocfs2_error(sb,
512 			    "Extended attribute block #%llu has an "
513 			    "invalid xb_blkno of %llu",
514 			    (unsigned long long)bh->b_blocknr,
515 			    (unsigned long long)le64_to_cpu(xb->xb_blkno));
516 		return -EINVAL;
517 	}
518 
519 	if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
520 		ocfs2_error(sb,
521 			    "Extended attribute block #%llu has an invalid "
522 			    "xb_fs_generation of #%u",
523 			    (unsigned long long)bh->b_blocknr,
524 			    le32_to_cpu(xb->xb_fs_generation));
525 		return -EINVAL;
526 	}
527 
528 	return 0;
529 }
530 
531 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
532 				  struct buffer_head **bh)
533 {
534 	int rc;
535 	struct buffer_head *tmp = *bh;
536 
537 	rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
538 			      ocfs2_validate_xattr_block);
539 
540 	/* If ocfs2_read_block() got us a new bh, pass it up. */
541 	if (!rc && !*bh)
542 		*bh = tmp;
543 
544 	return rc;
545 }
546 
547 static inline const char *ocfs2_xattr_prefix(int name_index)
548 {
549 	const struct xattr_handler *handler = NULL;
550 
551 	if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
552 		handler = ocfs2_xattr_handler_map[name_index];
553 
554 	return handler ? handler->prefix : NULL;
555 }
556 
557 static u32 ocfs2_xattr_name_hash(struct inode *inode,
558 				 const char *name,
559 				 int name_len)
560 {
561 	/* Get hash value of uuid from super block */
562 	u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
563 	int i;
564 
565 	/* hash extended attribute name */
566 	for (i = 0; i < name_len; i++) {
567 		hash = (hash << OCFS2_HASH_SHIFT) ^
568 		       (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
569 		       *name++;
570 	}
571 
572 	return hash;
573 }
574 
575 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
576 {
577 	return namevalue_size(name_len, value_len) +
578 		sizeof(struct ocfs2_xattr_entry);
579 }
580 
581 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
582 {
583 	return namevalue_size_xi(xi) +
584 		sizeof(struct ocfs2_xattr_entry);
585 }
586 
587 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
588 {
589 	return namevalue_size_xe(xe) +
590 		sizeof(struct ocfs2_xattr_entry);
591 }
592 
593 int ocfs2_calc_security_init(struct inode *dir,
594 			     struct ocfs2_security_xattr_info *si,
595 			     int *want_clusters,
596 			     int *xattr_credits,
597 			     struct ocfs2_alloc_context **xattr_ac)
598 {
599 	int ret = 0;
600 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
601 	int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
602 						 si->value_len);
603 
604 	/*
605 	 * The max space of security xattr taken inline is
606 	 * 256(name) + 80(value) + 16(entry) = 352 bytes,
607 	 * So reserve one metadata block for it is ok.
608 	 */
609 	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
610 	    s_size > OCFS2_XATTR_FREE_IN_IBODY) {
611 		ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
612 		if (ret) {
613 			mlog_errno(ret);
614 			return ret;
615 		}
616 		*xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
617 	}
618 
619 	/* reserve clusters for xattr value which will be set in B tree*/
620 	if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
621 		int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
622 							    si->value_len);
623 
624 		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
625 							   new_clusters);
626 		*want_clusters += new_clusters;
627 	}
628 	return ret;
629 }
630 
631 int ocfs2_calc_xattr_init(struct inode *dir,
632 			  struct buffer_head *dir_bh,
633 			  umode_t mode,
634 			  struct ocfs2_security_xattr_info *si,
635 			  int *want_clusters,
636 			  int *xattr_credits,
637 			  int *want_meta)
638 {
639 	int ret = 0;
640 	struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
641 	int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
642 
643 	if (si->enable)
644 		s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
645 						     si->value_len);
646 
647 	if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
648 		acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
649 					OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
650 					"", NULL, 0);
651 		if (acl_len > 0) {
652 			a_size = ocfs2_xattr_entry_real_size(0, acl_len);
653 			if (S_ISDIR(mode))
654 				a_size <<= 1;
655 		} else if (acl_len != 0 && acl_len != -ENODATA) {
656 			mlog_errno(ret);
657 			return ret;
658 		}
659 	}
660 
661 	if (!(s_size + a_size))
662 		return ret;
663 
664 	/*
665 	 * The max space of security xattr taken inline is
666 	 * 256(name) + 80(value) + 16(entry) = 352 bytes,
667 	 * The max space of acl xattr taken inline is
668 	 * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
669 	 * when blocksize = 512, may reserve one more cluser for
670 	 * xattr bucket, otherwise reserve one metadata block
671 	 * for them is ok.
672 	 * If this is a new directory with inline data,
673 	 * we choose to reserve the entire inline area for
674 	 * directory contents and force an external xattr block.
675 	 */
676 	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
677 	    (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
678 	    (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
679 		*want_meta = *want_meta + 1;
680 		*xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
681 	}
682 
683 	if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
684 	    (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
685 		*want_clusters += 1;
686 		*xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
687 	}
688 
689 	/*
690 	 * reserve credits and clusters for xattrs which has large value
691 	 * and have to be set outside
692 	 */
693 	if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
694 		new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
695 							si->value_len);
696 		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
697 							   new_clusters);
698 		*want_clusters += new_clusters;
699 	}
700 	if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
701 	    acl_len > OCFS2_XATTR_INLINE_SIZE) {
702 		/* for directory, it has DEFAULT and ACCESS two types of acls */
703 		new_clusters = (S_ISDIR(mode) ? 2 : 1) *
704 				ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
705 		*xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
706 							   new_clusters);
707 		*want_clusters += new_clusters;
708 	}
709 
710 	return ret;
711 }
712 
713 static int ocfs2_xattr_extend_allocation(struct inode *inode,
714 					 u32 clusters_to_add,
715 					 struct ocfs2_xattr_value_buf *vb,
716 					 struct ocfs2_xattr_set_ctxt *ctxt)
717 {
718 	int status = 0, credits;
719 	handle_t *handle = ctxt->handle;
720 	enum ocfs2_alloc_restarted why;
721 	u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
722 	struct ocfs2_extent_tree et;
723 
724 	ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
725 
726 	while (clusters_to_add) {
727 		trace_ocfs2_xattr_extend_allocation(clusters_to_add);
728 
729 		status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
730 				       OCFS2_JOURNAL_ACCESS_WRITE);
731 		if (status < 0) {
732 			mlog_errno(status);
733 			break;
734 		}
735 
736 		prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
737 		status = ocfs2_add_clusters_in_btree(handle,
738 						     &et,
739 						     &logical_start,
740 						     clusters_to_add,
741 						     0,
742 						     ctxt->data_ac,
743 						     ctxt->meta_ac,
744 						     &why);
745 		if ((status < 0) && (status != -EAGAIN)) {
746 			if (status != -ENOSPC)
747 				mlog_errno(status);
748 			break;
749 		}
750 
751 		ocfs2_journal_dirty(handle, vb->vb_bh);
752 
753 		clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
754 					 prev_clusters;
755 
756 		if (why != RESTART_NONE && clusters_to_add) {
757 			/*
758 			 * We can only fail in case the alloc file doesn't give
759 			 * up enough clusters.
760 			 */
761 			BUG_ON(why == RESTART_META);
762 
763 			credits = ocfs2_calc_extend_credits(inode->i_sb,
764 							    &vb->vb_xv->xr_list);
765 			status = ocfs2_extend_trans(handle, credits);
766 			if (status < 0) {
767 				status = -ENOMEM;
768 				mlog_errno(status);
769 				break;
770 			}
771 		}
772 	}
773 
774 	return status;
775 }
776 
777 static int __ocfs2_remove_xattr_range(struct inode *inode,
778 				      struct ocfs2_xattr_value_buf *vb,
779 				      u32 cpos, u32 phys_cpos, u32 len,
780 				      unsigned int ext_flags,
781 				      struct ocfs2_xattr_set_ctxt *ctxt)
782 {
783 	int ret;
784 	u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
785 	handle_t *handle = ctxt->handle;
786 	struct ocfs2_extent_tree et;
787 
788 	ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
789 
790 	ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
791 			    OCFS2_JOURNAL_ACCESS_WRITE);
792 	if (ret) {
793 		mlog_errno(ret);
794 		goto out;
795 	}
796 
797 	ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
798 				  &ctxt->dealloc);
799 	if (ret) {
800 		mlog_errno(ret);
801 		goto out;
802 	}
803 
804 	le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
805 	ocfs2_journal_dirty(handle, vb->vb_bh);
806 
807 	if (ext_flags & OCFS2_EXT_REFCOUNTED)
808 		ret = ocfs2_decrease_refcount(inode, handle,
809 					ocfs2_blocks_to_clusters(inode->i_sb,
810 								 phys_blkno),
811 					len, ctxt->meta_ac, &ctxt->dealloc, 1);
812 	else
813 		ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
814 						  phys_blkno, len);
815 	if (ret)
816 		mlog_errno(ret);
817 
818 out:
819 	return ret;
820 }
821 
822 static int ocfs2_xattr_shrink_size(struct inode *inode,
823 				   u32 old_clusters,
824 				   u32 new_clusters,
825 				   struct ocfs2_xattr_value_buf *vb,
826 				   struct ocfs2_xattr_set_ctxt *ctxt)
827 {
828 	int ret = 0;
829 	unsigned int ext_flags;
830 	u32 trunc_len, cpos, phys_cpos, alloc_size;
831 	u64 block;
832 
833 	if (old_clusters <= new_clusters)
834 		return 0;
835 
836 	cpos = new_clusters;
837 	trunc_len = old_clusters - new_clusters;
838 	while (trunc_len) {
839 		ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
840 					       &alloc_size,
841 					       &vb->vb_xv->xr_list, &ext_flags);
842 		if (ret) {
843 			mlog_errno(ret);
844 			goto out;
845 		}
846 
847 		if (alloc_size > trunc_len)
848 			alloc_size = trunc_len;
849 
850 		ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
851 						 phys_cpos, alloc_size,
852 						 ext_flags, ctxt);
853 		if (ret) {
854 			mlog_errno(ret);
855 			goto out;
856 		}
857 
858 		block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
859 		ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
860 						       block, alloc_size);
861 		cpos += alloc_size;
862 		trunc_len -= alloc_size;
863 	}
864 
865 out:
866 	return ret;
867 }
868 
869 static int ocfs2_xattr_value_truncate(struct inode *inode,
870 				      struct ocfs2_xattr_value_buf *vb,
871 				      int len,
872 				      struct ocfs2_xattr_set_ctxt *ctxt)
873 {
874 	int ret;
875 	u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
876 	u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
877 
878 	if (new_clusters == old_clusters)
879 		return 0;
880 
881 	if (new_clusters > old_clusters)
882 		ret = ocfs2_xattr_extend_allocation(inode,
883 						    new_clusters - old_clusters,
884 						    vb, ctxt);
885 	else
886 		ret = ocfs2_xattr_shrink_size(inode,
887 					      old_clusters, new_clusters,
888 					      vb, ctxt);
889 
890 	return ret;
891 }
892 
893 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
894 				  size_t *result, const char *prefix,
895 				  const char *name, int name_len)
896 {
897 	char *p = buffer + *result;
898 	int prefix_len = strlen(prefix);
899 	int total_len = prefix_len + name_len + 1;
900 
901 	*result += total_len;
902 
903 	/* we are just looking for how big our buffer needs to be */
904 	if (!size)
905 		return 0;
906 
907 	if (*result > size)
908 		return -ERANGE;
909 
910 	memcpy(p, prefix, prefix_len);
911 	memcpy(p + prefix_len, name, name_len);
912 	p[prefix_len + name_len] = '\0';
913 
914 	return 0;
915 }
916 
917 static int ocfs2_xattr_list_entries(struct inode *inode,
918 				    struct ocfs2_xattr_header *header,
919 				    char *buffer, size_t buffer_size)
920 {
921 	size_t result = 0;
922 	int i, type, ret;
923 	const char *prefix, *name;
924 
925 	for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
926 		struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
927 		type = ocfs2_xattr_get_type(entry);
928 		prefix = ocfs2_xattr_prefix(type);
929 
930 		if (prefix) {
931 			name = (const char *)header +
932 				le16_to_cpu(entry->xe_name_offset);
933 
934 			ret = ocfs2_xattr_list_entry(buffer, buffer_size,
935 						     &result, prefix, name,
936 						     entry->xe_name_len);
937 			if (ret)
938 				return ret;
939 		}
940 	}
941 
942 	return result;
943 }
944 
945 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
946 					 struct ocfs2_dinode *di)
947 {
948 	struct ocfs2_xattr_header *xh;
949 	int i;
950 
951 	xh = (struct ocfs2_xattr_header *)
952 		 ((void *)di + inode->i_sb->s_blocksize -
953 		 le16_to_cpu(di->i_xattr_inline_size));
954 
955 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
956 		if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
957 			return 1;
958 
959 	return 0;
960 }
961 
962 static int ocfs2_xattr_ibody_list(struct inode *inode,
963 				  struct ocfs2_dinode *di,
964 				  char *buffer,
965 				  size_t buffer_size)
966 {
967 	struct ocfs2_xattr_header *header = NULL;
968 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
969 	int ret = 0;
970 
971 	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
972 		return ret;
973 
974 	header = (struct ocfs2_xattr_header *)
975 		 ((void *)di + inode->i_sb->s_blocksize -
976 		 le16_to_cpu(di->i_xattr_inline_size));
977 
978 	ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
979 
980 	return ret;
981 }
982 
983 static int ocfs2_xattr_block_list(struct inode *inode,
984 				  struct ocfs2_dinode *di,
985 				  char *buffer,
986 				  size_t buffer_size)
987 {
988 	struct buffer_head *blk_bh = NULL;
989 	struct ocfs2_xattr_block *xb;
990 	int ret = 0;
991 
992 	if (!di->i_xattr_loc)
993 		return ret;
994 
995 	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
996 				     &blk_bh);
997 	if (ret < 0) {
998 		mlog_errno(ret);
999 		return ret;
1000 	}
1001 
1002 	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1003 	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1004 		struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1005 		ret = ocfs2_xattr_list_entries(inode, header,
1006 					       buffer, buffer_size);
1007 	} else
1008 		ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1009 						   buffer, buffer_size);
1010 
1011 	brelse(blk_bh);
1012 
1013 	return ret;
1014 }
1015 
1016 ssize_t ocfs2_listxattr(struct dentry *dentry,
1017 			char *buffer,
1018 			size_t size)
1019 {
1020 	int ret = 0, i_ret = 0, b_ret = 0;
1021 	struct buffer_head *di_bh = NULL;
1022 	struct ocfs2_dinode *di = NULL;
1023 	struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1024 
1025 	if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1026 		return -EOPNOTSUPP;
1027 
1028 	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1029 		return ret;
1030 
1031 	ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1032 	if (ret < 0) {
1033 		mlog_errno(ret);
1034 		return ret;
1035 	}
1036 
1037 	di = (struct ocfs2_dinode *)di_bh->b_data;
1038 
1039 	down_read(&oi->ip_xattr_sem);
1040 	i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1041 	if (i_ret < 0)
1042 		b_ret = 0;
1043 	else {
1044 		if (buffer) {
1045 			buffer += i_ret;
1046 			size -= i_ret;
1047 		}
1048 		b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1049 					       buffer, size);
1050 		if (b_ret < 0)
1051 			i_ret = 0;
1052 	}
1053 	up_read(&oi->ip_xattr_sem);
1054 	ocfs2_inode_unlock(dentry->d_inode, 0);
1055 
1056 	brelse(di_bh);
1057 
1058 	return i_ret + b_ret;
1059 }
1060 
1061 static int ocfs2_xattr_find_entry(int name_index,
1062 				  const char *name,
1063 				  struct ocfs2_xattr_search *xs)
1064 {
1065 	struct ocfs2_xattr_entry *entry;
1066 	size_t name_len;
1067 	int i, cmp = 1;
1068 
1069 	if (name == NULL)
1070 		return -EINVAL;
1071 
1072 	name_len = strlen(name);
1073 	entry = xs->here;
1074 	for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1075 		cmp = name_index - ocfs2_xattr_get_type(entry);
1076 		if (!cmp)
1077 			cmp = name_len - entry->xe_name_len;
1078 		if (!cmp)
1079 			cmp = memcmp(name, (xs->base +
1080 				     le16_to_cpu(entry->xe_name_offset)),
1081 				     name_len);
1082 		if (cmp == 0)
1083 			break;
1084 		entry += 1;
1085 	}
1086 	xs->here = entry;
1087 
1088 	return cmp ? -ENODATA : 0;
1089 }
1090 
1091 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1092 					 struct ocfs2_xattr_value_root *xv,
1093 					 void *buffer,
1094 					 size_t len)
1095 {
1096 	u32 cpos, p_cluster, num_clusters, bpc, clusters;
1097 	u64 blkno;
1098 	int i, ret = 0;
1099 	size_t cplen, blocksize;
1100 	struct buffer_head *bh = NULL;
1101 	struct ocfs2_extent_list *el;
1102 
1103 	el = &xv->xr_list;
1104 	clusters = le32_to_cpu(xv->xr_clusters);
1105 	bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1106 	blocksize = inode->i_sb->s_blocksize;
1107 
1108 	cpos = 0;
1109 	while (cpos < clusters) {
1110 		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1111 					       &num_clusters, el, NULL);
1112 		if (ret) {
1113 			mlog_errno(ret);
1114 			goto out;
1115 		}
1116 
1117 		blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1118 		/* Copy ocfs2_xattr_value */
1119 		for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1120 			ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1121 					       &bh, NULL);
1122 			if (ret) {
1123 				mlog_errno(ret);
1124 				goto out;
1125 			}
1126 
1127 			cplen = len >= blocksize ? blocksize : len;
1128 			memcpy(buffer, bh->b_data, cplen);
1129 			len -= cplen;
1130 			buffer += cplen;
1131 
1132 			brelse(bh);
1133 			bh = NULL;
1134 			if (len == 0)
1135 				break;
1136 		}
1137 		cpos += num_clusters;
1138 	}
1139 out:
1140 	return ret;
1141 }
1142 
1143 static int ocfs2_xattr_ibody_get(struct inode *inode,
1144 				 int name_index,
1145 				 const char *name,
1146 				 void *buffer,
1147 				 size_t buffer_size,
1148 				 struct ocfs2_xattr_search *xs)
1149 {
1150 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1151 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1152 	struct ocfs2_xattr_value_root *xv;
1153 	size_t size;
1154 	int ret = 0;
1155 
1156 	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1157 		return -ENODATA;
1158 
1159 	xs->end = (void *)di + inode->i_sb->s_blocksize;
1160 	xs->header = (struct ocfs2_xattr_header *)
1161 			(xs->end - le16_to_cpu(di->i_xattr_inline_size));
1162 	xs->base = (void *)xs->header;
1163 	xs->here = xs->header->xh_entries;
1164 
1165 	ret = ocfs2_xattr_find_entry(name_index, name, xs);
1166 	if (ret)
1167 		return ret;
1168 	size = le64_to_cpu(xs->here->xe_value_size);
1169 	if (buffer) {
1170 		if (size > buffer_size)
1171 			return -ERANGE;
1172 		if (ocfs2_xattr_is_local(xs->here)) {
1173 			memcpy(buffer, (void *)xs->base +
1174 			       le16_to_cpu(xs->here->xe_name_offset) +
1175 			       OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1176 		} else {
1177 			xv = (struct ocfs2_xattr_value_root *)
1178 				(xs->base + le16_to_cpu(
1179 				 xs->here->xe_name_offset) +
1180 				OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1181 			ret = ocfs2_xattr_get_value_outside(inode, xv,
1182 							    buffer, size);
1183 			if (ret < 0) {
1184 				mlog_errno(ret);
1185 				return ret;
1186 			}
1187 		}
1188 	}
1189 
1190 	return size;
1191 }
1192 
1193 static int ocfs2_xattr_block_get(struct inode *inode,
1194 				 int name_index,
1195 				 const char *name,
1196 				 void *buffer,
1197 				 size_t buffer_size,
1198 				 struct ocfs2_xattr_search *xs)
1199 {
1200 	struct ocfs2_xattr_block *xb;
1201 	struct ocfs2_xattr_value_root *xv;
1202 	size_t size;
1203 	int ret = -ENODATA, name_offset, name_len, i;
1204 	int uninitialized_var(block_off);
1205 
1206 	xs->bucket = ocfs2_xattr_bucket_new(inode);
1207 	if (!xs->bucket) {
1208 		ret = -ENOMEM;
1209 		mlog_errno(ret);
1210 		goto cleanup;
1211 	}
1212 
1213 	ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1214 	if (ret) {
1215 		mlog_errno(ret);
1216 		goto cleanup;
1217 	}
1218 
1219 	if (xs->not_found) {
1220 		ret = -ENODATA;
1221 		goto cleanup;
1222 	}
1223 
1224 	xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1225 	size = le64_to_cpu(xs->here->xe_value_size);
1226 	if (buffer) {
1227 		ret = -ERANGE;
1228 		if (size > buffer_size)
1229 			goto cleanup;
1230 
1231 		name_offset = le16_to_cpu(xs->here->xe_name_offset);
1232 		name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1233 		i = xs->here - xs->header->xh_entries;
1234 
1235 		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1236 			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1237 								bucket_xh(xs->bucket),
1238 								i,
1239 								&block_off,
1240 								&name_offset);
1241 			xs->base = bucket_block(xs->bucket, block_off);
1242 		}
1243 		if (ocfs2_xattr_is_local(xs->here)) {
1244 			memcpy(buffer, (void *)xs->base +
1245 			       name_offset + name_len, size);
1246 		} else {
1247 			xv = (struct ocfs2_xattr_value_root *)
1248 				(xs->base + name_offset + name_len);
1249 			ret = ocfs2_xattr_get_value_outside(inode, xv,
1250 							    buffer, size);
1251 			if (ret < 0) {
1252 				mlog_errno(ret);
1253 				goto cleanup;
1254 			}
1255 		}
1256 	}
1257 	ret = size;
1258 cleanup:
1259 	ocfs2_xattr_bucket_free(xs->bucket);
1260 
1261 	brelse(xs->xattr_bh);
1262 	xs->xattr_bh = NULL;
1263 	return ret;
1264 }
1265 
1266 int ocfs2_xattr_get_nolock(struct inode *inode,
1267 			   struct buffer_head *di_bh,
1268 			   int name_index,
1269 			   const char *name,
1270 			   void *buffer,
1271 			   size_t buffer_size)
1272 {
1273 	int ret;
1274 	struct ocfs2_dinode *di = NULL;
1275 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
1276 	struct ocfs2_xattr_search xis = {
1277 		.not_found = -ENODATA,
1278 	};
1279 	struct ocfs2_xattr_search xbs = {
1280 		.not_found = -ENODATA,
1281 	};
1282 
1283 	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1284 		return -EOPNOTSUPP;
1285 
1286 	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1287 		return -ENODATA;
1288 
1289 	xis.inode_bh = xbs.inode_bh = di_bh;
1290 	di = (struct ocfs2_dinode *)di_bh->b_data;
1291 
1292 	ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1293 				    buffer_size, &xis);
1294 	if (ret == -ENODATA && di->i_xattr_loc)
1295 		ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1296 					    buffer_size, &xbs);
1297 
1298 	return ret;
1299 }
1300 
1301 /* ocfs2_xattr_get()
1302  *
1303  * Copy an extended attribute into the buffer provided.
1304  * Buffer is NULL to compute the size of buffer required.
1305  */
1306 static int ocfs2_xattr_get(struct inode *inode,
1307 			   int name_index,
1308 			   const char *name,
1309 			   void *buffer,
1310 			   size_t buffer_size)
1311 {
1312 	int ret;
1313 	struct buffer_head *di_bh = NULL;
1314 
1315 	ret = ocfs2_inode_lock(inode, &di_bh, 0);
1316 	if (ret < 0) {
1317 		mlog_errno(ret);
1318 		return ret;
1319 	}
1320 	down_read(&OCFS2_I(inode)->ip_xattr_sem);
1321 	ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1322 				     name, buffer, buffer_size);
1323 	up_read(&OCFS2_I(inode)->ip_xattr_sem);
1324 
1325 	ocfs2_inode_unlock(inode, 0);
1326 
1327 	brelse(di_bh);
1328 
1329 	return ret;
1330 }
1331 
1332 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1333 					   handle_t *handle,
1334 					   struct ocfs2_xattr_value_buf *vb,
1335 					   const void *value,
1336 					   int value_len)
1337 {
1338 	int ret = 0, i, cp_len;
1339 	u16 blocksize = inode->i_sb->s_blocksize;
1340 	u32 p_cluster, num_clusters;
1341 	u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1342 	u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1343 	u64 blkno;
1344 	struct buffer_head *bh = NULL;
1345 	unsigned int ext_flags;
1346 	struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1347 
1348 	BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1349 
1350 	while (cpos < clusters) {
1351 		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1352 					       &num_clusters, &xv->xr_list,
1353 					       &ext_flags);
1354 		if (ret) {
1355 			mlog_errno(ret);
1356 			goto out;
1357 		}
1358 
1359 		BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1360 
1361 		blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1362 
1363 		for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1364 			ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1365 					       &bh, NULL);
1366 			if (ret) {
1367 				mlog_errno(ret);
1368 				goto out;
1369 			}
1370 
1371 			ret = ocfs2_journal_access(handle,
1372 						   INODE_CACHE(inode),
1373 						   bh,
1374 						   OCFS2_JOURNAL_ACCESS_WRITE);
1375 			if (ret < 0) {
1376 				mlog_errno(ret);
1377 				goto out;
1378 			}
1379 
1380 			cp_len = value_len > blocksize ? blocksize : value_len;
1381 			memcpy(bh->b_data, value, cp_len);
1382 			value_len -= cp_len;
1383 			value += cp_len;
1384 			if (cp_len < blocksize)
1385 				memset(bh->b_data + cp_len, 0,
1386 				       blocksize - cp_len);
1387 
1388 			ocfs2_journal_dirty(handle, bh);
1389 			brelse(bh);
1390 			bh = NULL;
1391 
1392 			/*
1393 			 * XXX: do we need to empty all the following
1394 			 * blocks in this cluster?
1395 			 */
1396 			if (!value_len)
1397 				break;
1398 		}
1399 		cpos += num_clusters;
1400 	}
1401 out:
1402 	brelse(bh);
1403 
1404 	return ret;
1405 }
1406 
1407 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1408 				       int num_entries)
1409 {
1410 	int free_space;
1411 
1412 	if (!needed_space)
1413 		return 0;
1414 
1415 	free_space = free_start -
1416 		sizeof(struct ocfs2_xattr_header) -
1417 		(num_entries * sizeof(struct ocfs2_xattr_entry)) -
1418 		OCFS2_XATTR_HEADER_GAP;
1419 	if (free_space < 0)
1420 		return -EIO;
1421 	if (free_space < needed_space)
1422 		return -ENOSPC;
1423 
1424 	return 0;
1425 }
1426 
1427 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1428 				   int type)
1429 {
1430 	return loc->xl_ops->xlo_journal_access(handle, loc, type);
1431 }
1432 
1433 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1434 {
1435 	loc->xl_ops->xlo_journal_dirty(handle, loc);
1436 }
1437 
1438 /* Give a pointer into the storage for the given offset */
1439 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1440 {
1441 	BUG_ON(offset >= loc->xl_size);
1442 	return loc->xl_ops->xlo_offset_pointer(loc, offset);
1443 }
1444 
1445 /*
1446  * Wipe the name+value pair and allow the storage to reclaim it.  This
1447  * must be followed by either removal of the entry or a call to
1448  * ocfs2_xa_add_namevalue().
1449  */
1450 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1451 {
1452 	loc->xl_ops->xlo_wipe_namevalue(loc);
1453 }
1454 
1455 /*
1456  * Find lowest offset to a name+value pair.  This is the start of our
1457  * downward-growing free space.
1458  */
1459 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1460 {
1461 	return loc->xl_ops->xlo_get_free_start(loc);
1462 }
1463 
1464 /* Can we reuse loc->xl_entry for xi? */
1465 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1466 				    struct ocfs2_xattr_info *xi)
1467 {
1468 	return loc->xl_ops->xlo_can_reuse(loc, xi);
1469 }
1470 
1471 /* How much free space is needed to set the new value */
1472 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1473 				struct ocfs2_xattr_info *xi)
1474 {
1475 	return loc->xl_ops->xlo_check_space(loc, xi);
1476 }
1477 
1478 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1479 {
1480 	loc->xl_ops->xlo_add_entry(loc, name_hash);
1481 	loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1482 	/*
1483 	 * We can't leave the new entry's xe_name_offset at zero or
1484 	 * add_namevalue() will go nuts.  We set it to the size of our
1485 	 * storage so that it can never be less than any other entry.
1486 	 */
1487 	loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1488 }
1489 
1490 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1491 				   struct ocfs2_xattr_info *xi)
1492 {
1493 	int size = namevalue_size_xi(xi);
1494 	int nameval_offset;
1495 	char *nameval_buf;
1496 
1497 	loc->xl_ops->xlo_add_namevalue(loc, size);
1498 	loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1499 	loc->xl_entry->xe_name_len = xi->xi_name_len;
1500 	ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1501 	ocfs2_xattr_set_local(loc->xl_entry,
1502 			      xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1503 
1504 	nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1505 	nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1506 	memset(nameval_buf, 0, size);
1507 	memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1508 }
1509 
1510 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1511 				    struct ocfs2_xattr_value_buf *vb)
1512 {
1513 	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1514 	int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1515 
1516 	/* Value bufs are for value trees */
1517 	BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1518 	BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1519 	       (name_size + OCFS2_XATTR_ROOT_SIZE));
1520 
1521 	loc->xl_ops->xlo_fill_value_buf(loc, vb);
1522 	vb->vb_xv =
1523 		(struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1524 							nameval_offset +
1525 							name_size);
1526 }
1527 
1528 static int ocfs2_xa_block_journal_access(handle_t *handle,
1529 					 struct ocfs2_xa_loc *loc, int type)
1530 {
1531 	struct buffer_head *bh = loc->xl_storage;
1532 	ocfs2_journal_access_func access;
1533 
1534 	if (loc->xl_size == (bh->b_size -
1535 			     offsetof(struct ocfs2_xattr_block,
1536 				      xb_attrs.xb_header)))
1537 		access = ocfs2_journal_access_xb;
1538 	else
1539 		access = ocfs2_journal_access_di;
1540 	return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1541 }
1542 
1543 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1544 					 struct ocfs2_xa_loc *loc)
1545 {
1546 	struct buffer_head *bh = loc->xl_storage;
1547 
1548 	ocfs2_journal_dirty(handle, bh);
1549 }
1550 
1551 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1552 					   int offset)
1553 {
1554 	return (char *)loc->xl_header + offset;
1555 }
1556 
1557 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1558 				    struct ocfs2_xattr_info *xi)
1559 {
1560 	/*
1561 	 * Block storage is strict.  If the sizes aren't exact, we will
1562 	 * remove the old one and reinsert the new.
1563 	 */
1564 	return namevalue_size_xe(loc->xl_entry) ==
1565 		namevalue_size_xi(xi);
1566 }
1567 
1568 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1569 {
1570 	struct ocfs2_xattr_header *xh = loc->xl_header;
1571 	int i, count = le16_to_cpu(xh->xh_count);
1572 	int offset, free_start = loc->xl_size;
1573 
1574 	for (i = 0; i < count; i++) {
1575 		offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1576 		if (offset < free_start)
1577 			free_start = offset;
1578 	}
1579 
1580 	return free_start;
1581 }
1582 
1583 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1584 				      struct ocfs2_xattr_info *xi)
1585 {
1586 	int count = le16_to_cpu(loc->xl_header->xh_count);
1587 	int free_start = ocfs2_xa_get_free_start(loc);
1588 	int needed_space = ocfs2_xi_entry_usage(xi);
1589 
1590 	/*
1591 	 * Block storage will reclaim the original entry before inserting
1592 	 * the new value, so we only need the difference.  If the new
1593 	 * entry is smaller than the old one, we don't need anything.
1594 	 */
1595 	if (loc->xl_entry) {
1596 		/* Don't need space if we're reusing! */
1597 		if (ocfs2_xa_can_reuse_entry(loc, xi))
1598 			needed_space = 0;
1599 		else
1600 			needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1601 	}
1602 	if (needed_space < 0)
1603 		needed_space = 0;
1604 	return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1605 }
1606 
1607 /*
1608  * Block storage for xattrs keeps the name+value pairs compacted.  When
1609  * we remove one, we have to shift any that preceded it towards the end.
1610  */
1611 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1612 {
1613 	int i, offset;
1614 	int namevalue_offset, first_namevalue_offset, namevalue_size;
1615 	struct ocfs2_xattr_entry *entry = loc->xl_entry;
1616 	struct ocfs2_xattr_header *xh = loc->xl_header;
1617 	int count = le16_to_cpu(xh->xh_count);
1618 
1619 	namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1620 	namevalue_size = namevalue_size_xe(entry);
1621 	first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1622 
1623 	/* Shift the name+value pairs */
1624 	memmove((char *)xh + first_namevalue_offset + namevalue_size,
1625 		(char *)xh + first_namevalue_offset,
1626 		namevalue_offset - first_namevalue_offset);
1627 	memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1628 
1629 	/* Now tell xh->xh_entries about it */
1630 	for (i = 0; i < count; i++) {
1631 		offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1632 		if (offset <= namevalue_offset)
1633 			le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1634 				     namevalue_size);
1635 	}
1636 
1637 	/*
1638 	 * Note that we don't update xh_free_start or xh_name_value_len
1639 	 * because they're not used in block-stored xattrs.
1640 	 */
1641 }
1642 
1643 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1644 {
1645 	int count = le16_to_cpu(loc->xl_header->xh_count);
1646 	loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1647 	le16_add_cpu(&loc->xl_header->xh_count, 1);
1648 	memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1649 }
1650 
1651 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1652 {
1653 	int free_start = ocfs2_xa_get_free_start(loc);
1654 
1655 	loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1656 }
1657 
1658 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1659 					  struct ocfs2_xattr_value_buf *vb)
1660 {
1661 	struct buffer_head *bh = loc->xl_storage;
1662 
1663 	if (loc->xl_size == (bh->b_size -
1664 			     offsetof(struct ocfs2_xattr_block,
1665 				      xb_attrs.xb_header)))
1666 		vb->vb_access = ocfs2_journal_access_xb;
1667 	else
1668 		vb->vb_access = ocfs2_journal_access_di;
1669 	vb->vb_bh = bh;
1670 }
1671 
1672 /*
1673  * Operations for xattrs stored in blocks.  This includes inline inode
1674  * storage and unindexed ocfs2_xattr_blocks.
1675  */
1676 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1677 	.xlo_journal_access	= ocfs2_xa_block_journal_access,
1678 	.xlo_journal_dirty	= ocfs2_xa_block_journal_dirty,
1679 	.xlo_offset_pointer	= ocfs2_xa_block_offset_pointer,
1680 	.xlo_check_space	= ocfs2_xa_block_check_space,
1681 	.xlo_can_reuse		= ocfs2_xa_block_can_reuse,
1682 	.xlo_get_free_start	= ocfs2_xa_block_get_free_start,
1683 	.xlo_wipe_namevalue	= ocfs2_xa_block_wipe_namevalue,
1684 	.xlo_add_entry		= ocfs2_xa_block_add_entry,
1685 	.xlo_add_namevalue	= ocfs2_xa_block_add_namevalue,
1686 	.xlo_fill_value_buf	= ocfs2_xa_block_fill_value_buf,
1687 };
1688 
1689 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1690 					  struct ocfs2_xa_loc *loc, int type)
1691 {
1692 	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1693 
1694 	return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1695 }
1696 
1697 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1698 					  struct ocfs2_xa_loc *loc)
1699 {
1700 	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1701 
1702 	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1703 }
1704 
1705 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1706 					    int offset)
1707 {
1708 	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1709 	int block, block_offset;
1710 
1711 	/* The header is at the front of the bucket */
1712 	block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1713 	block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1714 
1715 	return bucket_block(bucket, block) + block_offset;
1716 }
1717 
1718 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1719 				     struct ocfs2_xattr_info *xi)
1720 {
1721 	return namevalue_size_xe(loc->xl_entry) >=
1722 		namevalue_size_xi(xi);
1723 }
1724 
1725 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1726 {
1727 	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1728 	return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1729 }
1730 
1731 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1732 					 int free_start, int size)
1733 {
1734 	/*
1735 	 * We need to make sure that the name+value pair fits within
1736 	 * one block.
1737 	 */
1738 	if (((free_start - size) >> sb->s_blocksize_bits) !=
1739 	    ((free_start - 1) >> sb->s_blocksize_bits))
1740 		free_start -= free_start % sb->s_blocksize;
1741 
1742 	return free_start;
1743 }
1744 
1745 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1746 				       struct ocfs2_xattr_info *xi)
1747 {
1748 	int rc;
1749 	int count = le16_to_cpu(loc->xl_header->xh_count);
1750 	int free_start = ocfs2_xa_get_free_start(loc);
1751 	int needed_space = ocfs2_xi_entry_usage(xi);
1752 	int size = namevalue_size_xi(xi);
1753 	struct super_block *sb = loc->xl_inode->i_sb;
1754 
1755 	/*
1756 	 * Bucket storage does not reclaim name+value pairs it cannot
1757 	 * reuse.  They live as holes until the bucket fills, and then
1758 	 * the bucket is defragmented.  However, the bucket can reclaim
1759 	 * the ocfs2_xattr_entry.
1760 	 */
1761 	if (loc->xl_entry) {
1762 		/* Don't need space if we're reusing! */
1763 		if (ocfs2_xa_can_reuse_entry(loc, xi))
1764 			needed_space = 0;
1765 		else
1766 			needed_space -= sizeof(struct ocfs2_xattr_entry);
1767 	}
1768 	BUG_ON(needed_space < 0);
1769 
1770 	if (free_start < size) {
1771 		if (needed_space)
1772 			return -ENOSPC;
1773 	} else {
1774 		/*
1775 		 * First we check if it would fit in the first place.
1776 		 * Below, we align the free start to a block.  This may
1777 		 * slide us below the minimum gap.  By checking unaligned
1778 		 * first, we avoid that error.
1779 		 */
1780 		rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1781 						 count);
1782 		if (rc)
1783 			return rc;
1784 		free_start = ocfs2_bucket_align_free_start(sb, free_start,
1785 							   size);
1786 	}
1787 	return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1788 }
1789 
1790 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1791 {
1792 	le16_add_cpu(&loc->xl_header->xh_name_value_len,
1793 		     -namevalue_size_xe(loc->xl_entry));
1794 }
1795 
1796 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1797 {
1798 	struct ocfs2_xattr_header *xh = loc->xl_header;
1799 	int count = le16_to_cpu(xh->xh_count);
1800 	int low = 0, high = count - 1, tmp;
1801 	struct ocfs2_xattr_entry *tmp_xe;
1802 
1803 	/*
1804 	 * We keep buckets sorted by name_hash, so we need to find
1805 	 * our insert place.
1806 	 */
1807 	while (low <= high && count) {
1808 		tmp = (low + high) / 2;
1809 		tmp_xe = &xh->xh_entries[tmp];
1810 
1811 		if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1812 			low = tmp + 1;
1813 		else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1814 			high = tmp - 1;
1815 		else {
1816 			low = tmp;
1817 			break;
1818 		}
1819 	}
1820 
1821 	if (low != count)
1822 		memmove(&xh->xh_entries[low + 1],
1823 			&xh->xh_entries[low],
1824 			((count - low) * sizeof(struct ocfs2_xattr_entry)));
1825 
1826 	le16_add_cpu(&xh->xh_count, 1);
1827 	loc->xl_entry = &xh->xh_entries[low];
1828 	memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1829 }
1830 
1831 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1832 {
1833 	int free_start = ocfs2_xa_get_free_start(loc);
1834 	struct ocfs2_xattr_header *xh = loc->xl_header;
1835 	struct super_block *sb = loc->xl_inode->i_sb;
1836 	int nameval_offset;
1837 
1838 	free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1839 	nameval_offset = free_start - size;
1840 	loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1841 	xh->xh_free_start = cpu_to_le16(nameval_offset);
1842 	le16_add_cpu(&xh->xh_name_value_len, size);
1843 
1844 }
1845 
1846 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1847 					   struct ocfs2_xattr_value_buf *vb)
1848 {
1849 	struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1850 	struct super_block *sb = loc->xl_inode->i_sb;
1851 	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1852 	int size = namevalue_size_xe(loc->xl_entry);
1853 	int block_offset = nameval_offset >> sb->s_blocksize_bits;
1854 
1855 	/* Values are not allowed to straddle block boundaries */
1856 	BUG_ON(block_offset !=
1857 	       ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1858 	/* We expect the bucket to be filled in */
1859 	BUG_ON(!bucket->bu_bhs[block_offset]);
1860 
1861 	vb->vb_access = ocfs2_journal_access;
1862 	vb->vb_bh = bucket->bu_bhs[block_offset];
1863 }
1864 
1865 /* Operations for xattrs stored in buckets. */
1866 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1867 	.xlo_journal_access	= ocfs2_xa_bucket_journal_access,
1868 	.xlo_journal_dirty	= ocfs2_xa_bucket_journal_dirty,
1869 	.xlo_offset_pointer	= ocfs2_xa_bucket_offset_pointer,
1870 	.xlo_check_space	= ocfs2_xa_bucket_check_space,
1871 	.xlo_can_reuse		= ocfs2_xa_bucket_can_reuse,
1872 	.xlo_get_free_start	= ocfs2_xa_bucket_get_free_start,
1873 	.xlo_wipe_namevalue	= ocfs2_xa_bucket_wipe_namevalue,
1874 	.xlo_add_entry		= ocfs2_xa_bucket_add_entry,
1875 	.xlo_add_namevalue	= ocfs2_xa_bucket_add_namevalue,
1876 	.xlo_fill_value_buf	= ocfs2_xa_bucket_fill_value_buf,
1877 };
1878 
1879 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1880 {
1881 	struct ocfs2_xattr_value_buf vb;
1882 
1883 	if (ocfs2_xattr_is_local(loc->xl_entry))
1884 		return 0;
1885 
1886 	ocfs2_xa_fill_value_buf(loc, &vb);
1887 	return le32_to_cpu(vb.vb_xv->xr_clusters);
1888 }
1889 
1890 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1891 				   struct ocfs2_xattr_set_ctxt *ctxt)
1892 {
1893 	int trunc_rc, access_rc;
1894 	struct ocfs2_xattr_value_buf vb;
1895 
1896 	ocfs2_xa_fill_value_buf(loc, &vb);
1897 	trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1898 					      ctxt);
1899 
1900 	/*
1901 	 * The caller of ocfs2_xa_value_truncate() has already called
1902 	 * ocfs2_xa_journal_access on the loc.  However, The truncate code
1903 	 * calls ocfs2_extend_trans().  This may commit the previous
1904 	 * transaction and open a new one.  If this is a bucket, truncate
1905 	 * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1906 	 * the caller is expecting to dirty the entire bucket.  So we must
1907 	 * reset the journal work.  We do this even if truncate has failed,
1908 	 * as it could have failed after committing the extend.
1909 	 */
1910 	access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1911 					    OCFS2_JOURNAL_ACCESS_WRITE);
1912 
1913 	/* Errors in truncate take precedence */
1914 	return trunc_rc ? trunc_rc : access_rc;
1915 }
1916 
1917 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1918 {
1919 	int index, count;
1920 	struct ocfs2_xattr_header *xh = loc->xl_header;
1921 	struct ocfs2_xattr_entry *entry = loc->xl_entry;
1922 
1923 	ocfs2_xa_wipe_namevalue(loc);
1924 	loc->xl_entry = NULL;
1925 
1926 	le16_add_cpu(&xh->xh_count, -1);
1927 	count = le16_to_cpu(xh->xh_count);
1928 
1929 	/*
1930 	 * Only zero out the entry if there are more remaining.  This is
1931 	 * important for an empty bucket, as it keeps track of the
1932 	 * bucket's hash value.  It doesn't hurt empty block storage.
1933 	 */
1934 	if (count) {
1935 		index = ((char *)entry - (char *)&xh->xh_entries) /
1936 			sizeof(struct ocfs2_xattr_entry);
1937 		memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1938 			(count - index) * sizeof(struct ocfs2_xattr_entry));
1939 		memset(&xh->xh_entries[count], 0,
1940 		       sizeof(struct ocfs2_xattr_entry));
1941 	}
1942 }
1943 
1944 /*
1945  * If we have a problem adjusting the size of an external value during
1946  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1947  * in an intermediate state.  For example, the value may be partially
1948  * truncated.
1949  *
1950  * If the value tree hasn't changed, the extend/truncate went nowhere.
1951  * We have nothing to do.  The caller can treat it as a straight error.
1952  *
1953  * If the value tree got partially truncated, we now have a corrupted
1954  * extended attribute.  We're going to wipe its entry and leak the
1955  * clusters.  Better to leak some storage than leave a corrupt entry.
1956  *
1957  * If the value tree grew, it obviously didn't grow enough for the
1958  * new entry.  We're not going to try and reclaim those clusters either.
1959  * If there was already an external value there (orig_clusters != 0),
1960  * the new clusters are attached safely and we can just leave the old
1961  * value in place.  If there was no external value there, we remove
1962  * the entry.
1963  *
1964  * This way, the xattr block we store in the journal will be consistent.
1965  * If the size change broke because of the journal, no changes will hit
1966  * disk anyway.
1967  */
1968 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1969 					    const char *what,
1970 					    unsigned int orig_clusters)
1971 {
1972 	unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1973 	char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1974 				le16_to_cpu(loc->xl_entry->xe_name_offset));
1975 
1976 	if (new_clusters < orig_clusters) {
1977 		mlog(ML_ERROR,
1978 		     "Partial truncate while %s xattr %.*s.  Leaking "
1979 		     "%u clusters and removing the entry\n",
1980 		     what, loc->xl_entry->xe_name_len, nameval_buf,
1981 		     orig_clusters - new_clusters);
1982 		ocfs2_xa_remove_entry(loc);
1983 	} else if (!orig_clusters) {
1984 		mlog(ML_ERROR,
1985 		     "Unable to allocate an external value for xattr "
1986 		     "%.*s safely.  Leaking %u clusters and removing the "
1987 		     "entry\n",
1988 		     loc->xl_entry->xe_name_len, nameval_buf,
1989 		     new_clusters - orig_clusters);
1990 		ocfs2_xa_remove_entry(loc);
1991 	} else if (new_clusters > orig_clusters)
1992 		mlog(ML_ERROR,
1993 		     "Unable to grow xattr %.*s safely.  %u new clusters "
1994 		     "have been added, but the value will not be "
1995 		     "modified\n",
1996 		     loc->xl_entry->xe_name_len, nameval_buf,
1997 		     new_clusters - orig_clusters);
1998 }
1999 
2000 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2001 			   struct ocfs2_xattr_set_ctxt *ctxt)
2002 {
2003 	int rc = 0;
2004 	unsigned int orig_clusters;
2005 
2006 	if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2007 		orig_clusters = ocfs2_xa_value_clusters(loc);
2008 		rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2009 		if (rc) {
2010 			mlog_errno(rc);
2011 			/*
2012 			 * Since this is remove, we can return 0 if
2013 			 * ocfs2_xa_cleanup_value_truncate() is going to
2014 			 * wipe the entry anyway.  So we check the
2015 			 * cluster count as well.
2016 			 */
2017 			if (orig_clusters != ocfs2_xa_value_clusters(loc))
2018 				rc = 0;
2019 			ocfs2_xa_cleanup_value_truncate(loc, "removing",
2020 							orig_clusters);
2021 			if (rc)
2022 				goto out;
2023 		}
2024 	}
2025 
2026 	ocfs2_xa_remove_entry(loc);
2027 
2028 out:
2029 	return rc;
2030 }
2031 
2032 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2033 {
2034 	int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2035 	char *nameval_buf;
2036 
2037 	nameval_buf = ocfs2_xa_offset_pointer(loc,
2038 				le16_to_cpu(loc->xl_entry->xe_name_offset));
2039 	memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2040 }
2041 
2042 /*
2043  * Take an existing entry and make it ready for the new value.  This
2044  * won't allocate space, but it may free space.  It should be ready for
2045  * ocfs2_xa_prepare_entry() to finish the work.
2046  */
2047 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2048 				struct ocfs2_xattr_info *xi,
2049 				struct ocfs2_xattr_set_ctxt *ctxt)
2050 {
2051 	int rc = 0;
2052 	int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2053 	unsigned int orig_clusters;
2054 	char *nameval_buf;
2055 	int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2056 	int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2057 
2058 	BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2059 	       name_size);
2060 
2061 	nameval_buf = ocfs2_xa_offset_pointer(loc,
2062 				le16_to_cpu(loc->xl_entry->xe_name_offset));
2063 	if (xe_local) {
2064 		memset(nameval_buf + name_size, 0,
2065 		       namevalue_size_xe(loc->xl_entry) - name_size);
2066 		if (!xi_local)
2067 			ocfs2_xa_install_value_root(loc);
2068 	} else {
2069 		orig_clusters = ocfs2_xa_value_clusters(loc);
2070 		if (xi_local) {
2071 			rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2072 			if (rc < 0)
2073 				mlog_errno(rc);
2074 			else
2075 				memset(nameval_buf + name_size, 0,
2076 				       namevalue_size_xe(loc->xl_entry) -
2077 				       name_size);
2078 		} else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2079 			   xi->xi_value_len) {
2080 			rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2081 						     ctxt);
2082 			if (rc < 0)
2083 				mlog_errno(rc);
2084 		}
2085 
2086 		if (rc) {
2087 			ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2088 							orig_clusters);
2089 			goto out;
2090 		}
2091 	}
2092 
2093 	loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2094 	ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2095 
2096 out:
2097 	return rc;
2098 }
2099 
2100 /*
2101  * Prepares loc->xl_entry to receive the new xattr.  This includes
2102  * properly setting up the name+value pair region.  If loc->xl_entry
2103  * already exists, it will take care of modifying it appropriately.
2104  *
2105  * Note that this modifies the data.  You did journal_access already,
2106  * right?
2107  */
2108 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2109 				  struct ocfs2_xattr_info *xi,
2110 				  u32 name_hash,
2111 				  struct ocfs2_xattr_set_ctxt *ctxt)
2112 {
2113 	int rc = 0;
2114 	unsigned int orig_clusters;
2115 	__le64 orig_value_size = 0;
2116 
2117 	rc = ocfs2_xa_check_space(loc, xi);
2118 	if (rc)
2119 		goto out;
2120 
2121 	if (loc->xl_entry) {
2122 		if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2123 			orig_value_size = loc->xl_entry->xe_value_size;
2124 			rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2125 			if (rc)
2126 				goto out;
2127 			goto alloc_value;
2128 		}
2129 
2130 		if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2131 			orig_clusters = ocfs2_xa_value_clusters(loc);
2132 			rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2133 			if (rc) {
2134 				mlog_errno(rc);
2135 				ocfs2_xa_cleanup_value_truncate(loc,
2136 								"overwriting",
2137 								orig_clusters);
2138 				goto out;
2139 			}
2140 		}
2141 		ocfs2_xa_wipe_namevalue(loc);
2142 	} else
2143 		ocfs2_xa_add_entry(loc, name_hash);
2144 
2145 	/*
2146 	 * If we get here, we have a blank entry.  Fill it.  We grow our
2147 	 * name+value pair back from the end.
2148 	 */
2149 	ocfs2_xa_add_namevalue(loc, xi);
2150 	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2151 		ocfs2_xa_install_value_root(loc);
2152 
2153 alloc_value:
2154 	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2155 		orig_clusters = ocfs2_xa_value_clusters(loc);
2156 		rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2157 		if (rc < 0) {
2158 			ctxt->set_abort = 1;
2159 			ocfs2_xa_cleanup_value_truncate(loc, "growing",
2160 							orig_clusters);
2161 			/*
2162 			 * If we were growing an existing value,
2163 			 * ocfs2_xa_cleanup_value_truncate() won't remove
2164 			 * the entry. We need to restore the original value
2165 			 * size.
2166 			 */
2167 			if (loc->xl_entry) {
2168 				BUG_ON(!orig_value_size);
2169 				loc->xl_entry->xe_value_size = orig_value_size;
2170 			}
2171 			mlog_errno(rc);
2172 		}
2173 	}
2174 
2175 out:
2176 	return rc;
2177 }
2178 
2179 /*
2180  * Store the value portion of the name+value pair.  This will skip
2181  * values that are stored externally.  Their tree roots were set up
2182  * by ocfs2_xa_prepare_entry().
2183  */
2184 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2185 				struct ocfs2_xattr_info *xi,
2186 				struct ocfs2_xattr_set_ctxt *ctxt)
2187 {
2188 	int rc = 0;
2189 	int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2190 	int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2191 	char *nameval_buf;
2192 	struct ocfs2_xattr_value_buf vb;
2193 
2194 	nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2195 	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2196 		ocfs2_xa_fill_value_buf(loc, &vb);
2197 		rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2198 						     ctxt->handle, &vb,
2199 						     xi->xi_value,
2200 						     xi->xi_value_len);
2201 	} else
2202 		memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2203 
2204 	return rc;
2205 }
2206 
2207 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2208 			struct ocfs2_xattr_info *xi,
2209 			struct ocfs2_xattr_set_ctxt *ctxt)
2210 {
2211 	int ret;
2212 	u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2213 					      xi->xi_name_len);
2214 
2215 	ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2216 				      OCFS2_JOURNAL_ACCESS_WRITE);
2217 	if (ret) {
2218 		mlog_errno(ret);
2219 		goto out;
2220 	}
2221 
2222 	/*
2223 	 * From here on out, everything is going to modify the buffer a
2224 	 * little.  Errors are going to leave the xattr header in a
2225 	 * sane state.  Thus, even with errors we dirty the sucker.
2226 	 */
2227 
2228 	/* Don't worry, we are never called with !xi_value and !xl_entry */
2229 	if (!xi->xi_value) {
2230 		ret = ocfs2_xa_remove(loc, ctxt);
2231 		goto out_dirty;
2232 	}
2233 
2234 	ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2235 	if (ret) {
2236 		if (ret != -ENOSPC)
2237 			mlog_errno(ret);
2238 		goto out_dirty;
2239 	}
2240 
2241 	ret = ocfs2_xa_store_value(loc, xi, ctxt);
2242 	if (ret)
2243 		mlog_errno(ret);
2244 
2245 out_dirty:
2246 	ocfs2_xa_journal_dirty(ctxt->handle, loc);
2247 
2248 out:
2249 	return ret;
2250 }
2251 
2252 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2253 				     struct inode *inode,
2254 				     struct buffer_head *bh,
2255 				     struct ocfs2_xattr_entry *entry)
2256 {
2257 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2258 
2259 	BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2260 
2261 	loc->xl_inode = inode;
2262 	loc->xl_ops = &ocfs2_xa_block_loc_ops;
2263 	loc->xl_storage = bh;
2264 	loc->xl_entry = entry;
2265 	loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2266 	loc->xl_header =
2267 		(struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2268 					      loc->xl_size);
2269 }
2270 
2271 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2272 					  struct inode *inode,
2273 					  struct buffer_head *bh,
2274 					  struct ocfs2_xattr_entry *entry)
2275 {
2276 	struct ocfs2_xattr_block *xb =
2277 		(struct ocfs2_xattr_block *)bh->b_data;
2278 
2279 	BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2280 
2281 	loc->xl_inode = inode;
2282 	loc->xl_ops = &ocfs2_xa_block_loc_ops;
2283 	loc->xl_storage = bh;
2284 	loc->xl_header = &(xb->xb_attrs.xb_header);
2285 	loc->xl_entry = entry;
2286 	loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2287 					     xb_attrs.xb_header);
2288 }
2289 
2290 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2291 					   struct ocfs2_xattr_bucket *bucket,
2292 					   struct ocfs2_xattr_entry *entry)
2293 {
2294 	loc->xl_inode = bucket->bu_inode;
2295 	loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2296 	loc->xl_storage = bucket;
2297 	loc->xl_header = bucket_xh(bucket);
2298 	loc->xl_entry = entry;
2299 	loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2300 }
2301 
2302 /*
2303  * In xattr remove, if it is stored outside and refcounted, we may have
2304  * the chance to split the refcount tree. So need the allocators.
2305  */
2306 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2307 					struct ocfs2_xattr_value_root *xv,
2308 					struct ocfs2_caching_info *ref_ci,
2309 					struct buffer_head *ref_root_bh,
2310 					struct ocfs2_alloc_context **meta_ac,
2311 					int *ref_credits)
2312 {
2313 	int ret, meta_add = 0;
2314 	u32 p_cluster, num_clusters;
2315 	unsigned int ext_flags;
2316 
2317 	*ref_credits = 0;
2318 	ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2319 				       &num_clusters,
2320 				       &xv->xr_list,
2321 				       &ext_flags);
2322 	if (ret) {
2323 		mlog_errno(ret);
2324 		goto out;
2325 	}
2326 
2327 	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2328 		goto out;
2329 
2330 	ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2331 						 ref_root_bh, xv,
2332 						 &meta_add, ref_credits);
2333 	if (ret) {
2334 		mlog_errno(ret);
2335 		goto out;
2336 	}
2337 
2338 	ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2339 						meta_add, meta_ac);
2340 	if (ret)
2341 		mlog_errno(ret);
2342 
2343 out:
2344 	return ret;
2345 }
2346 
2347 static int ocfs2_remove_value_outside(struct inode*inode,
2348 				      struct ocfs2_xattr_value_buf *vb,
2349 				      struct ocfs2_xattr_header *header,
2350 				      struct ocfs2_caching_info *ref_ci,
2351 				      struct buffer_head *ref_root_bh)
2352 {
2353 	int ret = 0, i, ref_credits;
2354 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2355 	struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2356 	void *val;
2357 
2358 	ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2359 
2360 	for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2361 		struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2362 
2363 		if (ocfs2_xattr_is_local(entry))
2364 			continue;
2365 
2366 		val = (void *)header +
2367 			le16_to_cpu(entry->xe_name_offset);
2368 		vb->vb_xv = (struct ocfs2_xattr_value_root *)
2369 			(val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2370 
2371 		ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2372 							 ref_ci, ref_root_bh,
2373 							 &ctxt.meta_ac,
2374 							 &ref_credits);
2375 
2376 		ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2377 					ocfs2_remove_extent_credits(osb->sb));
2378 		if (IS_ERR(ctxt.handle)) {
2379 			ret = PTR_ERR(ctxt.handle);
2380 			mlog_errno(ret);
2381 			break;
2382 		}
2383 
2384 		ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2385 
2386 		ocfs2_commit_trans(osb, ctxt.handle);
2387 		if (ctxt.meta_ac) {
2388 			ocfs2_free_alloc_context(ctxt.meta_ac);
2389 			ctxt.meta_ac = NULL;
2390 		}
2391 
2392 		if (ret < 0) {
2393 			mlog_errno(ret);
2394 			break;
2395 		}
2396 
2397 	}
2398 
2399 	if (ctxt.meta_ac)
2400 		ocfs2_free_alloc_context(ctxt.meta_ac);
2401 	ocfs2_schedule_truncate_log_flush(osb, 1);
2402 	ocfs2_run_deallocs(osb, &ctxt.dealloc);
2403 	return ret;
2404 }
2405 
2406 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2407 				    struct buffer_head *di_bh,
2408 				    struct ocfs2_caching_info *ref_ci,
2409 				    struct buffer_head *ref_root_bh)
2410 {
2411 
2412 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2413 	struct ocfs2_xattr_header *header;
2414 	int ret;
2415 	struct ocfs2_xattr_value_buf vb = {
2416 		.vb_bh = di_bh,
2417 		.vb_access = ocfs2_journal_access_di,
2418 	};
2419 
2420 	header = (struct ocfs2_xattr_header *)
2421 		 ((void *)di + inode->i_sb->s_blocksize -
2422 		 le16_to_cpu(di->i_xattr_inline_size));
2423 
2424 	ret = ocfs2_remove_value_outside(inode, &vb, header,
2425 					 ref_ci, ref_root_bh);
2426 
2427 	return ret;
2428 }
2429 
2430 struct ocfs2_rm_xattr_bucket_para {
2431 	struct ocfs2_caching_info *ref_ci;
2432 	struct buffer_head *ref_root_bh;
2433 };
2434 
2435 static int ocfs2_xattr_block_remove(struct inode *inode,
2436 				    struct buffer_head *blk_bh,
2437 				    struct ocfs2_caching_info *ref_ci,
2438 				    struct buffer_head *ref_root_bh)
2439 {
2440 	struct ocfs2_xattr_block *xb;
2441 	int ret = 0;
2442 	struct ocfs2_xattr_value_buf vb = {
2443 		.vb_bh = blk_bh,
2444 		.vb_access = ocfs2_journal_access_xb,
2445 	};
2446 	struct ocfs2_rm_xattr_bucket_para args = {
2447 		.ref_ci = ref_ci,
2448 		.ref_root_bh = ref_root_bh,
2449 	};
2450 
2451 	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2452 	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2453 		struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2454 		ret = ocfs2_remove_value_outside(inode, &vb, header,
2455 						 ref_ci, ref_root_bh);
2456 	} else
2457 		ret = ocfs2_iterate_xattr_index_block(inode,
2458 						blk_bh,
2459 						ocfs2_rm_xattr_cluster,
2460 						&args);
2461 
2462 	return ret;
2463 }
2464 
2465 static int ocfs2_xattr_free_block(struct inode *inode,
2466 				  u64 block,
2467 				  struct ocfs2_caching_info *ref_ci,
2468 				  struct buffer_head *ref_root_bh)
2469 {
2470 	struct inode *xb_alloc_inode;
2471 	struct buffer_head *xb_alloc_bh = NULL;
2472 	struct buffer_head *blk_bh = NULL;
2473 	struct ocfs2_xattr_block *xb;
2474 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2475 	handle_t *handle;
2476 	int ret = 0;
2477 	u64 blk, bg_blkno;
2478 	u16 bit;
2479 
2480 	ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2481 	if (ret < 0) {
2482 		mlog_errno(ret);
2483 		goto out;
2484 	}
2485 
2486 	ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2487 	if (ret < 0) {
2488 		mlog_errno(ret);
2489 		goto out;
2490 	}
2491 
2492 	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2493 	blk = le64_to_cpu(xb->xb_blkno);
2494 	bit = le16_to_cpu(xb->xb_suballoc_bit);
2495 	if (xb->xb_suballoc_loc)
2496 		bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2497 	else
2498 		bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2499 
2500 	xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2501 				EXTENT_ALLOC_SYSTEM_INODE,
2502 				le16_to_cpu(xb->xb_suballoc_slot));
2503 	if (!xb_alloc_inode) {
2504 		ret = -ENOMEM;
2505 		mlog_errno(ret);
2506 		goto out;
2507 	}
2508 	mutex_lock(&xb_alloc_inode->i_mutex);
2509 
2510 	ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2511 	if (ret < 0) {
2512 		mlog_errno(ret);
2513 		goto out_mutex;
2514 	}
2515 
2516 	handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2517 	if (IS_ERR(handle)) {
2518 		ret = PTR_ERR(handle);
2519 		mlog_errno(ret);
2520 		goto out_unlock;
2521 	}
2522 
2523 	ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2524 				       bit, bg_blkno, 1);
2525 	if (ret < 0)
2526 		mlog_errno(ret);
2527 
2528 	ocfs2_commit_trans(osb, handle);
2529 out_unlock:
2530 	ocfs2_inode_unlock(xb_alloc_inode, 1);
2531 	brelse(xb_alloc_bh);
2532 out_mutex:
2533 	mutex_unlock(&xb_alloc_inode->i_mutex);
2534 	iput(xb_alloc_inode);
2535 out:
2536 	brelse(blk_bh);
2537 	return ret;
2538 }
2539 
2540 /*
2541  * ocfs2_xattr_remove()
2542  *
2543  * Free extended attribute resources associated with this inode.
2544  */
2545 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2546 {
2547 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2548 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2549 	struct ocfs2_refcount_tree *ref_tree = NULL;
2550 	struct buffer_head *ref_root_bh = NULL;
2551 	struct ocfs2_caching_info *ref_ci = NULL;
2552 	handle_t *handle;
2553 	int ret;
2554 
2555 	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2556 		return 0;
2557 
2558 	if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2559 		return 0;
2560 
2561 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2562 		ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2563 					       le64_to_cpu(di->i_refcount_loc),
2564 					       1, &ref_tree, &ref_root_bh);
2565 		if (ret) {
2566 			mlog_errno(ret);
2567 			goto out;
2568 		}
2569 		ref_ci = &ref_tree->rf_ci;
2570 
2571 	}
2572 
2573 	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2574 		ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2575 					       ref_ci, ref_root_bh);
2576 		if (ret < 0) {
2577 			mlog_errno(ret);
2578 			goto out;
2579 		}
2580 	}
2581 
2582 	if (di->i_xattr_loc) {
2583 		ret = ocfs2_xattr_free_block(inode,
2584 					     le64_to_cpu(di->i_xattr_loc),
2585 					     ref_ci, ref_root_bh);
2586 		if (ret < 0) {
2587 			mlog_errno(ret);
2588 			goto out;
2589 		}
2590 	}
2591 
2592 	handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2593 				   OCFS2_INODE_UPDATE_CREDITS);
2594 	if (IS_ERR(handle)) {
2595 		ret = PTR_ERR(handle);
2596 		mlog_errno(ret);
2597 		goto out;
2598 	}
2599 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2600 				      OCFS2_JOURNAL_ACCESS_WRITE);
2601 	if (ret) {
2602 		mlog_errno(ret);
2603 		goto out_commit;
2604 	}
2605 
2606 	di->i_xattr_loc = 0;
2607 
2608 	spin_lock(&oi->ip_lock);
2609 	oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2610 	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2611 	spin_unlock(&oi->ip_lock);
2612 	ocfs2_update_inode_fsync_trans(handle, inode, 0);
2613 
2614 	ocfs2_journal_dirty(handle, di_bh);
2615 out_commit:
2616 	ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2617 out:
2618 	if (ref_tree)
2619 		ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2620 	brelse(ref_root_bh);
2621 	return ret;
2622 }
2623 
2624 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2625 					struct ocfs2_dinode *di)
2626 {
2627 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2628 	unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2629 	int free;
2630 
2631 	if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2632 		return 0;
2633 
2634 	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2635 		struct ocfs2_inline_data *idata = &di->id2.i_data;
2636 		free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2637 	} else if (ocfs2_inode_is_fast_symlink(inode)) {
2638 		free = ocfs2_fast_symlink_chars(inode->i_sb) -
2639 			le64_to_cpu(di->i_size);
2640 	} else {
2641 		struct ocfs2_extent_list *el = &di->id2.i_list;
2642 		free = (le16_to_cpu(el->l_count) -
2643 			le16_to_cpu(el->l_next_free_rec)) *
2644 			sizeof(struct ocfs2_extent_rec);
2645 	}
2646 	if (free >= xattrsize)
2647 		return 1;
2648 
2649 	return 0;
2650 }
2651 
2652 /*
2653  * ocfs2_xattr_ibody_find()
2654  *
2655  * Find extended attribute in inode block and
2656  * fill search info into struct ocfs2_xattr_search.
2657  */
2658 static int ocfs2_xattr_ibody_find(struct inode *inode,
2659 				  int name_index,
2660 				  const char *name,
2661 				  struct ocfs2_xattr_search *xs)
2662 {
2663 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2664 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2665 	int ret;
2666 	int has_space = 0;
2667 
2668 	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2669 		return 0;
2670 
2671 	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2672 		down_read(&oi->ip_alloc_sem);
2673 		has_space = ocfs2_xattr_has_space_inline(inode, di);
2674 		up_read(&oi->ip_alloc_sem);
2675 		if (!has_space)
2676 			return 0;
2677 	}
2678 
2679 	xs->xattr_bh = xs->inode_bh;
2680 	xs->end = (void *)di + inode->i_sb->s_blocksize;
2681 	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2682 		xs->header = (struct ocfs2_xattr_header *)
2683 			(xs->end - le16_to_cpu(di->i_xattr_inline_size));
2684 	else
2685 		xs->header = (struct ocfs2_xattr_header *)
2686 			(xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2687 	xs->base = (void *)xs->header;
2688 	xs->here = xs->header->xh_entries;
2689 
2690 	/* Find the named attribute. */
2691 	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2692 		ret = ocfs2_xattr_find_entry(name_index, name, xs);
2693 		if (ret && ret != -ENODATA)
2694 			return ret;
2695 		xs->not_found = ret;
2696 	}
2697 
2698 	return 0;
2699 }
2700 
2701 static int ocfs2_xattr_ibody_init(struct inode *inode,
2702 				  struct buffer_head *di_bh,
2703 				  struct ocfs2_xattr_set_ctxt *ctxt)
2704 {
2705 	int ret;
2706 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2707 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2708 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2709 	unsigned int xattrsize = osb->s_xattr_inline_size;
2710 
2711 	if (!ocfs2_xattr_has_space_inline(inode, di)) {
2712 		ret = -ENOSPC;
2713 		goto out;
2714 	}
2715 
2716 	ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2717 				      OCFS2_JOURNAL_ACCESS_WRITE);
2718 	if (ret) {
2719 		mlog_errno(ret);
2720 		goto out;
2721 	}
2722 
2723 	/*
2724 	 * Adjust extent record count or inline data size
2725 	 * to reserve space for extended attribute.
2726 	 */
2727 	if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2728 		struct ocfs2_inline_data *idata = &di->id2.i_data;
2729 		le16_add_cpu(&idata->id_count, -xattrsize);
2730 	} else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2731 		struct ocfs2_extent_list *el = &di->id2.i_list;
2732 		le16_add_cpu(&el->l_count, -(xattrsize /
2733 					     sizeof(struct ocfs2_extent_rec)));
2734 	}
2735 	di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2736 
2737 	spin_lock(&oi->ip_lock);
2738 	oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2739 	di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2740 	spin_unlock(&oi->ip_lock);
2741 
2742 	ocfs2_journal_dirty(ctxt->handle, di_bh);
2743 
2744 out:
2745 	return ret;
2746 }
2747 
2748 /*
2749  * ocfs2_xattr_ibody_set()
2750  *
2751  * Set, replace or remove an extended attribute into inode block.
2752  *
2753  */
2754 static int ocfs2_xattr_ibody_set(struct inode *inode,
2755 				 struct ocfs2_xattr_info *xi,
2756 				 struct ocfs2_xattr_search *xs,
2757 				 struct ocfs2_xattr_set_ctxt *ctxt)
2758 {
2759 	int ret;
2760 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
2761 	struct ocfs2_xa_loc loc;
2762 
2763 	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2764 		return -ENOSPC;
2765 
2766 	down_write(&oi->ip_alloc_sem);
2767 	if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2768 		ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2769 		if (ret) {
2770 			if (ret != -ENOSPC)
2771 				mlog_errno(ret);
2772 			goto out;
2773 		}
2774 	}
2775 
2776 	ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2777 				 xs->not_found ? NULL : xs->here);
2778 	ret = ocfs2_xa_set(&loc, xi, ctxt);
2779 	if (ret) {
2780 		if (ret != -ENOSPC)
2781 			mlog_errno(ret);
2782 		goto out;
2783 	}
2784 	xs->here = loc.xl_entry;
2785 
2786 out:
2787 	up_write(&oi->ip_alloc_sem);
2788 
2789 	return ret;
2790 }
2791 
2792 /*
2793  * ocfs2_xattr_block_find()
2794  *
2795  * Find extended attribute in external block and
2796  * fill search info into struct ocfs2_xattr_search.
2797  */
2798 static int ocfs2_xattr_block_find(struct inode *inode,
2799 				  int name_index,
2800 				  const char *name,
2801 				  struct ocfs2_xattr_search *xs)
2802 {
2803 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2804 	struct buffer_head *blk_bh = NULL;
2805 	struct ocfs2_xattr_block *xb;
2806 	int ret = 0;
2807 
2808 	if (!di->i_xattr_loc)
2809 		return ret;
2810 
2811 	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2812 				     &blk_bh);
2813 	if (ret < 0) {
2814 		mlog_errno(ret);
2815 		return ret;
2816 	}
2817 
2818 	xs->xattr_bh = blk_bh;
2819 	xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2820 
2821 	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2822 		xs->header = &xb->xb_attrs.xb_header;
2823 		xs->base = (void *)xs->header;
2824 		xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2825 		xs->here = xs->header->xh_entries;
2826 
2827 		ret = ocfs2_xattr_find_entry(name_index, name, xs);
2828 	} else
2829 		ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2830 						   name_index,
2831 						   name, xs);
2832 
2833 	if (ret && ret != -ENODATA) {
2834 		xs->xattr_bh = NULL;
2835 		goto cleanup;
2836 	}
2837 	xs->not_found = ret;
2838 	return 0;
2839 cleanup:
2840 	brelse(blk_bh);
2841 
2842 	return ret;
2843 }
2844 
2845 static int ocfs2_create_xattr_block(struct inode *inode,
2846 				    struct buffer_head *inode_bh,
2847 				    struct ocfs2_xattr_set_ctxt *ctxt,
2848 				    int indexed,
2849 				    struct buffer_head **ret_bh)
2850 {
2851 	int ret;
2852 	u16 suballoc_bit_start;
2853 	u32 num_got;
2854 	u64 suballoc_loc, first_blkno;
2855 	struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2856 	struct buffer_head *new_bh = NULL;
2857 	struct ocfs2_xattr_block *xblk;
2858 
2859 	ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2860 				      inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2861 	if (ret < 0) {
2862 		mlog_errno(ret);
2863 		goto end;
2864 	}
2865 
2866 	ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2867 				   &suballoc_loc, &suballoc_bit_start,
2868 				   &num_got, &first_blkno);
2869 	if (ret < 0) {
2870 		mlog_errno(ret);
2871 		goto end;
2872 	}
2873 
2874 	new_bh = sb_getblk(inode->i_sb, first_blkno);
2875 	if (!new_bh) {
2876 		ret = -ENOMEM;
2877 		mlog_errno(ret);
2878 		goto end;
2879 	}
2880 
2881 	ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2882 
2883 	ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2884 				      new_bh,
2885 				      OCFS2_JOURNAL_ACCESS_CREATE);
2886 	if (ret < 0) {
2887 		mlog_errno(ret);
2888 		goto end;
2889 	}
2890 
2891 	/* Initialize ocfs2_xattr_block */
2892 	xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2893 	memset(xblk, 0, inode->i_sb->s_blocksize);
2894 	strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2895 	xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2896 	xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2897 	xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2898 	xblk->xb_fs_generation =
2899 		cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2900 	xblk->xb_blkno = cpu_to_le64(first_blkno);
2901 	if (indexed) {
2902 		struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2903 		xr->xt_clusters = cpu_to_le32(1);
2904 		xr->xt_last_eb_blk = 0;
2905 		xr->xt_list.l_tree_depth = 0;
2906 		xr->xt_list.l_count = cpu_to_le16(
2907 					ocfs2_xattr_recs_per_xb(inode->i_sb));
2908 		xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2909 		xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2910 	}
2911 	ocfs2_journal_dirty(ctxt->handle, new_bh);
2912 
2913 	/* Add it to the inode */
2914 	di->i_xattr_loc = cpu_to_le64(first_blkno);
2915 
2916 	spin_lock(&OCFS2_I(inode)->ip_lock);
2917 	OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2918 	di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2919 	spin_unlock(&OCFS2_I(inode)->ip_lock);
2920 
2921 	ocfs2_journal_dirty(ctxt->handle, inode_bh);
2922 
2923 	*ret_bh = new_bh;
2924 	new_bh = NULL;
2925 
2926 end:
2927 	brelse(new_bh);
2928 	return ret;
2929 }
2930 
2931 /*
2932  * ocfs2_xattr_block_set()
2933  *
2934  * Set, replace or remove an extended attribute into external block.
2935  *
2936  */
2937 static int ocfs2_xattr_block_set(struct inode *inode,
2938 				 struct ocfs2_xattr_info *xi,
2939 				 struct ocfs2_xattr_search *xs,
2940 				 struct ocfs2_xattr_set_ctxt *ctxt)
2941 {
2942 	struct buffer_head *new_bh = NULL;
2943 	struct ocfs2_xattr_block *xblk = NULL;
2944 	int ret;
2945 	struct ocfs2_xa_loc loc;
2946 
2947 	if (!xs->xattr_bh) {
2948 		ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2949 					       0, &new_bh);
2950 		if (ret) {
2951 			mlog_errno(ret);
2952 			goto end;
2953 		}
2954 
2955 		xs->xattr_bh = new_bh;
2956 		xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2957 		xs->header = &xblk->xb_attrs.xb_header;
2958 		xs->base = (void *)xs->header;
2959 		xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2960 		xs->here = xs->header->xh_entries;
2961 	} else
2962 		xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2963 
2964 	if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2965 		ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2966 					      xs->not_found ? NULL : xs->here);
2967 
2968 		ret = ocfs2_xa_set(&loc, xi, ctxt);
2969 		if (!ret)
2970 			xs->here = loc.xl_entry;
2971 		else if ((ret != -ENOSPC) || ctxt->set_abort)
2972 			goto end;
2973 		else {
2974 			ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2975 			if (ret)
2976 				goto end;
2977 		}
2978 	}
2979 
2980 	if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2981 		ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2982 
2983 end:
2984 	return ret;
2985 }
2986 
2987 /* Check whether the new xattr can be inserted into the inode. */
2988 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2989 				       struct ocfs2_xattr_info *xi,
2990 				       struct ocfs2_xattr_search *xs)
2991 {
2992 	struct ocfs2_xattr_entry *last;
2993 	int free, i;
2994 	size_t min_offs = xs->end - xs->base;
2995 
2996 	if (!xs->header)
2997 		return 0;
2998 
2999 	last = xs->header->xh_entries;
3000 
3001 	for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3002 		size_t offs = le16_to_cpu(last->xe_name_offset);
3003 		if (offs < min_offs)
3004 			min_offs = offs;
3005 		last += 1;
3006 	}
3007 
3008 	free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3009 	if (free < 0)
3010 		return 0;
3011 
3012 	BUG_ON(!xs->not_found);
3013 
3014 	if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3015 		return 1;
3016 
3017 	return 0;
3018 }
3019 
3020 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3021 				     struct ocfs2_dinode *di,
3022 				     struct ocfs2_xattr_info *xi,
3023 				     struct ocfs2_xattr_search *xis,
3024 				     struct ocfs2_xattr_search *xbs,
3025 				     int *clusters_need,
3026 				     int *meta_need,
3027 				     int *credits_need)
3028 {
3029 	int ret = 0, old_in_xb = 0;
3030 	int clusters_add = 0, meta_add = 0, credits = 0;
3031 	struct buffer_head *bh = NULL;
3032 	struct ocfs2_xattr_block *xb = NULL;
3033 	struct ocfs2_xattr_entry *xe = NULL;
3034 	struct ocfs2_xattr_value_root *xv = NULL;
3035 	char *base = NULL;
3036 	int name_offset, name_len = 0;
3037 	u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3038 						    xi->xi_value_len);
3039 	u64 value_size;
3040 
3041 	/*
3042 	 * Calculate the clusters we need to write.
3043 	 * No matter whether we replace an old one or add a new one,
3044 	 * we need this for writing.
3045 	 */
3046 	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3047 		credits += new_clusters *
3048 			   ocfs2_clusters_to_blocks(inode->i_sb, 1);
3049 
3050 	if (xis->not_found && xbs->not_found) {
3051 		credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3052 
3053 		if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3054 			clusters_add += new_clusters;
3055 			credits += ocfs2_calc_extend_credits(inode->i_sb,
3056 							&def_xv.xv.xr_list);
3057 		}
3058 
3059 		goto meta_guess;
3060 	}
3061 
3062 	if (!xis->not_found) {
3063 		xe = xis->here;
3064 		name_offset = le16_to_cpu(xe->xe_name_offset);
3065 		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3066 		base = xis->base;
3067 		credits += OCFS2_INODE_UPDATE_CREDITS;
3068 	} else {
3069 		int i, block_off = 0;
3070 		xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3071 		xe = xbs->here;
3072 		name_offset = le16_to_cpu(xe->xe_name_offset);
3073 		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3074 		i = xbs->here - xbs->header->xh_entries;
3075 		old_in_xb = 1;
3076 
3077 		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3078 			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3079 							bucket_xh(xbs->bucket),
3080 							i, &block_off,
3081 							&name_offset);
3082 			base = bucket_block(xbs->bucket, block_off);
3083 			credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3084 		} else {
3085 			base = xbs->base;
3086 			credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3087 		}
3088 	}
3089 
3090 	/*
3091 	 * delete a xattr doesn't need metadata and cluster allocation.
3092 	 * so just calculate the credits and return.
3093 	 *
3094 	 * The credits for removing the value tree will be extended
3095 	 * by ocfs2_remove_extent itself.
3096 	 */
3097 	if (!xi->xi_value) {
3098 		if (!ocfs2_xattr_is_local(xe))
3099 			credits += ocfs2_remove_extent_credits(inode->i_sb);
3100 
3101 		goto out;
3102 	}
3103 
3104 	/* do cluster allocation guess first. */
3105 	value_size = le64_to_cpu(xe->xe_value_size);
3106 
3107 	if (old_in_xb) {
3108 		/*
3109 		 * In xattr set, we always try to set the xe in inode first,
3110 		 * so if it can be inserted into inode successfully, the old
3111 		 * one will be removed from the xattr block, and this xattr
3112 		 * will be inserted into inode as a new xattr in inode.
3113 		 */
3114 		if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3115 			clusters_add += new_clusters;
3116 			credits += ocfs2_remove_extent_credits(inode->i_sb) +
3117 				    OCFS2_INODE_UPDATE_CREDITS;
3118 			if (!ocfs2_xattr_is_local(xe))
3119 				credits += ocfs2_calc_extend_credits(
3120 							inode->i_sb,
3121 							&def_xv.xv.xr_list);
3122 			goto out;
3123 		}
3124 	}
3125 
3126 	if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3127 		/* the new values will be stored outside. */
3128 		u32 old_clusters = 0;
3129 
3130 		if (!ocfs2_xattr_is_local(xe)) {
3131 			old_clusters =	ocfs2_clusters_for_bytes(inode->i_sb,
3132 								 value_size);
3133 			xv = (struct ocfs2_xattr_value_root *)
3134 			     (base + name_offset + name_len);
3135 			value_size = OCFS2_XATTR_ROOT_SIZE;
3136 		} else
3137 			xv = &def_xv.xv;
3138 
3139 		if (old_clusters >= new_clusters) {
3140 			credits += ocfs2_remove_extent_credits(inode->i_sb);
3141 			goto out;
3142 		} else {
3143 			meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3144 			clusters_add += new_clusters - old_clusters;
3145 			credits += ocfs2_calc_extend_credits(inode->i_sb,
3146 							     &xv->xr_list);
3147 			if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3148 				goto out;
3149 		}
3150 	} else {
3151 		/*
3152 		 * Now the new value will be stored inside. So if the new
3153 		 * value is smaller than the size of value root or the old
3154 		 * value, we don't need any allocation, otherwise we have
3155 		 * to guess metadata allocation.
3156 		 */
3157 		if ((ocfs2_xattr_is_local(xe) &&
3158 		     (value_size >= xi->xi_value_len)) ||
3159 		    (!ocfs2_xattr_is_local(xe) &&
3160 		     OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3161 			goto out;
3162 	}
3163 
3164 meta_guess:
3165 	/* calculate metadata allocation. */
3166 	if (di->i_xattr_loc) {
3167 		if (!xbs->xattr_bh) {
3168 			ret = ocfs2_read_xattr_block(inode,
3169 						     le64_to_cpu(di->i_xattr_loc),
3170 						     &bh);
3171 			if (ret) {
3172 				mlog_errno(ret);
3173 				goto out;
3174 			}
3175 
3176 			xb = (struct ocfs2_xattr_block *)bh->b_data;
3177 		} else
3178 			xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3179 
3180 		/*
3181 		 * If there is already an xattr tree, good, we can calculate
3182 		 * like other b-trees. Otherwise we may have the chance of
3183 		 * create a tree, the credit calculation is borrowed from
3184 		 * ocfs2_calc_extend_credits with root_el = NULL. And the
3185 		 * new tree will be cluster based, so no meta is needed.
3186 		 */
3187 		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3188 			struct ocfs2_extent_list *el =
3189 				 &xb->xb_attrs.xb_root.xt_list;
3190 			meta_add += ocfs2_extend_meta_needed(el);
3191 			credits += ocfs2_calc_extend_credits(inode->i_sb,
3192 							     el);
3193 		} else
3194 			credits += OCFS2_SUBALLOC_ALLOC + 1;
3195 
3196 		/*
3197 		 * This cluster will be used either for new bucket or for
3198 		 * new xattr block.
3199 		 * If the cluster size is the same as the bucket size, one
3200 		 * more is needed since we may need to extend the bucket
3201 		 * also.
3202 		 */
3203 		clusters_add += 1;
3204 		credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3205 		if (OCFS2_XATTR_BUCKET_SIZE ==
3206 			OCFS2_SB(inode->i_sb)->s_clustersize) {
3207 			credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3208 			clusters_add += 1;
3209 		}
3210 	} else {
3211 		credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3212 		if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3213 			struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3214 			meta_add += ocfs2_extend_meta_needed(el);
3215 			credits += ocfs2_calc_extend_credits(inode->i_sb,
3216 							     el);
3217 		} else {
3218 			meta_add += 1;
3219 		}
3220 	}
3221 out:
3222 	if (clusters_need)
3223 		*clusters_need = clusters_add;
3224 	if (meta_need)
3225 		*meta_need = meta_add;
3226 	if (credits_need)
3227 		*credits_need = credits;
3228 	brelse(bh);
3229 	return ret;
3230 }
3231 
3232 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3233 				     struct ocfs2_dinode *di,
3234 				     struct ocfs2_xattr_info *xi,
3235 				     struct ocfs2_xattr_search *xis,
3236 				     struct ocfs2_xattr_search *xbs,
3237 				     struct ocfs2_xattr_set_ctxt *ctxt,
3238 				     int extra_meta,
3239 				     int *credits)
3240 {
3241 	int clusters_add, meta_add, ret;
3242 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3243 
3244 	memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3245 
3246 	ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3247 
3248 	ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3249 					&clusters_add, &meta_add, credits);
3250 	if (ret) {
3251 		mlog_errno(ret);
3252 		return ret;
3253 	}
3254 
3255 	meta_add += extra_meta;
3256 	trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3257 					clusters_add, *credits);
3258 
3259 	if (meta_add) {
3260 		ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3261 							&ctxt->meta_ac);
3262 		if (ret) {
3263 			mlog_errno(ret);
3264 			goto out;
3265 		}
3266 	}
3267 
3268 	if (clusters_add) {
3269 		ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3270 		if (ret)
3271 			mlog_errno(ret);
3272 	}
3273 out:
3274 	if (ret) {
3275 		if (ctxt->meta_ac) {
3276 			ocfs2_free_alloc_context(ctxt->meta_ac);
3277 			ctxt->meta_ac = NULL;
3278 		}
3279 
3280 		/*
3281 		 * We cannot have an error and a non null ctxt->data_ac.
3282 		 */
3283 	}
3284 
3285 	return ret;
3286 }
3287 
3288 static int __ocfs2_xattr_set_handle(struct inode *inode,
3289 				    struct ocfs2_dinode *di,
3290 				    struct ocfs2_xattr_info *xi,
3291 				    struct ocfs2_xattr_search *xis,
3292 				    struct ocfs2_xattr_search *xbs,
3293 				    struct ocfs2_xattr_set_ctxt *ctxt)
3294 {
3295 	int ret = 0, credits, old_found;
3296 
3297 	if (!xi->xi_value) {
3298 		/* Remove existing extended attribute */
3299 		if (!xis->not_found)
3300 			ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3301 		else if (!xbs->not_found)
3302 			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3303 	} else {
3304 		/* We always try to set extended attribute into inode first*/
3305 		ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3306 		if (!ret && !xbs->not_found) {
3307 			/*
3308 			 * If succeed and that extended attribute existing in
3309 			 * external block, then we will remove it.
3310 			 */
3311 			xi->xi_value = NULL;
3312 			xi->xi_value_len = 0;
3313 
3314 			old_found = xis->not_found;
3315 			xis->not_found = -ENODATA;
3316 			ret = ocfs2_calc_xattr_set_need(inode,
3317 							di,
3318 							xi,
3319 							xis,
3320 							xbs,
3321 							NULL,
3322 							NULL,
3323 							&credits);
3324 			xis->not_found = old_found;
3325 			if (ret) {
3326 				mlog_errno(ret);
3327 				goto out;
3328 			}
3329 
3330 			ret = ocfs2_extend_trans(ctxt->handle, credits);
3331 			if (ret) {
3332 				mlog_errno(ret);
3333 				goto out;
3334 			}
3335 			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3336 		} else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3337 			if (di->i_xattr_loc && !xbs->xattr_bh) {
3338 				ret = ocfs2_xattr_block_find(inode,
3339 							     xi->xi_name_index,
3340 							     xi->xi_name, xbs);
3341 				if (ret)
3342 					goto out;
3343 
3344 				old_found = xis->not_found;
3345 				xis->not_found = -ENODATA;
3346 				ret = ocfs2_calc_xattr_set_need(inode,
3347 								di,
3348 								xi,
3349 								xis,
3350 								xbs,
3351 								NULL,
3352 								NULL,
3353 								&credits);
3354 				xis->not_found = old_found;
3355 				if (ret) {
3356 					mlog_errno(ret);
3357 					goto out;
3358 				}
3359 
3360 				ret = ocfs2_extend_trans(ctxt->handle, credits);
3361 				if (ret) {
3362 					mlog_errno(ret);
3363 					goto out;
3364 				}
3365 			}
3366 			/*
3367 			 * If no space in inode, we will set extended attribute
3368 			 * into external block.
3369 			 */
3370 			ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3371 			if (ret)
3372 				goto out;
3373 			if (!xis->not_found) {
3374 				/*
3375 				 * If succeed and that extended attribute
3376 				 * existing in inode, we will remove it.
3377 				 */
3378 				xi->xi_value = NULL;
3379 				xi->xi_value_len = 0;
3380 				xbs->not_found = -ENODATA;
3381 				ret = ocfs2_calc_xattr_set_need(inode,
3382 								di,
3383 								xi,
3384 								xis,
3385 								xbs,
3386 								NULL,
3387 								NULL,
3388 								&credits);
3389 				if (ret) {
3390 					mlog_errno(ret);
3391 					goto out;
3392 				}
3393 
3394 				ret = ocfs2_extend_trans(ctxt->handle, credits);
3395 				if (ret) {
3396 					mlog_errno(ret);
3397 					goto out;
3398 				}
3399 				ret = ocfs2_xattr_ibody_set(inode, xi,
3400 							    xis, ctxt);
3401 			}
3402 		}
3403 	}
3404 
3405 	if (!ret) {
3406 		/* Update inode ctime. */
3407 		ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3408 					      xis->inode_bh,
3409 					      OCFS2_JOURNAL_ACCESS_WRITE);
3410 		if (ret) {
3411 			mlog_errno(ret);
3412 			goto out;
3413 		}
3414 
3415 		inode->i_ctime = CURRENT_TIME;
3416 		di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3417 		di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3418 		ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3419 	}
3420 out:
3421 	return ret;
3422 }
3423 
3424 /*
3425  * This function only called duing creating inode
3426  * for init security/acl xattrs of the new inode.
3427  * All transanction credits have been reserved in mknod.
3428  */
3429 int ocfs2_xattr_set_handle(handle_t *handle,
3430 			   struct inode *inode,
3431 			   struct buffer_head *di_bh,
3432 			   int name_index,
3433 			   const char *name,
3434 			   const void *value,
3435 			   size_t value_len,
3436 			   int flags,
3437 			   struct ocfs2_alloc_context *meta_ac,
3438 			   struct ocfs2_alloc_context *data_ac)
3439 {
3440 	struct ocfs2_dinode *di;
3441 	int ret;
3442 
3443 	struct ocfs2_xattr_info xi = {
3444 		.xi_name_index = name_index,
3445 		.xi_name = name,
3446 		.xi_name_len = strlen(name),
3447 		.xi_value = value,
3448 		.xi_value_len = value_len,
3449 	};
3450 
3451 	struct ocfs2_xattr_search xis = {
3452 		.not_found = -ENODATA,
3453 	};
3454 
3455 	struct ocfs2_xattr_search xbs = {
3456 		.not_found = -ENODATA,
3457 	};
3458 
3459 	struct ocfs2_xattr_set_ctxt ctxt = {
3460 		.handle = handle,
3461 		.meta_ac = meta_ac,
3462 		.data_ac = data_ac,
3463 	};
3464 
3465 	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3466 		return -EOPNOTSUPP;
3467 
3468 	/*
3469 	 * In extreme situation, may need xattr bucket when
3470 	 * block size is too small. And we have already reserved
3471 	 * the credits for bucket in mknod.
3472 	 */
3473 	if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3474 		xbs.bucket = ocfs2_xattr_bucket_new(inode);
3475 		if (!xbs.bucket) {
3476 			mlog_errno(-ENOMEM);
3477 			return -ENOMEM;
3478 		}
3479 	}
3480 
3481 	xis.inode_bh = xbs.inode_bh = di_bh;
3482 	di = (struct ocfs2_dinode *)di_bh->b_data;
3483 
3484 	down_write(&OCFS2_I(inode)->ip_xattr_sem);
3485 
3486 	ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3487 	if (ret)
3488 		goto cleanup;
3489 	if (xis.not_found) {
3490 		ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3491 		if (ret)
3492 			goto cleanup;
3493 	}
3494 
3495 	ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3496 
3497 cleanup:
3498 	up_write(&OCFS2_I(inode)->ip_xattr_sem);
3499 	brelse(xbs.xattr_bh);
3500 	ocfs2_xattr_bucket_free(xbs.bucket);
3501 
3502 	return ret;
3503 }
3504 
3505 /*
3506  * ocfs2_xattr_set()
3507  *
3508  * Set, replace or remove an extended attribute for this inode.
3509  * value is NULL to remove an existing extended attribute, else either
3510  * create or replace an extended attribute.
3511  */
3512 int ocfs2_xattr_set(struct inode *inode,
3513 		    int name_index,
3514 		    const char *name,
3515 		    const void *value,
3516 		    size_t value_len,
3517 		    int flags)
3518 {
3519 	struct buffer_head *di_bh = NULL;
3520 	struct ocfs2_dinode *di;
3521 	int ret, credits, ref_meta = 0, ref_credits = 0;
3522 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3523 	struct inode *tl_inode = osb->osb_tl_inode;
3524 	struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, NULL, };
3525 	struct ocfs2_refcount_tree *ref_tree = NULL;
3526 
3527 	struct ocfs2_xattr_info xi = {
3528 		.xi_name_index = name_index,
3529 		.xi_name = name,
3530 		.xi_name_len = strlen(name),
3531 		.xi_value = value,
3532 		.xi_value_len = value_len,
3533 	};
3534 
3535 	struct ocfs2_xattr_search xis = {
3536 		.not_found = -ENODATA,
3537 	};
3538 
3539 	struct ocfs2_xattr_search xbs = {
3540 		.not_found = -ENODATA,
3541 	};
3542 
3543 	if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3544 		return -EOPNOTSUPP;
3545 
3546 	/*
3547 	 * Only xbs will be used on indexed trees.  xis doesn't need a
3548 	 * bucket.
3549 	 */
3550 	xbs.bucket = ocfs2_xattr_bucket_new(inode);
3551 	if (!xbs.bucket) {
3552 		mlog_errno(-ENOMEM);
3553 		return -ENOMEM;
3554 	}
3555 
3556 	ret = ocfs2_inode_lock(inode, &di_bh, 1);
3557 	if (ret < 0) {
3558 		mlog_errno(ret);
3559 		goto cleanup_nolock;
3560 	}
3561 	xis.inode_bh = xbs.inode_bh = di_bh;
3562 	di = (struct ocfs2_dinode *)di_bh->b_data;
3563 
3564 	down_write(&OCFS2_I(inode)->ip_xattr_sem);
3565 	/*
3566 	 * Scan inode and external block to find the same name
3567 	 * extended attribute and collect search information.
3568 	 */
3569 	ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3570 	if (ret)
3571 		goto cleanup;
3572 	if (xis.not_found) {
3573 		ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3574 		if (ret)
3575 			goto cleanup;
3576 	}
3577 
3578 	if (xis.not_found && xbs.not_found) {
3579 		ret = -ENODATA;
3580 		if (flags & XATTR_REPLACE)
3581 			goto cleanup;
3582 		ret = 0;
3583 		if (!value)
3584 			goto cleanup;
3585 	} else {
3586 		ret = -EEXIST;
3587 		if (flags & XATTR_CREATE)
3588 			goto cleanup;
3589 	}
3590 
3591 	/* Check whether the value is refcounted and do some preparation. */
3592 	if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3593 	    (!xis.not_found || !xbs.not_found)) {
3594 		ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3595 						   &xis, &xbs, &ref_tree,
3596 						   &ref_meta, &ref_credits);
3597 		if (ret) {
3598 			mlog_errno(ret);
3599 			goto cleanup;
3600 		}
3601 	}
3602 
3603 	mutex_lock(&tl_inode->i_mutex);
3604 
3605 	if (ocfs2_truncate_log_needs_flush(osb)) {
3606 		ret = __ocfs2_flush_truncate_log(osb);
3607 		if (ret < 0) {
3608 			mutex_unlock(&tl_inode->i_mutex);
3609 			mlog_errno(ret);
3610 			goto cleanup;
3611 		}
3612 	}
3613 	mutex_unlock(&tl_inode->i_mutex);
3614 
3615 	ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3616 					&xbs, &ctxt, ref_meta, &credits);
3617 	if (ret) {
3618 		mlog_errno(ret);
3619 		goto cleanup;
3620 	}
3621 
3622 	/* we need to update inode's ctime field, so add credit for it. */
3623 	credits += OCFS2_INODE_UPDATE_CREDITS;
3624 	ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3625 	if (IS_ERR(ctxt.handle)) {
3626 		ret = PTR_ERR(ctxt.handle);
3627 		mlog_errno(ret);
3628 		goto out_free_ac;
3629 	}
3630 
3631 	ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3632 	ocfs2_update_inode_fsync_trans(ctxt.handle, inode, 0);
3633 
3634 	ocfs2_commit_trans(osb, ctxt.handle);
3635 
3636 out_free_ac:
3637 	if (ctxt.data_ac)
3638 		ocfs2_free_alloc_context(ctxt.data_ac);
3639 	if (ctxt.meta_ac)
3640 		ocfs2_free_alloc_context(ctxt.meta_ac);
3641 	if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3642 		ocfs2_schedule_truncate_log_flush(osb, 1);
3643 	ocfs2_run_deallocs(osb, &ctxt.dealloc);
3644 
3645 cleanup:
3646 	if (ref_tree)
3647 		ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3648 	up_write(&OCFS2_I(inode)->ip_xattr_sem);
3649 	if (!value && !ret) {
3650 		ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3651 		if (ret)
3652 			mlog_errno(ret);
3653 	}
3654 	ocfs2_inode_unlock(inode, 1);
3655 cleanup_nolock:
3656 	brelse(di_bh);
3657 	brelse(xbs.xattr_bh);
3658 	ocfs2_xattr_bucket_free(xbs.bucket);
3659 
3660 	return ret;
3661 }
3662 
3663 /*
3664  * Find the xattr extent rec which may contains name_hash.
3665  * e_cpos will be the first name hash of the xattr rec.
3666  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3667  */
3668 static int ocfs2_xattr_get_rec(struct inode *inode,
3669 			       u32 name_hash,
3670 			       u64 *p_blkno,
3671 			       u32 *e_cpos,
3672 			       u32 *num_clusters,
3673 			       struct ocfs2_extent_list *el)
3674 {
3675 	int ret = 0, i;
3676 	struct buffer_head *eb_bh = NULL;
3677 	struct ocfs2_extent_block *eb;
3678 	struct ocfs2_extent_rec *rec = NULL;
3679 	u64 e_blkno = 0;
3680 
3681 	if (el->l_tree_depth) {
3682 		ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3683 				      &eb_bh);
3684 		if (ret) {
3685 			mlog_errno(ret);
3686 			goto out;
3687 		}
3688 
3689 		eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3690 		el = &eb->h_list;
3691 
3692 		if (el->l_tree_depth) {
3693 			ocfs2_error(inode->i_sb,
3694 				    "Inode %lu has non zero tree depth in "
3695 				    "xattr tree block %llu\n", inode->i_ino,
3696 				    (unsigned long long)eb_bh->b_blocknr);
3697 			ret = -EROFS;
3698 			goto out;
3699 		}
3700 	}
3701 
3702 	for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3703 		rec = &el->l_recs[i];
3704 
3705 		if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3706 			e_blkno = le64_to_cpu(rec->e_blkno);
3707 			break;
3708 		}
3709 	}
3710 
3711 	if (!e_blkno) {
3712 		ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3713 			    "record (%u, %u, 0) in xattr", inode->i_ino,
3714 			    le32_to_cpu(rec->e_cpos),
3715 			    ocfs2_rec_clusters(el, rec));
3716 		ret = -EROFS;
3717 		goto out;
3718 	}
3719 
3720 	*p_blkno = le64_to_cpu(rec->e_blkno);
3721 	*num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3722 	if (e_cpos)
3723 		*e_cpos = le32_to_cpu(rec->e_cpos);
3724 out:
3725 	brelse(eb_bh);
3726 	return ret;
3727 }
3728 
3729 typedef int (xattr_bucket_func)(struct inode *inode,
3730 				struct ocfs2_xattr_bucket *bucket,
3731 				void *para);
3732 
3733 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3734 				   struct ocfs2_xattr_bucket *bucket,
3735 				   int name_index,
3736 				   const char *name,
3737 				   u32 name_hash,
3738 				   u16 *xe_index,
3739 				   int *found)
3740 {
3741 	int i, ret = 0, cmp = 1, block_off, new_offset;
3742 	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3743 	size_t name_len = strlen(name);
3744 	struct ocfs2_xattr_entry *xe = NULL;
3745 	char *xe_name;
3746 
3747 	/*
3748 	 * We don't use binary search in the bucket because there
3749 	 * may be multiple entries with the same name hash.
3750 	 */
3751 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3752 		xe = &xh->xh_entries[i];
3753 
3754 		if (name_hash > le32_to_cpu(xe->xe_name_hash))
3755 			continue;
3756 		else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3757 			break;
3758 
3759 		cmp = name_index - ocfs2_xattr_get_type(xe);
3760 		if (!cmp)
3761 			cmp = name_len - xe->xe_name_len;
3762 		if (cmp)
3763 			continue;
3764 
3765 		ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3766 							xh,
3767 							i,
3768 							&block_off,
3769 							&new_offset);
3770 		if (ret) {
3771 			mlog_errno(ret);
3772 			break;
3773 		}
3774 
3775 
3776 		xe_name = bucket_block(bucket, block_off) + new_offset;
3777 		if (!memcmp(name, xe_name, name_len)) {
3778 			*xe_index = i;
3779 			*found = 1;
3780 			ret = 0;
3781 			break;
3782 		}
3783 	}
3784 
3785 	return ret;
3786 }
3787 
3788 /*
3789  * Find the specified xattr entry in a series of buckets.
3790  * This series start from p_blkno and last for num_clusters.
3791  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3792  * the num of the valid buckets.
3793  *
3794  * Return the buffer_head this xattr should reside in. And if the xattr's
3795  * hash is in the gap of 2 buckets, return the lower bucket.
3796  */
3797 static int ocfs2_xattr_bucket_find(struct inode *inode,
3798 				   int name_index,
3799 				   const char *name,
3800 				   u32 name_hash,
3801 				   u64 p_blkno,
3802 				   u32 first_hash,
3803 				   u32 num_clusters,
3804 				   struct ocfs2_xattr_search *xs)
3805 {
3806 	int ret, found = 0;
3807 	struct ocfs2_xattr_header *xh = NULL;
3808 	struct ocfs2_xattr_entry *xe = NULL;
3809 	u16 index = 0;
3810 	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3811 	int low_bucket = 0, bucket, high_bucket;
3812 	struct ocfs2_xattr_bucket *search;
3813 	u32 last_hash;
3814 	u64 blkno, lower_blkno = 0;
3815 
3816 	search = ocfs2_xattr_bucket_new(inode);
3817 	if (!search) {
3818 		ret = -ENOMEM;
3819 		mlog_errno(ret);
3820 		goto out;
3821 	}
3822 
3823 	ret = ocfs2_read_xattr_bucket(search, p_blkno);
3824 	if (ret) {
3825 		mlog_errno(ret);
3826 		goto out;
3827 	}
3828 
3829 	xh = bucket_xh(search);
3830 	high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3831 	while (low_bucket <= high_bucket) {
3832 		ocfs2_xattr_bucket_relse(search);
3833 
3834 		bucket = (low_bucket + high_bucket) / 2;
3835 		blkno = p_blkno + bucket * blk_per_bucket;
3836 		ret = ocfs2_read_xattr_bucket(search, blkno);
3837 		if (ret) {
3838 			mlog_errno(ret);
3839 			goto out;
3840 		}
3841 
3842 		xh = bucket_xh(search);
3843 		xe = &xh->xh_entries[0];
3844 		if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3845 			high_bucket = bucket - 1;
3846 			continue;
3847 		}
3848 
3849 		/*
3850 		 * Check whether the hash of the last entry in our
3851 		 * bucket is larger than the search one. for an empty
3852 		 * bucket, the last one is also the first one.
3853 		 */
3854 		if (xh->xh_count)
3855 			xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3856 
3857 		last_hash = le32_to_cpu(xe->xe_name_hash);
3858 
3859 		/* record lower_blkno which may be the insert place. */
3860 		lower_blkno = blkno;
3861 
3862 		if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3863 			low_bucket = bucket + 1;
3864 			continue;
3865 		}
3866 
3867 		/* the searched xattr should reside in this bucket if exists. */
3868 		ret = ocfs2_find_xe_in_bucket(inode, search,
3869 					      name_index, name, name_hash,
3870 					      &index, &found);
3871 		if (ret) {
3872 			mlog_errno(ret);
3873 			goto out;
3874 		}
3875 		break;
3876 	}
3877 
3878 	/*
3879 	 * Record the bucket we have found.
3880 	 * When the xattr's hash value is in the gap of 2 buckets, we will
3881 	 * always set it to the previous bucket.
3882 	 */
3883 	if (!lower_blkno)
3884 		lower_blkno = p_blkno;
3885 
3886 	/* This should be in cache - we just read it during the search */
3887 	ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3888 	if (ret) {
3889 		mlog_errno(ret);
3890 		goto out;
3891 	}
3892 
3893 	xs->header = bucket_xh(xs->bucket);
3894 	xs->base = bucket_block(xs->bucket, 0);
3895 	xs->end = xs->base + inode->i_sb->s_blocksize;
3896 
3897 	if (found) {
3898 		xs->here = &xs->header->xh_entries[index];
3899 		trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3900 			name, name_index, name_hash,
3901 			(unsigned long long)bucket_blkno(xs->bucket),
3902 			index);
3903 	} else
3904 		ret = -ENODATA;
3905 
3906 out:
3907 	ocfs2_xattr_bucket_free(search);
3908 	return ret;
3909 }
3910 
3911 static int ocfs2_xattr_index_block_find(struct inode *inode,
3912 					struct buffer_head *root_bh,
3913 					int name_index,
3914 					const char *name,
3915 					struct ocfs2_xattr_search *xs)
3916 {
3917 	int ret;
3918 	struct ocfs2_xattr_block *xb =
3919 			(struct ocfs2_xattr_block *)root_bh->b_data;
3920 	struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3921 	struct ocfs2_extent_list *el = &xb_root->xt_list;
3922 	u64 p_blkno = 0;
3923 	u32 first_hash, num_clusters = 0;
3924 	u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3925 
3926 	if (le16_to_cpu(el->l_next_free_rec) == 0)
3927 		return -ENODATA;
3928 
3929 	trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3930 					name, name_index, name_hash,
3931 					(unsigned long long)root_bh->b_blocknr,
3932 					-1);
3933 
3934 	ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3935 				  &num_clusters, el);
3936 	if (ret) {
3937 		mlog_errno(ret);
3938 		goto out;
3939 	}
3940 
3941 	BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3942 
3943 	trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3944 					name, name_index, first_hash,
3945 					(unsigned long long)p_blkno,
3946 					num_clusters);
3947 
3948 	ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3949 				      p_blkno, first_hash, num_clusters, xs);
3950 
3951 out:
3952 	return ret;
3953 }
3954 
3955 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3956 				       u64 blkno,
3957 				       u32 clusters,
3958 				       xattr_bucket_func *func,
3959 				       void *para)
3960 {
3961 	int i, ret = 0;
3962 	u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3963 	u32 num_buckets = clusters * bpc;
3964 	struct ocfs2_xattr_bucket *bucket;
3965 
3966 	bucket = ocfs2_xattr_bucket_new(inode);
3967 	if (!bucket) {
3968 		mlog_errno(-ENOMEM);
3969 		return -ENOMEM;
3970 	}
3971 
3972 	trace_ocfs2_iterate_xattr_buckets(
3973 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
3974 		(unsigned long long)blkno, clusters);
3975 
3976 	for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3977 		ret = ocfs2_read_xattr_bucket(bucket, blkno);
3978 		if (ret) {
3979 			mlog_errno(ret);
3980 			break;
3981 		}
3982 
3983 		/*
3984 		 * The real bucket num in this series of blocks is stored
3985 		 * in the 1st bucket.
3986 		 */
3987 		if (i == 0)
3988 			num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3989 
3990 		trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3991 		     le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3992 		if (func) {
3993 			ret = func(inode, bucket, para);
3994 			if (ret && ret != -ERANGE)
3995 				mlog_errno(ret);
3996 			/* Fall through to bucket_relse() */
3997 		}
3998 
3999 		ocfs2_xattr_bucket_relse(bucket);
4000 		if (ret)
4001 			break;
4002 	}
4003 
4004 	ocfs2_xattr_bucket_free(bucket);
4005 	return ret;
4006 }
4007 
4008 struct ocfs2_xattr_tree_list {
4009 	char *buffer;
4010 	size_t buffer_size;
4011 	size_t result;
4012 };
4013 
4014 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4015 					     struct ocfs2_xattr_header *xh,
4016 					     int index,
4017 					     int *block_off,
4018 					     int *new_offset)
4019 {
4020 	u16 name_offset;
4021 
4022 	if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4023 		return -EINVAL;
4024 
4025 	name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4026 
4027 	*block_off = name_offset >> sb->s_blocksize_bits;
4028 	*new_offset = name_offset % sb->s_blocksize;
4029 
4030 	return 0;
4031 }
4032 
4033 static int ocfs2_list_xattr_bucket(struct inode *inode,
4034 				   struct ocfs2_xattr_bucket *bucket,
4035 				   void *para)
4036 {
4037 	int ret = 0, type;
4038 	struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4039 	int i, block_off, new_offset;
4040 	const char *prefix, *name;
4041 
4042 	for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4043 		struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4044 		type = ocfs2_xattr_get_type(entry);
4045 		prefix = ocfs2_xattr_prefix(type);
4046 
4047 		if (prefix) {
4048 			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4049 								bucket_xh(bucket),
4050 								i,
4051 								&block_off,
4052 								&new_offset);
4053 			if (ret)
4054 				break;
4055 
4056 			name = (const char *)bucket_block(bucket, block_off) +
4057 				new_offset;
4058 			ret = ocfs2_xattr_list_entry(xl->buffer,
4059 						     xl->buffer_size,
4060 						     &xl->result,
4061 						     prefix, name,
4062 						     entry->xe_name_len);
4063 			if (ret)
4064 				break;
4065 		}
4066 	}
4067 
4068 	return ret;
4069 }
4070 
4071 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4072 					   struct buffer_head *blk_bh,
4073 					   xattr_tree_rec_func *rec_func,
4074 					   void *para)
4075 {
4076 	struct ocfs2_xattr_block *xb =
4077 			(struct ocfs2_xattr_block *)blk_bh->b_data;
4078 	struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4079 	int ret = 0;
4080 	u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4081 	u64 p_blkno = 0;
4082 
4083 	if (!el->l_next_free_rec || !rec_func)
4084 		return 0;
4085 
4086 	while (name_hash > 0) {
4087 		ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4088 					  &e_cpos, &num_clusters, el);
4089 		if (ret) {
4090 			mlog_errno(ret);
4091 			break;
4092 		}
4093 
4094 		ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4095 			       num_clusters, para);
4096 		if (ret) {
4097 			if (ret != -ERANGE)
4098 				mlog_errno(ret);
4099 			break;
4100 		}
4101 
4102 		if (e_cpos == 0)
4103 			break;
4104 
4105 		name_hash = e_cpos - 1;
4106 	}
4107 
4108 	return ret;
4109 
4110 }
4111 
4112 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4113 				     struct buffer_head *root_bh,
4114 				     u64 blkno, u32 cpos, u32 len, void *para)
4115 {
4116 	return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4117 					   ocfs2_list_xattr_bucket, para);
4118 }
4119 
4120 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4121 					     struct buffer_head *blk_bh,
4122 					     char *buffer,
4123 					     size_t buffer_size)
4124 {
4125 	int ret;
4126 	struct ocfs2_xattr_tree_list xl = {
4127 		.buffer = buffer,
4128 		.buffer_size = buffer_size,
4129 		.result = 0,
4130 	};
4131 
4132 	ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4133 					      ocfs2_list_xattr_tree_rec, &xl);
4134 	if (ret) {
4135 		mlog_errno(ret);
4136 		goto out;
4137 	}
4138 
4139 	ret = xl.result;
4140 out:
4141 	return ret;
4142 }
4143 
4144 static int cmp_xe(const void *a, const void *b)
4145 {
4146 	const struct ocfs2_xattr_entry *l = a, *r = b;
4147 	u32 l_hash = le32_to_cpu(l->xe_name_hash);
4148 	u32 r_hash = le32_to_cpu(r->xe_name_hash);
4149 
4150 	if (l_hash > r_hash)
4151 		return 1;
4152 	if (l_hash < r_hash)
4153 		return -1;
4154 	return 0;
4155 }
4156 
4157 static void swap_xe(void *a, void *b, int size)
4158 {
4159 	struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4160 
4161 	tmp = *l;
4162 	memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4163 	memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4164 }
4165 
4166 /*
4167  * When the ocfs2_xattr_block is filled up, new bucket will be created
4168  * and all the xattr entries will be moved to the new bucket.
4169  * The header goes at the start of the bucket, and the names+values are
4170  * filled from the end.  This is why *target starts as the last buffer.
4171  * Note: we need to sort the entries since they are not saved in order
4172  * in the ocfs2_xattr_block.
4173  */
4174 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4175 					   struct buffer_head *xb_bh,
4176 					   struct ocfs2_xattr_bucket *bucket)
4177 {
4178 	int i, blocksize = inode->i_sb->s_blocksize;
4179 	int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4180 	u16 offset, size, off_change;
4181 	struct ocfs2_xattr_entry *xe;
4182 	struct ocfs2_xattr_block *xb =
4183 				(struct ocfs2_xattr_block *)xb_bh->b_data;
4184 	struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4185 	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4186 	u16 count = le16_to_cpu(xb_xh->xh_count);
4187 	char *src = xb_bh->b_data;
4188 	char *target = bucket_block(bucket, blks - 1);
4189 
4190 	trace_ocfs2_cp_xattr_block_to_bucket_begin(
4191 				(unsigned long long)xb_bh->b_blocknr,
4192 				(unsigned long long)bucket_blkno(bucket));
4193 
4194 	for (i = 0; i < blks; i++)
4195 		memset(bucket_block(bucket, i), 0, blocksize);
4196 
4197 	/*
4198 	 * Since the xe_name_offset is based on ocfs2_xattr_header,
4199 	 * there is a offset change corresponding to the change of
4200 	 * ocfs2_xattr_header's position.
4201 	 */
4202 	off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4203 	xe = &xb_xh->xh_entries[count - 1];
4204 	offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4205 	size = blocksize - offset;
4206 
4207 	/* copy all the names and values. */
4208 	memcpy(target + offset, src + offset, size);
4209 
4210 	/* Init new header now. */
4211 	xh->xh_count = xb_xh->xh_count;
4212 	xh->xh_num_buckets = cpu_to_le16(1);
4213 	xh->xh_name_value_len = cpu_to_le16(size);
4214 	xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4215 
4216 	/* copy all the entries. */
4217 	target = bucket_block(bucket, 0);
4218 	offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4219 	size = count * sizeof(struct ocfs2_xattr_entry);
4220 	memcpy(target + offset, (char *)xb_xh + offset, size);
4221 
4222 	/* Change the xe offset for all the xe because of the move. */
4223 	off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4224 		 offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4225 	for (i = 0; i < count; i++)
4226 		le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4227 
4228 	trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4229 
4230 	sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4231 	     cmp_xe, swap_xe);
4232 }
4233 
4234 /*
4235  * After we move xattr from block to index btree, we have to
4236  * update ocfs2_xattr_search to the new xe and base.
4237  *
4238  * When the entry is in xattr block, xattr_bh indicates the storage place.
4239  * While if the entry is in index b-tree, "bucket" indicates the
4240  * real place of the xattr.
4241  */
4242 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4243 					    struct ocfs2_xattr_search *xs,
4244 					    struct buffer_head *old_bh)
4245 {
4246 	char *buf = old_bh->b_data;
4247 	struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4248 	struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4249 	int i;
4250 
4251 	xs->header = bucket_xh(xs->bucket);
4252 	xs->base = bucket_block(xs->bucket, 0);
4253 	xs->end = xs->base + inode->i_sb->s_blocksize;
4254 
4255 	if (xs->not_found)
4256 		return;
4257 
4258 	i = xs->here - old_xh->xh_entries;
4259 	xs->here = &xs->header->xh_entries[i];
4260 }
4261 
4262 static int ocfs2_xattr_create_index_block(struct inode *inode,
4263 					  struct ocfs2_xattr_search *xs,
4264 					  struct ocfs2_xattr_set_ctxt *ctxt)
4265 {
4266 	int ret;
4267 	u32 bit_off, len;
4268 	u64 blkno;
4269 	handle_t *handle = ctxt->handle;
4270 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
4271 	struct buffer_head *xb_bh = xs->xattr_bh;
4272 	struct ocfs2_xattr_block *xb =
4273 			(struct ocfs2_xattr_block *)xb_bh->b_data;
4274 	struct ocfs2_xattr_tree_root *xr;
4275 	u16 xb_flags = le16_to_cpu(xb->xb_flags);
4276 
4277 	trace_ocfs2_xattr_create_index_block_begin(
4278 				(unsigned long long)xb_bh->b_blocknr);
4279 
4280 	BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4281 	BUG_ON(!xs->bucket);
4282 
4283 	/*
4284 	 * XXX:
4285 	 * We can use this lock for now, and maybe move to a dedicated mutex
4286 	 * if performance becomes a problem later.
4287 	 */
4288 	down_write(&oi->ip_alloc_sem);
4289 
4290 	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4291 				      OCFS2_JOURNAL_ACCESS_WRITE);
4292 	if (ret) {
4293 		mlog_errno(ret);
4294 		goto out;
4295 	}
4296 
4297 	ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4298 				     1, 1, &bit_off, &len);
4299 	if (ret) {
4300 		mlog_errno(ret);
4301 		goto out;
4302 	}
4303 
4304 	/*
4305 	 * The bucket may spread in many blocks, and
4306 	 * we will only touch the 1st block and the last block
4307 	 * in the whole bucket(one for entry and one for data).
4308 	 */
4309 	blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4310 
4311 	trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4312 
4313 	ret = ocfs2_init_xattr_bucket(xs->bucket, blkno, 1);
4314 	if (ret) {
4315 		mlog_errno(ret);
4316 		goto out;
4317 	}
4318 
4319 	ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4320 						OCFS2_JOURNAL_ACCESS_CREATE);
4321 	if (ret) {
4322 		mlog_errno(ret);
4323 		goto out;
4324 	}
4325 
4326 	ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4327 	ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4328 
4329 	ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4330 
4331 	/* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4332 	memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4333 	       offsetof(struct ocfs2_xattr_block, xb_attrs));
4334 
4335 	xr = &xb->xb_attrs.xb_root;
4336 	xr->xt_clusters = cpu_to_le32(1);
4337 	xr->xt_last_eb_blk = 0;
4338 	xr->xt_list.l_tree_depth = 0;
4339 	xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4340 	xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4341 
4342 	xr->xt_list.l_recs[0].e_cpos = 0;
4343 	xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4344 	xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4345 
4346 	xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4347 
4348 	ocfs2_journal_dirty(handle, xb_bh);
4349 
4350 out:
4351 	up_write(&oi->ip_alloc_sem);
4352 
4353 	return ret;
4354 }
4355 
4356 static int cmp_xe_offset(const void *a, const void *b)
4357 {
4358 	const struct ocfs2_xattr_entry *l = a, *r = b;
4359 	u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4360 	u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4361 
4362 	if (l_name_offset < r_name_offset)
4363 		return 1;
4364 	if (l_name_offset > r_name_offset)
4365 		return -1;
4366 	return 0;
4367 }
4368 
4369 /*
4370  * defrag a xattr bucket if we find that the bucket has some
4371  * holes beteen name/value pairs.
4372  * We will move all the name/value pairs to the end of the bucket
4373  * so that we can spare some space for insertion.
4374  */
4375 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4376 				     handle_t *handle,
4377 				     struct ocfs2_xattr_bucket *bucket)
4378 {
4379 	int ret, i;
4380 	size_t end, offset, len;
4381 	struct ocfs2_xattr_header *xh;
4382 	char *entries, *buf, *bucket_buf = NULL;
4383 	u64 blkno = bucket_blkno(bucket);
4384 	u16 xh_free_start;
4385 	size_t blocksize = inode->i_sb->s_blocksize;
4386 	struct ocfs2_xattr_entry *xe;
4387 
4388 	/*
4389 	 * In order to make the operation more efficient and generic,
4390 	 * we copy all the blocks into a contiguous memory and do the
4391 	 * defragment there, so if anything is error, we will not touch
4392 	 * the real block.
4393 	 */
4394 	bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4395 	if (!bucket_buf) {
4396 		ret = -EIO;
4397 		goto out;
4398 	}
4399 
4400 	buf = bucket_buf;
4401 	for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4402 		memcpy(buf, bucket_block(bucket, i), blocksize);
4403 
4404 	ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4405 						OCFS2_JOURNAL_ACCESS_WRITE);
4406 	if (ret < 0) {
4407 		mlog_errno(ret);
4408 		goto out;
4409 	}
4410 
4411 	xh = (struct ocfs2_xattr_header *)bucket_buf;
4412 	entries = (char *)xh->xh_entries;
4413 	xh_free_start = le16_to_cpu(xh->xh_free_start);
4414 
4415 	trace_ocfs2_defrag_xattr_bucket(
4416 	     (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4417 	     xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4418 
4419 	/*
4420 	 * sort all the entries by their offset.
4421 	 * the largest will be the first, so that we can
4422 	 * move them to the end one by one.
4423 	 */
4424 	sort(entries, le16_to_cpu(xh->xh_count),
4425 	     sizeof(struct ocfs2_xattr_entry),
4426 	     cmp_xe_offset, swap_xe);
4427 
4428 	/* Move all name/values to the end of the bucket. */
4429 	xe = xh->xh_entries;
4430 	end = OCFS2_XATTR_BUCKET_SIZE;
4431 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4432 		offset = le16_to_cpu(xe->xe_name_offset);
4433 		len = namevalue_size_xe(xe);
4434 
4435 		/*
4436 		 * We must make sure that the name/value pair
4437 		 * exist in the same block. So adjust end to
4438 		 * the previous block end if needed.
4439 		 */
4440 		if (((end - len) / blocksize !=
4441 			(end - 1) / blocksize))
4442 			end = end - end % blocksize;
4443 
4444 		if (end > offset + len) {
4445 			memmove(bucket_buf + end - len,
4446 				bucket_buf + offset, len);
4447 			xe->xe_name_offset = cpu_to_le16(end - len);
4448 		}
4449 
4450 		mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4451 				"bucket %llu\n", (unsigned long long)blkno);
4452 
4453 		end -= len;
4454 	}
4455 
4456 	mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4457 			"bucket %llu\n", (unsigned long long)blkno);
4458 
4459 	if (xh_free_start == end)
4460 		goto out;
4461 
4462 	memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4463 	xh->xh_free_start = cpu_to_le16(end);
4464 
4465 	/* sort the entries by their name_hash. */
4466 	sort(entries, le16_to_cpu(xh->xh_count),
4467 	     sizeof(struct ocfs2_xattr_entry),
4468 	     cmp_xe, swap_xe);
4469 
4470 	buf = bucket_buf;
4471 	for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4472 		memcpy(bucket_block(bucket, i), buf, blocksize);
4473 	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4474 
4475 out:
4476 	kfree(bucket_buf);
4477 	return ret;
4478 }
4479 
4480 /*
4481  * prev_blkno points to the start of an existing extent.  new_blkno
4482  * points to a newly allocated extent.  Because we know each of our
4483  * clusters contains more than bucket, we can easily split one cluster
4484  * at a bucket boundary.  So we take the last cluster of the existing
4485  * extent and split it down the middle.  We move the last half of the
4486  * buckets in the last cluster of the existing extent over to the new
4487  * extent.
4488  *
4489  * first_bh is the buffer at prev_blkno so we can update the existing
4490  * extent's bucket count.  header_bh is the bucket were we were hoping
4491  * to insert our xattr.  If the bucket move places the target in the new
4492  * extent, we'll update first_bh and header_bh after modifying the old
4493  * extent.
4494  *
4495  * first_hash will be set as the 1st xe's name_hash in the new extent.
4496  */
4497 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4498 					       handle_t *handle,
4499 					       struct ocfs2_xattr_bucket *first,
4500 					       struct ocfs2_xattr_bucket *target,
4501 					       u64 new_blkno,
4502 					       u32 num_clusters,
4503 					       u32 *first_hash)
4504 {
4505 	int ret;
4506 	struct super_block *sb = inode->i_sb;
4507 	int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4508 	int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4509 	int to_move = num_buckets / 2;
4510 	u64 src_blkno;
4511 	u64 last_cluster_blkno = bucket_blkno(first) +
4512 		((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4513 
4514 	BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4515 	BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4516 
4517 	trace_ocfs2_mv_xattr_bucket_cross_cluster(
4518 				(unsigned long long)last_cluster_blkno,
4519 				(unsigned long long)new_blkno);
4520 
4521 	ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4522 				     last_cluster_blkno, new_blkno,
4523 				     to_move, first_hash);
4524 	if (ret) {
4525 		mlog_errno(ret);
4526 		goto out;
4527 	}
4528 
4529 	/* This is the first bucket that got moved */
4530 	src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4531 
4532 	/*
4533 	 * If the target bucket was part of the moved buckets, we need to
4534 	 * update first and target.
4535 	 */
4536 	if (bucket_blkno(target) >= src_blkno) {
4537 		/* Find the block for the new target bucket */
4538 		src_blkno = new_blkno +
4539 			(bucket_blkno(target) - src_blkno);
4540 
4541 		ocfs2_xattr_bucket_relse(first);
4542 		ocfs2_xattr_bucket_relse(target);
4543 
4544 		/*
4545 		 * These shouldn't fail - the buffers are in the
4546 		 * journal from ocfs2_cp_xattr_bucket().
4547 		 */
4548 		ret = ocfs2_read_xattr_bucket(first, new_blkno);
4549 		if (ret) {
4550 			mlog_errno(ret);
4551 			goto out;
4552 		}
4553 		ret = ocfs2_read_xattr_bucket(target, src_blkno);
4554 		if (ret)
4555 			mlog_errno(ret);
4556 
4557 	}
4558 
4559 out:
4560 	return ret;
4561 }
4562 
4563 /*
4564  * Find the suitable pos when we divide a bucket into 2.
4565  * We have to make sure the xattrs with the same hash value exist
4566  * in the same bucket.
4567  *
4568  * If this ocfs2_xattr_header covers more than one hash value, find a
4569  * place where the hash value changes.  Try to find the most even split.
4570  * The most common case is that all entries have different hash values,
4571  * and the first check we make will find a place to split.
4572  */
4573 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4574 {
4575 	struct ocfs2_xattr_entry *entries = xh->xh_entries;
4576 	int count = le16_to_cpu(xh->xh_count);
4577 	int delta, middle = count / 2;
4578 
4579 	/*
4580 	 * We start at the middle.  Each step gets farther away in both
4581 	 * directions.  We therefore hit the change in hash value
4582 	 * nearest to the middle.  Note that this loop does not execute for
4583 	 * count < 2.
4584 	 */
4585 	for (delta = 0; delta < middle; delta++) {
4586 		/* Let's check delta earlier than middle */
4587 		if (cmp_xe(&entries[middle - delta - 1],
4588 			   &entries[middle - delta]))
4589 			return middle - delta;
4590 
4591 		/* For even counts, don't walk off the end */
4592 		if ((middle + delta + 1) == count)
4593 			continue;
4594 
4595 		/* Now try delta past middle */
4596 		if (cmp_xe(&entries[middle + delta],
4597 			   &entries[middle + delta + 1]))
4598 			return middle + delta + 1;
4599 	}
4600 
4601 	/* Every entry had the same hash */
4602 	return count;
4603 }
4604 
4605 /*
4606  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4607  * first_hash will record the 1st hash of the new bucket.
4608  *
4609  * Normally half of the xattrs will be moved.  But we have to make
4610  * sure that the xattrs with the same hash value are stored in the
4611  * same bucket. If all the xattrs in this bucket have the same hash
4612  * value, the new bucket will be initialized as an empty one and the
4613  * first_hash will be initialized as (hash_value+1).
4614  */
4615 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4616 				    handle_t *handle,
4617 				    u64 blk,
4618 				    u64 new_blk,
4619 				    u32 *first_hash,
4620 				    int new_bucket_head)
4621 {
4622 	int ret, i;
4623 	int count, start, len, name_value_len = 0, name_offset = 0;
4624 	struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4625 	struct ocfs2_xattr_header *xh;
4626 	struct ocfs2_xattr_entry *xe;
4627 	int blocksize = inode->i_sb->s_blocksize;
4628 
4629 	trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4630 					      (unsigned long long)new_blk);
4631 
4632 	s_bucket = ocfs2_xattr_bucket_new(inode);
4633 	t_bucket = ocfs2_xattr_bucket_new(inode);
4634 	if (!s_bucket || !t_bucket) {
4635 		ret = -ENOMEM;
4636 		mlog_errno(ret);
4637 		goto out;
4638 	}
4639 
4640 	ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4641 	if (ret) {
4642 		mlog_errno(ret);
4643 		goto out;
4644 	}
4645 
4646 	ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4647 						OCFS2_JOURNAL_ACCESS_WRITE);
4648 	if (ret) {
4649 		mlog_errno(ret);
4650 		goto out;
4651 	}
4652 
4653 	/*
4654 	 * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4655 	 * there's no need to read it.
4656 	 */
4657 	ret = ocfs2_init_xattr_bucket(t_bucket, new_blk, new_bucket_head);
4658 	if (ret) {
4659 		mlog_errno(ret);
4660 		goto out;
4661 	}
4662 
4663 	/*
4664 	 * Hey, if we're overwriting t_bucket, what difference does
4665 	 * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4666 	 * same part of ocfs2_cp_xattr_bucket().
4667 	 */
4668 	ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4669 						new_bucket_head ?
4670 						OCFS2_JOURNAL_ACCESS_CREATE :
4671 						OCFS2_JOURNAL_ACCESS_WRITE);
4672 	if (ret) {
4673 		mlog_errno(ret);
4674 		goto out;
4675 	}
4676 
4677 	xh = bucket_xh(s_bucket);
4678 	count = le16_to_cpu(xh->xh_count);
4679 	start = ocfs2_xattr_find_divide_pos(xh);
4680 
4681 	if (start == count) {
4682 		xe = &xh->xh_entries[start-1];
4683 
4684 		/*
4685 		 * initialized a new empty bucket here.
4686 		 * The hash value is set as one larger than
4687 		 * that of the last entry in the previous bucket.
4688 		 */
4689 		for (i = 0; i < t_bucket->bu_blocks; i++)
4690 			memset(bucket_block(t_bucket, i), 0, blocksize);
4691 
4692 		xh = bucket_xh(t_bucket);
4693 		xh->xh_free_start = cpu_to_le16(blocksize);
4694 		xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4695 		le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4696 
4697 		goto set_num_buckets;
4698 	}
4699 
4700 	/* copy the whole bucket to the new first. */
4701 	ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4702 
4703 	/* update the new bucket. */
4704 	xh = bucket_xh(t_bucket);
4705 
4706 	/*
4707 	 * Calculate the total name/value len and xh_free_start for
4708 	 * the old bucket first.
4709 	 */
4710 	name_offset = OCFS2_XATTR_BUCKET_SIZE;
4711 	name_value_len = 0;
4712 	for (i = 0; i < start; i++) {
4713 		xe = &xh->xh_entries[i];
4714 		name_value_len += namevalue_size_xe(xe);
4715 		if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4716 			name_offset = le16_to_cpu(xe->xe_name_offset);
4717 	}
4718 
4719 	/*
4720 	 * Now begin the modification to the new bucket.
4721 	 *
4722 	 * In the new bucket, We just move the xattr entry to the beginning
4723 	 * and don't touch the name/value. So there will be some holes in the
4724 	 * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4725 	 * called.
4726 	 */
4727 	xe = &xh->xh_entries[start];
4728 	len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4729 	trace_ocfs2_divide_xattr_bucket_move(len,
4730 			(int)((char *)xe - (char *)xh),
4731 			(int)((char *)xh->xh_entries - (char *)xh));
4732 	memmove((char *)xh->xh_entries, (char *)xe, len);
4733 	xe = &xh->xh_entries[count - start];
4734 	len = sizeof(struct ocfs2_xattr_entry) * start;
4735 	memset((char *)xe, 0, len);
4736 
4737 	le16_add_cpu(&xh->xh_count, -start);
4738 	le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4739 
4740 	/* Calculate xh_free_start for the new bucket. */
4741 	xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4742 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4743 		xe = &xh->xh_entries[i];
4744 		if (le16_to_cpu(xe->xe_name_offset) <
4745 		    le16_to_cpu(xh->xh_free_start))
4746 			xh->xh_free_start = xe->xe_name_offset;
4747 	}
4748 
4749 set_num_buckets:
4750 	/* set xh->xh_num_buckets for the new xh. */
4751 	if (new_bucket_head)
4752 		xh->xh_num_buckets = cpu_to_le16(1);
4753 	else
4754 		xh->xh_num_buckets = 0;
4755 
4756 	ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4757 
4758 	/* store the first_hash of the new bucket. */
4759 	if (first_hash)
4760 		*first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4761 
4762 	/*
4763 	 * Now only update the 1st block of the old bucket.  If we
4764 	 * just added a new empty bucket, there is no need to modify
4765 	 * it.
4766 	 */
4767 	if (start == count)
4768 		goto out;
4769 
4770 	xh = bucket_xh(s_bucket);
4771 	memset(&xh->xh_entries[start], 0,
4772 	       sizeof(struct ocfs2_xattr_entry) * (count - start));
4773 	xh->xh_count = cpu_to_le16(start);
4774 	xh->xh_free_start = cpu_to_le16(name_offset);
4775 	xh->xh_name_value_len = cpu_to_le16(name_value_len);
4776 
4777 	ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4778 
4779 out:
4780 	ocfs2_xattr_bucket_free(s_bucket);
4781 	ocfs2_xattr_bucket_free(t_bucket);
4782 
4783 	return ret;
4784 }
4785 
4786 /*
4787  * Copy xattr from one bucket to another bucket.
4788  *
4789  * The caller must make sure that the journal transaction
4790  * has enough space for journaling.
4791  */
4792 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4793 				 handle_t *handle,
4794 				 u64 s_blkno,
4795 				 u64 t_blkno,
4796 				 int t_is_new)
4797 {
4798 	int ret;
4799 	struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4800 
4801 	BUG_ON(s_blkno == t_blkno);
4802 
4803 	trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4804 				    (unsigned long long)t_blkno,
4805 				    t_is_new);
4806 
4807 	s_bucket = ocfs2_xattr_bucket_new(inode);
4808 	t_bucket = ocfs2_xattr_bucket_new(inode);
4809 	if (!s_bucket || !t_bucket) {
4810 		ret = -ENOMEM;
4811 		mlog_errno(ret);
4812 		goto out;
4813 	}
4814 
4815 	ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4816 	if (ret)
4817 		goto out;
4818 
4819 	/*
4820 	 * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4821 	 * there's no need to read it.
4822 	 */
4823 	ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno, t_is_new);
4824 	if (ret)
4825 		goto out;
4826 
4827 	/*
4828 	 * Hey, if we're overwriting t_bucket, what difference does
4829 	 * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4830 	 * cluster to fill, we came here from
4831 	 * ocfs2_mv_xattr_buckets(), and it is really new -
4832 	 * ACCESS_CREATE is required.  But we also might have moved data
4833 	 * out of t_bucket before extending back into it.
4834 	 * ocfs2_add_new_xattr_bucket() can do this - its call to
4835 	 * ocfs2_add_new_xattr_cluster() may have created a new extent
4836 	 * and copied out the end of the old extent.  Then it re-extends
4837 	 * the old extent back to create space for new xattrs.  That's
4838 	 * how we get here, and the bucket isn't really new.
4839 	 */
4840 	ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4841 						t_is_new ?
4842 						OCFS2_JOURNAL_ACCESS_CREATE :
4843 						OCFS2_JOURNAL_ACCESS_WRITE);
4844 	if (ret)
4845 		goto out;
4846 
4847 	ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4848 	ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4849 
4850 out:
4851 	ocfs2_xattr_bucket_free(t_bucket);
4852 	ocfs2_xattr_bucket_free(s_bucket);
4853 
4854 	return ret;
4855 }
4856 
4857 /*
4858  * src_blk points to the start of an existing extent.  last_blk points to
4859  * last cluster in that extent.  to_blk points to a newly allocated
4860  * extent.  We copy the buckets from the cluster at last_blk to the new
4861  * extent.  If start_bucket is non-zero, we skip that many buckets before
4862  * we start copying.  The new extent's xh_num_buckets gets set to the
4863  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4864  * by the same amount.
4865  */
4866 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4867 				  u64 src_blk, u64 last_blk, u64 to_blk,
4868 				  unsigned int start_bucket,
4869 				  u32 *first_hash)
4870 {
4871 	int i, ret, credits;
4872 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4873 	int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4874 	int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4875 	struct ocfs2_xattr_bucket *old_first, *new_first;
4876 
4877 	trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4878 				     (unsigned long long)to_blk);
4879 
4880 	BUG_ON(start_bucket >= num_buckets);
4881 	if (start_bucket) {
4882 		num_buckets -= start_bucket;
4883 		last_blk += (start_bucket * blks_per_bucket);
4884 	}
4885 
4886 	/* The first bucket of the original extent */
4887 	old_first = ocfs2_xattr_bucket_new(inode);
4888 	/* The first bucket of the new extent */
4889 	new_first = ocfs2_xattr_bucket_new(inode);
4890 	if (!old_first || !new_first) {
4891 		ret = -ENOMEM;
4892 		mlog_errno(ret);
4893 		goto out;
4894 	}
4895 
4896 	ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4897 	if (ret) {
4898 		mlog_errno(ret);
4899 		goto out;
4900 	}
4901 
4902 	/*
4903 	 * We need to update the first bucket of the old extent and all
4904 	 * the buckets going to the new extent.
4905 	 */
4906 	credits = ((num_buckets + 1) * blks_per_bucket);
4907 	ret = ocfs2_extend_trans(handle, credits);
4908 	if (ret) {
4909 		mlog_errno(ret);
4910 		goto out;
4911 	}
4912 
4913 	ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4914 						OCFS2_JOURNAL_ACCESS_WRITE);
4915 	if (ret) {
4916 		mlog_errno(ret);
4917 		goto out;
4918 	}
4919 
4920 	for (i = 0; i < num_buckets; i++) {
4921 		ret = ocfs2_cp_xattr_bucket(inode, handle,
4922 					    last_blk + (i * blks_per_bucket),
4923 					    to_blk + (i * blks_per_bucket),
4924 					    1);
4925 		if (ret) {
4926 			mlog_errno(ret);
4927 			goto out;
4928 		}
4929 	}
4930 
4931 	/*
4932 	 * Get the new bucket ready before we dirty anything
4933 	 * (This actually shouldn't fail, because we already dirtied
4934 	 * it once in ocfs2_cp_xattr_bucket()).
4935 	 */
4936 	ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4937 	if (ret) {
4938 		mlog_errno(ret);
4939 		goto out;
4940 	}
4941 	ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4942 						OCFS2_JOURNAL_ACCESS_WRITE);
4943 	if (ret) {
4944 		mlog_errno(ret);
4945 		goto out;
4946 	}
4947 
4948 	/* Now update the headers */
4949 	le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4950 	ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4951 
4952 	bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4953 	ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4954 
4955 	if (first_hash)
4956 		*first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4957 
4958 out:
4959 	ocfs2_xattr_bucket_free(new_first);
4960 	ocfs2_xattr_bucket_free(old_first);
4961 	return ret;
4962 }
4963 
4964 /*
4965  * Move some xattrs in this cluster to the new cluster.
4966  * This function should only be called when bucket size == cluster size.
4967  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4968  */
4969 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4970 				      handle_t *handle,
4971 				      u64 prev_blk,
4972 				      u64 new_blk,
4973 				      u32 *first_hash)
4974 {
4975 	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4976 	int ret, credits = 2 * blk_per_bucket;
4977 
4978 	BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4979 
4980 	ret = ocfs2_extend_trans(handle, credits);
4981 	if (ret) {
4982 		mlog_errno(ret);
4983 		return ret;
4984 	}
4985 
4986 	/* Move half of the xattr in start_blk to the next bucket. */
4987 	return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4988 					  new_blk, first_hash, 1);
4989 }
4990 
4991 /*
4992  * Move some xattrs from the old cluster to the new one since they are not
4993  * contiguous in ocfs2 xattr tree.
4994  *
4995  * new_blk starts a new separate cluster, and we will move some xattrs from
4996  * prev_blk to it. v_start will be set as the first name hash value in this
4997  * new cluster so that it can be used as e_cpos during tree insertion and
4998  * don't collide with our original b-tree operations. first_bh and header_bh
4999  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
5000  * to extend the insert bucket.
5001  *
5002  * The problem is how much xattr should we move to the new one and when should
5003  * we update first_bh and header_bh?
5004  * 1. If cluster size > bucket size, that means the previous cluster has more
5005  *    than 1 bucket, so just move half nums of bucket into the new cluster and
5006  *    update the first_bh and header_bh if the insert bucket has been moved
5007  *    to the new cluster.
5008  * 2. If cluster_size == bucket_size:
5009  *    a) If the previous extent rec has more than one cluster and the insert
5010  *       place isn't in the last cluster, copy the entire last cluster to the
5011  *       new one. This time, we don't need to upate the first_bh and header_bh
5012  *       since they will not be moved into the new cluster.
5013  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
5014  *       the new one. And we set the extend flag to zero if the insert place is
5015  *       moved into the new allocated cluster since no extend is needed.
5016  */
5017 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5018 					    handle_t *handle,
5019 					    struct ocfs2_xattr_bucket *first,
5020 					    struct ocfs2_xattr_bucket *target,
5021 					    u64 new_blk,
5022 					    u32 prev_clusters,
5023 					    u32 *v_start,
5024 					    int *extend)
5025 {
5026 	int ret;
5027 
5028 	trace_ocfs2_adjust_xattr_cross_cluster(
5029 			(unsigned long long)bucket_blkno(first),
5030 			(unsigned long long)new_blk, prev_clusters);
5031 
5032 	if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5033 		ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5034 							  handle,
5035 							  first, target,
5036 							  new_blk,
5037 							  prev_clusters,
5038 							  v_start);
5039 		if (ret)
5040 			mlog_errno(ret);
5041 	} else {
5042 		/* The start of the last cluster in the first extent */
5043 		u64 last_blk = bucket_blkno(first) +
5044 			((prev_clusters - 1) *
5045 			 ocfs2_clusters_to_blocks(inode->i_sb, 1));
5046 
5047 		if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5048 			ret = ocfs2_mv_xattr_buckets(inode, handle,
5049 						     bucket_blkno(first),
5050 						     last_blk, new_blk, 0,
5051 						     v_start);
5052 			if (ret)
5053 				mlog_errno(ret);
5054 		} else {
5055 			ret = ocfs2_divide_xattr_cluster(inode, handle,
5056 							 last_blk, new_blk,
5057 							 v_start);
5058 			if (ret)
5059 				mlog_errno(ret);
5060 
5061 			if ((bucket_blkno(target) == last_blk) && extend)
5062 				*extend = 0;
5063 		}
5064 	}
5065 
5066 	return ret;
5067 }
5068 
5069 /*
5070  * Add a new cluster for xattr storage.
5071  *
5072  * If the new cluster is contiguous with the previous one, it will be
5073  * appended to the same extent record, and num_clusters will be updated.
5074  * If not, we will insert a new extent for it and move some xattrs in
5075  * the last cluster into the new allocated one.
5076  * We also need to limit the maximum size of a btree leaf, otherwise we'll
5077  * lose the benefits of hashing because we'll have to search large leaves.
5078  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5079  * if it's bigger).
5080  *
5081  * first_bh is the first block of the previous extent rec and header_bh
5082  * indicates the bucket we will insert the new xattrs. They will be updated
5083  * when the header_bh is moved into the new cluster.
5084  */
5085 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5086 				       struct buffer_head *root_bh,
5087 				       struct ocfs2_xattr_bucket *first,
5088 				       struct ocfs2_xattr_bucket *target,
5089 				       u32 *num_clusters,
5090 				       u32 prev_cpos,
5091 				       int *extend,
5092 				       struct ocfs2_xattr_set_ctxt *ctxt)
5093 {
5094 	int ret;
5095 	u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5096 	u32 prev_clusters = *num_clusters;
5097 	u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5098 	u64 block;
5099 	handle_t *handle = ctxt->handle;
5100 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5101 	struct ocfs2_extent_tree et;
5102 
5103 	trace_ocfs2_add_new_xattr_cluster_begin(
5104 		(unsigned long long)OCFS2_I(inode)->ip_blkno,
5105 		(unsigned long long)bucket_blkno(first),
5106 		prev_cpos, prev_clusters);
5107 
5108 	ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5109 
5110 	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5111 				      OCFS2_JOURNAL_ACCESS_WRITE);
5112 	if (ret < 0) {
5113 		mlog_errno(ret);
5114 		goto leave;
5115 	}
5116 
5117 	ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5118 				     clusters_to_add, &bit_off, &num_bits);
5119 	if (ret < 0) {
5120 		if (ret != -ENOSPC)
5121 			mlog_errno(ret);
5122 		goto leave;
5123 	}
5124 
5125 	BUG_ON(num_bits > clusters_to_add);
5126 
5127 	block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5128 	trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5129 
5130 	if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5131 	    (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5132 	     OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5133 		/*
5134 		 * If this cluster is contiguous with the old one and
5135 		 * adding this new cluster, we don't surpass the limit of
5136 		 * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5137 		 * initialized and used like other buckets in the previous
5138 		 * cluster.
5139 		 * So add it as a contiguous one. The caller will handle
5140 		 * its init process.
5141 		 */
5142 		v_start = prev_cpos + prev_clusters;
5143 		*num_clusters = prev_clusters + num_bits;
5144 	} else {
5145 		ret = ocfs2_adjust_xattr_cross_cluster(inode,
5146 						       handle,
5147 						       first,
5148 						       target,
5149 						       block,
5150 						       prev_clusters,
5151 						       &v_start,
5152 						       extend);
5153 		if (ret) {
5154 			mlog_errno(ret);
5155 			goto leave;
5156 		}
5157 	}
5158 
5159 	trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5160 						 v_start, num_bits);
5161 	ret = ocfs2_insert_extent(handle, &et, v_start, block,
5162 				  num_bits, 0, ctxt->meta_ac);
5163 	if (ret < 0) {
5164 		mlog_errno(ret);
5165 		goto leave;
5166 	}
5167 
5168 	ocfs2_journal_dirty(handle, root_bh);
5169 
5170 leave:
5171 	return ret;
5172 }
5173 
5174 /*
5175  * We are given an extent.  'first' is the bucket at the very front of
5176  * the extent.  The extent has space for an additional bucket past
5177  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
5178  * of the target bucket.  We wish to shift every bucket past the target
5179  * down one, filling in that additional space.  When we get back to the
5180  * target, we split the target between itself and the now-empty bucket
5181  * at target+1 (aka, target_blkno + blks_per_bucket).
5182  */
5183 static int ocfs2_extend_xattr_bucket(struct inode *inode,
5184 				     handle_t *handle,
5185 				     struct ocfs2_xattr_bucket *first,
5186 				     u64 target_blk,
5187 				     u32 num_clusters)
5188 {
5189 	int ret, credits;
5190 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5191 	u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5192 	u64 end_blk;
5193 	u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5194 
5195 	trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5196 					(unsigned long long)bucket_blkno(first),
5197 					num_clusters, new_bucket);
5198 
5199 	/* The extent must have room for an additional bucket */
5200 	BUG_ON(new_bucket >=
5201 	       (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5202 
5203 	/* end_blk points to the last existing bucket */
5204 	end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5205 
5206 	/*
5207 	 * end_blk is the start of the last existing bucket.
5208 	 * Thus, (end_blk - target_blk) covers the target bucket and
5209 	 * every bucket after it up to, but not including, the last
5210 	 * existing bucket.  Then we add the last existing bucket, the
5211 	 * new bucket, and the first bucket (3 * blk_per_bucket).
5212 	 */
5213 	credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5214 	ret = ocfs2_extend_trans(handle, credits);
5215 	if (ret) {
5216 		mlog_errno(ret);
5217 		goto out;
5218 	}
5219 
5220 	ret = ocfs2_xattr_bucket_journal_access(handle, first,
5221 						OCFS2_JOURNAL_ACCESS_WRITE);
5222 	if (ret) {
5223 		mlog_errno(ret);
5224 		goto out;
5225 	}
5226 
5227 	while (end_blk != target_blk) {
5228 		ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5229 					    end_blk + blk_per_bucket, 0);
5230 		if (ret)
5231 			goto out;
5232 		end_blk -= blk_per_bucket;
5233 	}
5234 
5235 	/* Move half of the xattr in target_blkno to the next bucket. */
5236 	ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5237 					target_blk + blk_per_bucket, NULL, 0);
5238 
5239 	le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5240 	ocfs2_xattr_bucket_journal_dirty(handle, first);
5241 
5242 out:
5243 	return ret;
5244 }
5245 
5246 /*
5247  * Add new xattr bucket in an extent record and adjust the buckets
5248  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
5249  * bucket we want to insert into.
5250  *
5251  * In the easy case, we will move all the buckets after target down by
5252  * one. Half of target's xattrs will be moved to the next bucket.
5253  *
5254  * If current cluster is full, we'll allocate a new one.  This may not
5255  * be contiguous.  The underlying calls will make sure that there is
5256  * space for the insert, shifting buckets around if necessary.
5257  * 'target' may be moved by those calls.
5258  */
5259 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5260 				      struct buffer_head *xb_bh,
5261 				      struct ocfs2_xattr_bucket *target,
5262 				      struct ocfs2_xattr_set_ctxt *ctxt)
5263 {
5264 	struct ocfs2_xattr_block *xb =
5265 			(struct ocfs2_xattr_block *)xb_bh->b_data;
5266 	struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5267 	struct ocfs2_extent_list *el = &xb_root->xt_list;
5268 	u32 name_hash =
5269 		le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5270 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5271 	int ret, num_buckets, extend = 1;
5272 	u64 p_blkno;
5273 	u32 e_cpos, num_clusters;
5274 	/* The bucket at the front of the extent */
5275 	struct ocfs2_xattr_bucket *first;
5276 
5277 	trace_ocfs2_add_new_xattr_bucket(
5278 				(unsigned long long)bucket_blkno(target));
5279 
5280 	/* The first bucket of the original extent */
5281 	first = ocfs2_xattr_bucket_new(inode);
5282 	if (!first) {
5283 		ret = -ENOMEM;
5284 		mlog_errno(ret);
5285 		goto out;
5286 	}
5287 
5288 	ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5289 				  &num_clusters, el);
5290 	if (ret) {
5291 		mlog_errno(ret);
5292 		goto out;
5293 	}
5294 
5295 	ret = ocfs2_read_xattr_bucket(first, p_blkno);
5296 	if (ret) {
5297 		mlog_errno(ret);
5298 		goto out;
5299 	}
5300 
5301 	num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5302 	if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5303 		/*
5304 		 * This can move first+target if the target bucket moves
5305 		 * to the new extent.
5306 		 */
5307 		ret = ocfs2_add_new_xattr_cluster(inode,
5308 						  xb_bh,
5309 						  first,
5310 						  target,
5311 						  &num_clusters,
5312 						  e_cpos,
5313 						  &extend,
5314 						  ctxt);
5315 		if (ret) {
5316 			mlog_errno(ret);
5317 			goto out;
5318 		}
5319 	}
5320 
5321 	if (extend) {
5322 		ret = ocfs2_extend_xattr_bucket(inode,
5323 						ctxt->handle,
5324 						first,
5325 						bucket_blkno(target),
5326 						num_clusters);
5327 		if (ret)
5328 			mlog_errno(ret);
5329 	}
5330 
5331 out:
5332 	ocfs2_xattr_bucket_free(first);
5333 
5334 	return ret;
5335 }
5336 
5337 /*
5338  * Truncate the specified xe_off entry in xattr bucket.
5339  * bucket is indicated by header_bh and len is the new length.
5340  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5341  *
5342  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5343  */
5344 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5345 					     struct ocfs2_xattr_bucket *bucket,
5346 					     int xe_off,
5347 					     int len,
5348 					     struct ocfs2_xattr_set_ctxt *ctxt)
5349 {
5350 	int ret, offset;
5351 	u64 value_blk;
5352 	struct ocfs2_xattr_entry *xe;
5353 	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5354 	size_t blocksize = inode->i_sb->s_blocksize;
5355 	struct ocfs2_xattr_value_buf vb = {
5356 		.vb_access = ocfs2_journal_access,
5357 	};
5358 
5359 	xe = &xh->xh_entries[xe_off];
5360 
5361 	BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5362 
5363 	offset = le16_to_cpu(xe->xe_name_offset) +
5364 		 OCFS2_XATTR_SIZE(xe->xe_name_len);
5365 
5366 	value_blk = offset / blocksize;
5367 
5368 	/* We don't allow ocfs2_xattr_value to be stored in different block. */
5369 	BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5370 
5371 	vb.vb_bh = bucket->bu_bhs[value_blk];
5372 	BUG_ON(!vb.vb_bh);
5373 
5374 	vb.vb_xv = (struct ocfs2_xattr_value_root *)
5375 		(vb.vb_bh->b_data + offset % blocksize);
5376 
5377 	/*
5378 	 * From here on out we have to dirty the bucket.  The generic
5379 	 * value calls only modify one of the bucket's bhs, but we need
5380 	 * to send the bucket at once.  So if they error, they *could* have
5381 	 * modified something.  We have to assume they did, and dirty
5382 	 * the whole bucket.  This leaves us in a consistent state.
5383 	 */
5384 	trace_ocfs2_xattr_bucket_value_truncate(
5385 			(unsigned long long)bucket_blkno(bucket), xe_off, len);
5386 	ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5387 	if (ret) {
5388 		mlog_errno(ret);
5389 		goto out;
5390 	}
5391 
5392 	ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5393 						OCFS2_JOURNAL_ACCESS_WRITE);
5394 	if (ret) {
5395 		mlog_errno(ret);
5396 		goto out;
5397 	}
5398 
5399 	xe->xe_value_size = cpu_to_le64(len);
5400 
5401 	ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5402 
5403 out:
5404 	return ret;
5405 }
5406 
5407 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5408 				  struct buffer_head *root_bh,
5409 				  u64 blkno,
5410 				  u32 cpos,
5411 				  u32 len,
5412 				  void *para)
5413 {
5414 	int ret;
5415 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5416 	struct inode *tl_inode = osb->osb_tl_inode;
5417 	handle_t *handle;
5418 	struct ocfs2_xattr_block *xb =
5419 			(struct ocfs2_xattr_block *)root_bh->b_data;
5420 	struct ocfs2_alloc_context *meta_ac = NULL;
5421 	struct ocfs2_cached_dealloc_ctxt dealloc;
5422 	struct ocfs2_extent_tree et;
5423 
5424 	ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5425 					  ocfs2_delete_xattr_in_bucket, para);
5426 	if (ret) {
5427 		mlog_errno(ret);
5428 		return ret;
5429 	}
5430 
5431 	ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5432 
5433 	ocfs2_init_dealloc_ctxt(&dealloc);
5434 
5435 	trace_ocfs2_rm_xattr_cluster(
5436 			(unsigned long long)OCFS2_I(inode)->ip_blkno,
5437 			(unsigned long long)blkno, cpos, len);
5438 
5439 	ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5440 					       len);
5441 
5442 	ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5443 	if (ret) {
5444 		mlog_errno(ret);
5445 		return ret;
5446 	}
5447 
5448 	mutex_lock(&tl_inode->i_mutex);
5449 
5450 	if (ocfs2_truncate_log_needs_flush(osb)) {
5451 		ret = __ocfs2_flush_truncate_log(osb);
5452 		if (ret < 0) {
5453 			mlog_errno(ret);
5454 			goto out;
5455 		}
5456 	}
5457 
5458 	handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5459 	if (IS_ERR(handle)) {
5460 		ret = -ENOMEM;
5461 		mlog_errno(ret);
5462 		goto out;
5463 	}
5464 
5465 	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5466 				      OCFS2_JOURNAL_ACCESS_WRITE);
5467 	if (ret) {
5468 		mlog_errno(ret);
5469 		goto out_commit;
5470 	}
5471 
5472 	ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5473 				  &dealloc);
5474 	if (ret) {
5475 		mlog_errno(ret);
5476 		goto out_commit;
5477 	}
5478 
5479 	le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5480 	ocfs2_journal_dirty(handle, root_bh);
5481 
5482 	ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5483 	if (ret)
5484 		mlog_errno(ret);
5485 	ocfs2_update_inode_fsync_trans(handle, inode, 0);
5486 
5487 out_commit:
5488 	ocfs2_commit_trans(osb, handle);
5489 out:
5490 	ocfs2_schedule_truncate_log_flush(osb, 1);
5491 
5492 	mutex_unlock(&tl_inode->i_mutex);
5493 
5494 	if (meta_ac)
5495 		ocfs2_free_alloc_context(meta_ac);
5496 
5497 	ocfs2_run_deallocs(osb, &dealloc);
5498 
5499 	return ret;
5500 }
5501 
5502 /*
5503  * check whether the xattr bucket is filled up with the same hash value.
5504  * If we want to insert the xattr with the same hash, return -ENOSPC.
5505  * If we want to insert a xattr with different hash value, go ahead
5506  * and ocfs2_divide_xattr_bucket will handle this.
5507  */
5508 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5509 					      struct ocfs2_xattr_bucket *bucket,
5510 					      const char *name)
5511 {
5512 	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5513 	u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5514 
5515 	if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5516 		return 0;
5517 
5518 	if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5519 	    xh->xh_entries[0].xe_name_hash) {
5520 		mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5521 		     "hash = %u\n",
5522 		     (unsigned long long)bucket_blkno(bucket),
5523 		     le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5524 		return -ENOSPC;
5525 	}
5526 
5527 	return 0;
5528 }
5529 
5530 /*
5531  * Try to set the entry in the current bucket.  If we fail, the caller
5532  * will handle getting us another bucket.
5533  */
5534 static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5535 					struct ocfs2_xattr_info *xi,
5536 					struct ocfs2_xattr_search *xs,
5537 					struct ocfs2_xattr_set_ctxt *ctxt)
5538 {
5539 	int ret;
5540 	struct ocfs2_xa_loc loc;
5541 
5542 	trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5543 
5544 	ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5545 				       xs->not_found ? NULL : xs->here);
5546 	ret = ocfs2_xa_set(&loc, xi, ctxt);
5547 	if (!ret) {
5548 		xs->here = loc.xl_entry;
5549 		goto out;
5550 	}
5551 	if (ret != -ENOSPC) {
5552 		mlog_errno(ret);
5553 		goto out;
5554 	}
5555 
5556 	/* Ok, we need space.  Let's try defragmenting the bucket. */
5557 	ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5558 					xs->bucket);
5559 	if (ret) {
5560 		mlog_errno(ret);
5561 		goto out;
5562 	}
5563 
5564 	ret = ocfs2_xa_set(&loc, xi, ctxt);
5565 	if (!ret) {
5566 		xs->here = loc.xl_entry;
5567 		goto out;
5568 	}
5569 	if (ret != -ENOSPC)
5570 		mlog_errno(ret);
5571 
5572 
5573 out:
5574 	return ret;
5575 }
5576 
5577 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5578 					     struct ocfs2_xattr_info *xi,
5579 					     struct ocfs2_xattr_search *xs,
5580 					     struct ocfs2_xattr_set_ctxt *ctxt)
5581 {
5582 	int ret;
5583 
5584 	trace_ocfs2_xattr_set_entry_index_block(xi->xi_name);
5585 
5586 	ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5587 	if (!ret)
5588 		goto out;
5589 	if (ret != -ENOSPC) {
5590 		mlog_errno(ret);
5591 		goto out;
5592 	}
5593 
5594 	/* Ack, need more space.  Let's try to get another bucket! */
5595 
5596 	/*
5597 	 * We do not allow for overlapping ranges between buckets. And
5598 	 * the maximum number of collisions we will allow for then is
5599 	 * one bucket's worth, so check it here whether we need to
5600 	 * add a new bucket for the insert.
5601 	 */
5602 	ret = ocfs2_check_xattr_bucket_collision(inode,
5603 						 xs->bucket,
5604 						 xi->xi_name);
5605 	if (ret) {
5606 		mlog_errno(ret);
5607 		goto out;
5608 	}
5609 
5610 	ret = ocfs2_add_new_xattr_bucket(inode,
5611 					 xs->xattr_bh,
5612 					 xs->bucket,
5613 					 ctxt);
5614 	if (ret) {
5615 		mlog_errno(ret);
5616 		goto out;
5617 	}
5618 
5619 	/*
5620 	 * ocfs2_add_new_xattr_bucket() will have updated
5621 	 * xs->bucket if it moved, but it will not have updated
5622 	 * any of the other search fields.  Thus, we drop it and
5623 	 * re-search.  Everything should be cached, so it'll be
5624 	 * quick.
5625 	 */
5626 	ocfs2_xattr_bucket_relse(xs->bucket);
5627 	ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5628 					   xi->xi_name_index,
5629 					   xi->xi_name, xs);
5630 	if (ret && ret != -ENODATA)
5631 		goto out;
5632 	xs->not_found = ret;
5633 
5634 	/* Ok, we have a new bucket, let's try again */
5635 	ret = ocfs2_xattr_set_entry_bucket(inode, xi, xs, ctxt);
5636 	if (ret && (ret != -ENOSPC))
5637 		mlog_errno(ret);
5638 
5639 out:
5640 	return ret;
5641 }
5642 
5643 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5644 					struct ocfs2_xattr_bucket *bucket,
5645 					void *para)
5646 {
5647 	int ret = 0, ref_credits;
5648 	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5649 	u16 i;
5650 	struct ocfs2_xattr_entry *xe;
5651 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5652 	struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5653 	int credits = ocfs2_remove_extent_credits(osb->sb) +
5654 		ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5655 	struct ocfs2_xattr_value_root *xv;
5656 	struct ocfs2_rm_xattr_bucket_para *args =
5657 			(struct ocfs2_rm_xattr_bucket_para *)para;
5658 
5659 	ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5660 
5661 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5662 		xe = &xh->xh_entries[i];
5663 		if (ocfs2_xattr_is_local(xe))
5664 			continue;
5665 
5666 		ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket,
5667 						      i, &xv, NULL);
5668 
5669 		ret = ocfs2_lock_xattr_remove_allocators(inode, xv,
5670 							 args->ref_ci,
5671 							 args->ref_root_bh,
5672 							 &ctxt.meta_ac,
5673 							 &ref_credits);
5674 
5675 		ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
5676 		if (IS_ERR(ctxt.handle)) {
5677 			ret = PTR_ERR(ctxt.handle);
5678 			mlog_errno(ret);
5679 			break;
5680 		}
5681 
5682 		ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5683 							i, 0, &ctxt);
5684 
5685 		ocfs2_commit_trans(osb, ctxt.handle);
5686 		if (ctxt.meta_ac) {
5687 			ocfs2_free_alloc_context(ctxt.meta_ac);
5688 			ctxt.meta_ac = NULL;
5689 		}
5690 		if (ret) {
5691 			mlog_errno(ret);
5692 			break;
5693 		}
5694 	}
5695 
5696 	if (ctxt.meta_ac)
5697 		ocfs2_free_alloc_context(ctxt.meta_ac);
5698 	ocfs2_schedule_truncate_log_flush(osb, 1);
5699 	ocfs2_run_deallocs(osb, &ctxt.dealloc);
5700 	return ret;
5701 }
5702 
5703 /*
5704  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5705  * or change the extent record flag), we need to recalculate
5706  * the metaecc for the whole bucket. So it is done here.
5707  *
5708  * Note:
5709  * We have to give the extra credits for the caller.
5710  */
5711 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5712 					    handle_t *handle,
5713 					    void *para)
5714 {
5715 	int ret;
5716 	struct ocfs2_xattr_bucket *bucket =
5717 			(struct ocfs2_xattr_bucket *)para;
5718 
5719 	ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5720 						OCFS2_JOURNAL_ACCESS_WRITE);
5721 	if (ret) {
5722 		mlog_errno(ret);
5723 		return ret;
5724 	}
5725 
5726 	ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5727 
5728 	return 0;
5729 }
5730 
5731 /*
5732  * Special action we need if the xattr value is refcounted.
5733  *
5734  * 1. If the xattr is refcounted, lock the tree.
5735  * 2. CoW the xattr if we are setting the new value and the value
5736  *    will be stored outside.
5737  * 3. In other case, decrease_refcount will work for us, so just
5738  *    lock the refcount tree, calculate the meta and credits is OK.
5739  *
5740  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5741  * currently CoW is a completed transaction, while this function
5742  * will also lock the allocators and let us deadlock. So we will
5743  * CoW the whole xattr value.
5744  */
5745 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5746 					struct ocfs2_dinode *di,
5747 					struct ocfs2_xattr_info *xi,
5748 					struct ocfs2_xattr_search *xis,
5749 					struct ocfs2_xattr_search *xbs,
5750 					struct ocfs2_refcount_tree **ref_tree,
5751 					int *meta_add,
5752 					int *credits)
5753 {
5754 	int ret = 0;
5755 	struct ocfs2_xattr_block *xb;
5756 	struct ocfs2_xattr_entry *xe;
5757 	char *base;
5758 	u32 p_cluster, num_clusters;
5759 	unsigned int ext_flags;
5760 	int name_offset, name_len;
5761 	struct ocfs2_xattr_value_buf vb;
5762 	struct ocfs2_xattr_bucket *bucket = NULL;
5763 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5764 	struct ocfs2_post_refcount refcount;
5765 	struct ocfs2_post_refcount *p = NULL;
5766 	struct buffer_head *ref_root_bh = NULL;
5767 
5768 	if (!xis->not_found) {
5769 		xe = xis->here;
5770 		name_offset = le16_to_cpu(xe->xe_name_offset);
5771 		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5772 		base = xis->base;
5773 		vb.vb_bh = xis->inode_bh;
5774 		vb.vb_access = ocfs2_journal_access_di;
5775 	} else {
5776 		int i, block_off = 0;
5777 		xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5778 		xe = xbs->here;
5779 		name_offset = le16_to_cpu(xe->xe_name_offset);
5780 		name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5781 		i = xbs->here - xbs->header->xh_entries;
5782 
5783 		if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5784 			ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5785 							bucket_xh(xbs->bucket),
5786 							i, &block_off,
5787 							&name_offset);
5788 			if (ret) {
5789 				mlog_errno(ret);
5790 				goto out;
5791 			}
5792 			base = bucket_block(xbs->bucket, block_off);
5793 			vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5794 			vb.vb_access = ocfs2_journal_access;
5795 
5796 			if (ocfs2_meta_ecc(osb)) {
5797 				/*create parameters for ocfs2_post_refcount. */
5798 				bucket = xbs->bucket;
5799 				refcount.credits = bucket->bu_blocks;
5800 				refcount.para = bucket;
5801 				refcount.func =
5802 					ocfs2_xattr_bucket_post_refcount;
5803 				p = &refcount;
5804 			}
5805 		} else {
5806 			base = xbs->base;
5807 			vb.vb_bh = xbs->xattr_bh;
5808 			vb.vb_access = ocfs2_journal_access_xb;
5809 		}
5810 	}
5811 
5812 	if (ocfs2_xattr_is_local(xe))
5813 		goto out;
5814 
5815 	vb.vb_xv = (struct ocfs2_xattr_value_root *)
5816 				(base + name_offset + name_len);
5817 
5818 	ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5819 				       &num_clusters, &vb.vb_xv->xr_list,
5820 				       &ext_flags);
5821 	if (ret) {
5822 		mlog_errno(ret);
5823 		goto out;
5824 	}
5825 
5826 	/*
5827 	 * We just need to check the 1st extent record, since we always
5828 	 * CoW the whole xattr. So there shouldn't be a xattr with
5829 	 * some REFCOUNT extent recs after the 1st one.
5830 	 */
5831 	if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5832 		goto out;
5833 
5834 	ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5835 				       1, ref_tree, &ref_root_bh);
5836 	if (ret) {
5837 		mlog_errno(ret);
5838 		goto out;
5839 	}
5840 
5841 	/*
5842 	 * If we are deleting the xattr or the new size will be stored inside,
5843 	 * cool, leave it there, the xattr truncate process will remove them
5844 	 * for us(it still needs the refcount tree lock and the meta, credits).
5845 	 * And the worse case is that every cluster truncate will split the
5846 	 * refcount tree, and make the original extent become 3. So we will need
5847 	 * 2 * cluster more extent recs at most.
5848 	 */
5849 	if (!xi->xi_value || xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE) {
5850 
5851 		ret = ocfs2_refcounted_xattr_delete_need(inode,
5852 							 &(*ref_tree)->rf_ci,
5853 							 ref_root_bh, vb.vb_xv,
5854 							 meta_add, credits);
5855 		if (ret)
5856 			mlog_errno(ret);
5857 		goto out;
5858 	}
5859 
5860 	ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5861 				       *ref_tree, ref_root_bh, 0,
5862 				       le32_to_cpu(vb.vb_xv->xr_clusters), p);
5863 	if (ret)
5864 		mlog_errno(ret);
5865 
5866 out:
5867 	brelse(ref_root_bh);
5868 	return ret;
5869 }
5870 
5871 /*
5872  * Add the REFCOUNTED flags for all the extent rec in ocfs2_xattr_value_root.
5873  * The physical clusters will be added to refcount tree.
5874  */
5875 static int ocfs2_xattr_value_attach_refcount(struct inode *inode,
5876 				struct ocfs2_xattr_value_root *xv,
5877 				struct ocfs2_extent_tree *value_et,
5878 				struct ocfs2_caching_info *ref_ci,
5879 				struct buffer_head *ref_root_bh,
5880 				struct ocfs2_cached_dealloc_ctxt *dealloc,
5881 				struct ocfs2_post_refcount *refcount)
5882 {
5883 	int ret = 0;
5884 	u32 clusters = le32_to_cpu(xv->xr_clusters);
5885 	u32 cpos, p_cluster, num_clusters;
5886 	struct ocfs2_extent_list *el = &xv->xr_list;
5887 	unsigned int ext_flags;
5888 
5889 	cpos = 0;
5890 	while (cpos < clusters) {
5891 		ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
5892 					       &num_clusters, el, &ext_flags);
5893 		if (ret) {
5894 			mlog_errno(ret);
5895 			break;
5896 		}
5897 
5898 		cpos += num_clusters;
5899 		if ((ext_flags & OCFS2_EXT_REFCOUNTED))
5900 			continue;
5901 
5902 		BUG_ON(!p_cluster);
5903 
5904 		ret = ocfs2_add_refcount_flag(inode, value_et,
5905 					      ref_ci, ref_root_bh,
5906 					      cpos - num_clusters,
5907 					      p_cluster, num_clusters,
5908 					      dealloc, refcount);
5909 		if (ret) {
5910 			mlog_errno(ret);
5911 			break;
5912 		}
5913 	}
5914 
5915 	return ret;
5916 }
5917 
5918 /*
5919  * Given a normal ocfs2_xattr_header, refcount all the entries which
5920  * have value stored outside.
5921  * Used for xattrs stored in inode and ocfs2_xattr_block.
5922  */
5923 static int ocfs2_xattr_attach_refcount_normal(struct inode *inode,
5924 				struct ocfs2_xattr_value_buf *vb,
5925 				struct ocfs2_xattr_header *header,
5926 				struct ocfs2_caching_info *ref_ci,
5927 				struct buffer_head *ref_root_bh,
5928 				struct ocfs2_cached_dealloc_ctxt *dealloc)
5929 {
5930 
5931 	struct ocfs2_xattr_entry *xe;
5932 	struct ocfs2_xattr_value_root *xv;
5933 	struct ocfs2_extent_tree et;
5934 	int i, ret = 0;
5935 
5936 	for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
5937 		xe = &header->xh_entries[i];
5938 
5939 		if (ocfs2_xattr_is_local(xe))
5940 			continue;
5941 
5942 		xv = (struct ocfs2_xattr_value_root *)((void *)header +
5943 			le16_to_cpu(xe->xe_name_offset) +
5944 			OCFS2_XATTR_SIZE(xe->xe_name_len));
5945 
5946 		vb->vb_xv = xv;
5947 		ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
5948 
5949 		ret = ocfs2_xattr_value_attach_refcount(inode, xv, &et,
5950 							ref_ci, ref_root_bh,
5951 							dealloc, NULL);
5952 		if (ret) {
5953 			mlog_errno(ret);
5954 			break;
5955 		}
5956 	}
5957 
5958 	return ret;
5959 }
5960 
5961 static int ocfs2_xattr_inline_attach_refcount(struct inode *inode,
5962 				struct buffer_head *fe_bh,
5963 				struct ocfs2_caching_info *ref_ci,
5964 				struct buffer_head *ref_root_bh,
5965 				struct ocfs2_cached_dealloc_ctxt *dealloc)
5966 {
5967 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
5968 	struct ocfs2_xattr_header *header = (struct ocfs2_xattr_header *)
5969 				(fe_bh->b_data + inode->i_sb->s_blocksize -
5970 				le16_to_cpu(di->i_xattr_inline_size));
5971 	struct ocfs2_xattr_value_buf vb = {
5972 		.vb_bh = fe_bh,
5973 		.vb_access = ocfs2_journal_access_di,
5974 	};
5975 
5976 	return ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
5977 						  ref_ci, ref_root_bh, dealloc);
5978 }
5979 
5980 struct ocfs2_xattr_tree_value_refcount_para {
5981 	struct ocfs2_caching_info *ref_ci;
5982 	struct buffer_head *ref_root_bh;
5983 	struct ocfs2_cached_dealloc_ctxt *dealloc;
5984 };
5985 
5986 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
5987 					   struct ocfs2_xattr_bucket *bucket,
5988 					   int offset,
5989 					   struct ocfs2_xattr_value_root **xv,
5990 					   struct buffer_head **bh)
5991 {
5992 	int ret, block_off, name_offset;
5993 	struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5994 	struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
5995 	void *base;
5996 
5997 	ret = ocfs2_xattr_bucket_get_name_value(sb,
5998 						bucket_xh(bucket),
5999 						offset,
6000 						&block_off,
6001 						&name_offset);
6002 	if (ret) {
6003 		mlog_errno(ret);
6004 		goto out;
6005 	}
6006 
6007 	base = bucket_block(bucket, block_off);
6008 
6009 	*xv = (struct ocfs2_xattr_value_root *)(base + name_offset +
6010 			 OCFS2_XATTR_SIZE(xe->xe_name_len));
6011 
6012 	if (bh)
6013 		*bh = bucket->bu_bhs[block_off];
6014 out:
6015 	return ret;
6016 }
6017 
6018 /*
6019  * For a given xattr bucket, refcount all the entries which
6020  * have value stored outside.
6021  */
6022 static int ocfs2_xattr_bucket_value_refcount(struct inode *inode,
6023 					     struct ocfs2_xattr_bucket *bucket,
6024 					     void *para)
6025 {
6026 	int i, ret = 0;
6027 	struct ocfs2_extent_tree et;
6028 	struct ocfs2_xattr_tree_value_refcount_para *ref =
6029 			(struct ocfs2_xattr_tree_value_refcount_para *)para;
6030 	struct ocfs2_xattr_header *xh =
6031 			(struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6032 	struct ocfs2_xattr_entry *xe;
6033 	struct ocfs2_xattr_value_buf vb = {
6034 		.vb_access = ocfs2_journal_access,
6035 	};
6036 	struct ocfs2_post_refcount refcount = {
6037 		.credits = bucket->bu_blocks,
6038 		.para = bucket,
6039 		.func = ocfs2_xattr_bucket_post_refcount,
6040 	};
6041 	struct ocfs2_post_refcount *p = NULL;
6042 
6043 	/* We only need post_refcount if we support metaecc. */
6044 	if (ocfs2_meta_ecc(OCFS2_SB(inode->i_sb)))
6045 		p = &refcount;
6046 
6047 	trace_ocfs2_xattr_bucket_value_refcount(
6048 				(unsigned long long)bucket_blkno(bucket),
6049 				le16_to_cpu(xh->xh_count));
6050 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6051 		xe = &xh->xh_entries[i];
6052 
6053 		if (ocfs2_xattr_is_local(xe))
6054 			continue;
6055 
6056 		ret = ocfs2_get_xattr_tree_value_root(inode->i_sb, bucket, i,
6057 						      &vb.vb_xv, &vb.vb_bh);
6058 		if (ret) {
6059 			mlog_errno(ret);
6060 			break;
6061 		}
6062 
6063 		ocfs2_init_xattr_value_extent_tree(&et,
6064 						   INODE_CACHE(inode), &vb);
6065 
6066 		ret = ocfs2_xattr_value_attach_refcount(inode, vb.vb_xv,
6067 							&et, ref->ref_ci,
6068 							ref->ref_root_bh,
6069 							ref->dealloc, p);
6070 		if (ret) {
6071 			mlog_errno(ret);
6072 			break;
6073 		}
6074 	}
6075 
6076 	return ret;
6077 
6078 }
6079 
6080 static int ocfs2_refcount_xattr_tree_rec(struct inode *inode,
6081 				     struct buffer_head *root_bh,
6082 				     u64 blkno, u32 cpos, u32 len, void *para)
6083 {
6084 	return ocfs2_iterate_xattr_buckets(inode, blkno, len,
6085 					   ocfs2_xattr_bucket_value_refcount,
6086 					   para);
6087 }
6088 
6089 static int ocfs2_xattr_block_attach_refcount(struct inode *inode,
6090 				struct buffer_head *blk_bh,
6091 				struct ocfs2_caching_info *ref_ci,
6092 				struct buffer_head *ref_root_bh,
6093 				struct ocfs2_cached_dealloc_ctxt *dealloc)
6094 {
6095 	int ret = 0;
6096 	struct ocfs2_xattr_block *xb =
6097 				(struct ocfs2_xattr_block *)blk_bh->b_data;
6098 
6099 	if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
6100 		struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
6101 		struct ocfs2_xattr_value_buf vb = {
6102 			.vb_bh = blk_bh,
6103 			.vb_access = ocfs2_journal_access_xb,
6104 		};
6105 
6106 		ret = ocfs2_xattr_attach_refcount_normal(inode, &vb, header,
6107 							 ref_ci, ref_root_bh,
6108 							 dealloc);
6109 	} else {
6110 		struct ocfs2_xattr_tree_value_refcount_para para = {
6111 			.ref_ci = ref_ci,
6112 			.ref_root_bh = ref_root_bh,
6113 			.dealloc = dealloc,
6114 		};
6115 
6116 		ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
6117 						ocfs2_refcount_xattr_tree_rec,
6118 						&para);
6119 	}
6120 
6121 	return ret;
6122 }
6123 
6124 int ocfs2_xattr_attach_refcount_tree(struct inode *inode,
6125 				     struct buffer_head *fe_bh,
6126 				     struct ocfs2_caching_info *ref_ci,
6127 				     struct buffer_head *ref_root_bh,
6128 				     struct ocfs2_cached_dealloc_ctxt *dealloc)
6129 {
6130 	int ret = 0;
6131 	struct ocfs2_inode_info *oi = OCFS2_I(inode);
6132 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)fe_bh->b_data;
6133 	struct buffer_head *blk_bh = NULL;
6134 
6135 	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
6136 		ret = ocfs2_xattr_inline_attach_refcount(inode, fe_bh,
6137 							 ref_ci, ref_root_bh,
6138 							 dealloc);
6139 		if (ret) {
6140 			mlog_errno(ret);
6141 			goto out;
6142 		}
6143 	}
6144 
6145 	if (!di->i_xattr_loc)
6146 		goto out;
6147 
6148 	ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
6149 				     &blk_bh);
6150 	if (ret < 0) {
6151 		mlog_errno(ret);
6152 		goto out;
6153 	}
6154 
6155 	ret = ocfs2_xattr_block_attach_refcount(inode, blk_bh, ref_ci,
6156 						ref_root_bh, dealloc);
6157 	if (ret)
6158 		mlog_errno(ret);
6159 
6160 	brelse(blk_bh);
6161 out:
6162 
6163 	return ret;
6164 }
6165 
6166 typedef int (should_xattr_reflinked)(struct ocfs2_xattr_entry *xe);
6167 /*
6168  * Store the information we need in xattr reflink.
6169  * old_bh and new_bh are inode bh for the old and new inode.
6170  */
6171 struct ocfs2_xattr_reflink {
6172 	struct inode *old_inode;
6173 	struct inode *new_inode;
6174 	struct buffer_head *old_bh;
6175 	struct buffer_head *new_bh;
6176 	struct ocfs2_caching_info *ref_ci;
6177 	struct buffer_head *ref_root_bh;
6178 	struct ocfs2_cached_dealloc_ctxt *dealloc;
6179 	should_xattr_reflinked *xattr_reflinked;
6180 };
6181 
6182 /*
6183  * Given a xattr header and xe offset,
6184  * return the proper xv and the corresponding bh.
6185  * xattr in inode, block and xattr tree have different implementaions.
6186  */
6187 typedef int (get_xattr_value_root)(struct super_block *sb,
6188 				   struct buffer_head *bh,
6189 				   struct ocfs2_xattr_header *xh,
6190 				   int offset,
6191 				   struct ocfs2_xattr_value_root **xv,
6192 				   struct buffer_head **ret_bh,
6193 				   void *para);
6194 
6195 /*
6196  * Calculate all the xattr value root metadata stored in this xattr header and
6197  * credits we need if we create them from the scratch.
6198  * We use get_xattr_value_root so that all types of xattr container can use it.
6199  */
6200 static int ocfs2_value_metas_in_xattr_header(struct super_block *sb,
6201 					     struct buffer_head *bh,
6202 					     struct ocfs2_xattr_header *xh,
6203 					     int *metas, int *credits,
6204 					     int *num_recs,
6205 					     get_xattr_value_root *func,
6206 					     void *para)
6207 {
6208 	int i, ret = 0;
6209 	struct ocfs2_xattr_value_root *xv;
6210 	struct ocfs2_xattr_entry *xe;
6211 
6212 	for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
6213 		xe = &xh->xh_entries[i];
6214 		if (ocfs2_xattr_is_local(xe))
6215 			continue;
6216 
6217 		ret = func(sb, bh, xh, i, &xv, NULL, para);
6218 		if (ret) {
6219 			mlog_errno(ret);
6220 			break;
6221 		}
6222 
6223 		*metas += le16_to_cpu(xv->xr_list.l_tree_depth) *
6224 			  le16_to_cpu(xv->xr_list.l_next_free_rec);
6225 
6226 		*credits += ocfs2_calc_extend_credits(sb,
6227 						&def_xv.xv.xr_list);
6228 
6229 		/*
6230 		 * If the value is a tree with depth > 1, We don't go deep
6231 		 * to the extent block, so just calculate a maximum record num.
6232 		 */
6233 		if (!xv->xr_list.l_tree_depth)
6234 			*num_recs += le16_to_cpu(xv->xr_list.l_next_free_rec);
6235 		else
6236 			*num_recs += ocfs2_clusters_for_bytes(sb,
6237 							      XATTR_SIZE_MAX);
6238 	}
6239 
6240 	return ret;
6241 }
6242 
6243 /* Used by xattr inode and block to return the right xv and buffer_head. */
6244 static int ocfs2_get_xattr_value_root(struct super_block *sb,
6245 				      struct buffer_head *bh,
6246 				      struct ocfs2_xattr_header *xh,
6247 				      int offset,
6248 				      struct ocfs2_xattr_value_root **xv,
6249 				      struct buffer_head **ret_bh,
6250 				      void *para)
6251 {
6252 	struct ocfs2_xattr_entry *xe = &xh->xh_entries[offset];
6253 
6254 	*xv = (struct ocfs2_xattr_value_root *)((void *)xh +
6255 		le16_to_cpu(xe->xe_name_offset) +
6256 		OCFS2_XATTR_SIZE(xe->xe_name_len));
6257 
6258 	if (ret_bh)
6259 		*ret_bh = bh;
6260 
6261 	return 0;
6262 }
6263 
6264 /*
6265  * Lock the meta_ac and caculate how much credits we need for reflink xattrs.
6266  * It is only used for inline xattr and xattr block.
6267  */
6268 static int ocfs2_reflink_lock_xattr_allocators(struct ocfs2_super *osb,
6269 					struct ocfs2_xattr_header *xh,
6270 					struct buffer_head *ref_root_bh,
6271 					int *credits,
6272 					struct ocfs2_alloc_context **meta_ac)
6273 {
6274 	int ret, meta_add = 0, num_recs = 0;
6275 	struct ocfs2_refcount_block *rb =
6276 			(struct ocfs2_refcount_block *)ref_root_bh->b_data;
6277 
6278 	*credits = 0;
6279 
6280 	ret = ocfs2_value_metas_in_xattr_header(osb->sb, NULL, xh,
6281 						&meta_add, credits, &num_recs,
6282 						ocfs2_get_xattr_value_root,
6283 						NULL);
6284 	if (ret) {
6285 		mlog_errno(ret);
6286 		goto out;
6287 	}
6288 
6289 	/*
6290 	 * We need to add/modify num_recs in refcount tree, so just calculate
6291 	 * an approximate number we need for refcount tree change.
6292 	 * Sometimes we need to split the tree, and after split,  half recs
6293 	 * will be moved to the new block, and a new block can only provide
6294 	 * half number of recs. So we multiple new blocks by 2.
6295 	 */
6296 	num_recs = num_recs / ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6297 	meta_add += num_recs;
6298 	*credits += num_recs + num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6299 	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6300 		*credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6301 			    le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6302 	else
6303 		*credits += 1;
6304 
6305 	ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add, meta_ac);
6306 	if (ret)
6307 		mlog_errno(ret);
6308 
6309 out:
6310 	return ret;
6311 }
6312 
6313 /*
6314  * Given a xattr header, reflink all the xattrs in this container.
6315  * It can be used for inode, block and bucket.
6316  *
6317  * NOTE:
6318  * Before we call this function, the caller has memcpy the xattr in
6319  * old_xh to the new_xh.
6320  *
6321  * If args.xattr_reflinked is set, call it to decide whether the xe should
6322  * be reflinked or not. If not, remove it from the new xattr header.
6323  */
6324 static int ocfs2_reflink_xattr_header(handle_t *handle,
6325 				      struct ocfs2_xattr_reflink *args,
6326 				      struct buffer_head *old_bh,
6327 				      struct ocfs2_xattr_header *xh,
6328 				      struct buffer_head *new_bh,
6329 				      struct ocfs2_xattr_header *new_xh,
6330 				      struct ocfs2_xattr_value_buf *vb,
6331 				      struct ocfs2_alloc_context *meta_ac,
6332 				      get_xattr_value_root *func,
6333 				      void *para)
6334 {
6335 	int ret = 0, i, j;
6336 	struct super_block *sb = args->old_inode->i_sb;
6337 	struct buffer_head *value_bh;
6338 	struct ocfs2_xattr_entry *xe, *last;
6339 	struct ocfs2_xattr_value_root *xv, *new_xv;
6340 	struct ocfs2_extent_tree data_et;
6341 	u32 clusters, cpos, p_cluster, num_clusters;
6342 	unsigned int ext_flags = 0;
6343 
6344 	trace_ocfs2_reflink_xattr_header((unsigned long long)old_bh->b_blocknr,
6345 					 le16_to_cpu(xh->xh_count));
6346 
6347 	last = &new_xh->xh_entries[le16_to_cpu(new_xh->xh_count)];
6348 	for (i = 0, j = 0; i < le16_to_cpu(xh->xh_count); i++, j++) {
6349 		xe = &xh->xh_entries[i];
6350 
6351 		if (args->xattr_reflinked && !args->xattr_reflinked(xe)) {
6352 			xe = &new_xh->xh_entries[j];
6353 
6354 			le16_add_cpu(&new_xh->xh_count, -1);
6355 			if (new_xh->xh_count) {
6356 				memmove(xe, xe + 1,
6357 					(void *)last - (void *)xe);
6358 				memset(last, 0,
6359 				       sizeof(struct ocfs2_xattr_entry));
6360 			}
6361 
6362 			/*
6363 			 * We don't want j to increase in the next round since
6364 			 * it is already moved ahead.
6365 			 */
6366 			j--;
6367 			continue;
6368 		}
6369 
6370 		if (ocfs2_xattr_is_local(xe))
6371 			continue;
6372 
6373 		ret = func(sb, old_bh, xh, i, &xv, NULL, para);
6374 		if (ret) {
6375 			mlog_errno(ret);
6376 			break;
6377 		}
6378 
6379 		ret = func(sb, new_bh, new_xh, j, &new_xv, &value_bh, para);
6380 		if (ret) {
6381 			mlog_errno(ret);
6382 			break;
6383 		}
6384 
6385 		/*
6386 		 * For the xattr which has l_tree_depth = 0, all the extent
6387 		 * recs have already be copied to the new xh with the
6388 		 * propriate OCFS2_EXT_REFCOUNTED flag we just need to
6389 		 * increase the refount count int the refcount tree.
6390 		 *
6391 		 * For the xattr which has l_tree_depth > 0, we need
6392 		 * to initialize it to the empty default value root,
6393 		 * and then insert the extents one by one.
6394 		 */
6395 		if (xv->xr_list.l_tree_depth) {
6396 			memcpy(new_xv, &def_xv, sizeof(def_xv));
6397 			vb->vb_xv = new_xv;
6398 			vb->vb_bh = value_bh;
6399 			ocfs2_init_xattr_value_extent_tree(&data_et,
6400 					INODE_CACHE(args->new_inode), vb);
6401 		}
6402 
6403 		clusters = le32_to_cpu(xv->xr_clusters);
6404 		cpos = 0;
6405 		while (cpos < clusters) {
6406 			ret = ocfs2_xattr_get_clusters(args->old_inode,
6407 						       cpos,
6408 						       &p_cluster,
6409 						       &num_clusters,
6410 						       &xv->xr_list,
6411 						       &ext_flags);
6412 			if (ret) {
6413 				mlog_errno(ret);
6414 				goto out;
6415 			}
6416 
6417 			BUG_ON(!p_cluster);
6418 
6419 			if (xv->xr_list.l_tree_depth) {
6420 				ret = ocfs2_insert_extent(handle,
6421 						&data_et, cpos,
6422 						ocfs2_clusters_to_blocks(
6423 							args->old_inode->i_sb,
6424 							p_cluster),
6425 						num_clusters, ext_flags,
6426 						meta_ac);
6427 				if (ret) {
6428 					mlog_errno(ret);
6429 					goto out;
6430 				}
6431 			}
6432 
6433 			ret = ocfs2_increase_refcount(handle, args->ref_ci,
6434 						      args->ref_root_bh,
6435 						      p_cluster, num_clusters,
6436 						      meta_ac, args->dealloc);
6437 			if (ret) {
6438 				mlog_errno(ret);
6439 				goto out;
6440 			}
6441 
6442 			cpos += num_clusters;
6443 		}
6444 	}
6445 
6446 out:
6447 	return ret;
6448 }
6449 
6450 static int ocfs2_reflink_xattr_inline(struct ocfs2_xattr_reflink *args)
6451 {
6452 	int ret = 0, credits = 0;
6453 	handle_t *handle;
6454 	struct ocfs2_super *osb = OCFS2_SB(args->old_inode->i_sb);
6455 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)args->old_bh->b_data;
6456 	int inline_size = le16_to_cpu(di->i_xattr_inline_size);
6457 	int header_off = osb->sb->s_blocksize - inline_size;
6458 	struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)
6459 					(args->old_bh->b_data + header_off);
6460 	struct ocfs2_xattr_header *new_xh = (struct ocfs2_xattr_header *)
6461 					(args->new_bh->b_data + header_off);
6462 	struct ocfs2_alloc_context *meta_ac = NULL;
6463 	struct ocfs2_inode_info *new_oi;
6464 	struct ocfs2_dinode *new_di;
6465 	struct ocfs2_xattr_value_buf vb = {
6466 		.vb_bh = args->new_bh,
6467 		.vb_access = ocfs2_journal_access_di,
6468 	};
6469 
6470 	ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6471 						  &credits, &meta_ac);
6472 	if (ret) {
6473 		mlog_errno(ret);
6474 		goto out;
6475 	}
6476 
6477 	handle = ocfs2_start_trans(osb, credits);
6478 	if (IS_ERR(handle)) {
6479 		ret = PTR_ERR(handle);
6480 		mlog_errno(ret);
6481 		goto out;
6482 	}
6483 
6484 	ret = ocfs2_journal_access_di(handle, INODE_CACHE(args->new_inode),
6485 				      args->new_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6486 	if (ret) {
6487 		mlog_errno(ret);
6488 		goto out_commit;
6489 	}
6490 
6491 	memcpy(args->new_bh->b_data + header_off,
6492 	       args->old_bh->b_data + header_off, inline_size);
6493 
6494 	new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6495 	new_di->i_xattr_inline_size = cpu_to_le16(inline_size);
6496 
6497 	ret = ocfs2_reflink_xattr_header(handle, args, args->old_bh, xh,
6498 					 args->new_bh, new_xh, &vb, meta_ac,
6499 					 ocfs2_get_xattr_value_root, NULL);
6500 	if (ret) {
6501 		mlog_errno(ret);
6502 		goto out_commit;
6503 	}
6504 
6505 	new_oi = OCFS2_I(args->new_inode);
6506 	/*
6507 	 * Adjust extent record count to reserve space for extended attribute.
6508 	 * Inline data count had been adjusted in ocfs2_duplicate_inline_data().
6509 	 */
6510 	if (!(new_oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) &&
6511 	    !(ocfs2_inode_is_fast_symlink(args->new_inode))) {
6512 		struct ocfs2_extent_list *el = &new_di->id2.i_list;
6513 		le16_add_cpu(&el->l_count, -(inline_size /
6514 					sizeof(struct ocfs2_extent_rec)));
6515 	}
6516 	spin_lock(&new_oi->ip_lock);
6517 	new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL | OCFS2_INLINE_XATTR_FL;
6518 	new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6519 	spin_unlock(&new_oi->ip_lock);
6520 
6521 	ocfs2_journal_dirty(handle, args->new_bh);
6522 
6523 out_commit:
6524 	ocfs2_commit_trans(osb, handle);
6525 
6526 out:
6527 	if (meta_ac)
6528 		ocfs2_free_alloc_context(meta_ac);
6529 	return ret;
6530 }
6531 
6532 static int ocfs2_create_empty_xattr_block(struct inode *inode,
6533 					  struct buffer_head *fe_bh,
6534 					  struct buffer_head **ret_bh,
6535 					  int indexed)
6536 {
6537 	int ret;
6538 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6539 	struct ocfs2_xattr_set_ctxt ctxt;
6540 
6541 	memset(&ctxt, 0, sizeof(ctxt));
6542 	ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &ctxt.meta_ac);
6543 	if (ret < 0) {
6544 		mlog_errno(ret);
6545 		return ret;
6546 	}
6547 
6548 	ctxt.handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_CREATE_CREDITS);
6549 	if (IS_ERR(ctxt.handle)) {
6550 		ret = PTR_ERR(ctxt.handle);
6551 		mlog_errno(ret);
6552 		goto out;
6553 	}
6554 
6555 	trace_ocfs2_create_empty_xattr_block(
6556 				(unsigned long long)fe_bh->b_blocknr, indexed);
6557 	ret = ocfs2_create_xattr_block(inode, fe_bh, &ctxt, indexed,
6558 				       ret_bh);
6559 	if (ret)
6560 		mlog_errno(ret);
6561 
6562 	ocfs2_commit_trans(osb, ctxt.handle);
6563 out:
6564 	ocfs2_free_alloc_context(ctxt.meta_ac);
6565 	return ret;
6566 }
6567 
6568 static int ocfs2_reflink_xattr_block(struct ocfs2_xattr_reflink *args,
6569 				     struct buffer_head *blk_bh,
6570 				     struct buffer_head *new_blk_bh)
6571 {
6572 	int ret = 0, credits = 0;
6573 	handle_t *handle;
6574 	struct ocfs2_inode_info *new_oi = OCFS2_I(args->new_inode);
6575 	struct ocfs2_dinode *new_di;
6576 	struct ocfs2_super *osb = OCFS2_SB(args->new_inode->i_sb);
6577 	int header_off = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
6578 	struct ocfs2_xattr_block *xb =
6579 			(struct ocfs2_xattr_block *)blk_bh->b_data;
6580 	struct ocfs2_xattr_header *xh = &xb->xb_attrs.xb_header;
6581 	struct ocfs2_xattr_block *new_xb =
6582 			(struct ocfs2_xattr_block *)new_blk_bh->b_data;
6583 	struct ocfs2_xattr_header *new_xh = &new_xb->xb_attrs.xb_header;
6584 	struct ocfs2_alloc_context *meta_ac;
6585 	struct ocfs2_xattr_value_buf vb = {
6586 		.vb_bh = new_blk_bh,
6587 		.vb_access = ocfs2_journal_access_xb,
6588 	};
6589 
6590 	ret = ocfs2_reflink_lock_xattr_allocators(osb, xh, args->ref_root_bh,
6591 						  &credits, &meta_ac);
6592 	if (ret) {
6593 		mlog_errno(ret);
6594 		return ret;
6595 	}
6596 
6597 	/* One more credits in case we need to add xattr flags in new inode. */
6598 	handle = ocfs2_start_trans(osb, credits + 1);
6599 	if (IS_ERR(handle)) {
6600 		ret = PTR_ERR(handle);
6601 		mlog_errno(ret);
6602 		goto out;
6603 	}
6604 
6605 	if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6606 		ret = ocfs2_journal_access_di(handle,
6607 					      INODE_CACHE(args->new_inode),
6608 					      args->new_bh,
6609 					      OCFS2_JOURNAL_ACCESS_WRITE);
6610 		if (ret) {
6611 			mlog_errno(ret);
6612 			goto out_commit;
6613 		}
6614 	}
6615 
6616 	ret = ocfs2_journal_access_xb(handle, INODE_CACHE(args->new_inode),
6617 				      new_blk_bh, OCFS2_JOURNAL_ACCESS_WRITE);
6618 	if (ret) {
6619 		mlog_errno(ret);
6620 		goto out_commit;
6621 	}
6622 
6623 	memcpy(new_blk_bh->b_data + header_off, blk_bh->b_data + header_off,
6624 	       osb->sb->s_blocksize - header_off);
6625 
6626 	ret = ocfs2_reflink_xattr_header(handle, args, blk_bh, xh,
6627 					 new_blk_bh, new_xh, &vb, meta_ac,
6628 					 ocfs2_get_xattr_value_root, NULL);
6629 	if (ret) {
6630 		mlog_errno(ret);
6631 		goto out_commit;
6632 	}
6633 
6634 	ocfs2_journal_dirty(handle, new_blk_bh);
6635 
6636 	if (!(new_oi->ip_dyn_features & OCFS2_HAS_XATTR_FL)) {
6637 		new_di = (struct ocfs2_dinode *)args->new_bh->b_data;
6638 		spin_lock(&new_oi->ip_lock);
6639 		new_oi->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
6640 		new_di->i_dyn_features = cpu_to_le16(new_oi->ip_dyn_features);
6641 		spin_unlock(&new_oi->ip_lock);
6642 
6643 		ocfs2_journal_dirty(handle, args->new_bh);
6644 	}
6645 
6646 out_commit:
6647 	ocfs2_commit_trans(osb, handle);
6648 
6649 out:
6650 	ocfs2_free_alloc_context(meta_ac);
6651 	return ret;
6652 }
6653 
6654 struct ocfs2_reflink_xattr_tree_args {
6655 	struct ocfs2_xattr_reflink *reflink;
6656 	struct buffer_head *old_blk_bh;
6657 	struct buffer_head *new_blk_bh;
6658 	struct ocfs2_xattr_bucket *old_bucket;
6659 	struct ocfs2_xattr_bucket *new_bucket;
6660 };
6661 
6662 /*
6663  * NOTE:
6664  * We have to handle the case that both old bucket and new bucket
6665  * will call this function to get the right ret_bh.
6666  * So The caller must give us the right bh.
6667  */
6668 static int ocfs2_get_reflink_xattr_value_root(struct super_block *sb,
6669 					struct buffer_head *bh,
6670 					struct ocfs2_xattr_header *xh,
6671 					int offset,
6672 					struct ocfs2_xattr_value_root **xv,
6673 					struct buffer_head **ret_bh,
6674 					void *para)
6675 {
6676 	struct ocfs2_reflink_xattr_tree_args *args =
6677 			(struct ocfs2_reflink_xattr_tree_args *)para;
6678 	struct ocfs2_xattr_bucket *bucket;
6679 
6680 	if (bh == args->old_bucket->bu_bhs[0])
6681 		bucket = args->old_bucket;
6682 	else
6683 		bucket = args->new_bucket;
6684 
6685 	return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6686 					       xv, ret_bh);
6687 }
6688 
6689 struct ocfs2_value_tree_metas {
6690 	int num_metas;
6691 	int credits;
6692 	int num_recs;
6693 };
6694 
6695 static int ocfs2_value_tree_metas_in_bucket(struct super_block *sb,
6696 					struct buffer_head *bh,
6697 					struct ocfs2_xattr_header *xh,
6698 					int offset,
6699 					struct ocfs2_xattr_value_root **xv,
6700 					struct buffer_head **ret_bh,
6701 					void *para)
6702 {
6703 	struct ocfs2_xattr_bucket *bucket =
6704 				(struct ocfs2_xattr_bucket *)para;
6705 
6706 	return ocfs2_get_xattr_tree_value_root(sb, bucket, offset,
6707 					       xv, ret_bh);
6708 }
6709 
6710 static int ocfs2_calc_value_tree_metas(struct inode *inode,
6711 				      struct ocfs2_xattr_bucket *bucket,
6712 				      void *para)
6713 {
6714 	struct ocfs2_value_tree_metas *metas =
6715 			(struct ocfs2_value_tree_metas *)para;
6716 	struct ocfs2_xattr_header *xh =
6717 			(struct ocfs2_xattr_header *)bucket->bu_bhs[0]->b_data;
6718 
6719 	/* Add the credits for this bucket first. */
6720 	metas->credits += bucket->bu_blocks;
6721 	return ocfs2_value_metas_in_xattr_header(inode->i_sb, bucket->bu_bhs[0],
6722 					xh, &metas->num_metas,
6723 					&metas->credits, &metas->num_recs,
6724 					ocfs2_value_tree_metas_in_bucket,
6725 					bucket);
6726 }
6727 
6728 /*
6729  * Given a xattr extent rec starting from blkno and having len clusters,
6730  * iterate all the buckets calculate how much metadata we need for reflinking
6731  * all the ocfs2_xattr_value_root and lock the allocators accordingly.
6732  */
6733 static int ocfs2_lock_reflink_xattr_rec_allocators(
6734 				struct ocfs2_reflink_xattr_tree_args *args,
6735 				struct ocfs2_extent_tree *xt_et,
6736 				u64 blkno, u32 len, int *credits,
6737 				struct ocfs2_alloc_context **meta_ac,
6738 				struct ocfs2_alloc_context **data_ac)
6739 {
6740 	int ret, num_free_extents;
6741 	struct ocfs2_value_tree_metas metas;
6742 	struct ocfs2_super *osb = OCFS2_SB(args->reflink->old_inode->i_sb);
6743 	struct ocfs2_refcount_block *rb;
6744 
6745 	memset(&metas, 0, sizeof(metas));
6746 
6747 	ret = ocfs2_iterate_xattr_buckets(args->reflink->old_inode, blkno, len,
6748 					  ocfs2_calc_value_tree_metas, &metas);
6749 	if (ret) {
6750 		mlog_errno(ret);
6751 		goto out;
6752 	}
6753 
6754 	*credits = metas.credits;
6755 
6756 	/*
6757 	 * Calculate we need for refcount tree change.
6758 	 *
6759 	 * We need to add/modify num_recs in refcount tree, so just calculate
6760 	 * an approximate number we need for refcount tree change.
6761 	 * Sometimes we need to split the tree, and after split,  half recs
6762 	 * will be moved to the new block, and a new block can only provide
6763 	 * half number of recs. So we multiple new blocks by 2.
6764 	 * In the end, we have to add credits for modifying the already
6765 	 * existed refcount block.
6766 	 */
6767 	rb = (struct ocfs2_refcount_block *)args->reflink->ref_root_bh->b_data;
6768 	metas.num_recs =
6769 		(metas.num_recs + ocfs2_refcount_recs_per_rb(osb->sb) - 1) /
6770 		 ocfs2_refcount_recs_per_rb(osb->sb) * 2;
6771 	metas.num_metas += metas.num_recs;
6772 	*credits += metas.num_recs +
6773 		    metas.num_recs * OCFS2_EXPAND_REFCOUNT_TREE_CREDITS;
6774 	if (le32_to_cpu(rb->rf_flags) & OCFS2_REFCOUNT_TREE_FL)
6775 		*credits += le16_to_cpu(rb->rf_list.l_tree_depth) *
6776 			    le16_to_cpu(rb->rf_list.l_next_free_rec) + 1;
6777 	else
6778 		*credits += 1;
6779 
6780 	/* count in the xattr tree change. */
6781 	num_free_extents = ocfs2_num_free_extents(osb, xt_et);
6782 	if (num_free_extents < 0) {
6783 		ret = num_free_extents;
6784 		mlog_errno(ret);
6785 		goto out;
6786 	}
6787 
6788 	if (num_free_extents < len)
6789 		metas.num_metas += ocfs2_extend_meta_needed(xt_et->et_root_el);
6790 
6791 	*credits += ocfs2_calc_extend_credits(osb->sb,
6792 					      xt_et->et_root_el);
6793 
6794 	if (metas.num_metas) {
6795 		ret = ocfs2_reserve_new_metadata_blocks(osb, metas.num_metas,
6796 							meta_ac);
6797 		if (ret) {
6798 			mlog_errno(ret);
6799 			goto out;
6800 		}
6801 	}
6802 
6803 	if (len) {
6804 		ret = ocfs2_reserve_clusters(osb, len, data_ac);
6805 		if (ret)
6806 			mlog_errno(ret);
6807 	}
6808 out:
6809 	if (ret) {
6810 		if (*meta_ac) {
6811 			ocfs2_free_alloc_context(*meta_ac);
6812 			*meta_ac = NULL;
6813 		}
6814 	}
6815 
6816 	return ret;
6817 }
6818 
6819 static int ocfs2_reflink_xattr_bucket(handle_t *handle,
6820 				u64 blkno, u64 new_blkno, u32 clusters,
6821 				u32 *cpos, int num_buckets,
6822 				struct ocfs2_alloc_context *meta_ac,
6823 				struct ocfs2_alloc_context *data_ac,
6824 				struct ocfs2_reflink_xattr_tree_args *args)
6825 {
6826 	int i, j, ret = 0;
6827 	struct super_block *sb = args->reflink->old_inode->i_sb;
6828 	int bpb = args->old_bucket->bu_blocks;
6829 	struct ocfs2_xattr_value_buf vb = {
6830 		.vb_access = ocfs2_journal_access,
6831 	};
6832 
6833 	for (i = 0; i < num_buckets; i++, blkno += bpb, new_blkno += bpb) {
6834 		ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6835 		if (ret) {
6836 			mlog_errno(ret);
6837 			break;
6838 		}
6839 
6840 		ret = ocfs2_init_xattr_bucket(args->new_bucket, new_blkno, 1);
6841 		if (ret) {
6842 			mlog_errno(ret);
6843 			break;
6844 		}
6845 
6846 		ret = ocfs2_xattr_bucket_journal_access(handle,
6847 						args->new_bucket,
6848 						OCFS2_JOURNAL_ACCESS_CREATE);
6849 		if (ret) {
6850 			mlog_errno(ret);
6851 			break;
6852 		}
6853 
6854 		for (j = 0; j < bpb; j++)
6855 			memcpy(bucket_block(args->new_bucket, j),
6856 			       bucket_block(args->old_bucket, j),
6857 			       sb->s_blocksize);
6858 
6859 		/*
6860 		 * Record the start cpos so that we can use it to initialize
6861 		 * our xattr tree we also set the xh_num_bucket for the new
6862 		 * bucket.
6863 		 */
6864 		if (i == 0) {
6865 			*cpos = le32_to_cpu(bucket_xh(args->new_bucket)->
6866 					    xh_entries[0].xe_name_hash);
6867 			bucket_xh(args->new_bucket)->xh_num_buckets =
6868 				cpu_to_le16(num_buckets);
6869 		}
6870 
6871 		ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6872 
6873 		ret = ocfs2_reflink_xattr_header(handle, args->reflink,
6874 					args->old_bucket->bu_bhs[0],
6875 					bucket_xh(args->old_bucket),
6876 					args->new_bucket->bu_bhs[0],
6877 					bucket_xh(args->new_bucket),
6878 					&vb, meta_ac,
6879 					ocfs2_get_reflink_xattr_value_root,
6880 					args);
6881 		if (ret) {
6882 			mlog_errno(ret);
6883 			break;
6884 		}
6885 
6886 		/*
6887 		 * Re-access and dirty the bucket to calculate metaecc.
6888 		 * Because we may extend the transaction in reflink_xattr_header
6889 		 * which will let the already accessed block gone.
6890 		 */
6891 		ret = ocfs2_xattr_bucket_journal_access(handle,
6892 						args->new_bucket,
6893 						OCFS2_JOURNAL_ACCESS_WRITE);
6894 		if (ret) {
6895 			mlog_errno(ret);
6896 			break;
6897 		}
6898 
6899 		ocfs2_xattr_bucket_journal_dirty(handle, args->new_bucket);
6900 
6901 		ocfs2_xattr_bucket_relse(args->old_bucket);
6902 		ocfs2_xattr_bucket_relse(args->new_bucket);
6903 	}
6904 
6905 	ocfs2_xattr_bucket_relse(args->old_bucket);
6906 	ocfs2_xattr_bucket_relse(args->new_bucket);
6907 	return ret;
6908 }
6909 
6910 static int ocfs2_reflink_xattr_buckets(handle_t *handle,
6911 				struct inode *inode,
6912 				struct ocfs2_reflink_xattr_tree_args *args,
6913 				struct ocfs2_extent_tree *et,
6914 				struct ocfs2_alloc_context *meta_ac,
6915 				struct ocfs2_alloc_context *data_ac,
6916 				u64 blkno, u32 cpos, u32 len)
6917 {
6918 	int ret, first_inserted = 0;
6919 	u32 p_cluster, num_clusters, reflink_cpos = 0;
6920 	u64 new_blkno;
6921 	unsigned int num_buckets, reflink_buckets;
6922 	unsigned int bpc =
6923 		ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
6924 
6925 	ret = ocfs2_read_xattr_bucket(args->old_bucket, blkno);
6926 	if (ret) {
6927 		mlog_errno(ret);
6928 		goto out;
6929 	}
6930 	num_buckets = le16_to_cpu(bucket_xh(args->old_bucket)->xh_num_buckets);
6931 	ocfs2_xattr_bucket_relse(args->old_bucket);
6932 
6933 	while (len && num_buckets) {
6934 		ret = ocfs2_claim_clusters(handle, data_ac,
6935 					   1, &p_cluster, &num_clusters);
6936 		if (ret) {
6937 			mlog_errno(ret);
6938 			goto out;
6939 		}
6940 
6941 		new_blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
6942 		reflink_buckets = min(num_buckets, bpc * num_clusters);
6943 
6944 		ret = ocfs2_reflink_xattr_bucket(handle, blkno,
6945 						 new_blkno, num_clusters,
6946 						 &reflink_cpos, reflink_buckets,
6947 						 meta_ac, data_ac, args);
6948 		if (ret) {
6949 			mlog_errno(ret);
6950 			goto out;
6951 		}
6952 
6953 		/*
6954 		 * For the 1st allocated cluster, we make it use the same cpos
6955 		 * so that the xattr tree looks the same as the original one
6956 		 * in the most case.
6957 		 */
6958 		if (!first_inserted) {
6959 			reflink_cpos = cpos;
6960 			first_inserted = 1;
6961 		}
6962 		ret = ocfs2_insert_extent(handle, et, reflink_cpos, new_blkno,
6963 					  num_clusters, 0, meta_ac);
6964 		if (ret)
6965 			mlog_errno(ret);
6966 
6967 		trace_ocfs2_reflink_xattr_buckets((unsigned long long)new_blkno,
6968 						  num_clusters, reflink_cpos);
6969 
6970 		len -= num_clusters;
6971 		blkno += ocfs2_clusters_to_blocks(inode->i_sb, num_clusters);
6972 		num_buckets -= reflink_buckets;
6973 	}
6974 out:
6975 	return ret;
6976 }
6977 
6978 /*
6979  * Create the same xattr extent record in the new inode's xattr tree.
6980  */
6981 static int ocfs2_reflink_xattr_rec(struct inode *inode,
6982 				   struct buffer_head *root_bh,
6983 				   u64 blkno,
6984 				   u32 cpos,
6985 				   u32 len,
6986 				   void *para)
6987 {
6988 	int ret, credits = 0;
6989 	handle_t *handle;
6990 	struct ocfs2_reflink_xattr_tree_args *args =
6991 			(struct ocfs2_reflink_xattr_tree_args *)para;
6992 	struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
6993 	struct ocfs2_alloc_context *meta_ac = NULL;
6994 	struct ocfs2_alloc_context *data_ac = NULL;
6995 	struct ocfs2_extent_tree et;
6996 
6997 	trace_ocfs2_reflink_xattr_rec((unsigned long long)blkno, len);
6998 
6999 	ocfs2_init_xattr_tree_extent_tree(&et,
7000 					  INODE_CACHE(args->reflink->new_inode),
7001 					  args->new_blk_bh);
7002 
7003 	ret = ocfs2_lock_reflink_xattr_rec_allocators(args, &et, blkno,
7004 						      len, &credits,
7005 						      &meta_ac, &data_ac);
7006 	if (ret) {
7007 		mlog_errno(ret);
7008 		goto out;
7009 	}
7010 
7011 	handle = ocfs2_start_trans(osb, credits);
7012 	if (IS_ERR(handle)) {
7013 		ret = PTR_ERR(handle);
7014 		mlog_errno(ret);
7015 		goto out;
7016 	}
7017 
7018 	ret = ocfs2_reflink_xattr_buckets(handle, inode, args, &et,
7019 					  meta_ac, data_ac,
7020 					  blkno, cpos, len);
7021 	if (ret)
7022 		mlog_errno(ret);
7023 
7024 	ocfs2_commit_trans(osb, handle);
7025 
7026 out:
7027 	if (meta_ac)
7028 		ocfs2_free_alloc_context(meta_ac);
7029 	if (data_ac)
7030 		ocfs2_free_alloc_context(data_ac);
7031 	return ret;
7032 }
7033 
7034 /*
7035  * Create reflinked xattr buckets.
7036  * We will add bucket one by one, and refcount all the xattrs in the bucket
7037  * if they are stored outside.
7038  */
7039 static int ocfs2_reflink_xattr_tree(struct ocfs2_xattr_reflink *args,
7040 				    struct buffer_head *blk_bh,
7041 				    struct buffer_head *new_blk_bh)
7042 {
7043 	int ret;
7044 	struct ocfs2_reflink_xattr_tree_args para;
7045 
7046 	memset(&para, 0, sizeof(para));
7047 	para.reflink = args;
7048 	para.old_blk_bh = blk_bh;
7049 	para.new_blk_bh = new_blk_bh;
7050 
7051 	para.old_bucket = ocfs2_xattr_bucket_new(args->old_inode);
7052 	if (!para.old_bucket) {
7053 		mlog_errno(-ENOMEM);
7054 		return -ENOMEM;
7055 	}
7056 
7057 	para.new_bucket = ocfs2_xattr_bucket_new(args->new_inode);
7058 	if (!para.new_bucket) {
7059 		ret = -ENOMEM;
7060 		mlog_errno(ret);
7061 		goto out;
7062 	}
7063 
7064 	ret = ocfs2_iterate_xattr_index_block(args->old_inode, blk_bh,
7065 					      ocfs2_reflink_xattr_rec,
7066 					      &para);
7067 	if (ret)
7068 		mlog_errno(ret);
7069 
7070 out:
7071 	ocfs2_xattr_bucket_free(para.old_bucket);
7072 	ocfs2_xattr_bucket_free(para.new_bucket);
7073 	return ret;
7074 }
7075 
7076 static int ocfs2_reflink_xattr_in_block(struct ocfs2_xattr_reflink *args,
7077 					struct buffer_head *blk_bh)
7078 {
7079 	int ret, indexed = 0;
7080 	struct buffer_head *new_blk_bh = NULL;
7081 	struct ocfs2_xattr_block *xb =
7082 			(struct ocfs2_xattr_block *)blk_bh->b_data;
7083 
7084 
7085 	if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)
7086 		indexed = 1;
7087 
7088 	ret = ocfs2_create_empty_xattr_block(args->new_inode, args->new_bh,
7089 					     &new_blk_bh, indexed);
7090 	if (ret) {
7091 		mlog_errno(ret);
7092 		goto out;
7093 	}
7094 
7095 	if (!indexed)
7096 		ret = ocfs2_reflink_xattr_block(args, blk_bh, new_blk_bh);
7097 	else
7098 		ret = ocfs2_reflink_xattr_tree(args, blk_bh, new_blk_bh);
7099 	if (ret)
7100 		mlog_errno(ret);
7101 
7102 out:
7103 	brelse(new_blk_bh);
7104 	return ret;
7105 }
7106 
7107 static int ocfs2_reflink_xattr_no_security(struct ocfs2_xattr_entry *xe)
7108 {
7109 	int type = ocfs2_xattr_get_type(xe);
7110 
7111 	return type != OCFS2_XATTR_INDEX_SECURITY &&
7112 	       type != OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS &&
7113 	       type != OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT;
7114 }
7115 
7116 int ocfs2_reflink_xattrs(struct inode *old_inode,
7117 			 struct buffer_head *old_bh,
7118 			 struct inode *new_inode,
7119 			 struct buffer_head *new_bh,
7120 			 bool preserve_security)
7121 {
7122 	int ret;
7123 	struct ocfs2_xattr_reflink args;
7124 	struct ocfs2_inode_info *oi = OCFS2_I(old_inode);
7125 	struct ocfs2_dinode *di = (struct ocfs2_dinode *)old_bh->b_data;
7126 	struct buffer_head *blk_bh = NULL;
7127 	struct ocfs2_cached_dealloc_ctxt dealloc;
7128 	struct ocfs2_refcount_tree *ref_tree;
7129 	struct buffer_head *ref_root_bh = NULL;
7130 
7131 	ret = ocfs2_lock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7132 				       le64_to_cpu(di->i_refcount_loc),
7133 				       1, &ref_tree, &ref_root_bh);
7134 	if (ret) {
7135 		mlog_errno(ret);
7136 		goto out;
7137 	}
7138 
7139 	ocfs2_init_dealloc_ctxt(&dealloc);
7140 
7141 	args.old_inode = old_inode;
7142 	args.new_inode = new_inode;
7143 	args.old_bh = old_bh;
7144 	args.new_bh = new_bh;
7145 	args.ref_ci = &ref_tree->rf_ci;
7146 	args.ref_root_bh = ref_root_bh;
7147 	args.dealloc = &dealloc;
7148 	if (preserve_security)
7149 		args.xattr_reflinked = NULL;
7150 	else
7151 		args.xattr_reflinked = ocfs2_reflink_xattr_no_security;
7152 
7153 	if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
7154 		ret = ocfs2_reflink_xattr_inline(&args);
7155 		if (ret) {
7156 			mlog_errno(ret);
7157 			goto out_unlock;
7158 		}
7159 	}
7160 
7161 	if (!di->i_xattr_loc)
7162 		goto out_unlock;
7163 
7164 	ret = ocfs2_read_xattr_block(old_inode, le64_to_cpu(di->i_xattr_loc),
7165 				     &blk_bh);
7166 	if (ret < 0) {
7167 		mlog_errno(ret);
7168 		goto out_unlock;
7169 	}
7170 
7171 	ret = ocfs2_reflink_xattr_in_block(&args, blk_bh);
7172 	if (ret)
7173 		mlog_errno(ret);
7174 
7175 	brelse(blk_bh);
7176 
7177 out_unlock:
7178 	ocfs2_unlock_refcount_tree(OCFS2_SB(old_inode->i_sb),
7179 				   ref_tree, 1);
7180 	brelse(ref_root_bh);
7181 
7182 	if (ocfs2_dealloc_has_cluster(&dealloc)) {
7183 		ocfs2_schedule_truncate_log_flush(OCFS2_SB(old_inode->i_sb), 1);
7184 		ocfs2_run_deallocs(OCFS2_SB(old_inode->i_sb), &dealloc);
7185 	}
7186 
7187 out:
7188 	return ret;
7189 }
7190 
7191 /*
7192  * Initialize security and acl for a already created inode.
7193  * Used for reflink a non-preserve-security file.
7194  *
7195  * It uses common api like ocfs2_xattr_set, so the caller
7196  * must not hold any lock expect i_mutex.
7197  */
7198 int ocfs2_init_security_and_acl(struct inode *dir,
7199 				struct inode *inode,
7200 				const struct qstr *qstr,
7201 				struct posix_acl *default_acl,
7202 				struct posix_acl *acl)
7203 {
7204 	struct buffer_head *dir_bh = NULL;
7205 	int ret = 0;
7206 
7207 	ret = ocfs2_init_security_get(inode, dir, qstr, NULL);
7208 	if (ret) {
7209 		mlog_errno(ret);
7210 		goto leave;
7211 	}
7212 
7213 	ret = ocfs2_inode_lock(dir, &dir_bh, 0);
7214 	if (ret) {
7215 		mlog_errno(ret);
7216 		goto leave;
7217 	}
7218 
7219 	if (!ret && default_acl)
7220 		ret = ocfs2_iop_set_acl(inode, default_acl, ACL_TYPE_DEFAULT);
7221 	if (!ret && acl)
7222 		ret = ocfs2_iop_set_acl(inode, acl, ACL_TYPE_ACCESS);
7223 
7224 	ocfs2_inode_unlock(dir, 0);
7225 	brelse(dir_bh);
7226 leave:
7227 	return ret;
7228 }
7229 /*
7230  * 'security' attributes support
7231  */
7232 static size_t ocfs2_xattr_security_list(struct dentry *dentry, char *list,
7233 					size_t list_size, const char *name,
7234 					size_t name_len, int type)
7235 {
7236 	const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
7237 	const size_t total_len = prefix_len + name_len + 1;
7238 
7239 	if (list && total_len <= list_size) {
7240 		memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
7241 		memcpy(list + prefix_len, name, name_len);
7242 		list[prefix_len + name_len] = '\0';
7243 	}
7244 	return total_len;
7245 }
7246 
7247 static int ocfs2_xattr_security_get(struct dentry *dentry, const char *name,
7248 				    void *buffer, size_t size, int type)
7249 {
7250 	if (strcmp(name, "") == 0)
7251 		return -EINVAL;
7252 	return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7253 			       name, buffer, size);
7254 }
7255 
7256 static int ocfs2_xattr_security_set(struct dentry *dentry, const char *name,
7257 		const void *value, size_t size, int flags, int type)
7258 {
7259 	if (strcmp(name, "") == 0)
7260 		return -EINVAL;
7261 
7262 	return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_SECURITY,
7263 			       name, value, size, flags);
7264 }
7265 
7266 int ocfs2_initxattrs(struct inode *inode, const struct xattr *xattr_array,
7267 		     void *fs_info)
7268 {
7269 	const struct xattr *xattr;
7270 	int err = 0;
7271 
7272 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
7273 		err = ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY,
7274 				      xattr->name, xattr->value,
7275 				      xattr->value_len, XATTR_CREATE);
7276 		if (err)
7277 			break;
7278 	}
7279 	return err;
7280 }
7281 
7282 int ocfs2_init_security_get(struct inode *inode,
7283 			    struct inode *dir,
7284 			    const struct qstr *qstr,
7285 			    struct ocfs2_security_xattr_info *si)
7286 {
7287 	/* check whether ocfs2 support feature xattr */
7288 	if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
7289 		return -EOPNOTSUPP;
7290 	if (si)
7291 		return security_old_inode_init_security(inode, dir, qstr,
7292 							&si->name, &si->value,
7293 							&si->value_len);
7294 
7295 	return security_inode_init_security(inode, dir, qstr,
7296 					    &ocfs2_initxattrs, NULL);
7297 }
7298 
7299 int ocfs2_init_security_set(handle_t *handle,
7300 			    struct inode *inode,
7301 			    struct buffer_head *di_bh,
7302 			    struct ocfs2_security_xattr_info *si,
7303 			    struct ocfs2_alloc_context *xattr_ac,
7304 			    struct ocfs2_alloc_context *data_ac)
7305 {
7306 	return ocfs2_xattr_set_handle(handle, inode, di_bh,
7307 				     OCFS2_XATTR_INDEX_SECURITY,
7308 				     si->name, si->value, si->value_len, 0,
7309 				     xattr_ac, data_ac);
7310 }
7311 
7312 const struct xattr_handler ocfs2_xattr_security_handler = {
7313 	.prefix	= XATTR_SECURITY_PREFIX,
7314 	.list	= ocfs2_xattr_security_list,
7315 	.get	= ocfs2_xattr_security_get,
7316 	.set	= ocfs2_xattr_security_set,
7317 };
7318 
7319 /*
7320  * 'trusted' attributes support
7321  */
7322 static size_t ocfs2_xattr_trusted_list(struct dentry *dentry, char *list,
7323 				       size_t list_size, const char *name,
7324 				       size_t name_len, int type)
7325 {
7326 	const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
7327 	const size_t total_len = prefix_len + name_len + 1;
7328 
7329 	if (list && total_len <= list_size) {
7330 		memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
7331 		memcpy(list + prefix_len, name, name_len);
7332 		list[prefix_len + name_len] = '\0';
7333 	}
7334 	return total_len;
7335 }
7336 
7337 static int ocfs2_xattr_trusted_get(struct dentry *dentry, const char *name,
7338 		void *buffer, size_t size, int type)
7339 {
7340 	if (strcmp(name, "") == 0)
7341 		return -EINVAL;
7342 	return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7343 			       name, buffer, size);
7344 }
7345 
7346 static int ocfs2_xattr_trusted_set(struct dentry *dentry, const char *name,
7347 		const void *value, size_t size, int flags, int type)
7348 {
7349 	if (strcmp(name, "") == 0)
7350 		return -EINVAL;
7351 
7352 	return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_TRUSTED,
7353 			       name, value, size, flags);
7354 }
7355 
7356 const struct xattr_handler ocfs2_xattr_trusted_handler = {
7357 	.prefix	= XATTR_TRUSTED_PREFIX,
7358 	.list	= ocfs2_xattr_trusted_list,
7359 	.get	= ocfs2_xattr_trusted_get,
7360 	.set	= ocfs2_xattr_trusted_set,
7361 };
7362 
7363 /*
7364  * 'user' attributes support
7365  */
7366 static size_t ocfs2_xattr_user_list(struct dentry *dentry, char *list,
7367 				    size_t list_size, const char *name,
7368 				    size_t name_len, int type)
7369 {
7370 	const size_t prefix_len = XATTR_USER_PREFIX_LEN;
7371 	const size_t total_len = prefix_len + name_len + 1;
7372 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7373 
7374 	if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7375 		return 0;
7376 
7377 	if (list && total_len <= list_size) {
7378 		memcpy(list, XATTR_USER_PREFIX, prefix_len);
7379 		memcpy(list + prefix_len, name, name_len);
7380 		list[prefix_len + name_len] = '\0';
7381 	}
7382 	return total_len;
7383 }
7384 
7385 static int ocfs2_xattr_user_get(struct dentry *dentry, const char *name,
7386 		void *buffer, size_t size, int type)
7387 {
7388 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7389 
7390 	if (strcmp(name, "") == 0)
7391 		return -EINVAL;
7392 	if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7393 		return -EOPNOTSUPP;
7394 	return ocfs2_xattr_get(dentry->d_inode, OCFS2_XATTR_INDEX_USER, name,
7395 			       buffer, size);
7396 }
7397 
7398 static int ocfs2_xattr_user_set(struct dentry *dentry, const char *name,
7399 		const void *value, size_t size, int flags, int type)
7400 {
7401 	struct ocfs2_super *osb = OCFS2_SB(dentry->d_sb);
7402 
7403 	if (strcmp(name, "") == 0)
7404 		return -EINVAL;
7405 	if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
7406 		return -EOPNOTSUPP;
7407 
7408 	return ocfs2_xattr_set(dentry->d_inode, OCFS2_XATTR_INDEX_USER,
7409 			       name, value, size, flags);
7410 }
7411 
7412 const struct xattr_handler ocfs2_xattr_user_handler = {
7413 	.prefix	= XATTR_USER_PREFIX,
7414 	.list	= ocfs2_xattr_user_list,
7415 	.get	= ocfs2_xattr_user_get,
7416 	.set	= ocfs2_xattr_user_set,
7417 };
7418