f3ea7efb48c63a1dff92f8b0e7eb33f12d02a9d9
[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         struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
65         struct ocfs2_xattr_header *xh;
66 };
67
68 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
69 #define OCFS2_XATTR_INLINE_SIZE 80
70
71 static struct ocfs2_xattr_def_value_root def_xv = {
72         .xv.xr_list.l_count = cpu_to_le16(1),
73 };
74
75 struct xattr_handler *ocfs2_xattr_handlers[] = {
76         &ocfs2_xattr_user_handler,
77         &ocfs2_xattr_trusted_handler,
78         NULL
79 };
80
81 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
82         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
83         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
84 };
85
86 struct ocfs2_xattr_info {
87         int name_index;
88         const char *name;
89         const void *value;
90         size_t value_len;
91 };
92
93 struct ocfs2_xattr_search {
94         struct buffer_head *inode_bh;
95         /*
96          * xattr_bh point to the block buffer head which has extended attribute
97          * when extended attribute in inode, xattr_bh is equal to inode_bh.
98          */
99         struct buffer_head *xattr_bh;
100         struct ocfs2_xattr_header *header;
101         struct ocfs2_xattr_bucket bucket;
102         void *base;
103         void *end;
104         struct ocfs2_xattr_entry *here;
105         int not_found;
106 };
107
108 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
109                                              struct ocfs2_xattr_header *xh,
110                                              int index,
111                                              int *block_off,
112                                              int *new_offset);
113
114 static int ocfs2_xattr_block_find(struct inode *inode,
115                                   int name_index,
116                                   const char *name,
117                                   struct ocfs2_xattr_search *xs);
118 static int ocfs2_xattr_index_block_find(struct inode *inode,
119                                         struct buffer_head *root_bh,
120                                         int name_index,
121                                         const char *name,
122                                         struct ocfs2_xattr_search *xs);
123
124 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
125                                         struct ocfs2_xattr_tree_root *xt,
126                                         char *buffer,
127                                         size_t buffer_size);
128
129 static int ocfs2_xattr_create_index_block(struct inode *inode,
130                                           struct ocfs2_xattr_search *xs);
131
132 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
133                                              struct ocfs2_xattr_info *xi,
134                                              struct ocfs2_xattr_search *xs);
135
136 static int ocfs2_delete_xattr_index_block(struct inode *inode,
137                                           struct buffer_head *xb_bh);
138
139 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
140 {
141         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
142 }
143
144 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
145 {
146         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
147 }
148
149 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
150 {
151         u16 len = sb->s_blocksize -
152                  offsetof(struct ocfs2_xattr_header, xh_entries);
153
154         return len / sizeof(struct ocfs2_xattr_entry);
155 }
156
157 static inline const char *ocfs2_xattr_prefix(int name_index)
158 {
159         struct xattr_handler *handler = NULL;
160
161         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
162                 handler = ocfs2_xattr_handler_map[name_index];
163
164         return handler ? handler->prefix : NULL;
165 }
166
167 static u32 ocfs2_xattr_name_hash(struct inode *inode,
168                                  const char *name,
169                                  int name_len)
170 {
171         /* Get hash value of uuid from super block */
172         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
173         int i;
174
175         /* hash extended attribute name */
176         for (i = 0; i < name_len; i++) {
177                 hash = (hash << OCFS2_HASH_SHIFT) ^
178                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
179                        *name++;
180         }
181
182         return hash;
183 }
184
185 /*
186  * ocfs2_xattr_hash_entry()
187  *
188  * Compute the hash of an extended attribute.
189  */
190 static void ocfs2_xattr_hash_entry(struct inode *inode,
191                                    struct ocfs2_xattr_header *header,
192                                    struct ocfs2_xattr_entry *entry)
193 {
194         u32 hash = 0;
195         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
196
197         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
198         entry->xe_name_hash = cpu_to_le32(hash);
199
200         return;
201 }
202
203 static int ocfs2_xattr_extend_allocation(struct inode *inode,
204                                          u32 clusters_to_add,
205                                          struct buffer_head *xattr_bh,
206                                          struct ocfs2_xattr_value_root *xv)
207 {
208         int status = 0;
209         int restart_func = 0;
210         int credits = 0;
211         handle_t *handle = NULL;
212         struct ocfs2_alloc_context *data_ac = NULL;
213         struct ocfs2_alloc_context *meta_ac = NULL;
214         enum ocfs2_alloc_restarted why;
215         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
216         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
217         struct ocfs2_extent_tree et;
218
219         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
220
221         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
222
223 restart_all:
224
225         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
226                                        &data_ac, &meta_ac);
227         if (status) {
228                 mlog_errno(status);
229                 goto leave;
230         }
231
232         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
233                                             clusters_to_add);
234         handle = ocfs2_start_trans(osb, credits);
235         if (IS_ERR(handle)) {
236                 status = PTR_ERR(handle);
237                 handle = NULL;
238                 mlog_errno(status);
239                 goto leave;
240         }
241
242 restarted_transaction:
243         status = ocfs2_journal_access(handle, inode, xattr_bh,
244                                       OCFS2_JOURNAL_ACCESS_WRITE);
245         if (status < 0) {
246                 mlog_errno(status);
247                 goto leave;
248         }
249
250         prev_clusters = le32_to_cpu(xv->xr_clusters);
251         status = ocfs2_add_clusters_in_btree(osb,
252                                              inode,
253                                              &logical_start,
254                                              clusters_to_add,
255                                              0,
256                                              &et,
257                                              handle,
258                                              data_ac,
259                                              meta_ac,
260                                              &why);
261         if ((status < 0) && (status != -EAGAIN)) {
262                 if (status != -ENOSPC)
263                         mlog_errno(status);
264                 goto leave;
265         }
266
267         status = ocfs2_journal_dirty(handle, xattr_bh);
268         if (status < 0) {
269                 mlog_errno(status);
270                 goto leave;
271         }
272
273         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
274
275         if (why != RESTART_NONE && clusters_to_add) {
276                 if (why == RESTART_META) {
277                         mlog(0, "restarting function.\n");
278                         restart_func = 1;
279                 } else {
280                         BUG_ON(why != RESTART_TRANS);
281
282                         mlog(0, "restarting transaction.\n");
283                         /* TODO: This can be more intelligent. */
284                         credits = ocfs2_calc_extend_credits(osb->sb,
285                                                             et.et_root_el,
286                                                             clusters_to_add);
287                         status = ocfs2_extend_trans(handle, credits);
288                         if (status < 0) {
289                                 /* handle still has to be committed at
290                                  * this point. */
291                                 status = -ENOMEM;
292                                 mlog_errno(status);
293                                 goto leave;
294                         }
295                         goto restarted_transaction;
296                 }
297         }
298
299 leave:
300         if (handle) {
301                 ocfs2_commit_trans(osb, handle);
302                 handle = NULL;
303         }
304         if (data_ac) {
305                 ocfs2_free_alloc_context(data_ac);
306                 data_ac = NULL;
307         }
308         if (meta_ac) {
309                 ocfs2_free_alloc_context(meta_ac);
310                 meta_ac = NULL;
311         }
312         if ((!status) && restart_func) {
313                 restart_func = 0;
314                 goto restart_all;
315         }
316
317         return status;
318 }
319
320 static int __ocfs2_remove_xattr_range(struct inode *inode,
321                                       struct buffer_head *root_bh,
322                                       struct ocfs2_xattr_value_root *xv,
323                                       u32 cpos, u32 phys_cpos, u32 len,
324                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
325 {
326         int ret;
327         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
328         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
329         struct inode *tl_inode = osb->osb_tl_inode;
330         handle_t *handle;
331         struct ocfs2_alloc_context *meta_ac = NULL;
332         struct ocfs2_extent_tree et;
333
334         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
335
336         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
337         if (ret) {
338                 mlog_errno(ret);
339                 return ret;
340         }
341
342         mutex_lock(&tl_inode->i_mutex);
343
344         if (ocfs2_truncate_log_needs_flush(osb)) {
345                 ret = __ocfs2_flush_truncate_log(osb);
346                 if (ret < 0) {
347                         mlog_errno(ret);
348                         goto out;
349                 }
350         }
351
352         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
353         if (IS_ERR(handle)) {
354                 ret = PTR_ERR(handle);
355                 mlog_errno(ret);
356                 goto out;
357         }
358
359         ret = ocfs2_journal_access(handle, inode, root_bh,
360                                    OCFS2_JOURNAL_ACCESS_WRITE);
361         if (ret) {
362                 mlog_errno(ret);
363                 goto out_commit;
364         }
365
366         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
367                                   dealloc);
368         if (ret) {
369                 mlog_errno(ret);
370                 goto out_commit;
371         }
372
373         le32_add_cpu(&xv->xr_clusters, -len);
374
375         ret = ocfs2_journal_dirty(handle, root_bh);
376         if (ret) {
377                 mlog_errno(ret);
378                 goto out_commit;
379         }
380
381         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
382         if (ret)
383                 mlog_errno(ret);
384
385 out_commit:
386         ocfs2_commit_trans(osb, handle);
387 out:
388         mutex_unlock(&tl_inode->i_mutex);
389
390         if (meta_ac)
391                 ocfs2_free_alloc_context(meta_ac);
392
393         return ret;
394 }
395
396 static int ocfs2_xattr_shrink_size(struct inode *inode,
397                                    u32 old_clusters,
398                                    u32 new_clusters,
399                                    struct buffer_head *root_bh,
400                                    struct ocfs2_xattr_value_root *xv)
401 {
402         int ret = 0;
403         u32 trunc_len, cpos, phys_cpos, alloc_size;
404         u64 block;
405         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
406         struct ocfs2_cached_dealloc_ctxt dealloc;
407
408         ocfs2_init_dealloc_ctxt(&dealloc);
409
410         if (old_clusters <= new_clusters)
411                 return 0;
412
413         cpos = new_clusters;
414         trunc_len = old_clusters - new_clusters;
415         while (trunc_len) {
416                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
417                                                &alloc_size, &xv->xr_list);
418                 if (ret) {
419                         mlog_errno(ret);
420                         goto out;
421                 }
422
423                 if (alloc_size > trunc_len)
424                         alloc_size = trunc_len;
425
426                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
427                                                  phys_cpos, alloc_size,
428                                                  &dealloc);
429                 if (ret) {
430                         mlog_errno(ret);
431                         goto out;
432                 }
433
434                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
435                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
436                                                        alloc_size);
437                 cpos += alloc_size;
438                 trunc_len -= alloc_size;
439         }
440
441 out:
442         ocfs2_schedule_truncate_log_flush(osb, 1);
443         ocfs2_run_deallocs(osb, &dealloc);
444
445         return ret;
446 }
447
448 static int ocfs2_xattr_value_truncate(struct inode *inode,
449                                       struct buffer_head *root_bh,
450                                       struct ocfs2_xattr_value_root *xv,
451                                       int len)
452 {
453         int ret;
454         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
455         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
456
457         if (new_clusters == old_clusters)
458                 return 0;
459
460         if (new_clusters > old_clusters)
461                 ret = ocfs2_xattr_extend_allocation(inode,
462                                                     new_clusters - old_clusters,
463                                                     root_bh, xv);
464         else
465                 ret = ocfs2_xattr_shrink_size(inode,
466                                               old_clusters, new_clusters,
467                                               root_bh, xv);
468
469         return ret;
470 }
471
472 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
473                                   size_t *result, const char *prefix,
474                                   const char *name, int name_len)
475 {
476         char *p = buffer + *result;
477         int prefix_len = strlen(prefix);
478         int total_len = prefix_len + name_len + 1;
479
480         *result += total_len;
481
482         /* we are just looking for how big our buffer needs to be */
483         if (!size)
484                 return 0;
485
486         if (*result > size)
487                 return -ERANGE;
488
489         memcpy(p, prefix, prefix_len);
490         memcpy(p + prefix_len, name, name_len);
491         p[prefix_len + name_len] = '\0';
492
493         return 0;
494 }
495
496 static int ocfs2_xattr_list_entries(struct inode *inode,
497                                     struct ocfs2_xattr_header *header,
498                                     char *buffer, size_t buffer_size)
499 {
500         size_t result = 0;
501         int i, type, ret;
502         const char *prefix, *name;
503
504         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
505                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
506                 type = ocfs2_xattr_get_type(entry);
507                 prefix = ocfs2_xattr_prefix(type);
508
509                 if (prefix) {
510                         name = (const char *)header +
511                                 le16_to_cpu(entry->xe_name_offset);
512
513                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
514                                                      &result, prefix, name,
515                                                      entry->xe_name_len);
516                         if (ret)
517                                 return ret;
518                 }
519         }
520
521         return result;
522 }
523
524 static int ocfs2_xattr_ibody_list(struct inode *inode,
525                                   struct ocfs2_dinode *di,
526                                   char *buffer,
527                                   size_t buffer_size)
528 {
529         struct ocfs2_xattr_header *header = NULL;
530         struct ocfs2_inode_info *oi = OCFS2_I(inode);
531         int ret = 0;
532
533         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
534                 return ret;
535
536         header = (struct ocfs2_xattr_header *)
537                  ((void *)di + inode->i_sb->s_blocksize -
538                  le16_to_cpu(di->i_xattr_inline_size));
539
540         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
541
542         return ret;
543 }
544
545 static int ocfs2_xattr_block_list(struct inode *inode,
546                                   struct ocfs2_dinode *di,
547                                   char *buffer,
548                                   size_t buffer_size)
549 {
550         struct buffer_head *blk_bh = NULL;
551         struct ocfs2_xattr_block *xb;
552         int ret = 0;
553
554         if (!di->i_xattr_loc)
555                 return ret;
556
557         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
558         if (ret < 0) {
559                 mlog_errno(ret);
560                 return ret;
561         }
562
563         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
564         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
565                 ret = -EIO;
566                 goto cleanup;
567         }
568
569         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
570                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
571                 ret = ocfs2_xattr_list_entries(inode, header,
572                                                buffer, buffer_size);
573         } else {
574                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
575                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
576                                                    buffer, buffer_size);
577         }
578 cleanup:
579         brelse(blk_bh);
580
581         return ret;
582 }
583
584 ssize_t ocfs2_listxattr(struct dentry *dentry,
585                         char *buffer,
586                         size_t size)
587 {
588         int ret = 0, i_ret = 0, b_ret = 0;
589         struct buffer_head *di_bh = NULL;
590         struct ocfs2_dinode *di = NULL;
591         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
592
593         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
594                 return -EOPNOTSUPP;
595
596         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
597                 return ret;
598
599         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
600         if (ret < 0) {
601                 mlog_errno(ret);
602                 return ret;
603         }
604
605         di = (struct ocfs2_dinode *)di_bh->b_data;
606
607         down_read(&oi->ip_xattr_sem);
608         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
609         if (i_ret < 0)
610                 b_ret = 0;
611         else {
612                 if (buffer) {
613                         buffer += i_ret;
614                         size -= i_ret;
615                 }
616                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
617                                                buffer, size);
618                 if (b_ret < 0)
619                         i_ret = 0;
620         }
621         up_read(&oi->ip_xattr_sem);
622         ocfs2_inode_unlock(dentry->d_inode, 0);
623
624         brelse(di_bh);
625
626         return i_ret + b_ret;
627 }
628
629 static int ocfs2_xattr_find_entry(int name_index,
630                                   const char *name,
631                                   struct ocfs2_xattr_search *xs)
632 {
633         struct ocfs2_xattr_entry *entry;
634         size_t name_len;
635         int i, cmp = 1;
636
637         if (name == NULL)
638                 return -EINVAL;
639
640         name_len = strlen(name);
641         entry = xs->here;
642         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
643                 cmp = name_index - ocfs2_xattr_get_type(entry);
644                 if (!cmp)
645                         cmp = name_len - entry->xe_name_len;
646                 if (!cmp)
647                         cmp = memcmp(name, (xs->base +
648                                      le16_to_cpu(entry->xe_name_offset)),
649                                      name_len);
650                 if (cmp == 0)
651                         break;
652                 entry += 1;
653         }
654         xs->here = entry;
655
656         return cmp ? -ENODATA : 0;
657 }
658
659 static int ocfs2_xattr_get_value_outside(struct inode *inode,
660                                          struct ocfs2_xattr_value_root *xv,
661                                          void *buffer,
662                                          size_t len)
663 {
664         u32 cpos, p_cluster, num_clusters, bpc, clusters;
665         u64 blkno;
666         int i, ret = 0;
667         size_t cplen, blocksize;
668         struct buffer_head *bh = NULL;
669         struct ocfs2_extent_list *el;
670
671         el = &xv->xr_list;
672         clusters = le32_to_cpu(xv->xr_clusters);
673         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
674         blocksize = inode->i_sb->s_blocksize;
675
676         cpos = 0;
677         while (cpos < clusters) {
678                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
679                                                &num_clusters, el);
680                 if (ret) {
681                         mlog_errno(ret);
682                         goto out;
683                 }
684
685                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
686                 /* Copy ocfs2_xattr_value */
687                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
688                         ret = ocfs2_read_block(inode, blkno, &bh);
689                         if (ret) {
690                                 mlog_errno(ret);
691                                 goto out;
692                         }
693
694                         cplen = len >= blocksize ? blocksize : len;
695                         memcpy(buffer, bh->b_data, cplen);
696                         len -= cplen;
697                         buffer += cplen;
698
699                         brelse(bh);
700                         bh = NULL;
701                         if (len == 0)
702                                 break;
703                 }
704                 cpos += num_clusters;
705         }
706 out:
707         return ret;
708 }
709
710 static int ocfs2_xattr_ibody_get(struct inode *inode,
711                                  int name_index,
712                                  const char *name,
713                                  void *buffer,
714                                  size_t buffer_size,
715                                  struct ocfs2_xattr_search *xs)
716 {
717         struct ocfs2_inode_info *oi = OCFS2_I(inode);
718         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
719         struct ocfs2_xattr_value_root *xv;
720         size_t size;
721         int ret = 0;
722
723         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
724                 return -ENODATA;
725
726         xs->end = (void *)di + inode->i_sb->s_blocksize;
727         xs->header = (struct ocfs2_xattr_header *)
728                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
729         xs->base = (void *)xs->header;
730         xs->here = xs->header->xh_entries;
731
732         ret = ocfs2_xattr_find_entry(name_index, name, xs);
733         if (ret)
734                 return ret;
735         size = le64_to_cpu(xs->here->xe_value_size);
736         if (buffer) {
737                 if (size > buffer_size)
738                         return -ERANGE;
739                 if (ocfs2_xattr_is_local(xs->here)) {
740                         memcpy(buffer, (void *)xs->base +
741                                le16_to_cpu(xs->here->xe_name_offset) +
742                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
743                 } else {
744                         xv = (struct ocfs2_xattr_value_root *)
745                                 (xs->base + le16_to_cpu(
746                                  xs->here->xe_name_offset) +
747                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
748                         ret = ocfs2_xattr_get_value_outside(inode, xv,
749                                                             buffer, size);
750                         if (ret < 0) {
751                                 mlog_errno(ret);
752                                 return ret;
753                         }
754                 }
755         }
756
757         return size;
758 }
759
760 static int ocfs2_xattr_block_get(struct inode *inode,
761                                  int name_index,
762                                  const char *name,
763                                  void *buffer,
764                                  size_t buffer_size,
765                                  struct ocfs2_xattr_search *xs)
766 {
767         struct ocfs2_xattr_block *xb;
768         struct ocfs2_xattr_value_root *xv;
769         size_t size;
770         int ret = -ENODATA, name_offset, name_len, block_off, i;
771
772         memset(&xs->bucket, 0, sizeof(xs->bucket));
773
774         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
775         if (ret) {
776                 mlog_errno(ret);
777                 goto cleanup;
778         }
779
780         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
781         size = le64_to_cpu(xs->here->xe_value_size);
782         if (buffer) {
783                 ret = -ERANGE;
784                 if (size > buffer_size)
785                         goto cleanup;
786
787                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
788                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
789                 i = xs->here - xs->header->xh_entries;
790
791                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
792                         ret = ocfs2_xattr_bucket_get_name_value(inode,
793                                                                 xs->bucket.xh,
794                                                                 i,
795                                                                 &block_off,
796                                                                 &name_offset);
797                         xs->base = xs->bucket.bhs[block_off]->b_data;
798                 }
799                 if (ocfs2_xattr_is_local(xs->here)) {
800                         memcpy(buffer, (void *)xs->base +
801                                name_offset + name_len, size);
802                 } else {
803                         xv = (struct ocfs2_xattr_value_root *)
804                                 (xs->base + name_offset + name_len);
805                         ret = ocfs2_xattr_get_value_outside(inode, xv,
806                                                             buffer, size);
807                         if (ret < 0) {
808                                 mlog_errno(ret);
809                                 goto cleanup;
810                         }
811                 }
812         }
813         ret = size;
814 cleanup:
815         for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
816                 brelse(xs->bucket.bhs[i]);
817         memset(&xs->bucket, 0, sizeof(xs->bucket));
818
819         brelse(xs->xattr_bh);
820         xs->xattr_bh = NULL;
821         return ret;
822 }
823
824 /* ocfs2_xattr_get()
825  *
826  * Copy an extended attribute into the buffer provided.
827  * Buffer is NULL to compute the size of buffer required.
828  */
829 static int ocfs2_xattr_get(struct inode *inode,
830                            int name_index,
831                            const char *name,
832                            void *buffer,
833                            size_t buffer_size)
834 {
835         int ret;
836         struct ocfs2_dinode *di = NULL;
837         struct buffer_head *di_bh = NULL;
838         struct ocfs2_inode_info *oi = OCFS2_I(inode);
839         struct ocfs2_xattr_search xis = {
840                 .not_found = -ENODATA,
841         };
842         struct ocfs2_xattr_search xbs = {
843                 .not_found = -ENODATA,
844         };
845
846         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
847                 return -EOPNOTSUPP;
848
849         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
850                 ret = -ENODATA;
851
852         ret = ocfs2_inode_lock(inode, &di_bh, 0);
853         if (ret < 0) {
854                 mlog_errno(ret);
855                 return ret;
856         }
857         xis.inode_bh = xbs.inode_bh = di_bh;
858         di = (struct ocfs2_dinode *)di_bh->b_data;
859
860         down_read(&oi->ip_xattr_sem);
861         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
862                                     buffer_size, &xis);
863         if (ret == -ENODATA)
864                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
865                                             buffer_size, &xbs);
866         up_read(&oi->ip_xattr_sem);
867         ocfs2_inode_unlock(inode, 0);
868
869         brelse(di_bh);
870
871         return ret;
872 }
873
874 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
875                                            struct ocfs2_xattr_value_root *xv,
876                                            const void *value,
877                                            int value_len)
878 {
879         int ret = 0, i, cp_len, credits;
880         u16 blocksize = inode->i_sb->s_blocksize;
881         u32 p_cluster, num_clusters;
882         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
883         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
884         u64 blkno;
885         struct buffer_head *bh = NULL;
886         handle_t *handle;
887
888         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
889
890         credits = clusters * bpc;
891         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
892         if (IS_ERR(handle)) {
893                 ret = PTR_ERR(handle);
894                 mlog_errno(ret);
895                 goto out;
896         }
897
898         while (cpos < clusters) {
899                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
900                                                &num_clusters, &xv->xr_list);
901                 if (ret) {
902                         mlog_errno(ret);
903                         goto out_commit;
904                 }
905
906                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
907
908                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
909                         ret = ocfs2_read_block(inode, blkno, &bh);
910                         if (ret) {
911                                 mlog_errno(ret);
912                                 goto out_commit;
913                         }
914
915                         ret = ocfs2_journal_access(handle,
916                                                    inode,
917                                                    bh,
918                                                    OCFS2_JOURNAL_ACCESS_WRITE);
919                         if (ret < 0) {
920                                 mlog_errno(ret);
921                                 goto out_commit;
922                         }
923
924                         cp_len = value_len > blocksize ? blocksize : value_len;
925                         memcpy(bh->b_data, value, cp_len);
926                         value_len -= cp_len;
927                         value += cp_len;
928                         if (cp_len < blocksize)
929                                 memset(bh->b_data + cp_len, 0,
930                                        blocksize - cp_len);
931
932                         ret = ocfs2_journal_dirty(handle, bh);
933                         if (ret < 0) {
934                                 mlog_errno(ret);
935                                 goto out_commit;
936                         }
937                         brelse(bh);
938                         bh = NULL;
939
940                         /*
941                          * XXX: do we need to empty all the following
942                          * blocks in this cluster?
943                          */
944                         if (!value_len)
945                                 break;
946                 }
947                 cpos += num_clusters;
948         }
949 out_commit:
950         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
951 out:
952         brelse(bh);
953
954         return ret;
955 }
956
957 static int ocfs2_xattr_cleanup(struct inode *inode,
958                                struct ocfs2_xattr_info *xi,
959                                struct ocfs2_xattr_search *xs,
960                                size_t offs)
961 {
962         handle_t *handle = NULL;
963         int ret = 0;
964         size_t name_len = strlen(xi->name);
965         void *val = xs->base + offs;
966         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
967
968         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
969                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
970         if (IS_ERR(handle)) {
971                 ret = PTR_ERR(handle);
972                 mlog_errno(ret);
973                 goto out;
974         }
975         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
976                                    OCFS2_JOURNAL_ACCESS_WRITE);
977         if (ret) {
978                 mlog_errno(ret);
979                 goto out_commit;
980         }
981         /* Decrease xattr count */
982         le16_add_cpu(&xs->header->xh_count, -1);
983         /* Remove the xattr entry and tree root which has already be set*/
984         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
985         memset(val, 0, size);
986
987         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
988         if (ret < 0)
989                 mlog_errno(ret);
990 out_commit:
991         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
992 out:
993         return ret;
994 }
995
996 static int ocfs2_xattr_update_entry(struct inode *inode,
997                                     struct ocfs2_xattr_info *xi,
998                                     struct ocfs2_xattr_search *xs,
999                                     size_t offs)
1000 {
1001         handle_t *handle = NULL;
1002         int ret = 0;
1003
1004         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1005                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1006         if (IS_ERR(handle)) {
1007                 ret = PTR_ERR(handle);
1008                 mlog_errno(ret);
1009                 goto out;
1010         }
1011         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1012                                    OCFS2_JOURNAL_ACCESS_WRITE);
1013         if (ret) {
1014                 mlog_errno(ret);
1015                 goto out_commit;
1016         }
1017
1018         xs->here->xe_name_offset = cpu_to_le16(offs);
1019         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1020         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1021                 ocfs2_xattr_set_local(xs->here, 1);
1022         else
1023                 ocfs2_xattr_set_local(xs->here, 0);
1024         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1025
1026         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1027         if (ret < 0)
1028                 mlog_errno(ret);
1029 out_commit:
1030         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1031 out:
1032         return ret;
1033 }
1034
1035 /*
1036  * ocfs2_xattr_set_value_outside()
1037  *
1038  * Set large size value in B tree.
1039  */
1040 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1041                                          struct ocfs2_xattr_info *xi,
1042                                          struct ocfs2_xattr_search *xs,
1043                                          size_t offs)
1044 {
1045         size_t name_len = strlen(xi->name);
1046         void *val = xs->base + offs;
1047         struct ocfs2_xattr_value_root *xv = NULL;
1048         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1049         int ret = 0;
1050
1051         memset(val, 0, size);
1052         memcpy(val, xi->name, name_len);
1053         xv = (struct ocfs2_xattr_value_root *)
1054                 (val + OCFS2_XATTR_SIZE(name_len));
1055         xv->xr_clusters = 0;
1056         xv->xr_last_eb_blk = 0;
1057         xv->xr_list.l_tree_depth = 0;
1058         xv->xr_list.l_count = cpu_to_le16(1);
1059         xv->xr_list.l_next_free_rec = 0;
1060
1061         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1062                                          xi->value_len);
1063         if (ret < 0) {
1064                 mlog_errno(ret);
1065                 return ret;
1066         }
1067         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1068                                               xi->value_len);
1069         if (ret < 0) {
1070                 mlog_errno(ret);
1071                 return ret;
1072         }
1073         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1074         if (ret < 0)
1075                 mlog_errno(ret);
1076
1077         return ret;
1078 }
1079
1080 /*
1081  * ocfs2_xattr_set_entry_local()
1082  *
1083  * Set, replace or remove extended attribute in local.
1084  */
1085 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1086                                         struct ocfs2_xattr_info *xi,
1087                                         struct ocfs2_xattr_search *xs,
1088                                         struct ocfs2_xattr_entry *last,
1089                                         size_t min_offs)
1090 {
1091         size_t name_len = strlen(xi->name);
1092         int i;
1093
1094         if (xi->value && xs->not_found) {
1095                 /* Insert the new xattr entry. */
1096                 le16_add_cpu(&xs->header->xh_count, 1);
1097                 ocfs2_xattr_set_type(last, xi->name_index);
1098                 ocfs2_xattr_set_local(last, 1);
1099                 last->xe_name_len = name_len;
1100         } else {
1101                 void *first_val;
1102                 void *val;
1103                 size_t offs, size;
1104
1105                 first_val = xs->base + min_offs;
1106                 offs = le16_to_cpu(xs->here->xe_name_offset);
1107                 val = xs->base + offs;
1108
1109                 if (le64_to_cpu(xs->here->xe_value_size) >
1110                     OCFS2_XATTR_INLINE_SIZE)
1111                         size = OCFS2_XATTR_SIZE(name_len) +
1112                                 OCFS2_XATTR_ROOT_SIZE;
1113                 else
1114                         size = OCFS2_XATTR_SIZE(name_len) +
1115                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1116
1117                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1118                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1119                         /* The old and the new value have the
1120                            same size. Just replace the value. */
1121                         ocfs2_xattr_set_local(xs->here, 1);
1122                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1123                         /* Clear value bytes. */
1124                         memset(val + OCFS2_XATTR_SIZE(name_len),
1125                                0,
1126                                OCFS2_XATTR_SIZE(xi->value_len));
1127                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1128                                xi->value,
1129                                xi->value_len);
1130                         return;
1131                 }
1132                 /* Remove the old name+value. */
1133                 memmove(first_val + size, first_val, val - first_val);
1134                 memset(first_val, 0, size);
1135                 xs->here->xe_name_hash = 0;
1136                 xs->here->xe_name_offset = 0;
1137                 ocfs2_xattr_set_local(xs->here, 1);
1138                 xs->here->xe_value_size = 0;
1139
1140                 min_offs += size;
1141
1142                 /* Adjust all value offsets. */
1143                 last = xs->header->xh_entries;
1144                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1145                         size_t o = le16_to_cpu(last->xe_name_offset);
1146
1147                         if (o < offs)
1148                                 last->xe_name_offset = cpu_to_le16(o + size);
1149                         last += 1;
1150                 }
1151
1152                 if (!xi->value) {
1153                         /* Remove the old entry. */
1154                         last -= 1;
1155                         memmove(xs->here, xs->here + 1,
1156                                 (void *)last - (void *)xs->here);
1157                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1158                         le16_add_cpu(&xs->header->xh_count, -1);
1159                 }
1160         }
1161         if (xi->value) {
1162                 /* Insert the new name+value. */
1163                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1164                                 OCFS2_XATTR_SIZE(xi->value_len);
1165                 void *val = xs->base + min_offs - size;
1166
1167                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1168                 memset(val, 0, size);
1169                 memcpy(val, xi->name, name_len);
1170                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1171                        xi->value,
1172                        xi->value_len);
1173                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1174                 ocfs2_xattr_set_local(xs->here, 1);
1175                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1176         }
1177
1178         return;
1179 }
1180
1181 /*
1182  * ocfs2_xattr_set_entry()
1183  *
1184  * Set extended attribute entry into inode or block.
1185  *
1186  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1187  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1188  * then set value in B tree with set_value_outside().
1189  */
1190 static int ocfs2_xattr_set_entry(struct inode *inode,
1191                                  struct ocfs2_xattr_info *xi,
1192                                  struct ocfs2_xattr_search *xs,
1193                                  int flag)
1194 {
1195         struct ocfs2_xattr_entry *last;
1196         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1197         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1198         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1199         size_t size_l = 0;
1200         handle_t *handle = NULL;
1201         int free, i, ret;
1202         struct ocfs2_xattr_info xi_l = {
1203                 .name_index = xi->name_index,
1204                 .name = xi->name,
1205                 .value = xi->value,
1206                 .value_len = xi->value_len,
1207         };
1208
1209         /* Compute min_offs, last and free space. */
1210         last = xs->header->xh_entries;
1211
1212         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1213                 size_t offs = le16_to_cpu(last->xe_name_offset);
1214                 if (offs < min_offs)
1215                         min_offs = offs;
1216                 last += 1;
1217         }
1218
1219         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1220         if (free < 0)
1221                 return -EIO;
1222
1223         if (!xs->not_found) {
1224                 size_t size = 0;
1225                 if (ocfs2_xattr_is_local(xs->here))
1226                         size = OCFS2_XATTR_SIZE(name_len) +
1227                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1228                 else
1229                         size = OCFS2_XATTR_SIZE(name_len) +
1230                                 OCFS2_XATTR_ROOT_SIZE;
1231                 free += (size + sizeof(struct ocfs2_xattr_entry));
1232         }
1233         /* Check free space in inode or block */
1234         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1235                 if (free < sizeof(struct ocfs2_xattr_entry) +
1236                            OCFS2_XATTR_SIZE(name_len) +
1237                            OCFS2_XATTR_ROOT_SIZE) {
1238                         ret = -ENOSPC;
1239                         goto out;
1240                 }
1241                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1242                 xi_l.value = (void *)&def_xv;
1243                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1244         } else if (xi->value) {
1245                 if (free < sizeof(struct ocfs2_xattr_entry) +
1246                            OCFS2_XATTR_SIZE(name_len) +
1247                            OCFS2_XATTR_SIZE(xi->value_len)) {
1248                         ret = -ENOSPC;
1249                         goto out;
1250                 }
1251         }
1252
1253         if (!xs->not_found) {
1254                 /* For existing extended attribute */
1255                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1256                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1257                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1258                 void *val = xs->base + offs;
1259
1260                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1261                         /* Replace existing local xattr with tree root */
1262                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1263                                                             offs);
1264                         if (ret < 0)
1265                                 mlog_errno(ret);
1266                         goto out;
1267                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1268                         /* For existing xattr which has value outside */
1269                         struct ocfs2_xattr_value_root *xv = NULL;
1270                         xv = (struct ocfs2_xattr_value_root *)(val +
1271                                 OCFS2_XATTR_SIZE(name_len));
1272
1273                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1274                                 /*
1275                                  * If new value need set outside also,
1276                                  * first truncate old value to new value,
1277                                  * then set new value with set_value_outside().
1278                                  */
1279                                 ret = ocfs2_xattr_value_truncate(inode,
1280                                                                  xs->xattr_bh,
1281                                                                  xv,
1282                                                                  xi->value_len);
1283                                 if (ret < 0) {
1284                                         mlog_errno(ret);
1285                                         goto out;
1286                                 }
1287
1288                                 ret = __ocfs2_xattr_set_value_outside(inode,
1289                                                                 xv,
1290                                                                 xi->value,
1291                                                                 xi->value_len);
1292                                 if (ret < 0) {
1293                                         mlog_errno(ret);
1294                                         goto out;
1295                                 }
1296
1297                                 ret = ocfs2_xattr_update_entry(inode,
1298                                                                xi,
1299                                                                xs,
1300                                                                offs);
1301                                 if (ret < 0)
1302                                         mlog_errno(ret);
1303                                 goto out;
1304                         } else {
1305                                 /*
1306                                  * If new value need set in local,
1307                                  * just trucate old value to zero.
1308                                  */
1309                                  ret = ocfs2_xattr_value_truncate(inode,
1310                                                                  xs->xattr_bh,
1311                                                                  xv,
1312                                                                  0);
1313                                 if (ret < 0)
1314                                         mlog_errno(ret);
1315                         }
1316                 }
1317         }
1318
1319         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1320                                    OCFS2_INODE_UPDATE_CREDITS);
1321         if (IS_ERR(handle)) {
1322                 ret = PTR_ERR(handle);
1323                 mlog_errno(ret);
1324                 goto out;
1325         }
1326
1327         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1328                                    OCFS2_JOURNAL_ACCESS_WRITE);
1329         if (ret) {
1330                 mlog_errno(ret);
1331                 goto out_commit;
1332         }
1333
1334         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1335                 /* set extended attribute in external block. */
1336                 ret = ocfs2_extend_trans(handle,
1337                                          OCFS2_INODE_UPDATE_CREDITS +
1338                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1339                 if (ret) {
1340                         mlog_errno(ret);
1341                         goto out_commit;
1342                 }
1343                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1344                                            OCFS2_JOURNAL_ACCESS_WRITE);
1345                 if (ret) {
1346                         mlog_errno(ret);
1347                         goto out_commit;
1348                 }
1349         }
1350
1351         /*
1352          * Set value in local, include set tree root in local.
1353          * This is the first step for value size >INLINE_SIZE.
1354          */
1355         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1356
1357         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1358                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1359                 if (ret < 0) {
1360                         mlog_errno(ret);
1361                         goto out_commit;
1362                 }
1363         }
1364
1365         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1366             (flag & OCFS2_INLINE_XATTR_FL)) {
1367                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1368                 unsigned int xattrsize = osb->s_xattr_inline_size;
1369
1370                 /*
1371                  * Adjust extent record count or inline data size
1372                  * to reserve space for extended attribute.
1373                  */
1374                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1375                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1376                         le16_add_cpu(&idata->id_count, -xattrsize);
1377                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1378                         struct ocfs2_extent_list *el = &di->id2.i_list;
1379                         le16_add_cpu(&el->l_count, -(xattrsize /
1380                                         sizeof(struct ocfs2_extent_rec)));
1381                 }
1382                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1383         }
1384         /* Update xattr flag */
1385         spin_lock(&oi->ip_lock);
1386         oi->ip_dyn_features |= flag;
1387         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1388         spin_unlock(&oi->ip_lock);
1389         /* Update inode ctime */
1390         inode->i_ctime = CURRENT_TIME;
1391         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1392         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1393
1394         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1395         if (ret < 0)
1396                 mlog_errno(ret);
1397
1398 out_commit:
1399         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1400
1401         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1402                 /*
1403                  * Set value outside in B tree.
1404                  * This is the second step for value size > INLINE_SIZE.
1405                  */
1406                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1407                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1408                 if (ret < 0) {
1409                         int ret2;
1410
1411                         mlog_errno(ret);
1412                         /*
1413                          * If set value outside failed, we have to clean
1414                          * the junk tree root we have already set in local.
1415                          */
1416                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1417                         if (ret2 < 0)
1418                                 mlog_errno(ret2);
1419                 }
1420         }
1421 out:
1422         return ret;
1423
1424 }
1425
1426 static int ocfs2_remove_value_outside(struct inode*inode,
1427                                       struct buffer_head *bh,
1428                                       struct ocfs2_xattr_header *header)
1429 {
1430         int ret = 0, i;
1431
1432         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1433                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1434
1435                 if (!ocfs2_xattr_is_local(entry)) {
1436                         struct ocfs2_xattr_value_root *xv;
1437                         void *val;
1438
1439                         val = (void *)header +
1440                                 le16_to_cpu(entry->xe_name_offset);
1441                         xv = (struct ocfs2_xattr_value_root *)
1442                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1443                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1444                         if (ret < 0) {
1445                                 mlog_errno(ret);
1446                                 return ret;
1447                         }
1448                 }
1449         }
1450
1451         return ret;
1452 }
1453
1454 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1455                                     struct buffer_head *di_bh)
1456 {
1457
1458         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1459         struct ocfs2_xattr_header *header;
1460         int ret;
1461
1462         header = (struct ocfs2_xattr_header *)
1463                  ((void *)di + inode->i_sb->s_blocksize -
1464                  le16_to_cpu(di->i_xattr_inline_size));
1465
1466         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1467
1468         return ret;
1469 }
1470
1471 static int ocfs2_xattr_block_remove(struct inode *inode,
1472                                     struct buffer_head *blk_bh)
1473 {
1474         struct ocfs2_xattr_block *xb;
1475         int ret = 0;
1476
1477         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1478         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1479                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1480                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1481         } else
1482                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1483
1484         return ret;
1485 }
1486
1487 static int ocfs2_xattr_free_block(struct inode *inode,
1488                                   u64 block)
1489 {
1490         struct inode *xb_alloc_inode;
1491         struct buffer_head *xb_alloc_bh = NULL;
1492         struct buffer_head *blk_bh = NULL;
1493         struct ocfs2_xattr_block *xb;
1494         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1495         handle_t *handle;
1496         int ret = 0;
1497         u64 blk, bg_blkno;
1498         u16 bit;
1499
1500         ret = ocfs2_read_block(inode, block, &blk_bh);
1501         if (ret < 0) {
1502                 mlog_errno(ret);
1503                 goto out;
1504         }
1505
1506         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1507         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1508                 ret = -EIO;
1509                 goto out;
1510         }
1511
1512         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1513         if (ret < 0) {
1514                 mlog_errno(ret);
1515                 goto out;
1516         }
1517
1518         blk = le64_to_cpu(xb->xb_blkno);
1519         bit = le16_to_cpu(xb->xb_suballoc_bit);
1520         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1521
1522         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1523                                 EXTENT_ALLOC_SYSTEM_INODE,
1524                                 le16_to_cpu(xb->xb_suballoc_slot));
1525         if (!xb_alloc_inode) {
1526                 ret = -ENOMEM;
1527                 mlog_errno(ret);
1528                 goto out;
1529         }
1530         mutex_lock(&xb_alloc_inode->i_mutex);
1531
1532         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1533         if (ret < 0) {
1534                 mlog_errno(ret);
1535                 goto out_mutex;
1536         }
1537
1538         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1539         if (IS_ERR(handle)) {
1540                 ret = PTR_ERR(handle);
1541                 mlog_errno(ret);
1542                 goto out_unlock;
1543         }
1544
1545         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1546                                        bit, bg_blkno, 1);
1547         if (ret < 0)
1548                 mlog_errno(ret);
1549
1550         ocfs2_commit_trans(osb, handle);
1551 out_unlock:
1552         ocfs2_inode_unlock(xb_alloc_inode, 1);
1553         brelse(xb_alloc_bh);
1554 out_mutex:
1555         mutex_unlock(&xb_alloc_inode->i_mutex);
1556         iput(xb_alloc_inode);
1557 out:
1558         brelse(blk_bh);
1559         return ret;
1560 }
1561
1562 /*
1563  * ocfs2_xattr_remove()
1564  *
1565  * Free extended attribute resources associated with this inode.
1566  */
1567 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1568 {
1569         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1570         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1571         handle_t *handle;
1572         int ret;
1573
1574         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1575                 return 0;
1576
1577         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1578                 return 0;
1579
1580         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1581                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1582                 if (ret < 0) {
1583                         mlog_errno(ret);
1584                         goto out;
1585                 }
1586         }
1587
1588         if (di->i_xattr_loc) {
1589                 ret = ocfs2_xattr_free_block(inode,
1590                                              le64_to_cpu(di->i_xattr_loc));
1591                 if (ret < 0) {
1592                         mlog_errno(ret);
1593                         goto out;
1594                 }
1595         }
1596
1597         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1598                                    OCFS2_INODE_UPDATE_CREDITS);
1599         if (IS_ERR(handle)) {
1600                 ret = PTR_ERR(handle);
1601                 mlog_errno(ret);
1602                 goto out;
1603         }
1604         ret = ocfs2_journal_access(handle, inode, di_bh,
1605                                    OCFS2_JOURNAL_ACCESS_WRITE);
1606         if (ret) {
1607                 mlog_errno(ret);
1608                 goto out_commit;
1609         }
1610
1611         di->i_xattr_loc = 0;
1612
1613         spin_lock(&oi->ip_lock);
1614         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1615         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1616         spin_unlock(&oi->ip_lock);
1617
1618         ret = ocfs2_journal_dirty(handle, di_bh);
1619         if (ret < 0)
1620                 mlog_errno(ret);
1621 out_commit:
1622         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1623 out:
1624         return ret;
1625 }
1626
1627 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1628                                         struct ocfs2_dinode *di)
1629 {
1630         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1631         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1632         int free;
1633
1634         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1635                 return 0;
1636
1637         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1638                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1639                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1640         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1641                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1642                         le64_to_cpu(di->i_size);
1643         } else {
1644                 struct ocfs2_extent_list *el = &di->id2.i_list;
1645                 free = (le16_to_cpu(el->l_count) -
1646                         le16_to_cpu(el->l_next_free_rec)) *
1647                         sizeof(struct ocfs2_extent_rec);
1648         }
1649         if (free >= xattrsize)
1650                 return 1;
1651
1652         return 0;
1653 }
1654
1655 /*
1656  * ocfs2_xattr_ibody_find()
1657  *
1658  * Find extended attribute in inode block and
1659  * fill search info into struct ocfs2_xattr_search.
1660  */
1661 static int ocfs2_xattr_ibody_find(struct inode *inode,
1662                                   int name_index,
1663                                   const char *name,
1664                                   struct ocfs2_xattr_search *xs)
1665 {
1666         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1667         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1668         int ret;
1669         int has_space = 0;
1670
1671         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1672                 return 0;
1673
1674         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1675                 down_read(&oi->ip_alloc_sem);
1676                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1677                 up_read(&oi->ip_alloc_sem);
1678                 if (!has_space)
1679                         return 0;
1680         }
1681
1682         xs->xattr_bh = xs->inode_bh;
1683         xs->end = (void *)di + inode->i_sb->s_blocksize;
1684         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1685                 xs->header = (struct ocfs2_xattr_header *)
1686                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1687         else
1688                 xs->header = (struct ocfs2_xattr_header *)
1689                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1690         xs->base = (void *)xs->header;
1691         xs->here = xs->header->xh_entries;
1692
1693         /* Find the named attribute. */
1694         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1695                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1696                 if (ret && ret != -ENODATA)
1697                         return ret;
1698                 xs->not_found = ret;
1699         }
1700
1701         return 0;
1702 }
1703
1704 /*
1705  * ocfs2_xattr_ibody_set()
1706  *
1707  * Set, replace or remove an extended attribute into inode block.
1708  *
1709  */
1710 static int ocfs2_xattr_ibody_set(struct inode *inode,
1711                                  struct ocfs2_xattr_info *xi,
1712                                  struct ocfs2_xattr_search *xs)
1713 {
1714         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1715         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1716         int ret;
1717
1718         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1719                 return -ENOSPC;
1720
1721         down_write(&oi->ip_alloc_sem);
1722         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1723                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1724                         ret = -ENOSPC;
1725                         goto out;
1726                 }
1727         }
1728
1729         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1730                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1731 out:
1732         up_write(&oi->ip_alloc_sem);
1733
1734         return ret;
1735 }
1736
1737 /*
1738  * ocfs2_xattr_block_find()
1739  *
1740  * Find extended attribute in external block and
1741  * fill search info into struct ocfs2_xattr_search.
1742  */
1743 static int ocfs2_xattr_block_find(struct inode *inode,
1744                                   int name_index,
1745                                   const char *name,
1746                                   struct ocfs2_xattr_search *xs)
1747 {
1748         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1749         struct buffer_head *blk_bh = NULL;
1750         struct ocfs2_xattr_block *xb;
1751         int ret = 0;
1752
1753         if (!di->i_xattr_loc)
1754                 return ret;
1755
1756         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
1757         if (ret < 0) {
1758                 mlog_errno(ret);
1759                 return ret;
1760         }
1761
1762         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1763         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1764                 ret = -EIO;
1765                 goto cleanup;
1766         }
1767
1768         xs->xattr_bh = blk_bh;
1769
1770         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1771                 xs->header = &xb->xb_attrs.xb_header;
1772                 xs->base = (void *)xs->header;
1773                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1774                 xs->here = xs->header->xh_entries;
1775
1776                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1777         } else
1778                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1779                                                    name_index,
1780                                                    name, xs);
1781
1782         if (ret && ret != -ENODATA) {
1783                 xs->xattr_bh = NULL;
1784                 goto cleanup;
1785         }
1786         xs->not_found = ret;
1787         return 0;
1788 cleanup:
1789         brelse(blk_bh);
1790
1791         return ret;
1792 }
1793
1794 /*
1795  * ocfs2_xattr_block_set()
1796  *
1797  * Set, replace or remove an extended attribute into external block.
1798  *
1799  */
1800 static int ocfs2_xattr_block_set(struct inode *inode,
1801                                  struct ocfs2_xattr_info *xi,
1802                                  struct ocfs2_xattr_search *xs)
1803 {
1804         struct buffer_head *new_bh = NULL;
1805         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1806         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1807         struct ocfs2_alloc_context *meta_ac = NULL;
1808         handle_t *handle = NULL;
1809         struct ocfs2_xattr_block *xblk = NULL;
1810         u16 suballoc_bit_start;
1811         u32 num_got;
1812         u64 first_blkno;
1813         int ret;
1814
1815         if (!xs->xattr_bh) {
1816                 /*
1817                  * Alloc one external block for extended attribute
1818                  * outside of inode.
1819                  */
1820                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1821                 if (ret < 0) {
1822                         mlog_errno(ret);
1823                         goto out;
1824                 }
1825                 handle = ocfs2_start_trans(osb,
1826                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1827                 if (IS_ERR(handle)) {
1828                         ret = PTR_ERR(handle);
1829                         mlog_errno(ret);
1830                         goto out;
1831                 }
1832                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1833                                            OCFS2_JOURNAL_ACCESS_CREATE);
1834                 if (ret < 0) {
1835                         mlog_errno(ret);
1836                         goto out_commit;
1837                 }
1838
1839                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1840                                            &suballoc_bit_start, &num_got,
1841                                            &first_blkno);
1842                 if (ret < 0) {
1843                         mlog_errno(ret);
1844                         goto out_commit;
1845                 }
1846
1847                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1848                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1849
1850                 ret = ocfs2_journal_access(handle, inode, new_bh,
1851                                            OCFS2_JOURNAL_ACCESS_CREATE);
1852                 if (ret < 0) {
1853                         mlog_errno(ret);
1854                         goto out_commit;
1855                 }
1856
1857                 /* Initialize ocfs2_xattr_block */
1858                 xs->xattr_bh = new_bh;
1859                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1860                 memset(xblk, 0, inode->i_sb->s_blocksize);
1861                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1862                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1863                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1864                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1865                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1866
1867                 xs->header = &xblk->xb_attrs.xb_header;
1868                 xs->base = (void *)xs->header;
1869                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1870                 xs->here = xs->header->xh_entries;
1871
1872
1873                 ret = ocfs2_journal_dirty(handle, new_bh);
1874                 if (ret < 0) {
1875                         mlog_errno(ret);
1876                         goto out_commit;
1877                 }
1878                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1879                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1880                 if (ret < 0)
1881                         mlog_errno(ret);
1882 out_commit:
1883                 ocfs2_commit_trans(osb, handle);
1884 out:
1885                 if (meta_ac)
1886                         ocfs2_free_alloc_context(meta_ac);
1887                 if (ret < 0)
1888                         return ret;
1889         } else
1890                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1891
1892         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1893                 /* Set extended attribute into external block */
1894                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1895                 if (!ret || ret != -ENOSPC)
1896                         goto end;
1897
1898                 ret = ocfs2_xattr_create_index_block(inode, xs);
1899                 if (ret)
1900                         goto end;
1901         }
1902
1903         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1904
1905 end:
1906
1907         return ret;
1908 }
1909
1910 /*
1911  * ocfs2_xattr_set()
1912  *
1913  * Set, replace or remove an extended attribute for this inode.
1914  * value is NULL to remove an existing extended attribute, else either
1915  * create or replace an extended attribute.
1916  */
1917 int ocfs2_xattr_set(struct inode *inode,
1918                     int name_index,
1919                     const char *name,
1920                     const void *value,
1921                     size_t value_len,
1922                     int flags)
1923 {
1924         struct buffer_head *di_bh = NULL;
1925         struct ocfs2_dinode *di;
1926         int ret;
1927         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1928
1929         struct ocfs2_xattr_info xi = {
1930                 .name_index = name_index,
1931                 .name = name,
1932                 .value = value,
1933                 .value_len = value_len,
1934         };
1935
1936         struct ocfs2_xattr_search xis = {
1937                 .not_found = -ENODATA,
1938         };
1939
1940         struct ocfs2_xattr_search xbs = {
1941                 .not_found = -ENODATA,
1942         };
1943
1944         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1945                 return -EOPNOTSUPP;
1946
1947         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1948         if (ret < 0) {
1949                 mlog_errno(ret);
1950                 return ret;
1951         }
1952         xis.inode_bh = xbs.inode_bh = di_bh;
1953         di = (struct ocfs2_dinode *)di_bh->b_data;
1954
1955         down_write(&OCFS2_I(inode)->ip_xattr_sem);
1956         /*
1957          * Scan inode and external block to find the same name
1958          * extended attribute and collect search infomation.
1959          */
1960         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
1961         if (ret)
1962                 goto cleanup;
1963         if (xis.not_found) {
1964                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
1965                 if (ret)
1966                         goto cleanup;
1967         }
1968
1969         if (xis.not_found && xbs.not_found) {
1970                 ret = -ENODATA;
1971                 if (flags & XATTR_REPLACE)
1972                         goto cleanup;
1973                 ret = 0;
1974                 if (!value)
1975                         goto cleanup;
1976         } else {
1977                 ret = -EEXIST;
1978                 if (flags & XATTR_CREATE)
1979                         goto cleanup;
1980         }
1981
1982         if (!value) {
1983                 /* Remove existing extended attribute */
1984                 if (!xis.not_found)
1985                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
1986                 else if (!xbs.not_found)
1987                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
1988         } else {
1989                 /* We always try to set extended attribute into inode first*/
1990                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
1991                 if (!ret && !xbs.not_found) {
1992                         /*
1993                          * If succeed and that extended attribute existing in
1994                          * external block, then we will remove it.
1995                          */
1996                         xi.value = NULL;
1997                         xi.value_len = 0;
1998                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
1999                 } else if (ret == -ENOSPC) {
2000                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2001                                 ret = ocfs2_xattr_block_find(inode, name_index,
2002                                                              name, &xbs);
2003                                 if (ret)
2004                                         goto cleanup;
2005                         }
2006                         /*
2007                          * If no space in inode, we will set extended attribute
2008                          * into external block.
2009                          */
2010                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2011                         if (ret)
2012                                 goto cleanup;
2013                         if (!xis.not_found) {
2014                                 /*
2015                                  * If succeed and that extended attribute
2016                                  * existing in inode, we will remove it.
2017                                  */
2018                                 xi.value = NULL;
2019                                 xi.value_len = 0;
2020                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2021                         }
2022                 }
2023         }
2024 cleanup:
2025         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2026         ocfs2_inode_unlock(inode, 1);
2027         brelse(di_bh);
2028         brelse(xbs.xattr_bh);
2029         for (i = 0; i < blk_per_bucket; i++)
2030                 brelse(xbs.bucket.bhs[i]);
2031
2032         return ret;
2033 }
2034
2035 /*
2036  * Find the xattr extent rec which may contains name_hash.
2037  * e_cpos will be the first name hash of the xattr rec.
2038  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2039  */
2040 static int ocfs2_xattr_get_rec(struct inode *inode,
2041                                u32 name_hash,
2042                                u64 *p_blkno,
2043                                u32 *e_cpos,
2044                                u32 *num_clusters,
2045                                struct ocfs2_extent_list *el)
2046 {
2047         int ret = 0, i;
2048         struct buffer_head *eb_bh = NULL;
2049         struct ocfs2_extent_block *eb;
2050         struct ocfs2_extent_rec *rec = NULL;
2051         u64 e_blkno = 0;
2052
2053         if (el->l_tree_depth) {
2054                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2055                 if (ret) {
2056                         mlog_errno(ret);
2057                         goto out;
2058                 }
2059
2060                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2061                 el = &eb->h_list;
2062
2063                 if (el->l_tree_depth) {
2064                         ocfs2_error(inode->i_sb,
2065                                     "Inode %lu has non zero tree depth in "
2066                                     "xattr tree block %llu\n", inode->i_ino,
2067                                     (unsigned long long)eb_bh->b_blocknr);
2068                         ret = -EROFS;
2069                         goto out;
2070                 }
2071         }
2072
2073         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2074                 rec = &el->l_recs[i];
2075
2076                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2077                         e_blkno = le64_to_cpu(rec->e_blkno);
2078                         break;
2079                 }
2080         }
2081
2082         if (!e_blkno) {
2083                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2084                             "record (%u, %u, 0) in xattr", inode->i_ino,
2085                             le32_to_cpu(rec->e_cpos),
2086                             ocfs2_rec_clusters(el, rec));
2087                 ret = -EROFS;
2088                 goto out;
2089         }
2090
2091         *p_blkno = le64_to_cpu(rec->e_blkno);
2092         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2093         if (e_cpos)
2094                 *e_cpos = le32_to_cpu(rec->e_cpos);
2095 out:
2096         brelse(eb_bh);
2097         return ret;
2098 }
2099
2100 typedef int (xattr_bucket_func)(struct inode *inode,
2101                                 struct ocfs2_xattr_bucket *bucket,
2102                                 void *para);
2103
2104 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2105                                    struct buffer_head *header_bh,
2106                                    int name_index,
2107                                    const char *name,
2108                                    u32 name_hash,
2109                                    u16 *xe_index,
2110                                    int *found)
2111 {
2112         int i, ret = 0, cmp = 1, block_off, new_offset;
2113         struct ocfs2_xattr_header *xh =
2114                         (struct ocfs2_xattr_header *)header_bh->b_data;
2115         size_t name_len = strlen(name);
2116         struct ocfs2_xattr_entry *xe = NULL;
2117         struct buffer_head *name_bh = NULL;
2118         char *xe_name;
2119
2120         /*
2121          * We don't use binary search in the bucket because there
2122          * may be multiple entries with the same name hash.
2123          */
2124         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2125                 xe = &xh->xh_entries[i];
2126
2127                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2128                         continue;
2129                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2130                         break;
2131
2132                 cmp = name_index - ocfs2_xattr_get_type(xe);
2133                 if (!cmp)
2134                         cmp = name_len - xe->xe_name_len;
2135                 if (cmp)
2136                         continue;
2137
2138                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2139                                                         xh,
2140                                                         i,
2141                                                         &block_off,
2142                                                         &new_offset);
2143                 if (ret) {
2144                         mlog_errno(ret);
2145                         break;
2146                 }
2147
2148                 ret = ocfs2_read_block(inode, header_bh->b_blocknr + block_off,
2149                                        &name_bh);
2150                 if (ret) {
2151                         mlog_errno(ret);
2152                         break;
2153                 }
2154                 xe_name = name_bh->b_data + new_offset;
2155
2156                 cmp = memcmp(name, xe_name, name_len);
2157                 brelse(name_bh);
2158                 name_bh = NULL;
2159
2160                 if (cmp == 0) {
2161                         *xe_index = i;
2162                         *found = 1;
2163                         ret = 0;
2164                         break;
2165                 }
2166         }
2167
2168         return ret;
2169 }
2170
2171 /*
2172  * Find the specified xattr entry in a series of buckets.
2173  * This series start from p_blkno and last for num_clusters.
2174  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2175  * the num of the valid buckets.
2176  *
2177  * Return the buffer_head this xattr should reside in. And if the xattr's
2178  * hash is in the gap of 2 buckets, return the lower bucket.
2179  */
2180 static int ocfs2_xattr_bucket_find(struct inode *inode,
2181                                    int name_index,
2182                                    const char *name,
2183                                    u32 name_hash,
2184                                    u64 p_blkno,
2185                                    u32 first_hash,
2186                                    u32 num_clusters,
2187                                    struct ocfs2_xattr_search *xs)
2188 {
2189         int ret, found = 0;
2190         struct buffer_head *bh = NULL;
2191         struct buffer_head *lower_bh = NULL;
2192         struct ocfs2_xattr_header *xh = NULL;
2193         struct ocfs2_xattr_entry *xe = NULL;
2194         u16 index = 0;
2195         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2196         int low_bucket = 0, bucket, high_bucket;
2197         u32 last_hash;
2198         u64 blkno;
2199
2200         ret = ocfs2_read_block(inode, p_blkno, &bh);
2201         if (ret) {
2202                 mlog_errno(ret);
2203                 goto out;
2204         }
2205
2206         xh = (struct ocfs2_xattr_header *)bh->b_data;
2207         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2208
2209         while (low_bucket <= high_bucket) {
2210                 brelse(bh);
2211                 bh = NULL;
2212                 bucket = (low_bucket + high_bucket) / 2;
2213
2214                 blkno = p_blkno + bucket * blk_per_bucket;
2215
2216                 ret = ocfs2_read_block(inode, blkno, &bh);
2217                 if (ret) {
2218                         mlog_errno(ret);
2219                         goto out;
2220                 }
2221
2222                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2223                 xe = &xh->xh_entries[0];
2224                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2225                         high_bucket = bucket - 1;
2226                         continue;
2227                 }
2228
2229                 /*
2230                  * Check whether the hash of the last entry in our
2231                  * bucket is larger than the search one. for an empty
2232                  * bucket, the last one is also the first one.
2233                  */
2234                 if (xh->xh_count)
2235                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2236
2237                 last_hash = le32_to_cpu(xe->xe_name_hash);
2238
2239                 /* record lower_bh which may be the insert place. */
2240                 brelse(lower_bh);
2241                 lower_bh = bh;
2242                 bh = NULL;
2243
2244                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2245                         low_bucket = bucket + 1;
2246                         continue;
2247                 }
2248
2249                 /* the searched xattr should reside in this bucket if exists. */
2250                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2251                                               name_index, name, name_hash,
2252                                               &index, &found);
2253                 if (ret) {
2254                         mlog_errno(ret);
2255                         goto out;
2256                 }
2257                 break;
2258         }
2259
2260         /*
2261          * Record the bucket we have found.
2262          * When the xattr's hash value is in the gap of 2 buckets, we will
2263          * always set it to the previous bucket.
2264          */
2265         if (!lower_bh) {
2266                 /*
2267                  * We can't find any bucket whose first name_hash is less
2268                  * than the find name_hash.
2269                  */
2270                 BUG_ON(bh->b_blocknr != p_blkno);
2271                 lower_bh = bh;
2272                 bh = NULL;
2273         }
2274         xs->bucket.bhs[0] = lower_bh;
2275         xs->bucket.xh = (struct ocfs2_xattr_header *)
2276                                         xs->bucket.bhs[0]->b_data;
2277         lower_bh = NULL;
2278
2279         xs->header = xs->bucket.xh;
2280         xs->base = xs->bucket.bhs[0]->b_data;
2281         xs->end = xs->base + inode->i_sb->s_blocksize;
2282
2283         if (found) {
2284                 /*
2285                  * If we have found the xattr enty, read all the blocks in
2286                  * this bucket.
2287                  */
2288                 ret = ocfs2_read_blocks(inode, xs->bucket.bhs[0]->b_blocknr + 1,
2289                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2290                                         0);
2291                 if (ret) {
2292                         mlog_errno(ret);
2293                         goto out;
2294                 }
2295
2296                 xs->here = &xs->header->xh_entries[index];
2297                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2298                      (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2299         } else
2300                 ret = -ENODATA;
2301
2302 out:
2303         brelse(bh);
2304         brelse(lower_bh);
2305         return ret;
2306 }
2307
2308 static int ocfs2_xattr_index_block_find(struct inode *inode,
2309                                         struct buffer_head *root_bh,
2310                                         int name_index,
2311                                         const char *name,
2312                                         struct ocfs2_xattr_search *xs)
2313 {
2314         int ret;
2315         struct ocfs2_xattr_block *xb =
2316                         (struct ocfs2_xattr_block *)root_bh->b_data;
2317         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2318         struct ocfs2_extent_list *el = &xb_root->xt_list;
2319         u64 p_blkno = 0;
2320         u32 first_hash, num_clusters = 0;
2321         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2322
2323         if (le16_to_cpu(el->l_next_free_rec) == 0)
2324                 return -ENODATA;
2325
2326         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2327              name, name_hash, name_index);
2328
2329         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2330                                   &num_clusters, el);
2331         if (ret) {
2332                 mlog_errno(ret);
2333                 goto out;
2334         }
2335
2336         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2337
2338         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2339              "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2340
2341         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2342                                       p_blkno, first_hash, num_clusters, xs);
2343
2344 out:
2345         return ret;
2346 }
2347
2348 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2349                                        u64 blkno,
2350                                        u32 clusters,
2351                                        xattr_bucket_func *func,
2352                                        void *para)
2353 {
2354         int i, j, ret = 0;
2355         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2356         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2357         u32 num_buckets = clusters * bpc;
2358         struct ocfs2_xattr_bucket bucket;
2359
2360         memset(&bucket, 0, sizeof(bucket));
2361
2362         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2363              clusters, blkno);
2364
2365         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2366                 ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket,
2367                                         bucket.bhs, 0);
2368                 if (ret) {
2369                         mlog_errno(ret);
2370                         goto out;
2371                 }
2372
2373                 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2374                 /*
2375                  * The real bucket num in this series of blocks is stored
2376                  * in the 1st bucket.
2377                  */
2378                 if (i == 0)
2379                         num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2380
2381                 mlog(0, "iterating xattr bucket %llu, first hash %u\n", blkno,
2382                      le32_to_cpu(bucket.xh->xh_entries[0].xe_name_hash));
2383                 if (func) {
2384                         ret = func(inode, &bucket, para);
2385                         if (ret) {
2386                                 mlog_errno(ret);
2387                                 break;
2388                         }
2389                 }
2390
2391                 for (j = 0; j < blk_per_bucket; j++)
2392                         brelse(bucket.bhs[j]);
2393                 memset(&bucket, 0, sizeof(bucket));
2394         }
2395
2396 out:
2397         for (j = 0; j < blk_per_bucket; j++)
2398                 brelse(bucket.bhs[j]);
2399
2400         return ret;
2401 }
2402
2403 struct ocfs2_xattr_tree_list {
2404         char *buffer;
2405         size_t buffer_size;
2406         size_t result;
2407 };
2408
2409 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2410                                              struct ocfs2_xattr_header *xh,
2411                                              int index,
2412                                              int *block_off,
2413                                              int *new_offset)
2414 {
2415         u16 name_offset;
2416
2417         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2418                 return -EINVAL;
2419
2420         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2421
2422         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2423         *new_offset = name_offset % inode->i_sb->s_blocksize;
2424
2425         return 0;
2426 }
2427
2428 static int ocfs2_list_xattr_bucket(struct inode *inode,
2429                                    struct ocfs2_xattr_bucket *bucket,
2430                                    void *para)
2431 {
2432         int ret = 0, type;
2433         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2434         int i, block_off, new_offset;
2435         const char *prefix, *name;
2436
2437         for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2438                 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2439                 type = ocfs2_xattr_get_type(entry);
2440                 prefix = ocfs2_xattr_prefix(type);
2441
2442                 if (prefix) {
2443                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2444                                                                 bucket->xh,
2445                                                                 i,
2446                                                                 &block_off,
2447                                                                 &new_offset);
2448                         if (ret)
2449                                 break;
2450
2451                         name = (const char *)bucket->bhs[block_off]->b_data +
2452                                 new_offset;
2453                         ret = ocfs2_xattr_list_entry(xl->buffer,
2454                                                      xl->buffer_size,
2455                                                      &xl->result,
2456                                                      prefix, name,
2457                                                      entry->xe_name_len);
2458                         if (ret)
2459                                 break;
2460                 }
2461         }
2462
2463         return ret;
2464 }
2465
2466 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2467                                              struct ocfs2_xattr_tree_root *xt,
2468                                              char *buffer,
2469                                              size_t buffer_size)
2470 {
2471         struct ocfs2_extent_list *el = &xt->xt_list;
2472         int ret = 0;
2473         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2474         u64 p_blkno = 0;
2475         struct ocfs2_xattr_tree_list xl = {
2476                 .buffer = buffer,
2477                 .buffer_size = buffer_size,
2478                 .result = 0,
2479         };
2480
2481         if (le16_to_cpu(el->l_next_free_rec) == 0)
2482                 return 0;
2483
2484         while (name_hash > 0) {
2485                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2486                                           &e_cpos, &num_clusters, el);
2487                 if (ret) {
2488                         mlog_errno(ret);
2489                         goto out;
2490                 }
2491
2492                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2493                                                   ocfs2_list_xattr_bucket,
2494                                                   &xl);
2495                 if (ret) {
2496                         mlog_errno(ret);
2497                         goto out;
2498                 }
2499
2500                 if (e_cpos == 0)
2501                         break;
2502
2503                 name_hash = e_cpos - 1;
2504         }
2505
2506         ret = xl.result;
2507 out:
2508         return ret;
2509 }
2510
2511 static int cmp_xe(const void *a, const void *b)
2512 {
2513         const struct ocfs2_xattr_entry *l = a, *r = b;
2514         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2515         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2516
2517         if (l_hash > r_hash)
2518                 return 1;
2519         if (l_hash < r_hash)
2520                 return -1;
2521         return 0;
2522 }
2523
2524 static void swap_xe(void *a, void *b, int size)
2525 {
2526         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2527
2528         tmp = *l;
2529         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2530         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2531 }
2532
2533 /*
2534  * When the ocfs2_xattr_block is filled up, new bucket will be created
2535  * and all the xattr entries will be moved to the new bucket.
2536  * Note: we need to sort the entries since they are not saved in order
2537  * in the ocfs2_xattr_block.
2538  */
2539 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2540                                            struct buffer_head *xb_bh,
2541                                            struct buffer_head *xh_bh,
2542                                            struct buffer_head *data_bh)
2543 {
2544         int i, blocksize = inode->i_sb->s_blocksize;
2545         u16 offset, size, off_change;
2546         struct ocfs2_xattr_entry *xe;
2547         struct ocfs2_xattr_block *xb =
2548                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2549         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2550         struct ocfs2_xattr_header *xh =
2551                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2552         u16 count = le16_to_cpu(xb_xh->xh_count);
2553         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2554
2555         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2556              (unsigned long long)xb_bh->b_blocknr,
2557              (unsigned long long)xh_bh->b_blocknr);
2558
2559         memset(xh_bh->b_data, 0, blocksize);
2560         if (data_bh)
2561                 memset(data_bh->b_data, 0, blocksize);
2562         /*
2563          * Since the xe_name_offset is based on ocfs2_xattr_header,
2564          * there is a offset change corresponding to the change of
2565          * ocfs2_xattr_header's position.
2566          */
2567         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2568         xe = &xb_xh->xh_entries[count - 1];
2569         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2570         size = blocksize - offset;
2571
2572         /* copy all the names and values. */
2573         if (data_bh)
2574                 target = data_bh->b_data;
2575         memcpy(target + offset, src + offset, size);
2576
2577         /* Init new header now. */
2578         xh->xh_count = xb_xh->xh_count;
2579         xh->xh_num_buckets = cpu_to_le16(1);
2580         xh->xh_name_value_len = cpu_to_le16(size);
2581         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2582
2583         /* copy all the entries. */
2584         target = xh_bh->b_data;
2585         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2586         size = count * sizeof(struct ocfs2_xattr_entry);
2587         memcpy(target + offset, (char *)xb_xh + offset, size);
2588
2589         /* Change the xe offset for all the xe because of the move. */
2590         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2591                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2592         for (i = 0; i < count; i++)
2593                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2594
2595         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2596              offset, size, off_change);
2597
2598         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2599              cmp_xe, swap_xe);
2600 }
2601
2602 /*
2603  * After we move xattr from block to index btree, we have to
2604  * update ocfs2_xattr_search to the new xe and base.
2605  *
2606  * When the entry is in xattr block, xattr_bh indicates the storage place.
2607  * While if the entry is in index b-tree, "bucket" indicates the
2608  * real place of the xattr.
2609  */
2610 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2611                                            struct ocfs2_xattr_search *xs,
2612                                            struct buffer_head *old_bh,
2613                                            struct buffer_head *new_bh)
2614 {
2615         int ret = 0;
2616         char *buf = old_bh->b_data;
2617         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2618         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2619         int i, blocksize = inode->i_sb->s_blocksize;
2620         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2621
2622         xs->bucket.bhs[0] = new_bh;
2623         get_bh(new_bh);
2624         xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2625         xs->header = xs->bucket.xh;
2626
2627         xs->base = new_bh->b_data;
2628         xs->end = xs->base + inode->i_sb->s_blocksize;
2629
2630         if (!xs->not_found) {
2631                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2632                         ret = ocfs2_read_blocks(inode,
2633                                         xs->bucket.bhs[0]->b_blocknr + 1,
2634                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2635                                         0);
2636                         if (ret) {
2637                                 mlog_errno(ret);
2638                                 return ret;
2639                         }
2640
2641                         i = xs->here - old_xh->xh_entries;
2642                         xs->here = &xs->header->xh_entries[i];
2643                 }
2644         }
2645
2646         return ret;
2647 }
2648
2649 static int ocfs2_xattr_create_index_block(struct inode *inode,
2650                                           struct ocfs2_xattr_search *xs)
2651 {
2652         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2653         u32 bit_off, len;
2654         u64 blkno;
2655         handle_t *handle;
2656         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2657         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2658         struct ocfs2_alloc_context *data_ac;
2659         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2660         struct buffer_head *xb_bh = xs->xattr_bh;
2661         struct ocfs2_xattr_block *xb =
2662                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2663         struct ocfs2_xattr_tree_root *xr;
2664         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2665         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2666
2667         mlog(0, "create xattr index block for %llu\n",
2668              (unsigned long long)xb_bh->b_blocknr);
2669
2670         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2671
2672         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2673         if (ret) {
2674                 mlog_errno(ret);
2675                 goto out;
2676         }
2677
2678         /*
2679          * XXX:
2680          * We can use this lock for now, and maybe move to a dedicated mutex
2681          * if performance becomes a problem later.
2682          */
2683         down_write(&oi->ip_alloc_sem);
2684
2685         /*
2686          * 3 more credits, one for xattr block update, one for the 1st block
2687          * of the new xattr bucket and one for the value/data.
2688          */
2689         credits += 3;
2690         handle = ocfs2_start_trans(osb, credits);
2691         if (IS_ERR(handle)) {
2692                 ret = PTR_ERR(handle);
2693                 mlog_errno(ret);
2694                 goto out_sem;
2695         }
2696
2697         ret = ocfs2_journal_access(handle, inode, xb_bh,
2698                                    OCFS2_JOURNAL_ACCESS_WRITE);
2699         if (ret) {
2700                 mlog_errno(ret);
2701                 goto out_commit;
2702         }
2703
2704         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2705         if (ret) {
2706                 mlog_errno(ret);
2707                 goto out_commit;
2708         }
2709
2710         /*
2711          * The bucket may spread in many blocks, and
2712          * we will only touch the 1st block and the last block
2713          * in the whole bucket(one for entry and one for data).
2714          */
2715         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2716
2717         mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2718
2719         xh_bh = sb_getblk(inode->i_sb, blkno);
2720         if (!xh_bh) {
2721                 ret = -EIO;
2722                 mlog_errno(ret);
2723                 goto out_commit;
2724         }
2725
2726         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2727
2728         ret = ocfs2_journal_access(handle, inode, xh_bh,
2729                                    OCFS2_JOURNAL_ACCESS_CREATE);
2730         if (ret) {
2731                 mlog_errno(ret);
2732                 goto out_commit;
2733         }
2734
2735         if (bpb > 1) {
2736                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2737                 if (!data_bh) {
2738                         ret = -EIO;
2739                         mlog_errno(ret);
2740                         goto out_commit;
2741                 }
2742
2743                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2744
2745                 ret = ocfs2_journal_access(handle, inode, data_bh,
2746                                            OCFS2_JOURNAL_ACCESS_CREATE);
2747                 if (ret) {
2748                         mlog_errno(ret);
2749                         goto out_commit;
2750                 }
2751         }
2752
2753         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2754
2755         ocfs2_journal_dirty(handle, xh_bh);
2756         if (data_bh)
2757                 ocfs2_journal_dirty(handle, data_bh);
2758
2759         ret = ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2760         if (ret) {
2761                 mlog_errno(ret);
2762                 goto out_commit;
2763         }
2764
2765         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2766         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2767                offsetof(struct ocfs2_xattr_block, xb_attrs));
2768
2769         xr = &xb->xb_attrs.xb_root;
2770         xr->xt_clusters = cpu_to_le32(1);
2771         xr->xt_last_eb_blk = 0;
2772         xr->xt_list.l_tree_depth = 0;
2773         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2774         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2775
2776         xr->xt_list.l_recs[0].e_cpos = 0;
2777         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2778         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2779
2780         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2781
2782         ret = ocfs2_journal_dirty(handle, xb_bh);
2783         if (ret) {
2784                 mlog_errno(ret);
2785                 goto out_commit;
2786         }
2787
2788 out_commit:
2789         ocfs2_commit_trans(osb, handle);
2790
2791 out_sem:
2792         up_write(&oi->ip_alloc_sem);
2793
2794 out:
2795         if (data_ac)
2796                 ocfs2_free_alloc_context(data_ac);
2797
2798         brelse(xh_bh);
2799         brelse(data_bh);
2800
2801         return ret;
2802 }
2803
2804 static int cmp_xe_offset(const void *a, const void *b)
2805 {
2806         const struct ocfs2_xattr_entry *l = a, *r = b;
2807         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2808         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2809
2810         if (l_name_offset < r_name_offset)
2811                 return 1;
2812         if (l_name_offset > r_name_offset)
2813                 return -1;
2814         return 0;
2815 }
2816
2817 /*
2818  * defrag a xattr bucket if we find that the bucket has some
2819  * holes beteen name/value pairs.
2820  * We will move all the name/value pairs to the end of the bucket
2821  * so that we can spare some space for insertion.
2822  */
2823 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2824                                      struct ocfs2_xattr_bucket *bucket)
2825 {
2826         int ret, i;
2827         size_t end, offset, len, value_len;
2828         struct ocfs2_xattr_header *xh;
2829         char *entries, *buf, *bucket_buf = NULL;
2830         u64 blkno = bucket->bhs[0]->b_blocknr;
2831         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2832         u16 xh_free_start;
2833         size_t blocksize = inode->i_sb->s_blocksize;
2834         handle_t *handle;
2835         struct buffer_head **bhs;
2836         struct ocfs2_xattr_entry *xe;
2837
2838         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2839                         GFP_NOFS);
2840         if (!bhs)
2841                 return -ENOMEM;
2842
2843         ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket, bhs, 0);
2844         if (ret)
2845                 goto out;
2846
2847         /*
2848          * In order to make the operation more efficient and generic,
2849          * we copy all the blocks into a contiguous memory and do the
2850          * defragment there, so if anything is error, we will not touch
2851          * the real block.
2852          */
2853         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2854         if (!bucket_buf) {
2855                 ret = -EIO;
2856                 goto out;
2857         }
2858
2859         buf = bucket_buf;
2860         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2861                 memcpy(buf, bhs[i]->b_data, blocksize);
2862
2863         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2864         if (IS_ERR(handle)) {
2865                 ret = PTR_ERR(handle);
2866                 handle = NULL;
2867                 mlog_errno(ret);
2868                 goto out;
2869         }
2870
2871         for (i = 0; i < blk_per_bucket; i++) {
2872                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2873                                            OCFS2_JOURNAL_ACCESS_WRITE);
2874                 if (ret < 0) {
2875                         mlog_errno(ret);
2876                         goto commit;
2877                 }
2878         }
2879
2880         xh = (struct ocfs2_xattr_header *)bucket_buf;
2881         entries = (char *)xh->xh_entries;
2882         xh_free_start = le16_to_cpu(xh->xh_free_start);
2883
2884         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2885              "xh_free_start = %u, xh_name_value_len = %u.\n",
2886              blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2887              le16_to_cpu(xh->xh_name_value_len));
2888
2889         /*
2890          * sort all the entries by their offset.
2891          * the largest will be the first, so that we can
2892          * move them to the end one by one.
2893          */
2894         sort(entries, le16_to_cpu(xh->xh_count),
2895              sizeof(struct ocfs2_xattr_entry),
2896              cmp_xe_offset, swap_xe);
2897
2898         /* Move all name/values to the end of the bucket. */
2899         xe = xh->xh_entries;
2900         end = OCFS2_XATTR_BUCKET_SIZE;
2901         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2902                 offset = le16_to_cpu(xe->xe_name_offset);
2903                 if (ocfs2_xattr_is_local(xe))
2904                         value_len = OCFS2_XATTR_SIZE(
2905                                         le64_to_cpu(xe->xe_value_size));
2906                 else
2907                         value_len = OCFS2_XATTR_ROOT_SIZE;
2908                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2909
2910                 /*
2911                  * We must make sure that the name/value pair
2912                  * exist in the same block. So adjust end to
2913                  * the previous block end if needed.
2914                  */
2915                 if (((end - len) / blocksize !=
2916                         (end - 1) / blocksize))
2917                         end = end - end % blocksize;
2918
2919                 if (end > offset + len) {
2920                         memmove(bucket_buf + end - len,
2921                                 bucket_buf + offset, len);
2922                         xe->xe_name_offset = cpu_to_le16(end - len);
2923                 }
2924
2925                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2926                                 "bucket %llu\n", (unsigned long long)blkno);
2927
2928                 end -= len;
2929         }
2930
2931         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2932                         "bucket %llu\n", (unsigned long long)blkno);
2933
2934         if (xh_free_start == end)
2935                 goto commit;
2936
2937         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2938         xh->xh_free_start = cpu_to_le16(end);
2939
2940         /* sort the entries by their name_hash. */
2941         sort(entries, le16_to_cpu(xh->xh_count),
2942              sizeof(struct ocfs2_xattr_entry),
2943              cmp_xe, swap_xe);
2944
2945         buf = bucket_buf;
2946         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
2947                 memcpy(bhs[i]->b_data, buf, blocksize);
2948                 ocfs2_journal_dirty(handle, bhs[i]);
2949         }
2950
2951 commit:
2952         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2953 out:
2954
2955         if (bhs) {
2956                 for (i = 0; i < blk_per_bucket; i++)
2957                         brelse(bhs[i]);
2958         }
2959         kfree(bhs);
2960
2961         kfree(bucket_buf);
2962         return ret;
2963 }
2964
2965 /*
2966  * Move half nums of the xattr bucket in the previous cluster to this new
2967  * cluster. We only touch the last cluster of the previous extend record.
2968  *
2969  * first_bh is the first buffer_head of a series of bucket in the same
2970  * extent rec and header_bh is the header of one bucket in this cluster.
2971  * They will be updated if we move the data header_bh contains to the new
2972  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
2973  */
2974 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
2975                                                handle_t *handle,
2976                                                struct buffer_head **first_bh,
2977                                                struct buffer_head **header_bh,
2978                                                u64 new_blkno,
2979                                                u64 prev_blkno,
2980                                                u32 num_clusters,
2981                                                u32 *first_hash)
2982 {
2983         int i, ret, credits;
2984         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2985         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
2986         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
2987         int blocksize = inode->i_sb->s_blocksize;
2988         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
2989         struct ocfs2_xattr_header *new_xh;
2990         struct ocfs2_xattr_header *xh =
2991                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
2992
2993         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
2994         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
2995
2996         prev_bh = *first_bh;
2997         get_bh(prev_bh);
2998         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
2999
3000         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3001
3002         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3003              prev_blkno, new_blkno);
3004
3005         /*
3006          * We need to update the 1st half of the new cluster and
3007          * 1 more for the update of the 1st bucket of the previous
3008          * extent record.
3009          */
3010         credits = bpc / 2 + 1;
3011         ret = ocfs2_extend_trans(handle, credits);
3012         if (ret) {
3013                 mlog_errno(ret);
3014                 goto out;
3015         }
3016
3017         ret = ocfs2_journal_access(handle, inode, prev_bh,
3018                                    OCFS2_JOURNAL_ACCESS_WRITE);
3019         if (ret) {
3020                 mlog_errno(ret);
3021                 goto out;
3022         }
3023
3024         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3025                 old_bh = new_bh = NULL;
3026                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3027                 if (!new_bh) {
3028                         ret = -EIO;
3029                         mlog_errno(ret);
3030                         goto out;
3031                 }
3032
3033                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3034
3035                 ret = ocfs2_journal_access(handle, inode, new_bh,
3036                                            OCFS2_JOURNAL_ACCESS_CREATE);
3037                 if (ret < 0) {
3038                         mlog_errno(ret);
3039                         brelse(new_bh);
3040                         goto out;
3041                 }
3042
3043                 ret = ocfs2_read_block(inode, prev_blkno, &old_bh);
3044                 if (ret < 0) {
3045                         mlog_errno(ret);
3046                         brelse(new_bh);
3047                         goto out;
3048                 }
3049
3050                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3051
3052                 if (i == 0) {
3053                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3054                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3055
3056                         if (first_hash)
3057                                 *first_hash = le32_to_cpu(
3058                                         new_xh->xh_entries[0].xe_name_hash);
3059                         new_first_bh = new_bh;
3060                         get_bh(new_first_bh);
3061                 }
3062
3063                 ocfs2_journal_dirty(handle, new_bh);
3064
3065                 if (*header_bh == old_bh) {
3066                         brelse(*header_bh);
3067                         *header_bh = new_bh;
3068                         get_bh(*header_bh);
3069
3070                         brelse(*first_bh);
3071                         *first_bh = new_first_bh;
3072                         get_bh(*first_bh);
3073                 }
3074                 brelse(new_bh);
3075                 brelse(old_bh);
3076         }
3077
3078         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3079
3080         ocfs2_journal_dirty(handle, prev_bh);
3081 out:
3082         brelse(prev_bh);
3083         brelse(new_first_bh);
3084         return ret;
3085 }
3086
3087 static int ocfs2_read_xattr_bucket(struct inode *inode,
3088                                    u64 blkno,
3089                                    struct buffer_head **bhs,
3090                                    int new)
3091 {
3092         int ret = 0;
3093         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3094
3095         if (!new)
3096                 return ocfs2_read_blocks(inode, blkno,
3097                                          blk_per_bucket, bhs, 0);
3098
3099         for (i = 0; i < blk_per_bucket; i++) {
3100                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3101                 if (bhs[i] == NULL) {
3102                         ret = -EIO;
3103                         mlog_errno(ret);
3104                         break;
3105                 }
3106                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3107         }
3108
3109         return ret;
3110 }
3111
3112 /*
3113  * Find the suitable pos when we divide a bucket into 2.
3114  * We have to make sure the xattrs with the same hash value exist
3115  * in the same bucket.
3116  *
3117  * If this ocfs2_xattr_header covers more than one hash value, find a
3118  * place where the hash value changes.  Try to find the most even split.
3119  * The most common case is that all entries have different hash values,
3120  * and the first check we make will find a place to split.
3121  */
3122 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3123 {
3124         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3125         int count = le16_to_cpu(xh->xh_count);
3126         int delta, middle = count / 2;
3127
3128         /*
3129          * We start at the middle.  Each step gets farther away in both
3130          * directions.  We therefore hit the change in hash value
3131          * nearest to the middle.  Note that this loop does not execute for
3132          * count < 2.
3133          */
3134         for (delta = 0; delta < middle; delta++) {
3135                 /* Let's check delta earlier than middle */
3136                 if (cmp_xe(&entries[middle - delta - 1],
3137                            &entries[middle - delta]))
3138                         return middle - delta;
3139
3140                 /* For even counts, don't walk off the end */
3141                 if ((middle + delta + 1) == count)
3142                         continue;
3143
3144                 /* Now try delta past middle */
3145                 if (cmp_xe(&entries[middle + delta],
3146                            &entries[middle + delta + 1]))
3147                         return middle + delta + 1;
3148         }
3149
3150         /* Every entry had the same hash */
3151         return count;
3152 }
3153
3154 /*
3155  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3156  * first_hash will record the 1st hash of the new bucket.
3157  *
3158  * Normally half of the xattrs will be moved.  But we have to make
3159  * sure that the xattrs with the same hash value are stored in the
3160  * same bucket. If all the xattrs in this bucket have the same hash
3161  * value, the new bucket will be initialized as an empty one and the
3162  * first_hash will be initialized as (hash_value+1).
3163  */
3164 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3165                                     handle_t *handle,
3166                                     u64 blk,
3167                                     u64 new_blk,
3168                                     u32 *first_hash,
3169                                     int new_bucket_head)
3170 {
3171         int ret, i;
3172         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3173         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3174         struct buffer_head **s_bhs, **t_bhs = NULL;
3175         struct ocfs2_xattr_header *xh;
3176         struct ocfs2_xattr_entry *xe;
3177         int blocksize = inode->i_sb->s_blocksize;
3178
3179         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3180              blk, new_blk);
3181
3182         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3183         if (!s_bhs)
3184                 return -ENOMEM;
3185
3186         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3187         if (ret) {
3188                 mlog_errno(ret);
3189                 goto out;
3190         }
3191
3192         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3193                                    OCFS2_JOURNAL_ACCESS_WRITE);
3194         if (ret) {
3195                 mlog_errno(ret);
3196                 goto out;
3197         }
3198
3199         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3200         if (!t_bhs) {
3201                 ret = -ENOMEM;
3202                 goto out;
3203         }
3204
3205         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3206         if (ret) {
3207                 mlog_errno(ret);
3208                 goto out;
3209         }
3210
3211         for (i = 0; i < blk_per_bucket; i++) {
3212                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3213                                            new_bucket_head ?
3214                                            OCFS2_JOURNAL_ACCESS_CREATE :
3215                                            OCFS2_JOURNAL_ACCESS_WRITE);
3216                 if (ret) {
3217                         mlog_errno(ret);
3218                         goto out;
3219                 }
3220         }
3221
3222         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3223         count = le16_to_cpu(xh->xh_count);
3224         start = ocfs2_xattr_find_divide_pos(xh);
3225
3226         if (start == count) {
3227                 xe = &xh->xh_entries[start-1];
3228
3229                 /*
3230                  * initialized a new empty bucket here.
3231                  * The hash value is set as one larger than
3232                  * that of the last entry in the previous bucket.
3233                  */
3234                 for (i = 0; i < blk_per_bucket; i++)
3235                         memset(t_bhs[i]->b_data, 0, blocksize);
3236
3237                 xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3238                 xh->xh_free_start = cpu_to_le16(blocksize);
3239                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3240                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3241
3242                 goto set_num_buckets;
3243         }
3244
3245         /* copy the whole bucket to the new first. */
3246         for (i = 0; i < blk_per_bucket; i++)
3247                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3248
3249         /* update the new bucket. */
3250         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3251
3252         /*
3253          * Calculate the total name/value len and xh_free_start for
3254          * the old bucket first.
3255          */
3256         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3257         name_value_len = 0;
3258         for (i = 0; i < start; i++) {
3259                 xe = &xh->xh_entries[i];
3260                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3261                 if (ocfs2_xattr_is_local(xe))
3262                         xe_len +=
3263                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3264                 else
3265                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3266                 name_value_len += xe_len;
3267                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3268                         name_offset = le16_to_cpu(xe->xe_name_offset);
3269         }
3270
3271         /*
3272          * Now begin the modification to the new bucket.
3273          *
3274          * In the new bucket, We just move the xattr entry to the beginning
3275          * and don't touch the name/value. So there will be some holes in the
3276          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3277          * called.
3278          */
3279         xe = &xh->xh_entries[start];
3280         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3281         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3282              (int)((char *)xe - (char *)xh),
3283              (int)((char *)xh->xh_entries - (char *)xh));
3284         memmove((char *)xh->xh_entries, (char *)xe, len);
3285         xe = &xh->xh_entries[count - start];
3286         len = sizeof(struct ocfs2_xattr_entry) * start;
3287         memset((char *)xe, 0, len);
3288
3289         le16_add_cpu(&xh->xh_count, -start);
3290         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3291
3292         /* Calculate xh_free_start for the new bucket. */
3293         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3294         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3295                 xe = &xh->xh_entries[i];
3296                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3297                 if (ocfs2_xattr_is_local(xe))
3298                         xe_len +=
3299                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3300                 else
3301                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3302                 if (le16_to_cpu(xe->xe_name_offset) <
3303                     le16_to_cpu(xh->xh_free_start))
3304                         xh->xh_free_start = xe->xe_name_offset;
3305         }
3306
3307 set_num_buckets:
3308         /* set xh->xh_num_buckets for the new xh. */
3309         if (new_bucket_head)
3310                 xh->xh_num_buckets = cpu_to_le16(1);
3311         else
3312                 xh->xh_num_buckets = 0;
3313
3314         for (i = 0; i < blk_per_bucket; i++) {
3315                 ocfs2_journal_dirty(handle, t_bhs[i]);
3316                 if (ret)
3317                         mlog_errno(ret);
3318         }
3319
3320         /* store the first_hash of the new bucket. */
3321         if (first_hash)
3322                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3323
3324         /*
3325          * Now only update the 1st block of the old bucket.  If we
3326          * just added a new empty bucket, there is no need to modify
3327          * it.
3328          */
3329         if (start == count)
3330                 goto out;
3331
3332         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3333         memset(&xh->xh_entries[start], 0,
3334                sizeof(struct ocfs2_xattr_entry) * (count - start));
3335         xh->xh_count = cpu_to_le16(start);
3336         xh->xh_free_start = cpu_to_le16(name_offset);
3337         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3338
3339         ocfs2_journal_dirty(handle, s_bhs[0]);
3340         if (ret)
3341                 mlog_errno(ret);
3342
3343 out:
3344         if (s_bhs) {
3345                 for (i = 0; i < blk_per_bucket; i++)
3346                         brelse(s_bhs[i]);
3347         }
3348         kfree(s_bhs);
3349
3350         if (t_bhs) {
3351                 for (i = 0; i < blk_per_bucket; i++)
3352                         brelse(t_bhs[i]);
3353         }
3354         kfree(t_bhs);
3355
3356         return ret;
3357 }
3358
3359 /*
3360  * Copy xattr from one bucket to another bucket.
3361  *
3362  * The caller must make sure that the journal transaction
3363  * has enough space for journaling.
3364  */
3365 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3366                                  handle_t *handle,
3367                                  u64 s_blkno,
3368                                  u64 t_blkno,
3369                                  int t_is_new)
3370 {
3371         int ret, i;
3372         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3373         int blocksize = inode->i_sb->s_blocksize;
3374         struct buffer_head **s_bhs, **t_bhs = NULL;
3375
3376         BUG_ON(s_blkno == t_blkno);
3377
3378         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3379              s_blkno, t_blkno, t_is_new);
3380
3381         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3382                         GFP_NOFS);
3383         if (!s_bhs)
3384                 return -ENOMEM;
3385
3386         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3387         if (ret)
3388                 goto out;
3389
3390         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3391                         GFP_NOFS);
3392         if (!t_bhs) {
3393                 ret = -ENOMEM;
3394                 goto out;
3395         }
3396
3397         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3398         if (ret)
3399                 goto out;
3400
3401         for (i = 0; i < blk_per_bucket; i++) {
3402                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3403                                            t_is_new ?
3404                                            OCFS2_JOURNAL_ACCESS_CREATE :
3405                                            OCFS2_JOURNAL_ACCESS_WRITE);
3406                 if (ret)
3407                         goto out;
3408         }
3409
3410         for (i = 0; i < blk_per_bucket; i++) {
3411                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3412                 ocfs2_journal_dirty(handle, t_bhs[i]);
3413         }
3414
3415 out:
3416         if (s_bhs) {
3417                 for (i = 0; i < blk_per_bucket; i++)
3418                         brelse(s_bhs[i]);
3419         }
3420         kfree(s_bhs);
3421
3422         if (t_bhs) {
3423                 for (i = 0; i < blk_per_bucket; i++)
3424                         brelse(t_bhs[i]);
3425         }
3426         kfree(t_bhs);
3427
3428         return ret;
3429 }
3430
3431 /*
3432  * Copy one xattr cluster from src_blk to to_blk.
3433  * The to_blk will become the first bucket header of the cluster, so its
3434  * xh_num_buckets will be initialized as the bucket num in the cluster.
3435  */
3436 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3437                                   handle_t *handle,
3438                                   struct buffer_head *first_bh,
3439                                   u64 src_blk,
3440                                   u64 to_blk,
3441                                   u32 *first_hash)
3442 {
3443         int i, ret, credits;
3444         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3445         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3446         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3447         struct buffer_head *bh = NULL;
3448         struct ocfs2_xattr_header *xh;
3449         u64 to_blk_start = to_blk;
3450
3451         mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3452
3453         /*
3454          * We need to update the new cluster and 1 more for the update of
3455          * the 1st bucket of the previous extent rec.
3456          */
3457         credits = bpc + 1;
3458         ret = ocfs2_extend_trans(handle, credits);
3459         if (ret) {
3460                 mlog_errno(ret);
3461                 goto out;
3462         }
3463
3464         ret = ocfs2_journal_access(handle, inode, first_bh,
3465                                    OCFS2_JOURNAL_ACCESS_WRITE);
3466         if (ret) {
3467                 mlog_errno(ret);
3468                 goto out;
3469         }
3470
3471         for (i = 0; i < num_buckets; i++) {
3472                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3473                                             src_blk, to_blk, 1);
3474                 if (ret) {
3475                         mlog_errno(ret);
3476                         goto out;
3477                 }
3478
3479                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3480                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3481         }
3482
3483         /* update the old bucket header. */
3484         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3485         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3486
3487         ocfs2_journal_dirty(handle, first_bh);
3488
3489         /* update the new bucket header. */
3490         ret = ocfs2_read_block(inode, to_blk_start, &bh);
3491         if (ret < 0) {
3492                 mlog_errno(ret);
3493                 goto out;
3494         }
3495
3496         ret = ocfs2_journal_access(handle, inode, bh,
3497                                    OCFS2_JOURNAL_ACCESS_WRITE);
3498         if (ret) {
3499                 mlog_errno(ret);
3500                 goto out;
3501         }
3502
3503         xh = (struct ocfs2_xattr_header *)bh->b_data;
3504         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3505
3506         ocfs2_journal_dirty(handle, bh);
3507
3508         if (first_hash)
3509                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3510 out:
3511         brelse(bh);
3512         return ret;
3513 }
3514
3515 /*
3516  * Move some xattrs in this cluster to the new cluster.
3517  * This function should only be called when bucket size == cluster size.
3518  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3519  */
3520 static int ocfs2_divide_xattr_cluster(struct inode *inode,
3521                                       handle_t *handle,
3522                                       u64 prev_blk,
3523                                       u64 new_blk,
3524                                       u32 *first_hash)
3525 {
3526         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3527         int ret, credits = 2 * blk_per_bucket;
3528
3529         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3530
3531         ret = ocfs2_extend_trans(handle, credits);
3532         if (ret) {
3533                 mlog_errno(ret);
3534                 return ret;
3535         }
3536
3537         /* Move half of the xattr in start_blk to the next bucket. */
3538         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
3539                                           new_blk, first_hash, 1);
3540 }
3541
3542 /*
3543  * Move some xattrs from the old cluster to the new one since they are not
3544  * contiguous in ocfs2 xattr tree.
3545  *
3546  * new_blk starts a new separate cluster, and we will move some xattrs from
3547  * prev_blk to it. v_start will be set as the first name hash value in this
3548  * new cluster so that it can be used as e_cpos during tree insertion and
3549  * don't collide with our original b-tree operations. first_bh and header_bh
3550  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3551  * to extend the insert bucket.
3552  *
3553  * The problem is how much xattr should we move to the new one and when should
3554  * we update first_bh and header_bh?
3555  * 1. If cluster size > bucket size, that means the previous cluster has more
3556  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3557  *    update the first_bh and header_bh if the insert bucket has been moved
3558  *    to the new cluster.
3559  * 2. If cluster_size == bucket_size:
3560  *    a) If the previous extent rec has more than one cluster and the insert
3561  *       place isn't in the last cluster, copy the entire last cluster to the
3562  *       new one. This time, we don't need to upate the first_bh and header_bh
3563  *       since they will not be moved into the new cluster.
3564  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3565  *       the new one. And we set the extend flag to zero if the insert place is
3566  *       moved into the new allocated cluster since no extend is needed.
3567  */
3568 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3569                                             handle_t *handle,
3570                                             struct buffer_head **first_bh,
3571                                             struct buffer_head **header_bh,
3572                                             u64 new_blk,
3573                                             u64 prev_blk,
3574                                             u32 prev_clusters,
3575                                             u32 *v_start,
3576                                             int *extend)
3577 {
3578         int ret = 0;
3579         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3580
3581         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3582              prev_blk, prev_clusters, new_blk);
3583
3584         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3585                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3586                                                           handle,
3587                                                           first_bh,
3588                                                           header_bh,
3589                                                           new_blk,
3590                                                           prev_blk,
3591                                                           prev_clusters,
3592                                                           v_start);
3593         else {
3594                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3595
3596                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3597                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3598                                                      last_blk, new_blk,
3599                                                      v_start);
3600                 else {
3601                         ret = ocfs2_divide_xattr_cluster(inode, handle,
3602                                                          last_blk, new_blk,
3603                                                          v_start);
3604
3605                         if ((*header_bh)->b_blocknr == last_blk && extend)
3606                                 *extend = 0;
3607                 }
3608         }
3609
3610         return ret;
3611 }
3612
3613 /*
3614  * Add a new cluster for xattr storage.
3615  *
3616  * If the new cluster is contiguous with the previous one, it will be
3617  * appended to the same extent record, and num_clusters will be updated.
3618  * If not, we will insert a new extent for it and move some xattrs in
3619  * the last cluster into the new allocated one.
3620  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3621  * lose the benefits of hashing because we'll have to search large leaves.
3622  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3623  * if it's bigger).
3624  *
3625  * first_bh is the first block of the previous extent rec and header_bh
3626  * indicates the bucket we will insert the new xattrs. They will be updated
3627  * when the header_bh is moved into the new cluster.
3628  */
3629 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3630                                        struct buffer_head *root_bh,
3631                                        struct buffer_head **first_bh,
3632                                        struct buffer_head **header_bh,
3633                                        u32 *num_clusters,
3634                                        u32 prev_cpos,
3635                                        u64 prev_blkno,
3636                                        int *extend)
3637 {
3638         int ret, credits;
3639         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3640         u32 prev_clusters = *num_clusters;
3641         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3642         u64 block;
3643         handle_t *handle = NULL;
3644         struct ocfs2_alloc_context *data_ac = NULL;
3645         struct ocfs2_alloc_context *meta_ac = NULL;
3646         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3647         struct ocfs2_extent_tree et;
3648
3649         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3650              "previous xattr blkno = %llu\n",
3651              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3652              prev_cpos, prev_blkno);
3653
3654         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3655
3656         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3657                                     &data_ac, &meta_ac);
3658         if (ret) {
3659                 mlog_errno(ret);
3660                 goto leave;
3661         }
3662
3663         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3664                                             clusters_to_add);
3665         handle = ocfs2_start_trans(osb, credits);
3666         if (IS_ERR(handle)) {
3667                 ret = PTR_ERR(handle);
3668                 handle = NULL;
3669                 mlog_errno(ret);
3670                 goto leave;
3671         }
3672
3673         ret = ocfs2_journal_access(handle, inode, root_bh,
3674                                    OCFS2_JOURNAL_ACCESS_WRITE);
3675         if (ret < 0) {
3676                 mlog_errno(ret);
3677                 goto leave;
3678         }
3679
3680         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3681                                      clusters_to_add, &bit_off, &num_bits);
3682         if (ret < 0) {
3683                 if (ret != -ENOSPC)
3684                         mlog_errno(ret);
3685                 goto leave;
3686         }
3687
3688         BUG_ON(num_bits > clusters_to_add);
3689
3690         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3691         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3692              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3693
3694         if (prev_blkno + prev_clusters * bpc == block &&
3695             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3696              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3697                 /*
3698                  * If this cluster is contiguous with the old one and
3699                  * adding this new cluster, we don't surpass the limit of
3700                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3701                  * initialized and used like other buckets in the previous
3702                  * cluster.
3703                  * So add it as a contiguous one. The caller will handle
3704                  * its init process.
3705                  */
3706                 v_start = prev_cpos + prev_clusters;
3707                 *num_clusters = prev_clusters + num_bits;
3708                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3709                      num_bits);
3710         } else {
3711                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3712                                                        handle,
3713                                                        first_bh,
3714                                                        header_bh,
3715                                                        block,
3716                                                        prev_blkno,
3717                                                        prev_clusters,
3718                                                        &v_start,
3719                                                        extend);
3720                 if (ret) {
3721                         mlog_errno(ret);
3722                         goto leave;
3723                 }
3724         }
3725
3726         if (handle->h_buffer_credits < credits) {
3727                 /*
3728                  * The journal has been restarted before, and don't
3729                  * have enough space for the insertion, so extend it
3730                  * here.
3731                  */
3732                 ret = ocfs2_extend_trans(handle, credits);
3733                 if (ret) {
3734                         mlog_errno(ret);
3735                         goto leave;
3736                 }
3737         }
3738         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3739              num_bits, block, v_start);
3740         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3741                                   num_bits, 0, meta_ac);
3742         if (ret < 0) {
3743                 mlog_errno(ret);
3744                 goto leave;
3745         }
3746
3747         ret = ocfs2_journal_dirty(handle, root_bh);
3748         if (ret < 0) {
3749                 mlog_errno(ret);
3750                 goto leave;
3751         }
3752
3753 leave:
3754         if (handle)
3755                 ocfs2_commit_trans(osb, handle);
3756         if (data_ac)
3757                 ocfs2_free_alloc_context(data_ac);
3758         if (meta_ac)
3759                 ocfs2_free_alloc_context(meta_ac);
3760
3761         return ret;
3762 }
3763
3764 /*
3765  * Extend a new xattr bucket and move xattrs to the end one by one until
3766  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3767  */
3768 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3769                                      struct buffer_head *first_bh,
3770                                      struct buffer_head *start_bh,
3771                                      u32 num_clusters)
3772 {
3773         int ret, credits;
3774         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3775         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3776         u64 start_blk = start_bh->b_blocknr, end_blk;
3777         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3778         handle_t *handle;
3779         struct ocfs2_xattr_header *first_xh =
3780                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3781         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3782
3783         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3784              "from %llu, len = %u\n", start_blk,
3785              (unsigned long long)first_bh->b_blocknr, num_clusters);
3786
3787         BUG_ON(bucket >= num_buckets);
3788
3789         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3790
3791         /*
3792          * We will touch all the buckets after the start_bh(include it).
3793          * Add one more bucket and modify the first_bh.
3794          */
3795         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3796         handle = ocfs2_start_trans(osb, credits);
3797         if (IS_ERR(handle)) {
3798                 ret = PTR_ERR(handle);
3799                 handle = NULL;
3800                 mlog_errno(ret);
3801                 goto out;
3802         }
3803
3804         ret = ocfs2_journal_access(handle, inode, first_bh,
3805                                    OCFS2_JOURNAL_ACCESS_WRITE);
3806         if (ret) {
3807                 mlog_errno(ret);
3808                 goto commit;
3809         }
3810
3811         while (end_blk != start_blk) {
3812                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3813                                             end_blk + blk_per_bucket, 0);
3814                 if (ret)
3815                         goto commit;
3816                 end_blk -= blk_per_bucket;
3817         }
3818
3819         /* Move half of the xattr in start_blk to the next bucket. */
3820         ret = ocfs2_divide_xattr_bucket(inode, handle, start_blk,
3821                                         start_blk + blk_per_bucket, NULL, 0);
3822
3823         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3824         ocfs2_journal_dirty(handle, first_bh);
3825
3826 commit:
3827         ocfs2_commit_trans(osb, handle);
3828 out:
3829         return ret;
3830 }
3831
3832 /*
3833  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3834  * xb_bh is the ocfs2_xattr_block.
3835  * We will move all the buckets starting from header_bh to the next place. As
3836  * for this one, half num of its xattrs will be moved to the next one.
3837  *
3838  * We will allocate a new cluster if current cluster is full and adjust
3839  * header_bh and first_bh if the insert place is moved to the new cluster.
3840  */
3841 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3842                                       struct buffer_head *xb_bh,
3843                                       struct buffer_head *header_bh)
3844 {
3845         struct ocfs2_xattr_header *first_xh = NULL;
3846         struct buffer_head *first_bh = NULL;
3847         struct ocfs2_xattr_block *xb =
3848                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3849         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3850         struct ocfs2_extent_list *el = &xb_root->xt_list;
3851         struct ocfs2_xattr_header *xh =
3852                         (struct ocfs2_xattr_header *)header_bh->b_data;
3853         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3854         struct super_block *sb = inode->i_sb;
3855         struct ocfs2_super *osb = OCFS2_SB(sb);
3856         int ret, num_buckets, extend = 1;
3857         u64 p_blkno;
3858         u32 e_cpos, num_clusters;
3859
3860         mlog(0, "Add new xattr bucket starting form %llu\n",
3861              (unsigned long long)header_bh->b_blocknr);
3862
3863         /*
3864          * Add refrence for header_bh here because it may be
3865          * changed in ocfs2_add_new_xattr_cluster and we need
3866          * to free it in the end.
3867          */
3868         get_bh(header_bh);
3869
3870         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3871                                   &num_clusters, el);
3872         if (ret) {
3873                 mlog_errno(ret);
3874                 goto out;
3875         }
3876
3877         ret = ocfs2_read_block(inode, p_blkno, &first_bh);
3878         if (ret) {
3879                 mlog_errno(ret);
3880                 goto out;
3881         }
3882
3883         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3884         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3885
3886         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3887                 ret = ocfs2_add_new_xattr_cluster(inode,
3888                                                   xb_bh,
3889                                                   &first_bh,
3890                                                   &header_bh,
3891                                                   &num_clusters,
3892                                                   e_cpos,
3893                                                   p_blkno,
3894                                                   &extend);
3895                 if (ret) {
3896                         mlog_errno(ret);
3897                         goto out;
3898                 }
3899         }
3900
3901         if (extend)
3902                 ret = ocfs2_extend_xattr_bucket(inode,
3903                                                 first_bh,
3904                                                 header_bh,
3905                                                 num_clusters);
3906         if (ret)
3907                 mlog_errno(ret);
3908 out:
3909         brelse(first_bh);
3910         brelse(header_bh);
3911         return ret;
3912 }
3913
3914 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3915                                         struct ocfs2_xattr_bucket *bucket,
3916                                         int offs)
3917 {
3918         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3919
3920         offs = offs % inode->i_sb->s_blocksize;
3921         return bucket->bhs[block_off]->b_data + offs;
3922 }
3923
3924 /*
3925  * Handle the normal xattr set, including replace, delete and new.
3926  *
3927  * Note: "local" indicates the real data's locality. So we can't
3928  * just its bucket locality by its length.
3929  */
3930 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3931                                          struct ocfs2_xattr_info *xi,
3932                                          struct ocfs2_xattr_search *xs,
3933                                          u32 name_hash,
3934                                          int local)
3935 {
3936         struct ocfs2_xattr_entry *last, *xe;
3937         int name_len = strlen(xi->name);
3938         struct ocfs2_xattr_header *xh = xs->header;
3939         u16 count = le16_to_cpu(xh->xh_count), start;
3940         size_t blocksize = inode->i_sb->s_blocksize;
3941         char *val;
3942         size_t offs, size, new_size;
3943
3944         last = &xh->xh_entries[count];
3945         if (!xs->not_found) {
3946                 xe = xs->here;
3947                 offs = le16_to_cpu(xe->xe_name_offset);
3948                 if (ocfs2_xattr_is_local(xe))
3949                         size = OCFS2_XATTR_SIZE(name_len) +
3950                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3951                 else
3952                         size = OCFS2_XATTR_SIZE(name_len) +
3953                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3954
3955                 /*
3956                  * If the new value will be stored outside, xi->value has been
3957                  * initalized as an empty ocfs2_xattr_value_root, and the same
3958                  * goes with xi->value_len, so we can set new_size safely here.
3959                  * See ocfs2_xattr_set_in_bucket.
3960                  */
3961                 new_size = OCFS2_XATTR_SIZE(name_len) +
3962                            OCFS2_XATTR_SIZE(xi->value_len);
3963
3964                 le16_add_cpu(&xh->xh_name_value_len, -size);
3965                 if (xi->value) {
3966                         if (new_size > size)
3967                                 goto set_new_name_value;
3968
3969                         /* Now replace the old value with new one. */
3970                         if (local)
3971                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3972                         else
3973                                 xe->xe_value_size = 0;
3974
3975                         val = ocfs2_xattr_bucket_get_val(inode,
3976                                                          &xs->bucket, offs);
3977                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3978                                size - OCFS2_XATTR_SIZE(name_len));
3979                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3980                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3981                                        xi->value, xi->value_len);
3982
3983                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3984                         ocfs2_xattr_set_local(xe, local);
3985                         return;
3986                 } else {
3987                         /*
3988                          * Remove the old entry if there is more than one.
3989                          * We don't remove the last entry so that we can
3990                          * use it to indicate the hash value of the empty
3991                          * bucket.
3992                          */
3993                         last -= 1;
3994                         le16_add_cpu(&xh->xh_count, -1);
3995                         if (xh->xh_count) {
3996                                 memmove(xe, xe + 1,
3997                                         (void *)last - (void *)xe);
3998                                 memset(last, 0,
3999                                        sizeof(struct ocfs2_xattr_entry));
4000                         } else
4001                                 xh->xh_free_start =
4002                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4003
4004                         return;
4005                 }
4006         } else {
4007                 /* find a new entry for insert. */
4008                 int low = 0, high = count - 1, tmp;
4009                 struct ocfs2_xattr_entry *tmp_xe;
4010
4011                 while (low <= high && count) {
4012                         tmp = (low + high) / 2;
4013                         tmp_xe = &xh->xh_entries[tmp];
4014
4015                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4016                                 low = tmp + 1;
4017                         else if (name_hash <
4018                                  le32_to_cpu(tmp_xe->xe_name_hash))
4019                                 high = tmp - 1;
4020                         else {
4021                                 low = tmp;
4022                                 break;
4023                         }
4024                 }
4025
4026                 xe = &xh->xh_entries[low];
4027                 if (low != count)
4028                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4029
4030                 le16_add_cpu(&xh->xh_count, 1);
4031                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4032                 xe->xe_name_hash = cpu_to_le32(name_hash);
4033                 xe->xe_name_len = name_len;
4034                 ocfs2_xattr_set_type(xe, xi->name_index);
4035         }
4036
4037 set_new_name_value:
4038         /* Insert the new name+value. */
4039         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4040
4041         /*
4042          * We must make sure that the name/value pair
4043          * exists in the same block.
4044          */
4045         offs = le16_to_cpu(xh->xh_free_start);
4046         start = offs - size;
4047
4048         if (start >> inode->i_sb->s_blocksize_bits !=
4049             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4050                 offs = offs - offs % blocksize;
4051                 xh->xh_free_start = cpu_to_le16(offs);
4052         }
4053
4054         val = ocfs2_xattr_bucket_get_val(inode,
4055                                          &xs->bucket, offs - size);
4056         xe->xe_name_offset = cpu_to_le16(offs - size);
4057
4058         memset(val, 0, size);
4059         memcpy(val, xi->name, name_len);
4060         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4061
4062         xe->xe_value_size = cpu_to_le64(xi->value_len);
4063         ocfs2_xattr_set_local(xe, local);
4064         xs->here = xe;
4065         le16_add_cpu(&xh->xh_free_start, -size);
4066         le16_add_cpu(&xh->xh_name_value_len, size);
4067
4068         return;
4069 }
4070
4071 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4072                                              handle_t *handle,
4073                                              struct ocfs2_xattr_search *xs,
4074                                              struct buffer_head **bhs,
4075                                              u16 bh_num)
4076 {
4077         int ret = 0, off, block_off;
4078         struct ocfs2_xattr_entry *xe = xs->here;
4079
4080         /*
4081          * First calculate all the blocks we should journal_access
4082          * and journal_dirty. The first block should always be touched.
4083          */
4084         ret = ocfs2_journal_dirty(handle, bhs[0]);
4085         if (ret)
4086                 mlog_errno(ret);
4087
4088         /* calc the data. */
4089         off = le16_to_cpu(xe->xe_name_offset);
4090         block_off = off >> inode->i_sb->s_blocksize_bits;
4091         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4092         if (ret)
4093                 mlog_errno(ret);
4094
4095         return ret;
4096 }
4097
4098 /*
4099  * Set the xattr entry in the specified bucket.
4100  * The bucket is indicated by xs->bucket and it should have the enough
4101  * space for the xattr insertion.
4102  */
4103 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4104                                            struct ocfs2_xattr_info *xi,
4105                                            struct ocfs2_xattr_search *xs,
4106                                            u32 name_hash,
4107                                            int local)
4108 {
4109         int i, ret;
4110         handle_t *handle = NULL;
4111         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4112         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4113
4114         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4115              (unsigned long)xi->value_len, xi->name_index,
4116              (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4117
4118         if (!xs->bucket.bhs[1]) {
4119                 ret = ocfs2_read_blocks(inode,
4120                                         xs->bucket.bhs[0]->b_blocknr + 1,
4121                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
4122                                         0);
4123                 if (ret) {
4124                         mlog_errno(ret);
4125                         goto out;
4126                 }
4127         }
4128
4129         handle = ocfs2_start_trans(osb, blk_per_bucket);
4130         if (IS_ERR(handle)) {
4131                 ret = PTR_ERR(handle);
4132                 handle = NULL;
4133                 mlog_errno(ret);
4134                 goto out;
4135         }
4136
4137         for (i = 0; i < blk_per_bucket; i++) {
4138                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4139                                            OCFS2_JOURNAL_ACCESS_WRITE);
4140                 if (ret < 0) {
4141                         mlog_errno(ret);
4142                         goto out;
4143                 }
4144         }
4145
4146         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4147
4148         /*Only dirty the blocks we have touched in set xattr. */
4149         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4150                                                 xs->bucket.bhs, blk_per_bucket);
4151         if (ret)
4152                 mlog_errno(ret);
4153 out:
4154         ocfs2_commit_trans(osb, handle);
4155
4156         return ret;
4157 }
4158
4159 static int ocfs2_xattr_value_update_size(struct inode *inode,
4160                                          struct buffer_head *xe_bh,
4161                                          struct ocfs2_xattr_entry *xe,
4162                                          u64 new_size)
4163 {
4164         int ret;
4165         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4166         handle_t *handle = NULL;
4167
4168         handle = ocfs2_start_trans(osb, 1);
4169         if (IS_ERR(handle)) {
4170                 ret = -ENOMEM;
4171                 mlog_errno(ret);
4172                 goto out;
4173         }
4174
4175         ret = ocfs2_journal_access(handle, inode, xe_bh,
4176                                    OCFS2_JOURNAL_ACCESS_WRITE);
4177         if (ret < 0) {
4178                 mlog_errno(ret);
4179                 goto out_commit;
4180         }
4181
4182         xe->xe_value_size = cpu_to_le64(new_size);
4183
4184         ret = ocfs2_journal_dirty(handle, xe_bh);
4185         if (ret < 0)
4186                 mlog_errno(ret);
4187
4188 out_commit:
4189         ocfs2_commit_trans(osb, handle);
4190 out:
4191         return ret;
4192 }
4193
4194 /*
4195  * Truncate the specified xe_off entry in xattr bucket.
4196  * bucket is indicated by header_bh and len is the new length.
4197  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4198  *
4199  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4200  */
4201 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4202                                              struct buffer_head *header_bh,
4203                                              int xe_off,
4204                                              int len)
4205 {
4206         int ret, offset;
4207         u64 value_blk;
4208         struct buffer_head *value_bh = NULL;
4209         struct ocfs2_xattr_value_root *xv;
4210         struct ocfs2_xattr_entry *xe;
4211         struct ocfs2_xattr_header *xh =
4212                         (struct ocfs2_xattr_header *)header_bh->b_data;
4213         size_t blocksize = inode->i_sb->s_blocksize;
4214
4215         xe = &xh->xh_entries[xe_off];
4216
4217         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4218
4219         offset = le16_to_cpu(xe->xe_name_offset) +
4220                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4221
4222         value_blk = offset / blocksize;
4223
4224         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4225         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4226         value_blk += header_bh->b_blocknr;
4227
4228         ret = ocfs2_read_block(inode, value_blk, &value_bh);
4229         if (ret) {
4230                 mlog_errno(ret);
4231                 goto out;
4232         }
4233
4234         xv = (struct ocfs2_xattr_value_root *)
4235                 (value_bh->b_data + offset % blocksize);
4236
4237         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4238              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4239         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4240         if (ret) {
4241                 mlog_errno(ret);
4242                 goto out;
4243         }
4244
4245         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4246         if (ret) {
4247                 mlog_errno(ret);
4248                 goto out;
4249         }
4250
4251 out:
4252         brelse(value_bh);
4253         return ret;
4254 }
4255
4256 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4257                                                 struct ocfs2_xattr_search *xs,
4258                                                 int len)
4259 {
4260         int ret, offset;
4261         struct ocfs2_xattr_entry *xe = xs->here;
4262         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4263
4264         BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4265
4266         offset = xe - xh->xh_entries;
4267         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4268                                                 offset, len);
4269         if (ret)
4270                 mlog_errno(ret);
4271
4272         return ret;
4273 }
4274
4275 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4276                                                 struct ocfs2_xattr_search *xs,
4277                                                 char *val,
4278                                                 int value_len)
4279 {
4280         int offset;
4281         struct ocfs2_xattr_value_root *xv;
4282         struct ocfs2_xattr_entry *xe = xs->here;
4283
4284         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4285
4286         offset = le16_to_cpu(xe->xe_name_offset) +
4287                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4288
4289         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4290
4291         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4292 }
4293
4294 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4295                                   struct buffer_head *root_bh,
4296                                   u64 blkno,
4297                                   u32 cpos,
4298                                   u32 len)
4299 {
4300         int ret;
4301         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4302         struct inode *tl_inode = osb->osb_tl_inode;
4303         handle_t *handle;
4304         struct ocfs2_xattr_block *xb =
4305                         (struct ocfs2_xattr_block *)root_bh->b_data;
4306         struct ocfs2_alloc_context *meta_ac = NULL;
4307         struct ocfs2_cached_dealloc_ctxt dealloc;
4308         struct ocfs2_extent_tree et;
4309
4310         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4311
4312         ocfs2_init_dealloc_ctxt(&dealloc);
4313
4314         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4315              cpos, len, (unsigned long long)blkno);
4316
4317         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4318
4319         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4320         if (ret) {
4321                 mlog_errno(ret);
4322                 return ret;
4323         }
4324
4325         mutex_lock(&tl_inode->i_mutex);
4326
4327         if (ocfs2_truncate_log_needs_flush(osb)) {
4328                 ret = __ocfs2_flush_truncate_log(osb);
4329                 if (ret < 0) {
4330                         mlog_errno(ret);
4331                         goto out;
4332                 }
4333         }
4334
4335         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4336         if (IS_ERR(handle)) {
4337                 ret = -ENOMEM;
4338                 mlog_errno(ret);
4339                 goto out;
4340         }
4341
4342         ret = ocfs2_journal_access(handle, inode, root_bh,
4343                                    OCFS2_JOURNAL_ACCESS_WRITE);
4344         if (ret) {
4345                 mlog_errno(ret);
4346                 goto out_commit;
4347         }
4348
4349         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4350                                   &dealloc);
4351         if (ret) {
4352                 mlog_errno(ret);
4353                 goto out_commit;
4354         }
4355
4356         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4357
4358         ret = ocfs2_journal_dirty(handle, root_bh);
4359         if (ret) {
4360                 mlog_errno(ret);
4361                 goto out_commit;
4362         }
4363
4364         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4365         if (ret)
4366                 mlog_errno(ret);
4367
4368 out_commit:
4369         ocfs2_commit_trans(osb, handle);
4370 out:
4371         ocfs2_schedule_truncate_log_flush(osb, 1);
4372
4373         mutex_unlock(&tl_inode->i_mutex);
4374
4375         if (meta_ac)
4376                 ocfs2_free_alloc_context(meta_ac);
4377
4378         ocfs2_run_deallocs(osb, &dealloc);
4379
4380         return ret;
4381 }
4382
4383 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4384                                          struct ocfs2_xattr_search *xs)
4385 {
4386         handle_t *handle = NULL;
4387         struct ocfs2_xattr_header *xh = xs->bucket.xh;
4388         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4389                                                 le16_to_cpu(xh->xh_count) - 1];
4390         int ret = 0;
4391
4392         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4393         if (IS_ERR(handle)) {
4394                 ret = PTR_ERR(handle);
4395                 mlog_errno(ret);
4396                 return;
4397         }
4398
4399         ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4400                                    OCFS2_JOURNAL_ACCESS_WRITE);
4401         if (ret) {
4402                 mlog_errno(ret);
4403                 goto out_commit;
4404         }
4405
4406         /* Remove the old entry. */
4407         memmove(xs->here, xs->here + 1,
4408                 (void *)last - (void *)xs->here);
4409         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4410         le16_add_cpu(&xh->xh_count, -1);
4411
4412         ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4413         if (ret < 0)
4414                 mlog_errno(ret);
4415 out_commit:
4416         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4417 }
4418
4419 /*
4420  * Set the xattr name/value in the bucket specified in xs.
4421  *
4422  * As the new value in xi may be stored in the bucket or in an outside cluster,
4423  * we divide the whole process into 3 steps:
4424  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4425  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4426  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4427  * 4. If the clusters for the new outside value can't be allocated, we need
4428  *    to free the xattr we allocated in set.
4429  */
4430 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4431                                      struct ocfs2_xattr_info *xi,
4432                                      struct ocfs2_xattr_search *xs)
4433 {
4434         int ret, local = 1;
4435         size_t value_len;
4436         char *val = (char *)xi->value;
4437         struct ocfs2_xattr_entry *xe = xs->here;
4438         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4439                                               strlen(xi->name));
4440
4441         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4442                 /*
4443                  * We need to truncate the xattr storage first.
4444                  *
4445                  * If both the old and new value are stored to
4446                  * outside block, we only need to truncate
4447                  * the storage and then set the value outside.
4448                  *
4449                  * If the new value should be stored within block,
4450                  * we should free all the outside block first and
4451                  * the modification to the xattr block will be done
4452                  * by following steps.
4453                  */
4454                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4455                         value_len = xi->value_len;
4456                 else
4457                         value_len = 0;
4458
4459                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4460                                                            value_len);
4461                 if (ret)
4462                         goto out;
4463
4464                 if (value_len)
4465                         goto set_value_outside;
4466         }
4467
4468         value_len = xi->value_len;
4469         /* So we have to handle the inside block change now. */
4470         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4471                 /*
4472                  * If the new value will be stored outside of block,
4473                  * initalize a new empty value root and insert it first.
4474                  */
4475                 local = 0;
4476                 xi->value = &def_xv;
4477                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4478         }
4479
4480         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4481         if (ret) {
4482                 mlog_errno(ret);
4483                 goto out;
4484         }
4485
4486         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4487                 goto out;
4488
4489         /* allocate the space now for the outside block storage. */
4490         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4491                                                    value_len);
4492         if (ret) {
4493                 mlog_errno(ret);
4494
4495                 if (xs->not_found) {
4496                         /*
4497                          * We can't allocate enough clusters for outside
4498                          * storage and we have allocated xattr already,
4499                          * so need to remove it.
4500                          */
4501                         ocfs2_xattr_bucket_remove_xs(inode, xs);
4502                 }
4503                 goto out;
4504         }
4505
4506 set_value_outside:
4507         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4508 out:
4509         return ret;
4510 }
4511
4512 /*
4513  * check whether the xattr bucket is filled up with the same hash value.
4514  * If we want to insert the xattr with the same hash, return -ENOSPC.
4515  * If we want to insert a xattr with different hash value, go ahead
4516  * and ocfs2_divide_xattr_bucket will handle this.
4517  */
4518 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4519                                               struct ocfs2_xattr_bucket *bucket,
4520                                               const char *name)
4521 {
4522         struct ocfs2_xattr_header *xh = bucket->xh;
4523         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4524
4525         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4526                 return 0;
4527
4528         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4529             xh->xh_entries[0].xe_name_hash) {
4530                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4531                      "hash = %u\n",
4532                      (unsigned long long)bucket->bhs[0]->b_blocknr,
4533                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4534                 return -ENOSPC;
4535         }
4536
4537         return 0;
4538 }
4539
4540 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4541                                              struct ocfs2_xattr_info *xi,
4542                                              struct ocfs2_xattr_search *xs)
4543 {
4544         struct ocfs2_xattr_header *xh;
4545         struct ocfs2_xattr_entry *xe;
4546         u16 count, header_size, xh_free_start;
4547         int i, free, max_free, need, old;
4548         size_t value_size = 0, name_len = strlen(xi->name);
4549         size_t blocksize = inode->i_sb->s_blocksize;
4550         int ret, allocation = 0;
4551         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4552
4553         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4554
4555 try_again:
4556         xh = xs->header;
4557         count = le16_to_cpu(xh->xh_count);
4558         xh_free_start = le16_to_cpu(xh->xh_free_start);
4559         header_size = sizeof(struct ocfs2_xattr_header) +
4560                         count * sizeof(struct ocfs2_xattr_entry);
4561         max_free = OCFS2_XATTR_BUCKET_SIZE -
4562                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4563
4564         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4565                         "of %u which exceed block size\n",
4566                         (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4567                         header_size);
4568
4569         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4570                 value_size = OCFS2_XATTR_ROOT_SIZE;
4571         else if (xi->value)
4572                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4573
4574         if (xs->not_found)
4575                 need = sizeof(struct ocfs2_xattr_entry) +
4576                         OCFS2_XATTR_SIZE(name_len) + value_size;
4577         else {
4578                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4579
4580                 /*
4581                  * We only replace the old value if the new length is smaller
4582                  * than the old one. Otherwise we will allocate new space in the
4583                  * bucket to store it.
4584                  */
4585                 xe = xs->here;
4586                 if (ocfs2_xattr_is_local(xe))
4587                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4588                 else
4589                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4590
4591                 if (old >= value_size)
4592                         need = 0;
4593         }
4594
4595         free = xh_free_start - header_size;
4596         /*
4597          * We need to make sure the new name/value pair
4598          * can exist in the same block.
4599          */
4600         if (xh_free_start % blocksize < need)
4601                 free -= xh_free_start % blocksize;
4602
4603         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4604              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4605              " %u\n", xs->not_found,
4606              (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4607              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4608              le16_to_cpu(xh->xh_name_value_len));
4609
4610         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4611                 if (need <= max_free &&
4612                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4613                         /*
4614                          * We can create the space by defragment. Since only the
4615                          * name/value will be moved, the xe shouldn't be changed
4616                          * in xs.
4617                          */
4618                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4619                         if (ret) {
4620                                 mlog_errno(ret);
4621                                 goto out;
4622                         }
4623
4624                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4625                         free = xh_free_start - header_size;
4626                         if (xh_free_start % blocksize < need)
4627                                 free -= xh_free_start % blocksize;
4628
4629                         if (free >= need)
4630                                 goto xattr_set;
4631
4632                         mlog(0, "Can't get enough space for xattr insert by "
4633                              "defragment. Need %u bytes, but we have %d, so "
4634                              "allocate new bucket for it.\n", need, free);
4635                 }
4636
4637                 /*
4638                  * We have to add new buckets or clusters and one
4639                  * allocation should leave us enough space for insert.
4640                  */
4641                 BUG_ON(allocation);
4642
4643                 /*
4644                  * We do not allow for overlapping ranges between buckets. And
4645                  * the maximum number of collisions we will allow for then is
4646                  * one bucket's worth, so check it here whether we need to
4647                  * add a new bucket for the insert.
4648                  */
4649                 ret = ocfs2_check_xattr_bucket_collision(inode,
4650                                                          &xs->bucket,
4651                                                          xi->name);
4652                 if (ret) {
4653                         mlog_errno(ret);
4654                         goto out;
4655                 }
4656
4657                 ret = ocfs2_add_new_xattr_bucket(inode,
4658                                                  xs->xattr_bh,
4659                                                  xs->bucket.bhs[0]);
4660                 if (ret) {
4661                         mlog_errno(ret);
4662                         goto out;
4663                 }
4664
4665                 for (i = 0; i < blk_per_bucket; i++)
4666                         brelse(xs->bucket.bhs[i]);
4667
4668                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4669
4670                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4671                                                    xi->name_index,
4672                                                    xi->name, xs);
4673                 if (ret && ret != -ENODATA)
4674                         goto out;
4675                 xs->not_found = ret;
4676                 allocation = 1;
4677                 goto try_again;
4678         }
4679
4680 xattr_set:
4681         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4682 out:
4683         mlog_exit(ret);
4684         return ret;
4685 }
4686
4687 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4688                                         struct ocfs2_xattr_bucket *bucket,
4689                                         void *para)
4690 {
4691         int ret = 0;
4692         struct ocfs2_xattr_header *xh = bucket->xh;
4693         u16 i;
4694         struct ocfs2_xattr_entry *xe;
4695
4696         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4697                 xe = &xh->xh_entries[i];
4698                 if (ocfs2_xattr_is_local(xe))
4699                         continue;
4700
4701                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4702                                                         bucket->bhs[0],
4703                                                         i, 0);
4704                 if (ret) {
4705                         mlog_errno(ret);
4706                         break;
4707                 }
4708         }
4709
4710         return ret;
4711 }
4712
4713 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4714                                           struct buffer_head *xb_bh)
4715 {
4716         struct ocfs2_xattr_block *xb =
4717                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4718         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4719         int ret = 0;
4720         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4721         u64 p_blkno;
4722
4723         if (le16_to_cpu(el->l_next_free_rec) == 0)
4724                 return 0;
4725
4726         while (name_hash > 0) {
4727                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4728                                           &e_cpos, &num_clusters, el);
4729                 if (ret) {
4730                         mlog_errno(ret);
4731                         goto out;
4732                 }
4733
4734                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4735                                                   ocfs2_delete_xattr_in_bucket,
4736                                                   NULL);
4737                 if (ret) {
4738                         mlog_errno(ret);
4739                         goto out;
4740                 }
4741
4742                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4743                                              p_blkno, e_cpos, num_clusters);
4744                 if (ret) {
4745                         mlog_errno(ret);
4746                         break;
4747                 }
4748
4749                 if (e_cpos == 0)
4750                         break;
4751
4752                 name_hash = e_cpos - 1;
4753         }
4754
4755 out:
4756         return ret;
4757 }
4758
4759 /*
4760  * 'trusted' attributes support
4761  */
4762 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4763                                        size_t list_size, const char *name,
4764                                        size_t name_len)
4765 {
4766         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
4767         const size_t total_len = prefix_len + name_len + 1;
4768
4769         if (list && total_len <= list_size) {
4770                 memcpy(list, XATTR_TRUSTED_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_trusted_get(struct inode *inode, const char *name,
4778                                    void *buffer, size_t size)
4779 {
4780         if (strcmp(name, "") == 0)
4781                 return -EINVAL;
4782         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4783                                buffer, size);
4784 }
4785
4786 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4787                                    const void *value, size_t size, int flags)
4788 {
4789         if (strcmp(name, "") == 0)
4790                 return -EINVAL;
4791
4792         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4793                                size, flags);
4794 }
4795
4796 struct xattr_handler ocfs2_xattr_trusted_handler = {
4797         .prefix = XATTR_TRUSTED_PREFIX,
4798         .list   = ocfs2_xattr_trusted_list,
4799         .get    = ocfs2_xattr_trusted_get,
4800         .set    = ocfs2_xattr_trusted_set,
4801 };
4802
4803 /*
4804  * 'user' attributes support
4805  */
4806 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4807                                     size_t list_size, const char *name,
4808                                     size_t name_len)
4809 {
4810         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
4811         const size_t total_len = prefix_len + name_len + 1;
4812         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4813
4814         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4815                 return 0;
4816
4817         if (list && total_len <= list_size) {
4818                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4819                 memcpy(list + prefix_len, name, name_len);
4820                 list[prefix_len + name_len] = '\0';
4821         }
4822         return total_len;
4823 }
4824
4825 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4826                                 void *buffer, size_t size)
4827 {
4828         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4829
4830         if (strcmp(name, "") == 0)
4831                 return -EINVAL;
4832         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4833                 return -EOPNOTSUPP;
4834         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4835                                buffer, size);
4836 }
4837
4838 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4839                                 const void *value, size_t size, int flags)
4840 {
4841         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4842
4843         if (strcmp(name, "") == 0)
4844                 return -EINVAL;
4845         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4846                 return -EOPNOTSUPP;
4847
4848         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4849                                size, flags);
4850 }
4851
4852 struct xattr_handler ocfs2_xattr_user_handler = {
4853         .prefix = XATTR_USER_PREFIX,
4854         .list   = ocfs2_xattr_user_list,
4855         .get    = ocfs2_xattr_user_get,
4856         .set    = ocfs2_xattr_user_set,
4857 };