ocfs2: Check xattr block signatures properly.
[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_index_block_find(struct inode *inode,
115                                         struct buffer_head *root_bh,
116                                         int name_index,
117                                         const char *name,
118                                         struct ocfs2_xattr_search *xs);
119
120 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
121                                         struct ocfs2_xattr_tree_root *xt,
122                                         char *buffer,
123                                         size_t buffer_size);
124
125 static int ocfs2_xattr_create_index_block(struct inode *inode,
126                                           struct ocfs2_xattr_search *xs);
127
128 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
129                                              struct ocfs2_xattr_info *xi,
130                                              struct ocfs2_xattr_search *xs);
131
132 static int ocfs2_delete_xattr_index_block(struct inode *inode,
133                                           struct buffer_head *xb_bh);
134
135 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
136 {
137         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
138 }
139
140 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
141 {
142         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
143 }
144
145 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
146 {
147         u16 len = sb->s_blocksize -
148                  offsetof(struct ocfs2_xattr_header, xh_entries);
149
150         return len / sizeof(struct ocfs2_xattr_entry);
151 }
152
153 static inline const char *ocfs2_xattr_prefix(int name_index)
154 {
155         struct xattr_handler *handler = NULL;
156
157         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
158                 handler = ocfs2_xattr_handler_map[name_index];
159
160         return handler ? handler->prefix : NULL;
161 }
162
163 static u32 ocfs2_xattr_name_hash(struct inode *inode,
164                                  const char *name,
165                                  int name_len)
166 {
167         /* Get hash value of uuid from super block */
168         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
169         int i;
170
171         /* hash extended attribute name */
172         for (i = 0; i < name_len; i++) {
173                 hash = (hash << OCFS2_HASH_SHIFT) ^
174                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
175                        *name++;
176         }
177
178         return hash;
179 }
180
181 /*
182  * ocfs2_xattr_hash_entry()
183  *
184  * Compute the hash of an extended attribute.
185  */
186 static void ocfs2_xattr_hash_entry(struct inode *inode,
187                                    struct ocfs2_xattr_header *header,
188                                    struct ocfs2_xattr_entry *entry)
189 {
190         u32 hash = 0;
191         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
192
193         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
194         entry->xe_name_hash = cpu_to_le32(hash);
195
196         return;
197 }
198
199 static int ocfs2_xattr_extend_allocation(struct inode *inode,
200                                          u32 clusters_to_add,
201                                          struct buffer_head *xattr_bh,
202                                          struct ocfs2_xattr_value_root *xv)
203 {
204         int status = 0;
205         int restart_func = 0;
206         int credits = 0;
207         handle_t *handle = NULL;
208         struct ocfs2_alloc_context *data_ac = NULL;
209         struct ocfs2_alloc_context *meta_ac = NULL;
210         enum ocfs2_alloc_restarted why;
211         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
212         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
213         struct ocfs2_extent_tree et;
214
215         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
216
217         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
218
219 restart_all:
220
221         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
222                                        &data_ac, &meta_ac);
223         if (status) {
224                 mlog_errno(status);
225                 goto leave;
226         }
227
228         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
229                                             clusters_to_add);
230         handle = ocfs2_start_trans(osb, credits);
231         if (IS_ERR(handle)) {
232                 status = PTR_ERR(handle);
233                 handle = NULL;
234                 mlog_errno(status);
235                 goto leave;
236         }
237
238 restarted_transaction:
239         status = ocfs2_journal_access(handle, inode, xattr_bh,
240                                       OCFS2_JOURNAL_ACCESS_WRITE);
241         if (status < 0) {
242                 mlog_errno(status);
243                 goto leave;
244         }
245
246         prev_clusters = le32_to_cpu(xv->xr_clusters);
247         status = ocfs2_add_clusters_in_btree(osb,
248                                              inode,
249                                              &logical_start,
250                                              clusters_to_add,
251                                              0,
252                                              &et,
253                                              handle,
254                                              data_ac,
255                                              meta_ac,
256                                              &why);
257         if ((status < 0) && (status != -EAGAIN)) {
258                 if (status != -ENOSPC)
259                         mlog_errno(status);
260                 goto leave;
261         }
262
263         status = ocfs2_journal_dirty(handle, xattr_bh);
264         if (status < 0) {
265                 mlog_errno(status);
266                 goto leave;
267         }
268
269         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
270
271         if (why != RESTART_NONE && clusters_to_add) {
272                 if (why == RESTART_META) {
273                         mlog(0, "restarting function.\n");
274                         restart_func = 1;
275                 } else {
276                         BUG_ON(why != RESTART_TRANS);
277
278                         mlog(0, "restarting transaction.\n");
279                         /* TODO: This can be more intelligent. */
280                         credits = ocfs2_calc_extend_credits(osb->sb,
281                                                             et.et_root_el,
282                                                             clusters_to_add);
283                         status = ocfs2_extend_trans(handle, credits);
284                         if (status < 0) {
285                                 /* handle still has to be committed at
286                                  * this point. */
287                                 status = -ENOMEM;
288                                 mlog_errno(status);
289                                 goto leave;
290                         }
291                         goto restarted_transaction;
292                 }
293         }
294
295 leave:
296         if (handle) {
297                 ocfs2_commit_trans(osb, handle);
298                 handle = NULL;
299         }
300         if (data_ac) {
301                 ocfs2_free_alloc_context(data_ac);
302                 data_ac = NULL;
303         }
304         if (meta_ac) {
305                 ocfs2_free_alloc_context(meta_ac);
306                 meta_ac = NULL;
307         }
308         if ((!status) && restart_func) {
309                 restart_func = 0;
310                 goto restart_all;
311         }
312
313         return status;
314 }
315
316 static int __ocfs2_remove_xattr_range(struct inode *inode,
317                                       struct buffer_head *root_bh,
318                                       struct ocfs2_xattr_value_root *xv,
319                                       u32 cpos, u32 phys_cpos, u32 len,
320                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
321 {
322         int ret;
323         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
324         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
325         struct inode *tl_inode = osb->osb_tl_inode;
326         handle_t *handle;
327         struct ocfs2_alloc_context *meta_ac = NULL;
328         struct ocfs2_extent_tree et;
329
330         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
331
332         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
333         if (ret) {
334                 mlog_errno(ret);
335                 return ret;
336         }
337
338         mutex_lock(&tl_inode->i_mutex);
339
340         if (ocfs2_truncate_log_needs_flush(osb)) {
341                 ret = __ocfs2_flush_truncate_log(osb);
342                 if (ret < 0) {
343                         mlog_errno(ret);
344                         goto out;
345                 }
346         }
347
348         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
349         if (IS_ERR(handle)) {
350                 ret = PTR_ERR(handle);
351                 mlog_errno(ret);
352                 goto out;
353         }
354
355         ret = ocfs2_journal_access(handle, inode, root_bh,
356                                    OCFS2_JOURNAL_ACCESS_WRITE);
357         if (ret) {
358                 mlog_errno(ret);
359                 goto out_commit;
360         }
361
362         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
363                                   dealloc);
364         if (ret) {
365                 mlog_errno(ret);
366                 goto out_commit;
367         }
368
369         le32_add_cpu(&xv->xr_clusters, -len);
370
371         ret = ocfs2_journal_dirty(handle, root_bh);
372         if (ret) {
373                 mlog_errno(ret);
374                 goto out_commit;
375         }
376
377         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
378         if (ret)
379                 mlog_errno(ret);
380
381 out_commit:
382         ocfs2_commit_trans(osb, handle);
383 out:
384         mutex_unlock(&tl_inode->i_mutex);
385
386         if (meta_ac)
387                 ocfs2_free_alloc_context(meta_ac);
388
389         return ret;
390 }
391
392 static int ocfs2_xattr_shrink_size(struct inode *inode,
393                                    u32 old_clusters,
394                                    u32 new_clusters,
395                                    struct buffer_head *root_bh,
396                                    struct ocfs2_xattr_value_root *xv)
397 {
398         int ret = 0;
399         u32 trunc_len, cpos, phys_cpos, alloc_size;
400         u64 block;
401         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
402         struct ocfs2_cached_dealloc_ctxt dealloc;
403
404         ocfs2_init_dealloc_ctxt(&dealloc);
405
406         if (old_clusters <= new_clusters)
407                 return 0;
408
409         cpos = new_clusters;
410         trunc_len = old_clusters - new_clusters;
411         while (trunc_len) {
412                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
413                                                &alloc_size, &xv->xr_list);
414                 if (ret) {
415                         mlog_errno(ret);
416                         goto out;
417                 }
418
419                 if (alloc_size > trunc_len)
420                         alloc_size = trunc_len;
421
422                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
423                                                  phys_cpos, alloc_size,
424                                                  &dealloc);
425                 if (ret) {
426                         mlog_errno(ret);
427                         goto out;
428                 }
429
430                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
431                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
432                                                        alloc_size);
433                 cpos += alloc_size;
434                 trunc_len -= alloc_size;
435         }
436
437 out:
438         ocfs2_schedule_truncate_log_flush(osb, 1);
439         ocfs2_run_deallocs(osb, &dealloc);
440
441         return ret;
442 }
443
444 static int ocfs2_xattr_value_truncate(struct inode *inode,
445                                       struct buffer_head *root_bh,
446                                       struct ocfs2_xattr_value_root *xv,
447                                       int len)
448 {
449         int ret;
450         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
451         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
452
453         if (new_clusters == old_clusters)
454                 return 0;
455
456         if (new_clusters > old_clusters)
457                 ret = ocfs2_xattr_extend_allocation(inode,
458                                                     new_clusters - old_clusters,
459                                                     root_bh, xv);
460         else
461                 ret = ocfs2_xattr_shrink_size(inode,
462                                               old_clusters, new_clusters,
463                                               root_bh, xv);
464
465         return ret;
466 }
467
468 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
469                                   size_t *result, const char *prefix,
470                                   const char *name, int name_len)
471 {
472         char *p = buffer + *result;
473         int prefix_len = strlen(prefix);
474         int total_len = prefix_len + name_len + 1;
475
476         *result += total_len;
477
478         /* we are just looking for how big our buffer needs to be */
479         if (!size)
480                 return 0;
481
482         if (*result > size)
483                 return -ERANGE;
484
485         memcpy(p, prefix, prefix_len);
486         memcpy(p + prefix_len, name, name_len);
487         p[prefix_len + name_len] = '\0';
488
489         return 0;
490 }
491
492 static int ocfs2_xattr_list_entries(struct inode *inode,
493                                     struct ocfs2_xattr_header *header,
494                                     char *buffer, size_t buffer_size)
495 {
496         size_t result = 0;
497         int i, type, ret;
498         const char *prefix, *name;
499
500         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
501                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
502                 type = ocfs2_xattr_get_type(entry);
503                 prefix = ocfs2_xattr_prefix(type);
504
505                 if (prefix) {
506                         name = (const char *)header +
507                                 le16_to_cpu(entry->xe_name_offset);
508
509                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
510                                                      &result, prefix, name,
511                                                      entry->xe_name_len);
512                         if (ret)
513                                 return ret;
514                 }
515         }
516
517         return result;
518 }
519
520 static int ocfs2_xattr_ibody_list(struct inode *inode,
521                                   struct ocfs2_dinode *di,
522                                   char *buffer,
523                                   size_t buffer_size)
524 {
525         struct ocfs2_xattr_header *header = NULL;
526         struct ocfs2_inode_info *oi = OCFS2_I(inode);
527         int ret = 0;
528
529         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
530                 return ret;
531
532         header = (struct ocfs2_xattr_header *)
533                  ((void *)di + inode->i_sb->s_blocksize -
534                  le16_to_cpu(di->i_xattr_inline_size));
535
536         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
537
538         return ret;
539 }
540
541 static int ocfs2_xattr_block_list(struct inode *inode,
542                                   struct ocfs2_dinode *di,
543                                   char *buffer,
544                                   size_t buffer_size)
545 {
546         struct buffer_head *blk_bh = NULL;
547         struct ocfs2_xattr_block *xb;
548         int ret = 0;
549
550         if (!di->i_xattr_loc)
551                 return ret;
552
553         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
554         if (ret < 0) {
555                 mlog_errno(ret);
556                 return ret;
557         }
558
559         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
560         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
561                 ret = -EIO;
562                 goto cleanup;
563         }
564
565         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
566                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
567                 ret = ocfs2_xattr_list_entries(inode, header,
568                                                buffer, buffer_size);
569         } else {
570                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
571                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
572                                                    buffer, buffer_size);
573         }
574 cleanup:
575         brelse(blk_bh);
576
577         return ret;
578 }
579
580 ssize_t ocfs2_listxattr(struct dentry *dentry,
581                         char *buffer,
582                         size_t size)
583 {
584         int ret = 0, i_ret = 0, b_ret = 0;
585         struct buffer_head *di_bh = NULL;
586         struct ocfs2_dinode *di = NULL;
587         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
588
589         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
590                 return -EOPNOTSUPP;
591
592         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
593                 return ret;
594
595         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
596         if (ret < 0) {
597                 mlog_errno(ret);
598                 return ret;
599         }
600
601         di = (struct ocfs2_dinode *)di_bh->b_data;
602
603         down_read(&oi->ip_xattr_sem);
604         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
605         if (i_ret < 0)
606                 b_ret = 0;
607         else {
608                 if (buffer) {
609                         buffer += i_ret;
610                         size -= i_ret;
611                 }
612                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
613                                                buffer, size);
614                 if (b_ret < 0)
615                         i_ret = 0;
616         }
617         up_read(&oi->ip_xattr_sem);
618         ocfs2_inode_unlock(dentry->d_inode, 0);
619
620         brelse(di_bh);
621
622         return i_ret + b_ret;
623 }
624
625 static int ocfs2_xattr_find_entry(int name_index,
626                                   const char *name,
627                                   struct ocfs2_xattr_search *xs)
628 {
629         struct ocfs2_xattr_entry *entry;
630         size_t name_len;
631         int i, cmp = 1;
632
633         if (name == NULL)
634                 return -EINVAL;
635
636         name_len = strlen(name);
637         entry = xs->here;
638         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
639                 cmp = name_index - ocfs2_xattr_get_type(entry);
640                 if (!cmp)
641                         cmp = name_len - entry->xe_name_len;
642                 if (!cmp)
643                         cmp = memcmp(name, (xs->base +
644                                      le16_to_cpu(entry->xe_name_offset)),
645                                      name_len);
646                 if (cmp == 0)
647                         break;
648                 entry += 1;
649         }
650         xs->here = entry;
651
652         return cmp ? -ENODATA : 0;
653 }
654
655 static int ocfs2_xattr_get_value_outside(struct inode *inode,
656                                          struct ocfs2_xattr_value_root *xv,
657                                          void *buffer,
658                                          size_t len)
659 {
660         u32 cpos, p_cluster, num_clusters, bpc, clusters;
661         u64 blkno;
662         int i, ret = 0;
663         size_t cplen, blocksize;
664         struct buffer_head *bh = NULL;
665         struct ocfs2_extent_list *el;
666
667         el = &xv->xr_list;
668         clusters = le32_to_cpu(xv->xr_clusters);
669         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
670         blocksize = inode->i_sb->s_blocksize;
671
672         cpos = 0;
673         while (cpos < clusters) {
674                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
675                                                &num_clusters, el);
676                 if (ret) {
677                         mlog_errno(ret);
678                         goto out;
679                 }
680
681                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
682                 /* Copy ocfs2_xattr_value */
683                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
684                         ret = ocfs2_read_block(inode, blkno, &bh);
685                         if (ret) {
686                                 mlog_errno(ret);
687                                 goto out;
688                         }
689
690                         cplen = len >= blocksize ? blocksize : len;
691                         memcpy(buffer, bh->b_data, cplen);
692                         len -= cplen;
693                         buffer += cplen;
694
695                         brelse(bh);
696                         bh = NULL;
697                         if (len == 0)
698                                 break;
699                 }
700                 cpos += num_clusters;
701         }
702 out:
703         return ret;
704 }
705
706 static int ocfs2_xattr_ibody_get(struct inode *inode,
707                                  int name_index,
708                                  const char *name,
709                                  void *buffer,
710                                  size_t buffer_size,
711                                  struct ocfs2_xattr_search *xs)
712 {
713         struct ocfs2_inode_info *oi = OCFS2_I(inode);
714         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
715         struct ocfs2_xattr_value_root *xv;
716         size_t size;
717         int ret = 0;
718
719         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
720                 return -ENODATA;
721
722         xs->end = (void *)di + inode->i_sb->s_blocksize;
723         xs->header = (struct ocfs2_xattr_header *)
724                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
725         xs->base = (void *)xs->header;
726         xs->here = xs->header->xh_entries;
727
728         ret = ocfs2_xattr_find_entry(name_index, name, xs);
729         if (ret)
730                 return ret;
731         size = le64_to_cpu(xs->here->xe_value_size);
732         if (buffer) {
733                 if (size > buffer_size)
734                         return -ERANGE;
735                 if (ocfs2_xattr_is_local(xs->here)) {
736                         memcpy(buffer, (void *)xs->base +
737                                le16_to_cpu(xs->here->xe_name_offset) +
738                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
739                 } else {
740                         xv = (struct ocfs2_xattr_value_root *)
741                                 (xs->base + le16_to_cpu(
742                                  xs->here->xe_name_offset) +
743                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
744                         ret = ocfs2_xattr_get_value_outside(inode, xv,
745                                                             buffer, size);
746                         if (ret < 0) {
747                                 mlog_errno(ret);
748                                 return ret;
749                         }
750                 }
751         }
752
753         return size;
754 }
755
756 static int ocfs2_xattr_block_get(struct inode *inode,
757                                  int name_index,
758                                  const char *name,
759                                  void *buffer,
760                                  size_t buffer_size,
761                                  struct ocfs2_xattr_search *xs)
762 {
763         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
764         struct buffer_head *blk_bh = NULL;
765         struct ocfs2_xattr_block *xb;
766         struct ocfs2_xattr_value_root *xv;
767         size_t size;
768         int ret = -ENODATA, name_offset, name_len, block_off, i;
769
770         if (!di->i_xattr_loc)
771                 return ret;
772
773         memset(&xs->bucket, 0, sizeof(xs->bucket));
774
775         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
776         if (ret < 0) {
777                 mlog_errno(ret);
778                 return ret;
779         }
780
781         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
782         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
783                 ret = -EIO;
784                 goto cleanup;
785         }
786
787         xs->xattr_bh = blk_bh;
788
789         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
790                 xs->header = &xb->xb_attrs.xb_header;
791                 xs->base = (void *)xs->header;
792                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
793                 xs->here = xs->header->xh_entries;
794
795                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
796         } else
797                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
798                                                    name_index,
799                                                    name, xs);
800
801         if (ret)
802                 goto cleanup;
803         size = le64_to_cpu(xs->here->xe_value_size);
804         if (buffer) {
805                 ret = -ERANGE;
806                 if (size > buffer_size)
807                         goto cleanup;
808
809                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
810                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
811                 i = xs->here - xs->header->xh_entries;
812
813                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
814                         ret = ocfs2_xattr_bucket_get_name_value(inode,
815                                                                 xs->bucket.xh,
816                                                                 i,
817                                                                 &block_off,
818                                                                 &name_offset);
819                         xs->base = xs->bucket.bhs[block_off]->b_data;
820                 }
821                 if (ocfs2_xattr_is_local(xs->here)) {
822                         memcpy(buffer, (void *)xs->base +
823                                name_offset + name_len, size);
824                 } else {
825                         xv = (struct ocfs2_xattr_value_root *)
826                                 (xs->base + name_offset + name_len);
827                         ret = ocfs2_xattr_get_value_outside(inode, xv,
828                                                             buffer, size);
829                         if (ret < 0) {
830                                 mlog_errno(ret);
831                                 goto cleanup;
832                         }
833                 }
834         }
835         ret = size;
836 cleanup:
837         for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
838                 brelse(xs->bucket.bhs[i]);
839         memset(&xs->bucket, 0, sizeof(xs->bucket));
840
841         brelse(blk_bh);
842         return ret;
843 }
844
845 /* ocfs2_xattr_get()
846  *
847  * Copy an extended attribute into the buffer provided.
848  * Buffer is NULL to compute the size of buffer required.
849  */
850 static int ocfs2_xattr_get(struct inode *inode,
851                            int name_index,
852                            const char *name,
853                            void *buffer,
854                            size_t buffer_size)
855 {
856         int ret;
857         struct ocfs2_dinode *di = NULL;
858         struct buffer_head *di_bh = NULL;
859         struct ocfs2_inode_info *oi = OCFS2_I(inode);
860         struct ocfs2_xattr_search xis = {
861                 .not_found = -ENODATA,
862         };
863         struct ocfs2_xattr_search xbs = {
864                 .not_found = -ENODATA,
865         };
866
867         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
868                 return -EOPNOTSUPP;
869
870         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
871                 ret = -ENODATA;
872
873         ret = ocfs2_inode_lock(inode, &di_bh, 0);
874         if (ret < 0) {
875                 mlog_errno(ret);
876                 return ret;
877         }
878         xis.inode_bh = xbs.inode_bh = di_bh;
879         di = (struct ocfs2_dinode *)di_bh->b_data;
880
881         down_read(&oi->ip_xattr_sem);
882         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
883                                     buffer_size, &xis);
884         if (ret == -ENODATA)
885                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
886                                             buffer_size, &xbs);
887         up_read(&oi->ip_xattr_sem);
888         ocfs2_inode_unlock(inode, 0);
889
890         brelse(di_bh);
891
892         return ret;
893 }
894
895 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
896                                            struct ocfs2_xattr_value_root *xv,
897                                            const void *value,
898                                            int value_len)
899 {
900         int ret = 0, i, cp_len, credits;
901         u16 blocksize = inode->i_sb->s_blocksize;
902         u32 p_cluster, num_clusters;
903         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
904         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
905         u64 blkno;
906         struct buffer_head *bh = NULL;
907         handle_t *handle;
908
909         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
910
911         credits = clusters * bpc;
912         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
913         if (IS_ERR(handle)) {
914                 ret = PTR_ERR(handle);
915                 mlog_errno(ret);
916                 goto out;
917         }
918
919         while (cpos < clusters) {
920                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
921                                                &num_clusters, &xv->xr_list);
922                 if (ret) {
923                         mlog_errno(ret);
924                         goto out_commit;
925                 }
926
927                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
928
929                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
930                         ret = ocfs2_read_block(inode, blkno, &bh);
931                         if (ret) {
932                                 mlog_errno(ret);
933                                 goto out_commit;
934                         }
935
936                         ret = ocfs2_journal_access(handle,
937                                                    inode,
938                                                    bh,
939                                                    OCFS2_JOURNAL_ACCESS_WRITE);
940                         if (ret < 0) {
941                                 mlog_errno(ret);
942                                 goto out_commit;
943                         }
944
945                         cp_len = value_len > blocksize ? blocksize : value_len;
946                         memcpy(bh->b_data, value, cp_len);
947                         value_len -= cp_len;
948                         value += cp_len;
949                         if (cp_len < blocksize)
950                                 memset(bh->b_data + cp_len, 0,
951                                        blocksize - cp_len);
952
953                         ret = ocfs2_journal_dirty(handle, bh);
954                         if (ret < 0) {
955                                 mlog_errno(ret);
956                                 goto out_commit;
957                         }
958                         brelse(bh);
959                         bh = NULL;
960
961                         /*
962                          * XXX: do we need to empty all the following
963                          * blocks in this cluster?
964                          */
965                         if (!value_len)
966                                 break;
967                 }
968                 cpos += num_clusters;
969         }
970 out_commit:
971         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
972 out:
973         brelse(bh);
974
975         return ret;
976 }
977
978 static int ocfs2_xattr_cleanup(struct inode *inode,
979                                struct ocfs2_xattr_info *xi,
980                                struct ocfs2_xattr_search *xs,
981                                size_t offs)
982 {
983         handle_t *handle = NULL;
984         int ret = 0;
985         size_t name_len = strlen(xi->name);
986         void *val = xs->base + offs;
987         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
988
989         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
990                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
991         if (IS_ERR(handle)) {
992                 ret = PTR_ERR(handle);
993                 mlog_errno(ret);
994                 goto out;
995         }
996         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
997                                    OCFS2_JOURNAL_ACCESS_WRITE);
998         if (ret) {
999                 mlog_errno(ret);
1000                 goto out_commit;
1001         }
1002         /* Decrease xattr count */
1003         le16_add_cpu(&xs->header->xh_count, -1);
1004         /* Remove the xattr entry and tree root which has already be set*/
1005         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1006         memset(val, 0, size);
1007
1008         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1009         if (ret < 0)
1010                 mlog_errno(ret);
1011 out_commit:
1012         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1013 out:
1014         return ret;
1015 }
1016
1017 static int ocfs2_xattr_update_entry(struct inode *inode,
1018                                     struct ocfs2_xattr_info *xi,
1019                                     struct ocfs2_xattr_search *xs,
1020                                     size_t offs)
1021 {
1022         handle_t *handle = NULL;
1023         int ret = 0;
1024
1025         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1026                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1027         if (IS_ERR(handle)) {
1028                 ret = PTR_ERR(handle);
1029                 mlog_errno(ret);
1030                 goto out;
1031         }
1032         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1033                                    OCFS2_JOURNAL_ACCESS_WRITE);
1034         if (ret) {
1035                 mlog_errno(ret);
1036                 goto out_commit;
1037         }
1038
1039         xs->here->xe_name_offset = cpu_to_le16(offs);
1040         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1041         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1042                 ocfs2_xattr_set_local(xs->here, 1);
1043         else
1044                 ocfs2_xattr_set_local(xs->here, 0);
1045         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1046
1047         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1048         if (ret < 0)
1049                 mlog_errno(ret);
1050 out_commit:
1051         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1052 out:
1053         return ret;
1054 }
1055
1056 /*
1057  * ocfs2_xattr_set_value_outside()
1058  *
1059  * Set large size value in B tree.
1060  */
1061 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1062                                          struct ocfs2_xattr_info *xi,
1063                                          struct ocfs2_xattr_search *xs,
1064                                          size_t offs)
1065 {
1066         size_t name_len = strlen(xi->name);
1067         void *val = xs->base + offs;
1068         struct ocfs2_xattr_value_root *xv = NULL;
1069         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1070         int ret = 0;
1071
1072         memset(val, 0, size);
1073         memcpy(val, xi->name, name_len);
1074         xv = (struct ocfs2_xattr_value_root *)
1075                 (val + OCFS2_XATTR_SIZE(name_len));
1076         xv->xr_clusters = 0;
1077         xv->xr_last_eb_blk = 0;
1078         xv->xr_list.l_tree_depth = 0;
1079         xv->xr_list.l_count = cpu_to_le16(1);
1080         xv->xr_list.l_next_free_rec = 0;
1081
1082         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1083                                          xi->value_len);
1084         if (ret < 0) {
1085                 mlog_errno(ret);
1086                 return ret;
1087         }
1088         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1089                                               xi->value_len);
1090         if (ret < 0) {
1091                 mlog_errno(ret);
1092                 return ret;
1093         }
1094         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1095         if (ret < 0)
1096                 mlog_errno(ret);
1097
1098         return ret;
1099 }
1100
1101 /*
1102  * ocfs2_xattr_set_entry_local()
1103  *
1104  * Set, replace or remove extended attribute in local.
1105  */
1106 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1107                                         struct ocfs2_xattr_info *xi,
1108                                         struct ocfs2_xattr_search *xs,
1109                                         struct ocfs2_xattr_entry *last,
1110                                         size_t min_offs)
1111 {
1112         size_t name_len = strlen(xi->name);
1113         int i;
1114
1115         if (xi->value && xs->not_found) {
1116                 /* Insert the new xattr entry. */
1117                 le16_add_cpu(&xs->header->xh_count, 1);
1118                 ocfs2_xattr_set_type(last, xi->name_index);
1119                 ocfs2_xattr_set_local(last, 1);
1120                 last->xe_name_len = name_len;
1121         } else {
1122                 void *first_val;
1123                 void *val;
1124                 size_t offs, size;
1125
1126                 first_val = xs->base + min_offs;
1127                 offs = le16_to_cpu(xs->here->xe_name_offset);
1128                 val = xs->base + offs;
1129
1130                 if (le64_to_cpu(xs->here->xe_value_size) >
1131                     OCFS2_XATTR_INLINE_SIZE)
1132                         size = OCFS2_XATTR_SIZE(name_len) +
1133                                 OCFS2_XATTR_ROOT_SIZE;
1134                 else
1135                         size = OCFS2_XATTR_SIZE(name_len) +
1136                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1137
1138                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1139                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1140                         /* The old and the new value have the
1141                            same size. Just replace the value. */
1142                         ocfs2_xattr_set_local(xs->here, 1);
1143                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1144                         /* Clear value bytes. */
1145                         memset(val + OCFS2_XATTR_SIZE(name_len),
1146                                0,
1147                                OCFS2_XATTR_SIZE(xi->value_len));
1148                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1149                                xi->value,
1150                                xi->value_len);
1151                         return;
1152                 }
1153                 /* Remove the old name+value. */
1154                 memmove(first_val + size, first_val, val - first_val);
1155                 memset(first_val, 0, size);
1156                 xs->here->xe_name_hash = 0;
1157                 xs->here->xe_name_offset = 0;
1158                 ocfs2_xattr_set_local(xs->here, 1);
1159                 xs->here->xe_value_size = 0;
1160
1161                 min_offs += size;
1162
1163                 /* Adjust all value offsets. */
1164                 last = xs->header->xh_entries;
1165                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1166                         size_t o = le16_to_cpu(last->xe_name_offset);
1167
1168                         if (o < offs)
1169                                 last->xe_name_offset = cpu_to_le16(o + size);
1170                         last += 1;
1171                 }
1172
1173                 if (!xi->value) {
1174                         /* Remove the old entry. */
1175                         last -= 1;
1176                         memmove(xs->here, xs->here + 1,
1177                                 (void *)last - (void *)xs->here);
1178                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1179                         le16_add_cpu(&xs->header->xh_count, -1);
1180                 }
1181         }
1182         if (xi->value) {
1183                 /* Insert the new name+value. */
1184                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1185                                 OCFS2_XATTR_SIZE(xi->value_len);
1186                 void *val = xs->base + min_offs - size;
1187
1188                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1189                 memset(val, 0, size);
1190                 memcpy(val, xi->name, name_len);
1191                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1192                        xi->value,
1193                        xi->value_len);
1194                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1195                 ocfs2_xattr_set_local(xs->here, 1);
1196                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1197         }
1198
1199         return;
1200 }
1201
1202 /*
1203  * ocfs2_xattr_set_entry()
1204  *
1205  * Set extended attribute entry into inode or block.
1206  *
1207  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1208  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1209  * then set value in B tree with set_value_outside().
1210  */
1211 static int ocfs2_xattr_set_entry(struct inode *inode,
1212                                  struct ocfs2_xattr_info *xi,
1213                                  struct ocfs2_xattr_search *xs,
1214                                  int flag)
1215 {
1216         struct ocfs2_xattr_entry *last;
1217         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1218         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1219         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1220         size_t size_l = 0;
1221         handle_t *handle = NULL;
1222         int free, i, ret;
1223         struct ocfs2_xattr_info xi_l = {
1224                 .name_index = xi->name_index,
1225                 .name = xi->name,
1226                 .value = xi->value,
1227                 .value_len = xi->value_len,
1228         };
1229
1230         /* Compute min_offs, last and free space. */
1231         last = xs->header->xh_entries;
1232
1233         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1234                 size_t offs = le16_to_cpu(last->xe_name_offset);
1235                 if (offs < min_offs)
1236                         min_offs = offs;
1237                 last += 1;
1238         }
1239
1240         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1241         if (free < 0)
1242                 return -EFAULT;
1243
1244         if (!xs->not_found) {
1245                 size_t size = 0;
1246                 if (ocfs2_xattr_is_local(xs->here))
1247                         size = OCFS2_XATTR_SIZE(name_len) +
1248                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1249                 else
1250                         size = OCFS2_XATTR_SIZE(name_len) +
1251                                 OCFS2_XATTR_ROOT_SIZE;
1252                 free += (size + sizeof(struct ocfs2_xattr_entry));
1253         }
1254         /* Check free space in inode or block */
1255         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1256                 if (free < sizeof(struct ocfs2_xattr_entry) +
1257                            OCFS2_XATTR_SIZE(name_len) +
1258                            OCFS2_XATTR_ROOT_SIZE) {
1259                         ret = -ENOSPC;
1260                         goto out;
1261                 }
1262                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1263                 xi_l.value = (void *)&def_xv;
1264                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1265         } else if (xi->value) {
1266                 if (free < sizeof(struct ocfs2_xattr_entry) +
1267                            OCFS2_XATTR_SIZE(name_len) +
1268                            OCFS2_XATTR_SIZE(xi->value_len)) {
1269                         ret = -ENOSPC;
1270                         goto out;
1271                 }
1272         }
1273
1274         if (!xs->not_found) {
1275                 /* For existing extended attribute */
1276                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1277                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1278                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1279                 void *val = xs->base + offs;
1280
1281                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1282                         /* Replace existing local xattr with tree root */
1283                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1284                                                             offs);
1285                         if (ret < 0)
1286                                 mlog_errno(ret);
1287                         goto out;
1288                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1289                         /* For existing xattr which has value outside */
1290                         struct ocfs2_xattr_value_root *xv = NULL;
1291                         xv = (struct ocfs2_xattr_value_root *)(val +
1292                                 OCFS2_XATTR_SIZE(name_len));
1293
1294                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1295                                 /*
1296                                  * If new value need set outside also,
1297                                  * first truncate old value to new value,
1298                                  * then set new value with set_value_outside().
1299                                  */
1300                                 ret = ocfs2_xattr_value_truncate(inode,
1301                                                                  xs->xattr_bh,
1302                                                                  xv,
1303                                                                  xi->value_len);
1304                                 if (ret < 0) {
1305                                         mlog_errno(ret);
1306                                         goto out;
1307                                 }
1308
1309                                 ret = __ocfs2_xattr_set_value_outside(inode,
1310                                                                 xv,
1311                                                                 xi->value,
1312                                                                 xi->value_len);
1313                                 if (ret < 0) {
1314                                         mlog_errno(ret);
1315                                         goto out;
1316                                 }
1317
1318                                 ret = ocfs2_xattr_update_entry(inode,
1319                                                                xi,
1320                                                                xs,
1321                                                                offs);
1322                                 if (ret < 0)
1323                                         mlog_errno(ret);
1324                                 goto out;
1325                         } else {
1326                                 /*
1327                                  * If new value need set in local,
1328                                  * just trucate old value to zero.
1329                                  */
1330                                  ret = ocfs2_xattr_value_truncate(inode,
1331                                                                  xs->xattr_bh,
1332                                                                  xv,
1333                                                                  0);
1334                                 if (ret < 0)
1335                                         mlog_errno(ret);
1336                         }
1337                 }
1338         }
1339
1340         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1341                                    OCFS2_INODE_UPDATE_CREDITS);
1342         if (IS_ERR(handle)) {
1343                 ret = PTR_ERR(handle);
1344                 mlog_errno(ret);
1345                 goto out;
1346         }
1347
1348         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1349                                    OCFS2_JOURNAL_ACCESS_WRITE);
1350         if (ret) {
1351                 mlog_errno(ret);
1352                 goto out_commit;
1353         }
1354
1355         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1356                 /* set extended attribute in external block. */
1357                 ret = ocfs2_extend_trans(handle,
1358                                          OCFS2_INODE_UPDATE_CREDITS +
1359                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1360                 if (ret) {
1361                         mlog_errno(ret);
1362                         goto out_commit;
1363                 }
1364                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1365                                            OCFS2_JOURNAL_ACCESS_WRITE);
1366                 if (ret) {
1367                         mlog_errno(ret);
1368                         goto out_commit;
1369                 }
1370         }
1371
1372         /*
1373          * Set value in local, include set tree root in local.
1374          * This is the first step for value size >INLINE_SIZE.
1375          */
1376         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1377
1378         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1379                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1380                 if (ret < 0) {
1381                         mlog_errno(ret);
1382                         goto out_commit;
1383                 }
1384         }
1385
1386         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1387             (flag & OCFS2_INLINE_XATTR_FL)) {
1388                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1389                 unsigned int xattrsize = osb->s_xattr_inline_size;
1390
1391                 /*
1392                  * Adjust extent record count or inline data size
1393                  * to reserve space for extended attribute.
1394                  */
1395                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1396                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1397                         le16_add_cpu(&idata->id_count, -xattrsize);
1398                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1399                         struct ocfs2_extent_list *el = &di->id2.i_list;
1400                         le16_add_cpu(&el->l_count, -(xattrsize /
1401                                         sizeof(struct ocfs2_extent_rec)));
1402                 }
1403                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1404         }
1405         /* Update xattr flag */
1406         spin_lock(&oi->ip_lock);
1407         oi->ip_dyn_features |= flag;
1408         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1409         spin_unlock(&oi->ip_lock);
1410         /* Update inode ctime */
1411         inode->i_ctime = CURRENT_TIME;
1412         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1413         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1414
1415         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1416         if (ret < 0)
1417                 mlog_errno(ret);
1418
1419 out_commit:
1420         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1421
1422         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1423                 /*
1424                  * Set value outside in B tree.
1425                  * This is the second step for value size > INLINE_SIZE.
1426                  */
1427                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1428                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1429                 if (ret < 0) {
1430                         int ret2;
1431
1432                         mlog_errno(ret);
1433                         /*
1434                          * If set value outside failed, we have to clean
1435                          * the junk tree root we have already set in local.
1436                          */
1437                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1438                         if (ret2 < 0)
1439                                 mlog_errno(ret2);
1440                 }
1441         }
1442 out:
1443         return ret;
1444
1445 }
1446
1447 static int ocfs2_remove_value_outside(struct inode*inode,
1448                                       struct buffer_head *bh,
1449                                       struct ocfs2_xattr_header *header)
1450 {
1451         int ret = 0, i;
1452
1453         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1454                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1455
1456                 if (!ocfs2_xattr_is_local(entry)) {
1457                         struct ocfs2_xattr_value_root *xv;
1458                         void *val;
1459
1460                         val = (void *)header +
1461                                 le16_to_cpu(entry->xe_name_offset);
1462                         xv = (struct ocfs2_xattr_value_root *)
1463                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1464                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1465                         if (ret < 0) {
1466                                 mlog_errno(ret);
1467                                 return ret;
1468                         }
1469                 }
1470         }
1471
1472         return ret;
1473 }
1474
1475 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1476                                     struct buffer_head *di_bh)
1477 {
1478
1479         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1480         struct ocfs2_xattr_header *header;
1481         int ret;
1482
1483         header = (struct ocfs2_xattr_header *)
1484                  ((void *)di + inode->i_sb->s_blocksize -
1485                  le16_to_cpu(di->i_xattr_inline_size));
1486
1487         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1488
1489         return ret;
1490 }
1491
1492 static int ocfs2_xattr_block_remove(struct inode *inode,
1493                                     struct buffer_head *blk_bh)
1494 {
1495         struct ocfs2_xattr_block *xb;
1496         int ret = 0;
1497
1498         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1499         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1500                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1501                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1502         } else
1503                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1504
1505         return ret;
1506 }
1507
1508 static int ocfs2_xattr_free_block(struct inode *inode,
1509                                   u64 block)
1510 {
1511         struct inode *xb_alloc_inode;
1512         struct buffer_head *xb_alloc_bh = NULL;
1513         struct buffer_head *blk_bh = NULL;
1514         struct ocfs2_xattr_block *xb;
1515         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1516         handle_t *handle;
1517         int ret = 0;
1518         u64 blk, bg_blkno;
1519         u16 bit;
1520
1521         ret = ocfs2_read_block(inode, block, &blk_bh);
1522         if (ret < 0) {
1523                 mlog_errno(ret);
1524                 goto out;
1525         }
1526
1527         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1528         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1529                 ret = -EIO;
1530                 goto out;
1531         }
1532
1533         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1534         if (ret < 0) {
1535                 mlog_errno(ret);
1536                 goto out;
1537         }
1538
1539         blk = le64_to_cpu(xb->xb_blkno);
1540         bit = le16_to_cpu(xb->xb_suballoc_bit);
1541         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1542
1543         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1544                                 EXTENT_ALLOC_SYSTEM_INODE,
1545                                 le16_to_cpu(xb->xb_suballoc_slot));
1546         if (!xb_alloc_inode) {
1547                 ret = -ENOMEM;
1548                 mlog_errno(ret);
1549                 goto out;
1550         }
1551         mutex_lock(&xb_alloc_inode->i_mutex);
1552
1553         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1554         if (ret < 0) {
1555                 mlog_errno(ret);
1556                 goto out_mutex;
1557         }
1558
1559         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1560         if (IS_ERR(handle)) {
1561                 ret = PTR_ERR(handle);
1562                 mlog_errno(ret);
1563                 goto out_unlock;
1564         }
1565
1566         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1567                                        bit, bg_blkno, 1);
1568         if (ret < 0)
1569                 mlog_errno(ret);
1570
1571         ocfs2_commit_trans(osb, handle);
1572 out_unlock:
1573         ocfs2_inode_unlock(xb_alloc_inode, 1);
1574         brelse(xb_alloc_bh);
1575 out_mutex:
1576         mutex_unlock(&xb_alloc_inode->i_mutex);
1577         iput(xb_alloc_inode);
1578 out:
1579         brelse(blk_bh);
1580         return ret;
1581 }
1582
1583 /*
1584  * ocfs2_xattr_remove()
1585  *
1586  * Free extended attribute resources associated with this inode.
1587  */
1588 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1589 {
1590         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1591         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1592         handle_t *handle;
1593         int ret;
1594
1595         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1596                 return 0;
1597
1598         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1599                 return 0;
1600
1601         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1602                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1603                 if (ret < 0) {
1604                         mlog_errno(ret);
1605                         goto out;
1606                 }
1607         }
1608
1609         if (di->i_xattr_loc) {
1610                 ret = ocfs2_xattr_free_block(inode,
1611                                              le64_to_cpu(di->i_xattr_loc));
1612                 if (ret < 0) {
1613                         mlog_errno(ret);
1614                         goto out;
1615                 }
1616         }
1617
1618         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1619                                    OCFS2_INODE_UPDATE_CREDITS);
1620         if (IS_ERR(handle)) {
1621                 ret = PTR_ERR(handle);
1622                 mlog_errno(ret);
1623                 goto out;
1624         }
1625         ret = ocfs2_journal_access(handle, inode, di_bh,
1626                                    OCFS2_JOURNAL_ACCESS_WRITE);
1627         if (ret) {
1628                 mlog_errno(ret);
1629                 goto out_commit;
1630         }
1631
1632         di->i_xattr_loc = 0;
1633
1634         spin_lock(&oi->ip_lock);
1635         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1636         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1637         spin_unlock(&oi->ip_lock);
1638
1639         ret = ocfs2_journal_dirty(handle, di_bh);
1640         if (ret < 0)
1641                 mlog_errno(ret);
1642 out_commit:
1643         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1644 out:
1645         return ret;
1646 }
1647
1648 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1649                                         struct ocfs2_dinode *di)
1650 {
1651         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1652         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1653         int free;
1654
1655         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1656                 return 0;
1657
1658         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1659                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1660                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1661         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1662                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1663                         le64_to_cpu(di->i_size);
1664         } else {
1665                 struct ocfs2_extent_list *el = &di->id2.i_list;
1666                 free = (le16_to_cpu(el->l_count) -
1667                         le16_to_cpu(el->l_next_free_rec)) *
1668                         sizeof(struct ocfs2_extent_rec);
1669         }
1670         if (free >= xattrsize)
1671                 return 1;
1672
1673         return 0;
1674 }
1675
1676 /*
1677  * ocfs2_xattr_ibody_find()
1678  *
1679  * Find extended attribute in inode block and
1680  * fill search info into struct ocfs2_xattr_search.
1681  */
1682 static int ocfs2_xattr_ibody_find(struct inode *inode,
1683                                   int name_index,
1684                                   const char *name,
1685                                   struct ocfs2_xattr_search *xs)
1686 {
1687         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1688         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1689         int ret;
1690         int has_space = 0;
1691
1692         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1693                 return 0;
1694
1695         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1696                 down_read(&oi->ip_alloc_sem);
1697                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1698                 up_read(&oi->ip_alloc_sem);
1699                 if (!has_space)
1700                         return 0;
1701         }
1702
1703         xs->xattr_bh = xs->inode_bh;
1704         xs->end = (void *)di + inode->i_sb->s_blocksize;
1705         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1706                 xs->header = (struct ocfs2_xattr_header *)
1707                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1708         else
1709                 xs->header = (struct ocfs2_xattr_header *)
1710                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1711         xs->base = (void *)xs->header;
1712         xs->here = xs->header->xh_entries;
1713
1714         /* Find the named attribute. */
1715         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1716                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1717                 if (ret && ret != -ENODATA)
1718                         return ret;
1719                 xs->not_found = ret;
1720         }
1721
1722         return 0;
1723 }
1724
1725 /*
1726  * ocfs2_xattr_ibody_set()
1727  *
1728  * Set, replace or remove an extended attribute into inode block.
1729  *
1730  */
1731 static int ocfs2_xattr_ibody_set(struct inode *inode,
1732                                  struct ocfs2_xattr_info *xi,
1733                                  struct ocfs2_xattr_search *xs)
1734 {
1735         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1736         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1737         int ret;
1738
1739         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1740                 return -ENOSPC;
1741
1742         down_write(&oi->ip_alloc_sem);
1743         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1744                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1745                         ret = -ENOSPC;
1746                         goto out;
1747                 }
1748         }
1749
1750         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1751                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1752 out:
1753         up_write(&oi->ip_alloc_sem);
1754
1755         return ret;
1756 }
1757
1758 /*
1759  * ocfs2_xattr_block_find()
1760  *
1761  * Find extended attribute in external block and
1762  * fill search info into struct ocfs2_xattr_search.
1763  */
1764 static int ocfs2_xattr_block_find(struct inode *inode,
1765                                   int name_index,
1766                                   const char *name,
1767                                   struct ocfs2_xattr_search *xs)
1768 {
1769         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1770         struct buffer_head *blk_bh = NULL;
1771         struct ocfs2_xattr_block *xb;
1772         int ret = 0;
1773
1774         if (!di->i_xattr_loc)
1775                 return ret;
1776
1777         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
1778         if (ret < 0) {
1779                 mlog_errno(ret);
1780                 return ret;
1781         }
1782
1783         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1784         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1785                 ret = -EIO;
1786                 goto cleanup;
1787         }
1788
1789         xs->xattr_bh = blk_bh;
1790
1791         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1792                 xs->header = &xb->xb_attrs.xb_header;
1793                 xs->base = (void *)xs->header;
1794                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1795                 xs->here = xs->header->xh_entries;
1796
1797                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1798         } else
1799                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1800                                                    name_index,
1801                                                    name, xs);
1802
1803         if (ret && ret != -ENODATA) {
1804                 xs->xattr_bh = NULL;
1805                 goto cleanup;
1806         }
1807         xs->not_found = ret;
1808         return 0;
1809 cleanup:
1810         brelse(blk_bh);
1811
1812         return ret;
1813 }
1814
1815 /*
1816  * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1817  * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1818  * re-initialized.
1819  */
1820 static int ocfs2_restore_xattr_block(struct inode *inode,
1821                                      struct ocfs2_xattr_search *xs)
1822 {
1823         int ret;
1824         handle_t *handle;
1825         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1826         struct ocfs2_xattr_block *xb =
1827                 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1828         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1829         u16 xb_flags = le16_to_cpu(xb->xb_flags);
1830
1831         BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1832                 le16_to_cpu(el->l_next_free_rec) != 0);
1833
1834         handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1835         if (IS_ERR(handle)) {
1836                 ret = PTR_ERR(handle);
1837                 handle = NULL;
1838                 goto out;
1839         }
1840
1841         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1842                                    OCFS2_JOURNAL_ACCESS_WRITE);
1843         if (ret < 0) {
1844                 mlog_errno(ret);
1845                 goto out_commit;
1846         }
1847
1848         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1849                offsetof(struct ocfs2_xattr_block, xb_attrs));
1850
1851         xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1852
1853         ocfs2_journal_dirty(handle, xs->xattr_bh);
1854
1855 out_commit:
1856         ocfs2_commit_trans(osb, handle);
1857 out:
1858         return ret;
1859 }
1860
1861 /*
1862  * ocfs2_xattr_block_set()
1863  *
1864  * Set, replace or remove an extended attribute into external block.
1865  *
1866  */
1867 static int ocfs2_xattr_block_set(struct inode *inode,
1868                                  struct ocfs2_xattr_info *xi,
1869                                  struct ocfs2_xattr_search *xs)
1870 {
1871         struct buffer_head *new_bh = NULL;
1872         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1873         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1874         struct ocfs2_alloc_context *meta_ac = NULL;
1875         handle_t *handle = NULL;
1876         struct ocfs2_xattr_block *xblk = NULL;
1877         u16 suballoc_bit_start;
1878         u32 num_got;
1879         u64 first_blkno;
1880         int ret;
1881
1882         if (!xs->xattr_bh) {
1883                 /*
1884                  * Alloc one external block for extended attribute
1885                  * outside of inode.
1886                  */
1887                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1888                 if (ret < 0) {
1889                         mlog_errno(ret);
1890                         goto out;
1891                 }
1892                 handle = ocfs2_start_trans(osb,
1893                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1894                 if (IS_ERR(handle)) {
1895                         ret = PTR_ERR(handle);
1896                         mlog_errno(ret);
1897                         goto out;
1898                 }
1899                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1900                                            OCFS2_JOURNAL_ACCESS_CREATE);
1901                 if (ret < 0) {
1902                         mlog_errno(ret);
1903                         goto out_commit;
1904                 }
1905
1906                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1907                                            &suballoc_bit_start, &num_got,
1908                                            &first_blkno);
1909                 if (ret < 0) {
1910                         mlog_errno(ret);
1911                         goto out_commit;
1912                 }
1913
1914                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1915                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1916
1917                 ret = ocfs2_journal_access(handle, inode, new_bh,
1918                                            OCFS2_JOURNAL_ACCESS_CREATE);
1919                 if (ret < 0) {
1920                         mlog_errno(ret);
1921                         goto out_commit;
1922                 }
1923
1924                 /* Initialize ocfs2_xattr_block */
1925                 xs->xattr_bh = new_bh;
1926                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1927                 memset(xblk, 0, inode->i_sb->s_blocksize);
1928                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1929                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1930                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1931                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1932                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1933
1934                 xs->header = &xblk->xb_attrs.xb_header;
1935                 xs->base = (void *)xs->header;
1936                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1937                 xs->here = xs->header->xh_entries;
1938
1939
1940                 ret = ocfs2_journal_dirty(handle, new_bh);
1941                 if (ret < 0) {
1942                         mlog_errno(ret);
1943                         goto out_commit;
1944                 }
1945                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1946                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1947                 if (ret < 0)
1948                         mlog_errno(ret);
1949 out_commit:
1950                 ocfs2_commit_trans(osb, handle);
1951 out:
1952                 if (meta_ac)
1953                         ocfs2_free_alloc_context(meta_ac);
1954                 if (ret < 0)
1955                         return ret;
1956         } else
1957                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1958
1959         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1960                 /* Set extended attribute into external block */
1961                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1962                 if (!ret || ret != -ENOSPC)
1963                         goto end;
1964
1965                 ret = ocfs2_xattr_create_index_block(inode, xs);
1966                 if (ret)
1967                         goto end;
1968         }
1969
1970         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1971         if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1972                 ret = ocfs2_restore_xattr_block(inode, xs);
1973
1974 end:
1975
1976         return ret;
1977 }
1978
1979 /*
1980  * ocfs2_xattr_set()
1981  *
1982  * Set, replace or remove an extended attribute for this inode.
1983  * value is NULL to remove an existing extended attribute, else either
1984  * create or replace an extended attribute.
1985  */
1986 int ocfs2_xattr_set(struct inode *inode,
1987                     int name_index,
1988                     const char *name,
1989                     const void *value,
1990                     size_t value_len,
1991                     int flags)
1992 {
1993         struct buffer_head *di_bh = NULL;
1994         struct ocfs2_dinode *di;
1995         int ret;
1996         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1997
1998         struct ocfs2_xattr_info xi = {
1999                 .name_index = name_index,
2000                 .name = name,
2001                 .value = value,
2002                 .value_len = value_len,
2003         };
2004
2005         struct ocfs2_xattr_search xis = {
2006                 .not_found = -ENODATA,
2007         };
2008
2009         struct ocfs2_xattr_search xbs = {
2010                 .not_found = -ENODATA,
2011         };
2012
2013         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2014                 return -EOPNOTSUPP;
2015
2016         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2017         if (ret < 0) {
2018                 mlog_errno(ret);
2019                 return ret;
2020         }
2021         xis.inode_bh = xbs.inode_bh = di_bh;
2022         di = (struct ocfs2_dinode *)di_bh->b_data;
2023
2024         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2025         /*
2026          * Scan inode and external block to find the same name
2027          * extended attribute and collect search infomation.
2028          */
2029         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2030         if (ret)
2031                 goto cleanup;
2032         if (xis.not_found) {
2033                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2034                 if (ret)
2035                         goto cleanup;
2036         }
2037
2038         if (xis.not_found && xbs.not_found) {
2039                 ret = -ENODATA;
2040                 if (flags & XATTR_REPLACE)
2041                         goto cleanup;
2042                 ret = 0;
2043                 if (!value)
2044                         goto cleanup;
2045         } else {
2046                 ret = -EEXIST;
2047                 if (flags & XATTR_CREATE)
2048                         goto cleanup;
2049         }
2050
2051         if (!value) {
2052                 /* Remove existing extended attribute */
2053                 if (!xis.not_found)
2054                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2055                 else if (!xbs.not_found)
2056                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2057         } else {
2058                 /* We always try to set extended attribute into inode first*/
2059                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2060                 if (!ret && !xbs.not_found) {
2061                         /*
2062                          * If succeed and that extended attribute existing in
2063                          * external block, then we will remove it.
2064                          */
2065                         xi.value = NULL;
2066                         xi.value_len = 0;
2067                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2068                 } else if (ret == -ENOSPC) {
2069                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2070                                 ret = ocfs2_xattr_block_find(inode, name_index,
2071                                                              name, &xbs);
2072                                 if (ret)
2073                                         goto cleanup;
2074                         }
2075                         /*
2076                          * If no space in inode, we will set extended attribute
2077                          * into external block.
2078                          */
2079                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2080                         if (ret)
2081                                 goto cleanup;
2082                         if (!xis.not_found) {
2083                                 /*
2084                                  * If succeed and that extended attribute
2085                                  * existing in inode, we will remove it.
2086                                  */
2087                                 xi.value = NULL;
2088                                 xi.value_len = 0;
2089                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2090                         }
2091                 }
2092         }
2093 cleanup:
2094         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2095         ocfs2_inode_unlock(inode, 1);
2096         brelse(di_bh);
2097         brelse(xbs.xattr_bh);
2098         for (i = 0; i < blk_per_bucket; i++)
2099                 brelse(xbs.bucket.bhs[i]);
2100
2101         return ret;
2102 }
2103
2104 /*
2105  * Find the xattr extent rec which may contains name_hash.
2106  * e_cpos will be the first name hash of the xattr rec.
2107  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2108  */
2109 static int ocfs2_xattr_get_rec(struct inode *inode,
2110                                u32 name_hash,
2111                                u64 *p_blkno,
2112                                u32 *e_cpos,
2113                                u32 *num_clusters,
2114                                struct ocfs2_extent_list *el)
2115 {
2116         int ret = 0, i;
2117         struct buffer_head *eb_bh = NULL;
2118         struct ocfs2_extent_block *eb;
2119         struct ocfs2_extent_rec *rec = NULL;
2120         u64 e_blkno = 0;
2121
2122         if (el->l_tree_depth) {
2123                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2124                 if (ret) {
2125                         mlog_errno(ret);
2126                         goto out;
2127                 }
2128
2129                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2130                 el = &eb->h_list;
2131
2132                 if (el->l_tree_depth) {
2133                         ocfs2_error(inode->i_sb,
2134                                     "Inode %lu has non zero tree depth in "
2135                                     "xattr tree block %llu\n", inode->i_ino,
2136                                     (unsigned long long)eb_bh->b_blocknr);
2137                         ret = -EROFS;
2138                         goto out;
2139                 }
2140         }
2141
2142         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2143                 rec = &el->l_recs[i];
2144
2145                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2146                         e_blkno = le64_to_cpu(rec->e_blkno);
2147                         break;
2148                 }
2149         }
2150
2151         if (!e_blkno) {
2152                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2153                             "record (%u, %u, 0) in xattr", inode->i_ino,
2154                             le32_to_cpu(rec->e_cpos),
2155                             ocfs2_rec_clusters(el, rec));
2156                 ret = -EROFS;
2157                 goto out;
2158         }
2159
2160         *p_blkno = le64_to_cpu(rec->e_blkno);
2161         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2162         if (e_cpos)
2163                 *e_cpos = le32_to_cpu(rec->e_cpos);
2164 out:
2165         brelse(eb_bh);
2166         return ret;
2167 }
2168
2169 typedef int (xattr_bucket_func)(struct inode *inode,
2170                                 struct ocfs2_xattr_bucket *bucket,
2171                                 void *para);
2172
2173 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2174                                    struct buffer_head *header_bh,
2175                                    int name_index,
2176                                    const char *name,
2177                                    u32 name_hash,
2178                                    u16 *xe_index,
2179                                    int *found)
2180 {
2181         int i, ret = 0, cmp = 1, block_off, new_offset;
2182         struct ocfs2_xattr_header *xh =
2183                         (struct ocfs2_xattr_header *)header_bh->b_data;
2184         size_t name_len = strlen(name);
2185         struct ocfs2_xattr_entry *xe = NULL;
2186         struct buffer_head *name_bh = NULL;
2187         char *xe_name;
2188
2189         /*
2190          * We don't use binary search in the bucket because there
2191          * may be multiple entries with the same name hash.
2192          */
2193         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2194                 xe = &xh->xh_entries[i];
2195
2196                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2197                         continue;
2198                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2199                         break;
2200
2201                 cmp = name_index - ocfs2_xattr_get_type(xe);
2202                 if (!cmp)
2203                         cmp = name_len - xe->xe_name_len;
2204                 if (cmp)
2205                         continue;
2206
2207                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2208                                                         xh,
2209                                                         i,
2210                                                         &block_off,
2211                                                         &new_offset);
2212                 if (ret) {
2213                         mlog_errno(ret);
2214                         break;
2215                 }
2216
2217                 ret = ocfs2_read_block(inode, header_bh->b_blocknr + block_off,
2218                                        &name_bh);
2219                 if (ret) {
2220                         mlog_errno(ret);
2221                         break;
2222                 }
2223                 xe_name = name_bh->b_data + new_offset;
2224
2225                 cmp = memcmp(name, xe_name, name_len);
2226                 brelse(name_bh);
2227                 name_bh = NULL;
2228
2229                 if (cmp == 0) {
2230                         *xe_index = i;
2231                         *found = 1;
2232                         ret = 0;
2233                         break;
2234                 }
2235         }
2236
2237         return ret;
2238 }
2239
2240 /*
2241  * Find the specified xattr entry in a series of buckets.
2242  * This series start from p_blkno and last for num_clusters.
2243  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2244  * the num of the valid buckets.
2245  *
2246  * Return the buffer_head this xattr should reside in. And if the xattr's
2247  * hash is in the gap of 2 buckets, return the lower bucket.
2248  */
2249 static int ocfs2_xattr_bucket_find(struct inode *inode,
2250                                    int name_index,
2251                                    const char *name,
2252                                    u32 name_hash,
2253                                    u64 p_blkno,
2254                                    u32 first_hash,
2255                                    u32 num_clusters,
2256                                    struct ocfs2_xattr_search *xs)
2257 {
2258         int ret, found = 0;
2259         struct buffer_head *bh = NULL;
2260         struct buffer_head *lower_bh = NULL;
2261         struct ocfs2_xattr_header *xh = NULL;
2262         struct ocfs2_xattr_entry *xe = NULL;
2263         u16 index = 0;
2264         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2265         int low_bucket = 0, bucket, high_bucket;
2266         u32 last_hash;
2267         u64 blkno;
2268
2269         ret = ocfs2_read_block(inode, p_blkno, &bh);
2270         if (ret) {
2271                 mlog_errno(ret);
2272                 goto out;
2273         }
2274
2275         xh = (struct ocfs2_xattr_header *)bh->b_data;
2276         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2277
2278         while (low_bucket <= high_bucket) {
2279                 brelse(bh);
2280                 bh = NULL;
2281                 bucket = (low_bucket + high_bucket) / 2;
2282
2283                 blkno = p_blkno + bucket * blk_per_bucket;
2284
2285                 ret = ocfs2_read_block(inode, blkno, &bh);
2286                 if (ret) {
2287                         mlog_errno(ret);
2288                         goto out;
2289                 }
2290
2291                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2292                 xe = &xh->xh_entries[0];
2293                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2294                         high_bucket = bucket - 1;
2295                         continue;
2296                 }
2297
2298                 /*
2299                  * Check whether the hash of the last entry in our
2300                  * bucket is larger than the search one. for an empty
2301                  * bucket, the last one is also the first one.
2302                  */
2303                 if (xh->xh_count)
2304                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2305
2306                 last_hash = le32_to_cpu(xe->xe_name_hash);
2307
2308                 /* record lower_bh which may be the insert place. */
2309                 brelse(lower_bh);
2310                 lower_bh = bh;
2311                 bh = NULL;
2312
2313                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2314                         low_bucket = bucket + 1;
2315                         continue;
2316                 }
2317
2318                 /* the searched xattr should reside in this bucket if exists. */
2319                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2320                                               name_index, name, name_hash,
2321                                               &index, &found);
2322                 if (ret) {
2323                         mlog_errno(ret);
2324                         goto out;
2325                 }
2326                 break;
2327         }
2328
2329         /*
2330          * Record the bucket we have found.
2331          * When the xattr's hash value is in the gap of 2 buckets, we will
2332          * always set it to the previous bucket.
2333          */
2334         if (!lower_bh) {
2335                 /*
2336                  * We can't find any bucket whose first name_hash is less
2337                  * than the find name_hash.
2338                  */
2339                 BUG_ON(bh->b_blocknr != p_blkno);
2340                 lower_bh = bh;
2341                 bh = NULL;
2342         }
2343         xs->bucket.bhs[0] = lower_bh;
2344         xs->bucket.xh = (struct ocfs2_xattr_header *)
2345                                         xs->bucket.bhs[0]->b_data;
2346         lower_bh = NULL;
2347
2348         xs->header = xs->bucket.xh;
2349         xs->base = xs->bucket.bhs[0]->b_data;
2350         xs->end = xs->base + inode->i_sb->s_blocksize;
2351
2352         if (found) {
2353                 /*
2354                  * If we have found the xattr enty, read all the blocks in
2355                  * this bucket.
2356                  */
2357                 ret = ocfs2_read_blocks(inode, xs->bucket.bhs[0]->b_blocknr + 1,
2358                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2359                                         0);
2360                 if (ret) {
2361                         mlog_errno(ret);
2362                         goto out;
2363                 }
2364
2365                 xs->here = &xs->header->xh_entries[index];
2366                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2367                      (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2368         } else
2369                 ret = -ENODATA;
2370
2371 out:
2372         brelse(bh);
2373         brelse(lower_bh);
2374         return ret;
2375 }
2376
2377 static int ocfs2_xattr_index_block_find(struct inode *inode,
2378                                         struct buffer_head *root_bh,
2379                                         int name_index,
2380                                         const char *name,
2381                                         struct ocfs2_xattr_search *xs)
2382 {
2383         int ret;
2384         struct ocfs2_xattr_block *xb =
2385                         (struct ocfs2_xattr_block *)root_bh->b_data;
2386         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2387         struct ocfs2_extent_list *el = &xb_root->xt_list;
2388         u64 p_blkno = 0;
2389         u32 first_hash, num_clusters = 0;
2390         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2391
2392         if (le16_to_cpu(el->l_next_free_rec) == 0)
2393                 return -ENODATA;
2394
2395         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2396              name, name_hash, name_index);
2397
2398         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2399                                   &num_clusters, el);
2400         if (ret) {
2401                 mlog_errno(ret);
2402                 goto out;
2403         }
2404
2405         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2406
2407         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2408              "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2409
2410         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2411                                       p_blkno, first_hash, num_clusters, xs);
2412
2413 out:
2414         return ret;
2415 }
2416
2417 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2418                                        u64 blkno,
2419                                        u32 clusters,
2420                                        xattr_bucket_func *func,
2421                                        void *para)
2422 {
2423         int i, j, ret = 0;
2424         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2425         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2426         u32 num_buckets = clusters * bpc;
2427         struct ocfs2_xattr_bucket bucket;
2428
2429         memset(&bucket, 0, sizeof(bucket));
2430
2431         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2432              clusters, blkno);
2433
2434         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2435                 ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket,
2436                                         bucket.bhs, 0);
2437                 if (ret) {
2438                         mlog_errno(ret);
2439                         goto out;
2440                 }
2441
2442                 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2443                 /*
2444                  * The real bucket num in this series of blocks is stored
2445                  * in the 1st bucket.
2446                  */
2447                 if (i == 0)
2448                         num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2449
2450                 mlog(0, "iterating xattr bucket %llu, first hash %u\n", blkno,
2451                      le32_to_cpu(bucket.xh->xh_entries[0].xe_name_hash));
2452                 if (func) {
2453                         ret = func(inode, &bucket, para);
2454                         if (ret) {
2455                                 mlog_errno(ret);
2456                                 break;
2457                         }
2458                 }
2459
2460                 for (j = 0; j < blk_per_bucket; j++)
2461                         brelse(bucket.bhs[j]);
2462                 memset(&bucket, 0, sizeof(bucket));
2463         }
2464
2465 out:
2466         for (j = 0; j < blk_per_bucket; j++)
2467                 brelse(bucket.bhs[j]);
2468
2469         return ret;
2470 }
2471
2472 struct ocfs2_xattr_tree_list {
2473         char *buffer;
2474         size_t buffer_size;
2475         size_t result;
2476 };
2477
2478 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2479                                              struct ocfs2_xattr_header *xh,
2480                                              int index,
2481                                              int *block_off,
2482                                              int *new_offset)
2483 {
2484         u16 name_offset;
2485
2486         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2487                 return -EINVAL;
2488
2489         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2490
2491         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2492         *new_offset = name_offset % inode->i_sb->s_blocksize;
2493
2494         return 0;
2495 }
2496
2497 static int ocfs2_list_xattr_bucket(struct inode *inode,
2498                                    struct ocfs2_xattr_bucket *bucket,
2499                                    void *para)
2500 {
2501         int ret = 0, type;
2502         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2503         int i, block_off, new_offset;
2504         const char *prefix, *name;
2505
2506         for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2507                 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2508                 type = ocfs2_xattr_get_type(entry);
2509                 prefix = ocfs2_xattr_prefix(type);
2510
2511                 if (prefix) {
2512                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2513                                                                 bucket->xh,
2514                                                                 i,
2515                                                                 &block_off,
2516                                                                 &new_offset);
2517                         if (ret)
2518                                 break;
2519
2520                         name = (const char *)bucket->bhs[block_off]->b_data +
2521                                 new_offset;
2522                         ret = ocfs2_xattr_list_entry(xl->buffer,
2523                                                      xl->buffer_size,
2524                                                      &xl->result,
2525                                                      prefix, name,
2526                                                      entry->xe_name_len);
2527                         if (ret)
2528                                 break;
2529                 }
2530         }
2531
2532         return ret;
2533 }
2534
2535 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2536                                              struct ocfs2_xattr_tree_root *xt,
2537                                              char *buffer,
2538                                              size_t buffer_size)
2539 {
2540         struct ocfs2_extent_list *el = &xt->xt_list;
2541         int ret = 0;
2542         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2543         u64 p_blkno = 0;
2544         struct ocfs2_xattr_tree_list xl = {
2545                 .buffer = buffer,
2546                 .buffer_size = buffer_size,
2547                 .result = 0,
2548         };
2549
2550         if (le16_to_cpu(el->l_next_free_rec) == 0)
2551                 return 0;
2552
2553         while (name_hash > 0) {
2554                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2555                                           &e_cpos, &num_clusters, el);
2556                 if (ret) {
2557                         mlog_errno(ret);
2558                         goto out;
2559                 }
2560
2561                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2562                                                   ocfs2_list_xattr_bucket,
2563                                                   &xl);
2564                 if (ret) {
2565                         mlog_errno(ret);
2566                         goto out;
2567                 }
2568
2569                 if (e_cpos == 0)
2570                         break;
2571
2572                 name_hash = e_cpos - 1;
2573         }
2574
2575         ret = xl.result;
2576 out:
2577         return ret;
2578 }
2579
2580 static int cmp_xe(const void *a, const void *b)
2581 {
2582         const struct ocfs2_xattr_entry *l = a, *r = b;
2583         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2584         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2585
2586         if (l_hash > r_hash)
2587                 return 1;
2588         if (l_hash < r_hash)
2589                 return -1;
2590         return 0;
2591 }
2592
2593 static void swap_xe(void *a, void *b, int size)
2594 {
2595         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2596
2597         tmp = *l;
2598         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2599         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2600 }
2601
2602 /*
2603  * When the ocfs2_xattr_block is filled up, new bucket will be created
2604  * and all the xattr entries will be moved to the new bucket.
2605  * Note: we need to sort the entries since they are not saved in order
2606  * in the ocfs2_xattr_block.
2607  */
2608 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2609                                            struct buffer_head *xb_bh,
2610                                            struct buffer_head *xh_bh,
2611                                            struct buffer_head *data_bh)
2612 {
2613         int i, blocksize = inode->i_sb->s_blocksize;
2614         u16 offset, size, off_change;
2615         struct ocfs2_xattr_entry *xe;
2616         struct ocfs2_xattr_block *xb =
2617                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2618         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2619         struct ocfs2_xattr_header *xh =
2620                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2621         u16 count = le16_to_cpu(xb_xh->xh_count);
2622         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2623
2624         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2625              (unsigned long long)xb_bh->b_blocknr,
2626              (unsigned long long)xh_bh->b_blocknr);
2627
2628         memset(xh_bh->b_data, 0, blocksize);
2629         if (data_bh)
2630                 memset(data_bh->b_data, 0, blocksize);
2631         /*
2632          * Since the xe_name_offset is based on ocfs2_xattr_header,
2633          * there is a offset change corresponding to the change of
2634          * ocfs2_xattr_header's position.
2635          */
2636         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2637         xe = &xb_xh->xh_entries[count - 1];
2638         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2639         size = blocksize - offset;
2640
2641         /* copy all the names and values. */
2642         if (data_bh)
2643                 target = data_bh->b_data;
2644         memcpy(target + offset, src + offset, size);
2645
2646         /* Init new header now. */
2647         xh->xh_count = xb_xh->xh_count;
2648         xh->xh_num_buckets = cpu_to_le16(1);
2649         xh->xh_name_value_len = cpu_to_le16(size);
2650         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2651
2652         /* copy all the entries. */
2653         target = xh_bh->b_data;
2654         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2655         size = count * sizeof(struct ocfs2_xattr_entry);
2656         memcpy(target + offset, (char *)xb_xh + offset, size);
2657
2658         /* Change the xe offset for all the xe because of the move. */
2659         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2660                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2661         for (i = 0; i < count; i++)
2662                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2663
2664         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2665              offset, size, off_change);
2666
2667         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2668              cmp_xe, swap_xe);
2669 }
2670
2671 /*
2672  * After we move xattr from block to index btree, we have to
2673  * update ocfs2_xattr_search to the new xe and base.
2674  *
2675  * When the entry is in xattr block, xattr_bh indicates the storage place.
2676  * While if the entry is in index b-tree, "bucket" indicates the
2677  * real place of the xattr.
2678  */
2679 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2680                                            struct ocfs2_xattr_search *xs,
2681                                            struct buffer_head *old_bh,
2682                                            struct buffer_head *new_bh)
2683 {
2684         int ret = 0;
2685         char *buf = old_bh->b_data;
2686         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2687         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2688         int i, blocksize = inode->i_sb->s_blocksize;
2689         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2690
2691         xs->bucket.bhs[0] = new_bh;
2692         get_bh(new_bh);
2693         xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2694         xs->header = xs->bucket.xh;
2695
2696         xs->base = new_bh->b_data;
2697         xs->end = xs->base + inode->i_sb->s_blocksize;
2698
2699         if (!xs->not_found) {
2700                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2701                         ret = ocfs2_read_blocks(inode,
2702                                         xs->bucket.bhs[0]->b_blocknr + 1,
2703                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2704                                         0);
2705                         if (ret) {
2706                                 mlog_errno(ret);
2707                                 return ret;
2708                         }
2709
2710                         i = xs->here - old_xh->xh_entries;
2711                         xs->here = &xs->header->xh_entries[i];
2712                 }
2713         }
2714
2715         return ret;
2716 }
2717
2718 static int ocfs2_xattr_create_index_block(struct inode *inode,
2719                                           struct ocfs2_xattr_search *xs)
2720 {
2721         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2722         u32 bit_off, len;
2723         u64 blkno;
2724         handle_t *handle;
2725         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2726         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2727         struct ocfs2_alloc_context *data_ac;
2728         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2729         struct buffer_head *xb_bh = xs->xattr_bh;
2730         struct ocfs2_xattr_block *xb =
2731                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2732         struct ocfs2_xattr_tree_root *xr;
2733         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2734         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2735
2736         mlog(0, "create xattr index block for %llu\n",
2737              (unsigned long long)xb_bh->b_blocknr);
2738
2739         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2740
2741         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2742         if (ret) {
2743                 mlog_errno(ret);
2744                 goto out;
2745         }
2746
2747         /*
2748          * XXX:
2749          * We can use this lock for now, and maybe move to a dedicated mutex
2750          * if performance becomes a problem later.
2751          */
2752         down_write(&oi->ip_alloc_sem);
2753
2754         /*
2755          * 3 more credits, one for xattr block update, one for the 1st block
2756          * of the new xattr bucket and one for the value/data.
2757          */
2758         credits += 3;
2759         handle = ocfs2_start_trans(osb, credits);
2760         if (IS_ERR(handle)) {
2761                 ret = PTR_ERR(handle);
2762                 mlog_errno(ret);
2763                 goto out_sem;
2764         }
2765
2766         ret = ocfs2_journal_access(handle, inode, xb_bh,
2767                                    OCFS2_JOURNAL_ACCESS_WRITE);
2768         if (ret) {
2769                 mlog_errno(ret);
2770                 goto out_commit;
2771         }
2772
2773         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2774         if (ret) {
2775                 mlog_errno(ret);
2776                 goto out_commit;
2777         }
2778
2779         /*
2780          * The bucket may spread in many blocks, and
2781          * we will only touch the 1st block and the last block
2782          * in the whole bucket(one for entry and one for data).
2783          */
2784         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2785
2786         mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2787
2788         xh_bh = sb_getblk(inode->i_sb, blkno);
2789         if (!xh_bh) {
2790                 ret = -EIO;
2791                 mlog_errno(ret);
2792                 goto out_commit;
2793         }
2794
2795         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2796
2797         ret = ocfs2_journal_access(handle, inode, xh_bh,
2798                                    OCFS2_JOURNAL_ACCESS_CREATE);
2799         if (ret) {
2800                 mlog_errno(ret);
2801                 goto out_commit;
2802         }
2803
2804         if (bpb > 1) {
2805                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2806                 if (!data_bh) {
2807                         ret = -EIO;
2808                         mlog_errno(ret);
2809                         goto out_commit;
2810                 }
2811
2812                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2813
2814                 ret = ocfs2_journal_access(handle, inode, data_bh,
2815                                            OCFS2_JOURNAL_ACCESS_CREATE);
2816                 if (ret) {
2817                         mlog_errno(ret);
2818                         goto out_commit;
2819                 }
2820         }
2821
2822         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2823
2824         ocfs2_journal_dirty(handle, xh_bh);
2825         if (data_bh)
2826                 ocfs2_journal_dirty(handle, data_bh);
2827
2828         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2829
2830         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2831         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2832                offsetof(struct ocfs2_xattr_block, xb_attrs));
2833
2834         xr = &xb->xb_attrs.xb_root;
2835         xr->xt_clusters = cpu_to_le32(1);
2836         xr->xt_last_eb_blk = 0;
2837         xr->xt_list.l_tree_depth = 0;
2838         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2839         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2840
2841         xr->xt_list.l_recs[0].e_cpos = 0;
2842         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2843         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2844
2845         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2846
2847         ret = ocfs2_journal_dirty(handle, xb_bh);
2848         if (ret) {
2849                 mlog_errno(ret);
2850                 goto out_commit;
2851         }
2852
2853 out_commit:
2854         ocfs2_commit_trans(osb, handle);
2855
2856 out_sem:
2857         up_write(&oi->ip_alloc_sem);
2858
2859 out:
2860         if (data_ac)
2861                 ocfs2_free_alloc_context(data_ac);
2862
2863         brelse(xh_bh);
2864         brelse(data_bh);
2865
2866         return ret;
2867 }
2868
2869 static int cmp_xe_offset(const void *a, const void *b)
2870 {
2871         const struct ocfs2_xattr_entry *l = a, *r = b;
2872         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2873         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2874
2875         if (l_name_offset < r_name_offset)
2876                 return 1;
2877         if (l_name_offset > r_name_offset)
2878                 return -1;
2879         return 0;
2880 }
2881
2882 /*
2883  * defrag a xattr bucket if we find that the bucket has some
2884  * holes beteen name/value pairs.
2885  * We will move all the name/value pairs to the end of the bucket
2886  * so that we can spare some space for insertion.
2887  */
2888 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2889                                      struct ocfs2_xattr_bucket *bucket)
2890 {
2891         int ret, i;
2892         size_t end, offset, len, value_len;
2893         struct ocfs2_xattr_header *xh;
2894         char *entries, *buf, *bucket_buf = NULL;
2895         u64 blkno = bucket->bhs[0]->b_blocknr;
2896         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2897         u16 xh_free_start;
2898         size_t blocksize = inode->i_sb->s_blocksize;
2899         handle_t *handle;
2900         struct buffer_head **bhs;
2901         struct ocfs2_xattr_entry *xe;
2902
2903         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2904                         GFP_NOFS);
2905         if (!bhs)
2906                 return -ENOMEM;
2907
2908         ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket, bhs, 0);
2909         if (ret)
2910                 goto out;
2911
2912         /*
2913          * In order to make the operation more efficient and generic,
2914          * we copy all the blocks into a contiguous memory and do the
2915          * defragment there, so if anything is error, we will not touch
2916          * the real block.
2917          */
2918         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2919         if (!bucket_buf) {
2920                 ret = -EIO;
2921                 goto out;
2922         }
2923
2924         buf = bucket_buf;
2925         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2926                 memcpy(buf, bhs[i]->b_data, blocksize);
2927
2928         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2929         if (IS_ERR(handle)) {
2930                 ret = PTR_ERR(handle);
2931                 handle = NULL;
2932                 mlog_errno(ret);
2933                 goto out;
2934         }
2935
2936         for (i = 0; i < blk_per_bucket; i++) {
2937                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2938                                            OCFS2_JOURNAL_ACCESS_WRITE);
2939                 if (ret < 0) {
2940                         mlog_errno(ret);
2941                         goto commit;
2942                 }
2943         }
2944
2945         xh = (struct ocfs2_xattr_header *)bucket_buf;
2946         entries = (char *)xh->xh_entries;
2947         xh_free_start = le16_to_cpu(xh->xh_free_start);
2948
2949         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2950              "xh_free_start = %u, xh_name_value_len = %u.\n",
2951              blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2952              le16_to_cpu(xh->xh_name_value_len));
2953
2954         /*
2955          * sort all the entries by their offset.
2956          * the largest will be the first, so that we can
2957          * move them to the end one by one.
2958          */
2959         sort(entries, le16_to_cpu(xh->xh_count),
2960              sizeof(struct ocfs2_xattr_entry),
2961              cmp_xe_offset, swap_xe);
2962
2963         /* Move all name/values to the end of the bucket. */
2964         xe = xh->xh_entries;
2965         end = OCFS2_XATTR_BUCKET_SIZE;
2966         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2967                 offset = le16_to_cpu(xe->xe_name_offset);
2968                 if (ocfs2_xattr_is_local(xe))
2969                         value_len = OCFS2_XATTR_SIZE(
2970                                         le64_to_cpu(xe->xe_value_size));
2971                 else
2972                         value_len = OCFS2_XATTR_ROOT_SIZE;
2973                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2974
2975                 /*
2976                  * We must make sure that the name/value pair
2977                  * exist in the same block. So adjust end to
2978                  * the previous block end if needed.
2979                  */
2980                 if (((end - len) / blocksize !=
2981                         (end - 1) / blocksize))
2982                         end = end - end % blocksize;
2983
2984                 if (end > offset + len) {
2985                         memmove(bucket_buf + end - len,
2986                                 bucket_buf + offset, len);
2987                         xe->xe_name_offset = cpu_to_le16(end - len);
2988                 }
2989
2990                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2991                                 "bucket %llu\n", (unsigned long long)blkno);
2992
2993                 end -= len;
2994         }
2995
2996         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2997                         "bucket %llu\n", (unsigned long long)blkno);
2998
2999         if (xh_free_start == end)
3000                 goto commit;
3001
3002         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3003         xh->xh_free_start = cpu_to_le16(end);
3004
3005         /* sort the entries by their name_hash. */
3006         sort(entries, le16_to_cpu(xh->xh_count),
3007              sizeof(struct ocfs2_xattr_entry),
3008              cmp_xe, swap_xe);
3009
3010         buf = bucket_buf;
3011         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3012                 memcpy(bhs[i]->b_data, buf, blocksize);
3013                 ocfs2_journal_dirty(handle, bhs[i]);
3014         }
3015
3016 commit:
3017         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3018 out:
3019
3020         if (bhs) {
3021                 for (i = 0; i < blk_per_bucket; i++)
3022                         brelse(bhs[i]);
3023         }
3024         kfree(bhs);
3025
3026         kfree(bucket_buf);
3027         return ret;
3028 }
3029
3030 /*
3031  * Move half nums of the xattr bucket in the previous cluster to this new
3032  * cluster. We only touch the last cluster of the previous extend record.
3033  *
3034  * first_bh is the first buffer_head of a series of bucket in the same
3035  * extent rec and header_bh is the header of one bucket in this cluster.
3036  * They will be updated if we move the data header_bh contains to the new
3037  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3038  */
3039 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3040                                                handle_t *handle,
3041                                                struct buffer_head **first_bh,
3042                                                struct buffer_head **header_bh,
3043                                                u64 new_blkno,
3044                                                u64 prev_blkno,
3045                                                u32 num_clusters,
3046                                                u32 *first_hash)
3047 {
3048         int i, ret, credits;
3049         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3050         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3051         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3052         int blocksize = inode->i_sb->s_blocksize;
3053         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3054         struct ocfs2_xattr_header *new_xh;
3055         struct ocfs2_xattr_header *xh =
3056                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3057
3058         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3059         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3060
3061         prev_bh = *first_bh;
3062         get_bh(prev_bh);
3063         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3064
3065         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3066
3067         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3068              prev_blkno, new_blkno);
3069
3070         /*
3071          * We need to update the 1st half of the new cluster and
3072          * 1 more for the update of the 1st bucket of the previous
3073          * extent record.
3074          */
3075         credits = bpc / 2 + 1;
3076         ret = ocfs2_extend_trans(handle, credits);
3077         if (ret) {
3078                 mlog_errno(ret);
3079                 goto out;
3080         }
3081
3082         ret = ocfs2_journal_access(handle, inode, prev_bh,
3083                                    OCFS2_JOURNAL_ACCESS_WRITE);
3084         if (ret) {
3085                 mlog_errno(ret);
3086                 goto out;
3087         }
3088
3089         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3090                 old_bh = new_bh = NULL;
3091                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3092                 if (!new_bh) {
3093                         ret = -EIO;
3094                         mlog_errno(ret);
3095                         goto out;
3096                 }
3097
3098                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3099
3100                 ret = ocfs2_journal_access(handle, inode, new_bh,
3101                                            OCFS2_JOURNAL_ACCESS_CREATE);
3102                 if (ret < 0) {
3103                         mlog_errno(ret);
3104                         brelse(new_bh);
3105                         goto out;
3106                 }
3107
3108                 ret = ocfs2_read_block(inode, prev_blkno, &old_bh);
3109                 if (ret < 0) {
3110                         mlog_errno(ret);
3111                         brelse(new_bh);
3112                         goto out;
3113                 }
3114
3115                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3116
3117                 if (i == 0) {
3118                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3119                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3120
3121                         if (first_hash)
3122                                 *first_hash = le32_to_cpu(
3123                                         new_xh->xh_entries[0].xe_name_hash);
3124                         new_first_bh = new_bh;
3125                         get_bh(new_first_bh);
3126                 }
3127
3128                 ocfs2_journal_dirty(handle, new_bh);
3129
3130                 if (*header_bh == old_bh) {
3131                         brelse(*header_bh);
3132                         *header_bh = new_bh;
3133                         get_bh(*header_bh);
3134
3135                         brelse(*first_bh);
3136                         *first_bh = new_first_bh;
3137                         get_bh(*first_bh);
3138                 }
3139                 brelse(new_bh);
3140                 brelse(old_bh);
3141         }
3142
3143         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3144
3145         ocfs2_journal_dirty(handle, prev_bh);
3146 out:
3147         brelse(prev_bh);
3148         brelse(new_first_bh);
3149         return ret;
3150 }
3151
3152 static int ocfs2_read_xattr_bucket(struct inode *inode,
3153                                    u64 blkno,
3154                                    struct buffer_head **bhs,
3155                                    int new)
3156 {
3157         int ret = 0;
3158         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3159
3160         if (!new)
3161                 return ocfs2_read_blocks(inode, blkno,
3162                                          blk_per_bucket, bhs, 0);
3163
3164         for (i = 0; i < blk_per_bucket; i++) {
3165                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3166                 if (bhs[i] == NULL) {
3167                         ret = -EIO;
3168                         mlog_errno(ret);
3169                         break;
3170                 }
3171                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3172         }
3173
3174         return ret;
3175 }
3176
3177 /*
3178  * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3179  * first_hash will record the 1st hash of the new bucket.
3180  */
3181 static int ocfs2_half_xattr_bucket(struct inode *inode,
3182                                    handle_t *handle,
3183                                    u64 blk,
3184                                    u64 new_blk,
3185                                    u32 *first_hash,
3186                                    int new_bucket_head)
3187 {
3188         int ret, i;
3189         u16 count, start, len, name_value_len, xe_len, name_offset;
3190         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3191         struct buffer_head **s_bhs, **t_bhs = NULL;
3192         struct ocfs2_xattr_header *xh;
3193         struct ocfs2_xattr_entry *xe;
3194         int blocksize = inode->i_sb->s_blocksize;
3195
3196         mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3197              blk, new_blk);
3198
3199         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3200         if (!s_bhs)
3201                 return -ENOMEM;
3202
3203         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3204         if (ret) {
3205                 mlog_errno(ret);
3206                 goto out;
3207         }
3208
3209         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3210                                    OCFS2_JOURNAL_ACCESS_WRITE);
3211         if (ret) {
3212                 mlog_errno(ret);
3213                 goto out;
3214         }
3215
3216         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3217         if (!t_bhs) {
3218                 ret = -ENOMEM;
3219                 goto out;
3220         }
3221
3222         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3223         if (ret) {
3224                 mlog_errno(ret);
3225                 goto out;
3226         }
3227
3228         for (i = 0; i < blk_per_bucket; i++) {
3229                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3230                                            OCFS2_JOURNAL_ACCESS_CREATE);
3231                 if (ret) {
3232                         mlog_errno(ret);
3233                         goto out;
3234                 }
3235         }
3236
3237         /* copy the whole bucket to the new first. */
3238         for (i = 0; i < blk_per_bucket; i++)
3239                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3240
3241         /* update the new bucket. */
3242         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3243         count = le16_to_cpu(xh->xh_count);
3244         start = count / 2;
3245
3246         /*
3247          * Calculate the total name/value len and xh_free_start for
3248          * the old bucket first.
3249          */
3250         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3251         name_value_len = 0;
3252         for (i = 0; i < start; i++) {
3253                 xe = &xh->xh_entries[i];
3254                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3255                 if (ocfs2_xattr_is_local(xe))
3256                         xe_len +=
3257                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3258                 else
3259                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3260                 name_value_len += xe_len;
3261                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3262                         name_offset = le16_to_cpu(xe->xe_name_offset);
3263         }
3264
3265         /*
3266          * Now begin the modification to the new bucket.
3267          *
3268          * In the new bucket, We just move the xattr entry to the beginning
3269          * and don't touch the name/value. So there will be some holes in the
3270          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3271          * called.
3272          */
3273         xe = &xh->xh_entries[start];
3274         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3275         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3276              (int)((char *)xe - (char *)xh),
3277              (int)((char *)xh->xh_entries - (char *)xh));
3278         memmove((char *)xh->xh_entries, (char *)xe, len);
3279         xe = &xh->xh_entries[count - start];
3280         len = sizeof(struct ocfs2_xattr_entry) * start;
3281         memset((char *)xe, 0, len);
3282
3283         le16_add_cpu(&xh->xh_count, -start);
3284         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3285
3286         /* Calculate xh_free_start for the new bucket. */
3287         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3288         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3289                 xe = &xh->xh_entries[i];
3290                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3291                 if (ocfs2_xattr_is_local(xe))
3292                         xe_len +=
3293                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3294                 else
3295                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3296                 if (le16_to_cpu(xe->xe_name_offset) <
3297                     le16_to_cpu(xh->xh_free_start))
3298                         xh->xh_free_start = xe->xe_name_offset;
3299         }
3300
3301         /* set xh->xh_num_buckets for the new xh. */
3302         if (new_bucket_head)
3303                 xh->xh_num_buckets = cpu_to_le16(1);
3304         else
3305                 xh->xh_num_buckets = 0;
3306
3307         for (i = 0; i < blk_per_bucket; i++) {
3308                 ocfs2_journal_dirty(handle, t_bhs[i]);
3309                 if (ret)
3310                         mlog_errno(ret);
3311         }
3312
3313         /* store the first_hash of the new bucket. */
3314         if (first_hash)
3315                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3316
3317         /*
3318          * Now only update the 1st block of the old bucket.
3319          * Please note that the entry has been sorted already above.
3320          */
3321         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3322         memset(&xh->xh_entries[start], 0,
3323                sizeof(struct ocfs2_xattr_entry) * (count - start));
3324         xh->xh_count = cpu_to_le16(start);
3325         xh->xh_free_start = cpu_to_le16(name_offset);
3326         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3327
3328         ocfs2_journal_dirty(handle, s_bhs[0]);
3329         if (ret)
3330                 mlog_errno(ret);
3331
3332 out:
3333         if (s_bhs) {
3334                 for (i = 0; i < blk_per_bucket; i++)
3335                         brelse(s_bhs[i]);
3336         }
3337         kfree(s_bhs);
3338
3339         if (t_bhs) {
3340                 for (i = 0; i < blk_per_bucket; i++)
3341                         brelse(t_bhs[i]);
3342         }
3343         kfree(t_bhs);
3344
3345         return ret;
3346 }
3347
3348 /*
3349  * Copy xattr from one bucket to another bucket.
3350  *
3351  * The caller must make sure that the journal transaction
3352  * has enough space for journaling.
3353  */
3354 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3355                                  handle_t *handle,
3356                                  u64 s_blkno,
3357                                  u64 t_blkno,
3358                                  int t_is_new)
3359 {
3360         int ret, i;
3361         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3362         int blocksize = inode->i_sb->s_blocksize;
3363         struct buffer_head **s_bhs, **t_bhs = NULL;
3364
3365         BUG_ON(s_blkno == t_blkno);
3366
3367         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3368              s_blkno, t_blkno, t_is_new);
3369
3370         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3371                         GFP_NOFS);
3372         if (!s_bhs)
3373                 return -ENOMEM;
3374
3375         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3376         if (ret)
3377                 goto out;
3378
3379         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3380                         GFP_NOFS);
3381         if (!t_bhs) {
3382                 ret = -ENOMEM;
3383                 goto out;
3384         }
3385
3386         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3387         if (ret)
3388                 goto out;
3389
3390         for (i = 0; i < blk_per_bucket; i++) {
3391                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3392                                            OCFS2_JOURNAL_ACCESS_WRITE);
3393                 if (ret)
3394                         goto out;
3395         }
3396
3397         for (i = 0; i < blk_per_bucket; i++) {
3398                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3399                 ocfs2_journal_dirty(handle, t_bhs[i]);
3400         }
3401
3402 out:
3403         if (s_bhs) {
3404                 for (i = 0; i < blk_per_bucket; i++)
3405                         brelse(s_bhs[i]);
3406         }
3407         kfree(s_bhs);
3408
3409         if (t_bhs) {
3410                 for (i = 0; i < blk_per_bucket; i++)
3411                         brelse(t_bhs[i]);
3412         }
3413         kfree(t_bhs);
3414
3415         return ret;
3416 }
3417
3418 /*
3419  * Copy one xattr cluster from src_blk to to_blk.
3420  * The to_blk will become the first bucket header of the cluster, so its
3421  * xh_num_buckets will be initialized as the bucket num in the cluster.
3422  */
3423 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3424                                   handle_t *handle,
3425                                   struct buffer_head *first_bh,
3426                                   u64 src_blk,
3427                                   u64 to_blk,
3428                                   u32 *first_hash)
3429 {
3430         int i, ret, credits;
3431         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3432         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3433         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3434         struct buffer_head *bh = NULL;
3435         struct ocfs2_xattr_header *xh;
3436         u64 to_blk_start = to_blk;
3437
3438         mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3439
3440         /*
3441          * We need to update the new cluster and 1 more for the update of
3442          * the 1st bucket of the previous extent rec.
3443          */
3444         credits = bpc + 1;
3445         ret = ocfs2_extend_trans(handle, credits);
3446         if (ret) {
3447                 mlog_errno(ret);
3448                 goto out;
3449         }
3450
3451         ret = ocfs2_journal_access(handle, inode, first_bh,
3452                                    OCFS2_JOURNAL_ACCESS_WRITE);
3453         if (ret) {
3454                 mlog_errno(ret);
3455                 goto out;
3456         }
3457
3458         for (i = 0; i < num_buckets; i++) {
3459                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3460                                             src_blk, to_blk, 1);
3461                 if (ret) {
3462                         mlog_errno(ret);
3463                         goto out;
3464                 }
3465
3466                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3467                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3468         }
3469
3470         /* update the old bucket header. */
3471         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3472         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3473
3474         ocfs2_journal_dirty(handle, first_bh);
3475
3476         /* update the new bucket header. */
3477         ret = ocfs2_read_block(inode, to_blk_start, &bh);
3478         if (ret < 0) {
3479                 mlog_errno(ret);
3480                 goto out;
3481         }
3482
3483         ret = ocfs2_journal_access(handle, inode, bh,
3484                                    OCFS2_JOURNAL_ACCESS_WRITE);
3485         if (ret) {
3486                 mlog_errno(ret);
3487                 goto out;
3488         }
3489
3490         xh = (struct ocfs2_xattr_header *)bh->b_data;
3491         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3492
3493         ocfs2_journal_dirty(handle, bh);
3494
3495         if (first_hash)
3496                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3497 out:
3498         brelse(bh);
3499         return ret;
3500 }
3501
3502 /*
3503  * Move half of the xattrs in this cluster to the new cluster.
3504  * This function should only be called when bucket size == cluster size.
3505  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3506  */
3507 static int ocfs2_half_xattr_cluster(struct inode *inode,
3508                                     handle_t *handle,
3509                                     u64 prev_blk,
3510                                     u64 new_blk,
3511                                     u32 *first_hash)
3512 {
3513         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3514         int ret, credits = 2 * blk_per_bucket;
3515
3516         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3517
3518         ret = ocfs2_extend_trans(handle, credits);
3519         if (ret) {
3520                 mlog_errno(ret);
3521                 return ret;
3522         }
3523
3524         /* Move half of the xattr in start_blk to the next bucket. */
3525         return  ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3526                                         new_blk, first_hash, 1);
3527 }
3528
3529 /*
3530  * Move some xattrs from the old cluster to the new one since they are not
3531  * contiguous in ocfs2 xattr tree.
3532  *
3533  * new_blk starts a new separate cluster, and we will move some xattrs from
3534  * prev_blk to it. v_start will be set as the first name hash value in this
3535  * new cluster so that it can be used as e_cpos during tree insertion and
3536  * don't collide with our original b-tree operations. first_bh and header_bh
3537  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3538  * to extend the insert bucket.
3539  *
3540  * The problem is how much xattr should we move to the new one and when should
3541  * we update first_bh and header_bh?
3542  * 1. If cluster size > bucket size, that means the previous cluster has more
3543  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3544  *    update the first_bh and header_bh if the insert bucket has been moved
3545  *    to the new cluster.
3546  * 2. If cluster_size == bucket_size:
3547  *    a) If the previous extent rec has more than one cluster and the insert
3548  *       place isn't in the last cluster, copy the entire last cluster to the
3549  *       new one. This time, we don't need to upate the first_bh and header_bh
3550  *       since they will not be moved into the new cluster.
3551  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3552  *       the new one. And we set the extend flag to zero if the insert place is
3553  *       moved into the new allocated cluster since no extend is needed.
3554  */
3555 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3556                                             handle_t *handle,
3557                                             struct buffer_head **first_bh,
3558                                             struct buffer_head **header_bh,
3559                                             u64 new_blk,
3560                                             u64 prev_blk,
3561                                             u32 prev_clusters,
3562                                             u32 *v_start,
3563                                             int *extend)
3564 {
3565         int ret = 0;
3566         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3567
3568         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3569              prev_blk, prev_clusters, new_blk);
3570
3571         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3572                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3573                                                           handle,
3574                                                           first_bh,
3575                                                           header_bh,
3576                                                           new_blk,
3577                                                           prev_blk,
3578                                                           prev_clusters,
3579                                                           v_start);
3580         else {
3581                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3582
3583                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3584                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3585                                                      last_blk, new_blk,
3586                                                      v_start);
3587                 else {
3588                         ret = ocfs2_half_xattr_cluster(inode, handle,
3589                                                        last_blk, new_blk,
3590                                                        v_start);
3591
3592                         if ((*header_bh)->b_blocknr == last_blk && extend)
3593                                 *extend = 0;
3594                 }
3595         }
3596
3597         return ret;
3598 }
3599
3600 /*
3601  * Add a new cluster for xattr storage.
3602  *
3603  * If the new cluster is contiguous with the previous one, it will be
3604  * appended to the same extent record, and num_clusters will be updated.
3605  * If not, we will insert a new extent for it and move some xattrs in
3606  * the last cluster into the new allocated one.
3607  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3608  * lose the benefits of hashing because we'll have to search large leaves.
3609  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3610  * if it's bigger).
3611  *
3612  * first_bh is the first block of the previous extent rec and header_bh
3613  * indicates the bucket we will insert the new xattrs. They will be updated
3614  * when the header_bh is moved into the new cluster.
3615  */
3616 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3617                                        struct buffer_head *root_bh,
3618                                        struct buffer_head **first_bh,
3619                                        struct buffer_head **header_bh,
3620                                        u32 *num_clusters,
3621                                        u32 prev_cpos,
3622                                        u64 prev_blkno,
3623                                        int *extend)
3624 {
3625         int ret, credits;
3626         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3627         u32 prev_clusters = *num_clusters;
3628         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3629         u64 block;
3630         handle_t *handle = NULL;
3631         struct ocfs2_alloc_context *data_ac = NULL;
3632         struct ocfs2_alloc_context *meta_ac = NULL;
3633         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3634         struct ocfs2_extent_tree et;
3635
3636         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3637              "previous xattr blkno = %llu\n",
3638              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3639              prev_cpos, prev_blkno);
3640
3641         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3642
3643         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3644                                     &data_ac, &meta_ac);
3645         if (ret) {
3646                 mlog_errno(ret);
3647                 goto leave;
3648         }
3649
3650         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3651                                             clusters_to_add);
3652         handle = ocfs2_start_trans(osb, credits);
3653         if (IS_ERR(handle)) {
3654                 ret = PTR_ERR(handle);
3655                 handle = NULL;
3656                 mlog_errno(ret);
3657                 goto leave;
3658         }
3659
3660         ret = ocfs2_journal_access(handle, inode, root_bh,
3661                                    OCFS2_JOURNAL_ACCESS_WRITE);
3662         if (ret < 0) {
3663                 mlog_errno(ret);
3664                 goto leave;
3665         }
3666
3667         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3668                                      clusters_to_add, &bit_off, &num_bits);
3669         if (ret < 0) {
3670                 if (ret != -ENOSPC)
3671                         mlog_errno(ret);
3672                 goto leave;
3673         }
3674
3675         BUG_ON(num_bits > clusters_to_add);
3676
3677         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3678         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3679              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3680
3681         if (prev_blkno + prev_clusters * bpc == block &&
3682             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3683              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3684                 /*
3685                  * If this cluster is contiguous with the old one and
3686                  * adding this new cluster, we don't surpass the limit of
3687                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3688                  * initialized and used like other buckets in the previous
3689                  * cluster.
3690                  * So add it as a contiguous one. The caller will handle
3691                  * its init process.
3692                  */
3693                 v_start = prev_cpos + prev_clusters;
3694                 *num_clusters = prev_clusters + num_bits;
3695                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3696                      num_bits);
3697         } else {
3698                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3699                                                        handle,
3700                                                        first_bh,
3701                                                        header_bh,
3702                                                        block,
3703                                                        prev_blkno,
3704                                                        prev_clusters,
3705                                                        &v_start,
3706                                                        extend);
3707                 if (ret) {
3708                         mlog_errno(ret);
3709                         goto leave;
3710                 }
3711         }
3712
3713         if (handle->h_buffer_credits < credits) {
3714                 /*
3715                  * The journal has been restarted before, and don't
3716                  * have enough space for the insertion, so extend it
3717                  * here.
3718                  */
3719                 ret = ocfs2_extend_trans(handle, credits);
3720                 if (ret) {
3721                         mlog_errno(ret);
3722                         goto leave;
3723                 }
3724         }
3725         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3726              num_bits, block, v_start);
3727         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3728                                   num_bits, 0, meta_ac);
3729         if (ret < 0) {
3730                 mlog_errno(ret);
3731                 goto leave;
3732         }
3733
3734         ret = ocfs2_journal_dirty(handle, root_bh);
3735         if (ret < 0) {
3736                 mlog_errno(ret);
3737                 goto leave;
3738         }
3739
3740 leave:
3741         if (handle)
3742                 ocfs2_commit_trans(osb, handle);
3743         if (data_ac)
3744                 ocfs2_free_alloc_context(data_ac);
3745         if (meta_ac)
3746                 ocfs2_free_alloc_context(meta_ac);
3747
3748         return ret;
3749 }
3750
3751 /*
3752  * Extend a new xattr bucket and move xattrs to the end one by one until
3753  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3754  */
3755 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3756                                      struct buffer_head *first_bh,
3757                                      struct buffer_head *start_bh,
3758                                      u32 num_clusters)
3759 {
3760         int ret, credits;
3761         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3762         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3763         u64 start_blk = start_bh->b_blocknr, end_blk;
3764         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3765         handle_t *handle;
3766         struct ocfs2_xattr_header *first_xh =
3767                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3768         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3769
3770         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3771              "from %llu, len = %u\n", start_blk,
3772              (unsigned long long)first_bh->b_blocknr, num_clusters);
3773
3774         BUG_ON(bucket >= num_buckets);
3775
3776         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3777
3778         /*
3779          * We will touch all the buckets after the start_bh(include it).
3780          * Add one more bucket and modify the first_bh.
3781          */
3782         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3783         handle = ocfs2_start_trans(osb, credits);
3784         if (IS_ERR(handle)) {
3785                 ret = PTR_ERR(handle);
3786                 handle = NULL;
3787                 mlog_errno(ret);
3788                 goto out;
3789         }
3790
3791         ret = ocfs2_journal_access(handle, inode, first_bh,
3792                                    OCFS2_JOURNAL_ACCESS_WRITE);
3793         if (ret) {
3794                 mlog_errno(ret);
3795                 goto commit;
3796         }
3797
3798         while (end_blk != start_blk) {
3799                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3800                                             end_blk + blk_per_bucket, 0);
3801                 if (ret)
3802                         goto commit;
3803                 end_blk -= blk_per_bucket;
3804         }
3805
3806         /* Move half of the xattr in start_blk to the next bucket. */
3807         ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3808                                       start_blk + blk_per_bucket, NULL, 0);
3809
3810         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3811         ocfs2_journal_dirty(handle, first_bh);
3812
3813 commit:
3814         ocfs2_commit_trans(osb, handle);
3815 out:
3816         return ret;
3817 }
3818
3819 /*
3820  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3821  * xb_bh is the ocfs2_xattr_block.
3822  * We will move all the buckets starting from header_bh to the next place. As
3823  * for this one, half num of its xattrs will be moved to the next one.
3824  *
3825  * We will allocate a new cluster if current cluster is full and adjust
3826  * header_bh and first_bh if the insert place is moved to the new cluster.
3827  */
3828 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3829                                       struct buffer_head *xb_bh,
3830                                       struct buffer_head *header_bh)
3831 {
3832         struct ocfs2_xattr_header *first_xh = NULL;
3833         struct buffer_head *first_bh = NULL;
3834         struct ocfs2_xattr_block *xb =
3835                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3836         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3837         struct ocfs2_extent_list *el = &xb_root->xt_list;
3838         struct ocfs2_xattr_header *xh =
3839                         (struct ocfs2_xattr_header *)header_bh->b_data;
3840         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3841         struct super_block *sb = inode->i_sb;
3842         struct ocfs2_super *osb = OCFS2_SB(sb);
3843         int ret, num_buckets, extend = 1;
3844         u64 p_blkno;
3845         u32 e_cpos, num_clusters;
3846
3847         mlog(0, "Add new xattr bucket starting form %llu\n",
3848              (unsigned long long)header_bh->b_blocknr);
3849
3850         /*
3851          * Add refrence for header_bh here because it may be
3852          * changed in ocfs2_add_new_xattr_cluster and we need
3853          * to free it in the end.
3854          */
3855         get_bh(header_bh);
3856
3857         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3858                                   &num_clusters, el);
3859         if (ret) {
3860                 mlog_errno(ret);
3861                 goto out;
3862         }
3863
3864         ret = ocfs2_read_block(inode, p_blkno, &first_bh);
3865         if (ret) {
3866                 mlog_errno(ret);
3867                 goto out;
3868         }
3869
3870         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3871         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3872
3873         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3874                 ret = ocfs2_add_new_xattr_cluster(inode,
3875                                                   xb_bh,
3876                                                   &first_bh,
3877                                                   &header_bh,
3878                                                   &num_clusters,
3879                                                   e_cpos,
3880                                                   p_blkno,
3881                                                   &extend);
3882                 if (ret) {
3883                         mlog_errno(ret);
3884                         goto out;
3885                 }
3886         }
3887
3888         if (extend)
3889                 ret = ocfs2_extend_xattr_bucket(inode,
3890                                                 first_bh,
3891                                                 header_bh,
3892                                                 num_clusters);
3893         if (ret)
3894                 mlog_errno(ret);
3895 out:
3896         brelse(first_bh);
3897         brelse(header_bh);
3898         return ret;
3899 }
3900
3901 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3902                                         struct ocfs2_xattr_bucket *bucket,
3903                                         int offs)
3904 {
3905         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3906
3907         offs = offs % inode->i_sb->s_blocksize;
3908         return bucket->bhs[block_off]->b_data + offs;
3909 }
3910
3911 /*
3912  * Handle the normal xattr set, including replace, delete and new.
3913  *
3914  * Note: "local" indicates the real data's locality. So we can't
3915  * just its bucket locality by its length.
3916  */
3917 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3918                                          struct ocfs2_xattr_info *xi,
3919                                          struct ocfs2_xattr_search *xs,
3920                                          u32 name_hash,
3921                                          int local)
3922 {
3923         struct ocfs2_xattr_entry *last, *xe;
3924         int name_len = strlen(xi->name);
3925         struct ocfs2_xattr_header *xh = xs->header;
3926         u16 count = le16_to_cpu(xh->xh_count), start;
3927         size_t blocksize = inode->i_sb->s_blocksize;
3928         char *val;
3929         size_t offs, size, new_size;
3930
3931         last = &xh->xh_entries[count];
3932         if (!xs->not_found) {
3933                 xe = xs->here;
3934                 offs = le16_to_cpu(xe->xe_name_offset);
3935                 if (ocfs2_xattr_is_local(xe))
3936                         size = OCFS2_XATTR_SIZE(name_len) +
3937                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3938                 else
3939                         size = OCFS2_XATTR_SIZE(name_len) +
3940                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3941
3942                 /*
3943                  * If the new value will be stored outside, xi->value has been
3944                  * initalized as an empty ocfs2_xattr_value_root, and the same
3945                  * goes with xi->value_len, so we can set new_size safely here.
3946                  * See ocfs2_xattr_set_in_bucket.
3947                  */
3948                 new_size = OCFS2_XATTR_SIZE(name_len) +
3949                            OCFS2_XATTR_SIZE(xi->value_len);
3950
3951                 le16_add_cpu(&xh->xh_name_value_len, -size);
3952                 if (xi->value) {
3953                         if (new_size > size)
3954                                 goto set_new_name_value;
3955
3956                         /* Now replace the old value with new one. */
3957                         if (local)
3958                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3959                         else
3960                                 xe->xe_value_size = 0;
3961
3962                         val = ocfs2_xattr_bucket_get_val(inode,
3963                                                          &xs->bucket, offs);
3964                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3965                                size - OCFS2_XATTR_SIZE(name_len));
3966                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3967                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3968                                        xi->value, xi->value_len);
3969
3970                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3971                         ocfs2_xattr_set_local(xe, local);
3972                         return;
3973                 } else {
3974                         /*
3975                          * Remove the old entry if there is more than one.
3976                          * We don't remove the last entry so that we can
3977                          * use it to indicate the hash value of the empty
3978                          * bucket.
3979                          */
3980                         last -= 1;
3981                         le16_add_cpu(&xh->xh_count, -1);
3982                         if (xh->xh_count) {
3983                                 memmove(xe, xe + 1,
3984                                         (void *)last - (void *)xe);
3985                                 memset(last, 0,
3986                                        sizeof(struct ocfs2_xattr_entry));
3987                         } else
3988                                 xh->xh_free_start =
3989                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3990
3991                         return;
3992                 }
3993         } else {
3994                 /* find a new entry for insert. */
3995                 int low = 0, high = count - 1, tmp;
3996                 struct ocfs2_xattr_entry *tmp_xe;
3997
3998                 while (low <= high && count) {
3999                         tmp = (low + high) / 2;
4000                         tmp_xe = &xh->xh_entries[tmp];
4001
4002                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4003                                 low = tmp + 1;
4004                         else if (name_hash <
4005                                  le32_to_cpu(tmp_xe->xe_name_hash))
4006                                 high = tmp - 1;
4007                         else {
4008                                 low = tmp;
4009                                 break;
4010                         }
4011                 }
4012
4013                 xe = &xh->xh_entries[low];
4014                 if (low != count)
4015                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4016
4017                 le16_add_cpu(&xh->xh_count, 1);
4018                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4019                 xe->xe_name_hash = cpu_to_le32(name_hash);
4020                 xe->xe_name_len = name_len;
4021                 ocfs2_xattr_set_type(xe, xi->name_index);
4022         }
4023
4024 set_new_name_value:
4025         /* Insert the new name+value. */
4026         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4027
4028         /*
4029          * We must make sure that the name/value pair
4030          * exists in the same block.
4031          */
4032         offs = le16_to_cpu(xh->xh_free_start);
4033         start = offs - size;
4034
4035         if (start >> inode->i_sb->s_blocksize_bits !=
4036             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4037                 offs = offs - offs % blocksize;
4038                 xh->xh_free_start = cpu_to_le16(offs);
4039         }
4040
4041         val = ocfs2_xattr_bucket_get_val(inode,
4042                                          &xs->bucket, offs - size);
4043         xe->xe_name_offset = cpu_to_le16(offs - size);
4044
4045         memset(val, 0, size);
4046         memcpy(val, xi->name, name_len);
4047         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4048
4049         xe->xe_value_size = cpu_to_le64(xi->value_len);
4050         ocfs2_xattr_set_local(xe, local);
4051         xs->here = xe;
4052         le16_add_cpu(&xh->xh_free_start, -size);
4053         le16_add_cpu(&xh->xh_name_value_len, size);
4054
4055         return;
4056 }
4057
4058 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4059                                              handle_t *handle,
4060                                              struct ocfs2_xattr_search *xs,
4061                                              struct buffer_head **bhs,
4062                                              u16 bh_num)
4063 {
4064         int ret = 0, off, block_off;
4065         struct ocfs2_xattr_entry *xe = xs->here;
4066
4067         /*
4068          * First calculate all the blocks we should journal_access
4069          * and journal_dirty. The first block should always be touched.
4070          */
4071         ret = ocfs2_journal_dirty(handle, bhs[0]);
4072         if (ret)
4073                 mlog_errno(ret);
4074
4075         /* calc the data. */
4076         off = le16_to_cpu(xe->xe_name_offset);
4077         block_off = off >> inode->i_sb->s_blocksize_bits;
4078         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4079         if (ret)
4080                 mlog_errno(ret);
4081
4082         return ret;
4083 }
4084
4085 /*
4086  * Set the xattr entry in the specified bucket.
4087  * The bucket is indicated by xs->bucket and it should have the enough
4088  * space for the xattr insertion.
4089  */
4090 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4091                                            struct ocfs2_xattr_info *xi,
4092                                            struct ocfs2_xattr_search *xs,
4093                                            u32 name_hash,
4094                                            int local)
4095 {
4096         int i, ret;
4097         handle_t *handle = NULL;
4098         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4099         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4100
4101         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4102              (unsigned long)xi->value_len, xi->name_index,
4103              (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4104
4105         if (!xs->bucket.bhs[1]) {
4106                 ret = ocfs2_read_blocks(inode,
4107                                         xs->bucket.bhs[0]->b_blocknr + 1,
4108                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
4109                                         0);
4110                 if (ret) {
4111                         mlog_errno(ret);
4112                         goto out;
4113                 }
4114         }
4115
4116         handle = ocfs2_start_trans(osb, blk_per_bucket);
4117         if (IS_ERR(handle)) {
4118                 ret = PTR_ERR(handle);
4119                 handle = NULL;
4120                 mlog_errno(ret);
4121                 goto out;
4122         }
4123
4124         for (i = 0; i < blk_per_bucket; i++) {
4125                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4126                                            OCFS2_JOURNAL_ACCESS_WRITE);
4127                 if (ret < 0) {
4128                         mlog_errno(ret);
4129                         goto out;
4130                 }
4131         }
4132
4133         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4134
4135         /*Only dirty the blocks we have touched in set xattr. */
4136         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4137                                                 xs->bucket.bhs, blk_per_bucket);
4138         if (ret)
4139                 mlog_errno(ret);
4140 out:
4141         ocfs2_commit_trans(osb, handle);
4142
4143         return ret;
4144 }
4145
4146 static int ocfs2_xattr_value_update_size(struct inode *inode,
4147                                          struct buffer_head *xe_bh,
4148                                          struct ocfs2_xattr_entry *xe,
4149                                          u64 new_size)
4150 {
4151         int ret;
4152         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4153         handle_t *handle = NULL;
4154
4155         handle = ocfs2_start_trans(osb, 1);
4156         if (handle == NULL) {
4157                 ret = -ENOMEM;
4158                 mlog_errno(ret);
4159                 goto out;
4160         }
4161
4162         ret = ocfs2_journal_access(handle, inode, xe_bh,
4163                                    OCFS2_JOURNAL_ACCESS_WRITE);
4164         if (ret < 0) {
4165                 mlog_errno(ret);
4166                 goto out_commit;
4167         }
4168
4169         xe->xe_value_size = cpu_to_le64(new_size);
4170
4171         ret = ocfs2_journal_dirty(handle, xe_bh);
4172         if (ret < 0)
4173                 mlog_errno(ret);
4174
4175 out_commit:
4176         ocfs2_commit_trans(osb, handle);
4177 out:
4178         return ret;
4179 }
4180
4181 /*
4182  * Truncate the specified xe_off entry in xattr bucket.
4183  * bucket is indicated by header_bh and len is the new length.
4184  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4185  *
4186  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4187  */
4188 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4189                                              struct buffer_head *header_bh,
4190                                              int xe_off,
4191                                              int len)
4192 {
4193         int ret, offset;
4194         u64 value_blk;
4195         struct buffer_head *value_bh = NULL;
4196         struct ocfs2_xattr_value_root *xv;
4197         struct ocfs2_xattr_entry *xe;
4198         struct ocfs2_xattr_header *xh =
4199                         (struct ocfs2_xattr_header *)header_bh->b_data;
4200         size_t blocksize = inode->i_sb->s_blocksize;
4201
4202         xe = &xh->xh_entries[xe_off];
4203
4204         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4205
4206         offset = le16_to_cpu(xe->xe_name_offset) +
4207                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4208
4209         value_blk = offset / blocksize;
4210
4211         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4212         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4213         value_blk += header_bh->b_blocknr;
4214
4215         ret = ocfs2_read_block(inode, value_blk, &value_bh);
4216         if (ret) {
4217                 mlog_errno(ret);
4218                 goto out;
4219         }
4220
4221         xv = (struct ocfs2_xattr_value_root *)
4222                 (value_bh->b_data + offset % blocksize);
4223
4224         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4225              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4226         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4227         if (ret) {
4228                 mlog_errno(ret);
4229                 goto out;
4230         }
4231
4232         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4233         if (ret) {
4234                 mlog_errno(ret);
4235                 goto out;
4236         }
4237
4238 out:
4239         brelse(value_bh);
4240         return ret;
4241 }
4242
4243 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4244                                                 struct ocfs2_xattr_search *xs,
4245                                                 int len)
4246 {
4247         int ret, offset;
4248         struct ocfs2_xattr_entry *xe = xs->here;
4249         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4250
4251         BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4252
4253         offset = xe - xh->xh_entries;
4254         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4255                                                 offset, len);
4256         if (ret)
4257                 mlog_errno(ret);
4258
4259         return ret;
4260 }
4261
4262 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4263                                                 struct ocfs2_xattr_search *xs,
4264                                                 char *val,
4265                                                 int value_len)
4266 {
4267         int offset;
4268         struct ocfs2_xattr_value_root *xv;
4269         struct ocfs2_xattr_entry *xe = xs->here;
4270
4271         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4272
4273         offset = le16_to_cpu(xe->xe_name_offset) +
4274                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4275
4276         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4277
4278         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4279 }
4280
4281 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4282                                   struct buffer_head *root_bh,
4283                                   u64 blkno,
4284                                   u32 cpos,
4285                                   u32 len)
4286 {
4287         int ret;
4288         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4289         struct inode *tl_inode = osb->osb_tl_inode;
4290         handle_t *handle;
4291         struct ocfs2_xattr_block *xb =
4292                         (struct ocfs2_xattr_block *)root_bh->b_data;
4293         struct ocfs2_alloc_context *meta_ac = NULL;
4294         struct ocfs2_cached_dealloc_ctxt dealloc;
4295         struct ocfs2_extent_tree et;
4296
4297         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4298
4299         ocfs2_init_dealloc_ctxt(&dealloc);
4300
4301         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4302              cpos, len, (unsigned long long)blkno);
4303
4304         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4305
4306         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4307         if (ret) {
4308                 mlog_errno(ret);
4309                 return ret;
4310         }
4311
4312         mutex_lock(&tl_inode->i_mutex);
4313
4314         if (ocfs2_truncate_log_needs_flush(osb)) {
4315                 ret = __ocfs2_flush_truncate_log(osb);
4316                 if (ret < 0) {
4317                         mlog_errno(ret);
4318                         goto out;
4319                 }
4320         }
4321
4322         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4323         if (handle == NULL) {
4324                 ret = -ENOMEM;
4325                 mlog_errno(ret);
4326                 goto out;
4327         }
4328
4329         ret = ocfs2_journal_access(handle, inode, root_bh,
4330                                    OCFS2_JOURNAL_ACCESS_WRITE);
4331         if (ret) {
4332                 mlog_errno(ret);
4333                 goto out_commit;
4334         }
4335
4336         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4337                                   &dealloc);
4338         if (ret) {
4339                 mlog_errno(ret);
4340                 goto out_commit;
4341         }
4342
4343         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4344
4345         ret = ocfs2_journal_dirty(handle, root_bh);
4346         if (ret) {
4347                 mlog_errno(ret);
4348                 goto out_commit;
4349         }
4350
4351         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4352         if (ret)
4353                 mlog_errno(ret);
4354
4355 out_commit:
4356         ocfs2_commit_trans(osb, handle);
4357 out:
4358         ocfs2_schedule_truncate_log_flush(osb, 1);
4359
4360         mutex_unlock(&tl_inode->i_mutex);
4361
4362         if (meta_ac)
4363                 ocfs2_free_alloc_context(meta_ac);
4364
4365         ocfs2_run_deallocs(osb, &dealloc);
4366
4367         return ret;
4368 }
4369
4370 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4371                                          struct ocfs2_xattr_search *xs)
4372 {
4373         handle_t *handle = NULL;
4374         struct ocfs2_xattr_header *xh = xs->bucket.xh;
4375         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4376                                                 le16_to_cpu(xh->xh_count) - 1];
4377         int ret = 0;
4378
4379         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4380         if (IS_ERR(handle)) {
4381                 ret = PTR_ERR(handle);
4382                 mlog_errno(ret);
4383                 return;
4384         }
4385
4386         ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4387                                    OCFS2_JOURNAL_ACCESS_WRITE);
4388         if (ret) {
4389                 mlog_errno(ret);
4390                 goto out_commit;
4391         }
4392
4393         /* Remove the old entry. */
4394         memmove(xs->here, xs->here + 1,
4395                 (void *)last - (void *)xs->here);
4396         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4397         le16_add_cpu(&xh->xh_count, -1);
4398
4399         ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4400         if (ret < 0)
4401                 mlog_errno(ret);
4402 out_commit:
4403         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4404 }
4405
4406 /*
4407  * Set the xattr name/value in the bucket specified in xs.
4408  *
4409  * As the new value in xi may be stored in the bucket or in an outside cluster,
4410  * we divide the whole process into 3 steps:
4411  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4412  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4413  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4414  * 4. If the clusters for the new outside value can't be allocated, we need
4415  *    to free the xattr we allocated in set.
4416  */
4417 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4418                                      struct ocfs2_xattr_info *xi,
4419                                      struct ocfs2_xattr_search *xs)
4420 {
4421         int ret, local = 1;
4422         size_t value_len;
4423         char *val = (char *)xi->value;
4424         struct ocfs2_xattr_entry *xe = xs->here;
4425         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4426                                               strlen(xi->name));
4427
4428         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4429                 /*
4430                  * We need to truncate the xattr storage first.
4431                  *
4432                  * If both the old and new value are stored to
4433                  * outside block, we only need to truncate
4434                  * the storage and then set the value outside.
4435                  *
4436                  * If the new value should be stored within block,
4437                  * we should free all the outside block first and
4438                  * the modification to the xattr block will be done
4439                  * by following steps.
4440                  */
4441                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4442                         value_len = xi->value_len;
4443                 else
4444                         value_len = 0;
4445
4446                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4447                                                            value_len);
4448                 if (ret)
4449                         goto out;
4450
4451                 if (value_len)
4452                         goto set_value_outside;
4453         }
4454
4455         value_len = xi->value_len;
4456         /* So we have to handle the inside block change now. */
4457         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4458                 /*
4459                  * If the new value will be stored outside of block,
4460                  * initalize a new empty value root and insert it first.
4461                  */
4462                 local = 0;
4463                 xi->value = &def_xv;
4464                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4465         }
4466
4467         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4468         if (ret) {
4469                 mlog_errno(ret);
4470                 goto out;
4471         }
4472
4473         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4474                 goto out;
4475
4476         /* allocate the space now for the outside block storage. */
4477         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4478                                                    value_len);
4479         if (ret) {
4480                 mlog_errno(ret);
4481
4482                 if (xs->not_found) {
4483                         /*
4484                          * We can't allocate enough clusters for outside
4485                          * storage and we have allocated xattr already,
4486                          * so need to remove it.
4487                          */
4488                         ocfs2_xattr_bucket_remove_xs(inode, xs);
4489                 }
4490                 goto out;
4491         }
4492
4493 set_value_outside:
4494         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4495 out:
4496         return ret;
4497 }
4498
4499 /* check whether the xattr bucket is filled up with the same hash value. */
4500 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4501                                               struct ocfs2_xattr_bucket *bucket)
4502 {
4503         struct ocfs2_xattr_header *xh = bucket->xh;
4504
4505         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4506             xh->xh_entries[0].xe_name_hash) {
4507                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4508                      "hash = %u\n",
4509                      (unsigned long long)bucket->bhs[0]->b_blocknr,
4510                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4511                 return -ENOSPC;
4512         }
4513
4514         return 0;
4515 }
4516
4517 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4518                                              struct ocfs2_xattr_info *xi,
4519                                              struct ocfs2_xattr_search *xs)
4520 {
4521         struct ocfs2_xattr_header *xh;
4522         struct ocfs2_xattr_entry *xe;
4523         u16 count, header_size, xh_free_start;
4524         int i, free, max_free, need, old;
4525         size_t value_size = 0, name_len = strlen(xi->name);
4526         size_t blocksize = inode->i_sb->s_blocksize;
4527         int ret, allocation = 0;
4528         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4529
4530         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4531
4532 try_again:
4533         xh = xs->header;
4534         count = le16_to_cpu(xh->xh_count);
4535         xh_free_start = le16_to_cpu(xh->xh_free_start);
4536         header_size = sizeof(struct ocfs2_xattr_header) +
4537                         count * sizeof(struct ocfs2_xattr_entry);
4538         max_free = OCFS2_XATTR_BUCKET_SIZE -
4539                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4540
4541         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4542                         "of %u which exceed block size\n",
4543                         (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4544                         header_size);
4545
4546         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4547                 value_size = OCFS2_XATTR_ROOT_SIZE;
4548         else if (xi->value)
4549                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4550
4551         if (xs->not_found)
4552                 need = sizeof(struct ocfs2_xattr_entry) +
4553                         OCFS2_XATTR_SIZE(name_len) + value_size;
4554         else {
4555                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4556
4557                 /*
4558                  * We only replace the old value if the new length is smaller
4559                  * than the old one. Otherwise we will allocate new space in the
4560                  * bucket to store it.
4561                  */
4562                 xe = xs->here;
4563                 if (ocfs2_xattr_is_local(xe))
4564                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4565                 else
4566                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4567
4568                 if (old >= value_size)
4569                         need = 0;
4570         }
4571
4572         free = xh_free_start - header_size;
4573         /*
4574          * We need to make sure the new name/value pair
4575          * can exist in the same block.
4576          */
4577         if (xh_free_start % blocksize < need)
4578                 free -= xh_free_start % blocksize;
4579
4580         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4581              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4582              " %u\n", xs->not_found,
4583              (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4584              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4585              le16_to_cpu(xh->xh_name_value_len));
4586
4587         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4588                 if (need <= max_free &&
4589                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4590                         /*
4591                          * We can create the space by defragment. Since only the
4592                          * name/value will be moved, the xe shouldn't be changed
4593                          * in xs.
4594                          */
4595                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4596                         if (ret) {
4597                                 mlog_errno(ret);
4598                                 goto out;
4599                         }
4600
4601                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4602                         free = xh_free_start - header_size;
4603                         if (xh_free_start % blocksize < need)
4604                                 free -= xh_free_start % blocksize;
4605
4606                         if (free >= need)
4607                                 goto xattr_set;
4608
4609                         mlog(0, "Can't get enough space for xattr insert by "
4610                              "defragment. Need %u bytes, but we have %d, so "
4611                              "allocate new bucket for it.\n", need, free);
4612                 }
4613
4614                 /*
4615                  * We have to add new buckets or clusters and one
4616                  * allocation should leave us enough space for insert.
4617                  */
4618                 BUG_ON(allocation);
4619
4620                 /*
4621                  * We do not allow for overlapping ranges between buckets. And
4622                  * the maximum number of collisions we will allow for then is
4623                  * one bucket's worth, so check it here whether we need to
4624                  * add a new bucket for the insert.
4625                  */
4626                 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4627                 if (ret) {
4628                         mlog_errno(ret);
4629                         goto out;
4630                 }
4631
4632                 ret = ocfs2_add_new_xattr_bucket(inode,
4633                                                  xs->xattr_bh,
4634                                                  xs->bucket.bhs[0]);
4635                 if (ret) {
4636                         mlog_errno(ret);
4637                         goto out;
4638                 }
4639
4640                 for (i = 0; i < blk_per_bucket; i++)
4641                         brelse(xs->bucket.bhs[i]);
4642
4643                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4644
4645                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4646                                                    xi->name_index,
4647                                                    xi->name, xs);
4648                 if (ret && ret != -ENODATA)
4649                         goto out;
4650                 xs->not_found = ret;
4651                 allocation = 1;
4652                 goto try_again;
4653         }
4654
4655 xattr_set:
4656         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4657 out:
4658         mlog_exit(ret);
4659         return ret;
4660 }
4661
4662 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4663                                         struct ocfs2_xattr_bucket *bucket,
4664                                         void *para)
4665 {
4666         int ret = 0;
4667         struct ocfs2_xattr_header *xh = bucket->xh;
4668         u16 i;
4669         struct ocfs2_xattr_entry *xe;
4670
4671         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4672                 xe = &xh->xh_entries[i];
4673                 if (ocfs2_xattr_is_local(xe))
4674                         continue;
4675
4676                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4677                                                         bucket->bhs[0],
4678                                                         i, 0);
4679                 if (ret) {
4680                         mlog_errno(ret);
4681                         break;
4682                 }
4683         }
4684
4685         return ret;
4686 }
4687
4688 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4689                                           struct buffer_head *xb_bh)
4690 {
4691         struct ocfs2_xattr_block *xb =
4692                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4693         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4694         int ret = 0;
4695         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4696         u64 p_blkno;
4697
4698         if (le16_to_cpu(el->l_next_free_rec) == 0)
4699                 return 0;
4700
4701         while (name_hash > 0) {
4702                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4703                                           &e_cpos, &num_clusters, el);
4704                 if (ret) {
4705                         mlog_errno(ret);
4706                         goto out;
4707                 }
4708
4709                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4710                                                   ocfs2_delete_xattr_in_bucket,
4711                                                   NULL);
4712                 if (ret) {
4713                         mlog_errno(ret);
4714                         goto out;
4715                 }
4716
4717                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4718                                              p_blkno, e_cpos, num_clusters);
4719                 if (ret) {
4720                         mlog_errno(ret);
4721                         break;
4722                 }
4723
4724                 if (e_cpos == 0)
4725                         break;
4726
4727                 name_hash = e_cpos - 1;
4728         }
4729
4730 out:
4731         return ret;
4732 }
4733
4734 /*
4735  * 'trusted' attributes support
4736  */
4737 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4738                                        size_t list_size, const char *name,
4739                                        size_t name_len)
4740 {
4741         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
4742         const size_t total_len = prefix_len + name_len + 1;
4743
4744         if (list && total_len <= list_size) {
4745                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4746                 memcpy(list + prefix_len, name, name_len);
4747                 list[prefix_len + name_len] = '\0';
4748         }
4749         return total_len;
4750 }
4751
4752 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4753                                    void *buffer, size_t size)
4754 {
4755         if (strcmp(name, "") == 0)
4756                 return -EINVAL;
4757         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4758                                buffer, size);
4759 }
4760
4761 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4762                                    const void *value, size_t size, int flags)
4763 {
4764         if (strcmp(name, "") == 0)
4765                 return -EINVAL;
4766
4767         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4768                                size, flags);
4769 }
4770
4771 struct xattr_handler ocfs2_xattr_trusted_handler = {
4772         .prefix = XATTR_TRUSTED_PREFIX,
4773         .list   = ocfs2_xattr_trusted_list,
4774         .get    = ocfs2_xattr_trusted_get,
4775         .set    = ocfs2_xattr_trusted_set,
4776 };
4777
4778 /*
4779  * 'user' attributes support
4780  */
4781 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4782                                     size_t list_size, const char *name,
4783                                     size_t name_len)
4784 {
4785         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
4786         const size_t total_len = prefix_len + name_len + 1;
4787         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4788
4789         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4790                 return 0;
4791
4792         if (list && total_len <= list_size) {
4793                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4794                 memcpy(list + prefix_len, name, name_len);
4795                 list[prefix_len + name_len] = '\0';
4796         }
4797         return total_len;
4798 }
4799
4800 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4801                                 void *buffer, size_t size)
4802 {
4803         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4804
4805         if (strcmp(name, "") == 0)
4806                 return -EINVAL;
4807         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4808                 return -EOPNOTSUPP;
4809         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4810                                buffer, size);
4811 }
4812
4813 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4814                                 const void *value, size_t size, int flags)
4815 {
4816         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4817
4818         if (strcmp(name, "") == 0)
4819                 return -EINVAL;
4820         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4821                 return -EOPNOTSUPP;
4822
4823         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4824                                size, flags);
4825 }
4826
4827 struct xattr_handler ocfs2_xattr_user_handler = {
4828         .prefix = XATTR_USER_PREFIX,
4829         .list   = ocfs2_xattr_user_list,
4830         .get    = ocfs2_xattr_user_get,
4831         .set    = ocfs2_xattr_user_set,
4832 };