ocfs2: Copy xattr buckets with a dedicated function.
[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 *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
65 };
66
67 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
68 #define OCFS2_XATTR_INLINE_SIZE 80
69
70 static struct ocfs2_xattr_def_value_root def_xv = {
71         .xv.xr_list.l_count = cpu_to_le16(1),
72 };
73
74 struct xattr_handler *ocfs2_xattr_handlers[] = {
75         &ocfs2_xattr_user_handler,
76         &ocfs2_xattr_trusted_handler,
77         NULL
78 };
79
80 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
81         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
82         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
83 };
84
85 struct ocfs2_xattr_info {
86         int name_index;
87         const char *name;
88         const void *value;
89         size_t value_len;
90 };
91
92 struct ocfs2_xattr_search {
93         struct buffer_head *inode_bh;
94         /*
95          * xattr_bh point to the block buffer head which has extended attribute
96          * when extended attribute in inode, xattr_bh is equal to inode_bh.
97          */
98         struct buffer_head *xattr_bh;
99         struct ocfs2_xattr_header *header;
100         struct ocfs2_xattr_bucket bucket;
101         void *base;
102         void *end;
103         struct ocfs2_xattr_entry *here;
104         int not_found;
105 };
106
107 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
108                                              struct ocfs2_xattr_header *xh,
109                                              int index,
110                                              int *block_off,
111                                              int *new_offset);
112
113 static int ocfs2_xattr_block_find(struct inode *inode,
114                                   int name_index,
115                                   const char *name,
116                                   struct ocfs2_xattr_search *xs);
117 static int ocfs2_xattr_index_block_find(struct inode *inode,
118                                         struct buffer_head *root_bh,
119                                         int name_index,
120                                         const char *name,
121                                         struct ocfs2_xattr_search *xs);
122
123 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
124                                         struct ocfs2_xattr_tree_root *xt,
125                                         char *buffer,
126                                         size_t buffer_size);
127
128 static int ocfs2_xattr_create_index_block(struct inode *inode,
129                                           struct ocfs2_xattr_search *xs);
130
131 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
132                                              struct ocfs2_xattr_info *xi,
133                                              struct ocfs2_xattr_search *xs);
134
135 static int ocfs2_delete_xattr_index_block(struct inode *inode,
136                                           struct buffer_head *xb_bh);
137
138 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
139 {
140         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
141 }
142
143 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
144 {
145         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
146 }
147
148 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
149 {
150         u16 len = sb->s_blocksize -
151                  offsetof(struct ocfs2_xattr_header, xh_entries);
152
153         return len / sizeof(struct ocfs2_xattr_entry);
154 }
155
156 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
157 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
158 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
159
160 static void ocfs2_xattr_bucket_relse(struct inode *inode,
161                                      struct ocfs2_xattr_bucket *bucket)
162 {
163         int i, blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
164
165         for (i = 0; i < blks; i++) {
166                 brelse(bucket->bu_bhs[i]);
167                 bucket->bu_bhs[i] = NULL;
168         }
169 }
170
171 /*
172  * A bucket that has never been written to disk doesn't need to be
173  * read.  We just need the buffer_heads.  Don't call this for
174  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
175  * them fully.
176  */
177 static int ocfs2_init_xattr_bucket(struct inode *inode,
178                                    struct ocfs2_xattr_bucket *bucket,
179                                    u64 xb_blkno)
180 {
181         int i, rc = 0;
182         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
183
184         for (i = 0; i < blks; i++) {
185                 bucket->bu_bhs[i] = sb_getblk(inode->i_sb, xb_blkno + i);
186                 if (!bucket->bu_bhs[i]) {
187                         rc = -EIO;
188                         mlog_errno(rc);
189                         break;
190                 }
191
192                 ocfs2_set_new_buffer_uptodate(inode, bucket->bu_bhs[i]);
193         }
194
195         if (rc)
196                 ocfs2_xattr_bucket_relse(inode, bucket);
197         return rc;
198 }
199
200 /* Read the xattr bucket at xb_blkno */
201 static int ocfs2_read_xattr_bucket(struct inode *inode,
202                                    struct ocfs2_xattr_bucket *bucket,
203                                    u64 xb_blkno)
204 {
205         int rc, blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
206
207         rc = ocfs2_read_blocks(inode, xb_blkno, blks, bucket->bu_bhs, 0);
208         if (rc)
209                 ocfs2_xattr_bucket_relse(inode, bucket);
210         return rc;
211 }
212
213 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
214                                              struct inode *inode,
215                                              struct ocfs2_xattr_bucket *bucket,
216                                              int type)
217 {
218         int i, rc = 0;
219         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
220
221         for (i = 0; i < blks; i++) {
222                 rc = ocfs2_journal_access(handle, inode,
223                                           bucket->bu_bhs[i], type);
224                 if (rc) {
225                         mlog_errno(rc);
226                         break;
227                 }
228         }
229
230         return rc;
231 }
232
233 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
234                                              struct inode *inode,
235                                              struct ocfs2_xattr_bucket *bucket)
236 {
237         int i, blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
238
239         for (i = 0; i < blks; i++)
240                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
241 }
242
243 static void ocfs2_xattr_bucket_copy_data(struct inode *inode,
244                                          struct ocfs2_xattr_bucket *dest,
245                                          struct ocfs2_xattr_bucket *src)
246 {
247         int i;
248         int blocksize = inode->i_sb->s_blocksize;
249         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
250
251         for (i = 0; i < blks; i++) {
252                 memcpy(bucket_block(dest, i), bucket_block(src, i),
253                        blocksize);
254         }
255 }
256
257 static inline const char *ocfs2_xattr_prefix(int name_index)
258 {
259         struct xattr_handler *handler = NULL;
260
261         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
262                 handler = ocfs2_xattr_handler_map[name_index];
263
264         return handler ? handler->prefix : NULL;
265 }
266
267 static u32 ocfs2_xattr_name_hash(struct inode *inode,
268                                  const char *name,
269                                  int name_len)
270 {
271         /* Get hash value of uuid from super block */
272         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
273         int i;
274
275         /* hash extended attribute name */
276         for (i = 0; i < name_len; i++) {
277                 hash = (hash << OCFS2_HASH_SHIFT) ^
278                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
279                        *name++;
280         }
281
282         return hash;
283 }
284
285 /*
286  * ocfs2_xattr_hash_entry()
287  *
288  * Compute the hash of an extended attribute.
289  */
290 static void ocfs2_xattr_hash_entry(struct inode *inode,
291                                    struct ocfs2_xattr_header *header,
292                                    struct ocfs2_xattr_entry *entry)
293 {
294         u32 hash = 0;
295         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
296
297         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
298         entry->xe_name_hash = cpu_to_le32(hash);
299
300         return;
301 }
302
303 static int ocfs2_xattr_extend_allocation(struct inode *inode,
304                                          u32 clusters_to_add,
305                                          struct buffer_head *xattr_bh,
306                                          struct ocfs2_xattr_value_root *xv)
307 {
308         int status = 0;
309         int restart_func = 0;
310         int credits = 0;
311         handle_t *handle = NULL;
312         struct ocfs2_alloc_context *data_ac = NULL;
313         struct ocfs2_alloc_context *meta_ac = NULL;
314         enum ocfs2_alloc_restarted why;
315         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
316         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
317         struct ocfs2_extent_tree et;
318
319         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
320
321         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
322
323 restart_all:
324
325         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
326                                        &data_ac, &meta_ac);
327         if (status) {
328                 mlog_errno(status);
329                 goto leave;
330         }
331
332         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
333                                             clusters_to_add);
334         handle = ocfs2_start_trans(osb, credits);
335         if (IS_ERR(handle)) {
336                 status = PTR_ERR(handle);
337                 handle = NULL;
338                 mlog_errno(status);
339                 goto leave;
340         }
341
342 restarted_transaction:
343         status = ocfs2_journal_access(handle, inode, xattr_bh,
344                                       OCFS2_JOURNAL_ACCESS_WRITE);
345         if (status < 0) {
346                 mlog_errno(status);
347                 goto leave;
348         }
349
350         prev_clusters = le32_to_cpu(xv->xr_clusters);
351         status = ocfs2_add_clusters_in_btree(osb,
352                                              inode,
353                                              &logical_start,
354                                              clusters_to_add,
355                                              0,
356                                              &et,
357                                              handle,
358                                              data_ac,
359                                              meta_ac,
360                                              &why);
361         if ((status < 0) && (status != -EAGAIN)) {
362                 if (status != -ENOSPC)
363                         mlog_errno(status);
364                 goto leave;
365         }
366
367         status = ocfs2_journal_dirty(handle, xattr_bh);
368         if (status < 0) {
369                 mlog_errno(status);
370                 goto leave;
371         }
372
373         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
374
375         if (why != RESTART_NONE && clusters_to_add) {
376                 if (why == RESTART_META) {
377                         mlog(0, "restarting function.\n");
378                         restart_func = 1;
379                 } else {
380                         BUG_ON(why != RESTART_TRANS);
381
382                         mlog(0, "restarting transaction.\n");
383                         /* TODO: This can be more intelligent. */
384                         credits = ocfs2_calc_extend_credits(osb->sb,
385                                                             et.et_root_el,
386                                                             clusters_to_add);
387                         status = ocfs2_extend_trans(handle, credits);
388                         if (status < 0) {
389                                 /* handle still has to be committed at
390                                  * this point. */
391                                 status = -ENOMEM;
392                                 mlog_errno(status);
393                                 goto leave;
394                         }
395                         goto restarted_transaction;
396                 }
397         }
398
399 leave:
400         if (handle) {
401                 ocfs2_commit_trans(osb, handle);
402                 handle = NULL;
403         }
404         if (data_ac) {
405                 ocfs2_free_alloc_context(data_ac);
406                 data_ac = NULL;
407         }
408         if (meta_ac) {
409                 ocfs2_free_alloc_context(meta_ac);
410                 meta_ac = NULL;
411         }
412         if ((!status) && restart_func) {
413                 restart_func = 0;
414                 goto restart_all;
415         }
416
417         return status;
418 }
419
420 static int __ocfs2_remove_xattr_range(struct inode *inode,
421                                       struct buffer_head *root_bh,
422                                       struct ocfs2_xattr_value_root *xv,
423                                       u32 cpos, u32 phys_cpos, u32 len,
424                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
425 {
426         int ret;
427         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
428         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
429         struct inode *tl_inode = osb->osb_tl_inode;
430         handle_t *handle;
431         struct ocfs2_alloc_context *meta_ac = NULL;
432         struct ocfs2_extent_tree et;
433
434         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
435
436         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
437         if (ret) {
438                 mlog_errno(ret);
439                 return ret;
440         }
441
442         mutex_lock(&tl_inode->i_mutex);
443
444         if (ocfs2_truncate_log_needs_flush(osb)) {
445                 ret = __ocfs2_flush_truncate_log(osb);
446                 if (ret < 0) {
447                         mlog_errno(ret);
448                         goto out;
449                 }
450         }
451
452         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
453         if (IS_ERR(handle)) {
454                 ret = PTR_ERR(handle);
455                 mlog_errno(ret);
456                 goto out;
457         }
458
459         ret = ocfs2_journal_access(handle, inode, root_bh,
460                                    OCFS2_JOURNAL_ACCESS_WRITE);
461         if (ret) {
462                 mlog_errno(ret);
463                 goto out_commit;
464         }
465
466         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
467                                   dealloc);
468         if (ret) {
469                 mlog_errno(ret);
470                 goto out_commit;
471         }
472
473         le32_add_cpu(&xv->xr_clusters, -len);
474
475         ret = ocfs2_journal_dirty(handle, root_bh);
476         if (ret) {
477                 mlog_errno(ret);
478                 goto out_commit;
479         }
480
481         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
482         if (ret)
483                 mlog_errno(ret);
484
485 out_commit:
486         ocfs2_commit_trans(osb, handle);
487 out:
488         mutex_unlock(&tl_inode->i_mutex);
489
490         if (meta_ac)
491                 ocfs2_free_alloc_context(meta_ac);
492
493         return ret;
494 }
495
496 static int ocfs2_xattr_shrink_size(struct inode *inode,
497                                    u32 old_clusters,
498                                    u32 new_clusters,
499                                    struct buffer_head *root_bh,
500                                    struct ocfs2_xattr_value_root *xv)
501 {
502         int ret = 0;
503         u32 trunc_len, cpos, phys_cpos, alloc_size;
504         u64 block;
505         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
506         struct ocfs2_cached_dealloc_ctxt dealloc;
507
508         ocfs2_init_dealloc_ctxt(&dealloc);
509
510         if (old_clusters <= new_clusters)
511                 return 0;
512
513         cpos = new_clusters;
514         trunc_len = old_clusters - new_clusters;
515         while (trunc_len) {
516                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
517                                                &alloc_size, &xv->xr_list);
518                 if (ret) {
519                         mlog_errno(ret);
520                         goto out;
521                 }
522
523                 if (alloc_size > trunc_len)
524                         alloc_size = trunc_len;
525
526                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
527                                                  phys_cpos, alloc_size,
528                                                  &dealloc);
529                 if (ret) {
530                         mlog_errno(ret);
531                         goto out;
532                 }
533
534                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
535                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
536                                                        alloc_size);
537                 cpos += alloc_size;
538                 trunc_len -= alloc_size;
539         }
540
541 out:
542         ocfs2_schedule_truncate_log_flush(osb, 1);
543         ocfs2_run_deallocs(osb, &dealloc);
544
545         return ret;
546 }
547
548 static int ocfs2_xattr_value_truncate(struct inode *inode,
549                                       struct buffer_head *root_bh,
550                                       struct ocfs2_xattr_value_root *xv,
551                                       int len)
552 {
553         int ret;
554         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
555         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
556
557         if (new_clusters == old_clusters)
558                 return 0;
559
560         if (new_clusters > old_clusters)
561                 ret = ocfs2_xattr_extend_allocation(inode,
562                                                     new_clusters - old_clusters,
563                                                     root_bh, xv);
564         else
565                 ret = ocfs2_xattr_shrink_size(inode,
566                                               old_clusters, new_clusters,
567                                               root_bh, xv);
568
569         return ret;
570 }
571
572 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
573                                   size_t *result, const char *prefix,
574                                   const char *name, int name_len)
575 {
576         char *p = buffer + *result;
577         int prefix_len = strlen(prefix);
578         int total_len = prefix_len + name_len + 1;
579
580         *result += total_len;
581
582         /* we are just looking for how big our buffer needs to be */
583         if (!size)
584                 return 0;
585
586         if (*result > size)
587                 return -ERANGE;
588
589         memcpy(p, prefix, prefix_len);
590         memcpy(p + prefix_len, name, name_len);
591         p[prefix_len + name_len] = '\0';
592
593         return 0;
594 }
595
596 static int ocfs2_xattr_list_entries(struct inode *inode,
597                                     struct ocfs2_xattr_header *header,
598                                     char *buffer, size_t buffer_size)
599 {
600         size_t result = 0;
601         int i, type, ret;
602         const char *prefix, *name;
603
604         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
605                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
606                 type = ocfs2_xattr_get_type(entry);
607                 prefix = ocfs2_xattr_prefix(type);
608
609                 if (prefix) {
610                         name = (const char *)header +
611                                 le16_to_cpu(entry->xe_name_offset);
612
613                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
614                                                      &result, prefix, name,
615                                                      entry->xe_name_len);
616                         if (ret)
617                                 return ret;
618                 }
619         }
620
621         return result;
622 }
623
624 static int ocfs2_xattr_ibody_list(struct inode *inode,
625                                   struct ocfs2_dinode *di,
626                                   char *buffer,
627                                   size_t buffer_size)
628 {
629         struct ocfs2_xattr_header *header = NULL;
630         struct ocfs2_inode_info *oi = OCFS2_I(inode);
631         int ret = 0;
632
633         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
634                 return ret;
635
636         header = (struct ocfs2_xattr_header *)
637                  ((void *)di + inode->i_sb->s_blocksize -
638                  le16_to_cpu(di->i_xattr_inline_size));
639
640         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
641
642         return ret;
643 }
644
645 static int ocfs2_xattr_block_list(struct inode *inode,
646                                   struct ocfs2_dinode *di,
647                                   char *buffer,
648                                   size_t buffer_size)
649 {
650         struct buffer_head *blk_bh = NULL;
651         struct ocfs2_xattr_block *xb;
652         int ret = 0;
653
654         if (!di->i_xattr_loc)
655                 return ret;
656
657         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
658         if (ret < 0) {
659                 mlog_errno(ret);
660                 return ret;
661         }
662
663         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
664         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
665                 ret = -EIO;
666                 goto cleanup;
667         }
668
669         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
670                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
671                 ret = ocfs2_xattr_list_entries(inode, header,
672                                                buffer, buffer_size);
673         } else {
674                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
675                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
676                                                    buffer, buffer_size);
677         }
678 cleanup:
679         brelse(blk_bh);
680
681         return ret;
682 }
683
684 ssize_t ocfs2_listxattr(struct dentry *dentry,
685                         char *buffer,
686                         size_t size)
687 {
688         int ret = 0, i_ret = 0, b_ret = 0;
689         struct buffer_head *di_bh = NULL;
690         struct ocfs2_dinode *di = NULL;
691         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
692
693         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
694                 return -EOPNOTSUPP;
695
696         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
697                 return ret;
698
699         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
700         if (ret < 0) {
701                 mlog_errno(ret);
702                 return ret;
703         }
704
705         di = (struct ocfs2_dinode *)di_bh->b_data;
706
707         down_read(&oi->ip_xattr_sem);
708         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
709         if (i_ret < 0)
710                 b_ret = 0;
711         else {
712                 if (buffer) {
713                         buffer += i_ret;
714                         size -= i_ret;
715                 }
716                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
717                                                buffer, size);
718                 if (b_ret < 0)
719                         i_ret = 0;
720         }
721         up_read(&oi->ip_xattr_sem);
722         ocfs2_inode_unlock(dentry->d_inode, 0);
723
724         brelse(di_bh);
725
726         return i_ret + b_ret;
727 }
728
729 static int ocfs2_xattr_find_entry(int name_index,
730                                   const char *name,
731                                   struct ocfs2_xattr_search *xs)
732 {
733         struct ocfs2_xattr_entry *entry;
734         size_t name_len;
735         int i, cmp = 1;
736
737         if (name == NULL)
738                 return -EINVAL;
739
740         name_len = strlen(name);
741         entry = xs->here;
742         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
743                 cmp = name_index - ocfs2_xattr_get_type(entry);
744                 if (!cmp)
745                         cmp = name_len - entry->xe_name_len;
746                 if (!cmp)
747                         cmp = memcmp(name, (xs->base +
748                                      le16_to_cpu(entry->xe_name_offset)),
749                                      name_len);
750                 if (cmp == 0)
751                         break;
752                 entry += 1;
753         }
754         xs->here = entry;
755
756         return cmp ? -ENODATA : 0;
757 }
758
759 static int ocfs2_xattr_get_value_outside(struct inode *inode,
760                                          struct ocfs2_xattr_value_root *xv,
761                                          void *buffer,
762                                          size_t len)
763 {
764         u32 cpos, p_cluster, num_clusters, bpc, clusters;
765         u64 blkno;
766         int i, ret = 0;
767         size_t cplen, blocksize;
768         struct buffer_head *bh = NULL;
769         struct ocfs2_extent_list *el;
770
771         el = &xv->xr_list;
772         clusters = le32_to_cpu(xv->xr_clusters);
773         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
774         blocksize = inode->i_sb->s_blocksize;
775
776         cpos = 0;
777         while (cpos < clusters) {
778                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
779                                                &num_clusters, el);
780                 if (ret) {
781                         mlog_errno(ret);
782                         goto out;
783                 }
784
785                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
786                 /* Copy ocfs2_xattr_value */
787                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
788                         ret = ocfs2_read_block(inode, blkno, &bh);
789                         if (ret) {
790                                 mlog_errno(ret);
791                                 goto out;
792                         }
793
794                         cplen = len >= blocksize ? blocksize : len;
795                         memcpy(buffer, bh->b_data, cplen);
796                         len -= cplen;
797                         buffer += cplen;
798
799                         brelse(bh);
800                         bh = NULL;
801                         if (len == 0)
802                                 break;
803                 }
804                 cpos += num_clusters;
805         }
806 out:
807         return ret;
808 }
809
810 static int ocfs2_xattr_ibody_get(struct inode *inode,
811                                  int name_index,
812                                  const char *name,
813                                  void *buffer,
814                                  size_t buffer_size,
815                                  struct ocfs2_xattr_search *xs)
816 {
817         struct ocfs2_inode_info *oi = OCFS2_I(inode);
818         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
819         struct ocfs2_xattr_value_root *xv;
820         size_t size;
821         int ret = 0;
822
823         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
824                 return -ENODATA;
825
826         xs->end = (void *)di + inode->i_sb->s_blocksize;
827         xs->header = (struct ocfs2_xattr_header *)
828                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
829         xs->base = (void *)xs->header;
830         xs->here = xs->header->xh_entries;
831
832         ret = ocfs2_xattr_find_entry(name_index, name, xs);
833         if (ret)
834                 return ret;
835         size = le64_to_cpu(xs->here->xe_value_size);
836         if (buffer) {
837                 if (size > buffer_size)
838                         return -ERANGE;
839                 if (ocfs2_xattr_is_local(xs->here)) {
840                         memcpy(buffer, (void *)xs->base +
841                                le16_to_cpu(xs->here->xe_name_offset) +
842                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
843                 } else {
844                         xv = (struct ocfs2_xattr_value_root *)
845                                 (xs->base + le16_to_cpu(
846                                  xs->here->xe_name_offset) +
847                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
848                         ret = ocfs2_xattr_get_value_outside(inode, xv,
849                                                             buffer, size);
850                         if (ret < 0) {
851                                 mlog_errno(ret);
852                                 return ret;
853                         }
854                 }
855         }
856
857         return size;
858 }
859
860 static int ocfs2_xattr_block_get(struct inode *inode,
861                                  int name_index,
862                                  const char *name,
863                                  void *buffer,
864                                  size_t buffer_size,
865                                  struct ocfs2_xattr_search *xs)
866 {
867         struct ocfs2_xattr_block *xb;
868         struct ocfs2_xattr_value_root *xv;
869         size_t size;
870         int ret = -ENODATA, name_offset, name_len, block_off, i;
871
872         memset(&xs->bucket, 0, sizeof(xs->bucket));
873
874         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
875         if (ret) {
876                 mlog_errno(ret);
877                 goto cleanup;
878         }
879
880         if (xs->not_found) {
881                 ret = -ENODATA;
882                 goto cleanup;
883         }
884
885         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
886         size = le64_to_cpu(xs->here->xe_value_size);
887         if (buffer) {
888                 ret = -ERANGE;
889                 if (size > buffer_size)
890                         goto cleanup;
891
892                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
893                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
894                 i = xs->here - xs->header->xh_entries;
895
896                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
897                         ret = ocfs2_xattr_bucket_get_name_value(inode,
898                                                                 bucket_xh(&xs->bucket),
899                                                                 i,
900                                                                 &block_off,
901                                                                 &name_offset);
902                         xs->base = bucket_block(&xs->bucket, block_off);
903                 }
904                 if (ocfs2_xattr_is_local(xs->here)) {
905                         memcpy(buffer, (void *)xs->base +
906                                name_offset + name_len, size);
907                 } else {
908                         xv = (struct ocfs2_xattr_value_root *)
909                                 (xs->base + name_offset + name_len);
910                         ret = ocfs2_xattr_get_value_outside(inode, xv,
911                                                             buffer, size);
912                         if (ret < 0) {
913                                 mlog_errno(ret);
914                                 goto cleanup;
915                         }
916                 }
917         }
918         ret = size;
919 cleanup:
920         ocfs2_xattr_bucket_relse(inode, &xs->bucket);
921         memset(&xs->bucket, 0, sizeof(xs->bucket));
922
923         brelse(xs->xattr_bh);
924         xs->xattr_bh = NULL;
925         return ret;
926 }
927
928 /* ocfs2_xattr_get()
929  *
930  * Copy an extended attribute into the buffer provided.
931  * Buffer is NULL to compute the size of buffer required.
932  */
933 static int ocfs2_xattr_get(struct inode *inode,
934                            int name_index,
935                            const char *name,
936                            void *buffer,
937                            size_t buffer_size)
938 {
939         int ret;
940         struct ocfs2_dinode *di = NULL;
941         struct buffer_head *di_bh = NULL;
942         struct ocfs2_inode_info *oi = OCFS2_I(inode);
943         struct ocfs2_xattr_search xis = {
944                 .not_found = -ENODATA,
945         };
946         struct ocfs2_xattr_search xbs = {
947                 .not_found = -ENODATA,
948         };
949
950         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
951                 return -EOPNOTSUPP;
952
953         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
954                 ret = -ENODATA;
955
956         ret = ocfs2_inode_lock(inode, &di_bh, 0);
957         if (ret < 0) {
958                 mlog_errno(ret);
959                 return ret;
960         }
961         xis.inode_bh = xbs.inode_bh = di_bh;
962         di = (struct ocfs2_dinode *)di_bh->b_data;
963
964         down_read(&oi->ip_xattr_sem);
965         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
966                                     buffer_size, &xis);
967         if (ret == -ENODATA && di->i_xattr_loc)
968                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
969                                             buffer_size, &xbs);
970         up_read(&oi->ip_xattr_sem);
971         ocfs2_inode_unlock(inode, 0);
972
973         brelse(di_bh);
974
975         return ret;
976 }
977
978 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
979                                            struct ocfs2_xattr_value_root *xv,
980                                            const void *value,
981                                            int value_len)
982 {
983         int ret = 0, i, cp_len, credits;
984         u16 blocksize = inode->i_sb->s_blocksize;
985         u32 p_cluster, num_clusters;
986         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
987         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
988         u64 blkno;
989         struct buffer_head *bh = NULL;
990         handle_t *handle;
991
992         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
993
994         credits = clusters * bpc;
995         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
996         if (IS_ERR(handle)) {
997                 ret = PTR_ERR(handle);
998                 mlog_errno(ret);
999                 goto out;
1000         }
1001
1002         while (cpos < clusters) {
1003                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1004                                                &num_clusters, &xv->xr_list);
1005                 if (ret) {
1006                         mlog_errno(ret);
1007                         goto out_commit;
1008                 }
1009
1010                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1011
1012                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1013                         ret = ocfs2_read_block(inode, blkno, &bh);
1014                         if (ret) {
1015                                 mlog_errno(ret);
1016                                 goto out_commit;
1017                         }
1018
1019                         ret = ocfs2_journal_access(handle,
1020                                                    inode,
1021                                                    bh,
1022                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1023                         if (ret < 0) {
1024                                 mlog_errno(ret);
1025                                 goto out_commit;
1026                         }
1027
1028                         cp_len = value_len > blocksize ? blocksize : value_len;
1029                         memcpy(bh->b_data, value, cp_len);
1030                         value_len -= cp_len;
1031                         value += cp_len;
1032                         if (cp_len < blocksize)
1033                                 memset(bh->b_data + cp_len, 0,
1034                                        blocksize - cp_len);
1035
1036                         ret = ocfs2_journal_dirty(handle, bh);
1037                         if (ret < 0) {
1038                                 mlog_errno(ret);
1039                                 goto out_commit;
1040                         }
1041                         brelse(bh);
1042                         bh = NULL;
1043
1044                         /*
1045                          * XXX: do we need to empty all the following
1046                          * blocks in this cluster?
1047                          */
1048                         if (!value_len)
1049                                 break;
1050                 }
1051                 cpos += num_clusters;
1052         }
1053 out_commit:
1054         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1055 out:
1056         brelse(bh);
1057
1058         return ret;
1059 }
1060
1061 static int ocfs2_xattr_cleanup(struct inode *inode,
1062                                struct ocfs2_xattr_info *xi,
1063                                struct ocfs2_xattr_search *xs,
1064                                size_t offs)
1065 {
1066         handle_t *handle = NULL;
1067         int ret = 0;
1068         size_t name_len = strlen(xi->name);
1069         void *val = xs->base + offs;
1070         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1071
1072         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1073                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1074         if (IS_ERR(handle)) {
1075                 ret = PTR_ERR(handle);
1076                 mlog_errno(ret);
1077                 goto out;
1078         }
1079         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1080                                    OCFS2_JOURNAL_ACCESS_WRITE);
1081         if (ret) {
1082                 mlog_errno(ret);
1083                 goto out_commit;
1084         }
1085         /* Decrease xattr count */
1086         le16_add_cpu(&xs->header->xh_count, -1);
1087         /* Remove the xattr entry and tree root which has already be set*/
1088         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1089         memset(val, 0, size);
1090
1091         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1092         if (ret < 0)
1093                 mlog_errno(ret);
1094 out_commit:
1095         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1096 out:
1097         return ret;
1098 }
1099
1100 static int ocfs2_xattr_update_entry(struct inode *inode,
1101                                     struct ocfs2_xattr_info *xi,
1102                                     struct ocfs2_xattr_search *xs,
1103                                     size_t offs)
1104 {
1105         handle_t *handle = NULL;
1106         int ret = 0;
1107
1108         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1109                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1110         if (IS_ERR(handle)) {
1111                 ret = PTR_ERR(handle);
1112                 mlog_errno(ret);
1113                 goto out;
1114         }
1115         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1116                                    OCFS2_JOURNAL_ACCESS_WRITE);
1117         if (ret) {
1118                 mlog_errno(ret);
1119                 goto out_commit;
1120         }
1121
1122         xs->here->xe_name_offset = cpu_to_le16(offs);
1123         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1124         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1125                 ocfs2_xattr_set_local(xs->here, 1);
1126         else
1127                 ocfs2_xattr_set_local(xs->here, 0);
1128         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1129
1130         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1131         if (ret < 0)
1132                 mlog_errno(ret);
1133 out_commit:
1134         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1135 out:
1136         return ret;
1137 }
1138
1139 /*
1140  * ocfs2_xattr_set_value_outside()
1141  *
1142  * Set large size value in B tree.
1143  */
1144 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1145                                          struct ocfs2_xattr_info *xi,
1146                                          struct ocfs2_xattr_search *xs,
1147                                          size_t offs)
1148 {
1149         size_t name_len = strlen(xi->name);
1150         void *val = xs->base + offs;
1151         struct ocfs2_xattr_value_root *xv = NULL;
1152         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1153         int ret = 0;
1154
1155         memset(val, 0, size);
1156         memcpy(val, xi->name, name_len);
1157         xv = (struct ocfs2_xattr_value_root *)
1158                 (val + OCFS2_XATTR_SIZE(name_len));
1159         xv->xr_clusters = 0;
1160         xv->xr_last_eb_blk = 0;
1161         xv->xr_list.l_tree_depth = 0;
1162         xv->xr_list.l_count = cpu_to_le16(1);
1163         xv->xr_list.l_next_free_rec = 0;
1164
1165         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1166                                          xi->value_len);
1167         if (ret < 0) {
1168                 mlog_errno(ret);
1169                 return ret;
1170         }
1171         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1172                                               xi->value_len);
1173         if (ret < 0) {
1174                 mlog_errno(ret);
1175                 return ret;
1176         }
1177         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1178         if (ret < 0)
1179                 mlog_errno(ret);
1180
1181         return ret;
1182 }
1183
1184 /*
1185  * ocfs2_xattr_set_entry_local()
1186  *
1187  * Set, replace or remove extended attribute in local.
1188  */
1189 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1190                                         struct ocfs2_xattr_info *xi,
1191                                         struct ocfs2_xattr_search *xs,
1192                                         struct ocfs2_xattr_entry *last,
1193                                         size_t min_offs)
1194 {
1195         size_t name_len = strlen(xi->name);
1196         int i;
1197
1198         if (xi->value && xs->not_found) {
1199                 /* Insert the new xattr entry. */
1200                 le16_add_cpu(&xs->header->xh_count, 1);
1201                 ocfs2_xattr_set_type(last, xi->name_index);
1202                 ocfs2_xattr_set_local(last, 1);
1203                 last->xe_name_len = name_len;
1204         } else {
1205                 void *first_val;
1206                 void *val;
1207                 size_t offs, size;
1208
1209                 first_val = xs->base + min_offs;
1210                 offs = le16_to_cpu(xs->here->xe_name_offset);
1211                 val = xs->base + offs;
1212
1213                 if (le64_to_cpu(xs->here->xe_value_size) >
1214                     OCFS2_XATTR_INLINE_SIZE)
1215                         size = OCFS2_XATTR_SIZE(name_len) +
1216                                 OCFS2_XATTR_ROOT_SIZE;
1217                 else
1218                         size = OCFS2_XATTR_SIZE(name_len) +
1219                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1220
1221                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1222                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1223                         /* The old and the new value have the
1224                            same size. Just replace the value. */
1225                         ocfs2_xattr_set_local(xs->here, 1);
1226                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1227                         /* Clear value bytes. */
1228                         memset(val + OCFS2_XATTR_SIZE(name_len),
1229                                0,
1230                                OCFS2_XATTR_SIZE(xi->value_len));
1231                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1232                                xi->value,
1233                                xi->value_len);
1234                         return;
1235                 }
1236                 /* Remove the old name+value. */
1237                 memmove(first_val + size, first_val, val - first_val);
1238                 memset(first_val, 0, size);
1239                 xs->here->xe_name_hash = 0;
1240                 xs->here->xe_name_offset = 0;
1241                 ocfs2_xattr_set_local(xs->here, 1);
1242                 xs->here->xe_value_size = 0;
1243
1244                 min_offs += size;
1245
1246                 /* Adjust all value offsets. */
1247                 last = xs->header->xh_entries;
1248                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1249                         size_t o = le16_to_cpu(last->xe_name_offset);
1250
1251                         if (o < offs)
1252                                 last->xe_name_offset = cpu_to_le16(o + size);
1253                         last += 1;
1254                 }
1255
1256                 if (!xi->value) {
1257                         /* Remove the old entry. */
1258                         last -= 1;
1259                         memmove(xs->here, xs->here + 1,
1260                                 (void *)last - (void *)xs->here);
1261                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1262                         le16_add_cpu(&xs->header->xh_count, -1);
1263                 }
1264         }
1265         if (xi->value) {
1266                 /* Insert the new name+value. */
1267                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1268                                 OCFS2_XATTR_SIZE(xi->value_len);
1269                 void *val = xs->base + min_offs - size;
1270
1271                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1272                 memset(val, 0, size);
1273                 memcpy(val, xi->name, name_len);
1274                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1275                        xi->value,
1276                        xi->value_len);
1277                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1278                 ocfs2_xattr_set_local(xs->here, 1);
1279                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1280         }
1281
1282         return;
1283 }
1284
1285 /*
1286  * ocfs2_xattr_set_entry()
1287  *
1288  * Set extended attribute entry into inode or block.
1289  *
1290  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1291  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1292  * then set value in B tree with set_value_outside().
1293  */
1294 static int ocfs2_xattr_set_entry(struct inode *inode,
1295                                  struct ocfs2_xattr_info *xi,
1296                                  struct ocfs2_xattr_search *xs,
1297                                  int flag)
1298 {
1299         struct ocfs2_xattr_entry *last;
1300         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1301         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1302         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1303         size_t size_l = 0;
1304         handle_t *handle = NULL;
1305         int free, i, ret;
1306         struct ocfs2_xattr_info xi_l = {
1307                 .name_index = xi->name_index,
1308                 .name = xi->name,
1309                 .value = xi->value,
1310                 .value_len = xi->value_len,
1311         };
1312
1313         /* Compute min_offs, last and free space. */
1314         last = xs->header->xh_entries;
1315
1316         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1317                 size_t offs = le16_to_cpu(last->xe_name_offset);
1318                 if (offs < min_offs)
1319                         min_offs = offs;
1320                 last += 1;
1321         }
1322
1323         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1324         if (free < 0)
1325                 return -EIO;
1326
1327         if (!xs->not_found) {
1328                 size_t size = 0;
1329                 if (ocfs2_xattr_is_local(xs->here))
1330                         size = OCFS2_XATTR_SIZE(name_len) +
1331                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1332                 else
1333                         size = OCFS2_XATTR_SIZE(name_len) +
1334                                 OCFS2_XATTR_ROOT_SIZE;
1335                 free += (size + sizeof(struct ocfs2_xattr_entry));
1336         }
1337         /* Check free space in inode or block */
1338         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1339                 if (free < sizeof(struct ocfs2_xattr_entry) +
1340                            OCFS2_XATTR_SIZE(name_len) +
1341                            OCFS2_XATTR_ROOT_SIZE) {
1342                         ret = -ENOSPC;
1343                         goto out;
1344                 }
1345                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1346                 xi_l.value = (void *)&def_xv;
1347                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1348         } else if (xi->value) {
1349                 if (free < sizeof(struct ocfs2_xattr_entry) +
1350                            OCFS2_XATTR_SIZE(name_len) +
1351                            OCFS2_XATTR_SIZE(xi->value_len)) {
1352                         ret = -ENOSPC;
1353                         goto out;
1354                 }
1355         }
1356
1357         if (!xs->not_found) {
1358                 /* For existing extended attribute */
1359                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1360                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1361                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1362                 void *val = xs->base + offs;
1363
1364                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1365                         /* Replace existing local xattr with tree root */
1366                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1367                                                             offs);
1368                         if (ret < 0)
1369                                 mlog_errno(ret);
1370                         goto out;
1371                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1372                         /* For existing xattr which has value outside */
1373                         struct ocfs2_xattr_value_root *xv = NULL;
1374                         xv = (struct ocfs2_xattr_value_root *)(val +
1375                                 OCFS2_XATTR_SIZE(name_len));
1376
1377                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1378                                 /*
1379                                  * If new value need set outside also,
1380                                  * first truncate old value to new value,
1381                                  * then set new value with set_value_outside().
1382                                  */
1383                                 ret = ocfs2_xattr_value_truncate(inode,
1384                                                                  xs->xattr_bh,
1385                                                                  xv,
1386                                                                  xi->value_len);
1387                                 if (ret < 0) {
1388                                         mlog_errno(ret);
1389                                         goto out;
1390                                 }
1391
1392                                 ret = __ocfs2_xattr_set_value_outside(inode,
1393                                                                 xv,
1394                                                                 xi->value,
1395                                                                 xi->value_len);
1396                                 if (ret < 0) {
1397                                         mlog_errno(ret);
1398                                         goto out;
1399                                 }
1400
1401                                 ret = ocfs2_xattr_update_entry(inode,
1402                                                                xi,
1403                                                                xs,
1404                                                                offs);
1405                                 if (ret < 0)
1406                                         mlog_errno(ret);
1407                                 goto out;
1408                         } else {
1409                                 /*
1410                                  * If new value need set in local,
1411                                  * just trucate old value to zero.
1412                                  */
1413                                  ret = ocfs2_xattr_value_truncate(inode,
1414                                                                  xs->xattr_bh,
1415                                                                  xv,
1416                                                                  0);
1417                                 if (ret < 0)
1418                                         mlog_errno(ret);
1419                         }
1420                 }
1421         }
1422
1423         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1424                                    OCFS2_INODE_UPDATE_CREDITS);
1425         if (IS_ERR(handle)) {
1426                 ret = PTR_ERR(handle);
1427                 mlog_errno(ret);
1428                 goto out;
1429         }
1430
1431         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1432                                    OCFS2_JOURNAL_ACCESS_WRITE);
1433         if (ret) {
1434                 mlog_errno(ret);
1435                 goto out_commit;
1436         }
1437
1438         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1439                 /* set extended attribute in external block. */
1440                 ret = ocfs2_extend_trans(handle,
1441                                          OCFS2_INODE_UPDATE_CREDITS +
1442                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1443                 if (ret) {
1444                         mlog_errno(ret);
1445                         goto out_commit;
1446                 }
1447                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1448                                            OCFS2_JOURNAL_ACCESS_WRITE);
1449                 if (ret) {
1450                         mlog_errno(ret);
1451                         goto out_commit;
1452                 }
1453         }
1454
1455         /*
1456          * Set value in local, include set tree root in local.
1457          * This is the first step for value size >INLINE_SIZE.
1458          */
1459         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1460
1461         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1462                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1463                 if (ret < 0) {
1464                         mlog_errno(ret);
1465                         goto out_commit;
1466                 }
1467         }
1468
1469         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1470             (flag & OCFS2_INLINE_XATTR_FL)) {
1471                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1472                 unsigned int xattrsize = osb->s_xattr_inline_size;
1473
1474                 /*
1475                  * Adjust extent record count or inline data size
1476                  * to reserve space for extended attribute.
1477                  */
1478                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1479                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1480                         le16_add_cpu(&idata->id_count, -xattrsize);
1481                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1482                         struct ocfs2_extent_list *el = &di->id2.i_list;
1483                         le16_add_cpu(&el->l_count, -(xattrsize /
1484                                         sizeof(struct ocfs2_extent_rec)));
1485                 }
1486                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1487         }
1488         /* Update xattr flag */
1489         spin_lock(&oi->ip_lock);
1490         oi->ip_dyn_features |= flag;
1491         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1492         spin_unlock(&oi->ip_lock);
1493         /* Update inode ctime */
1494         inode->i_ctime = CURRENT_TIME;
1495         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1496         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1497
1498         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1499         if (ret < 0)
1500                 mlog_errno(ret);
1501
1502 out_commit:
1503         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1504
1505         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1506                 /*
1507                  * Set value outside in B tree.
1508                  * This is the second step for value size > INLINE_SIZE.
1509                  */
1510                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1511                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1512                 if (ret < 0) {
1513                         int ret2;
1514
1515                         mlog_errno(ret);
1516                         /*
1517                          * If set value outside failed, we have to clean
1518                          * the junk tree root we have already set in local.
1519                          */
1520                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1521                         if (ret2 < 0)
1522                                 mlog_errno(ret2);
1523                 }
1524         }
1525 out:
1526         return ret;
1527
1528 }
1529
1530 static int ocfs2_remove_value_outside(struct inode*inode,
1531                                       struct buffer_head *bh,
1532                                       struct ocfs2_xattr_header *header)
1533 {
1534         int ret = 0, i;
1535
1536         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1537                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1538
1539                 if (!ocfs2_xattr_is_local(entry)) {
1540                         struct ocfs2_xattr_value_root *xv;
1541                         void *val;
1542
1543                         val = (void *)header +
1544                                 le16_to_cpu(entry->xe_name_offset);
1545                         xv = (struct ocfs2_xattr_value_root *)
1546                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1547                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1548                         if (ret < 0) {
1549                                 mlog_errno(ret);
1550                                 return ret;
1551                         }
1552                 }
1553         }
1554
1555         return ret;
1556 }
1557
1558 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1559                                     struct buffer_head *di_bh)
1560 {
1561
1562         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1563         struct ocfs2_xattr_header *header;
1564         int ret;
1565
1566         header = (struct ocfs2_xattr_header *)
1567                  ((void *)di + inode->i_sb->s_blocksize -
1568                  le16_to_cpu(di->i_xattr_inline_size));
1569
1570         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1571
1572         return ret;
1573 }
1574
1575 static int ocfs2_xattr_block_remove(struct inode *inode,
1576                                     struct buffer_head *blk_bh)
1577 {
1578         struct ocfs2_xattr_block *xb;
1579         int ret = 0;
1580
1581         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1582         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1583                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1584                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1585         } else
1586                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1587
1588         return ret;
1589 }
1590
1591 static int ocfs2_xattr_free_block(struct inode *inode,
1592                                   u64 block)
1593 {
1594         struct inode *xb_alloc_inode;
1595         struct buffer_head *xb_alloc_bh = NULL;
1596         struct buffer_head *blk_bh = NULL;
1597         struct ocfs2_xattr_block *xb;
1598         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1599         handle_t *handle;
1600         int ret = 0;
1601         u64 blk, bg_blkno;
1602         u16 bit;
1603
1604         ret = ocfs2_read_block(inode, block, &blk_bh);
1605         if (ret < 0) {
1606                 mlog_errno(ret);
1607                 goto out;
1608         }
1609
1610         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1611         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1612                 ret = -EIO;
1613                 goto out;
1614         }
1615
1616         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1617         if (ret < 0) {
1618                 mlog_errno(ret);
1619                 goto out;
1620         }
1621
1622         blk = le64_to_cpu(xb->xb_blkno);
1623         bit = le16_to_cpu(xb->xb_suballoc_bit);
1624         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1625
1626         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1627                                 EXTENT_ALLOC_SYSTEM_INODE,
1628                                 le16_to_cpu(xb->xb_suballoc_slot));
1629         if (!xb_alloc_inode) {
1630                 ret = -ENOMEM;
1631                 mlog_errno(ret);
1632                 goto out;
1633         }
1634         mutex_lock(&xb_alloc_inode->i_mutex);
1635
1636         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1637         if (ret < 0) {
1638                 mlog_errno(ret);
1639                 goto out_mutex;
1640         }
1641
1642         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1643         if (IS_ERR(handle)) {
1644                 ret = PTR_ERR(handle);
1645                 mlog_errno(ret);
1646                 goto out_unlock;
1647         }
1648
1649         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1650                                        bit, bg_blkno, 1);
1651         if (ret < 0)
1652                 mlog_errno(ret);
1653
1654         ocfs2_commit_trans(osb, handle);
1655 out_unlock:
1656         ocfs2_inode_unlock(xb_alloc_inode, 1);
1657         brelse(xb_alloc_bh);
1658 out_mutex:
1659         mutex_unlock(&xb_alloc_inode->i_mutex);
1660         iput(xb_alloc_inode);
1661 out:
1662         brelse(blk_bh);
1663         return ret;
1664 }
1665
1666 /*
1667  * ocfs2_xattr_remove()
1668  *
1669  * Free extended attribute resources associated with this inode.
1670  */
1671 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1672 {
1673         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1674         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1675         handle_t *handle;
1676         int ret;
1677
1678         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1679                 return 0;
1680
1681         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1682                 return 0;
1683
1684         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1685                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1686                 if (ret < 0) {
1687                         mlog_errno(ret);
1688                         goto out;
1689                 }
1690         }
1691
1692         if (di->i_xattr_loc) {
1693                 ret = ocfs2_xattr_free_block(inode,
1694                                              le64_to_cpu(di->i_xattr_loc));
1695                 if (ret < 0) {
1696                         mlog_errno(ret);
1697                         goto out;
1698                 }
1699         }
1700
1701         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1702                                    OCFS2_INODE_UPDATE_CREDITS);
1703         if (IS_ERR(handle)) {
1704                 ret = PTR_ERR(handle);
1705                 mlog_errno(ret);
1706                 goto out;
1707         }
1708         ret = ocfs2_journal_access(handle, inode, di_bh,
1709                                    OCFS2_JOURNAL_ACCESS_WRITE);
1710         if (ret) {
1711                 mlog_errno(ret);
1712                 goto out_commit;
1713         }
1714
1715         di->i_xattr_loc = 0;
1716
1717         spin_lock(&oi->ip_lock);
1718         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1719         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1720         spin_unlock(&oi->ip_lock);
1721
1722         ret = ocfs2_journal_dirty(handle, di_bh);
1723         if (ret < 0)
1724                 mlog_errno(ret);
1725 out_commit:
1726         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1727 out:
1728         return ret;
1729 }
1730
1731 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1732                                         struct ocfs2_dinode *di)
1733 {
1734         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1735         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1736         int free;
1737
1738         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1739                 return 0;
1740
1741         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1742                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1743                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1744         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1745                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1746                         le64_to_cpu(di->i_size);
1747         } else {
1748                 struct ocfs2_extent_list *el = &di->id2.i_list;
1749                 free = (le16_to_cpu(el->l_count) -
1750                         le16_to_cpu(el->l_next_free_rec)) *
1751                         sizeof(struct ocfs2_extent_rec);
1752         }
1753         if (free >= xattrsize)
1754                 return 1;
1755
1756         return 0;
1757 }
1758
1759 /*
1760  * ocfs2_xattr_ibody_find()
1761  *
1762  * Find extended attribute in inode block and
1763  * fill search info into struct ocfs2_xattr_search.
1764  */
1765 static int ocfs2_xattr_ibody_find(struct inode *inode,
1766                                   int name_index,
1767                                   const char *name,
1768                                   struct ocfs2_xattr_search *xs)
1769 {
1770         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1771         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1772         int ret;
1773         int has_space = 0;
1774
1775         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1776                 return 0;
1777
1778         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1779                 down_read(&oi->ip_alloc_sem);
1780                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1781                 up_read(&oi->ip_alloc_sem);
1782                 if (!has_space)
1783                         return 0;
1784         }
1785
1786         xs->xattr_bh = xs->inode_bh;
1787         xs->end = (void *)di + inode->i_sb->s_blocksize;
1788         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1789                 xs->header = (struct ocfs2_xattr_header *)
1790                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1791         else
1792                 xs->header = (struct ocfs2_xattr_header *)
1793                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1794         xs->base = (void *)xs->header;
1795         xs->here = xs->header->xh_entries;
1796
1797         /* Find the named attribute. */
1798         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1799                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1800                 if (ret && ret != -ENODATA)
1801                         return ret;
1802                 xs->not_found = ret;
1803         }
1804
1805         return 0;
1806 }
1807
1808 /*
1809  * ocfs2_xattr_ibody_set()
1810  *
1811  * Set, replace or remove an extended attribute into inode block.
1812  *
1813  */
1814 static int ocfs2_xattr_ibody_set(struct inode *inode,
1815                                  struct ocfs2_xattr_info *xi,
1816                                  struct ocfs2_xattr_search *xs)
1817 {
1818         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1819         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1820         int ret;
1821
1822         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1823                 return -ENOSPC;
1824
1825         down_write(&oi->ip_alloc_sem);
1826         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1827                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1828                         ret = -ENOSPC;
1829                         goto out;
1830                 }
1831         }
1832
1833         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1834                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1835 out:
1836         up_write(&oi->ip_alloc_sem);
1837
1838         return ret;
1839 }
1840
1841 /*
1842  * ocfs2_xattr_block_find()
1843  *
1844  * Find extended attribute in external block and
1845  * fill search info into struct ocfs2_xattr_search.
1846  */
1847 static int ocfs2_xattr_block_find(struct inode *inode,
1848                                   int name_index,
1849                                   const char *name,
1850                                   struct ocfs2_xattr_search *xs)
1851 {
1852         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1853         struct buffer_head *blk_bh = NULL;
1854         struct ocfs2_xattr_block *xb;
1855         int ret = 0;
1856
1857         if (!di->i_xattr_loc)
1858                 return ret;
1859
1860         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
1861         if (ret < 0) {
1862                 mlog_errno(ret);
1863                 return ret;
1864         }
1865
1866         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1867         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1868                 ret = -EIO;
1869                 goto cleanup;
1870         }
1871
1872         xs->xattr_bh = blk_bh;
1873
1874         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1875                 xs->header = &xb->xb_attrs.xb_header;
1876                 xs->base = (void *)xs->header;
1877                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1878                 xs->here = xs->header->xh_entries;
1879
1880                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1881         } else
1882                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1883                                                    name_index,
1884                                                    name, xs);
1885
1886         if (ret && ret != -ENODATA) {
1887                 xs->xattr_bh = NULL;
1888                 goto cleanup;
1889         }
1890         xs->not_found = ret;
1891         return 0;
1892 cleanup:
1893         brelse(blk_bh);
1894
1895         return ret;
1896 }
1897
1898 /*
1899  * ocfs2_xattr_block_set()
1900  *
1901  * Set, replace or remove an extended attribute into external block.
1902  *
1903  */
1904 static int ocfs2_xattr_block_set(struct inode *inode,
1905                                  struct ocfs2_xattr_info *xi,
1906                                  struct ocfs2_xattr_search *xs)
1907 {
1908         struct buffer_head *new_bh = NULL;
1909         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1910         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1911         struct ocfs2_alloc_context *meta_ac = NULL;
1912         handle_t *handle = NULL;
1913         struct ocfs2_xattr_block *xblk = NULL;
1914         u16 suballoc_bit_start;
1915         u32 num_got;
1916         u64 first_blkno;
1917         int ret;
1918
1919         if (!xs->xattr_bh) {
1920                 /*
1921                  * Alloc one external block for extended attribute
1922                  * outside of inode.
1923                  */
1924                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1925                 if (ret < 0) {
1926                         mlog_errno(ret);
1927                         goto out;
1928                 }
1929                 handle = ocfs2_start_trans(osb,
1930                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1931                 if (IS_ERR(handle)) {
1932                         ret = PTR_ERR(handle);
1933                         mlog_errno(ret);
1934                         goto out;
1935                 }
1936                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1937                                            OCFS2_JOURNAL_ACCESS_CREATE);
1938                 if (ret < 0) {
1939                         mlog_errno(ret);
1940                         goto out_commit;
1941                 }
1942
1943                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1944                                            &suballoc_bit_start, &num_got,
1945                                            &first_blkno);
1946                 if (ret < 0) {
1947                         mlog_errno(ret);
1948                         goto out_commit;
1949                 }
1950
1951                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1952                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1953
1954                 ret = ocfs2_journal_access(handle, inode, new_bh,
1955                                            OCFS2_JOURNAL_ACCESS_CREATE);
1956                 if (ret < 0) {
1957                         mlog_errno(ret);
1958                         goto out_commit;
1959                 }
1960
1961                 /* Initialize ocfs2_xattr_block */
1962                 xs->xattr_bh = new_bh;
1963                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1964                 memset(xblk, 0, inode->i_sb->s_blocksize);
1965                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1966                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1967                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1968                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1969                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1970
1971                 xs->header = &xblk->xb_attrs.xb_header;
1972                 xs->base = (void *)xs->header;
1973                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1974                 xs->here = xs->header->xh_entries;
1975
1976
1977                 ret = ocfs2_journal_dirty(handle, new_bh);
1978                 if (ret < 0) {
1979                         mlog_errno(ret);
1980                         goto out_commit;
1981                 }
1982                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1983                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1984                 if (ret < 0)
1985                         mlog_errno(ret);
1986 out_commit:
1987                 ocfs2_commit_trans(osb, handle);
1988 out:
1989                 if (meta_ac)
1990                         ocfs2_free_alloc_context(meta_ac);
1991                 if (ret < 0)
1992                         return ret;
1993         } else
1994                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1995
1996         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1997                 /* Set extended attribute into external block */
1998                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1999                 if (!ret || ret != -ENOSPC)
2000                         goto end;
2001
2002                 ret = ocfs2_xattr_create_index_block(inode, xs);
2003                 if (ret)
2004                         goto end;
2005         }
2006
2007         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
2008
2009 end:
2010
2011         return ret;
2012 }
2013
2014 /*
2015  * ocfs2_xattr_set()
2016  *
2017  * Set, replace or remove an extended attribute for this inode.
2018  * value is NULL to remove an existing extended attribute, else either
2019  * create or replace an extended attribute.
2020  */
2021 int ocfs2_xattr_set(struct inode *inode,
2022                     int name_index,
2023                     const char *name,
2024                     const void *value,
2025                     size_t value_len,
2026                     int flags)
2027 {
2028         struct buffer_head *di_bh = NULL;
2029         struct ocfs2_dinode *di;
2030         int ret;
2031
2032         struct ocfs2_xattr_info xi = {
2033                 .name_index = name_index,
2034                 .name = name,
2035                 .value = value,
2036                 .value_len = value_len,
2037         };
2038
2039         struct ocfs2_xattr_search xis = {
2040                 .not_found = -ENODATA,
2041         };
2042
2043         struct ocfs2_xattr_search xbs = {
2044                 .not_found = -ENODATA,
2045         };
2046
2047         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2048                 return -EOPNOTSUPP;
2049
2050         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2051         if (ret < 0) {
2052                 mlog_errno(ret);
2053                 return ret;
2054         }
2055         xis.inode_bh = xbs.inode_bh = di_bh;
2056         di = (struct ocfs2_dinode *)di_bh->b_data;
2057
2058         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2059         /*
2060          * Scan inode and external block to find the same name
2061          * extended attribute and collect search infomation.
2062          */
2063         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2064         if (ret)
2065                 goto cleanup;
2066         if (xis.not_found) {
2067                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2068                 if (ret)
2069                         goto cleanup;
2070         }
2071
2072         if (xis.not_found && xbs.not_found) {
2073                 ret = -ENODATA;
2074                 if (flags & XATTR_REPLACE)
2075                         goto cleanup;
2076                 ret = 0;
2077                 if (!value)
2078                         goto cleanup;
2079         } else {
2080                 ret = -EEXIST;
2081                 if (flags & XATTR_CREATE)
2082                         goto cleanup;
2083         }
2084
2085         if (!value) {
2086                 /* Remove existing extended attribute */
2087                 if (!xis.not_found)
2088                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2089                 else if (!xbs.not_found)
2090                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2091         } else {
2092                 /* We always try to set extended attribute into inode first*/
2093                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2094                 if (!ret && !xbs.not_found) {
2095                         /*
2096                          * If succeed and that extended attribute existing in
2097                          * external block, then we will remove it.
2098                          */
2099                         xi.value = NULL;
2100                         xi.value_len = 0;
2101                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2102                 } else if (ret == -ENOSPC) {
2103                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2104                                 ret = ocfs2_xattr_block_find(inode, name_index,
2105                                                              name, &xbs);
2106                                 if (ret)
2107                                         goto cleanup;
2108                         }
2109                         /*
2110                          * If no space in inode, we will set extended attribute
2111                          * into external block.
2112                          */
2113                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2114                         if (ret)
2115                                 goto cleanup;
2116                         if (!xis.not_found) {
2117                                 /*
2118                                  * If succeed and that extended attribute
2119                                  * existing in inode, we will remove it.
2120                                  */
2121                                 xi.value = NULL;
2122                                 xi.value_len = 0;
2123                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2124                         }
2125                 }
2126         }
2127 cleanup:
2128         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2129         ocfs2_inode_unlock(inode, 1);
2130         brelse(di_bh);
2131         brelse(xbs.xattr_bh);
2132         ocfs2_xattr_bucket_relse(inode, &xbs.bucket);
2133
2134         return ret;
2135 }
2136
2137 /*
2138  * Find the xattr extent rec which may contains name_hash.
2139  * e_cpos will be the first name hash of the xattr rec.
2140  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2141  */
2142 static int ocfs2_xattr_get_rec(struct inode *inode,
2143                                u32 name_hash,
2144                                u64 *p_blkno,
2145                                u32 *e_cpos,
2146                                u32 *num_clusters,
2147                                struct ocfs2_extent_list *el)
2148 {
2149         int ret = 0, i;
2150         struct buffer_head *eb_bh = NULL;
2151         struct ocfs2_extent_block *eb;
2152         struct ocfs2_extent_rec *rec = NULL;
2153         u64 e_blkno = 0;
2154
2155         if (el->l_tree_depth) {
2156                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2157                 if (ret) {
2158                         mlog_errno(ret);
2159                         goto out;
2160                 }
2161
2162                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2163                 el = &eb->h_list;
2164
2165                 if (el->l_tree_depth) {
2166                         ocfs2_error(inode->i_sb,
2167                                     "Inode %lu has non zero tree depth in "
2168                                     "xattr tree block %llu\n", inode->i_ino,
2169                                     (unsigned long long)eb_bh->b_blocknr);
2170                         ret = -EROFS;
2171                         goto out;
2172                 }
2173         }
2174
2175         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2176                 rec = &el->l_recs[i];
2177
2178                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2179                         e_blkno = le64_to_cpu(rec->e_blkno);
2180                         break;
2181                 }
2182         }
2183
2184         if (!e_blkno) {
2185                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2186                             "record (%u, %u, 0) in xattr", inode->i_ino,
2187                             le32_to_cpu(rec->e_cpos),
2188                             ocfs2_rec_clusters(el, rec));
2189                 ret = -EROFS;
2190                 goto out;
2191         }
2192
2193         *p_blkno = le64_to_cpu(rec->e_blkno);
2194         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2195         if (e_cpos)
2196                 *e_cpos = le32_to_cpu(rec->e_cpos);
2197 out:
2198         brelse(eb_bh);
2199         return ret;
2200 }
2201
2202 typedef int (xattr_bucket_func)(struct inode *inode,
2203                                 struct ocfs2_xattr_bucket *bucket,
2204                                 void *para);
2205
2206 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2207                                    struct buffer_head *header_bh,
2208                                    int name_index,
2209                                    const char *name,
2210                                    u32 name_hash,
2211                                    u16 *xe_index,
2212                                    int *found)
2213 {
2214         int i, ret = 0, cmp = 1, block_off, new_offset;
2215         struct ocfs2_xattr_header *xh =
2216                         (struct ocfs2_xattr_header *)header_bh->b_data;
2217         size_t name_len = strlen(name);
2218         struct ocfs2_xattr_entry *xe = NULL;
2219         struct buffer_head *name_bh = NULL;
2220         char *xe_name;
2221
2222         /*
2223          * We don't use binary search in the bucket because there
2224          * may be multiple entries with the same name hash.
2225          */
2226         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2227                 xe = &xh->xh_entries[i];
2228
2229                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2230                         continue;
2231                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2232                         break;
2233
2234                 cmp = name_index - ocfs2_xattr_get_type(xe);
2235                 if (!cmp)
2236                         cmp = name_len - xe->xe_name_len;
2237                 if (cmp)
2238                         continue;
2239
2240                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2241                                                         xh,
2242                                                         i,
2243                                                         &block_off,
2244                                                         &new_offset);
2245                 if (ret) {
2246                         mlog_errno(ret);
2247                         break;
2248                 }
2249
2250                 ret = ocfs2_read_block(inode, header_bh->b_blocknr + block_off,
2251                                        &name_bh);
2252                 if (ret) {
2253                         mlog_errno(ret);
2254                         break;
2255                 }
2256                 xe_name = name_bh->b_data + new_offset;
2257
2258                 cmp = memcmp(name, xe_name, name_len);
2259                 brelse(name_bh);
2260                 name_bh = NULL;
2261
2262                 if (cmp == 0) {
2263                         *xe_index = i;
2264                         *found = 1;
2265                         ret = 0;
2266                         break;
2267                 }
2268         }
2269
2270         return ret;
2271 }
2272
2273 /*
2274  * Find the specified xattr entry in a series of buckets.
2275  * This series start from p_blkno and last for num_clusters.
2276  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2277  * the num of the valid buckets.
2278  *
2279  * Return the buffer_head this xattr should reside in. And if the xattr's
2280  * hash is in the gap of 2 buckets, return the lower bucket.
2281  */
2282 static int ocfs2_xattr_bucket_find(struct inode *inode,
2283                                    int name_index,
2284                                    const char *name,
2285                                    u32 name_hash,
2286                                    u64 p_blkno,
2287                                    u32 first_hash,
2288                                    u32 num_clusters,
2289                                    struct ocfs2_xattr_search *xs)
2290 {
2291         int ret, found = 0;
2292         struct buffer_head *bh = NULL;
2293         struct buffer_head *lower_bh = NULL;
2294         struct ocfs2_xattr_header *xh = NULL;
2295         struct ocfs2_xattr_entry *xe = NULL;
2296         u16 index = 0;
2297         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2298         int low_bucket = 0, bucket, high_bucket;
2299         u32 last_hash;
2300         u64 blkno;
2301
2302         ret = ocfs2_read_block(inode, p_blkno, &bh);
2303         if (ret) {
2304                 mlog_errno(ret);
2305                 goto out;
2306         }
2307
2308         xh = (struct ocfs2_xattr_header *)bh->b_data;
2309         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2310
2311         while (low_bucket <= high_bucket) {
2312                 brelse(bh);
2313                 bh = NULL;
2314                 bucket = (low_bucket + high_bucket) / 2;
2315
2316                 blkno = p_blkno + bucket * blk_per_bucket;
2317
2318                 ret = ocfs2_read_block(inode, blkno, &bh);
2319                 if (ret) {
2320                         mlog_errno(ret);
2321                         goto out;
2322                 }
2323
2324                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2325                 xe = &xh->xh_entries[0];
2326                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2327                         high_bucket = bucket - 1;
2328                         continue;
2329                 }
2330
2331                 /*
2332                  * Check whether the hash of the last entry in our
2333                  * bucket is larger than the search one. for an empty
2334                  * bucket, the last one is also the first one.
2335                  */
2336                 if (xh->xh_count)
2337                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2338
2339                 last_hash = le32_to_cpu(xe->xe_name_hash);
2340
2341                 /* record lower_bh which may be the insert place. */
2342                 brelse(lower_bh);
2343                 lower_bh = bh;
2344                 bh = NULL;
2345
2346                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2347                         low_bucket = bucket + 1;
2348                         continue;
2349                 }
2350
2351                 /* the searched xattr should reside in this bucket if exists. */
2352                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2353                                               name_index, name, name_hash,
2354                                               &index, &found);
2355                 if (ret) {
2356                         mlog_errno(ret);
2357                         goto out;
2358                 }
2359                 break;
2360         }
2361
2362         /*
2363          * Record the bucket we have found.
2364          * When the xattr's hash value is in the gap of 2 buckets, we will
2365          * always set it to the previous bucket.
2366          */
2367         if (!lower_bh) {
2368                 /*
2369                  * We can't find any bucket whose first name_hash is less
2370                  * than the find name_hash.
2371                  */
2372                 BUG_ON(bh->b_blocknr != p_blkno);
2373                 lower_bh = bh;
2374                 bh = NULL;
2375         }
2376         xs->bucket.bu_bhs[0] = lower_bh;
2377         lower_bh = NULL;
2378
2379         xs->header = bucket_xh(&xs->bucket);
2380         xs->base = bucket_block(&xs->bucket, 0);
2381         xs->end = xs->base + inode->i_sb->s_blocksize;
2382
2383         if (found) {
2384                 /*
2385                  * If we have found the xattr enty, read all the blocks in
2386                  * this bucket.
2387                  */
2388                 ret = ocfs2_read_blocks(inode, bucket_blkno(&xs->bucket) + 1,
2389                                         blk_per_bucket - 1, &xs->bucket.bu_bhs[1],
2390                                         0);
2391                 if (ret) {
2392                         mlog_errno(ret);
2393                         goto out;
2394                 }
2395
2396                 xs->here = &xs->header->xh_entries[index];
2397                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2398                      (unsigned long long)bucket_blkno(&xs->bucket), index);
2399         } else
2400                 ret = -ENODATA;
2401
2402 out:
2403         brelse(bh);
2404         brelse(lower_bh);
2405         return ret;
2406 }
2407
2408 static int ocfs2_xattr_index_block_find(struct inode *inode,
2409                                         struct buffer_head *root_bh,
2410                                         int name_index,
2411                                         const char *name,
2412                                         struct ocfs2_xattr_search *xs)
2413 {
2414         int ret;
2415         struct ocfs2_xattr_block *xb =
2416                         (struct ocfs2_xattr_block *)root_bh->b_data;
2417         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2418         struct ocfs2_extent_list *el = &xb_root->xt_list;
2419         u64 p_blkno = 0;
2420         u32 first_hash, num_clusters = 0;
2421         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2422
2423         if (le16_to_cpu(el->l_next_free_rec) == 0)
2424                 return -ENODATA;
2425
2426         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2427              name, name_hash, name_index);
2428
2429         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2430                                   &num_clusters, el);
2431         if (ret) {
2432                 mlog_errno(ret);
2433                 goto out;
2434         }
2435
2436         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2437
2438         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2439              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
2440              first_hash);
2441
2442         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2443                                       p_blkno, first_hash, num_clusters, xs);
2444
2445 out:
2446         return ret;
2447 }
2448
2449 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2450                                        u64 blkno,
2451                                        u32 clusters,
2452                                        xattr_bucket_func *func,
2453                                        void *para)
2454 {
2455         int i, ret = 0;
2456         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2457         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2458         u32 num_buckets = clusters * bpc;
2459         struct ocfs2_xattr_bucket bucket;
2460
2461         memset(&bucket, 0, sizeof(bucket));
2462
2463         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2464              clusters, (unsigned long long)blkno);
2465
2466         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2467                 ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket,
2468                                         bucket.bu_bhs, 0);
2469                 if (ret) {
2470                         mlog_errno(ret);
2471                         goto out;
2472                 }
2473
2474                 /*
2475                  * The real bucket num in this series of blocks is stored
2476                  * in the 1st bucket.
2477                  */
2478                 if (i == 0)
2479                         num_buckets = le16_to_cpu(bucket_xh(&bucket)->xh_num_buckets);
2480
2481                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
2482                      (unsigned long long)blkno,
2483                      le32_to_cpu(bucket_xh(&bucket)->xh_entries[0].xe_name_hash));
2484                 if (func) {
2485                         ret = func(inode, &bucket, para);
2486                         if (ret) {
2487                                 mlog_errno(ret);
2488                                 break;
2489                         }
2490                 }
2491
2492                 ocfs2_xattr_bucket_relse(inode, &bucket);
2493                 memset(&bucket, 0, sizeof(bucket));
2494         }
2495
2496 out:
2497         ocfs2_xattr_bucket_relse(inode, &bucket);
2498
2499         return ret;
2500 }
2501
2502 struct ocfs2_xattr_tree_list {
2503         char *buffer;
2504         size_t buffer_size;
2505         size_t result;
2506 };
2507
2508 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2509                                              struct ocfs2_xattr_header *xh,
2510                                              int index,
2511                                              int *block_off,
2512                                              int *new_offset)
2513 {
2514         u16 name_offset;
2515
2516         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2517                 return -EINVAL;
2518
2519         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2520
2521         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2522         *new_offset = name_offset % inode->i_sb->s_blocksize;
2523
2524         return 0;
2525 }
2526
2527 static int ocfs2_list_xattr_bucket(struct inode *inode,
2528                                    struct ocfs2_xattr_bucket *bucket,
2529                                    void *para)
2530 {
2531         int ret = 0, type;
2532         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2533         int i, block_off, new_offset;
2534         const char *prefix, *name;
2535
2536         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
2537                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
2538                 type = ocfs2_xattr_get_type(entry);
2539                 prefix = ocfs2_xattr_prefix(type);
2540
2541                 if (prefix) {
2542                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2543                                                                 bucket_xh(bucket),
2544                                                                 i,
2545                                                                 &block_off,
2546                                                                 &new_offset);
2547                         if (ret)
2548                                 break;
2549
2550                         name = (const char *)bucket_block(bucket, block_off) +
2551                                 new_offset;
2552                         ret = ocfs2_xattr_list_entry(xl->buffer,
2553                                                      xl->buffer_size,
2554                                                      &xl->result,
2555                                                      prefix, name,
2556                                                      entry->xe_name_len);
2557                         if (ret)
2558                                 break;
2559                 }
2560         }
2561
2562         return ret;
2563 }
2564
2565 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2566                                              struct ocfs2_xattr_tree_root *xt,
2567                                              char *buffer,
2568                                              size_t buffer_size)
2569 {
2570         struct ocfs2_extent_list *el = &xt->xt_list;
2571         int ret = 0;
2572         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2573         u64 p_blkno = 0;
2574         struct ocfs2_xattr_tree_list xl = {
2575                 .buffer = buffer,
2576                 .buffer_size = buffer_size,
2577                 .result = 0,
2578         };
2579
2580         if (le16_to_cpu(el->l_next_free_rec) == 0)
2581                 return 0;
2582
2583         while (name_hash > 0) {
2584                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2585                                           &e_cpos, &num_clusters, el);
2586                 if (ret) {
2587                         mlog_errno(ret);
2588                         goto out;
2589                 }
2590
2591                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2592                                                   ocfs2_list_xattr_bucket,
2593                                                   &xl);
2594                 if (ret) {
2595                         mlog_errno(ret);
2596                         goto out;
2597                 }
2598
2599                 if (e_cpos == 0)
2600                         break;
2601
2602                 name_hash = e_cpos - 1;
2603         }
2604
2605         ret = xl.result;
2606 out:
2607         return ret;
2608 }
2609
2610 static int cmp_xe(const void *a, const void *b)
2611 {
2612         const struct ocfs2_xattr_entry *l = a, *r = b;
2613         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2614         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2615
2616         if (l_hash > r_hash)
2617                 return 1;
2618         if (l_hash < r_hash)
2619                 return -1;
2620         return 0;
2621 }
2622
2623 static void swap_xe(void *a, void *b, int size)
2624 {
2625         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2626
2627         tmp = *l;
2628         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2629         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2630 }
2631
2632 /*
2633  * When the ocfs2_xattr_block is filled up, new bucket will be created
2634  * and all the xattr entries will be moved to the new bucket.
2635  * Note: we need to sort the entries since they are not saved in order
2636  * in the ocfs2_xattr_block.
2637  */
2638 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2639                                            struct buffer_head *xb_bh,
2640                                            struct buffer_head *xh_bh,
2641                                            struct buffer_head *data_bh)
2642 {
2643         int i, blocksize = inode->i_sb->s_blocksize;
2644         u16 offset, size, off_change;
2645         struct ocfs2_xattr_entry *xe;
2646         struct ocfs2_xattr_block *xb =
2647                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2648         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2649         struct ocfs2_xattr_header *xh =
2650                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2651         u16 count = le16_to_cpu(xb_xh->xh_count);
2652         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2653
2654         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2655              (unsigned long long)xb_bh->b_blocknr,
2656              (unsigned long long)xh_bh->b_blocknr);
2657
2658         memset(xh_bh->b_data, 0, blocksize);
2659         if (data_bh)
2660                 memset(data_bh->b_data, 0, blocksize);
2661         /*
2662          * Since the xe_name_offset is based on ocfs2_xattr_header,
2663          * there is a offset change corresponding to the change of
2664          * ocfs2_xattr_header's position.
2665          */
2666         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2667         xe = &xb_xh->xh_entries[count - 1];
2668         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2669         size = blocksize - offset;
2670
2671         /* copy all the names and values. */
2672         if (data_bh)
2673                 target = data_bh->b_data;
2674         memcpy(target + offset, src + offset, size);
2675
2676         /* Init new header now. */
2677         xh->xh_count = xb_xh->xh_count;
2678         xh->xh_num_buckets = cpu_to_le16(1);
2679         xh->xh_name_value_len = cpu_to_le16(size);
2680         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2681
2682         /* copy all the entries. */
2683         target = xh_bh->b_data;
2684         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2685         size = count * sizeof(struct ocfs2_xattr_entry);
2686         memcpy(target + offset, (char *)xb_xh + offset, size);
2687
2688         /* Change the xe offset for all the xe because of the move. */
2689         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2690                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2691         for (i = 0; i < count; i++)
2692                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2693
2694         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2695              offset, size, off_change);
2696
2697         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2698              cmp_xe, swap_xe);
2699 }
2700
2701 /*
2702  * After we move xattr from block to index btree, we have to
2703  * update ocfs2_xattr_search to the new xe and base.
2704  *
2705  * When the entry is in xattr block, xattr_bh indicates the storage place.
2706  * While if the entry is in index b-tree, "bucket" indicates the
2707  * real place of the xattr.
2708  */
2709 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2710                                            struct ocfs2_xattr_search *xs,
2711                                            struct buffer_head *old_bh,
2712                                            struct buffer_head *new_bh)
2713 {
2714         int ret = 0;
2715         char *buf = old_bh->b_data;
2716         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2717         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2718         int i, blocksize = inode->i_sb->s_blocksize;
2719         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2720
2721         xs->bucket.bu_bhs[0] = new_bh;
2722         get_bh(new_bh);
2723         xs->header = bucket_xh(&xs->bucket);
2724
2725         xs->base = new_bh->b_data;
2726         xs->end = xs->base + inode->i_sb->s_blocksize;
2727
2728         if (!xs->not_found) {
2729                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2730                         ret = ocfs2_read_blocks(inode,
2731                                         bucket_blkno(&xs->bucket) + 1,
2732                                         blk_per_bucket - 1, &xs->bucket.bu_bhs[1],
2733                                         0);
2734                         if (ret) {
2735                                 mlog_errno(ret);
2736                                 return ret;
2737                         }
2738
2739                 }
2740                 i = xs->here - old_xh->xh_entries;
2741                 xs->here = &xs->header->xh_entries[i];
2742         }
2743
2744         return ret;
2745 }
2746
2747 static int ocfs2_xattr_create_index_block(struct inode *inode,
2748                                           struct ocfs2_xattr_search *xs)
2749 {
2750         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2751         u32 bit_off, len;
2752         u64 blkno;
2753         handle_t *handle;
2754         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2755         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2756         struct ocfs2_alloc_context *data_ac;
2757         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2758         struct buffer_head *xb_bh = xs->xattr_bh;
2759         struct ocfs2_xattr_block *xb =
2760                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2761         struct ocfs2_xattr_tree_root *xr;
2762         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2763         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2764
2765         mlog(0, "create xattr index block for %llu\n",
2766              (unsigned long long)xb_bh->b_blocknr);
2767
2768         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2769
2770         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2771         if (ret) {
2772                 mlog_errno(ret);
2773                 goto out;
2774         }
2775
2776         /*
2777          * XXX:
2778          * We can use this lock for now, and maybe move to a dedicated mutex
2779          * if performance becomes a problem later.
2780          */
2781         down_write(&oi->ip_alloc_sem);
2782
2783         /*
2784          * 3 more credits, one for xattr block update, one for the 1st block
2785          * of the new xattr bucket and one for the value/data.
2786          */
2787         credits += 3;
2788         handle = ocfs2_start_trans(osb, credits);
2789         if (IS_ERR(handle)) {
2790                 ret = PTR_ERR(handle);
2791                 mlog_errno(ret);
2792                 goto out_sem;
2793         }
2794
2795         ret = ocfs2_journal_access(handle, inode, xb_bh,
2796                                    OCFS2_JOURNAL_ACCESS_WRITE);
2797         if (ret) {
2798                 mlog_errno(ret);
2799                 goto out_commit;
2800         }
2801
2802         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2803         if (ret) {
2804                 mlog_errno(ret);
2805                 goto out_commit;
2806         }
2807
2808         /*
2809          * The bucket may spread in many blocks, and
2810          * we will only touch the 1st block and the last block
2811          * in the whole bucket(one for entry and one for data).
2812          */
2813         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2814
2815         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
2816              (unsigned long long)blkno);
2817
2818         xh_bh = sb_getblk(inode->i_sb, blkno);
2819         if (!xh_bh) {
2820                 ret = -EIO;
2821                 mlog_errno(ret);
2822                 goto out_commit;
2823         }
2824
2825         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2826
2827         ret = ocfs2_journal_access(handle, inode, xh_bh,
2828                                    OCFS2_JOURNAL_ACCESS_CREATE);
2829         if (ret) {
2830                 mlog_errno(ret);
2831                 goto out_commit;
2832         }
2833
2834         if (bpb > 1) {
2835                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2836                 if (!data_bh) {
2837                         ret = -EIO;
2838                         mlog_errno(ret);
2839                         goto out_commit;
2840                 }
2841
2842                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2843
2844                 ret = ocfs2_journal_access(handle, inode, data_bh,
2845                                            OCFS2_JOURNAL_ACCESS_CREATE);
2846                 if (ret) {
2847                         mlog_errno(ret);
2848                         goto out_commit;
2849                 }
2850         }
2851
2852         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2853
2854         ocfs2_journal_dirty(handle, xh_bh);
2855         if (data_bh)
2856                 ocfs2_journal_dirty(handle, data_bh);
2857
2858         ret = ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2859         if (ret) {
2860                 mlog_errno(ret);
2861                 goto out_commit;
2862         }
2863
2864         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2865         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2866                offsetof(struct ocfs2_xattr_block, xb_attrs));
2867
2868         xr = &xb->xb_attrs.xb_root;
2869         xr->xt_clusters = cpu_to_le32(1);
2870         xr->xt_last_eb_blk = 0;
2871         xr->xt_list.l_tree_depth = 0;
2872         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2873         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2874
2875         xr->xt_list.l_recs[0].e_cpos = 0;
2876         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2877         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2878
2879         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2880
2881         ret = ocfs2_journal_dirty(handle, xb_bh);
2882         if (ret) {
2883                 mlog_errno(ret);
2884                 goto out_commit;
2885         }
2886
2887 out_commit:
2888         ocfs2_commit_trans(osb, handle);
2889
2890 out_sem:
2891         up_write(&oi->ip_alloc_sem);
2892
2893 out:
2894         if (data_ac)
2895                 ocfs2_free_alloc_context(data_ac);
2896
2897         brelse(xh_bh);
2898         brelse(data_bh);
2899
2900         return ret;
2901 }
2902
2903 static int cmp_xe_offset(const void *a, const void *b)
2904 {
2905         const struct ocfs2_xattr_entry *l = a, *r = b;
2906         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2907         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2908
2909         if (l_name_offset < r_name_offset)
2910                 return 1;
2911         if (l_name_offset > r_name_offset)
2912                 return -1;
2913         return 0;
2914 }
2915
2916 /*
2917  * defrag a xattr bucket if we find that the bucket has some
2918  * holes beteen name/value pairs.
2919  * We will move all the name/value pairs to the end of the bucket
2920  * so that we can spare some space for insertion.
2921  */
2922 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2923                                      struct ocfs2_xattr_bucket *bucket)
2924 {
2925         int ret, i;
2926         size_t end, offset, len, value_len;
2927         struct ocfs2_xattr_header *xh;
2928         char *entries, *buf, *bucket_buf = NULL;
2929         u64 blkno = bucket_blkno(bucket);
2930         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2931         u16 xh_free_start;
2932         size_t blocksize = inode->i_sb->s_blocksize;
2933         handle_t *handle;
2934         struct buffer_head **bhs;
2935         struct ocfs2_xattr_entry *xe;
2936
2937         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2938                         GFP_NOFS);
2939         if (!bhs)
2940                 return -ENOMEM;
2941
2942         ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket, bhs, 0);
2943         if (ret)
2944                 goto out;
2945
2946         /*
2947          * In order to make the operation more efficient and generic,
2948          * we copy all the blocks into a contiguous memory and do the
2949          * defragment there, so if anything is error, we will not touch
2950          * the real block.
2951          */
2952         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2953         if (!bucket_buf) {
2954                 ret = -EIO;
2955                 goto out;
2956         }
2957
2958         buf = bucket_buf;
2959         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2960                 memcpy(buf, bhs[i]->b_data, blocksize);
2961
2962         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2963         if (IS_ERR(handle)) {
2964                 ret = PTR_ERR(handle);
2965                 handle = NULL;
2966                 mlog_errno(ret);
2967                 goto out;
2968         }
2969
2970         for (i = 0; i < blk_per_bucket; i++) {
2971                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2972                                            OCFS2_JOURNAL_ACCESS_WRITE);
2973                 if (ret < 0) {
2974                         mlog_errno(ret);
2975                         goto commit;
2976                 }
2977         }
2978
2979         xh = (struct ocfs2_xattr_header *)bucket_buf;
2980         entries = (char *)xh->xh_entries;
2981         xh_free_start = le16_to_cpu(xh->xh_free_start);
2982
2983         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2984              "xh_free_start = %u, xh_name_value_len = %u.\n",
2985              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
2986              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
2987
2988         /*
2989          * sort all the entries by their offset.
2990          * the largest will be the first, so that we can
2991          * move them to the end one by one.
2992          */
2993         sort(entries, le16_to_cpu(xh->xh_count),
2994              sizeof(struct ocfs2_xattr_entry),
2995              cmp_xe_offset, swap_xe);
2996
2997         /* Move all name/values to the end of the bucket. */
2998         xe = xh->xh_entries;
2999         end = OCFS2_XATTR_BUCKET_SIZE;
3000         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3001                 offset = le16_to_cpu(xe->xe_name_offset);
3002                 if (ocfs2_xattr_is_local(xe))
3003                         value_len = OCFS2_XATTR_SIZE(
3004                                         le64_to_cpu(xe->xe_value_size));
3005                 else
3006                         value_len = OCFS2_XATTR_ROOT_SIZE;
3007                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3008
3009                 /*
3010                  * We must make sure that the name/value pair
3011                  * exist in the same block. So adjust end to
3012                  * the previous block end if needed.
3013                  */
3014                 if (((end - len) / blocksize !=
3015                         (end - 1) / blocksize))
3016                         end = end - end % blocksize;
3017
3018                 if (end > offset + len) {
3019                         memmove(bucket_buf + end - len,
3020                                 bucket_buf + offset, len);
3021                         xe->xe_name_offset = cpu_to_le16(end - len);
3022                 }
3023
3024                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3025                                 "bucket %llu\n", (unsigned long long)blkno);
3026
3027                 end -= len;
3028         }
3029
3030         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3031                         "bucket %llu\n", (unsigned long long)blkno);
3032
3033         if (xh_free_start == end)
3034                 goto commit;
3035
3036         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3037         xh->xh_free_start = cpu_to_le16(end);
3038
3039         /* sort the entries by their name_hash. */
3040         sort(entries, le16_to_cpu(xh->xh_count),
3041              sizeof(struct ocfs2_xattr_entry),
3042              cmp_xe, swap_xe);
3043
3044         buf = bucket_buf;
3045         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3046                 memcpy(bhs[i]->b_data, buf, blocksize);
3047                 ocfs2_journal_dirty(handle, bhs[i]);
3048         }
3049
3050 commit:
3051         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3052 out:
3053
3054         if (bhs) {
3055                 for (i = 0; i < blk_per_bucket; i++)
3056                         brelse(bhs[i]);
3057         }
3058         kfree(bhs);
3059
3060         kfree(bucket_buf);
3061         return ret;
3062 }
3063
3064 /*
3065  * Move half nums of the xattr bucket in the previous cluster to this new
3066  * cluster. We only touch the last cluster of the previous extend record.
3067  *
3068  * first_bh is the first buffer_head of a series of bucket in the same
3069  * extent rec and header_bh is the header of one bucket in this cluster.
3070  * They will be updated if we move the data header_bh contains to the new
3071  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3072  */
3073 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3074                                                handle_t *handle,
3075                                                struct buffer_head **first_bh,
3076                                                struct buffer_head **header_bh,
3077                                                u64 new_blkno,
3078                                                u64 prev_blkno,
3079                                                u32 num_clusters,
3080                                                u32 *first_hash)
3081 {
3082         int i, ret, credits;
3083         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3084         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3085         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3086         int blocksize = inode->i_sb->s_blocksize;
3087         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3088         struct ocfs2_xattr_header *new_xh;
3089         struct ocfs2_xattr_header *xh =
3090                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3091
3092         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3093         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3094
3095         prev_bh = *first_bh;
3096         get_bh(prev_bh);
3097         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3098
3099         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3100
3101         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3102              (unsigned long long)prev_blkno, (unsigned long long)new_blkno);
3103
3104         /*
3105          * We need to update the 1st half of the new cluster and
3106          * 1 more for the update of the 1st bucket of the previous
3107          * extent record.
3108          */
3109         credits = bpc / 2 + 1;
3110         ret = ocfs2_extend_trans(handle, credits);
3111         if (ret) {
3112                 mlog_errno(ret);
3113                 goto out;
3114         }
3115
3116         ret = ocfs2_journal_access(handle, inode, prev_bh,
3117                                    OCFS2_JOURNAL_ACCESS_WRITE);
3118         if (ret) {
3119                 mlog_errno(ret);
3120                 goto out;
3121         }
3122
3123         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3124                 old_bh = new_bh = NULL;
3125                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3126                 if (!new_bh) {
3127                         ret = -EIO;
3128                         mlog_errno(ret);
3129                         goto out;
3130                 }
3131
3132                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3133
3134                 ret = ocfs2_journal_access(handle, inode, new_bh,
3135                                            OCFS2_JOURNAL_ACCESS_CREATE);
3136                 if (ret < 0) {
3137                         mlog_errno(ret);
3138                         brelse(new_bh);
3139                         goto out;
3140                 }
3141
3142                 ret = ocfs2_read_block(inode, prev_blkno, &old_bh);
3143                 if (ret < 0) {
3144                         mlog_errno(ret);
3145                         brelse(new_bh);
3146                         goto out;
3147                 }
3148
3149                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3150
3151                 if (i == 0) {
3152                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3153                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3154
3155                         if (first_hash)
3156                                 *first_hash = le32_to_cpu(
3157                                         new_xh->xh_entries[0].xe_name_hash);
3158                         new_first_bh = new_bh;
3159                         get_bh(new_first_bh);
3160                 }
3161
3162                 ocfs2_journal_dirty(handle, new_bh);
3163
3164                 if (*header_bh == old_bh) {
3165                         brelse(*header_bh);
3166                         *header_bh = new_bh;
3167                         get_bh(*header_bh);
3168
3169                         brelse(*first_bh);
3170                         *first_bh = new_first_bh;
3171                         get_bh(*first_bh);
3172                 }
3173                 brelse(new_bh);
3174                 brelse(old_bh);
3175         }
3176
3177         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3178
3179         ocfs2_journal_dirty(handle, prev_bh);
3180 out:
3181         brelse(prev_bh);
3182         brelse(new_first_bh);
3183         return ret;
3184 }
3185
3186 /*
3187  * Find the suitable pos when we divide a bucket into 2.
3188  * We have to make sure the xattrs with the same hash value exist
3189  * in the same bucket.
3190  *
3191  * If this ocfs2_xattr_header covers more than one hash value, find a
3192  * place where the hash value changes.  Try to find the most even split.
3193  * The most common case is that all entries have different hash values,
3194  * and the first check we make will find a place to split.
3195  */
3196 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3197 {
3198         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3199         int count = le16_to_cpu(xh->xh_count);
3200         int delta, middle = count / 2;
3201
3202         /*
3203          * We start at the middle.  Each step gets farther away in both
3204          * directions.  We therefore hit the change in hash value
3205          * nearest to the middle.  Note that this loop does not execute for
3206          * count < 2.
3207          */
3208         for (delta = 0; delta < middle; delta++) {
3209                 /* Let's check delta earlier than middle */
3210                 if (cmp_xe(&entries[middle - delta - 1],
3211                            &entries[middle - delta]))
3212                         return middle - delta;
3213
3214                 /* For even counts, don't walk off the end */
3215                 if ((middle + delta + 1) == count)
3216                         continue;
3217
3218                 /* Now try delta past middle */
3219                 if (cmp_xe(&entries[middle + delta],
3220                            &entries[middle + delta + 1]))
3221                         return middle + delta + 1;
3222         }
3223
3224         /* Every entry had the same hash */
3225         return count;
3226 }
3227
3228 /*
3229  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3230  * first_hash will record the 1st hash of the new bucket.
3231  *
3232  * Normally half of the xattrs will be moved.  But we have to make
3233  * sure that the xattrs with the same hash value are stored in the
3234  * same bucket. If all the xattrs in this bucket have the same hash
3235  * value, the new bucket will be initialized as an empty one and the
3236  * first_hash will be initialized as (hash_value+1).
3237  */
3238 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3239                                     handle_t *handle,
3240                                     u64 blk,
3241                                     u64 new_blk,
3242                                     u32 *first_hash,
3243                                     int new_bucket_head)
3244 {
3245         int ret, i;
3246         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3247         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3248         struct ocfs2_xattr_bucket s_bucket, t_bucket;
3249         struct ocfs2_xattr_header *xh;
3250         struct ocfs2_xattr_entry *xe;
3251         int blocksize = inode->i_sb->s_blocksize;
3252
3253         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3254              (unsigned long long)blk, (unsigned long long)new_blk);
3255
3256         memset(&s_bucket, 0, sizeof(struct ocfs2_xattr_bucket));
3257         memset(&t_bucket, 0, sizeof(struct ocfs2_xattr_bucket));
3258
3259         ret = ocfs2_read_xattr_bucket(inode, &s_bucket, blk);
3260         if (ret) {
3261                 mlog_errno(ret);
3262                 goto out;
3263         }
3264
3265         ret = ocfs2_xattr_bucket_journal_access(handle, inode, &s_bucket,
3266                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3267         if (ret) {
3268                 mlog_errno(ret);
3269                 goto out;
3270         }
3271
3272         /*
3273          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3274          * there's no need to read it.
3275          */
3276         ret = ocfs2_init_xattr_bucket(inode, &t_bucket, new_blk);
3277         if (ret) {
3278                 mlog_errno(ret);
3279                 goto out;
3280         }
3281
3282         ret = ocfs2_xattr_bucket_journal_access(handle, inode, &t_bucket,
3283                                                 new_bucket_head ?
3284                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3285                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3286         if (ret) {
3287                 mlog_errno(ret);
3288                 goto out;
3289         }
3290
3291         xh = bucket_xh(&s_bucket);
3292         count = le16_to_cpu(xh->xh_count);
3293         start = ocfs2_xattr_find_divide_pos(xh);
3294
3295         if (start == count) {
3296                 xe = &xh->xh_entries[start-1];
3297
3298                 /*
3299                  * initialized a new empty bucket here.
3300                  * The hash value is set as one larger than
3301                  * that of the last entry in the previous bucket.
3302                  */
3303                 for (i = 0; i < blk_per_bucket; i++)
3304                         memset(bucket_block(&t_bucket, i), 0, blocksize);
3305
3306                 xh = bucket_xh(&t_bucket);
3307                 xh->xh_free_start = cpu_to_le16(blocksize);
3308                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3309                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3310
3311                 goto set_num_buckets;
3312         }
3313
3314         /* copy the whole bucket to the new first. */
3315         ocfs2_xattr_bucket_copy_data(inode, &t_bucket, &s_bucket);
3316
3317         /* update the new bucket. */
3318         xh = bucket_xh(&t_bucket);
3319
3320         /*
3321          * Calculate the total name/value len and xh_free_start for
3322          * the old bucket first.
3323          */
3324         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3325         name_value_len = 0;
3326         for (i = 0; i < start; i++) {
3327                 xe = &xh->xh_entries[i];
3328                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3329                 if (ocfs2_xattr_is_local(xe))
3330                         xe_len +=
3331                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3332                 else
3333                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3334                 name_value_len += xe_len;
3335                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3336                         name_offset = le16_to_cpu(xe->xe_name_offset);
3337         }
3338
3339         /*
3340          * Now begin the modification to the new bucket.
3341          *
3342          * In the new bucket, We just move the xattr entry to the beginning
3343          * and don't touch the name/value. So there will be some holes in the
3344          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3345          * called.
3346          */
3347         xe = &xh->xh_entries[start];
3348         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3349         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3350              (int)((char *)xe - (char *)xh),
3351              (int)((char *)xh->xh_entries - (char *)xh));
3352         memmove((char *)xh->xh_entries, (char *)xe, len);
3353         xe = &xh->xh_entries[count - start];
3354         len = sizeof(struct ocfs2_xattr_entry) * start;
3355         memset((char *)xe, 0, len);
3356
3357         le16_add_cpu(&xh->xh_count, -start);
3358         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3359
3360         /* Calculate xh_free_start for the new bucket. */
3361         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3362         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3363                 xe = &xh->xh_entries[i];
3364                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3365                 if (ocfs2_xattr_is_local(xe))
3366                         xe_len +=
3367                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3368                 else
3369                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3370                 if (le16_to_cpu(xe->xe_name_offset) <
3371                     le16_to_cpu(xh->xh_free_start))
3372                         xh->xh_free_start = xe->xe_name_offset;
3373         }
3374
3375 set_num_buckets:
3376         /* set xh->xh_num_buckets for the new xh. */
3377         if (new_bucket_head)
3378                 xh->xh_num_buckets = cpu_to_le16(1);
3379         else
3380                 xh->xh_num_buckets = 0;
3381
3382         ocfs2_xattr_bucket_journal_dirty(handle, inode, &t_bucket);
3383
3384         /* store the first_hash of the new bucket. */
3385         if (first_hash)
3386                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3387
3388         /*
3389          * Now only update the 1st block of the old bucket.  If we
3390          * just added a new empty bucket, there is no need to modify
3391          * it.
3392          */
3393         if (start == count)
3394                 goto out;
3395
3396         xh = bucket_xh(&s_bucket);
3397         memset(&xh->xh_entries[start], 0,
3398                sizeof(struct ocfs2_xattr_entry) * (count - start));
3399         xh->xh_count = cpu_to_le16(start);
3400         xh->xh_free_start = cpu_to_le16(name_offset);
3401         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3402
3403         ocfs2_xattr_bucket_journal_dirty(handle, inode, &s_bucket);
3404
3405 out:
3406         ocfs2_xattr_bucket_relse(inode, &s_bucket);
3407         ocfs2_xattr_bucket_relse(inode, &t_bucket);
3408
3409         return ret;
3410 }
3411
3412 /*
3413  * Copy xattr from one bucket to another bucket.
3414  *
3415  * The caller must make sure that the journal transaction
3416  * has enough space for journaling.
3417  */
3418 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3419                                  handle_t *handle,
3420                                  u64 s_blkno,
3421                                  u64 t_blkno,
3422                                  int t_is_new)
3423 {
3424         int ret;
3425         struct ocfs2_xattr_bucket s_bucket, t_bucket;
3426
3427         BUG_ON(s_blkno == t_blkno);
3428
3429         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3430              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3431              t_is_new);
3432
3433         memset(&s_bucket, 0, sizeof(struct ocfs2_xattr_bucket));
3434         memset(&t_bucket, 0, sizeof(struct ocfs2_xattr_bucket));
3435
3436         ret = ocfs2_read_xattr_bucket(inode, &s_bucket, s_blkno);
3437         if (ret)
3438                 goto out;
3439
3440         /*
3441          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3442          * there's no need to read it.
3443          */
3444         ret = ocfs2_init_xattr_bucket(inode, &t_bucket, t_blkno);
3445         if (ret)
3446                 goto out;
3447
3448         ret = ocfs2_xattr_bucket_journal_access(handle, inode, &t_bucket,
3449                                                 t_is_new ?
3450                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3451                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3452         if (ret)
3453                 goto out;
3454
3455         ocfs2_xattr_bucket_copy_data(inode, &t_bucket, &s_bucket);
3456         ocfs2_xattr_bucket_journal_dirty(handle, inode, &t_bucket);
3457
3458 out:
3459         ocfs2_xattr_bucket_relse(inode, &s_bucket);
3460         ocfs2_xattr_bucket_relse(inode, &t_bucket);
3461
3462         return ret;
3463 }
3464
3465 /*
3466  * Copy one xattr cluster from src_blk to to_blk.
3467  * The to_blk will become the first bucket header of the cluster, so its
3468  * xh_num_buckets will be initialized as the bucket num in the cluster.
3469  */
3470 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3471                                   handle_t *handle,
3472                                   struct buffer_head *first_bh,
3473                                   u64 src_blk,
3474                                   u64 to_blk,
3475                                   u32 *first_hash)
3476 {
3477         int i, ret, credits;
3478         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3479         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3480         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3481         struct buffer_head *bh = NULL;
3482         struct ocfs2_xattr_header *xh;
3483         u64 to_blk_start = to_blk;
3484
3485         mlog(0, "cp xattrs from cluster %llu to %llu\n",
3486              (unsigned long long)src_blk, (unsigned long long)to_blk);
3487
3488         /*
3489          * We need to update the new cluster and 1 more for the update of
3490          * the 1st bucket of the previous extent rec.
3491          */
3492         credits = bpc + 1;
3493         ret = ocfs2_extend_trans(handle, credits);
3494         if (ret) {
3495                 mlog_errno(ret);
3496                 goto out;
3497         }
3498
3499         ret = ocfs2_journal_access(handle, inode, first_bh,
3500                                    OCFS2_JOURNAL_ACCESS_WRITE);
3501         if (ret) {
3502                 mlog_errno(ret);
3503                 goto out;
3504         }
3505
3506         for (i = 0; i < num_buckets; i++) {
3507                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3508                                             src_blk, to_blk, 1);
3509                 if (ret) {
3510                         mlog_errno(ret);
3511                         goto out;
3512                 }
3513
3514                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3515                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3516         }
3517
3518         /* update the old bucket header. */
3519         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3520         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3521
3522         ocfs2_journal_dirty(handle, first_bh);
3523
3524         /* update the new bucket header. */
3525         ret = ocfs2_read_block(inode, to_blk_start, &bh);
3526         if (ret < 0) {
3527                 mlog_errno(ret);
3528                 goto out;
3529         }
3530
3531         ret = ocfs2_journal_access(handle, inode, bh,
3532                                    OCFS2_JOURNAL_ACCESS_WRITE);
3533         if (ret) {
3534                 mlog_errno(ret);
3535                 goto out;
3536         }
3537
3538         xh = (struct ocfs2_xattr_header *)bh->b_data;
3539         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3540
3541         ocfs2_journal_dirty(handle, bh);
3542
3543         if (first_hash)
3544                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3545 out:
3546         brelse(bh);
3547         return ret;
3548 }
3549
3550 /*
3551  * Move some xattrs in this cluster to the new cluster.
3552  * This function should only be called when bucket size == cluster size.
3553  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3554  */
3555 static int ocfs2_divide_xattr_cluster(struct inode *inode,
3556                                       handle_t *handle,
3557                                       u64 prev_blk,
3558                                       u64 new_blk,
3559                                       u32 *first_hash)
3560 {
3561         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3562         int ret, credits = 2 * blk_per_bucket;
3563
3564         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3565
3566         ret = ocfs2_extend_trans(handle, credits);
3567         if (ret) {
3568                 mlog_errno(ret);
3569                 return ret;
3570         }
3571
3572         /* Move half of the xattr in start_blk to the next bucket. */
3573         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
3574                                           new_blk, first_hash, 1);
3575 }
3576
3577 /*
3578  * Move some xattrs from the old cluster to the new one since they are not
3579  * contiguous in ocfs2 xattr tree.
3580  *
3581  * new_blk starts a new separate cluster, and we will move some xattrs from
3582  * prev_blk to it. v_start will be set as the first name hash value in this
3583  * new cluster so that it can be used as e_cpos during tree insertion and
3584  * don't collide with our original b-tree operations. first_bh and header_bh
3585  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3586  * to extend the insert bucket.
3587  *
3588  * The problem is how much xattr should we move to the new one and when should
3589  * we update first_bh and header_bh?
3590  * 1. If cluster size > bucket size, that means the previous cluster has more
3591  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3592  *    update the first_bh and header_bh if the insert bucket has been moved
3593  *    to the new cluster.
3594  * 2. If cluster_size == bucket_size:
3595  *    a) If the previous extent rec has more than one cluster and the insert
3596  *       place isn't in the last cluster, copy the entire last cluster to the
3597  *       new one. This time, we don't need to upate the first_bh and header_bh
3598  *       since they will not be moved into the new cluster.
3599  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3600  *       the new one. And we set the extend flag to zero if the insert place is
3601  *       moved into the new allocated cluster since no extend is needed.
3602  */
3603 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3604                                             handle_t *handle,
3605                                             struct buffer_head **first_bh,
3606                                             struct buffer_head **header_bh,
3607                                             u64 new_blk,
3608                                             u64 prev_blk,
3609                                             u32 prev_clusters,
3610                                             u32 *v_start,
3611                                             int *extend)
3612 {
3613         int ret = 0;
3614         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3615
3616         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3617              (unsigned long long)prev_blk, prev_clusters,
3618              (unsigned long long)new_blk);
3619
3620         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3621                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3622                                                           handle,
3623                                                           first_bh,
3624                                                           header_bh,
3625                                                           new_blk,
3626                                                           prev_blk,
3627                                                           prev_clusters,
3628                                                           v_start);
3629         else {
3630                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3631
3632                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3633                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3634                                                      last_blk, new_blk,
3635                                                      v_start);
3636                 else {
3637                         ret = ocfs2_divide_xattr_cluster(inode, handle,
3638                                                          last_blk, new_blk,
3639                                                          v_start);
3640
3641                         if ((*header_bh)->b_blocknr == last_blk && extend)
3642                                 *extend = 0;
3643                 }
3644         }
3645
3646         return ret;
3647 }
3648
3649 /*
3650  * Add a new cluster for xattr storage.
3651  *
3652  * If the new cluster is contiguous with the previous one, it will be
3653  * appended to the same extent record, and num_clusters will be updated.
3654  * If not, we will insert a new extent for it and move some xattrs in
3655  * the last cluster into the new allocated one.
3656  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3657  * lose the benefits of hashing because we'll have to search large leaves.
3658  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3659  * if it's bigger).
3660  *
3661  * first_bh is the first block of the previous extent rec and header_bh
3662  * indicates the bucket we will insert the new xattrs. They will be updated
3663  * when the header_bh is moved into the new cluster.
3664  */
3665 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3666                                        struct buffer_head *root_bh,
3667                                        struct buffer_head **first_bh,
3668                                        struct buffer_head **header_bh,
3669                                        u32 *num_clusters,
3670                                        u32 prev_cpos,
3671                                        u64 prev_blkno,
3672                                        int *extend)
3673 {
3674         int ret, credits;
3675         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3676         u32 prev_clusters = *num_clusters;
3677         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3678         u64 block;
3679         handle_t *handle = NULL;
3680         struct ocfs2_alloc_context *data_ac = NULL;
3681         struct ocfs2_alloc_context *meta_ac = NULL;
3682         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3683         struct ocfs2_extent_tree et;
3684
3685         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3686              "previous xattr blkno = %llu\n",
3687              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3688              prev_cpos, (unsigned long long)prev_blkno);
3689
3690         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3691
3692         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3693                                     &data_ac, &meta_ac);
3694         if (ret) {
3695                 mlog_errno(ret);
3696                 goto leave;
3697         }
3698
3699         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3700                                             clusters_to_add);
3701         handle = ocfs2_start_trans(osb, credits);
3702         if (IS_ERR(handle)) {
3703                 ret = PTR_ERR(handle);
3704                 handle = NULL;
3705                 mlog_errno(ret);
3706                 goto leave;
3707         }
3708
3709         ret = ocfs2_journal_access(handle, inode, root_bh,
3710                                    OCFS2_JOURNAL_ACCESS_WRITE);
3711         if (ret < 0) {
3712                 mlog_errno(ret);
3713                 goto leave;
3714         }
3715
3716         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3717                                      clusters_to_add, &bit_off, &num_bits);
3718         if (ret < 0) {
3719                 if (ret != -ENOSPC)
3720                         mlog_errno(ret);
3721                 goto leave;
3722         }
3723
3724         BUG_ON(num_bits > clusters_to_add);
3725
3726         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3727         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3728              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3729
3730         if (prev_blkno + prev_clusters * bpc == block &&
3731             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3732              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3733                 /*
3734                  * If this cluster is contiguous with the old one and
3735                  * adding this new cluster, we don't surpass the limit of
3736                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3737                  * initialized and used like other buckets in the previous
3738                  * cluster.
3739                  * So add it as a contiguous one. The caller will handle
3740                  * its init process.
3741                  */
3742                 v_start = prev_cpos + prev_clusters;
3743                 *num_clusters = prev_clusters + num_bits;
3744                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3745                      num_bits);
3746         } else {
3747                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3748                                                        handle,
3749                                                        first_bh,
3750                                                        header_bh,
3751                                                        block,
3752                                                        prev_blkno,
3753                                                        prev_clusters,
3754                                                        &v_start,
3755                                                        extend);
3756                 if (ret) {
3757                         mlog_errno(ret);
3758                         goto leave;
3759                 }
3760         }
3761
3762         if (handle->h_buffer_credits < credits) {
3763                 /*
3764                  * The journal has been restarted before, and don't
3765                  * have enough space for the insertion, so extend it
3766                  * here.
3767                  */
3768                 ret = ocfs2_extend_trans(handle, credits);
3769                 if (ret) {
3770                         mlog_errno(ret);
3771                         goto leave;
3772                 }
3773         }
3774         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3775              num_bits, (unsigned long long)block, v_start);
3776         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3777                                   num_bits, 0, meta_ac);
3778         if (ret < 0) {
3779                 mlog_errno(ret);
3780                 goto leave;
3781         }
3782
3783         ret = ocfs2_journal_dirty(handle, root_bh);
3784         if (ret < 0) {
3785                 mlog_errno(ret);
3786                 goto leave;
3787         }
3788
3789 leave:
3790         if (handle)
3791                 ocfs2_commit_trans(osb, handle);
3792         if (data_ac)
3793                 ocfs2_free_alloc_context(data_ac);
3794         if (meta_ac)
3795                 ocfs2_free_alloc_context(meta_ac);
3796
3797         return ret;
3798 }
3799
3800 /*
3801  * Extend a new xattr bucket and move xattrs to the end one by one until
3802  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3803  */
3804 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3805                                      struct buffer_head *first_bh,
3806                                      struct buffer_head *start_bh,
3807                                      u32 num_clusters)
3808 {
3809         int ret, credits;
3810         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3811         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3812         u64 start_blk = start_bh->b_blocknr, end_blk;
3813         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3814         handle_t *handle;
3815         struct ocfs2_xattr_header *first_xh =
3816                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3817         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3818
3819         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3820              "from %llu, len = %u\n", (unsigned long long)start_blk,
3821              (unsigned long long)first_bh->b_blocknr, num_clusters);
3822
3823         BUG_ON(bucket >= num_buckets);
3824
3825         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3826
3827         /*
3828          * We will touch all the buckets after the start_bh(include it).
3829          * Then we add one more bucket.
3830          */
3831         credits = end_blk - start_blk + 3 * blk_per_bucket + 1;
3832         handle = ocfs2_start_trans(osb, credits);
3833         if (IS_ERR(handle)) {
3834                 ret = PTR_ERR(handle);
3835                 handle = NULL;
3836                 mlog_errno(ret);
3837                 goto out;
3838         }
3839
3840         ret = ocfs2_journal_access(handle, inode, first_bh,
3841                                    OCFS2_JOURNAL_ACCESS_WRITE);
3842         if (ret) {
3843                 mlog_errno(ret);
3844                 goto commit;
3845         }
3846
3847         while (end_blk != start_blk) {
3848                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3849                                             end_blk + blk_per_bucket, 0);
3850                 if (ret)
3851                         goto commit;
3852                 end_blk -= blk_per_bucket;
3853         }
3854
3855         /* Move half of the xattr in start_blk to the next bucket. */
3856         ret = ocfs2_divide_xattr_bucket(inode, handle, start_blk,
3857                                         start_blk + blk_per_bucket, NULL, 0);
3858
3859         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3860         ocfs2_journal_dirty(handle, first_bh);
3861
3862 commit:
3863         ocfs2_commit_trans(osb, handle);
3864 out:
3865         return ret;
3866 }
3867
3868 /*
3869  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3870  * xb_bh is the ocfs2_xattr_block.
3871  * We will move all the buckets starting from header_bh to the next place. As
3872  * for this one, half num of its xattrs will be moved to the next one.
3873  *
3874  * We will allocate a new cluster if current cluster is full and adjust
3875  * header_bh and first_bh if the insert place is moved to the new cluster.
3876  */
3877 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3878                                       struct buffer_head *xb_bh,
3879                                       struct buffer_head *header_bh)
3880 {
3881         struct ocfs2_xattr_header *first_xh = NULL;
3882         struct buffer_head *first_bh = NULL;
3883         struct ocfs2_xattr_block *xb =
3884                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3885         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3886         struct ocfs2_extent_list *el = &xb_root->xt_list;
3887         struct ocfs2_xattr_header *xh =
3888                         (struct ocfs2_xattr_header *)header_bh->b_data;
3889         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3890         struct super_block *sb = inode->i_sb;
3891         struct ocfs2_super *osb = OCFS2_SB(sb);
3892         int ret, num_buckets, extend = 1;
3893         u64 p_blkno;
3894         u32 e_cpos, num_clusters;
3895
3896         mlog(0, "Add new xattr bucket starting form %llu\n",
3897              (unsigned long long)header_bh->b_blocknr);
3898
3899         /*
3900          * Add refrence for header_bh here because it may be
3901          * changed in ocfs2_add_new_xattr_cluster and we need
3902          * to free it in the end.
3903          */
3904         get_bh(header_bh);
3905
3906         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3907                                   &num_clusters, el);
3908         if (ret) {
3909                 mlog_errno(ret);
3910                 goto out;
3911         }
3912
3913         ret = ocfs2_read_block(inode, p_blkno, &first_bh);
3914         if (ret) {
3915                 mlog_errno(ret);
3916                 goto out;
3917         }
3918
3919         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3920         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3921
3922         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3923                 ret = ocfs2_add_new_xattr_cluster(inode,
3924                                                   xb_bh,
3925                                                   &first_bh,
3926                                                   &header_bh,
3927                                                   &num_clusters,
3928                                                   e_cpos,
3929                                                   p_blkno,
3930                                                   &extend);
3931                 if (ret) {
3932                         mlog_errno(ret);
3933                         goto out;
3934                 }
3935         }
3936
3937         if (extend)
3938                 ret = ocfs2_extend_xattr_bucket(inode,
3939                                                 first_bh,
3940                                                 header_bh,
3941                                                 num_clusters);
3942         if (ret)
3943                 mlog_errno(ret);
3944 out:
3945         brelse(first_bh);
3946         brelse(header_bh);
3947         return ret;
3948 }
3949
3950 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3951                                         struct ocfs2_xattr_bucket *bucket,
3952                                         int offs)
3953 {
3954         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3955
3956         offs = offs % inode->i_sb->s_blocksize;
3957         return bucket_block(bucket, block_off) + offs;
3958 }
3959
3960 /*
3961  * Handle the normal xattr set, including replace, delete and new.
3962  *
3963  * Note: "local" indicates the real data's locality. So we can't
3964  * just its bucket locality by its length.
3965  */
3966 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3967                                          struct ocfs2_xattr_info *xi,
3968                                          struct ocfs2_xattr_search *xs,
3969                                          u32 name_hash,
3970                                          int local)
3971 {
3972         struct ocfs2_xattr_entry *last, *xe;
3973         int name_len = strlen(xi->name);
3974         struct ocfs2_xattr_header *xh = xs->header;
3975         u16 count = le16_to_cpu(xh->xh_count), start;
3976         size_t blocksize = inode->i_sb->s_blocksize;
3977         char *val;
3978         size_t offs, size, new_size;
3979
3980         last = &xh->xh_entries[count];
3981         if (!xs->not_found) {
3982                 xe = xs->here;
3983                 offs = le16_to_cpu(xe->xe_name_offset);
3984                 if (ocfs2_xattr_is_local(xe))
3985                         size = OCFS2_XATTR_SIZE(name_len) +
3986                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3987                 else
3988                         size = OCFS2_XATTR_SIZE(name_len) +
3989                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3990
3991                 /*
3992                  * If the new value will be stored outside, xi->value has been
3993                  * initalized as an empty ocfs2_xattr_value_root, and the same
3994                  * goes with xi->value_len, so we can set new_size safely here.
3995                  * See ocfs2_xattr_set_in_bucket.
3996                  */
3997                 new_size = OCFS2_XATTR_SIZE(name_len) +
3998                            OCFS2_XATTR_SIZE(xi->value_len);
3999
4000                 le16_add_cpu(&xh->xh_name_value_len, -size);
4001                 if (xi->value) {
4002                         if (new_size > size)
4003                                 goto set_new_name_value;
4004
4005                         /* Now replace the old value with new one. */
4006                         if (local)
4007                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4008                         else
4009                                 xe->xe_value_size = 0;
4010
4011                         val = ocfs2_xattr_bucket_get_val(inode,
4012                                                          &xs->bucket, offs);
4013                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4014                                size - OCFS2_XATTR_SIZE(name_len));
4015                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4016                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4017                                        xi->value, xi->value_len);
4018
4019                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4020                         ocfs2_xattr_set_local(xe, local);
4021                         return;
4022                 } else {
4023                         /*
4024                          * Remove the old entry if there is more than one.
4025                          * We don't remove the last entry so that we can
4026                          * use it to indicate the hash value of the empty
4027                          * bucket.
4028                          */
4029                         last -= 1;
4030                         le16_add_cpu(&xh->xh_count, -1);
4031                         if (xh->xh_count) {
4032                                 memmove(xe, xe + 1,
4033                                         (void *)last - (void *)xe);
4034                                 memset(last, 0,
4035                                        sizeof(struct ocfs2_xattr_entry));
4036                         } else
4037                                 xh->xh_free_start =
4038                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4039
4040                         return;
4041                 }
4042         } else {
4043                 /* find a new entry for insert. */
4044                 int low = 0, high = count - 1, tmp;
4045                 struct ocfs2_xattr_entry *tmp_xe;
4046
4047                 while (low <= high && count) {
4048                         tmp = (low + high) / 2;
4049                         tmp_xe = &xh->xh_entries[tmp];
4050
4051                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4052                                 low = tmp + 1;
4053                         else if (name_hash <
4054                                  le32_to_cpu(tmp_xe->xe_name_hash))
4055                                 high = tmp - 1;
4056                         else {
4057                                 low = tmp;
4058                                 break;
4059                         }
4060                 }
4061
4062                 xe = &xh->xh_entries[low];
4063                 if (low != count)
4064                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4065
4066                 le16_add_cpu(&xh->xh_count, 1);
4067                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4068                 xe->xe_name_hash = cpu_to_le32(name_hash);
4069                 xe->xe_name_len = name_len;
4070                 ocfs2_xattr_set_type(xe, xi->name_index);
4071         }
4072
4073 set_new_name_value:
4074         /* Insert the new name+value. */
4075         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4076
4077         /*
4078          * We must make sure that the name/value pair
4079          * exists in the same block.
4080          */
4081         offs = le16_to_cpu(xh->xh_free_start);
4082         start = offs - size;
4083
4084         if (start >> inode->i_sb->s_blocksize_bits !=
4085             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4086                 offs = offs - offs % blocksize;
4087                 xh->xh_free_start = cpu_to_le16(offs);
4088         }
4089
4090         val = ocfs2_xattr_bucket_get_val(inode,
4091                                          &xs->bucket, offs - size);
4092         xe->xe_name_offset = cpu_to_le16(offs - size);
4093
4094         memset(val, 0, size);
4095         memcpy(val, xi->name, name_len);
4096         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4097
4098         xe->xe_value_size = cpu_to_le64(xi->value_len);
4099         ocfs2_xattr_set_local(xe, local);
4100         xs->here = xe;
4101         le16_add_cpu(&xh->xh_free_start, -size);
4102         le16_add_cpu(&xh->xh_name_value_len, size);
4103
4104         return;
4105 }
4106
4107 /*
4108  * Set the xattr entry in the specified bucket.
4109  * The bucket is indicated by xs->bucket and it should have the enough
4110  * space for the xattr insertion.
4111  */
4112 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4113                                            struct ocfs2_xattr_info *xi,
4114                                            struct ocfs2_xattr_search *xs,
4115                                            u32 name_hash,
4116                                            int local)
4117 {
4118         int ret;
4119         handle_t *handle = NULL;
4120         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4121         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4122
4123         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4124              (unsigned long)xi->value_len, xi->name_index,
4125              (unsigned long long)bucket_blkno(&xs->bucket));
4126
4127         if (!xs->bucket.bu_bhs[1]) {
4128                 ret = ocfs2_read_blocks(inode,
4129                                         bucket_blkno(&xs->bucket) + 1,
4130                                         blk_per_bucket - 1, &xs->bucket.bu_bhs[1],
4131                                         0);
4132                 if (ret) {
4133                         mlog_errno(ret);
4134                         goto out;
4135                 }
4136         }
4137
4138         handle = ocfs2_start_trans(osb, blk_per_bucket);
4139         if (IS_ERR(handle)) {
4140                 ret = PTR_ERR(handle);
4141                 handle = NULL;
4142                 mlog_errno(ret);
4143                 goto out;
4144         }
4145
4146         ret = ocfs2_xattr_bucket_journal_access(handle, inode, &xs->bucket,
4147                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4148         if (ret < 0) {
4149                 mlog_errno(ret);
4150                 goto out;
4151         }
4152
4153         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4154         ocfs2_xattr_bucket_journal_dirty(handle, inode, &xs->bucket);
4155
4156 out:
4157         ocfs2_commit_trans(osb, handle);
4158
4159         return ret;
4160 }
4161
4162 static int ocfs2_xattr_value_update_size(struct inode *inode,
4163                                          struct buffer_head *xe_bh,
4164                                          struct ocfs2_xattr_entry *xe,
4165                                          u64 new_size)
4166 {
4167         int ret;
4168         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4169         handle_t *handle = NULL;
4170
4171         handle = ocfs2_start_trans(osb, 1);
4172         if (IS_ERR(handle)) {
4173                 ret = -ENOMEM;
4174                 mlog_errno(ret);
4175                 goto out;
4176         }
4177
4178         ret = ocfs2_journal_access(handle, inode, xe_bh,
4179                                    OCFS2_JOURNAL_ACCESS_WRITE);
4180         if (ret < 0) {
4181                 mlog_errno(ret);
4182                 goto out_commit;
4183         }
4184
4185         xe->xe_value_size = cpu_to_le64(new_size);
4186
4187         ret = ocfs2_journal_dirty(handle, xe_bh);
4188         if (ret < 0)
4189                 mlog_errno(ret);
4190
4191 out_commit:
4192         ocfs2_commit_trans(osb, handle);
4193 out:
4194         return ret;
4195 }
4196
4197 /*
4198  * Truncate the specified xe_off entry in xattr bucket.
4199  * bucket is indicated by header_bh and len is the new length.
4200  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4201  *
4202  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4203  */
4204 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4205                                              struct buffer_head *header_bh,
4206                                              int xe_off,
4207                                              int len)
4208 {
4209         int ret, offset;
4210         u64 value_blk;
4211         struct buffer_head *value_bh = NULL;
4212         struct ocfs2_xattr_value_root *xv;
4213         struct ocfs2_xattr_entry *xe;
4214         struct ocfs2_xattr_header *xh =
4215                         (struct ocfs2_xattr_header *)header_bh->b_data;
4216         size_t blocksize = inode->i_sb->s_blocksize;
4217
4218         xe = &xh->xh_entries[xe_off];
4219
4220         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4221
4222         offset = le16_to_cpu(xe->xe_name_offset) +
4223                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4224
4225         value_blk = offset / blocksize;
4226
4227         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4228         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4229         value_blk += header_bh->b_blocknr;
4230
4231         ret = ocfs2_read_block(inode, value_blk, &value_bh);
4232         if (ret) {
4233                 mlog_errno(ret);
4234                 goto out;
4235         }
4236
4237         xv = (struct ocfs2_xattr_value_root *)
4238                 (value_bh->b_data + offset % blocksize);
4239
4240         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4241              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4242         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4243         if (ret) {
4244                 mlog_errno(ret);
4245                 goto out;
4246         }
4247
4248         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4249         if (ret) {
4250                 mlog_errno(ret);
4251                 goto out;
4252         }
4253
4254 out:
4255         brelse(value_bh);
4256         return ret;
4257 }
4258
4259 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4260                                                 struct ocfs2_xattr_search *xs,
4261                                                 int len)
4262 {
4263         int ret, offset;
4264         struct ocfs2_xattr_entry *xe = xs->here;
4265         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4266
4267         BUG_ON(!xs->bucket.bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4268
4269         offset = xe - xh->xh_entries;
4270         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bu_bhs[0],
4271                                                 offset, len);
4272         if (ret)
4273                 mlog_errno(ret);
4274
4275         return ret;
4276 }
4277
4278 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4279                                                 struct ocfs2_xattr_search *xs,
4280                                                 char *val,
4281                                                 int value_len)
4282 {
4283         int offset;
4284         struct ocfs2_xattr_value_root *xv;
4285         struct ocfs2_xattr_entry *xe = xs->here;
4286
4287         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4288
4289         offset = le16_to_cpu(xe->xe_name_offset) +
4290                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4291
4292         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4293
4294         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4295 }
4296
4297 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4298                                   struct buffer_head *root_bh,
4299                                   u64 blkno,
4300                                   u32 cpos,
4301                                   u32 len)
4302 {
4303         int ret;
4304         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4305         struct inode *tl_inode = osb->osb_tl_inode;
4306         handle_t *handle;
4307         struct ocfs2_xattr_block *xb =
4308                         (struct ocfs2_xattr_block *)root_bh->b_data;
4309         struct ocfs2_alloc_context *meta_ac = NULL;
4310         struct ocfs2_cached_dealloc_ctxt dealloc;
4311         struct ocfs2_extent_tree et;
4312
4313         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4314
4315         ocfs2_init_dealloc_ctxt(&dealloc);
4316
4317         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4318              cpos, len, (unsigned long long)blkno);
4319
4320         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4321
4322         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4323         if (ret) {
4324                 mlog_errno(ret);
4325                 return ret;
4326         }
4327
4328         mutex_lock(&tl_inode->i_mutex);
4329
4330         if (ocfs2_truncate_log_needs_flush(osb)) {
4331                 ret = __ocfs2_flush_truncate_log(osb);
4332                 if (ret < 0) {
4333                         mlog_errno(ret);
4334                         goto out;
4335                 }
4336         }
4337
4338         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4339         if (IS_ERR(handle)) {
4340                 ret = -ENOMEM;
4341                 mlog_errno(ret);
4342                 goto out;
4343         }
4344
4345         ret = ocfs2_journal_access(handle, inode, root_bh,
4346                                    OCFS2_JOURNAL_ACCESS_WRITE);
4347         if (ret) {
4348                 mlog_errno(ret);
4349                 goto out_commit;
4350         }
4351
4352         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4353                                   &dealloc);
4354         if (ret) {
4355                 mlog_errno(ret);
4356                 goto out_commit;
4357         }
4358
4359         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4360
4361         ret = ocfs2_journal_dirty(handle, root_bh);
4362         if (ret) {
4363                 mlog_errno(ret);
4364                 goto out_commit;
4365         }
4366
4367         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4368         if (ret)
4369                 mlog_errno(ret);
4370
4371 out_commit:
4372         ocfs2_commit_trans(osb, handle);
4373 out:
4374         ocfs2_schedule_truncate_log_flush(osb, 1);
4375
4376         mutex_unlock(&tl_inode->i_mutex);
4377
4378         if (meta_ac)
4379                 ocfs2_free_alloc_context(meta_ac);
4380
4381         ocfs2_run_deallocs(osb, &dealloc);
4382
4383         return ret;
4384 }
4385
4386 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4387                                          struct ocfs2_xattr_search *xs)
4388 {
4389         handle_t *handle = NULL;
4390         struct ocfs2_xattr_header *xh = bucket_xh(&xs->bucket);
4391         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4392                                                 le16_to_cpu(xh->xh_count) - 1];
4393         int ret = 0;
4394
4395         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
4396                                    ocfs2_blocks_per_xattr_bucket(inode->i_sb));
4397         if (IS_ERR(handle)) {
4398                 ret = PTR_ERR(handle);
4399                 mlog_errno(ret);
4400                 return;
4401         }
4402
4403         ret = ocfs2_xattr_bucket_journal_access(handle, inode, &xs->bucket,
4404                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4405         if (ret) {
4406                 mlog_errno(ret);
4407                 goto out_commit;
4408         }
4409
4410         /* Remove the old entry. */
4411         memmove(xs->here, xs->here + 1,
4412                 (void *)last - (void *)xs->here);
4413         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4414         le16_add_cpu(&xh->xh_count, -1);
4415
4416         ocfs2_xattr_bucket_journal_dirty(handle, inode, &xs->bucket);
4417
4418 out_commit:
4419         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4420 }
4421
4422 /*
4423  * Set the xattr name/value in the bucket specified in xs.
4424  *
4425  * As the new value in xi may be stored in the bucket or in an outside cluster,
4426  * we divide the whole process into 3 steps:
4427  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4428  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4429  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4430  * 4. If the clusters for the new outside value can't be allocated, we need
4431  *    to free the xattr we allocated in set.
4432  */
4433 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4434                                      struct ocfs2_xattr_info *xi,
4435                                      struct ocfs2_xattr_search *xs)
4436 {
4437         int ret, local = 1;
4438         size_t value_len;
4439         char *val = (char *)xi->value;
4440         struct ocfs2_xattr_entry *xe = xs->here;
4441         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4442                                               strlen(xi->name));
4443
4444         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4445                 /*
4446                  * We need to truncate the xattr storage first.
4447                  *
4448                  * If both the old and new value are stored to
4449                  * outside block, we only need to truncate
4450                  * the storage and then set the value outside.
4451                  *
4452                  * If the new value should be stored within block,
4453                  * we should free all the outside block first and
4454                  * the modification to the xattr block will be done
4455                  * by following steps.
4456                  */
4457                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4458                         value_len = xi->value_len;
4459                 else
4460                         value_len = 0;
4461
4462                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4463                                                            value_len);
4464                 if (ret)
4465                         goto out;
4466
4467                 if (value_len)
4468                         goto set_value_outside;
4469         }
4470
4471         value_len = xi->value_len;
4472         /* So we have to handle the inside block change now. */
4473         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4474                 /*
4475                  * If the new value will be stored outside of block,
4476                  * initalize a new empty value root and insert it first.
4477                  */
4478                 local = 0;
4479                 xi->value = &def_xv;
4480                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4481         }
4482
4483         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4484         if (ret) {
4485                 mlog_errno(ret);
4486                 goto out;
4487         }
4488
4489         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4490                 goto out;
4491
4492         /* allocate the space now for the outside block storage. */
4493         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4494                                                    value_len);
4495         if (ret) {
4496                 mlog_errno(ret);
4497
4498                 if (xs->not_found) {
4499                         /*
4500                          * We can't allocate enough clusters for outside
4501                          * storage and we have allocated xattr already,
4502                          * so need to remove it.
4503                          */
4504                         ocfs2_xattr_bucket_remove_xs(inode, xs);
4505                 }
4506                 goto out;
4507         }
4508
4509 set_value_outside:
4510         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4511 out:
4512         return ret;
4513 }
4514
4515 /*
4516  * check whether the xattr bucket is filled up with the same hash value.
4517  * If we want to insert the xattr with the same hash, return -ENOSPC.
4518  * If we want to insert a xattr with different hash value, go ahead
4519  * and ocfs2_divide_xattr_bucket will handle this.
4520  */
4521 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4522                                               struct ocfs2_xattr_bucket *bucket,
4523                                               const char *name)
4524 {
4525         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4526         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4527
4528         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4529                 return 0;
4530
4531         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4532             xh->xh_entries[0].xe_name_hash) {
4533                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4534                      "hash = %u\n",
4535                      (unsigned long long)bucket_blkno(bucket),
4536                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4537                 return -ENOSPC;
4538         }
4539
4540         return 0;
4541 }
4542
4543 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4544                                              struct ocfs2_xattr_info *xi,
4545                                              struct ocfs2_xattr_search *xs)
4546 {
4547         struct ocfs2_xattr_header *xh;
4548         struct ocfs2_xattr_entry *xe;
4549         u16 count, header_size, xh_free_start;
4550         int free, max_free, need, old;
4551         size_t value_size = 0, name_len = strlen(xi->name);
4552         size_t blocksize = inode->i_sb->s_blocksize;
4553         int ret, allocation = 0;
4554
4555         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4556
4557 try_again:
4558         xh = xs->header;
4559         count = le16_to_cpu(xh->xh_count);
4560         xh_free_start = le16_to_cpu(xh->xh_free_start);
4561         header_size = sizeof(struct ocfs2_xattr_header) +
4562                         count * sizeof(struct ocfs2_xattr_entry);
4563         max_free = OCFS2_XATTR_BUCKET_SIZE -
4564                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4565
4566         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4567                         "of %u which exceed block size\n",
4568                         (unsigned long long)bucket_blkno(&xs->bucket),
4569                         header_size);
4570
4571         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4572                 value_size = OCFS2_XATTR_ROOT_SIZE;
4573         else if (xi->value)
4574                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4575
4576         if (xs->not_found)
4577                 need = sizeof(struct ocfs2_xattr_entry) +
4578                         OCFS2_XATTR_SIZE(name_len) + value_size;
4579         else {
4580                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4581
4582                 /*
4583                  * We only replace the old value if the new length is smaller
4584                  * than the old one. Otherwise we will allocate new space in the
4585                  * bucket to store it.
4586                  */
4587                 xe = xs->here;
4588                 if (ocfs2_xattr_is_local(xe))
4589                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4590                 else
4591                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4592
4593                 if (old >= value_size)
4594                         need = 0;
4595         }
4596
4597         free = xh_free_start - header_size;
4598         /*
4599          * We need to make sure the new name/value pair
4600          * can exist in the same block.
4601          */
4602         if (xh_free_start % blocksize < need)
4603                 free -= xh_free_start % blocksize;
4604
4605         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4606              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4607              " %u\n", xs->not_found,
4608              (unsigned long long)bucket_blkno(&xs->bucket),
4609              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4610              le16_to_cpu(xh->xh_name_value_len));
4611
4612         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4613                 if (need <= max_free &&
4614                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4615                         /*
4616                          * We can create the space by defragment. Since only the
4617                          * name/value will be moved, the xe shouldn't be changed
4618                          * in xs.
4619                          */
4620                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4621                         if (ret) {
4622                                 mlog_errno(ret);
4623                                 goto out;
4624                         }
4625
4626                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4627                         free = xh_free_start - header_size;
4628                         if (xh_free_start % blocksize < need)
4629                                 free -= xh_free_start % blocksize;
4630
4631                         if (free >= need)
4632                                 goto xattr_set;
4633
4634                         mlog(0, "Can't get enough space for xattr insert by "
4635                              "defragment. Need %u bytes, but we have %d, so "
4636                              "allocate new bucket for it.\n", need, free);
4637                 }
4638
4639                 /*
4640                  * We have to add new buckets or clusters and one
4641                  * allocation should leave us enough space for insert.
4642                  */
4643                 BUG_ON(allocation);
4644
4645                 /*
4646                  * We do not allow for overlapping ranges between buckets. And
4647                  * the maximum number of collisions we will allow for then is
4648                  * one bucket's worth, so check it here whether we need to
4649                  * add a new bucket for the insert.
4650                  */
4651                 ret = ocfs2_check_xattr_bucket_collision(inode,
4652                                                          &xs->bucket,
4653                                                          xi->name);
4654                 if (ret) {
4655                         mlog_errno(ret);
4656                         goto out;
4657                 }
4658
4659                 ret = ocfs2_add_new_xattr_bucket(inode,
4660                                                  xs->xattr_bh,
4661                                                  xs->bucket.bu_bhs[0]);
4662                 if (ret) {
4663                         mlog_errno(ret);
4664                         goto out;
4665                 }
4666
4667                 ocfs2_xattr_bucket_relse(inode, &xs->bucket);
4668                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4669
4670                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4671                                                    xi->name_index,
4672                                                    xi->name, xs);
4673                 if (ret && ret != -ENODATA)
4674                         goto out;
4675                 xs->not_found = ret;
4676                 allocation = 1;
4677                 goto try_again;
4678         }
4679
4680 xattr_set:
4681         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4682 out:
4683         mlog_exit(ret);
4684         return ret;
4685 }
4686
4687 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4688                                         struct ocfs2_xattr_bucket *bucket,
4689                                         void *para)
4690 {
4691         int ret = 0;
4692         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4693         u16 i;
4694         struct ocfs2_xattr_entry *xe;
4695
4696         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4697                 xe = &xh->xh_entries[i];
4698                 if (ocfs2_xattr_is_local(xe))
4699                         continue;
4700
4701                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4702                                                         bucket->bu_bhs[0],
4703                                                         i, 0);
4704                 if (ret) {
4705                         mlog_errno(ret);
4706                         break;
4707                 }
4708         }
4709
4710         return ret;
4711 }
4712
4713 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4714                                           struct buffer_head *xb_bh)
4715 {
4716         struct ocfs2_xattr_block *xb =
4717                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4718         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4719         int ret = 0;
4720         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4721         u64 p_blkno;
4722
4723         if (le16_to_cpu(el->l_next_free_rec) == 0)
4724                 return 0;
4725
4726         while (name_hash > 0) {
4727                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4728                                           &e_cpos, &num_clusters, el);
4729                 if (ret) {
4730                         mlog_errno(ret);
4731                         goto out;
4732                 }
4733
4734                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4735                                                   ocfs2_delete_xattr_in_bucket,
4736                                                   NULL);
4737                 if (ret) {
4738                         mlog_errno(ret);
4739                         goto out;
4740                 }
4741
4742                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4743                                              p_blkno, e_cpos, num_clusters);
4744                 if (ret) {
4745                         mlog_errno(ret);
4746                         break;
4747                 }
4748
4749                 if (e_cpos == 0)
4750                         break;
4751
4752                 name_hash = e_cpos - 1;
4753         }
4754
4755 out:
4756         return ret;
4757 }
4758
4759 /*
4760  * 'trusted' attributes support
4761  */
4762 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4763                                        size_t list_size, const char *name,
4764                                        size_t name_len)
4765 {
4766         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
4767         const size_t total_len = prefix_len + name_len + 1;
4768
4769         if (list && total_len <= list_size) {
4770                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4771                 memcpy(list + prefix_len, name, name_len);
4772                 list[prefix_len + name_len] = '\0';
4773         }
4774         return total_len;
4775 }
4776
4777 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4778                                    void *buffer, size_t size)
4779 {
4780         if (strcmp(name, "") == 0)
4781                 return -EINVAL;
4782         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4783                                buffer, size);
4784 }
4785
4786 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4787                                    const void *value, size_t size, int flags)
4788 {
4789         if (strcmp(name, "") == 0)
4790                 return -EINVAL;
4791
4792         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4793                                size, flags);
4794 }
4795
4796 struct xattr_handler ocfs2_xattr_trusted_handler = {
4797         .prefix = XATTR_TRUSTED_PREFIX,
4798         .list   = ocfs2_xattr_trusted_list,
4799         .get    = ocfs2_xattr_trusted_get,
4800         .set    = ocfs2_xattr_trusted_set,
4801 };
4802
4803 /*
4804  * 'user' attributes support
4805  */
4806 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4807                                     size_t list_size, const char *name,
4808                                     size_t name_len)
4809 {
4810         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
4811         const size_t total_len = prefix_len + name_len + 1;
4812         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4813
4814         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4815                 return 0;
4816
4817         if (list && total_len <= list_size) {
4818                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4819                 memcpy(list + prefix_len, name, name_len);
4820                 list[prefix_len + name_len] = '\0';
4821         }
4822         return total_len;
4823 }
4824
4825 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4826                                 void *buffer, size_t size)
4827 {
4828         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4829
4830         if (strcmp(name, "") == 0)
4831                 return -EINVAL;
4832         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4833                 return -EOPNOTSUPP;
4834         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4835                                buffer, size);
4836 }
4837
4838 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4839                                 const void *value, size_t size, int flags)
4840 {
4841         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4842
4843         if (strcmp(name, "") == 0)
4844                 return -EINVAL;
4845         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4846                 return -EOPNOTSUPP;
4847
4848         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4849                                size, flags);
4850 }
4851
4852 struct xattr_handler ocfs2_xattr_user_handler = {
4853         .prefix = XATTR_USER_PREFIX,
4854         .list   = ocfs2_xattr_user_list,
4855         .get    = ocfs2_xattr_user_get,
4856         .set    = ocfs2_xattr_user_set,
4857 };