ocfs2/xattr: Remove additional bucket allocation in bucket defragment.
[linux-2.6.git] / fs / ocfs2 / xattr.c
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
39 #define MLOG_MASK_PREFIX ML_XATTR
40 #include <cluster/masklog.h>
41
42 #include "ocfs2.h"
43 #include "alloc.h"
44 #include "dlmglue.h"
45 #include "file.h"
46 #include "symlink.h"
47 #include "sysfile.h"
48 #include "inode.h"
49 #include "journal.h"
50 #include "ocfs2_fs.h"
51 #include "suballoc.h"
52 #include "uptodate.h"
53 #include "buffer_head_io.h"
54 #include "super.h"
55 #include "xattr.h"
56
57
58 struct ocfs2_xattr_def_value_root {
59         struct ocfs2_xattr_value_root   xv;
60         struct ocfs2_extent_rec         er;
61 };
62
63 struct ocfs2_xattr_bucket {
64         /* The inode these xattrs are associated with */
65         struct inode *bu_inode;
66
67         /* The actual buffers that make up the bucket */
68         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
69
70         /* How many blocks make up one bucket for this filesystem */
71         int bu_blocks;
72 };
73
74 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
75 #define OCFS2_XATTR_INLINE_SIZE 80
76
77 static struct ocfs2_xattr_def_value_root def_xv = {
78         .xv.xr_list.l_count = cpu_to_le16(1),
79 };
80
81 struct xattr_handler *ocfs2_xattr_handlers[] = {
82         &ocfs2_xattr_user_handler,
83         &ocfs2_xattr_trusted_handler,
84         NULL
85 };
86
87 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
88         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
89         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
90 };
91
92 struct ocfs2_xattr_info {
93         int name_index;
94         const char *name;
95         const void *value;
96         size_t value_len;
97 };
98
99 struct ocfs2_xattr_search {
100         struct buffer_head *inode_bh;
101         /*
102          * xattr_bh point to the block buffer head which has extended attribute
103          * when extended attribute in inode, xattr_bh is equal to inode_bh.
104          */
105         struct buffer_head *xattr_bh;
106         struct ocfs2_xattr_header *header;
107         struct ocfs2_xattr_bucket *bucket;
108         void *base;
109         void *end;
110         struct ocfs2_xattr_entry *here;
111         int not_found;
112 };
113
114 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
115                                              struct ocfs2_xattr_header *xh,
116                                              int index,
117                                              int *block_off,
118                                              int *new_offset);
119
120 static int ocfs2_xattr_block_find(struct inode *inode,
121                                   int name_index,
122                                   const char *name,
123                                   struct ocfs2_xattr_search *xs);
124 static int ocfs2_xattr_index_block_find(struct inode *inode,
125                                         struct buffer_head *root_bh,
126                                         int name_index,
127                                         const char *name,
128                                         struct ocfs2_xattr_search *xs);
129
130 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
131                                         struct ocfs2_xattr_tree_root *xt,
132                                         char *buffer,
133                                         size_t buffer_size);
134
135 static int ocfs2_xattr_create_index_block(struct inode *inode,
136                                           struct ocfs2_xattr_search *xs);
137
138 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
139                                              struct ocfs2_xattr_info *xi,
140                                              struct ocfs2_xattr_search *xs);
141
142 static int ocfs2_delete_xattr_index_block(struct inode *inode,
143                                           struct buffer_head *xb_bh);
144
145 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
146 {
147         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
148 }
149
150 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
151 {
152         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
153 }
154
155 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
156 {
157         u16 len = sb->s_blocksize -
158                  offsetof(struct ocfs2_xattr_header, xh_entries);
159
160         return len / sizeof(struct ocfs2_xattr_entry);
161 }
162
163 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
164 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
165 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
166
167 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
168 {
169         struct ocfs2_xattr_bucket *bucket;
170         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
171
172         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
173
174         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
175         if (bucket) {
176                 bucket->bu_inode = inode;
177                 bucket->bu_blocks = blks;
178         }
179
180         return bucket;
181 }
182
183 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
184 {
185         int i;
186
187         for (i = 0; i < bucket->bu_blocks; i++) {
188                 brelse(bucket->bu_bhs[i]);
189                 bucket->bu_bhs[i] = NULL;
190         }
191 }
192
193 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
194 {
195         if (bucket) {
196                 ocfs2_xattr_bucket_relse(bucket);
197                 bucket->bu_inode = NULL;
198                 kfree(bucket);
199         }
200 }
201
202 /*
203  * A bucket that has never been written to disk doesn't need to be
204  * read.  We just need the buffer_heads.  Don't call this for
205  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
206  * them fully.
207  */
208 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
209                                    u64 xb_blkno)
210 {
211         int i, rc = 0;
212
213         for (i = 0; i < bucket->bu_blocks; i++) {
214                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
215                                               xb_blkno + i);
216                 if (!bucket->bu_bhs[i]) {
217                         rc = -EIO;
218                         mlog_errno(rc);
219                         break;
220                 }
221
222                 ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
223                                               bucket->bu_bhs[i]);
224         }
225
226         if (rc)
227                 ocfs2_xattr_bucket_relse(bucket);
228         return rc;
229 }
230
231 /* Read the xattr bucket at xb_blkno */
232 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
233                                    u64 xb_blkno)
234 {
235         int rc;
236
237         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
238                                bucket->bu_blocks, bucket->bu_bhs, 0);
239         if (rc)
240                 ocfs2_xattr_bucket_relse(bucket);
241         return rc;
242 }
243
244 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
245                                              struct ocfs2_xattr_bucket *bucket,
246                                              int type)
247 {
248         int i, rc = 0;
249
250         for (i = 0; i < bucket->bu_blocks; i++) {
251                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
252                                           bucket->bu_bhs[i], type);
253                 if (rc) {
254                         mlog_errno(rc);
255                         break;
256                 }
257         }
258
259         return rc;
260 }
261
262 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
263                                              struct ocfs2_xattr_bucket *bucket)
264 {
265         int i;
266
267         for (i = 0; i < bucket->bu_blocks; i++)
268                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
269 }
270
271 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
272                                          struct ocfs2_xattr_bucket *src)
273 {
274         int i;
275         int blocksize = src->bu_inode->i_sb->s_blocksize;
276
277         BUG_ON(dest->bu_blocks != src->bu_blocks);
278         BUG_ON(dest->bu_inode != src->bu_inode);
279
280         for (i = 0; i < src->bu_blocks; i++) {
281                 memcpy(bucket_block(dest, i), bucket_block(src, i),
282                        blocksize);
283         }
284 }
285
286 static inline const char *ocfs2_xattr_prefix(int name_index)
287 {
288         struct xattr_handler *handler = NULL;
289
290         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
291                 handler = ocfs2_xattr_handler_map[name_index];
292
293         return handler ? handler->prefix : NULL;
294 }
295
296 static u32 ocfs2_xattr_name_hash(struct inode *inode,
297                                  const char *name,
298                                  int name_len)
299 {
300         /* Get hash value of uuid from super block */
301         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
302         int i;
303
304         /* hash extended attribute name */
305         for (i = 0; i < name_len; i++) {
306                 hash = (hash << OCFS2_HASH_SHIFT) ^
307                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
308                        *name++;
309         }
310
311         return hash;
312 }
313
314 /*
315  * ocfs2_xattr_hash_entry()
316  *
317  * Compute the hash of an extended attribute.
318  */
319 static void ocfs2_xattr_hash_entry(struct inode *inode,
320                                    struct ocfs2_xattr_header *header,
321                                    struct ocfs2_xattr_entry *entry)
322 {
323         u32 hash = 0;
324         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
325
326         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
327         entry->xe_name_hash = cpu_to_le32(hash);
328
329         return;
330 }
331
332 static int ocfs2_xattr_extend_allocation(struct inode *inode,
333                                          u32 clusters_to_add,
334                                          struct buffer_head *xattr_bh,
335                                          struct ocfs2_xattr_value_root *xv)
336 {
337         int status = 0;
338         int restart_func = 0;
339         int credits = 0;
340         handle_t *handle = NULL;
341         struct ocfs2_alloc_context *data_ac = NULL;
342         struct ocfs2_alloc_context *meta_ac = NULL;
343         enum ocfs2_alloc_restarted why;
344         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
345         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
346         struct ocfs2_extent_tree et;
347
348         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
349
350         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
351
352 restart_all:
353
354         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
355                                        &data_ac, &meta_ac);
356         if (status) {
357                 mlog_errno(status);
358                 goto leave;
359         }
360
361         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
362                                             clusters_to_add);
363         handle = ocfs2_start_trans(osb, credits);
364         if (IS_ERR(handle)) {
365                 status = PTR_ERR(handle);
366                 handle = NULL;
367                 mlog_errno(status);
368                 goto leave;
369         }
370
371 restarted_transaction:
372         status = ocfs2_journal_access(handle, inode, xattr_bh,
373                                       OCFS2_JOURNAL_ACCESS_WRITE);
374         if (status < 0) {
375                 mlog_errno(status);
376                 goto leave;
377         }
378
379         prev_clusters = le32_to_cpu(xv->xr_clusters);
380         status = ocfs2_add_clusters_in_btree(osb,
381                                              inode,
382                                              &logical_start,
383                                              clusters_to_add,
384                                              0,
385                                              &et,
386                                              handle,
387                                              data_ac,
388                                              meta_ac,
389                                              &why);
390         if ((status < 0) && (status != -EAGAIN)) {
391                 if (status != -ENOSPC)
392                         mlog_errno(status);
393                 goto leave;
394         }
395
396         status = ocfs2_journal_dirty(handle, xattr_bh);
397         if (status < 0) {
398                 mlog_errno(status);
399                 goto leave;
400         }
401
402         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
403
404         if (why != RESTART_NONE && clusters_to_add) {
405                 if (why == RESTART_META) {
406                         mlog(0, "restarting function.\n");
407                         restart_func = 1;
408                 } else {
409                         BUG_ON(why != RESTART_TRANS);
410
411                         mlog(0, "restarting transaction.\n");
412                         /* TODO: This can be more intelligent. */
413                         credits = ocfs2_calc_extend_credits(osb->sb,
414                                                             et.et_root_el,
415                                                             clusters_to_add);
416                         status = ocfs2_extend_trans(handle, credits);
417                         if (status < 0) {
418                                 /* handle still has to be committed at
419                                  * this point. */
420                                 status = -ENOMEM;
421                                 mlog_errno(status);
422                                 goto leave;
423                         }
424                         goto restarted_transaction;
425                 }
426         }
427
428 leave:
429         if (handle) {
430                 ocfs2_commit_trans(osb, handle);
431                 handle = NULL;
432         }
433         if (data_ac) {
434                 ocfs2_free_alloc_context(data_ac);
435                 data_ac = NULL;
436         }
437         if (meta_ac) {
438                 ocfs2_free_alloc_context(meta_ac);
439                 meta_ac = NULL;
440         }
441         if ((!status) && restart_func) {
442                 restart_func = 0;
443                 goto restart_all;
444         }
445
446         return status;
447 }
448
449 static int __ocfs2_remove_xattr_range(struct inode *inode,
450                                       struct buffer_head *root_bh,
451                                       struct ocfs2_xattr_value_root *xv,
452                                       u32 cpos, u32 phys_cpos, u32 len,
453                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
454 {
455         int ret;
456         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
457         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
458         struct inode *tl_inode = osb->osb_tl_inode;
459         handle_t *handle;
460         struct ocfs2_alloc_context *meta_ac = NULL;
461         struct ocfs2_extent_tree et;
462
463         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
464
465         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
466         if (ret) {
467                 mlog_errno(ret);
468                 return ret;
469         }
470
471         mutex_lock(&tl_inode->i_mutex);
472
473         if (ocfs2_truncate_log_needs_flush(osb)) {
474                 ret = __ocfs2_flush_truncate_log(osb);
475                 if (ret < 0) {
476                         mlog_errno(ret);
477                         goto out;
478                 }
479         }
480
481         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
482         if (IS_ERR(handle)) {
483                 ret = PTR_ERR(handle);
484                 mlog_errno(ret);
485                 goto out;
486         }
487
488         ret = ocfs2_journal_access(handle, inode, root_bh,
489                                    OCFS2_JOURNAL_ACCESS_WRITE);
490         if (ret) {
491                 mlog_errno(ret);
492                 goto out_commit;
493         }
494
495         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
496                                   dealloc);
497         if (ret) {
498                 mlog_errno(ret);
499                 goto out_commit;
500         }
501
502         le32_add_cpu(&xv->xr_clusters, -len);
503
504         ret = ocfs2_journal_dirty(handle, root_bh);
505         if (ret) {
506                 mlog_errno(ret);
507                 goto out_commit;
508         }
509
510         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
511         if (ret)
512                 mlog_errno(ret);
513
514 out_commit:
515         ocfs2_commit_trans(osb, handle);
516 out:
517         mutex_unlock(&tl_inode->i_mutex);
518
519         if (meta_ac)
520                 ocfs2_free_alloc_context(meta_ac);
521
522         return ret;
523 }
524
525 static int ocfs2_xattr_shrink_size(struct inode *inode,
526                                    u32 old_clusters,
527                                    u32 new_clusters,
528                                    struct buffer_head *root_bh,
529                                    struct ocfs2_xattr_value_root *xv)
530 {
531         int ret = 0;
532         u32 trunc_len, cpos, phys_cpos, alloc_size;
533         u64 block;
534         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
535         struct ocfs2_cached_dealloc_ctxt dealloc;
536
537         ocfs2_init_dealloc_ctxt(&dealloc);
538
539         if (old_clusters <= new_clusters)
540                 return 0;
541
542         cpos = new_clusters;
543         trunc_len = old_clusters - new_clusters;
544         while (trunc_len) {
545                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
546                                                &alloc_size, &xv->xr_list);
547                 if (ret) {
548                         mlog_errno(ret);
549                         goto out;
550                 }
551
552                 if (alloc_size > trunc_len)
553                         alloc_size = trunc_len;
554
555                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
556                                                  phys_cpos, alloc_size,
557                                                  &dealloc);
558                 if (ret) {
559                         mlog_errno(ret);
560                         goto out;
561                 }
562
563                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
564                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
565                                                        alloc_size);
566                 cpos += alloc_size;
567                 trunc_len -= alloc_size;
568         }
569
570 out:
571         ocfs2_schedule_truncate_log_flush(osb, 1);
572         ocfs2_run_deallocs(osb, &dealloc);
573
574         return ret;
575 }
576
577 static int ocfs2_xattr_value_truncate(struct inode *inode,
578                                       struct buffer_head *root_bh,
579                                       struct ocfs2_xattr_value_root *xv,
580                                       int len)
581 {
582         int ret;
583         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
584         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
585
586         if (new_clusters == old_clusters)
587                 return 0;
588
589         if (new_clusters > old_clusters)
590                 ret = ocfs2_xattr_extend_allocation(inode,
591                                                     new_clusters - old_clusters,
592                                                     root_bh, xv);
593         else
594                 ret = ocfs2_xattr_shrink_size(inode,
595                                               old_clusters, new_clusters,
596                                               root_bh, xv);
597
598         return ret;
599 }
600
601 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
602                                   size_t *result, const char *prefix,
603                                   const char *name, int name_len)
604 {
605         char *p = buffer + *result;
606         int prefix_len = strlen(prefix);
607         int total_len = prefix_len + name_len + 1;
608
609         *result += total_len;
610
611         /* we are just looking for how big our buffer needs to be */
612         if (!size)
613                 return 0;
614
615         if (*result > size)
616                 return -ERANGE;
617
618         memcpy(p, prefix, prefix_len);
619         memcpy(p + prefix_len, name, name_len);
620         p[prefix_len + name_len] = '\0';
621
622         return 0;
623 }
624
625 static int ocfs2_xattr_list_entries(struct inode *inode,
626                                     struct ocfs2_xattr_header *header,
627                                     char *buffer, size_t buffer_size)
628 {
629         size_t result = 0;
630         int i, type, ret;
631         const char *prefix, *name;
632
633         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
634                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
635                 type = ocfs2_xattr_get_type(entry);
636                 prefix = ocfs2_xattr_prefix(type);
637
638                 if (prefix) {
639                         name = (const char *)header +
640                                 le16_to_cpu(entry->xe_name_offset);
641
642                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
643                                                      &result, prefix, name,
644                                                      entry->xe_name_len);
645                         if (ret)
646                                 return ret;
647                 }
648         }
649
650         return result;
651 }
652
653 static int ocfs2_xattr_ibody_list(struct inode *inode,
654                                   struct ocfs2_dinode *di,
655                                   char *buffer,
656                                   size_t buffer_size)
657 {
658         struct ocfs2_xattr_header *header = NULL;
659         struct ocfs2_inode_info *oi = OCFS2_I(inode);
660         int ret = 0;
661
662         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
663                 return ret;
664
665         header = (struct ocfs2_xattr_header *)
666                  ((void *)di + inode->i_sb->s_blocksize -
667                  le16_to_cpu(di->i_xattr_inline_size));
668
669         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
670
671         return ret;
672 }
673
674 static int ocfs2_xattr_block_list(struct inode *inode,
675                                   struct ocfs2_dinode *di,
676                                   char *buffer,
677                                   size_t buffer_size)
678 {
679         struct buffer_head *blk_bh = NULL;
680         struct ocfs2_xattr_block *xb;
681         int ret = 0;
682
683         if (!di->i_xattr_loc)
684                 return ret;
685
686         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
687         if (ret < 0) {
688                 mlog_errno(ret);
689                 return ret;
690         }
691
692         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
693         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
694                 ret = -EIO;
695                 goto cleanup;
696         }
697
698         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
699                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
700                 ret = ocfs2_xattr_list_entries(inode, header,
701                                                buffer, buffer_size);
702         } else {
703                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
704                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
705                                                    buffer, buffer_size);
706         }
707 cleanup:
708         brelse(blk_bh);
709
710         return ret;
711 }
712
713 ssize_t ocfs2_listxattr(struct dentry *dentry,
714                         char *buffer,
715                         size_t size)
716 {
717         int ret = 0, i_ret = 0, b_ret = 0;
718         struct buffer_head *di_bh = NULL;
719         struct ocfs2_dinode *di = NULL;
720         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
721
722         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
723                 return -EOPNOTSUPP;
724
725         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
726                 return ret;
727
728         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
729         if (ret < 0) {
730                 mlog_errno(ret);
731                 return ret;
732         }
733
734         di = (struct ocfs2_dinode *)di_bh->b_data;
735
736         down_read(&oi->ip_xattr_sem);
737         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
738         if (i_ret < 0)
739                 b_ret = 0;
740         else {
741                 if (buffer) {
742                         buffer += i_ret;
743                         size -= i_ret;
744                 }
745                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
746                                                buffer, size);
747                 if (b_ret < 0)
748                         i_ret = 0;
749         }
750         up_read(&oi->ip_xattr_sem);
751         ocfs2_inode_unlock(dentry->d_inode, 0);
752
753         brelse(di_bh);
754
755         return i_ret + b_ret;
756 }
757
758 static int ocfs2_xattr_find_entry(int name_index,
759                                   const char *name,
760                                   struct ocfs2_xattr_search *xs)
761 {
762         struct ocfs2_xattr_entry *entry;
763         size_t name_len;
764         int i, cmp = 1;
765
766         if (name == NULL)
767                 return -EINVAL;
768
769         name_len = strlen(name);
770         entry = xs->here;
771         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
772                 cmp = name_index - ocfs2_xattr_get_type(entry);
773                 if (!cmp)
774                         cmp = name_len - entry->xe_name_len;
775                 if (!cmp)
776                         cmp = memcmp(name, (xs->base +
777                                      le16_to_cpu(entry->xe_name_offset)),
778                                      name_len);
779                 if (cmp == 0)
780                         break;
781                 entry += 1;
782         }
783         xs->here = entry;
784
785         return cmp ? -ENODATA : 0;
786 }
787
788 static int ocfs2_xattr_get_value_outside(struct inode *inode,
789                                          struct ocfs2_xattr_value_root *xv,
790                                          void *buffer,
791                                          size_t len)
792 {
793         u32 cpos, p_cluster, num_clusters, bpc, clusters;
794         u64 blkno;
795         int i, ret = 0;
796         size_t cplen, blocksize;
797         struct buffer_head *bh = NULL;
798         struct ocfs2_extent_list *el;
799
800         el = &xv->xr_list;
801         clusters = le32_to_cpu(xv->xr_clusters);
802         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
803         blocksize = inode->i_sb->s_blocksize;
804
805         cpos = 0;
806         while (cpos < clusters) {
807                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
808                                                &num_clusters, el);
809                 if (ret) {
810                         mlog_errno(ret);
811                         goto out;
812                 }
813
814                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
815                 /* Copy ocfs2_xattr_value */
816                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
817                         ret = ocfs2_read_block(inode, blkno, &bh);
818                         if (ret) {
819                                 mlog_errno(ret);
820                                 goto out;
821                         }
822
823                         cplen = len >= blocksize ? blocksize : len;
824                         memcpy(buffer, bh->b_data, cplen);
825                         len -= cplen;
826                         buffer += cplen;
827
828                         brelse(bh);
829                         bh = NULL;
830                         if (len == 0)
831                                 break;
832                 }
833                 cpos += num_clusters;
834         }
835 out:
836         return ret;
837 }
838
839 static int ocfs2_xattr_ibody_get(struct inode *inode,
840                                  int name_index,
841                                  const char *name,
842                                  void *buffer,
843                                  size_t buffer_size,
844                                  struct ocfs2_xattr_search *xs)
845 {
846         struct ocfs2_inode_info *oi = OCFS2_I(inode);
847         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
848         struct ocfs2_xattr_value_root *xv;
849         size_t size;
850         int ret = 0;
851
852         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
853                 return -ENODATA;
854
855         xs->end = (void *)di + inode->i_sb->s_blocksize;
856         xs->header = (struct ocfs2_xattr_header *)
857                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
858         xs->base = (void *)xs->header;
859         xs->here = xs->header->xh_entries;
860
861         ret = ocfs2_xattr_find_entry(name_index, name, xs);
862         if (ret)
863                 return ret;
864         size = le64_to_cpu(xs->here->xe_value_size);
865         if (buffer) {
866                 if (size > buffer_size)
867                         return -ERANGE;
868                 if (ocfs2_xattr_is_local(xs->here)) {
869                         memcpy(buffer, (void *)xs->base +
870                                le16_to_cpu(xs->here->xe_name_offset) +
871                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
872                 } else {
873                         xv = (struct ocfs2_xattr_value_root *)
874                                 (xs->base + le16_to_cpu(
875                                  xs->here->xe_name_offset) +
876                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
877                         ret = ocfs2_xattr_get_value_outside(inode, xv,
878                                                             buffer, size);
879                         if (ret < 0) {
880                                 mlog_errno(ret);
881                                 return ret;
882                         }
883                 }
884         }
885
886         return size;
887 }
888
889 static int ocfs2_xattr_block_get(struct inode *inode,
890                                  int name_index,
891                                  const char *name,
892                                  void *buffer,
893                                  size_t buffer_size,
894                                  struct ocfs2_xattr_search *xs)
895 {
896         struct ocfs2_xattr_block *xb;
897         struct ocfs2_xattr_value_root *xv;
898         size_t size;
899         int ret = -ENODATA, name_offset, name_len, block_off, i;
900
901         xs->bucket = ocfs2_xattr_bucket_new(inode);
902         if (!xs->bucket) {
903                 ret = -ENOMEM;
904                 mlog_errno(ret);
905                 goto cleanup;
906         }
907
908         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
909         if (ret) {
910                 mlog_errno(ret);
911                 goto cleanup;
912         }
913
914         if (xs->not_found) {
915                 ret = -ENODATA;
916                 goto cleanup;
917         }
918
919         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
920         size = le64_to_cpu(xs->here->xe_value_size);
921         if (buffer) {
922                 ret = -ERANGE;
923                 if (size > buffer_size)
924                         goto cleanup;
925
926                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
927                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
928                 i = xs->here - xs->header->xh_entries;
929
930                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
931                         ret = ocfs2_xattr_bucket_get_name_value(inode,
932                                                                 bucket_xh(xs->bucket),
933                                                                 i,
934                                                                 &block_off,
935                                                                 &name_offset);
936                         xs->base = bucket_block(xs->bucket, block_off);
937                 }
938                 if (ocfs2_xattr_is_local(xs->here)) {
939                         memcpy(buffer, (void *)xs->base +
940                                name_offset + name_len, size);
941                 } else {
942                         xv = (struct ocfs2_xattr_value_root *)
943                                 (xs->base + name_offset + name_len);
944                         ret = ocfs2_xattr_get_value_outside(inode, xv,
945                                                             buffer, size);
946                         if (ret < 0) {
947                                 mlog_errno(ret);
948                                 goto cleanup;
949                         }
950                 }
951         }
952         ret = size;
953 cleanup:
954         ocfs2_xattr_bucket_free(xs->bucket);
955
956         brelse(xs->xattr_bh);
957         xs->xattr_bh = NULL;
958         return ret;
959 }
960
961 /* ocfs2_xattr_get()
962  *
963  * Copy an extended attribute into the buffer provided.
964  * Buffer is NULL to compute the size of buffer required.
965  */
966 static int ocfs2_xattr_get(struct inode *inode,
967                            int name_index,
968                            const char *name,
969                            void *buffer,
970                            size_t buffer_size)
971 {
972         int ret;
973         struct ocfs2_dinode *di = NULL;
974         struct buffer_head *di_bh = NULL;
975         struct ocfs2_inode_info *oi = OCFS2_I(inode);
976         struct ocfs2_xattr_search xis = {
977                 .not_found = -ENODATA,
978         };
979         struct ocfs2_xattr_search xbs = {
980                 .not_found = -ENODATA,
981         };
982
983         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
984                 return -EOPNOTSUPP;
985
986         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
987                 ret = -ENODATA;
988
989         ret = ocfs2_inode_lock(inode, &di_bh, 0);
990         if (ret < 0) {
991                 mlog_errno(ret);
992                 return ret;
993         }
994         xis.inode_bh = xbs.inode_bh = di_bh;
995         di = (struct ocfs2_dinode *)di_bh->b_data;
996
997         down_read(&oi->ip_xattr_sem);
998         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
999                                     buffer_size, &xis);
1000         if (ret == -ENODATA && di->i_xattr_loc)
1001                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1002                                             buffer_size, &xbs);
1003         up_read(&oi->ip_xattr_sem);
1004         ocfs2_inode_unlock(inode, 0);
1005
1006         brelse(di_bh);
1007
1008         return ret;
1009 }
1010
1011 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1012                                            struct ocfs2_xattr_value_root *xv,
1013                                            const void *value,
1014                                            int value_len)
1015 {
1016         int ret = 0, i, cp_len, credits;
1017         u16 blocksize = inode->i_sb->s_blocksize;
1018         u32 p_cluster, num_clusters;
1019         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1020         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1021         u64 blkno;
1022         struct buffer_head *bh = NULL;
1023         handle_t *handle;
1024
1025         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1026
1027         credits = clusters * bpc;
1028         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
1029         if (IS_ERR(handle)) {
1030                 ret = PTR_ERR(handle);
1031                 mlog_errno(ret);
1032                 goto out;
1033         }
1034
1035         while (cpos < clusters) {
1036                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1037                                                &num_clusters, &xv->xr_list);
1038                 if (ret) {
1039                         mlog_errno(ret);
1040                         goto out_commit;
1041                 }
1042
1043                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1044
1045                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1046                         ret = ocfs2_read_block(inode, blkno, &bh);
1047                         if (ret) {
1048                                 mlog_errno(ret);
1049                                 goto out_commit;
1050                         }
1051
1052                         ret = ocfs2_journal_access(handle,
1053                                                    inode,
1054                                                    bh,
1055                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1056                         if (ret < 0) {
1057                                 mlog_errno(ret);
1058                                 goto out_commit;
1059                         }
1060
1061                         cp_len = value_len > blocksize ? blocksize : value_len;
1062                         memcpy(bh->b_data, value, cp_len);
1063                         value_len -= cp_len;
1064                         value += cp_len;
1065                         if (cp_len < blocksize)
1066                                 memset(bh->b_data + cp_len, 0,
1067                                        blocksize - cp_len);
1068
1069                         ret = ocfs2_journal_dirty(handle, bh);
1070                         if (ret < 0) {
1071                                 mlog_errno(ret);
1072                                 goto out_commit;
1073                         }
1074                         brelse(bh);
1075                         bh = NULL;
1076
1077                         /*
1078                          * XXX: do we need to empty all the following
1079                          * blocks in this cluster?
1080                          */
1081                         if (!value_len)
1082                                 break;
1083                 }
1084                 cpos += num_clusters;
1085         }
1086 out_commit:
1087         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1088 out:
1089         brelse(bh);
1090
1091         return ret;
1092 }
1093
1094 static int ocfs2_xattr_cleanup(struct inode *inode,
1095                                struct ocfs2_xattr_info *xi,
1096                                struct ocfs2_xattr_search *xs,
1097                                size_t offs)
1098 {
1099         handle_t *handle = NULL;
1100         int ret = 0;
1101         size_t name_len = strlen(xi->name);
1102         void *val = xs->base + offs;
1103         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1104
1105         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1106                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1107         if (IS_ERR(handle)) {
1108                 ret = PTR_ERR(handle);
1109                 mlog_errno(ret);
1110                 goto out;
1111         }
1112         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1113                                    OCFS2_JOURNAL_ACCESS_WRITE);
1114         if (ret) {
1115                 mlog_errno(ret);
1116                 goto out_commit;
1117         }
1118         /* Decrease xattr count */
1119         le16_add_cpu(&xs->header->xh_count, -1);
1120         /* Remove the xattr entry and tree root which has already be set*/
1121         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1122         memset(val, 0, size);
1123
1124         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1125         if (ret < 0)
1126                 mlog_errno(ret);
1127 out_commit:
1128         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1129 out:
1130         return ret;
1131 }
1132
1133 static int ocfs2_xattr_update_entry(struct inode *inode,
1134                                     struct ocfs2_xattr_info *xi,
1135                                     struct ocfs2_xattr_search *xs,
1136                                     size_t offs)
1137 {
1138         handle_t *handle = NULL;
1139         int ret = 0;
1140
1141         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1142                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1143         if (IS_ERR(handle)) {
1144                 ret = PTR_ERR(handle);
1145                 mlog_errno(ret);
1146                 goto out;
1147         }
1148         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1149                                    OCFS2_JOURNAL_ACCESS_WRITE);
1150         if (ret) {
1151                 mlog_errno(ret);
1152                 goto out_commit;
1153         }
1154
1155         xs->here->xe_name_offset = cpu_to_le16(offs);
1156         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1157         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1158                 ocfs2_xattr_set_local(xs->here, 1);
1159         else
1160                 ocfs2_xattr_set_local(xs->here, 0);
1161         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1162
1163         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1164         if (ret < 0)
1165                 mlog_errno(ret);
1166 out_commit:
1167         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1168 out:
1169         return ret;
1170 }
1171
1172 /*
1173  * ocfs2_xattr_set_value_outside()
1174  *
1175  * Set large size value in B tree.
1176  */
1177 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1178                                          struct ocfs2_xattr_info *xi,
1179                                          struct ocfs2_xattr_search *xs,
1180                                          size_t offs)
1181 {
1182         size_t name_len = strlen(xi->name);
1183         void *val = xs->base + offs;
1184         struct ocfs2_xattr_value_root *xv = NULL;
1185         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1186         int ret = 0;
1187
1188         memset(val, 0, size);
1189         memcpy(val, xi->name, name_len);
1190         xv = (struct ocfs2_xattr_value_root *)
1191                 (val + OCFS2_XATTR_SIZE(name_len));
1192         xv->xr_clusters = 0;
1193         xv->xr_last_eb_blk = 0;
1194         xv->xr_list.l_tree_depth = 0;
1195         xv->xr_list.l_count = cpu_to_le16(1);
1196         xv->xr_list.l_next_free_rec = 0;
1197
1198         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1199                                          xi->value_len);
1200         if (ret < 0) {
1201                 mlog_errno(ret);
1202                 return ret;
1203         }
1204         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1205                                               xi->value_len);
1206         if (ret < 0) {
1207                 mlog_errno(ret);
1208                 return ret;
1209         }
1210         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1211         if (ret < 0)
1212                 mlog_errno(ret);
1213
1214         return ret;
1215 }
1216
1217 /*
1218  * ocfs2_xattr_set_entry_local()
1219  *
1220  * Set, replace or remove extended attribute in local.
1221  */
1222 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1223                                         struct ocfs2_xattr_info *xi,
1224                                         struct ocfs2_xattr_search *xs,
1225                                         struct ocfs2_xattr_entry *last,
1226                                         size_t min_offs)
1227 {
1228         size_t name_len = strlen(xi->name);
1229         int i;
1230
1231         if (xi->value && xs->not_found) {
1232                 /* Insert the new xattr entry. */
1233                 le16_add_cpu(&xs->header->xh_count, 1);
1234                 ocfs2_xattr_set_type(last, xi->name_index);
1235                 ocfs2_xattr_set_local(last, 1);
1236                 last->xe_name_len = name_len;
1237         } else {
1238                 void *first_val;
1239                 void *val;
1240                 size_t offs, size;
1241
1242                 first_val = xs->base + min_offs;
1243                 offs = le16_to_cpu(xs->here->xe_name_offset);
1244                 val = xs->base + offs;
1245
1246                 if (le64_to_cpu(xs->here->xe_value_size) >
1247                     OCFS2_XATTR_INLINE_SIZE)
1248                         size = OCFS2_XATTR_SIZE(name_len) +
1249                                 OCFS2_XATTR_ROOT_SIZE;
1250                 else
1251                         size = OCFS2_XATTR_SIZE(name_len) +
1252                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1253
1254                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1255                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1256                         /* The old and the new value have the
1257                            same size. Just replace the value. */
1258                         ocfs2_xattr_set_local(xs->here, 1);
1259                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1260                         /* Clear value bytes. */
1261                         memset(val + OCFS2_XATTR_SIZE(name_len),
1262                                0,
1263                                OCFS2_XATTR_SIZE(xi->value_len));
1264                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1265                                xi->value,
1266                                xi->value_len);
1267                         return;
1268                 }
1269                 /* Remove the old name+value. */
1270                 memmove(first_val + size, first_val, val - first_val);
1271                 memset(first_val, 0, size);
1272                 xs->here->xe_name_hash = 0;
1273                 xs->here->xe_name_offset = 0;
1274                 ocfs2_xattr_set_local(xs->here, 1);
1275                 xs->here->xe_value_size = 0;
1276
1277                 min_offs += size;
1278
1279                 /* Adjust all value offsets. */
1280                 last = xs->header->xh_entries;
1281                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1282                         size_t o = le16_to_cpu(last->xe_name_offset);
1283
1284                         if (o < offs)
1285                                 last->xe_name_offset = cpu_to_le16(o + size);
1286                         last += 1;
1287                 }
1288
1289                 if (!xi->value) {
1290                         /* Remove the old entry. */
1291                         last -= 1;
1292                         memmove(xs->here, xs->here + 1,
1293                                 (void *)last - (void *)xs->here);
1294                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1295                         le16_add_cpu(&xs->header->xh_count, -1);
1296                 }
1297         }
1298         if (xi->value) {
1299                 /* Insert the new name+value. */
1300                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1301                                 OCFS2_XATTR_SIZE(xi->value_len);
1302                 void *val = xs->base + min_offs - size;
1303
1304                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1305                 memset(val, 0, size);
1306                 memcpy(val, xi->name, name_len);
1307                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1308                        xi->value,
1309                        xi->value_len);
1310                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1311                 ocfs2_xattr_set_local(xs->here, 1);
1312                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1313         }
1314
1315         return;
1316 }
1317
1318 /*
1319  * ocfs2_xattr_set_entry()
1320  *
1321  * Set extended attribute entry into inode or block.
1322  *
1323  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1324  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1325  * then set value in B tree with set_value_outside().
1326  */
1327 static int ocfs2_xattr_set_entry(struct inode *inode,
1328                                  struct ocfs2_xattr_info *xi,
1329                                  struct ocfs2_xattr_search *xs,
1330                                  int flag)
1331 {
1332         struct ocfs2_xattr_entry *last;
1333         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1334         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1335         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1336         size_t size_l = 0;
1337         handle_t *handle = NULL;
1338         int free, i, ret;
1339         struct ocfs2_xattr_info xi_l = {
1340                 .name_index = xi->name_index,
1341                 .name = xi->name,
1342                 .value = xi->value,
1343                 .value_len = xi->value_len,
1344         };
1345
1346         /* Compute min_offs, last and free space. */
1347         last = xs->header->xh_entries;
1348
1349         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1350                 size_t offs = le16_to_cpu(last->xe_name_offset);
1351                 if (offs < min_offs)
1352                         min_offs = offs;
1353                 last += 1;
1354         }
1355
1356         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1357         if (free < 0)
1358                 return -EIO;
1359
1360         if (!xs->not_found) {
1361                 size_t size = 0;
1362                 if (ocfs2_xattr_is_local(xs->here))
1363                         size = OCFS2_XATTR_SIZE(name_len) +
1364                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1365                 else
1366                         size = OCFS2_XATTR_SIZE(name_len) +
1367                                 OCFS2_XATTR_ROOT_SIZE;
1368                 free += (size + sizeof(struct ocfs2_xattr_entry));
1369         }
1370         /* Check free space in inode or block */
1371         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1372                 if (free < sizeof(struct ocfs2_xattr_entry) +
1373                            OCFS2_XATTR_SIZE(name_len) +
1374                            OCFS2_XATTR_ROOT_SIZE) {
1375                         ret = -ENOSPC;
1376                         goto out;
1377                 }
1378                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1379                 xi_l.value = (void *)&def_xv;
1380                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1381         } else if (xi->value) {
1382                 if (free < sizeof(struct ocfs2_xattr_entry) +
1383                            OCFS2_XATTR_SIZE(name_len) +
1384                            OCFS2_XATTR_SIZE(xi->value_len)) {
1385                         ret = -ENOSPC;
1386                         goto out;
1387                 }
1388         }
1389
1390         if (!xs->not_found) {
1391                 /* For existing extended attribute */
1392                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1393                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1394                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1395                 void *val = xs->base + offs;
1396
1397                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1398                         /* Replace existing local xattr with tree root */
1399                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1400                                                             offs);
1401                         if (ret < 0)
1402                                 mlog_errno(ret);
1403                         goto out;
1404                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1405                         /* For existing xattr which has value outside */
1406                         struct ocfs2_xattr_value_root *xv = NULL;
1407                         xv = (struct ocfs2_xattr_value_root *)(val +
1408                                 OCFS2_XATTR_SIZE(name_len));
1409
1410                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1411                                 /*
1412                                  * If new value need set outside also,
1413                                  * first truncate old value to new value,
1414                                  * then set new value with set_value_outside().
1415                                  */
1416                                 ret = ocfs2_xattr_value_truncate(inode,
1417                                                                  xs->xattr_bh,
1418                                                                  xv,
1419                                                                  xi->value_len);
1420                                 if (ret < 0) {
1421                                         mlog_errno(ret);
1422                                         goto out;
1423                                 }
1424
1425                                 ret = __ocfs2_xattr_set_value_outside(inode,
1426                                                                 xv,
1427                                                                 xi->value,
1428                                                                 xi->value_len);
1429                                 if (ret < 0) {
1430                                         mlog_errno(ret);
1431                                         goto out;
1432                                 }
1433
1434                                 ret = ocfs2_xattr_update_entry(inode,
1435                                                                xi,
1436                                                                xs,
1437                                                                offs);
1438                                 if (ret < 0)
1439                                         mlog_errno(ret);
1440                                 goto out;
1441                         } else {
1442                                 /*
1443                                  * If new value need set in local,
1444                                  * just trucate old value to zero.
1445                                  */
1446                                  ret = ocfs2_xattr_value_truncate(inode,
1447                                                                  xs->xattr_bh,
1448                                                                  xv,
1449                                                                  0);
1450                                 if (ret < 0)
1451                                         mlog_errno(ret);
1452                         }
1453                 }
1454         }
1455
1456         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1457                                    OCFS2_INODE_UPDATE_CREDITS);
1458         if (IS_ERR(handle)) {
1459                 ret = PTR_ERR(handle);
1460                 mlog_errno(ret);
1461                 goto out;
1462         }
1463
1464         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1465                                    OCFS2_JOURNAL_ACCESS_WRITE);
1466         if (ret) {
1467                 mlog_errno(ret);
1468                 goto out_commit;
1469         }
1470
1471         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1472                 /* set extended attribute in external block. */
1473                 ret = ocfs2_extend_trans(handle,
1474                                          OCFS2_INODE_UPDATE_CREDITS +
1475                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1476                 if (ret) {
1477                         mlog_errno(ret);
1478                         goto out_commit;
1479                 }
1480                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1481                                            OCFS2_JOURNAL_ACCESS_WRITE);
1482                 if (ret) {
1483                         mlog_errno(ret);
1484                         goto out_commit;
1485                 }
1486         }
1487
1488         /*
1489          * Set value in local, include set tree root in local.
1490          * This is the first step for value size >INLINE_SIZE.
1491          */
1492         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1493
1494         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1495                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1496                 if (ret < 0) {
1497                         mlog_errno(ret);
1498                         goto out_commit;
1499                 }
1500         }
1501
1502         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1503             (flag & OCFS2_INLINE_XATTR_FL)) {
1504                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1505                 unsigned int xattrsize = osb->s_xattr_inline_size;
1506
1507                 /*
1508                  * Adjust extent record count or inline data size
1509                  * to reserve space for extended attribute.
1510                  */
1511                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1512                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1513                         le16_add_cpu(&idata->id_count, -xattrsize);
1514                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1515                         struct ocfs2_extent_list *el = &di->id2.i_list;
1516                         le16_add_cpu(&el->l_count, -(xattrsize /
1517                                         sizeof(struct ocfs2_extent_rec)));
1518                 }
1519                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1520         }
1521         /* Update xattr flag */
1522         spin_lock(&oi->ip_lock);
1523         oi->ip_dyn_features |= flag;
1524         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1525         spin_unlock(&oi->ip_lock);
1526         /* Update inode ctime */
1527         inode->i_ctime = CURRENT_TIME;
1528         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1529         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1530
1531         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1532         if (ret < 0)
1533                 mlog_errno(ret);
1534
1535 out_commit:
1536         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1537
1538         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1539                 /*
1540                  * Set value outside in B tree.
1541                  * This is the second step for value size > INLINE_SIZE.
1542                  */
1543                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1544                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1545                 if (ret < 0) {
1546                         int ret2;
1547
1548                         mlog_errno(ret);
1549                         /*
1550                          * If set value outside failed, we have to clean
1551                          * the junk tree root we have already set in local.
1552                          */
1553                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1554                         if (ret2 < 0)
1555                                 mlog_errno(ret2);
1556                 }
1557         }
1558 out:
1559         return ret;
1560
1561 }
1562
1563 static int ocfs2_remove_value_outside(struct inode*inode,
1564                                       struct buffer_head *bh,
1565                                       struct ocfs2_xattr_header *header)
1566 {
1567         int ret = 0, i;
1568
1569         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1570                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1571
1572                 if (!ocfs2_xattr_is_local(entry)) {
1573                         struct ocfs2_xattr_value_root *xv;
1574                         void *val;
1575
1576                         val = (void *)header +
1577                                 le16_to_cpu(entry->xe_name_offset);
1578                         xv = (struct ocfs2_xattr_value_root *)
1579                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1580                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1581                         if (ret < 0) {
1582                                 mlog_errno(ret);
1583                                 return ret;
1584                         }
1585                 }
1586         }
1587
1588         return ret;
1589 }
1590
1591 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1592                                     struct buffer_head *di_bh)
1593 {
1594
1595         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1596         struct ocfs2_xattr_header *header;
1597         int ret;
1598
1599         header = (struct ocfs2_xattr_header *)
1600                  ((void *)di + inode->i_sb->s_blocksize -
1601                  le16_to_cpu(di->i_xattr_inline_size));
1602
1603         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1604
1605         return ret;
1606 }
1607
1608 static int ocfs2_xattr_block_remove(struct inode *inode,
1609                                     struct buffer_head *blk_bh)
1610 {
1611         struct ocfs2_xattr_block *xb;
1612         int ret = 0;
1613
1614         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1615         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1616                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1617                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1618         } else
1619                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1620
1621         return ret;
1622 }
1623
1624 static int ocfs2_xattr_free_block(struct inode *inode,
1625                                   u64 block)
1626 {
1627         struct inode *xb_alloc_inode;
1628         struct buffer_head *xb_alloc_bh = NULL;
1629         struct buffer_head *blk_bh = NULL;
1630         struct ocfs2_xattr_block *xb;
1631         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1632         handle_t *handle;
1633         int ret = 0;
1634         u64 blk, bg_blkno;
1635         u16 bit;
1636
1637         ret = ocfs2_read_block(inode, block, &blk_bh);
1638         if (ret < 0) {
1639                 mlog_errno(ret);
1640                 goto out;
1641         }
1642
1643         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1644         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1645                 ret = -EIO;
1646                 goto out;
1647         }
1648
1649         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1650         if (ret < 0) {
1651                 mlog_errno(ret);
1652                 goto out;
1653         }
1654
1655         blk = le64_to_cpu(xb->xb_blkno);
1656         bit = le16_to_cpu(xb->xb_suballoc_bit);
1657         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1658
1659         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1660                                 EXTENT_ALLOC_SYSTEM_INODE,
1661                                 le16_to_cpu(xb->xb_suballoc_slot));
1662         if (!xb_alloc_inode) {
1663                 ret = -ENOMEM;
1664                 mlog_errno(ret);
1665                 goto out;
1666         }
1667         mutex_lock(&xb_alloc_inode->i_mutex);
1668
1669         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1670         if (ret < 0) {
1671                 mlog_errno(ret);
1672                 goto out_mutex;
1673         }
1674
1675         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1676         if (IS_ERR(handle)) {
1677                 ret = PTR_ERR(handle);
1678                 mlog_errno(ret);
1679                 goto out_unlock;
1680         }
1681
1682         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1683                                        bit, bg_blkno, 1);
1684         if (ret < 0)
1685                 mlog_errno(ret);
1686
1687         ocfs2_commit_trans(osb, handle);
1688 out_unlock:
1689         ocfs2_inode_unlock(xb_alloc_inode, 1);
1690         brelse(xb_alloc_bh);
1691 out_mutex:
1692         mutex_unlock(&xb_alloc_inode->i_mutex);
1693         iput(xb_alloc_inode);
1694 out:
1695         brelse(blk_bh);
1696         return ret;
1697 }
1698
1699 /*
1700  * ocfs2_xattr_remove()
1701  *
1702  * Free extended attribute resources associated with this inode.
1703  */
1704 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1705 {
1706         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1707         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1708         handle_t *handle;
1709         int ret;
1710
1711         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1712                 return 0;
1713
1714         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1715                 return 0;
1716
1717         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1718                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1719                 if (ret < 0) {
1720                         mlog_errno(ret);
1721                         goto out;
1722                 }
1723         }
1724
1725         if (di->i_xattr_loc) {
1726                 ret = ocfs2_xattr_free_block(inode,
1727                                              le64_to_cpu(di->i_xattr_loc));
1728                 if (ret < 0) {
1729                         mlog_errno(ret);
1730                         goto out;
1731                 }
1732         }
1733
1734         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1735                                    OCFS2_INODE_UPDATE_CREDITS);
1736         if (IS_ERR(handle)) {
1737                 ret = PTR_ERR(handle);
1738                 mlog_errno(ret);
1739                 goto out;
1740         }
1741         ret = ocfs2_journal_access(handle, inode, di_bh,
1742                                    OCFS2_JOURNAL_ACCESS_WRITE);
1743         if (ret) {
1744                 mlog_errno(ret);
1745                 goto out_commit;
1746         }
1747
1748         di->i_xattr_loc = 0;
1749
1750         spin_lock(&oi->ip_lock);
1751         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1752         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1753         spin_unlock(&oi->ip_lock);
1754
1755         ret = ocfs2_journal_dirty(handle, di_bh);
1756         if (ret < 0)
1757                 mlog_errno(ret);
1758 out_commit:
1759         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1760 out:
1761         return ret;
1762 }
1763
1764 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1765                                         struct ocfs2_dinode *di)
1766 {
1767         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1768         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1769         int free;
1770
1771         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1772                 return 0;
1773
1774         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1775                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1776                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1777         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1778                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1779                         le64_to_cpu(di->i_size);
1780         } else {
1781                 struct ocfs2_extent_list *el = &di->id2.i_list;
1782                 free = (le16_to_cpu(el->l_count) -
1783                         le16_to_cpu(el->l_next_free_rec)) *
1784                         sizeof(struct ocfs2_extent_rec);
1785         }
1786         if (free >= xattrsize)
1787                 return 1;
1788
1789         return 0;
1790 }
1791
1792 /*
1793  * ocfs2_xattr_ibody_find()
1794  *
1795  * Find extended attribute in inode block and
1796  * fill search info into struct ocfs2_xattr_search.
1797  */
1798 static int ocfs2_xattr_ibody_find(struct inode *inode,
1799                                   int name_index,
1800                                   const char *name,
1801                                   struct ocfs2_xattr_search *xs)
1802 {
1803         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1804         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1805         int ret;
1806         int has_space = 0;
1807
1808         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1809                 return 0;
1810
1811         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1812                 down_read(&oi->ip_alloc_sem);
1813                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1814                 up_read(&oi->ip_alloc_sem);
1815                 if (!has_space)
1816                         return 0;
1817         }
1818
1819         xs->xattr_bh = xs->inode_bh;
1820         xs->end = (void *)di + inode->i_sb->s_blocksize;
1821         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1822                 xs->header = (struct ocfs2_xattr_header *)
1823                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1824         else
1825                 xs->header = (struct ocfs2_xattr_header *)
1826                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1827         xs->base = (void *)xs->header;
1828         xs->here = xs->header->xh_entries;
1829
1830         /* Find the named attribute. */
1831         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1832                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1833                 if (ret && ret != -ENODATA)
1834                         return ret;
1835                 xs->not_found = ret;
1836         }
1837
1838         return 0;
1839 }
1840
1841 /*
1842  * ocfs2_xattr_ibody_set()
1843  *
1844  * Set, replace or remove an extended attribute into inode block.
1845  *
1846  */
1847 static int ocfs2_xattr_ibody_set(struct inode *inode,
1848                                  struct ocfs2_xattr_info *xi,
1849                                  struct ocfs2_xattr_search *xs)
1850 {
1851         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1852         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1853         int ret;
1854
1855         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1856                 return -ENOSPC;
1857
1858         down_write(&oi->ip_alloc_sem);
1859         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1860                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1861                         ret = -ENOSPC;
1862                         goto out;
1863                 }
1864         }
1865
1866         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1867                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1868 out:
1869         up_write(&oi->ip_alloc_sem);
1870
1871         return ret;
1872 }
1873
1874 /*
1875  * ocfs2_xattr_block_find()
1876  *
1877  * Find extended attribute in external block and
1878  * fill search info into struct ocfs2_xattr_search.
1879  */
1880 static int ocfs2_xattr_block_find(struct inode *inode,
1881                                   int name_index,
1882                                   const char *name,
1883                                   struct ocfs2_xattr_search *xs)
1884 {
1885         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1886         struct buffer_head *blk_bh = NULL;
1887         struct ocfs2_xattr_block *xb;
1888         int ret = 0;
1889
1890         if (!di->i_xattr_loc)
1891                 return ret;
1892
1893         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
1894         if (ret < 0) {
1895                 mlog_errno(ret);
1896                 return ret;
1897         }
1898
1899         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1900         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1901                 ret = -EIO;
1902                 goto cleanup;
1903         }
1904
1905         xs->xattr_bh = blk_bh;
1906
1907         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1908                 xs->header = &xb->xb_attrs.xb_header;
1909                 xs->base = (void *)xs->header;
1910                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1911                 xs->here = xs->header->xh_entries;
1912
1913                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1914         } else
1915                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1916                                                    name_index,
1917                                                    name, xs);
1918
1919         if (ret && ret != -ENODATA) {
1920                 xs->xattr_bh = NULL;
1921                 goto cleanup;
1922         }
1923         xs->not_found = ret;
1924         return 0;
1925 cleanup:
1926         brelse(blk_bh);
1927
1928         return ret;
1929 }
1930
1931 /*
1932  * ocfs2_xattr_block_set()
1933  *
1934  * Set, replace or remove an extended attribute into external block.
1935  *
1936  */
1937 static int ocfs2_xattr_block_set(struct inode *inode,
1938                                  struct ocfs2_xattr_info *xi,
1939                                  struct ocfs2_xattr_search *xs)
1940 {
1941         struct buffer_head *new_bh = NULL;
1942         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1943         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1944         struct ocfs2_alloc_context *meta_ac = NULL;
1945         handle_t *handle = NULL;
1946         struct ocfs2_xattr_block *xblk = NULL;
1947         u16 suballoc_bit_start;
1948         u32 num_got;
1949         u64 first_blkno;
1950         int ret;
1951
1952         if (!xs->xattr_bh) {
1953                 /*
1954                  * Alloc one external block for extended attribute
1955                  * outside of inode.
1956                  */
1957                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1958                 if (ret < 0) {
1959                         mlog_errno(ret);
1960                         goto out;
1961                 }
1962                 handle = ocfs2_start_trans(osb,
1963                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1964                 if (IS_ERR(handle)) {
1965                         ret = PTR_ERR(handle);
1966                         mlog_errno(ret);
1967                         goto out;
1968                 }
1969                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1970                                            OCFS2_JOURNAL_ACCESS_CREATE);
1971                 if (ret < 0) {
1972                         mlog_errno(ret);
1973                         goto out_commit;
1974                 }
1975
1976                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1977                                            &suballoc_bit_start, &num_got,
1978                                            &first_blkno);
1979                 if (ret < 0) {
1980                         mlog_errno(ret);
1981                         goto out_commit;
1982                 }
1983
1984                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1985                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1986
1987                 ret = ocfs2_journal_access(handle, inode, new_bh,
1988                                            OCFS2_JOURNAL_ACCESS_CREATE);
1989                 if (ret < 0) {
1990                         mlog_errno(ret);
1991                         goto out_commit;
1992                 }
1993
1994                 /* Initialize ocfs2_xattr_block */
1995                 xs->xattr_bh = new_bh;
1996                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1997                 memset(xblk, 0, inode->i_sb->s_blocksize);
1998                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1999                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2000                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2001                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2002                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2003
2004                 xs->header = &xblk->xb_attrs.xb_header;
2005                 xs->base = (void *)xs->header;
2006                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2007                 xs->here = xs->header->xh_entries;
2008
2009
2010                 ret = ocfs2_journal_dirty(handle, new_bh);
2011                 if (ret < 0) {
2012                         mlog_errno(ret);
2013                         goto out_commit;
2014                 }
2015                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2016                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
2017                 if (ret < 0)
2018                         mlog_errno(ret);
2019 out_commit:
2020                 ocfs2_commit_trans(osb, handle);
2021 out:
2022                 if (meta_ac)
2023                         ocfs2_free_alloc_context(meta_ac);
2024                 if (ret < 0)
2025                         return ret;
2026         } else
2027                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2028
2029         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2030                 /* Set extended attribute into external block */
2031                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
2032                 if (!ret || ret != -ENOSPC)
2033                         goto end;
2034
2035                 ret = ocfs2_xattr_create_index_block(inode, xs);
2036                 if (ret)
2037                         goto end;
2038         }
2039
2040         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
2041
2042 end:
2043
2044         return ret;
2045 }
2046
2047 /*
2048  * ocfs2_xattr_set()
2049  *
2050  * Set, replace or remove an extended attribute for this inode.
2051  * value is NULL to remove an existing extended attribute, else either
2052  * create or replace an extended attribute.
2053  */
2054 int ocfs2_xattr_set(struct inode *inode,
2055                     int name_index,
2056                     const char *name,
2057                     const void *value,
2058                     size_t value_len,
2059                     int flags)
2060 {
2061         struct buffer_head *di_bh = NULL;
2062         struct ocfs2_dinode *di;
2063         int ret;
2064
2065         struct ocfs2_xattr_info xi = {
2066                 .name_index = name_index,
2067                 .name = name,
2068                 .value = value,
2069                 .value_len = value_len,
2070         };
2071
2072         struct ocfs2_xattr_search xis = {
2073                 .not_found = -ENODATA,
2074         };
2075
2076         struct ocfs2_xattr_search xbs = {
2077                 .not_found = -ENODATA,
2078         };
2079
2080         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2081                 return -EOPNOTSUPP;
2082
2083         /*
2084          * Only xbs will be used on indexed trees.  xis doesn't need a
2085          * bucket.
2086          */
2087         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2088         if (!xbs.bucket) {
2089                 mlog_errno(-ENOMEM);
2090                 return -ENOMEM;
2091         }
2092
2093         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2094         if (ret < 0) {
2095                 mlog_errno(ret);
2096                 goto cleanup_nolock;
2097         }
2098         xis.inode_bh = xbs.inode_bh = di_bh;
2099         di = (struct ocfs2_dinode *)di_bh->b_data;
2100
2101         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2102         /*
2103          * Scan inode and external block to find the same name
2104          * extended attribute and collect search infomation.
2105          */
2106         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2107         if (ret)
2108                 goto cleanup;
2109         if (xis.not_found) {
2110                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2111                 if (ret)
2112                         goto cleanup;
2113         }
2114
2115         if (xis.not_found && xbs.not_found) {
2116                 ret = -ENODATA;
2117                 if (flags & XATTR_REPLACE)
2118                         goto cleanup;
2119                 ret = 0;
2120                 if (!value)
2121                         goto cleanup;
2122         } else {
2123                 ret = -EEXIST;
2124                 if (flags & XATTR_CREATE)
2125                         goto cleanup;
2126         }
2127
2128         if (!value) {
2129                 /* Remove existing extended attribute */
2130                 if (!xis.not_found)
2131                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2132                 else if (!xbs.not_found)
2133                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2134         } else {
2135                 /* We always try to set extended attribute into inode first*/
2136                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2137                 if (!ret && !xbs.not_found) {
2138                         /*
2139                          * If succeed and that extended attribute existing in
2140                          * external block, then we will remove it.
2141                          */
2142                         xi.value = NULL;
2143                         xi.value_len = 0;
2144                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2145                 } else if (ret == -ENOSPC) {
2146                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2147                                 ret = ocfs2_xattr_block_find(inode, name_index,
2148                                                              name, &xbs);
2149                                 if (ret)
2150                                         goto cleanup;
2151                         }
2152                         /*
2153                          * If no space in inode, we will set extended attribute
2154                          * into external block.
2155                          */
2156                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2157                         if (ret)
2158                                 goto cleanup;
2159                         if (!xis.not_found) {
2160                                 /*
2161                                  * If succeed and that extended attribute
2162                                  * existing in inode, we will remove it.
2163                                  */
2164                                 xi.value = NULL;
2165                                 xi.value_len = 0;
2166                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2167                         }
2168                 }
2169         }
2170 cleanup:
2171         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2172         ocfs2_inode_unlock(inode, 1);
2173 cleanup_nolock:
2174         brelse(di_bh);
2175         brelse(xbs.xattr_bh);
2176         ocfs2_xattr_bucket_free(xbs.bucket);
2177
2178         return ret;
2179 }
2180
2181 /*
2182  * Find the xattr extent rec which may contains name_hash.
2183  * e_cpos will be the first name hash of the xattr rec.
2184  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2185  */
2186 static int ocfs2_xattr_get_rec(struct inode *inode,
2187                                u32 name_hash,
2188                                u64 *p_blkno,
2189                                u32 *e_cpos,
2190                                u32 *num_clusters,
2191                                struct ocfs2_extent_list *el)
2192 {
2193         int ret = 0, i;
2194         struct buffer_head *eb_bh = NULL;
2195         struct ocfs2_extent_block *eb;
2196         struct ocfs2_extent_rec *rec = NULL;
2197         u64 e_blkno = 0;
2198
2199         if (el->l_tree_depth) {
2200                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2201                 if (ret) {
2202                         mlog_errno(ret);
2203                         goto out;
2204                 }
2205
2206                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2207                 el = &eb->h_list;
2208
2209                 if (el->l_tree_depth) {
2210                         ocfs2_error(inode->i_sb,
2211                                     "Inode %lu has non zero tree depth in "
2212                                     "xattr tree block %llu\n", inode->i_ino,
2213                                     (unsigned long long)eb_bh->b_blocknr);
2214                         ret = -EROFS;
2215                         goto out;
2216                 }
2217         }
2218
2219         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2220                 rec = &el->l_recs[i];
2221
2222                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2223                         e_blkno = le64_to_cpu(rec->e_blkno);
2224                         break;
2225                 }
2226         }
2227
2228         if (!e_blkno) {
2229                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2230                             "record (%u, %u, 0) in xattr", inode->i_ino,
2231                             le32_to_cpu(rec->e_cpos),
2232                             ocfs2_rec_clusters(el, rec));
2233                 ret = -EROFS;
2234                 goto out;
2235         }
2236
2237         *p_blkno = le64_to_cpu(rec->e_blkno);
2238         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2239         if (e_cpos)
2240                 *e_cpos = le32_to_cpu(rec->e_cpos);
2241 out:
2242         brelse(eb_bh);
2243         return ret;
2244 }
2245
2246 typedef int (xattr_bucket_func)(struct inode *inode,
2247                                 struct ocfs2_xattr_bucket *bucket,
2248                                 void *para);
2249
2250 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2251                                    struct ocfs2_xattr_bucket *bucket,
2252                                    int name_index,
2253                                    const char *name,
2254                                    u32 name_hash,
2255                                    u16 *xe_index,
2256                                    int *found)
2257 {
2258         int i, ret = 0, cmp = 1, block_off, new_offset;
2259         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2260         size_t name_len = strlen(name);
2261         struct ocfs2_xattr_entry *xe = NULL;
2262         char *xe_name;
2263
2264         /*
2265          * We don't use binary search in the bucket because there
2266          * may be multiple entries with the same name hash.
2267          */
2268         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2269                 xe = &xh->xh_entries[i];
2270
2271                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2272                         continue;
2273                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2274                         break;
2275
2276                 cmp = name_index - ocfs2_xattr_get_type(xe);
2277                 if (!cmp)
2278                         cmp = name_len - xe->xe_name_len;
2279                 if (cmp)
2280                         continue;
2281
2282                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2283                                                         xh,
2284                                                         i,
2285                                                         &block_off,
2286                                                         &new_offset);
2287                 if (ret) {
2288                         mlog_errno(ret);
2289                         break;
2290                 }
2291
2292                 xe_name = bucket_block(bucket, block_off) + new_offset;
2293                 if (!memcmp(name, xe_name, name_len)) {
2294                         *xe_index = i;
2295                         *found = 1;
2296                         ret = 0;
2297                         break;
2298                 }
2299         }
2300
2301         return ret;
2302 }
2303
2304 /*
2305  * Find the specified xattr entry in a series of buckets.
2306  * This series start from p_blkno and last for num_clusters.
2307  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2308  * the num of the valid buckets.
2309  *
2310  * Return the buffer_head this xattr should reside in. And if the xattr's
2311  * hash is in the gap of 2 buckets, return the lower bucket.
2312  */
2313 static int ocfs2_xattr_bucket_find(struct inode *inode,
2314                                    int name_index,
2315                                    const char *name,
2316                                    u32 name_hash,
2317                                    u64 p_blkno,
2318                                    u32 first_hash,
2319                                    u32 num_clusters,
2320                                    struct ocfs2_xattr_search *xs)
2321 {
2322         int ret, found = 0;
2323         struct ocfs2_xattr_header *xh = NULL;
2324         struct ocfs2_xattr_entry *xe = NULL;
2325         u16 index = 0;
2326         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2327         int low_bucket = 0, bucket, high_bucket;
2328         struct ocfs2_xattr_bucket *search;
2329         u32 last_hash;
2330         u64 blkno, lower_blkno = 0;
2331
2332         search = ocfs2_xattr_bucket_new(inode);
2333         if (!search) {
2334                 ret = -ENOMEM;
2335                 mlog_errno(ret);
2336                 goto out;
2337         }
2338
2339         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2340         if (ret) {
2341                 mlog_errno(ret);
2342                 goto out;
2343         }
2344
2345         xh = bucket_xh(search);
2346         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2347         while (low_bucket <= high_bucket) {
2348                 ocfs2_xattr_bucket_relse(search);
2349
2350                 bucket = (low_bucket + high_bucket) / 2;
2351                 blkno = p_blkno + bucket * blk_per_bucket;
2352                 ret = ocfs2_read_xattr_bucket(search, blkno);
2353                 if (ret) {
2354                         mlog_errno(ret);
2355                         goto out;
2356                 }
2357
2358                 xh = bucket_xh(search);
2359                 xe = &xh->xh_entries[0];
2360                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2361                         high_bucket = bucket - 1;
2362                         continue;
2363                 }
2364
2365                 /*
2366                  * Check whether the hash of the last entry in our
2367                  * bucket is larger than the search one. for an empty
2368                  * bucket, the last one is also the first one.
2369                  */
2370                 if (xh->xh_count)
2371                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2372
2373                 last_hash = le32_to_cpu(xe->xe_name_hash);
2374
2375                 /* record lower_blkno which may be the insert place. */
2376                 lower_blkno = blkno;
2377
2378                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2379                         low_bucket = bucket + 1;
2380                         continue;
2381                 }
2382
2383                 /* the searched xattr should reside in this bucket if exists. */
2384                 ret = ocfs2_find_xe_in_bucket(inode, search,
2385                                               name_index, name, name_hash,
2386                                               &index, &found);
2387                 if (ret) {
2388                         mlog_errno(ret);
2389                         goto out;
2390                 }
2391                 break;
2392         }
2393
2394         /*
2395          * Record the bucket we have found.
2396          * When the xattr's hash value is in the gap of 2 buckets, we will
2397          * always set it to the previous bucket.
2398          */
2399         if (!lower_blkno)
2400                 lower_blkno = p_blkno;
2401
2402         /* This should be in cache - we just read it during the search */
2403         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
2404         if (ret) {
2405                 mlog_errno(ret);
2406                 goto out;
2407         }
2408
2409         xs->header = bucket_xh(xs->bucket);
2410         xs->base = bucket_block(xs->bucket, 0);
2411         xs->end = xs->base + inode->i_sb->s_blocksize;
2412
2413         if (found) {
2414                 xs->here = &xs->header->xh_entries[index];
2415                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2416                      (unsigned long long)bucket_blkno(xs->bucket), index);
2417         } else
2418                 ret = -ENODATA;
2419
2420 out:
2421         ocfs2_xattr_bucket_free(search);
2422         return ret;
2423 }
2424
2425 static int ocfs2_xattr_index_block_find(struct inode *inode,
2426                                         struct buffer_head *root_bh,
2427                                         int name_index,
2428                                         const char *name,
2429                                         struct ocfs2_xattr_search *xs)
2430 {
2431         int ret;
2432         struct ocfs2_xattr_block *xb =
2433                         (struct ocfs2_xattr_block *)root_bh->b_data;
2434         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2435         struct ocfs2_extent_list *el = &xb_root->xt_list;
2436         u64 p_blkno = 0;
2437         u32 first_hash, num_clusters = 0;
2438         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2439
2440         if (le16_to_cpu(el->l_next_free_rec) == 0)
2441                 return -ENODATA;
2442
2443         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2444              name, name_hash, name_index);
2445
2446         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2447                                   &num_clusters, el);
2448         if (ret) {
2449                 mlog_errno(ret);
2450                 goto out;
2451         }
2452
2453         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2454
2455         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2456              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
2457              first_hash);
2458
2459         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2460                                       p_blkno, first_hash, num_clusters, xs);
2461
2462 out:
2463         return ret;
2464 }
2465
2466 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2467                                        u64 blkno,
2468                                        u32 clusters,
2469                                        xattr_bucket_func *func,
2470                                        void *para)
2471 {
2472         int i, ret = 0;
2473         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2474         u32 num_buckets = clusters * bpc;
2475         struct ocfs2_xattr_bucket *bucket;
2476
2477         bucket = ocfs2_xattr_bucket_new(inode);
2478         if (!bucket) {
2479                 mlog_errno(-ENOMEM);
2480                 return -ENOMEM;
2481         }
2482
2483         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2484              clusters, (unsigned long long)blkno);
2485
2486         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
2487                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
2488                 if (ret) {
2489                         mlog_errno(ret);
2490                         break;
2491                 }
2492
2493                 /*
2494                  * The real bucket num in this series of blocks is stored
2495                  * in the 1st bucket.
2496                  */
2497                 if (i == 0)
2498                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
2499
2500                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
2501                      (unsigned long long)blkno,
2502                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
2503                 if (func) {
2504                         ret = func(inode, bucket, para);
2505                         if (ret)
2506                                 mlog_errno(ret);
2507                         /* Fall through to bucket_relse() */
2508                 }
2509
2510                 ocfs2_xattr_bucket_relse(bucket);
2511                 if (ret)
2512                         break;
2513         }
2514
2515         ocfs2_xattr_bucket_free(bucket);
2516         return ret;
2517 }
2518
2519 struct ocfs2_xattr_tree_list {
2520         char *buffer;
2521         size_t buffer_size;
2522         size_t result;
2523 };
2524
2525 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2526                                              struct ocfs2_xattr_header *xh,
2527                                              int index,
2528                                              int *block_off,
2529                                              int *new_offset)
2530 {
2531         u16 name_offset;
2532
2533         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2534                 return -EINVAL;
2535
2536         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2537
2538         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2539         *new_offset = name_offset % inode->i_sb->s_blocksize;
2540
2541         return 0;
2542 }
2543
2544 static int ocfs2_list_xattr_bucket(struct inode *inode,
2545                                    struct ocfs2_xattr_bucket *bucket,
2546                                    void *para)
2547 {
2548         int ret = 0, type;
2549         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2550         int i, block_off, new_offset;
2551         const char *prefix, *name;
2552
2553         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
2554                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
2555                 type = ocfs2_xattr_get_type(entry);
2556                 prefix = ocfs2_xattr_prefix(type);
2557
2558                 if (prefix) {
2559                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2560                                                                 bucket_xh(bucket),
2561                                                                 i,
2562                                                                 &block_off,
2563                                                                 &new_offset);
2564                         if (ret)
2565                                 break;
2566
2567                         name = (const char *)bucket_block(bucket, block_off) +
2568                                 new_offset;
2569                         ret = ocfs2_xattr_list_entry(xl->buffer,
2570                                                      xl->buffer_size,
2571                                                      &xl->result,
2572                                                      prefix, name,
2573                                                      entry->xe_name_len);
2574                         if (ret)
2575                                 break;
2576                 }
2577         }
2578
2579         return ret;
2580 }
2581
2582 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2583                                              struct ocfs2_xattr_tree_root *xt,
2584                                              char *buffer,
2585                                              size_t buffer_size)
2586 {
2587         struct ocfs2_extent_list *el = &xt->xt_list;
2588         int ret = 0;
2589         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2590         u64 p_blkno = 0;
2591         struct ocfs2_xattr_tree_list xl = {
2592                 .buffer = buffer,
2593                 .buffer_size = buffer_size,
2594                 .result = 0,
2595         };
2596
2597         if (le16_to_cpu(el->l_next_free_rec) == 0)
2598                 return 0;
2599
2600         while (name_hash > 0) {
2601                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2602                                           &e_cpos, &num_clusters, el);
2603                 if (ret) {
2604                         mlog_errno(ret);
2605                         goto out;
2606                 }
2607
2608                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2609                                                   ocfs2_list_xattr_bucket,
2610                                                   &xl);
2611                 if (ret) {
2612                         mlog_errno(ret);
2613                         goto out;
2614                 }
2615
2616                 if (e_cpos == 0)
2617                         break;
2618
2619                 name_hash = e_cpos - 1;
2620         }
2621
2622         ret = xl.result;
2623 out:
2624         return ret;
2625 }
2626
2627 static int cmp_xe(const void *a, const void *b)
2628 {
2629         const struct ocfs2_xattr_entry *l = a, *r = b;
2630         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2631         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2632
2633         if (l_hash > r_hash)
2634                 return 1;
2635         if (l_hash < r_hash)
2636                 return -1;
2637         return 0;
2638 }
2639
2640 static void swap_xe(void *a, void *b, int size)
2641 {
2642         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2643
2644         tmp = *l;
2645         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2646         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2647 }
2648
2649 /*
2650  * When the ocfs2_xattr_block is filled up, new bucket will be created
2651  * and all the xattr entries will be moved to the new bucket.
2652  * The header goes at the start of the bucket, and the names+values are
2653  * filled from the end.  This is why *target starts as the last buffer.
2654  * Note: we need to sort the entries since they are not saved in order
2655  * in the ocfs2_xattr_block.
2656  */
2657 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2658                                            struct buffer_head *xb_bh,
2659                                            struct ocfs2_xattr_bucket *bucket)
2660 {
2661         int i, blocksize = inode->i_sb->s_blocksize;
2662         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2663         u16 offset, size, off_change;
2664         struct ocfs2_xattr_entry *xe;
2665         struct ocfs2_xattr_block *xb =
2666                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2667         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2668         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2669         u16 count = le16_to_cpu(xb_xh->xh_count);
2670         char *src = xb_bh->b_data;
2671         char *target = bucket_block(bucket, blks - 1);
2672
2673         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2674              (unsigned long long)xb_bh->b_blocknr,
2675              (unsigned long long)bucket_blkno(bucket));
2676
2677         for (i = 0; i < blks; i++)
2678                 memset(bucket_block(bucket, i), 0, blocksize);
2679
2680         /*
2681          * Since the xe_name_offset is based on ocfs2_xattr_header,
2682          * there is a offset change corresponding to the change of
2683          * ocfs2_xattr_header's position.
2684          */
2685         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2686         xe = &xb_xh->xh_entries[count - 1];
2687         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2688         size = blocksize - offset;
2689
2690         /* copy all the names and values. */
2691         memcpy(target + offset, src + offset, size);
2692
2693         /* Init new header now. */
2694         xh->xh_count = xb_xh->xh_count;
2695         xh->xh_num_buckets = cpu_to_le16(1);
2696         xh->xh_name_value_len = cpu_to_le16(size);
2697         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2698
2699         /* copy all the entries. */
2700         target = bucket_block(bucket, 0);
2701         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2702         size = count * sizeof(struct ocfs2_xattr_entry);
2703         memcpy(target + offset, (char *)xb_xh + offset, size);
2704
2705         /* Change the xe offset for all the xe because of the move. */
2706         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2707                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2708         for (i = 0; i < count; i++)
2709                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2710
2711         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2712              offset, size, off_change);
2713
2714         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2715              cmp_xe, swap_xe);
2716 }
2717
2718 /*
2719  * After we move xattr from block to index btree, we have to
2720  * update ocfs2_xattr_search to the new xe and base.
2721  *
2722  * When the entry is in xattr block, xattr_bh indicates the storage place.
2723  * While if the entry is in index b-tree, "bucket" indicates the
2724  * real place of the xattr.
2725  */
2726 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
2727                                             struct ocfs2_xattr_search *xs,
2728                                             struct buffer_head *old_bh)
2729 {
2730         char *buf = old_bh->b_data;
2731         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2732         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2733         int i;
2734
2735         xs->header = bucket_xh(xs->bucket);
2736         xs->base = bucket_block(xs->bucket, 0);
2737         xs->end = xs->base + inode->i_sb->s_blocksize;
2738
2739         if (xs->not_found)
2740                 return;
2741
2742         i = xs->here - old_xh->xh_entries;
2743         xs->here = &xs->header->xh_entries[i];
2744 }
2745
2746 static int ocfs2_xattr_create_index_block(struct inode *inode,
2747                                           struct ocfs2_xattr_search *xs)
2748 {
2749         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2750         u32 bit_off, len;
2751         u64 blkno;
2752         handle_t *handle;
2753         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2754         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2755         struct ocfs2_alloc_context *data_ac;
2756         struct buffer_head *xb_bh = xs->xattr_bh;
2757         struct ocfs2_xattr_block *xb =
2758                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2759         struct ocfs2_xattr_tree_root *xr;
2760         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2761
2762         mlog(0, "create xattr index block for %llu\n",
2763              (unsigned long long)xb_bh->b_blocknr);
2764
2765         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2766         BUG_ON(!xs->bucket);
2767
2768         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2769         if (ret) {
2770                 mlog_errno(ret);
2771                 goto out;
2772         }
2773
2774         /*
2775          * XXX:
2776          * We can use this lock for now, and maybe move to a dedicated mutex
2777          * if performance becomes a problem later.
2778          */
2779         down_write(&oi->ip_alloc_sem);
2780
2781         /*
2782          * We need more credits.  One for the xattr block update and one
2783          * for each block of the new xattr bucket.
2784          */
2785         credits += 1 + ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2786         handle = ocfs2_start_trans(osb, credits);
2787         if (IS_ERR(handle)) {
2788                 ret = PTR_ERR(handle);
2789                 mlog_errno(ret);
2790                 goto out_sem;
2791         }
2792
2793         ret = ocfs2_journal_access(handle, inode, xb_bh,
2794                                    OCFS2_JOURNAL_ACCESS_WRITE);
2795         if (ret) {
2796                 mlog_errno(ret);
2797                 goto out_commit;
2798         }
2799
2800         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2801         if (ret) {
2802                 mlog_errno(ret);
2803                 goto out_commit;
2804         }
2805
2806         /*
2807          * The bucket may spread in many blocks, and
2808          * we will only touch the 1st block and the last block
2809          * in the whole bucket(one for entry and one for data).
2810          */
2811         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2812
2813         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
2814              (unsigned long long)blkno);
2815
2816         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
2817         if (ret) {
2818                 mlog_errno(ret);
2819                 goto out_commit;
2820         }
2821
2822         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
2823                                                 OCFS2_JOURNAL_ACCESS_CREATE);
2824         if (ret) {
2825                 mlog_errno(ret);
2826                 goto out_commit;
2827         }
2828
2829         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
2830         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
2831
2832         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
2833
2834         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2835         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2836                offsetof(struct ocfs2_xattr_block, xb_attrs));
2837
2838         xr = &xb->xb_attrs.xb_root;
2839         xr->xt_clusters = cpu_to_le32(1);
2840         xr->xt_last_eb_blk = 0;
2841         xr->xt_list.l_tree_depth = 0;
2842         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2843         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2844
2845         xr->xt_list.l_recs[0].e_cpos = 0;
2846         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2847         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2848
2849         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2850
2851         ret = ocfs2_journal_dirty(handle, xb_bh);
2852         if (ret) {
2853                 mlog_errno(ret);
2854                 goto out_commit;
2855         }
2856
2857 out_commit:
2858         ocfs2_commit_trans(osb, handle);
2859
2860 out_sem:
2861         up_write(&oi->ip_alloc_sem);
2862
2863 out:
2864         if (data_ac)
2865                 ocfs2_free_alloc_context(data_ac);
2866
2867         return ret;
2868 }
2869
2870 static int cmp_xe_offset(const void *a, const void *b)
2871 {
2872         const struct ocfs2_xattr_entry *l = a, *r = b;
2873         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2874         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2875
2876         if (l_name_offset < r_name_offset)
2877                 return 1;
2878         if (l_name_offset > r_name_offset)
2879                 return -1;
2880         return 0;
2881 }
2882
2883 /*
2884  * defrag a xattr bucket if we find that the bucket has some
2885  * holes beteen name/value pairs.
2886  * We will move all the name/value pairs to the end of the bucket
2887  * so that we can spare some space for insertion.
2888  */
2889 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2890                                      struct ocfs2_xattr_bucket *bucket)
2891 {
2892         int ret, i;
2893         size_t end, offset, len, value_len;
2894         struct ocfs2_xattr_header *xh;
2895         char *entries, *buf, *bucket_buf = NULL;
2896         u64 blkno = bucket_blkno(bucket);
2897         u16 xh_free_start;
2898         size_t blocksize = inode->i_sb->s_blocksize;
2899         handle_t *handle;
2900         struct ocfs2_xattr_entry *xe;
2901
2902         /*
2903          * In order to make the operation more efficient and generic,
2904          * we copy all the blocks into a contiguous memory and do the
2905          * defragment there, so if anything is error, we will not touch
2906          * the real block.
2907          */
2908         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2909         if (!bucket_buf) {
2910                 ret = -EIO;
2911                 goto out;
2912         }
2913
2914         buf = bucket_buf;
2915         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
2916                 memcpy(buf, bucket_block(bucket, i), blocksize);
2917
2918         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), bucket->bu_blocks);
2919         if (IS_ERR(handle)) {
2920                 ret = PTR_ERR(handle);
2921                 handle = NULL;
2922                 mlog_errno(ret);
2923                 goto out;
2924         }
2925
2926         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
2927                                                 OCFS2_JOURNAL_ACCESS_WRITE);
2928         if (ret < 0) {
2929                 mlog_errno(ret);
2930                 goto commit;
2931         }
2932
2933         xh = (struct ocfs2_xattr_header *)bucket_buf;
2934         entries = (char *)xh->xh_entries;
2935         xh_free_start = le16_to_cpu(xh->xh_free_start);
2936
2937         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2938              "xh_free_start = %u, xh_name_value_len = %u.\n",
2939              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
2940              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
2941
2942         /*
2943          * sort all the entries by their offset.
2944          * the largest will be the first, so that we can
2945          * move them to the end one by one.
2946          */
2947         sort(entries, le16_to_cpu(xh->xh_count),
2948              sizeof(struct ocfs2_xattr_entry),
2949              cmp_xe_offset, swap_xe);
2950
2951         /* Move all name/values to the end of the bucket. */
2952         xe = xh->xh_entries;
2953         end = OCFS2_XATTR_BUCKET_SIZE;
2954         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2955                 offset = le16_to_cpu(xe->xe_name_offset);
2956                 if (ocfs2_xattr_is_local(xe))
2957                         value_len = OCFS2_XATTR_SIZE(
2958                                         le64_to_cpu(xe->xe_value_size));
2959                 else
2960                         value_len = OCFS2_XATTR_ROOT_SIZE;
2961                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2962
2963                 /*
2964                  * We must make sure that the name/value pair
2965                  * exist in the same block. So adjust end to
2966                  * the previous block end if needed.
2967                  */
2968                 if (((end - len) / blocksize !=
2969                         (end - 1) / blocksize))
2970                         end = end - end % blocksize;
2971
2972                 if (end > offset + len) {
2973                         memmove(bucket_buf + end - len,
2974                                 bucket_buf + offset, len);
2975                         xe->xe_name_offset = cpu_to_le16(end - len);
2976                 }
2977
2978                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2979                                 "bucket %llu\n", (unsigned long long)blkno);
2980
2981                 end -= len;
2982         }
2983
2984         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2985                         "bucket %llu\n", (unsigned long long)blkno);
2986
2987         if (xh_free_start == end)
2988                 goto commit;
2989
2990         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2991         xh->xh_free_start = cpu_to_le16(end);
2992
2993         /* sort the entries by their name_hash. */
2994         sort(entries, le16_to_cpu(xh->xh_count),
2995              sizeof(struct ocfs2_xattr_entry),
2996              cmp_xe, swap_xe);
2997
2998         buf = bucket_buf;
2999         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3000                 memcpy(bucket_block(bucket, i), buf, blocksize);
3001         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3002
3003 commit:
3004         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3005 out:
3006         kfree(bucket_buf);
3007         return ret;
3008 }
3009
3010 /*
3011  * Move half nums of the xattr bucket in the previous cluster to this new
3012  * cluster. We only touch the last cluster of the previous extend record.
3013  *
3014  * first_bh is the first buffer_head of a series of bucket in the same
3015  * extent rec and header_bh is the header of one bucket in this cluster.
3016  * They will be updated if we move the data header_bh contains to the new
3017  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3018  */
3019 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3020                                                handle_t *handle,
3021                                                struct buffer_head **first_bh,
3022                                                struct buffer_head **header_bh,
3023                                                u64 new_blkno,
3024                                                u64 prev_blkno,
3025                                                u32 num_clusters,
3026                                                u32 *first_hash)
3027 {
3028         int i, ret, credits;
3029         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3030         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3031         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3032         int blocksize = inode->i_sb->s_blocksize;
3033         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3034         struct ocfs2_xattr_header *new_xh;
3035         struct ocfs2_xattr_header *xh =
3036                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3037
3038         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3039         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3040
3041         prev_bh = *first_bh;
3042         get_bh(prev_bh);
3043         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3044
3045         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3046
3047         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3048              (unsigned long long)prev_blkno, (unsigned long long)new_blkno);
3049
3050         /*
3051          * We need to update the 1st half of the new cluster and
3052          * 1 more for the update of the 1st bucket of the previous
3053          * extent record.
3054          */
3055         credits = bpc / 2 + 1;
3056         ret = ocfs2_extend_trans(handle, credits);
3057         if (ret) {
3058                 mlog_errno(ret);
3059                 goto out;
3060         }
3061
3062         ret = ocfs2_journal_access(handle, inode, prev_bh,
3063                                    OCFS2_JOURNAL_ACCESS_WRITE);
3064         if (ret) {
3065                 mlog_errno(ret);
3066                 goto out;
3067         }
3068
3069         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3070                 old_bh = new_bh = NULL;
3071                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3072                 if (!new_bh) {
3073                         ret = -EIO;
3074                         mlog_errno(ret);
3075                         goto out;
3076                 }
3077
3078                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3079
3080                 ret = ocfs2_journal_access(handle, inode, new_bh,
3081                                            OCFS2_JOURNAL_ACCESS_CREATE);
3082                 if (ret < 0) {
3083                         mlog_errno(ret);
3084                         brelse(new_bh);
3085                         goto out;
3086                 }
3087
3088                 ret = ocfs2_read_block(inode, prev_blkno, &old_bh);
3089                 if (ret < 0) {
3090                         mlog_errno(ret);
3091                         brelse(new_bh);
3092                         goto out;
3093                 }
3094
3095                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3096
3097                 if (i == 0) {
3098                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3099                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3100
3101                         if (first_hash)
3102                                 *first_hash = le32_to_cpu(
3103                                         new_xh->xh_entries[0].xe_name_hash);
3104                         new_first_bh = new_bh;
3105                         get_bh(new_first_bh);
3106                 }
3107
3108                 ocfs2_journal_dirty(handle, new_bh);
3109
3110                 if (*header_bh == old_bh) {
3111                         brelse(*header_bh);
3112                         *header_bh = new_bh;
3113                         get_bh(*header_bh);
3114
3115                         brelse(*first_bh);
3116                         *first_bh = new_first_bh;
3117                         get_bh(*first_bh);
3118                 }
3119                 brelse(new_bh);
3120                 brelse(old_bh);
3121         }
3122
3123         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3124
3125         ocfs2_journal_dirty(handle, prev_bh);
3126 out:
3127         brelse(prev_bh);
3128         brelse(new_first_bh);
3129         return ret;
3130 }
3131
3132 /*
3133  * Find the suitable pos when we divide a bucket into 2.
3134  * We have to make sure the xattrs with the same hash value exist
3135  * in the same bucket.
3136  *
3137  * If this ocfs2_xattr_header covers more than one hash value, find a
3138  * place where the hash value changes.  Try to find the most even split.
3139  * The most common case is that all entries have different hash values,
3140  * and the first check we make will find a place to split.
3141  */
3142 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3143 {
3144         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3145         int count = le16_to_cpu(xh->xh_count);
3146         int delta, middle = count / 2;
3147
3148         /*
3149          * We start at the middle.  Each step gets farther away in both
3150          * directions.  We therefore hit the change in hash value
3151          * nearest to the middle.  Note that this loop does not execute for
3152          * count < 2.
3153          */
3154         for (delta = 0; delta < middle; delta++) {
3155                 /* Let's check delta earlier than middle */
3156                 if (cmp_xe(&entries[middle - delta - 1],
3157                            &entries[middle - delta]))
3158                         return middle - delta;
3159
3160                 /* For even counts, don't walk off the end */
3161                 if ((middle + delta + 1) == count)
3162                         continue;
3163
3164                 /* Now try delta past middle */
3165                 if (cmp_xe(&entries[middle + delta],
3166                            &entries[middle + delta + 1]))
3167                         return middle + delta + 1;
3168         }
3169
3170         /* Every entry had the same hash */
3171         return count;
3172 }
3173
3174 /*
3175  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3176  * first_hash will record the 1st hash of the new bucket.
3177  *
3178  * Normally half of the xattrs will be moved.  But we have to make
3179  * sure that the xattrs with the same hash value are stored in the
3180  * same bucket. If all the xattrs in this bucket have the same hash
3181  * value, the new bucket will be initialized as an empty one and the
3182  * first_hash will be initialized as (hash_value+1).
3183  */
3184 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3185                                     handle_t *handle,
3186                                     u64 blk,
3187                                     u64 new_blk,
3188                                     u32 *first_hash,
3189                                     int new_bucket_head)
3190 {
3191         int ret, i;
3192         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3193         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3194         struct ocfs2_xattr_header *xh;
3195         struct ocfs2_xattr_entry *xe;
3196         int blocksize = inode->i_sb->s_blocksize;
3197
3198         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3199              (unsigned long long)blk, (unsigned long long)new_blk);
3200
3201         s_bucket = ocfs2_xattr_bucket_new(inode);
3202         t_bucket = ocfs2_xattr_bucket_new(inode);
3203         if (!s_bucket || !t_bucket) {
3204                 ret = -ENOMEM;
3205                 mlog_errno(ret);
3206                 goto out;
3207         }
3208
3209         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3210         if (ret) {
3211                 mlog_errno(ret);
3212                 goto out;
3213         }
3214
3215         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3216                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3217         if (ret) {
3218                 mlog_errno(ret);
3219                 goto out;
3220         }
3221
3222         /*
3223          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3224          * there's no need to read it.
3225          */
3226         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3227         if (ret) {
3228                 mlog_errno(ret);
3229                 goto out;
3230         }
3231
3232         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3233                                                 new_bucket_head ?
3234                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3235                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3236         if (ret) {
3237                 mlog_errno(ret);
3238                 goto out;
3239         }
3240
3241         xh = bucket_xh(s_bucket);
3242         count = le16_to_cpu(xh->xh_count);
3243         start = ocfs2_xattr_find_divide_pos(xh);
3244
3245         if (start == count) {
3246                 xe = &xh->xh_entries[start-1];
3247
3248                 /*
3249                  * initialized a new empty bucket here.
3250                  * The hash value is set as one larger than
3251                  * that of the last entry in the previous bucket.
3252                  */
3253                 for (i = 0; i < t_bucket->bu_blocks; i++)
3254                         memset(bucket_block(t_bucket, i), 0, blocksize);
3255
3256                 xh = bucket_xh(t_bucket);
3257                 xh->xh_free_start = cpu_to_le16(blocksize);
3258                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3259                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3260
3261                 goto set_num_buckets;
3262         }
3263
3264         /* copy the whole bucket to the new first. */
3265         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3266
3267         /* update the new bucket. */
3268         xh = bucket_xh(t_bucket);
3269
3270         /*
3271          * Calculate the total name/value len and xh_free_start for
3272          * the old bucket first.
3273          */
3274         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3275         name_value_len = 0;
3276         for (i = 0; i < start; i++) {
3277                 xe = &xh->xh_entries[i];
3278                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3279                 if (ocfs2_xattr_is_local(xe))
3280                         xe_len +=
3281                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3282                 else
3283                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3284                 name_value_len += xe_len;
3285                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3286                         name_offset = le16_to_cpu(xe->xe_name_offset);
3287         }
3288
3289         /*
3290          * Now begin the modification to the new bucket.
3291          *
3292          * In the new bucket, We just move the xattr entry to the beginning
3293          * and don't touch the name/value. So there will be some holes in the
3294          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3295          * called.
3296          */
3297         xe = &xh->xh_entries[start];
3298         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3299         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3300              (int)((char *)xe - (char *)xh),
3301              (int)((char *)xh->xh_entries - (char *)xh));
3302         memmove((char *)xh->xh_entries, (char *)xe, len);
3303         xe = &xh->xh_entries[count - start];
3304         len = sizeof(struct ocfs2_xattr_entry) * start;
3305         memset((char *)xe, 0, len);
3306
3307         le16_add_cpu(&xh->xh_count, -start);
3308         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3309
3310         /* Calculate xh_free_start for the new bucket. */
3311         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3312         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3313                 xe = &xh->xh_entries[i];
3314                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3315                 if (ocfs2_xattr_is_local(xe))
3316                         xe_len +=
3317                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3318                 else
3319                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3320                 if (le16_to_cpu(xe->xe_name_offset) <
3321                     le16_to_cpu(xh->xh_free_start))
3322                         xh->xh_free_start = xe->xe_name_offset;
3323         }
3324
3325 set_num_buckets:
3326         /* set xh->xh_num_buckets for the new xh. */
3327         if (new_bucket_head)
3328                 xh->xh_num_buckets = cpu_to_le16(1);
3329         else
3330                 xh->xh_num_buckets = 0;
3331
3332         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3333
3334         /* store the first_hash of the new bucket. */
3335         if (first_hash)
3336                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3337
3338         /*
3339          * Now only update the 1st block of the old bucket.  If we
3340          * just added a new empty bucket, there is no need to modify
3341          * it.
3342          */
3343         if (start == count)
3344                 goto out;
3345
3346         xh = bucket_xh(s_bucket);
3347         memset(&xh->xh_entries[start], 0,
3348                sizeof(struct ocfs2_xattr_entry) * (count - start));
3349         xh->xh_count = cpu_to_le16(start);
3350         xh->xh_free_start = cpu_to_le16(name_offset);
3351         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3352
3353         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3354
3355 out:
3356         ocfs2_xattr_bucket_free(s_bucket);
3357         ocfs2_xattr_bucket_free(t_bucket);
3358
3359         return ret;
3360 }
3361
3362 /*
3363  * Copy xattr from one bucket to another bucket.
3364  *
3365  * The caller must make sure that the journal transaction
3366  * has enough space for journaling.
3367  */
3368 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3369                                  handle_t *handle,
3370                                  u64 s_blkno,
3371                                  u64 t_blkno,
3372                                  int t_is_new)
3373 {
3374         int ret;
3375         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3376
3377         BUG_ON(s_blkno == t_blkno);
3378
3379         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3380              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3381              t_is_new);
3382
3383         s_bucket = ocfs2_xattr_bucket_new(inode);
3384         t_bucket = ocfs2_xattr_bucket_new(inode);
3385         if (!s_bucket || !t_bucket) {
3386                 ret = -ENOMEM;
3387                 mlog_errno(ret);
3388                 goto out;
3389         }
3390   
3391         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3392         if (ret)
3393                 goto out;
3394
3395         /*
3396          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3397          * there's no need to read it.
3398          */
3399         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3400         if (ret)
3401                 goto out;
3402
3403         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3404                                                 t_is_new ?
3405                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3406                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3407         if (ret)
3408                 goto out;
3409
3410         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3411         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3412
3413 out:
3414         ocfs2_xattr_bucket_free(t_bucket);
3415         ocfs2_xattr_bucket_free(s_bucket);
3416
3417         return ret;
3418 }
3419
3420 /*
3421  * Copy one xattr cluster from src_blk to to_blk.
3422  * The to_blk will become the first bucket header of the cluster, so its
3423  * xh_num_buckets will be initialized as the bucket num in the cluster.
3424  */
3425 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3426                                   handle_t *handle,
3427                                   struct buffer_head *first_bh,
3428                                   u64 src_blk,
3429                                   u64 to_blk,
3430                                   u32 *first_hash)
3431 {
3432         int i, ret, credits;
3433         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3434         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3435         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3436         struct buffer_head *bh = NULL;
3437         struct ocfs2_xattr_header *xh;
3438         u64 to_blk_start = to_blk;
3439
3440         mlog(0, "cp xattrs from cluster %llu to %llu\n",
3441              (unsigned long long)src_blk, (unsigned long long)to_blk);
3442
3443         /*
3444          * We need to update the new cluster and 1 more for the update of
3445          * the 1st bucket of the previous extent rec.
3446          */
3447         credits = bpc + 1;
3448         ret = ocfs2_extend_trans(handle, credits);
3449         if (ret) {
3450                 mlog_errno(ret);
3451                 goto out;
3452         }
3453
3454         ret = ocfs2_journal_access(handle, inode, first_bh,
3455                                    OCFS2_JOURNAL_ACCESS_WRITE);
3456         if (ret) {
3457                 mlog_errno(ret);
3458                 goto out;
3459         }
3460
3461         for (i = 0; i < num_buckets; i++) {
3462                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3463                                             src_blk, to_blk, 1);
3464                 if (ret) {
3465                         mlog_errno(ret);
3466                         goto out;
3467                 }
3468
3469                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3470                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3471         }
3472
3473         /* update the old bucket header. */
3474         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3475         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3476
3477         ocfs2_journal_dirty(handle, first_bh);
3478
3479         /* update the new bucket header. */
3480         ret = ocfs2_read_block(inode, to_blk_start, &bh);
3481         if (ret < 0) {
3482                 mlog_errno(ret);
3483                 goto out;
3484         }
3485
3486         ret = ocfs2_journal_access(handle, inode, bh,
3487                                    OCFS2_JOURNAL_ACCESS_WRITE);
3488         if (ret) {
3489                 mlog_errno(ret);
3490                 goto out;
3491         }
3492
3493         xh = (struct ocfs2_xattr_header *)bh->b_data;
3494         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3495
3496         ocfs2_journal_dirty(handle, bh);
3497
3498         if (first_hash)
3499                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3500 out:
3501         brelse(bh);
3502         return ret;
3503 }
3504
3505 /*
3506  * Move some xattrs in this cluster to the new cluster.
3507  * This function should only be called when bucket size == cluster size.
3508  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3509  */
3510 static int ocfs2_divide_xattr_cluster(struct inode *inode,
3511                                       handle_t *handle,
3512                                       u64 prev_blk,
3513                                       u64 new_blk,
3514                                       u32 *first_hash)
3515 {
3516         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3517         int ret, credits = 2 * blk_per_bucket;
3518
3519         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3520
3521         ret = ocfs2_extend_trans(handle, credits);
3522         if (ret) {
3523                 mlog_errno(ret);
3524                 return ret;
3525         }
3526
3527         /* Move half of the xattr in start_blk to the next bucket. */
3528         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
3529                                           new_blk, first_hash, 1);
3530 }
3531
3532 /*
3533  * Move some xattrs from the old cluster to the new one since they are not
3534  * contiguous in ocfs2 xattr tree.
3535  *
3536  * new_blk starts a new separate cluster, and we will move some xattrs from
3537  * prev_blk to it. v_start will be set as the first name hash value in this
3538  * new cluster so that it can be used as e_cpos during tree insertion and
3539  * don't collide with our original b-tree operations. first_bh and header_bh
3540  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3541  * to extend the insert bucket.
3542  *
3543  * The problem is how much xattr should we move to the new one and when should
3544  * we update first_bh and header_bh?
3545  * 1. If cluster size > bucket size, that means the previous cluster has more
3546  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3547  *    update the first_bh and header_bh if the insert bucket has been moved
3548  *    to the new cluster.
3549  * 2. If cluster_size == bucket_size:
3550  *    a) If the previous extent rec has more than one cluster and the insert
3551  *       place isn't in the last cluster, copy the entire last cluster to the
3552  *       new one. This time, we don't need to upate the first_bh and header_bh
3553  *       since they will not be moved into the new cluster.
3554  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3555  *       the new one. And we set the extend flag to zero if the insert place is
3556  *       moved into the new allocated cluster since no extend is needed.
3557  */
3558 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3559                                             handle_t *handle,
3560                                             struct buffer_head **first_bh,
3561                                             struct buffer_head **header_bh,
3562                                             u64 new_blk,
3563                                             u64 prev_blk,
3564                                             u32 prev_clusters,
3565                                             u32 *v_start,
3566                                             int *extend)
3567 {
3568         int ret = 0;
3569         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3570
3571         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3572              (unsigned long long)prev_blk, prev_clusters,
3573              (unsigned long long)new_blk);
3574
3575         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3576                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3577                                                           handle,
3578                                                           first_bh,
3579                                                           header_bh,
3580                                                           new_blk,
3581                                                           prev_blk,
3582                                                           prev_clusters,
3583                                                           v_start);
3584         else {
3585                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3586
3587                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3588                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3589                                                      last_blk, new_blk,
3590                                                      v_start);
3591                 else {
3592                         ret = ocfs2_divide_xattr_cluster(inode, handle,
3593                                                          last_blk, new_blk,
3594                                                          v_start);
3595
3596                         if ((*header_bh)->b_blocknr == last_blk && extend)
3597                                 *extend = 0;
3598                 }
3599         }
3600
3601         return ret;
3602 }
3603
3604 /*
3605  * Add a new cluster for xattr storage.
3606  *
3607  * If the new cluster is contiguous with the previous one, it will be
3608  * appended to the same extent record, and num_clusters will be updated.
3609  * If not, we will insert a new extent for it and move some xattrs in
3610  * the last cluster into the new allocated one.
3611  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3612  * lose the benefits of hashing because we'll have to search large leaves.
3613  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3614  * if it's bigger).
3615  *
3616  * first_bh is the first block of the previous extent rec and header_bh
3617  * indicates the bucket we will insert the new xattrs. They will be updated
3618  * when the header_bh is moved into the new cluster.
3619  */
3620 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3621                                        struct buffer_head *root_bh,
3622                                        struct buffer_head **first_bh,
3623                                        struct buffer_head **header_bh,
3624                                        u32 *num_clusters,
3625                                        u32 prev_cpos,
3626                                        u64 prev_blkno,
3627                                        int *extend)
3628 {
3629         int ret, credits;
3630         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3631         u32 prev_clusters = *num_clusters;
3632         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3633         u64 block;
3634         handle_t *handle = NULL;
3635         struct ocfs2_alloc_context *data_ac = NULL;
3636         struct ocfs2_alloc_context *meta_ac = NULL;
3637         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3638         struct ocfs2_extent_tree et;
3639
3640         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3641              "previous xattr blkno = %llu\n",
3642              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3643              prev_cpos, (unsigned long long)prev_blkno);
3644
3645         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3646
3647         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3648                                     &data_ac, &meta_ac);
3649         if (ret) {
3650                 mlog_errno(ret);
3651                 goto leave;
3652         }
3653
3654         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3655                                             clusters_to_add);
3656         handle = ocfs2_start_trans(osb, credits);
3657         if (IS_ERR(handle)) {
3658                 ret = PTR_ERR(handle);
3659                 handle = NULL;
3660                 mlog_errno(ret);
3661                 goto leave;
3662         }
3663
3664         ret = ocfs2_journal_access(handle, inode, root_bh,
3665                                    OCFS2_JOURNAL_ACCESS_WRITE);
3666         if (ret < 0) {
3667                 mlog_errno(ret);
3668                 goto leave;
3669         }
3670
3671         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3672                                      clusters_to_add, &bit_off, &num_bits);
3673         if (ret < 0) {
3674                 if (ret != -ENOSPC)
3675                         mlog_errno(ret);
3676                 goto leave;
3677         }
3678
3679         BUG_ON(num_bits > clusters_to_add);
3680
3681         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3682         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3683              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3684
3685         if (prev_blkno + prev_clusters * bpc == block &&
3686             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3687              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3688                 /*
3689                  * If this cluster is contiguous with the old one and
3690                  * adding this new cluster, we don't surpass the limit of
3691                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3692                  * initialized and used like other buckets in the previous
3693                  * cluster.
3694                  * So add it as a contiguous one. The caller will handle
3695                  * its init process.
3696                  */
3697                 v_start = prev_cpos + prev_clusters;
3698                 *num_clusters = prev_clusters + num_bits;
3699                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3700                      num_bits);
3701         } else {
3702                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3703                                                        handle,
3704                                                        first_bh,
3705                                                        header_bh,
3706                                                        block,
3707                                                        prev_blkno,
3708                                                        prev_clusters,
3709                                                        &v_start,
3710                                                        extend);
3711                 if (ret) {
3712                         mlog_errno(ret);
3713                         goto leave;
3714                 }
3715         }
3716
3717         if (handle->h_buffer_credits < credits) {
3718                 /*
3719                  * The journal has been restarted before, and don't
3720                  * have enough space for the insertion, so extend it
3721                  * here.
3722                  */
3723                 ret = ocfs2_extend_trans(handle, credits);
3724                 if (ret) {
3725                         mlog_errno(ret);
3726                         goto leave;
3727                 }
3728         }
3729         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3730              num_bits, (unsigned long long)block, v_start);
3731         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3732                                   num_bits, 0, meta_ac);
3733         if (ret < 0) {
3734                 mlog_errno(ret);
3735                 goto leave;
3736         }
3737
3738         ret = ocfs2_journal_dirty(handle, root_bh);
3739         if (ret < 0) {
3740                 mlog_errno(ret);
3741                 goto leave;
3742         }
3743
3744 leave:
3745         if (handle)
3746                 ocfs2_commit_trans(osb, handle);
3747         if (data_ac)
3748                 ocfs2_free_alloc_context(data_ac);
3749         if (meta_ac)
3750                 ocfs2_free_alloc_context(meta_ac);
3751
3752         return ret;
3753 }
3754
3755 /*
3756  * Extend a new xattr bucket and move xattrs to the end one by one until
3757  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3758  */
3759 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3760                                      struct buffer_head *first_bh,
3761                                      struct buffer_head *start_bh,
3762                                      u32 num_clusters)
3763 {
3764         int ret, credits;
3765         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3766         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3767         u64 start_blk = start_bh->b_blocknr, end_blk;
3768         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3769         handle_t *handle;
3770         struct ocfs2_xattr_header *first_xh =
3771                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3772         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3773
3774         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3775              "from %llu, len = %u\n", (unsigned long long)start_blk,
3776              (unsigned long long)first_bh->b_blocknr, num_clusters);
3777
3778         BUG_ON(bucket >= num_buckets);
3779
3780         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3781
3782         /*
3783          * We will touch all the buckets after the start_bh(include it).
3784          * Then we add one more bucket.
3785          */
3786         credits = end_blk - start_blk + 3 * blk_per_bucket + 1;
3787         handle = ocfs2_start_trans(osb, credits);
3788         if (IS_ERR(handle)) {
3789                 ret = PTR_ERR(handle);
3790                 handle = NULL;
3791                 mlog_errno(ret);
3792                 goto out;
3793         }
3794
3795         ret = ocfs2_journal_access(handle, inode, first_bh,
3796                                    OCFS2_JOURNAL_ACCESS_WRITE);
3797         if (ret) {
3798                 mlog_errno(ret);
3799                 goto commit;
3800         }
3801
3802         while (end_blk != start_blk) {
3803                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3804                                             end_blk + blk_per_bucket, 0);
3805                 if (ret)
3806                         goto commit;
3807                 end_blk -= blk_per_bucket;
3808         }
3809
3810         /* Move half of the xattr in start_blk to the next bucket. */
3811         ret = ocfs2_divide_xattr_bucket(inode, handle, start_blk,
3812                                         start_blk + blk_per_bucket, NULL, 0);
3813
3814         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3815         ocfs2_journal_dirty(handle, first_bh);
3816
3817 commit:
3818         ocfs2_commit_trans(osb, handle);
3819 out:
3820         return ret;
3821 }
3822
3823 /*
3824  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3825  * xb_bh is the ocfs2_xattr_block.
3826  * We will move all the buckets starting from header_bh to the next place. As
3827  * for this one, half num of its xattrs will be moved to the next one.
3828  *
3829  * We will allocate a new cluster if current cluster is full and adjust
3830  * header_bh and first_bh if the insert place is moved to the new cluster.
3831  */
3832 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3833                                       struct buffer_head *xb_bh,
3834                                       struct buffer_head *header_bh)
3835 {
3836         struct ocfs2_xattr_header *first_xh = NULL;
3837         struct buffer_head *first_bh = NULL;
3838         struct ocfs2_xattr_block *xb =
3839                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3840         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3841         struct ocfs2_extent_list *el = &xb_root->xt_list;
3842         struct ocfs2_xattr_header *xh =
3843                         (struct ocfs2_xattr_header *)header_bh->b_data;
3844         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3845         struct super_block *sb = inode->i_sb;
3846         struct ocfs2_super *osb = OCFS2_SB(sb);
3847         int ret, num_buckets, extend = 1;
3848         u64 p_blkno;
3849         u32 e_cpos, num_clusters;
3850
3851         mlog(0, "Add new xattr bucket starting form %llu\n",
3852              (unsigned long long)header_bh->b_blocknr);
3853
3854         /*
3855          * Add refrence for header_bh here because it may be
3856          * changed in ocfs2_add_new_xattr_cluster and we need
3857          * to free it in the end.
3858          */
3859         get_bh(header_bh);
3860
3861         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3862                                   &num_clusters, el);
3863         if (ret) {
3864                 mlog_errno(ret);
3865                 goto out;
3866         }
3867
3868         ret = ocfs2_read_block(inode, p_blkno, &first_bh);
3869         if (ret) {
3870                 mlog_errno(ret);
3871                 goto out;
3872         }
3873
3874         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3875         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3876
3877         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3878                 ret = ocfs2_add_new_xattr_cluster(inode,
3879                                                   xb_bh,
3880                                                   &first_bh,
3881                                                   &header_bh,
3882                                                   &num_clusters,
3883                                                   e_cpos,
3884                                                   p_blkno,
3885                                                   &extend);
3886                 if (ret) {
3887                         mlog_errno(ret);
3888                         goto out;
3889                 }
3890         }
3891
3892         if (extend)
3893                 ret = ocfs2_extend_xattr_bucket(inode,
3894                                                 first_bh,
3895                                                 header_bh,
3896                                                 num_clusters);
3897         if (ret)
3898                 mlog_errno(ret);
3899 out:
3900         brelse(first_bh);
3901         brelse(header_bh);
3902         return ret;
3903 }
3904
3905 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3906                                         struct ocfs2_xattr_bucket *bucket,
3907                                         int offs)
3908 {
3909         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3910
3911         offs = offs % inode->i_sb->s_blocksize;
3912         return bucket_block(bucket, block_off) + offs;
3913 }
3914
3915 /*
3916  * Handle the normal xattr set, including replace, delete and new.
3917  *
3918  * Note: "local" indicates the real data's locality. So we can't
3919  * just its bucket locality by its length.
3920  */
3921 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3922                                          struct ocfs2_xattr_info *xi,
3923                                          struct ocfs2_xattr_search *xs,
3924                                          u32 name_hash,
3925                                          int local)
3926 {
3927         struct ocfs2_xattr_entry *last, *xe;
3928         int name_len = strlen(xi->name);
3929         struct ocfs2_xattr_header *xh = xs->header;
3930         u16 count = le16_to_cpu(xh->xh_count), start;
3931         size_t blocksize = inode->i_sb->s_blocksize;
3932         char *val;
3933         size_t offs, size, new_size;
3934
3935         last = &xh->xh_entries[count];
3936         if (!xs->not_found) {
3937                 xe = xs->here;
3938                 offs = le16_to_cpu(xe->xe_name_offset);
3939                 if (ocfs2_xattr_is_local(xe))
3940                         size = OCFS2_XATTR_SIZE(name_len) +
3941                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3942                 else
3943                         size = OCFS2_XATTR_SIZE(name_len) +
3944                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3945
3946                 /*
3947                  * If the new value will be stored outside, xi->value has been
3948                  * initalized as an empty ocfs2_xattr_value_root, and the same
3949                  * goes with xi->value_len, so we can set new_size safely here.
3950                  * See ocfs2_xattr_set_in_bucket.
3951                  */
3952                 new_size = OCFS2_XATTR_SIZE(name_len) +
3953                            OCFS2_XATTR_SIZE(xi->value_len);
3954
3955                 le16_add_cpu(&xh->xh_name_value_len, -size);
3956                 if (xi->value) {
3957                         if (new_size > size)
3958                                 goto set_new_name_value;
3959
3960                         /* Now replace the old value with new one. */
3961                         if (local)
3962                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3963                         else
3964                                 xe->xe_value_size = 0;
3965
3966                         val = ocfs2_xattr_bucket_get_val(inode,
3967                                                          xs->bucket, offs);
3968                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3969                                size - OCFS2_XATTR_SIZE(name_len));
3970                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3971                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3972                                        xi->value, xi->value_len);
3973
3974                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3975                         ocfs2_xattr_set_local(xe, local);
3976                         return;
3977                 } else {
3978                         /*
3979                          * Remove the old entry if there is more than one.
3980                          * We don't remove the last entry so that we can
3981                          * use it to indicate the hash value of the empty
3982                          * bucket.
3983                          */
3984                         last -= 1;
3985                         le16_add_cpu(&xh->xh_count, -1);
3986                         if (xh->xh_count) {
3987                                 memmove(xe, xe + 1,
3988                                         (void *)last - (void *)xe);
3989                                 memset(last, 0,
3990                                        sizeof(struct ocfs2_xattr_entry));
3991                         } else
3992                                 xh->xh_free_start =
3993                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3994
3995                         return;
3996                 }
3997         } else {
3998                 /* find a new entry for insert. */
3999                 int low = 0, high = count - 1, tmp;
4000                 struct ocfs2_xattr_entry *tmp_xe;
4001
4002                 while (low <= high && count) {
4003                         tmp = (low + high) / 2;
4004                         tmp_xe = &xh->xh_entries[tmp];
4005
4006                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4007                                 low = tmp + 1;
4008                         else if (name_hash <
4009                                  le32_to_cpu(tmp_xe->xe_name_hash))
4010                                 high = tmp - 1;
4011                         else {
4012                                 low = tmp;
4013                                 break;
4014                         }
4015                 }
4016
4017                 xe = &xh->xh_entries[low];
4018                 if (low != count)
4019                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4020
4021                 le16_add_cpu(&xh->xh_count, 1);
4022                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4023                 xe->xe_name_hash = cpu_to_le32(name_hash);
4024                 xe->xe_name_len = name_len;
4025                 ocfs2_xattr_set_type(xe, xi->name_index);
4026         }
4027
4028 set_new_name_value:
4029         /* Insert the new name+value. */
4030         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4031
4032         /*
4033          * We must make sure that the name/value pair
4034          * exists in the same block.
4035          */
4036         offs = le16_to_cpu(xh->xh_free_start);
4037         start = offs - size;
4038
4039         if (start >> inode->i_sb->s_blocksize_bits !=
4040             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4041                 offs = offs - offs % blocksize;
4042                 xh->xh_free_start = cpu_to_le16(offs);
4043         }
4044
4045         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4046         xe->xe_name_offset = cpu_to_le16(offs - size);
4047
4048         memset(val, 0, size);
4049         memcpy(val, xi->name, name_len);
4050         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4051
4052         xe->xe_value_size = cpu_to_le64(xi->value_len);
4053         ocfs2_xattr_set_local(xe, local);
4054         xs->here = xe;
4055         le16_add_cpu(&xh->xh_free_start, -size);
4056         le16_add_cpu(&xh->xh_name_value_len, size);
4057
4058         return;
4059 }
4060
4061 /*
4062  * Set the xattr entry in the specified bucket.
4063  * The bucket is indicated by xs->bucket and it should have the enough
4064  * space for the xattr insertion.
4065  */
4066 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4067                                            struct ocfs2_xattr_info *xi,
4068                                            struct ocfs2_xattr_search *xs,
4069                                            u32 name_hash,
4070                                            int local)
4071 {
4072         int ret;
4073         handle_t *handle = NULL;
4074         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4075         u64 blkno;
4076
4077         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4078              (unsigned long)xi->value_len, xi->name_index,
4079              (unsigned long long)bucket_blkno(xs->bucket));
4080
4081         if (!xs->bucket->bu_bhs[1]) {
4082                 blkno = bucket_blkno(xs->bucket);
4083                 ocfs2_xattr_bucket_relse(xs->bucket);
4084                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4085                 if (ret) {
4086                         mlog_errno(ret);
4087                         goto out;
4088                 }
4089         }
4090
4091         handle = ocfs2_start_trans(osb, xs->bucket->bu_blocks);
4092         if (IS_ERR(handle)) {
4093                 ret = PTR_ERR(handle);
4094                 handle = NULL;
4095                 mlog_errno(ret);
4096                 goto out;
4097         }
4098
4099         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4100                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4101         if (ret < 0) {
4102                 mlog_errno(ret);
4103                 goto out;
4104         }
4105
4106         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4107         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4108
4109 out:
4110         ocfs2_commit_trans(osb, handle);
4111
4112         return ret;
4113 }
4114
4115 static int ocfs2_xattr_value_update_size(struct inode *inode,
4116                                          struct buffer_head *xe_bh,
4117                                          struct ocfs2_xattr_entry *xe,
4118                                          u64 new_size)
4119 {
4120         int ret;
4121         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4122         handle_t *handle = NULL;
4123
4124         handle = ocfs2_start_trans(osb, 1);
4125         if (IS_ERR(handle)) {
4126                 ret = -ENOMEM;
4127                 mlog_errno(ret);
4128                 goto out;
4129         }
4130
4131         ret = ocfs2_journal_access(handle, inode, xe_bh,
4132                                    OCFS2_JOURNAL_ACCESS_WRITE);
4133         if (ret < 0) {
4134                 mlog_errno(ret);
4135                 goto out_commit;
4136         }
4137
4138         xe->xe_value_size = cpu_to_le64(new_size);
4139
4140         ret = ocfs2_journal_dirty(handle, xe_bh);
4141         if (ret < 0)
4142                 mlog_errno(ret);
4143
4144 out_commit:
4145         ocfs2_commit_trans(osb, handle);
4146 out:
4147         return ret;
4148 }
4149
4150 /*
4151  * Truncate the specified xe_off entry in xattr bucket.
4152  * bucket is indicated by header_bh and len is the new length.
4153  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4154  *
4155  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4156  */
4157 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4158                                              struct buffer_head *header_bh,
4159                                              int xe_off,
4160                                              int len)
4161 {
4162         int ret, offset;
4163         u64 value_blk;
4164         struct buffer_head *value_bh = NULL;
4165         struct ocfs2_xattr_value_root *xv;
4166         struct ocfs2_xattr_entry *xe;
4167         struct ocfs2_xattr_header *xh =
4168                         (struct ocfs2_xattr_header *)header_bh->b_data;
4169         size_t blocksize = inode->i_sb->s_blocksize;
4170
4171         xe = &xh->xh_entries[xe_off];
4172
4173         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4174
4175         offset = le16_to_cpu(xe->xe_name_offset) +
4176                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4177
4178         value_blk = offset / blocksize;
4179
4180         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4181         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4182         value_blk += header_bh->b_blocknr;
4183
4184         ret = ocfs2_read_block(inode, value_blk, &value_bh);
4185         if (ret) {
4186                 mlog_errno(ret);
4187                 goto out;
4188         }
4189
4190         xv = (struct ocfs2_xattr_value_root *)
4191                 (value_bh->b_data + offset % blocksize);
4192
4193         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4194              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4195         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4196         if (ret) {
4197                 mlog_errno(ret);
4198                 goto out;
4199         }
4200
4201         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4202         if (ret) {
4203                 mlog_errno(ret);
4204                 goto out;
4205         }
4206
4207 out:
4208         brelse(value_bh);
4209         return ret;
4210 }
4211
4212 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4213                                                 struct ocfs2_xattr_search *xs,
4214                                                 int len)
4215 {
4216         int ret, offset;
4217         struct ocfs2_xattr_entry *xe = xs->here;
4218         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4219
4220         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4221
4222         offset = xe - xh->xh_entries;
4223         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket->bu_bhs[0],
4224                                                 offset, len);
4225         if (ret)
4226                 mlog_errno(ret);
4227
4228         return ret;
4229 }
4230
4231 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4232                                                 struct ocfs2_xattr_search *xs,
4233                                                 char *val,
4234                                                 int value_len)
4235 {
4236         int offset;
4237         struct ocfs2_xattr_value_root *xv;
4238         struct ocfs2_xattr_entry *xe = xs->here;
4239
4240         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4241
4242         offset = le16_to_cpu(xe->xe_name_offset) +
4243                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4244
4245         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4246
4247         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4248 }
4249
4250 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4251                                   struct buffer_head *root_bh,
4252                                   u64 blkno,
4253                                   u32 cpos,
4254                                   u32 len)
4255 {
4256         int ret;
4257         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4258         struct inode *tl_inode = osb->osb_tl_inode;
4259         handle_t *handle;
4260         struct ocfs2_xattr_block *xb =
4261                         (struct ocfs2_xattr_block *)root_bh->b_data;
4262         struct ocfs2_alloc_context *meta_ac = NULL;
4263         struct ocfs2_cached_dealloc_ctxt dealloc;
4264         struct ocfs2_extent_tree et;
4265
4266         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4267
4268         ocfs2_init_dealloc_ctxt(&dealloc);
4269
4270         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4271              cpos, len, (unsigned long long)blkno);
4272
4273         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4274
4275         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4276         if (ret) {
4277                 mlog_errno(ret);
4278                 return ret;
4279         }
4280
4281         mutex_lock(&tl_inode->i_mutex);
4282
4283         if (ocfs2_truncate_log_needs_flush(osb)) {
4284                 ret = __ocfs2_flush_truncate_log(osb);
4285                 if (ret < 0) {
4286                         mlog_errno(ret);
4287                         goto out;
4288                 }
4289         }
4290
4291         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4292         if (IS_ERR(handle)) {
4293                 ret = -ENOMEM;
4294                 mlog_errno(ret);
4295                 goto out;
4296         }
4297
4298         ret = ocfs2_journal_access(handle, inode, root_bh,
4299                                    OCFS2_JOURNAL_ACCESS_WRITE);
4300         if (ret) {
4301                 mlog_errno(ret);
4302                 goto out_commit;
4303         }
4304
4305         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4306                                   &dealloc);
4307         if (ret) {
4308                 mlog_errno(ret);
4309                 goto out_commit;
4310         }
4311
4312         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4313
4314         ret = ocfs2_journal_dirty(handle, root_bh);
4315         if (ret) {
4316                 mlog_errno(ret);
4317                 goto out_commit;
4318         }
4319
4320         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4321         if (ret)
4322                 mlog_errno(ret);
4323
4324 out_commit:
4325         ocfs2_commit_trans(osb, handle);
4326 out:
4327         ocfs2_schedule_truncate_log_flush(osb, 1);
4328
4329         mutex_unlock(&tl_inode->i_mutex);
4330
4331         if (meta_ac)
4332                 ocfs2_free_alloc_context(meta_ac);
4333
4334         ocfs2_run_deallocs(osb, &dealloc);
4335
4336         return ret;
4337 }
4338
4339 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4340                                          struct ocfs2_xattr_search *xs)
4341 {
4342         handle_t *handle = NULL;
4343         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4344         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4345                                                 le16_to_cpu(xh->xh_count) - 1];
4346         int ret = 0;
4347
4348         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
4349                                    ocfs2_blocks_per_xattr_bucket(inode->i_sb));
4350         if (IS_ERR(handle)) {
4351                 ret = PTR_ERR(handle);
4352                 mlog_errno(ret);
4353                 return;
4354         }
4355
4356         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4357                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4358         if (ret) {
4359                 mlog_errno(ret);
4360                 goto out_commit;
4361         }
4362
4363         /* Remove the old entry. */
4364         memmove(xs->here, xs->here + 1,
4365                 (void *)last - (void *)xs->here);
4366         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4367         le16_add_cpu(&xh->xh_count, -1);
4368
4369         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4370
4371 out_commit:
4372         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4373 }
4374
4375 /*
4376  * Set the xattr name/value in the bucket specified in xs.
4377  *
4378  * As the new value in xi may be stored in the bucket or in an outside cluster,
4379  * we divide the whole process into 3 steps:
4380  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4381  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4382  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4383  * 4. If the clusters for the new outside value can't be allocated, we need
4384  *    to free the xattr we allocated in set.
4385  */
4386 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4387                                      struct ocfs2_xattr_info *xi,
4388                                      struct ocfs2_xattr_search *xs)
4389 {
4390         int ret, local = 1;
4391         size_t value_len;
4392         char *val = (char *)xi->value;
4393         struct ocfs2_xattr_entry *xe = xs->here;
4394         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4395                                               strlen(xi->name));
4396
4397         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4398                 /*
4399                  * We need to truncate the xattr storage first.
4400                  *
4401                  * If both the old and new value are stored to
4402                  * outside block, we only need to truncate
4403                  * the storage and then set the value outside.
4404                  *
4405                  * If the new value should be stored within block,
4406                  * we should free all the outside block first and
4407                  * the modification to the xattr block will be done
4408                  * by following steps.
4409                  */
4410                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4411                         value_len = xi->value_len;
4412                 else
4413                         value_len = 0;
4414
4415                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4416                                                            value_len);
4417                 if (ret)
4418                         goto out;
4419
4420                 if (value_len)
4421                         goto set_value_outside;
4422         }
4423
4424         value_len = xi->value_len;
4425         /* So we have to handle the inside block change now. */
4426         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4427                 /*
4428                  * If the new value will be stored outside of block,
4429                  * initalize a new empty value root and insert it first.
4430                  */
4431                 local = 0;
4432                 xi->value = &def_xv;
4433                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4434         }
4435
4436         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4437         if (ret) {
4438                 mlog_errno(ret);
4439                 goto out;
4440         }
4441
4442         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4443                 goto out;
4444
4445         /* allocate the space now for the outside block storage. */
4446         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4447                                                    value_len);
4448         if (ret) {
4449                 mlog_errno(ret);
4450
4451                 if (xs->not_found) {
4452                         /*
4453                          * We can't allocate enough clusters for outside
4454                          * storage and we have allocated xattr already,
4455                          * so need to remove it.
4456                          */
4457                         ocfs2_xattr_bucket_remove_xs(inode, xs);
4458                 }
4459                 goto out;
4460         }
4461
4462 set_value_outside:
4463         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4464 out:
4465         return ret;
4466 }
4467
4468 /*
4469  * check whether the xattr bucket is filled up with the same hash value.
4470  * If we want to insert the xattr with the same hash, return -ENOSPC.
4471  * If we want to insert a xattr with different hash value, go ahead
4472  * and ocfs2_divide_xattr_bucket will handle this.
4473  */
4474 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4475                                               struct ocfs2_xattr_bucket *bucket,
4476                                               const char *name)
4477 {
4478         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4479         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4480
4481         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4482                 return 0;
4483
4484         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4485             xh->xh_entries[0].xe_name_hash) {
4486                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4487                      "hash = %u\n",
4488                      (unsigned long long)bucket_blkno(bucket),
4489                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4490                 return -ENOSPC;
4491         }
4492
4493         return 0;
4494 }
4495
4496 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4497                                              struct ocfs2_xattr_info *xi,
4498                                              struct ocfs2_xattr_search *xs)
4499 {
4500         struct ocfs2_xattr_header *xh;
4501         struct ocfs2_xattr_entry *xe;
4502         u16 count, header_size, xh_free_start;
4503         int free, max_free, need, old;
4504         size_t value_size = 0, name_len = strlen(xi->name);
4505         size_t blocksize = inode->i_sb->s_blocksize;
4506         int ret, allocation = 0;
4507
4508         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4509
4510 try_again:
4511         xh = xs->header;
4512         count = le16_to_cpu(xh->xh_count);
4513         xh_free_start = le16_to_cpu(xh->xh_free_start);
4514         header_size = sizeof(struct ocfs2_xattr_header) +
4515                         count * sizeof(struct ocfs2_xattr_entry);
4516         max_free = OCFS2_XATTR_BUCKET_SIZE -
4517                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4518
4519         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4520                         "of %u which exceed block size\n",
4521                         (unsigned long long)bucket_blkno(xs->bucket),
4522                         header_size);
4523
4524         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4525                 value_size = OCFS2_XATTR_ROOT_SIZE;
4526         else if (xi->value)
4527                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4528
4529         if (xs->not_found)
4530                 need = sizeof(struct ocfs2_xattr_entry) +
4531                         OCFS2_XATTR_SIZE(name_len) + value_size;
4532         else {
4533                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4534
4535                 /*
4536                  * We only replace the old value if the new length is smaller
4537                  * than the old one. Otherwise we will allocate new space in the
4538                  * bucket to store it.
4539                  */
4540                 xe = xs->here;
4541                 if (ocfs2_xattr_is_local(xe))
4542                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4543                 else
4544                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4545
4546                 if (old >= value_size)
4547                         need = 0;
4548         }
4549
4550         free = xh_free_start - header_size;
4551         /*
4552          * We need to make sure the new name/value pair
4553          * can exist in the same block.
4554          */
4555         if (xh_free_start % blocksize < need)
4556                 free -= xh_free_start % blocksize;
4557
4558         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4559              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4560              " %u\n", xs->not_found,
4561              (unsigned long long)bucket_blkno(xs->bucket),
4562              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4563              le16_to_cpu(xh->xh_name_value_len));
4564
4565         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4566                 if (need <= max_free &&
4567                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4568                         /*
4569                          * We can create the space by defragment. Since only the
4570                          * name/value will be moved, the xe shouldn't be changed
4571                          * in xs.
4572                          */
4573                         ret = ocfs2_defrag_xattr_bucket(inode, xs->bucket);
4574                         if (ret) {
4575                                 mlog_errno(ret);
4576                                 goto out;
4577                         }
4578
4579                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4580                         free = xh_free_start - header_size;
4581                         if (xh_free_start % blocksize < need)
4582                                 free -= xh_free_start % blocksize;
4583
4584                         if (free >= need)
4585                                 goto xattr_set;
4586
4587                         mlog(0, "Can't get enough space for xattr insert by "
4588                              "defragment. Need %u bytes, but we have %d, so "
4589                              "allocate new bucket for it.\n", need, free);
4590                 }
4591
4592                 /*
4593                  * We have to add new buckets or clusters and one
4594                  * allocation should leave us enough space for insert.
4595                  */
4596                 BUG_ON(allocation);
4597
4598                 /*
4599                  * We do not allow for overlapping ranges between buckets. And
4600                  * the maximum number of collisions we will allow for then is
4601                  * one bucket's worth, so check it here whether we need to
4602                  * add a new bucket for the insert.
4603                  */
4604                 ret = ocfs2_check_xattr_bucket_collision(inode,
4605                                                          xs->bucket,
4606                                                          xi->name);
4607                 if (ret) {
4608                         mlog_errno(ret);
4609                         goto out;
4610                 }
4611
4612                 ret = ocfs2_add_new_xattr_bucket(inode,
4613                                                  xs->xattr_bh,
4614                                                  xs->bucket->bu_bhs[0]);
4615                 if (ret) {
4616                         mlog_errno(ret);
4617                         goto out;
4618                 }
4619
4620                 ocfs2_xattr_bucket_relse(xs->bucket);
4621
4622                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4623                                                    xi->name_index,
4624                                                    xi->name, xs);
4625                 if (ret && ret != -ENODATA)
4626                         goto out;
4627                 xs->not_found = ret;
4628                 allocation = 1;
4629                 goto try_again;
4630         }
4631
4632 xattr_set:
4633         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4634 out:
4635         mlog_exit(ret);
4636         return ret;
4637 }
4638
4639 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4640                                         struct ocfs2_xattr_bucket *bucket,
4641                                         void *para)
4642 {
4643         int ret = 0;
4644         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4645         u16 i;
4646         struct ocfs2_xattr_entry *xe;
4647
4648         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4649                 xe = &xh->xh_entries[i];
4650                 if (ocfs2_xattr_is_local(xe))
4651                         continue;
4652
4653                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4654                                                         bucket->bu_bhs[0],
4655                                                         i, 0);
4656                 if (ret) {
4657                         mlog_errno(ret);
4658                         break;
4659                 }
4660         }
4661
4662         return ret;
4663 }
4664
4665 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4666                                           struct buffer_head *xb_bh)
4667 {
4668         struct ocfs2_xattr_block *xb =
4669                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4670         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4671         int ret = 0;
4672         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4673         u64 p_blkno;
4674
4675         if (le16_to_cpu(el->l_next_free_rec) == 0)
4676                 return 0;
4677
4678         while (name_hash > 0) {
4679                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4680                                           &e_cpos, &num_clusters, el);
4681                 if (ret) {
4682                         mlog_errno(ret);
4683                         goto out;
4684                 }
4685
4686                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4687                                                   ocfs2_delete_xattr_in_bucket,
4688                                                   NULL);
4689                 if (ret) {
4690                         mlog_errno(ret);
4691                         goto out;
4692                 }
4693
4694                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4695                                              p_blkno, e_cpos, num_clusters);
4696                 if (ret) {
4697                         mlog_errno(ret);
4698                         break;
4699                 }
4700
4701                 if (e_cpos == 0)
4702                         break;
4703
4704                 name_hash = e_cpos - 1;
4705         }
4706
4707 out:
4708         return ret;
4709 }
4710
4711 /*
4712  * 'trusted' attributes support
4713  */
4714 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4715                                        size_t list_size, const char *name,
4716                                        size_t name_len)
4717 {
4718         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
4719         const size_t total_len = prefix_len + name_len + 1;
4720
4721         if (list && total_len <= list_size) {
4722                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4723                 memcpy(list + prefix_len, name, name_len);
4724                 list[prefix_len + name_len] = '\0';
4725         }
4726         return total_len;
4727 }
4728
4729 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4730                                    void *buffer, size_t size)
4731 {
4732         if (strcmp(name, "") == 0)
4733                 return -EINVAL;
4734         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4735                                buffer, size);
4736 }
4737
4738 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4739                                    const void *value, size_t size, int flags)
4740 {
4741         if (strcmp(name, "") == 0)
4742                 return -EINVAL;
4743
4744         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4745                                size, flags);
4746 }
4747
4748 struct xattr_handler ocfs2_xattr_trusted_handler = {
4749         .prefix = XATTR_TRUSTED_PREFIX,
4750         .list   = ocfs2_xattr_trusted_list,
4751         .get    = ocfs2_xattr_trusted_get,
4752         .set    = ocfs2_xattr_trusted_set,
4753 };
4754
4755 /*
4756  * 'user' attributes support
4757  */
4758 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4759                                     size_t list_size, const char *name,
4760                                     size_t name_len)
4761 {
4762         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
4763         const size_t total_len = prefix_len + name_len + 1;
4764         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4765
4766         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4767                 return 0;
4768
4769         if (list && total_len <= list_size) {
4770                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4771                 memcpy(list + prefix_len, name, name_len);
4772                 list[prefix_len + name_len] = '\0';
4773         }
4774         return total_len;
4775 }
4776
4777 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4778                                 void *buffer, size_t size)
4779 {
4780         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4781
4782         if (strcmp(name, "") == 0)
4783                 return -EINVAL;
4784         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4785                 return -EOPNOTSUPP;
4786         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4787                                buffer, size);
4788 }
4789
4790 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4791                                 const void *value, size_t size, int flags)
4792 {
4793         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4794
4795         if (strcmp(name, "") == 0)
4796                 return -EINVAL;
4797         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4798                 return -EOPNOTSUPP;
4799
4800         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4801                                size, flags);
4802 }
4803
4804 struct xattr_handler ocfs2_xattr_user_handler = {
4805         .prefix = XATTR_USER_PREFIX,
4806         .list   = ocfs2_xattr_user_list,
4807         .get    = ocfs2_xattr_user_get,
4808         .set    = ocfs2_xattr_user_set,
4809 };