ocfs2/xattr: Merge xattr set transaction.
[safe/jmp/linux-2.6] / fs / ocfs2 / xattr.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * xattr.c
5  *
6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
7  *
8  * CREDITS:
9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public
14  * License version 2 as published by the Free Software Foundation.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  */
21
22 #include <linux/capability.h>
23 #include <linux/fs.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38
39 #define MLOG_MASK_PREFIX ML_XATTR
40 #include <cluster/masklog.h>
41
42 #include "ocfs2.h"
43 #include "alloc.h"
44 #include "dlmglue.h"
45 #include "file.h"
46 #include "symlink.h"
47 #include "sysfile.h"
48 #include "inode.h"
49 #include "journal.h"
50 #include "ocfs2_fs.h"
51 #include "suballoc.h"
52 #include "uptodate.h"
53 #include "buffer_head_io.h"
54 #include "super.h"
55 #include "xattr.h"
56
57
58 struct ocfs2_xattr_def_value_root {
59         struct ocfs2_xattr_value_root   xv;
60         struct ocfs2_extent_rec         er;
61 };
62
63 struct ocfs2_xattr_bucket {
64         /* The inode these xattrs are associated with */
65         struct inode *bu_inode;
66
67         /* The actual buffers that make up the bucket */
68         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
69
70         /* How many blocks make up one bucket for this filesystem */
71         int bu_blocks;
72 };
73
74 struct ocfs2_xattr_set_ctxt {
75         handle_t *handle;
76         struct ocfs2_alloc_context *meta_ac;
77         struct ocfs2_alloc_context *data_ac;
78         struct ocfs2_cached_dealloc_ctxt dealloc;
79 };
80
81 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
82 #define OCFS2_XATTR_INLINE_SIZE 80
83
84 static struct ocfs2_xattr_def_value_root def_xv = {
85         .xv.xr_list.l_count = cpu_to_le16(1),
86 };
87
88 struct xattr_handler *ocfs2_xattr_handlers[] = {
89         &ocfs2_xattr_user_handler,
90         &ocfs2_xattr_trusted_handler,
91         NULL
92 };
93
94 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
95         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
96         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
97 };
98
99 struct ocfs2_xattr_info {
100         int name_index;
101         const char *name;
102         const void *value;
103         size_t value_len;
104 };
105
106 struct ocfs2_xattr_search {
107         struct buffer_head *inode_bh;
108         /*
109          * xattr_bh point to the block buffer head which has extended attribute
110          * when extended attribute in inode, xattr_bh is equal to inode_bh.
111          */
112         struct buffer_head *xattr_bh;
113         struct ocfs2_xattr_header *header;
114         struct ocfs2_xattr_bucket *bucket;
115         void *base;
116         void *end;
117         struct ocfs2_xattr_entry *here;
118         int not_found;
119 };
120
121 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
122                                              struct ocfs2_xattr_header *xh,
123                                              int index,
124                                              int *block_off,
125                                              int *new_offset);
126
127 static int ocfs2_xattr_block_find(struct inode *inode,
128                                   int name_index,
129                                   const char *name,
130                                   struct ocfs2_xattr_search *xs);
131 static int ocfs2_xattr_index_block_find(struct inode *inode,
132                                         struct buffer_head *root_bh,
133                                         int name_index,
134                                         const char *name,
135                                         struct ocfs2_xattr_search *xs);
136
137 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
138                                         struct ocfs2_xattr_tree_root *xt,
139                                         char *buffer,
140                                         size_t buffer_size);
141
142 static int ocfs2_xattr_create_index_block(struct inode *inode,
143                                           struct ocfs2_xattr_search *xs,
144                                           struct ocfs2_xattr_set_ctxt *ctxt);
145
146 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
147                                              struct ocfs2_xattr_info *xi,
148                                              struct ocfs2_xattr_search *xs,
149                                              struct ocfs2_xattr_set_ctxt *ctxt);
150
151 static int ocfs2_delete_xattr_index_block(struct inode *inode,
152                                           struct buffer_head *xb_bh);
153
154 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
155 {
156         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
157 }
158
159 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
160 {
161         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
162 }
163
164 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
165 {
166         u16 len = sb->s_blocksize -
167                  offsetof(struct ocfs2_xattr_header, xh_entries);
168
169         return len / sizeof(struct ocfs2_xattr_entry);
170 }
171
172 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
173 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
174 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
175
176 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
177 {
178         struct ocfs2_xattr_bucket *bucket;
179         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
180
181         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
182
183         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
184         if (bucket) {
185                 bucket->bu_inode = inode;
186                 bucket->bu_blocks = blks;
187         }
188
189         return bucket;
190 }
191
192 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
193 {
194         int i;
195
196         for (i = 0; i < bucket->bu_blocks; i++) {
197                 brelse(bucket->bu_bhs[i]);
198                 bucket->bu_bhs[i] = NULL;
199         }
200 }
201
202 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
203 {
204         if (bucket) {
205                 ocfs2_xattr_bucket_relse(bucket);
206                 bucket->bu_inode = NULL;
207                 kfree(bucket);
208         }
209 }
210
211 /*
212  * A bucket that has never been written to disk doesn't need to be
213  * read.  We just need the buffer_heads.  Don't call this for
214  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
215  * them fully.
216  */
217 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
218                                    u64 xb_blkno)
219 {
220         int i, rc = 0;
221
222         for (i = 0; i < bucket->bu_blocks; i++) {
223                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
224                                               xb_blkno + i);
225                 if (!bucket->bu_bhs[i]) {
226                         rc = -EIO;
227                         mlog_errno(rc);
228                         break;
229                 }
230
231                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
232                                            bucket->bu_bhs[i]))
233                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
234                                                       bucket->bu_bhs[i]);
235         }
236
237         if (rc)
238                 ocfs2_xattr_bucket_relse(bucket);
239         return rc;
240 }
241
242 /* Read the xattr bucket at xb_blkno */
243 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
244                                    u64 xb_blkno)
245 {
246         int rc;
247
248         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
249                                bucket->bu_blocks, bucket->bu_bhs, 0);
250         if (rc)
251                 ocfs2_xattr_bucket_relse(bucket);
252         return rc;
253 }
254
255 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
256                                              struct ocfs2_xattr_bucket *bucket,
257                                              int type)
258 {
259         int i, rc = 0;
260
261         for (i = 0; i < bucket->bu_blocks; i++) {
262                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
263                                           bucket->bu_bhs[i], type);
264                 if (rc) {
265                         mlog_errno(rc);
266                         break;
267                 }
268         }
269
270         return rc;
271 }
272
273 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
274                                              struct ocfs2_xattr_bucket *bucket)
275 {
276         int i;
277
278         for (i = 0; i < bucket->bu_blocks; i++)
279                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
280 }
281
282 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
283                                          struct ocfs2_xattr_bucket *src)
284 {
285         int i;
286         int blocksize = src->bu_inode->i_sb->s_blocksize;
287
288         BUG_ON(dest->bu_blocks != src->bu_blocks);
289         BUG_ON(dest->bu_inode != src->bu_inode);
290
291         for (i = 0; i < src->bu_blocks; i++) {
292                 memcpy(bucket_block(dest, i), bucket_block(src, i),
293                        blocksize);
294         }
295 }
296
297 static inline const char *ocfs2_xattr_prefix(int name_index)
298 {
299         struct xattr_handler *handler = NULL;
300
301         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
302                 handler = ocfs2_xattr_handler_map[name_index];
303
304         return handler ? handler->prefix : NULL;
305 }
306
307 static u32 ocfs2_xattr_name_hash(struct inode *inode,
308                                  const char *name,
309                                  int name_len)
310 {
311         /* Get hash value of uuid from super block */
312         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
313         int i;
314
315         /* hash extended attribute name */
316         for (i = 0; i < name_len; i++) {
317                 hash = (hash << OCFS2_HASH_SHIFT) ^
318                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
319                        *name++;
320         }
321
322         return hash;
323 }
324
325 /*
326  * ocfs2_xattr_hash_entry()
327  *
328  * Compute the hash of an extended attribute.
329  */
330 static void ocfs2_xattr_hash_entry(struct inode *inode,
331                                    struct ocfs2_xattr_header *header,
332                                    struct ocfs2_xattr_entry *entry)
333 {
334         u32 hash = 0;
335         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
336
337         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
338         entry->xe_name_hash = cpu_to_le32(hash);
339
340         return;
341 }
342
343 static int ocfs2_xattr_extend_allocation(struct inode *inode,
344                                          u32 clusters_to_add,
345                                          struct buffer_head *xattr_bh,
346                                          struct ocfs2_xattr_value_root *xv,
347                                          struct ocfs2_xattr_set_ctxt *ctxt)
348 {
349         int status = 0;
350         handle_t *handle = ctxt->handle;
351         enum ocfs2_alloc_restarted why;
352         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
353         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
354         struct ocfs2_extent_tree et;
355
356         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
357
358         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
359
360         status = ocfs2_journal_access(handle, inode, xattr_bh,
361                                       OCFS2_JOURNAL_ACCESS_WRITE);
362         if (status < 0) {
363                 mlog_errno(status);
364                 goto leave;
365         }
366
367         prev_clusters = le32_to_cpu(xv->xr_clusters);
368         status = ocfs2_add_clusters_in_btree(osb,
369                                              inode,
370                                              &logical_start,
371                                              clusters_to_add,
372                                              0,
373                                              &et,
374                                              handle,
375                                              ctxt->data_ac,
376                                              ctxt->meta_ac,
377                                              &why);
378         if (status < 0) {
379                 mlog_errno(status);
380                 goto leave;
381         }
382
383         status = ocfs2_journal_dirty(handle, xattr_bh);
384         if (status < 0) {
385                 mlog_errno(status);
386                 goto leave;
387         }
388
389         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
390
391         /*
392          * We should have already allocated enough space before the transaction,
393          * so no need to restart.
394          */
395         BUG_ON(why != RESTART_NONE || clusters_to_add);
396
397 leave:
398
399         return status;
400 }
401
402 static int __ocfs2_remove_xattr_range(struct inode *inode,
403                                       struct buffer_head *root_bh,
404                                       struct ocfs2_xattr_value_root *xv,
405                                       u32 cpos, u32 phys_cpos, u32 len,
406                                       struct ocfs2_xattr_set_ctxt *ctxt)
407 {
408         int ret;
409         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
410         handle_t *handle = ctxt->handle;
411         struct ocfs2_extent_tree et;
412
413         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
414
415         ret = ocfs2_journal_access(handle, inode, root_bh,
416                                    OCFS2_JOURNAL_ACCESS_WRITE);
417         if (ret) {
418                 mlog_errno(ret);
419                 goto out;
420         }
421
422         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
423                                   &ctxt->dealloc);
424         if (ret) {
425                 mlog_errno(ret);
426                 goto out;
427         }
428
429         le32_add_cpu(&xv->xr_clusters, -len);
430
431         ret = ocfs2_journal_dirty(handle, root_bh);
432         if (ret) {
433                 mlog_errno(ret);
434                 goto out;
435         }
436
437         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
438         if (ret)
439                 mlog_errno(ret);
440
441 out:
442         return ret;
443 }
444
445 static int ocfs2_xattr_shrink_size(struct inode *inode,
446                                    u32 old_clusters,
447                                    u32 new_clusters,
448                                    struct buffer_head *root_bh,
449                                    struct ocfs2_xattr_value_root *xv,
450                                    struct ocfs2_xattr_set_ctxt *ctxt)
451 {
452         int ret = 0;
453         u32 trunc_len, cpos, phys_cpos, alloc_size;
454         u64 block;
455
456         if (old_clusters <= new_clusters)
457                 return 0;
458
459         cpos = new_clusters;
460         trunc_len = old_clusters - new_clusters;
461         while (trunc_len) {
462                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
463                                                &alloc_size, &xv->xr_list);
464                 if (ret) {
465                         mlog_errno(ret);
466                         goto out;
467                 }
468
469                 if (alloc_size > trunc_len)
470                         alloc_size = trunc_len;
471
472                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
473                                                  phys_cpos, alloc_size,
474                                                  ctxt);
475                 if (ret) {
476                         mlog_errno(ret);
477                         goto out;
478                 }
479
480                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
481                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
482                                                        alloc_size);
483                 cpos += alloc_size;
484                 trunc_len -= alloc_size;
485         }
486
487 out:
488         return ret;
489 }
490
491 static int ocfs2_xattr_value_truncate(struct inode *inode,
492                                       struct buffer_head *root_bh,
493                                       struct ocfs2_xattr_value_root *xv,
494                                       int len,
495                                       struct ocfs2_xattr_set_ctxt *ctxt)
496 {
497         int ret;
498         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
499         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
500
501         if (new_clusters == old_clusters)
502                 return 0;
503
504         if (new_clusters > old_clusters)
505                 ret = ocfs2_xattr_extend_allocation(inode,
506                                                     new_clusters - old_clusters,
507                                                     root_bh, xv, ctxt);
508         else
509                 ret = ocfs2_xattr_shrink_size(inode,
510                                               old_clusters, new_clusters,
511                                               root_bh, xv, ctxt);
512
513         return ret;
514 }
515
516 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
517                                   size_t *result, const char *prefix,
518                                   const char *name, int name_len)
519 {
520         char *p = buffer + *result;
521         int prefix_len = strlen(prefix);
522         int total_len = prefix_len + name_len + 1;
523
524         *result += total_len;
525
526         /* we are just looking for how big our buffer needs to be */
527         if (!size)
528                 return 0;
529
530         if (*result > size)
531                 return -ERANGE;
532
533         memcpy(p, prefix, prefix_len);
534         memcpy(p + prefix_len, name, name_len);
535         p[prefix_len + name_len] = '\0';
536
537         return 0;
538 }
539
540 static int ocfs2_xattr_list_entries(struct inode *inode,
541                                     struct ocfs2_xattr_header *header,
542                                     char *buffer, size_t buffer_size)
543 {
544         size_t result = 0;
545         int i, type, ret;
546         const char *prefix, *name;
547
548         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
549                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
550                 type = ocfs2_xattr_get_type(entry);
551                 prefix = ocfs2_xattr_prefix(type);
552
553                 if (prefix) {
554                         name = (const char *)header +
555                                 le16_to_cpu(entry->xe_name_offset);
556
557                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
558                                                      &result, prefix, name,
559                                                      entry->xe_name_len);
560                         if (ret)
561                                 return ret;
562                 }
563         }
564
565         return result;
566 }
567
568 static int ocfs2_xattr_ibody_list(struct inode *inode,
569                                   struct ocfs2_dinode *di,
570                                   char *buffer,
571                                   size_t buffer_size)
572 {
573         struct ocfs2_xattr_header *header = NULL;
574         struct ocfs2_inode_info *oi = OCFS2_I(inode);
575         int ret = 0;
576
577         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
578                 return ret;
579
580         header = (struct ocfs2_xattr_header *)
581                  ((void *)di + inode->i_sb->s_blocksize -
582                  le16_to_cpu(di->i_xattr_inline_size));
583
584         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
585
586         return ret;
587 }
588
589 static int ocfs2_xattr_block_list(struct inode *inode,
590                                   struct ocfs2_dinode *di,
591                                   char *buffer,
592                                   size_t buffer_size)
593 {
594         struct buffer_head *blk_bh = NULL;
595         struct ocfs2_xattr_block *xb;
596         int ret = 0;
597
598         if (!di->i_xattr_loc)
599                 return ret;
600
601         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
602         if (ret < 0) {
603                 mlog_errno(ret);
604                 return ret;
605         }
606
607         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
608         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
609                 ret = -EIO;
610                 goto cleanup;
611         }
612
613         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
614                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
615                 ret = ocfs2_xattr_list_entries(inode, header,
616                                                buffer, buffer_size);
617         } else {
618                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
619                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
620                                                    buffer, buffer_size);
621         }
622 cleanup:
623         brelse(blk_bh);
624
625         return ret;
626 }
627
628 ssize_t ocfs2_listxattr(struct dentry *dentry,
629                         char *buffer,
630                         size_t size)
631 {
632         int ret = 0, i_ret = 0, b_ret = 0;
633         struct buffer_head *di_bh = NULL;
634         struct ocfs2_dinode *di = NULL;
635         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
636
637         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
638                 return -EOPNOTSUPP;
639
640         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
641                 return ret;
642
643         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
644         if (ret < 0) {
645                 mlog_errno(ret);
646                 return ret;
647         }
648
649         di = (struct ocfs2_dinode *)di_bh->b_data;
650
651         down_read(&oi->ip_xattr_sem);
652         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
653         if (i_ret < 0)
654                 b_ret = 0;
655         else {
656                 if (buffer) {
657                         buffer += i_ret;
658                         size -= i_ret;
659                 }
660                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
661                                                buffer, size);
662                 if (b_ret < 0)
663                         i_ret = 0;
664         }
665         up_read(&oi->ip_xattr_sem);
666         ocfs2_inode_unlock(dentry->d_inode, 0);
667
668         brelse(di_bh);
669
670         return i_ret + b_ret;
671 }
672
673 static int ocfs2_xattr_find_entry(int name_index,
674                                   const char *name,
675                                   struct ocfs2_xattr_search *xs)
676 {
677         struct ocfs2_xattr_entry *entry;
678         size_t name_len;
679         int i, cmp = 1;
680
681         if (name == NULL)
682                 return -EINVAL;
683
684         name_len = strlen(name);
685         entry = xs->here;
686         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
687                 cmp = name_index - ocfs2_xattr_get_type(entry);
688                 if (!cmp)
689                         cmp = name_len - entry->xe_name_len;
690                 if (!cmp)
691                         cmp = memcmp(name, (xs->base +
692                                      le16_to_cpu(entry->xe_name_offset)),
693                                      name_len);
694                 if (cmp == 0)
695                         break;
696                 entry += 1;
697         }
698         xs->here = entry;
699
700         return cmp ? -ENODATA : 0;
701 }
702
703 static int ocfs2_xattr_get_value_outside(struct inode *inode,
704                                          struct ocfs2_xattr_value_root *xv,
705                                          void *buffer,
706                                          size_t len)
707 {
708         u32 cpos, p_cluster, num_clusters, bpc, clusters;
709         u64 blkno;
710         int i, ret = 0;
711         size_t cplen, blocksize;
712         struct buffer_head *bh = NULL;
713         struct ocfs2_extent_list *el;
714
715         el = &xv->xr_list;
716         clusters = le32_to_cpu(xv->xr_clusters);
717         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
718         blocksize = inode->i_sb->s_blocksize;
719
720         cpos = 0;
721         while (cpos < clusters) {
722                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
723                                                &num_clusters, el);
724                 if (ret) {
725                         mlog_errno(ret);
726                         goto out;
727                 }
728
729                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
730                 /* Copy ocfs2_xattr_value */
731                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
732                         ret = ocfs2_read_block(inode, blkno, &bh);
733                         if (ret) {
734                                 mlog_errno(ret);
735                                 goto out;
736                         }
737
738                         cplen = len >= blocksize ? blocksize : len;
739                         memcpy(buffer, bh->b_data, cplen);
740                         len -= cplen;
741                         buffer += cplen;
742
743                         brelse(bh);
744                         bh = NULL;
745                         if (len == 0)
746                                 break;
747                 }
748                 cpos += num_clusters;
749         }
750 out:
751         return ret;
752 }
753
754 static int ocfs2_xattr_ibody_get(struct inode *inode,
755                                  int name_index,
756                                  const char *name,
757                                  void *buffer,
758                                  size_t buffer_size,
759                                  struct ocfs2_xattr_search *xs)
760 {
761         struct ocfs2_inode_info *oi = OCFS2_I(inode);
762         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
763         struct ocfs2_xattr_value_root *xv;
764         size_t size;
765         int ret = 0;
766
767         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
768                 return -ENODATA;
769
770         xs->end = (void *)di + inode->i_sb->s_blocksize;
771         xs->header = (struct ocfs2_xattr_header *)
772                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
773         xs->base = (void *)xs->header;
774         xs->here = xs->header->xh_entries;
775
776         ret = ocfs2_xattr_find_entry(name_index, name, xs);
777         if (ret)
778                 return ret;
779         size = le64_to_cpu(xs->here->xe_value_size);
780         if (buffer) {
781                 if (size > buffer_size)
782                         return -ERANGE;
783                 if (ocfs2_xattr_is_local(xs->here)) {
784                         memcpy(buffer, (void *)xs->base +
785                                le16_to_cpu(xs->here->xe_name_offset) +
786                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
787                 } else {
788                         xv = (struct ocfs2_xattr_value_root *)
789                                 (xs->base + le16_to_cpu(
790                                  xs->here->xe_name_offset) +
791                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
792                         ret = ocfs2_xattr_get_value_outside(inode, xv,
793                                                             buffer, size);
794                         if (ret < 0) {
795                                 mlog_errno(ret);
796                                 return ret;
797                         }
798                 }
799         }
800
801         return size;
802 }
803
804 static int ocfs2_xattr_block_get(struct inode *inode,
805                                  int name_index,
806                                  const char *name,
807                                  void *buffer,
808                                  size_t buffer_size,
809                                  struct ocfs2_xattr_search *xs)
810 {
811         struct ocfs2_xattr_block *xb;
812         struct ocfs2_xattr_value_root *xv;
813         size_t size;
814         int ret = -ENODATA, name_offset, name_len, block_off, i;
815
816         xs->bucket = ocfs2_xattr_bucket_new(inode);
817         if (!xs->bucket) {
818                 ret = -ENOMEM;
819                 mlog_errno(ret);
820                 goto cleanup;
821         }
822
823         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
824         if (ret) {
825                 mlog_errno(ret);
826                 goto cleanup;
827         }
828
829         if (xs->not_found) {
830                 ret = -ENODATA;
831                 goto cleanup;
832         }
833
834         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
835         size = le64_to_cpu(xs->here->xe_value_size);
836         if (buffer) {
837                 ret = -ERANGE;
838                 if (size > buffer_size)
839                         goto cleanup;
840
841                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
842                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
843                 i = xs->here - xs->header->xh_entries;
844
845                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
846                         ret = ocfs2_xattr_bucket_get_name_value(inode,
847                                                                 bucket_xh(xs->bucket),
848                                                                 i,
849                                                                 &block_off,
850                                                                 &name_offset);
851                         xs->base = bucket_block(xs->bucket, block_off);
852                 }
853                 if (ocfs2_xattr_is_local(xs->here)) {
854                         memcpy(buffer, (void *)xs->base +
855                                name_offset + name_len, size);
856                 } else {
857                         xv = (struct ocfs2_xattr_value_root *)
858                                 (xs->base + name_offset + name_len);
859                         ret = ocfs2_xattr_get_value_outside(inode, xv,
860                                                             buffer, size);
861                         if (ret < 0) {
862                                 mlog_errno(ret);
863                                 goto cleanup;
864                         }
865                 }
866         }
867         ret = size;
868 cleanup:
869         ocfs2_xattr_bucket_free(xs->bucket);
870
871         brelse(xs->xattr_bh);
872         xs->xattr_bh = NULL;
873         return ret;
874 }
875
876 /* ocfs2_xattr_get()
877  *
878  * Copy an extended attribute into the buffer provided.
879  * Buffer is NULL to compute the size of buffer required.
880  */
881 static int ocfs2_xattr_get(struct inode *inode,
882                            int name_index,
883                            const char *name,
884                            void *buffer,
885                            size_t buffer_size)
886 {
887         int ret;
888         struct ocfs2_dinode *di = NULL;
889         struct buffer_head *di_bh = NULL;
890         struct ocfs2_inode_info *oi = OCFS2_I(inode);
891         struct ocfs2_xattr_search xis = {
892                 .not_found = -ENODATA,
893         };
894         struct ocfs2_xattr_search xbs = {
895                 .not_found = -ENODATA,
896         };
897
898         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
899                 return -EOPNOTSUPP;
900
901         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
902                 ret = -ENODATA;
903
904         ret = ocfs2_inode_lock(inode, &di_bh, 0);
905         if (ret < 0) {
906                 mlog_errno(ret);
907                 return ret;
908         }
909         xis.inode_bh = xbs.inode_bh = di_bh;
910         di = (struct ocfs2_dinode *)di_bh->b_data;
911
912         down_read(&oi->ip_xattr_sem);
913         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
914                                     buffer_size, &xis);
915         if (ret == -ENODATA && di->i_xattr_loc)
916                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
917                                             buffer_size, &xbs);
918         up_read(&oi->ip_xattr_sem);
919         ocfs2_inode_unlock(inode, 0);
920
921         brelse(di_bh);
922
923         return ret;
924 }
925
926 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
927                                            handle_t *handle,
928                                            struct ocfs2_xattr_value_root *xv,
929                                            const void *value,
930                                            int value_len)
931 {
932         int ret = 0, i, cp_len, credits;
933         u16 blocksize = inode->i_sb->s_blocksize;
934         u32 p_cluster, num_clusters;
935         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
936         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
937         u64 blkno;
938         struct buffer_head *bh = NULL;
939
940         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
941
942         /*
943          * In __ocfs2_xattr_set_value_outside has already been dirtied,
944          * so we don't need to worry about whether ocfs2_extend_trans
945          * will create a new transactio for us or not.
946          */
947         credits = clusters * bpc;
948         ret = ocfs2_extend_trans(handle, credits);
949         if (ret) {
950                 mlog_errno(ret);
951                 goto out;
952         }
953
954         while (cpos < clusters) {
955                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
956                                                &num_clusters, &xv->xr_list);
957                 if (ret) {
958                         mlog_errno(ret);
959                         goto out;
960                 }
961
962                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
963
964                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
965                         ret = ocfs2_read_block(inode, blkno, &bh);
966                         if (ret) {
967                                 mlog_errno(ret);
968                                 goto out;
969                         }
970
971                         ret = ocfs2_journal_access(handle,
972                                                    inode,
973                                                    bh,
974                                                    OCFS2_JOURNAL_ACCESS_WRITE);
975                         if (ret < 0) {
976                                 mlog_errno(ret);
977                                 goto out;
978                         }
979
980                         cp_len = value_len > blocksize ? blocksize : value_len;
981                         memcpy(bh->b_data, value, cp_len);
982                         value_len -= cp_len;
983                         value += cp_len;
984                         if (cp_len < blocksize)
985                                 memset(bh->b_data + cp_len, 0,
986                                        blocksize - cp_len);
987
988                         ret = ocfs2_journal_dirty(handle, bh);
989                         if (ret < 0) {
990                                 mlog_errno(ret);
991                                 goto out;
992                         }
993                         brelse(bh);
994                         bh = NULL;
995
996                         /*
997                          * XXX: do we need to empty all the following
998                          * blocks in this cluster?
999                          */
1000                         if (!value_len)
1001                                 break;
1002                 }
1003                 cpos += num_clusters;
1004         }
1005 out:
1006         brelse(bh);
1007
1008         return ret;
1009 }
1010
1011 static int ocfs2_xattr_cleanup(struct inode *inode,
1012                                handle_t *handle,
1013                                struct ocfs2_xattr_info *xi,
1014                                struct ocfs2_xattr_search *xs,
1015                                size_t offs)
1016 {
1017         int ret = 0;
1018         size_t name_len = strlen(xi->name);
1019         void *val = xs->base + offs;
1020         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1021
1022         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1023                                    OCFS2_JOURNAL_ACCESS_WRITE);
1024         if (ret) {
1025                 mlog_errno(ret);
1026                 goto out;
1027         }
1028         /* Decrease xattr count */
1029         le16_add_cpu(&xs->header->xh_count, -1);
1030         /* Remove the xattr entry and tree root which has already be set*/
1031         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1032         memset(val, 0, size);
1033
1034         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1035         if (ret < 0)
1036                 mlog_errno(ret);
1037 out:
1038         return ret;
1039 }
1040
1041 static int ocfs2_xattr_update_entry(struct inode *inode,
1042                                     handle_t *handle,
1043                                     struct ocfs2_xattr_info *xi,
1044                                     struct ocfs2_xattr_search *xs,
1045                                     size_t offs)
1046 {
1047         int ret;
1048
1049         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1050                                    OCFS2_JOURNAL_ACCESS_WRITE);
1051         if (ret) {
1052                 mlog_errno(ret);
1053                 goto out;
1054         }
1055
1056         xs->here->xe_name_offset = cpu_to_le16(offs);
1057         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1058         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1059                 ocfs2_xattr_set_local(xs->here, 1);
1060         else
1061                 ocfs2_xattr_set_local(xs->here, 0);
1062         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1063
1064         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1065         if (ret < 0)
1066                 mlog_errno(ret);
1067 out:
1068         return ret;
1069 }
1070
1071 /*
1072  * ocfs2_xattr_set_value_outside()
1073  *
1074  * Set large size value in B tree.
1075  */
1076 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1077                                          struct ocfs2_xattr_info *xi,
1078                                          struct ocfs2_xattr_search *xs,
1079                                          struct ocfs2_xattr_set_ctxt *ctxt,
1080                                          size_t offs)
1081 {
1082         size_t name_len = strlen(xi->name);
1083         void *val = xs->base + offs;
1084         struct ocfs2_xattr_value_root *xv = NULL;
1085         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1086         int ret = 0;
1087
1088         memset(val, 0, size);
1089         memcpy(val, xi->name, name_len);
1090         xv = (struct ocfs2_xattr_value_root *)
1091                 (val + OCFS2_XATTR_SIZE(name_len));
1092         xv->xr_clusters = 0;
1093         xv->xr_last_eb_blk = 0;
1094         xv->xr_list.l_tree_depth = 0;
1095         xv->xr_list.l_count = cpu_to_le16(1);
1096         xv->xr_list.l_next_free_rec = 0;
1097
1098         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1099                                          xi->value_len, ctxt);
1100         if (ret < 0) {
1101                 mlog_errno(ret);
1102                 return ret;
1103         }
1104         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, offs);
1105         if (ret < 0) {
1106                 mlog_errno(ret);
1107                 return ret;
1108         }
1109         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, xv,
1110                                               xi->value, xi->value_len);
1111         if (ret < 0)
1112                 mlog_errno(ret);
1113
1114         return ret;
1115 }
1116
1117 /*
1118  * ocfs2_xattr_set_entry_local()
1119  *
1120  * Set, replace or remove extended attribute in local.
1121  */
1122 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1123                                         struct ocfs2_xattr_info *xi,
1124                                         struct ocfs2_xattr_search *xs,
1125                                         struct ocfs2_xattr_entry *last,
1126                                         size_t min_offs)
1127 {
1128         size_t name_len = strlen(xi->name);
1129         int i;
1130
1131         if (xi->value && xs->not_found) {
1132                 /* Insert the new xattr entry. */
1133                 le16_add_cpu(&xs->header->xh_count, 1);
1134                 ocfs2_xattr_set_type(last, xi->name_index);
1135                 ocfs2_xattr_set_local(last, 1);
1136                 last->xe_name_len = name_len;
1137         } else {
1138                 void *first_val;
1139                 void *val;
1140                 size_t offs, size;
1141
1142                 first_val = xs->base + min_offs;
1143                 offs = le16_to_cpu(xs->here->xe_name_offset);
1144                 val = xs->base + offs;
1145
1146                 if (le64_to_cpu(xs->here->xe_value_size) >
1147                     OCFS2_XATTR_INLINE_SIZE)
1148                         size = OCFS2_XATTR_SIZE(name_len) +
1149                                 OCFS2_XATTR_ROOT_SIZE;
1150                 else
1151                         size = OCFS2_XATTR_SIZE(name_len) +
1152                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1153
1154                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1155                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1156                         /* The old and the new value have the
1157                            same size. Just replace the value. */
1158                         ocfs2_xattr_set_local(xs->here, 1);
1159                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1160                         /* Clear value bytes. */
1161                         memset(val + OCFS2_XATTR_SIZE(name_len),
1162                                0,
1163                                OCFS2_XATTR_SIZE(xi->value_len));
1164                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1165                                xi->value,
1166                                xi->value_len);
1167                         return;
1168                 }
1169                 /* Remove the old name+value. */
1170                 memmove(first_val + size, first_val, val - first_val);
1171                 memset(first_val, 0, size);
1172                 xs->here->xe_name_hash = 0;
1173                 xs->here->xe_name_offset = 0;
1174                 ocfs2_xattr_set_local(xs->here, 1);
1175                 xs->here->xe_value_size = 0;
1176
1177                 min_offs += size;
1178
1179                 /* Adjust all value offsets. */
1180                 last = xs->header->xh_entries;
1181                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1182                         size_t o = le16_to_cpu(last->xe_name_offset);
1183
1184                         if (o < offs)
1185                                 last->xe_name_offset = cpu_to_le16(o + size);
1186                         last += 1;
1187                 }
1188
1189                 if (!xi->value) {
1190                         /* Remove the old entry. */
1191                         last -= 1;
1192                         memmove(xs->here, xs->here + 1,
1193                                 (void *)last - (void *)xs->here);
1194                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1195                         le16_add_cpu(&xs->header->xh_count, -1);
1196                 }
1197         }
1198         if (xi->value) {
1199                 /* Insert the new name+value. */
1200                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1201                                 OCFS2_XATTR_SIZE(xi->value_len);
1202                 void *val = xs->base + min_offs - size;
1203
1204                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1205                 memset(val, 0, size);
1206                 memcpy(val, xi->name, name_len);
1207                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1208                        xi->value,
1209                        xi->value_len);
1210                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1211                 ocfs2_xattr_set_local(xs->here, 1);
1212                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1213         }
1214
1215         return;
1216 }
1217
1218 /*
1219  * ocfs2_xattr_set_entry()
1220  *
1221  * Set extended attribute entry into inode or block.
1222  *
1223  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1224  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1225  * then set value in B tree with set_value_outside().
1226  */
1227 static int ocfs2_xattr_set_entry(struct inode *inode,
1228                                  struct ocfs2_xattr_info *xi,
1229                                  struct ocfs2_xattr_search *xs,
1230                                  struct ocfs2_xattr_set_ctxt *ctxt,
1231                                  int flag)
1232 {
1233         struct ocfs2_xattr_entry *last;
1234         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1235         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1236         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1237         size_t size_l = 0;
1238         handle_t *handle = ctxt->handle;
1239         int free, i, ret;
1240         struct ocfs2_xattr_info xi_l = {
1241                 .name_index = xi->name_index,
1242                 .name = xi->name,
1243                 .value = xi->value,
1244                 .value_len = xi->value_len,
1245         };
1246
1247         /* Compute min_offs, last and free space. */
1248         last = xs->header->xh_entries;
1249
1250         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1251                 size_t offs = le16_to_cpu(last->xe_name_offset);
1252                 if (offs < min_offs)
1253                         min_offs = offs;
1254                 last += 1;
1255         }
1256
1257         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1258         if (free < 0)
1259                 return -EIO;
1260
1261         if (!xs->not_found) {
1262                 size_t size = 0;
1263                 if (ocfs2_xattr_is_local(xs->here))
1264                         size = OCFS2_XATTR_SIZE(name_len) +
1265                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1266                 else
1267                         size = OCFS2_XATTR_SIZE(name_len) +
1268                                 OCFS2_XATTR_ROOT_SIZE;
1269                 free += (size + sizeof(struct ocfs2_xattr_entry));
1270         }
1271         /* Check free space in inode or block */
1272         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1273                 if (free < sizeof(struct ocfs2_xattr_entry) +
1274                            OCFS2_XATTR_SIZE(name_len) +
1275                            OCFS2_XATTR_ROOT_SIZE) {
1276                         ret = -ENOSPC;
1277                         goto out;
1278                 }
1279                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1280                 xi_l.value = (void *)&def_xv;
1281                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1282         } else if (xi->value) {
1283                 if (free < sizeof(struct ocfs2_xattr_entry) +
1284                            OCFS2_XATTR_SIZE(name_len) +
1285                            OCFS2_XATTR_SIZE(xi->value_len)) {
1286                         ret = -ENOSPC;
1287                         goto out;
1288                 }
1289         }
1290
1291         if (!xs->not_found) {
1292                 /* For existing extended attribute */
1293                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1294                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1295                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1296                 void *val = xs->base + offs;
1297
1298                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1299                         /* Replace existing local xattr with tree root */
1300                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1301                                                             ctxt, offs);
1302                         if (ret < 0)
1303                                 mlog_errno(ret);
1304                         goto out;
1305                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1306                         /* For existing xattr which has value outside */
1307                         struct ocfs2_xattr_value_root *xv = NULL;
1308                         xv = (struct ocfs2_xattr_value_root *)(val +
1309                                 OCFS2_XATTR_SIZE(name_len));
1310
1311                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1312                                 /*
1313                                  * If new value need set outside also,
1314                                  * first truncate old value to new value,
1315                                  * then set new value with set_value_outside().
1316                                  */
1317                                 ret = ocfs2_xattr_value_truncate(inode,
1318                                                                  xs->xattr_bh,
1319                                                                  xv,
1320                                                                  xi->value_len,
1321                                                                  ctxt);
1322                                 if (ret < 0) {
1323                                         mlog_errno(ret);
1324                                         goto out;
1325                                 }
1326
1327                                 ret = ocfs2_xattr_update_entry(inode,
1328                                                                handle,
1329                                                                xi,
1330                                                                xs,
1331                                                                offs);
1332                                 if (ret < 0) {
1333                                         mlog_errno(ret);
1334                                         goto out;
1335                                 }
1336
1337                                 ret = __ocfs2_xattr_set_value_outside(inode,
1338                                                                 handle,
1339                                                                 xv,
1340                                                                 xi->value,
1341                                                                 xi->value_len);
1342                                 if (ret < 0)
1343                                         mlog_errno(ret);
1344                                 goto out;
1345                         } else {
1346                                 /*
1347                                  * If new value need set in local,
1348                                  * just trucate old value to zero.
1349                                  */
1350                                  ret = ocfs2_xattr_value_truncate(inode,
1351                                                                   xs->xattr_bh,
1352                                                                   xv,
1353                                                                   0,
1354                                                                   ctxt);
1355                                 if (ret < 0)
1356                                         mlog_errno(ret);
1357                         }
1358                 }
1359         }
1360
1361         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1362                                    OCFS2_JOURNAL_ACCESS_WRITE);
1363         if (ret) {
1364                 mlog_errno(ret);
1365                 goto out;
1366         }
1367
1368         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1369                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1370                                            OCFS2_JOURNAL_ACCESS_WRITE);
1371                 if (ret) {
1372                         mlog_errno(ret);
1373                         goto out;
1374                 }
1375         }
1376
1377         /*
1378          * Set value in local, include set tree root in local.
1379          * This is the first step for value size >INLINE_SIZE.
1380          */
1381         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1382
1383         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1384                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1385                 if (ret < 0) {
1386                         mlog_errno(ret);
1387                         goto out;
1388                 }
1389         }
1390
1391         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1392             (flag & OCFS2_INLINE_XATTR_FL)) {
1393                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1394                 unsigned int xattrsize = osb->s_xattr_inline_size;
1395
1396                 /*
1397                  * Adjust extent record count or inline data size
1398                  * to reserve space for extended attribute.
1399                  */
1400                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1401                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1402                         le16_add_cpu(&idata->id_count, -xattrsize);
1403                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1404                         struct ocfs2_extent_list *el = &di->id2.i_list;
1405                         le16_add_cpu(&el->l_count, -(xattrsize /
1406                                         sizeof(struct ocfs2_extent_rec)));
1407                 }
1408                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1409         }
1410         /* Update xattr flag */
1411         spin_lock(&oi->ip_lock);
1412         oi->ip_dyn_features |= flag;
1413         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1414         spin_unlock(&oi->ip_lock);
1415         /* Update inode ctime */
1416         inode->i_ctime = CURRENT_TIME;
1417         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1418         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1419
1420         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1421         if (ret < 0)
1422                 mlog_errno(ret);
1423
1424         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1425                 /*
1426                  * Set value outside in B tree.
1427                  * This is the second step for value size > INLINE_SIZE.
1428                  */
1429                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1430                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt, offs);
1431                 if (ret < 0) {
1432                         int ret2;
1433
1434                         mlog_errno(ret);
1435                         /*
1436                          * If set value outside failed, we have to clean
1437                          * the junk tree root we have already set in local.
1438                          */
1439                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1440                                                    xi, xs, offs);
1441                         if (ret2 < 0)
1442                                 mlog_errno(ret2);
1443                 }
1444         }
1445 out:
1446         return ret;
1447 }
1448
1449 static int ocfs2_remove_value_outside(struct inode*inode,
1450                                       struct buffer_head *bh,
1451                                       struct ocfs2_xattr_header *header)
1452 {
1453         int ret = 0, i;
1454         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1455         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1456
1457         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1458
1459         ctxt.handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
1460         if (IS_ERR(ctxt.handle)) {
1461                 ret = PTR_ERR(ctxt.handle);
1462                 mlog_errno(ret);
1463                 goto out;
1464         }
1465
1466         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1467                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1468
1469                 if (!ocfs2_xattr_is_local(entry)) {
1470                         struct ocfs2_xattr_value_root *xv;
1471                         void *val;
1472
1473                         val = (void *)header +
1474                                 le16_to_cpu(entry->xe_name_offset);
1475                         xv = (struct ocfs2_xattr_value_root *)
1476                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1477                         ret = ocfs2_xattr_value_truncate(inode, bh, xv,
1478                                                          0, &ctxt);
1479                         if (ret < 0) {
1480                                 mlog_errno(ret);
1481                                 break;
1482                         }
1483                 }
1484         }
1485
1486         ocfs2_commit_trans(osb, ctxt.handle);
1487         ocfs2_schedule_truncate_log_flush(osb, 1);
1488         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1489 out:
1490         return ret;
1491 }
1492
1493 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1494                                     struct buffer_head *di_bh)
1495 {
1496
1497         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1498         struct ocfs2_xattr_header *header;
1499         int ret;
1500
1501         header = (struct ocfs2_xattr_header *)
1502                  ((void *)di + inode->i_sb->s_blocksize -
1503                  le16_to_cpu(di->i_xattr_inline_size));
1504
1505         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1506
1507         return ret;
1508 }
1509
1510 static int ocfs2_xattr_block_remove(struct inode *inode,
1511                                     struct buffer_head *blk_bh)
1512 {
1513         struct ocfs2_xattr_block *xb;
1514         int ret = 0;
1515
1516         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1517         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1518                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1519                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1520         } else
1521                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1522
1523         return ret;
1524 }
1525
1526 static int ocfs2_xattr_free_block(struct inode *inode,
1527                                   u64 block)
1528 {
1529         struct inode *xb_alloc_inode;
1530         struct buffer_head *xb_alloc_bh = NULL;
1531         struct buffer_head *blk_bh = NULL;
1532         struct ocfs2_xattr_block *xb;
1533         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1534         handle_t *handle;
1535         int ret = 0;
1536         u64 blk, bg_blkno;
1537         u16 bit;
1538
1539         ret = ocfs2_read_block(inode, block, &blk_bh);
1540         if (ret < 0) {
1541                 mlog_errno(ret);
1542                 goto out;
1543         }
1544
1545         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1546         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1547                 ret = -EIO;
1548                 goto out;
1549         }
1550
1551         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1552         if (ret < 0) {
1553                 mlog_errno(ret);
1554                 goto out;
1555         }
1556
1557         blk = le64_to_cpu(xb->xb_blkno);
1558         bit = le16_to_cpu(xb->xb_suballoc_bit);
1559         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1560
1561         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1562                                 EXTENT_ALLOC_SYSTEM_INODE,
1563                                 le16_to_cpu(xb->xb_suballoc_slot));
1564         if (!xb_alloc_inode) {
1565                 ret = -ENOMEM;
1566                 mlog_errno(ret);
1567                 goto out;
1568         }
1569         mutex_lock(&xb_alloc_inode->i_mutex);
1570
1571         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1572         if (ret < 0) {
1573                 mlog_errno(ret);
1574                 goto out_mutex;
1575         }
1576
1577         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1578         if (IS_ERR(handle)) {
1579                 ret = PTR_ERR(handle);
1580                 mlog_errno(ret);
1581                 goto out_unlock;
1582         }
1583
1584         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1585                                        bit, bg_blkno, 1);
1586         if (ret < 0)
1587                 mlog_errno(ret);
1588
1589         ocfs2_commit_trans(osb, handle);
1590 out_unlock:
1591         ocfs2_inode_unlock(xb_alloc_inode, 1);
1592         brelse(xb_alloc_bh);
1593 out_mutex:
1594         mutex_unlock(&xb_alloc_inode->i_mutex);
1595         iput(xb_alloc_inode);
1596 out:
1597         brelse(blk_bh);
1598         return ret;
1599 }
1600
1601 /*
1602  * ocfs2_xattr_remove()
1603  *
1604  * Free extended attribute resources associated with this inode.
1605  */
1606 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1607 {
1608         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1609         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1610         handle_t *handle;
1611         int ret;
1612
1613         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1614                 return 0;
1615
1616         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1617                 return 0;
1618
1619         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1620                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1621                 if (ret < 0) {
1622                         mlog_errno(ret);
1623                         goto out;
1624                 }
1625         }
1626
1627         if (di->i_xattr_loc) {
1628                 ret = ocfs2_xattr_free_block(inode,
1629                                              le64_to_cpu(di->i_xattr_loc));
1630                 if (ret < 0) {
1631                         mlog_errno(ret);
1632                         goto out;
1633                 }
1634         }
1635
1636         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1637                                    OCFS2_INODE_UPDATE_CREDITS);
1638         if (IS_ERR(handle)) {
1639                 ret = PTR_ERR(handle);
1640                 mlog_errno(ret);
1641                 goto out;
1642         }
1643         ret = ocfs2_journal_access(handle, inode, di_bh,
1644                                    OCFS2_JOURNAL_ACCESS_WRITE);
1645         if (ret) {
1646                 mlog_errno(ret);
1647                 goto out_commit;
1648         }
1649
1650         di->i_xattr_loc = 0;
1651
1652         spin_lock(&oi->ip_lock);
1653         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1654         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1655         spin_unlock(&oi->ip_lock);
1656
1657         ret = ocfs2_journal_dirty(handle, di_bh);
1658         if (ret < 0)
1659                 mlog_errno(ret);
1660 out_commit:
1661         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1662 out:
1663         return ret;
1664 }
1665
1666 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1667                                         struct ocfs2_dinode *di)
1668 {
1669         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1670         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1671         int free;
1672
1673         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1674                 return 0;
1675
1676         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1677                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1678                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1679         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1680                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1681                         le64_to_cpu(di->i_size);
1682         } else {
1683                 struct ocfs2_extent_list *el = &di->id2.i_list;
1684                 free = (le16_to_cpu(el->l_count) -
1685                         le16_to_cpu(el->l_next_free_rec)) *
1686                         sizeof(struct ocfs2_extent_rec);
1687         }
1688         if (free >= xattrsize)
1689                 return 1;
1690
1691         return 0;
1692 }
1693
1694 /*
1695  * ocfs2_xattr_ibody_find()
1696  *
1697  * Find extended attribute in inode block and
1698  * fill search info into struct ocfs2_xattr_search.
1699  */
1700 static int ocfs2_xattr_ibody_find(struct inode *inode,
1701                                   int name_index,
1702                                   const char *name,
1703                                   struct ocfs2_xattr_search *xs)
1704 {
1705         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1706         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1707         int ret;
1708         int has_space = 0;
1709
1710         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1711                 return 0;
1712
1713         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1714                 down_read(&oi->ip_alloc_sem);
1715                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1716                 up_read(&oi->ip_alloc_sem);
1717                 if (!has_space)
1718                         return 0;
1719         }
1720
1721         xs->xattr_bh = xs->inode_bh;
1722         xs->end = (void *)di + inode->i_sb->s_blocksize;
1723         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1724                 xs->header = (struct ocfs2_xattr_header *)
1725                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1726         else
1727                 xs->header = (struct ocfs2_xattr_header *)
1728                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1729         xs->base = (void *)xs->header;
1730         xs->here = xs->header->xh_entries;
1731
1732         /* Find the named attribute. */
1733         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1734                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1735                 if (ret && ret != -ENODATA)
1736                         return ret;
1737                 xs->not_found = ret;
1738         }
1739
1740         return 0;
1741 }
1742
1743 /*
1744  * ocfs2_xattr_ibody_set()
1745  *
1746  * Set, replace or remove an extended attribute into inode block.
1747  *
1748  */
1749 static int ocfs2_xattr_ibody_set(struct inode *inode,
1750                                  struct ocfs2_xattr_info *xi,
1751                                  struct ocfs2_xattr_search *xs,
1752                                  struct ocfs2_xattr_set_ctxt *ctxt)
1753 {
1754         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1755         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1756         int ret;
1757
1758         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1759                 return -ENOSPC;
1760
1761         down_write(&oi->ip_alloc_sem);
1762         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1763                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1764                         ret = -ENOSPC;
1765                         goto out;
1766                 }
1767         }
1768
1769         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
1770                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1771 out:
1772         up_write(&oi->ip_alloc_sem);
1773
1774         return ret;
1775 }
1776
1777 /*
1778  * ocfs2_xattr_block_find()
1779  *
1780  * Find extended attribute in external block and
1781  * fill search info into struct ocfs2_xattr_search.
1782  */
1783 static int ocfs2_xattr_block_find(struct inode *inode,
1784                                   int name_index,
1785                                   const char *name,
1786                                   struct ocfs2_xattr_search *xs)
1787 {
1788         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1789         struct buffer_head *blk_bh = NULL;
1790         struct ocfs2_xattr_block *xb;
1791         int ret = 0;
1792
1793         if (!di->i_xattr_loc)
1794                 return ret;
1795
1796         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
1797         if (ret < 0) {
1798                 mlog_errno(ret);
1799                 return ret;
1800         }
1801
1802         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1803         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1804                 ret = -EIO;
1805                 goto cleanup;
1806         }
1807
1808         xs->xattr_bh = blk_bh;
1809
1810         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1811                 xs->header = &xb->xb_attrs.xb_header;
1812                 xs->base = (void *)xs->header;
1813                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1814                 xs->here = xs->header->xh_entries;
1815
1816                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1817         } else
1818                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1819                                                    name_index,
1820                                                    name, xs);
1821
1822         if (ret && ret != -ENODATA) {
1823                 xs->xattr_bh = NULL;
1824                 goto cleanup;
1825         }
1826         xs->not_found = ret;
1827         return 0;
1828 cleanup:
1829         brelse(blk_bh);
1830
1831         return ret;
1832 }
1833
1834 /*
1835  * ocfs2_xattr_block_set()
1836  *
1837  * Set, replace or remove an extended attribute into external block.
1838  *
1839  */
1840 static int ocfs2_xattr_block_set(struct inode *inode,
1841                                  struct ocfs2_xattr_info *xi,
1842                                  struct ocfs2_xattr_search *xs,
1843                                  struct ocfs2_xattr_set_ctxt *ctxt)
1844 {
1845         struct buffer_head *new_bh = NULL;
1846         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1847         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1848         handle_t *handle = ctxt->handle;
1849         struct ocfs2_xattr_block *xblk = NULL;
1850         u16 suballoc_bit_start;
1851         u32 num_got;
1852         u64 first_blkno;
1853         int ret;
1854
1855         if (!xs->xattr_bh) {
1856                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1857                                            OCFS2_JOURNAL_ACCESS_CREATE);
1858                 if (ret < 0) {
1859                         mlog_errno(ret);
1860                         goto end;
1861                 }
1862
1863                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
1864                                            &suballoc_bit_start, &num_got,
1865                                            &first_blkno);
1866                 if (ret < 0) {
1867                         mlog_errno(ret);
1868                         goto end;
1869                 }
1870
1871                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1872                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1873
1874                 ret = ocfs2_journal_access(handle, inode, new_bh,
1875                                            OCFS2_JOURNAL_ACCESS_CREATE);
1876                 if (ret < 0) {
1877                         mlog_errno(ret);
1878                         goto end;
1879                 }
1880
1881                 /* Initialize ocfs2_xattr_block */
1882                 xs->xattr_bh = new_bh;
1883                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1884                 memset(xblk, 0, inode->i_sb->s_blocksize);
1885                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1886                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1887                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1888                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1889                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1890
1891                 xs->header = &xblk->xb_attrs.xb_header;
1892                 xs->base = (void *)xs->header;
1893                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1894                 xs->here = xs->header->xh_entries;
1895
1896                 ret = ocfs2_journal_dirty(handle, new_bh);
1897                 if (ret < 0) {
1898                         mlog_errno(ret);
1899                         goto end;
1900                 }
1901                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1902                 ocfs2_journal_dirty(handle, xs->inode_bh);
1903         } else
1904                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1905
1906         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1907                 /* Set extended attribute into external block */
1908                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
1909                                             OCFS2_HAS_XATTR_FL);
1910                 if (!ret || ret != -ENOSPC)
1911                         goto end;
1912
1913                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
1914                 if (ret)
1915                         goto end;
1916         }
1917
1918         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
1919
1920 end:
1921
1922         return ret;
1923 }
1924
1925 /* Check whether the new xattr can be inserted into the inode. */
1926 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
1927                                        struct ocfs2_xattr_info *xi,
1928                                        struct ocfs2_xattr_search *xs)
1929 {
1930         u64 value_size;
1931         struct ocfs2_xattr_entry *last;
1932         int free, i;
1933         size_t min_offs = xs->end - xs->base;
1934
1935         if (!xs->header)
1936                 return 0;
1937
1938         last = xs->header->xh_entries;
1939
1940         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1941                 size_t offs = le16_to_cpu(last->xe_name_offset);
1942                 if (offs < min_offs)
1943                         min_offs = offs;
1944                 last += 1;
1945         }
1946
1947         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1948         if (free < 0)
1949                 return 0;
1950
1951         BUG_ON(!xs->not_found);
1952
1953         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
1954                 value_size = OCFS2_XATTR_ROOT_SIZE;
1955         else
1956                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
1957
1958         if (free >= sizeof(struct ocfs2_xattr_entry) +
1959                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
1960                 return 1;
1961
1962         return 0;
1963 }
1964
1965 static int ocfs2_calc_xattr_set_need(struct inode *inode,
1966                                      struct ocfs2_dinode *di,
1967                                      struct ocfs2_xattr_info *xi,
1968                                      struct ocfs2_xattr_search *xis,
1969                                      struct ocfs2_xattr_search *xbs,
1970                                      int *clusters_need,
1971                                      int *meta_need,
1972                                      int *credits_need)
1973 {
1974         int ret = 0, old_in_xb = 0;
1975         int clusters_add = 0, meta_add = 0, credits = 0;
1976         struct buffer_head *bh = NULL;
1977         struct ocfs2_xattr_block *xb = NULL;
1978         struct ocfs2_xattr_entry *xe = NULL;
1979         struct ocfs2_xattr_value_root *xv = NULL;
1980         char *base = NULL;
1981         int name_offset, name_len = 0;
1982         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
1983                                                     xi->value_len);
1984         u64 value_size;
1985
1986         if (xis->not_found && xbs->not_found) {
1987                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1988
1989                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1990                         clusters_add += new_clusters;
1991                         credits += ocfs2_calc_extend_credits(inode->i_sb,
1992                                                         &def_xv.xv.xr_list,
1993                                                         new_clusters);
1994                 }
1995
1996                 goto meta_guess;
1997         }
1998
1999         if (!xis->not_found) {
2000                 xe = xis->here;
2001                 name_offset = le16_to_cpu(xe->xe_name_offset);
2002                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2003                 base = xis->base;
2004                 credits += OCFS2_INODE_UPDATE_CREDITS;
2005         } else {
2006                 int i, block_off;
2007                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2008                 xe = xbs->here;
2009                 name_offset = le16_to_cpu(xe->xe_name_offset);
2010                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2011                 i = xbs->here - xbs->header->xh_entries;
2012                 old_in_xb = 1;
2013
2014                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2015                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2016                                                         bucket_xh(xbs->bucket),
2017                                                         i, &block_off,
2018                                                         &name_offset);
2019                         base = bucket_block(xbs->bucket, block_off);
2020                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2021                 } else {
2022                         base = xbs->base;
2023                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2024                 }
2025         }
2026
2027         /*
2028          * delete a xattr doesn't need metadata and cluster allocation.
2029          * so just calculate the credits and return.
2030          *
2031          * The credits for removing the value tree will be extended
2032          * by ocfs2_remove_extent itself.
2033          */
2034         if (!xi->value) {
2035                 if (!ocfs2_xattr_is_local(xe))
2036                         credits += OCFS2_REMOVE_EXTENT_CREDITS;
2037
2038                 goto out;
2039         }
2040
2041         /* do cluster allocation guess first. */
2042         value_size = le64_to_cpu(xe->xe_value_size);
2043
2044         if (old_in_xb) {
2045                 /*
2046                  * In xattr set, we always try to set the xe in inode first,
2047                  * so if it can be inserted into inode successfully, the old
2048                  * one will be removed from the xattr block, and this xattr
2049                  * will be inserted into inode as a new xattr in inode.
2050                  */
2051                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2052                         clusters_add += new_clusters;
2053                         credits += OCFS2_REMOVE_EXTENT_CREDITS +
2054                                     OCFS2_INODE_UPDATE_CREDITS;
2055                         if (!ocfs2_xattr_is_local(xe))
2056                                 credits += ocfs2_calc_extend_credits(
2057                                                         inode->i_sb,
2058                                                         &def_xv.xv.xr_list,
2059                                                         new_clusters);
2060                         goto out;
2061                 }
2062         }
2063
2064         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2065                 /* the new values will be stored outside. */
2066                 u32 old_clusters = 0;
2067
2068                 if (!ocfs2_xattr_is_local(xe)) {
2069                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2070                                                                  value_size);
2071                         xv = (struct ocfs2_xattr_value_root *)
2072                              (base + name_offset + name_len);
2073                 } else
2074                         xv = &def_xv.xv;
2075
2076                 if (old_clusters >= new_clusters) {
2077                         credits += OCFS2_REMOVE_EXTENT_CREDITS;
2078                         goto out;
2079                 } else {
2080                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2081                         clusters_add += new_clusters - old_clusters;
2082                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2083                                                              &xv->xr_list,
2084                                                              new_clusters -
2085                                                              old_clusters);
2086                         goto out;
2087                 }
2088         } else {
2089                 /*
2090                  * Now the new value will be stored inside. So if the new
2091                  * value is smaller than the size of value root or the old
2092                  * value, we don't need any allocation, otherwise we have
2093                  * to guess metadata allocation.
2094                  */
2095                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2096                     (!ocfs2_xattr_is_local(xe) &&
2097                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2098                         goto out;
2099         }
2100
2101 meta_guess:
2102         /* calculate metadata allocation. */
2103         if (di->i_xattr_loc) {
2104                 if (!xbs->xattr_bh) {
2105                         ret = ocfs2_read_block(inode,
2106                                                le64_to_cpu(di->i_xattr_loc),
2107                                                &bh);
2108                         if (ret) {
2109                                 mlog_errno(ret);
2110                                 goto out;
2111                         }
2112
2113                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2114                 } else
2115                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2116
2117                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2118                         struct ocfs2_extent_list *el =
2119                                  &xb->xb_attrs.xb_root.xt_list;
2120                         meta_add += ocfs2_extend_meta_needed(el);
2121                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2122                                                              el, 1);
2123                 }
2124
2125                 /*
2126                  * This cluster will be used either for new bucket or for
2127                  * new xattr block.
2128                  * If the cluster size is the same as the bucket size, one
2129                  * more is needed since we may need to extend the bucket
2130                  * also.
2131                  */
2132                 clusters_add += 1;
2133                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2134                 if (OCFS2_XATTR_BUCKET_SIZE ==
2135                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2136                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2137                         clusters_add += 1;
2138                 }
2139         } else {
2140                 meta_add += 1;
2141                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2142         }
2143 out:
2144         if (clusters_need)
2145                 *clusters_need = clusters_add;
2146         if (meta_need)
2147                 *meta_need = meta_add;
2148         if (credits_need)
2149                 *credits_need = credits;
2150         brelse(bh);
2151         return ret;
2152 }
2153
2154 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2155                                      struct ocfs2_dinode *di,
2156                                      struct ocfs2_xattr_info *xi,
2157                                      struct ocfs2_xattr_search *xis,
2158                                      struct ocfs2_xattr_search *xbs,
2159                                      struct ocfs2_xattr_set_ctxt *ctxt,
2160                                      int *credits)
2161 {
2162         int clusters_add, meta_add, ret;
2163         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2164
2165         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2166
2167         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2168
2169         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2170                                         &clusters_add, &meta_add, credits);
2171         if (ret) {
2172                 mlog_errno(ret);
2173                 return ret;
2174         }
2175
2176         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2177              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2178
2179         if (meta_add) {
2180                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2181                                                         &ctxt->meta_ac);
2182                 if (ret) {
2183                         mlog_errno(ret);
2184                         goto out;
2185                 }
2186         }
2187
2188         if (clusters_add) {
2189                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2190                 if (ret)
2191                         mlog_errno(ret);
2192         }
2193 out:
2194         if (ret) {
2195                 if (ctxt->meta_ac) {
2196                         ocfs2_free_alloc_context(ctxt->meta_ac);
2197                         ctxt->meta_ac = NULL;
2198                 }
2199
2200                 /*
2201                  * We cannot have an error and a non null ctxt->data_ac.
2202                  */
2203         }
2204
2205         return ret;
2206 }
2207
2208 static int __ocfs2_xattr_set_handle(struct inode *inode,
2209                                     struct ocfs2_dinode *di,
2210                                     struct ocfs2_xattr_info *xi,
2211                                     struct ocfs2_xattr_search *xis,
2212                                     struct ocfs2_xattr_search *xbs,
2213                                     struct ocfs2_xattr_set_ctxt *ctxt)
2214 {
2215         int ret = 0, credits;
2216
2217         if (!xi->value) {
2218                 /* Remove existing extended attribute */
2219                 if (!xis->not_found)
2220                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2221                 else if (!xbs->not_found)
2222                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2223         } else {
2224                 /* We always try to set extended attribute into inode first*/
2225                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2226                 if (!ret && !xbs->not_found) {
2227                         /*
2228                          * If succeed and that extended attribute existing in
2229                          * external block, then we will remove it.
2230                          */
2231                         xi->value = NULL;
2232                         xi->value_len = 0;
2233
2234                         xis->not_found = -ENODATA;
2235                         ret = ocfs2_calc_xattr_set_need(inode,
2236                                                         di,
2237                                                         xi,
2238                                                         xis,
2239                                                         xbs,
2240                                                         NULL,
2241                                                         NULL,
2242                                                         &credits);
2243                         if (ret) {
2244                                 mlog_errno(ret);
2245                                 goto out;
2246                         }
2247
2248                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2249                                         ctxt->handle->h_buffer_credits);
2250                         if (ret) {
2251                                 mlog_errno(ret);
2252                                 goto out;
2253                         }
2254                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2255                 } else if (ret == -ENOSPC) {
2256                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2257                                 ret = ocfs2_xattr_block_find(inode,
2258                                                              xi->name_index,
2259                                                              xi->name, xbs);
2260                                 if (ret)
2261                                         goto out;
2262
2263                                 xis->not_found = -ENODATA;
2264                                 ret = ocfs2_calc_xattr_set_need(inode,
2265                                                                 di,
2266                                                                 xi,
2267                                                                 xis,
2268                                                                 xbs,
2269                                                                 NULL,
2270                                                                 NULL,
2271                                                                 &credits);
2272                                 if (ret) {
2273                                         mlog_errno(ret);
2274                                         goto out;
2275                                 }
2276
2277                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2278                                         ctxt->handle->h_buffer_credits);
2279                                 if (ret) {
2280                                         mlog_errno(ret);
2281                                         goto out;
2282                                 }
2283                         }
2284                         /*
2285                          * If no space in inode, we will set extended attribute
2286                          * into external block.
2287                          */
2288                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2289                         if (ret)
2290                                 goto out;
2291                         if (!xis->not_found) {
2292                                 /*
2293                                  * If succeed and that extended attribute
2294                                  * existing in inode, we will remove it.
2295                                  */
2296                                 xi->value = NULL;
2297                                 xi->value_len = 0;
2298                                 xbs->not_found = -ENODATA;
2299                                 ret = ocfs2_calc_xattr_set_need(inode,
2300                                                                 di,
2301                                                                 xi,
2302                                                                 xis,
2303                                                                 xbs,
2304                                                                 NULL,
2305                                                                 NULL,
2306                                                                 &credits);
2307                                 if (ret) {
2308                                         mlog_errno(ret);
2309                                         goto out;
2310                                 }
2311
2312                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2313                                                 ctxt->handle->h_buffer_credits);
2314                                 if (ret) {
2315                                         mlog_errno(ret);
2316                                         goto out;
2317                                 }
2318                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2319                                                             xis, ctxt);
2320                         }
2321                 }
2322         }
2323
2324 out:
2325         return ret;
2326 }
2327
2328 /*
2329  * ocfs2_xattr_set()
2330  *
2331  * Set, replace or remove an extended attribute for this inode.
2332  * value is NULL to remove an existing extended attribute, else either
2333  * create or replace an extended attribute.
2334  */
2335 int ocfs2_xattr_set(struct inode *inode,
2336                     int name_index,
2337                     const char *name,
2338                     const void *value,
2339                     size_t value_len,
2340                     int flags)
2341 {
2342         struct buffer_head *di_bh = NULL;
2343         struct ocfs2_dinode *di;
2344         int ret, credits;
2345         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2346         struct inode *tl_inode = osb->osb_tl_inode;
2347         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2348
2349         struct ocfs2_xattr_info xi = {
2350                 .name_index = name_index,
2351                 .name = name,
2352                 .value = value,
2353                 .value_len = value_len,
2354         };
2355
2356         struct ocfs2_xattr_search xis = {
2357                 .not_found = -ENODATA,
2358         };
2359
2360         struct ocfs2_xattr_search xbs = {
2361                 .not_found = -ENODATA,
2362         };
2363
2364         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2365                 return -EOPNOTSUPP;
2366
2367         /*
2368          * Only xbs will be used on indexed trees.  xis doesn't need a
2369          * bucket.
2370          */
2371         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2372         if (!xbs.bucket) {
2373                 mlog_errno(-ENOMEM);
2374                 return -ENOMEM;
2375         }
2376
2377         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2378         if (ret < 0) {
2379                 mlog_errno(ret);
2380                 goto cleanup_nolock;
2381         }
2382         xis.inode_bh = xbs.inode_bh = di_bh;
2383         di = (struct ocfs2_dinode *)di_bh->b_data;
2384
2385         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2386         /*
2387          * Scan inode and external block to find the same name
2388          * extended attribute and collect search infomation.
2389          */
2390         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2391         if (ret)
2392                 goto cleanup;
2393         if (xis.not_found) {
2394                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2395                 if (ret)
2396                         goto cleanup;
2397         }
2398
2399         if (xis.not_found && xbs.not_found) {
2400                 ret = -ENODATA;
2401                 if (flags & XATTR_REPLACE)
2402                         goto cleanup;
2403                 ret = 0;
2404                 if (!value)
2405                         goto cleanup;
2406         } else {
2407                 ret = -EEXIST;
2408                 if (flags & XATTR_CREATE)
2409                         goto cleanup;
2410         }
2411
2412
2413         mutex_lock(&tl_inode->i_mutex);
2414
2415         if (ocfs2_truncate_log_needs_flush(osb)) {
2416                 ret = __ocfs2_flush_truncate_log(osb);
2417                 if (ret < 0) {
2418                         mutex_unlock(&tl_inode->i_mutex);
2419                         mlog_errno(ret);
2420                         goto cleanup;
2421                 }
2422         }
2423         mutex_unlock(&tl_inode->i_mutex);
2424
2425         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2426                                         &xbs, &ctxt, &credits);
2427         if (ret) {
2428                 mlog_errno(ret);
2429                 goto cleanup;
2430         }
2431
2432         ctxt.handle = ocfs2_start_trans(osb, credits);
2433         if (IS_ERR(ctxt.handle)) {
2434                 ret = PTR_ERR(ctxt.handle);
2435                 mlog_errno(ret);
2436                 goto cleanup;
2437         }
2438
2439         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2440
2441         ocfs2_commit_trans(osb, ctxt.handle);
2442
2443         if (ctxt.data_ac)
2444                 ocfs2_free_alloc_context(ctxt.data_ac);
2445         if (ctxt.meta_ac)
2446                 ocfs2_free_alloc_context(ctxt.meta_ac);
2447         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2448                 ocfs2_schedule_truncate_log_flush(osb, 1);
2449         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2450 cleanup:
2451         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2452         ocfs2_inode_unlock(inode, 1);
2453 cleanup_nolock:
2454         brelse(di_bh);
2455         brelse(xbs.xattr_bh);
2456         ocfs2_xattr_bucket_free(xbs.bucket);
2457
2458         return ret;
2459 }
2460
2461 /*
2462  * Find the xattr extent rec which may contains name_hash.
2463  * e_cpos will be the first name hash of the xattr rec.
2464  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2465  */
2466 static int ocfs2_xattr_get_rec(struct inode *inode,
2467                                u32 name_hash,
2468                                u64 *p_blkno,
2469                                u32 *e_cpos,
2470                                u32 *num_clusters,
2471                                struct ocfs2_extent_list *el)
2472 {
2473         int ret = 0, i;
2474         struct buffer_head *eb_bh = NULL;
2475         struct ocfs2_extent_block *eb;
2476         struct ocfs2_extent_rec *rec = NULL;
2477         u64 e_blkno = 0;
2478
2479         if (el->l_tree_depth) {
2480                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2481                 if (ret) {
2482                         mlog_errno(ret);
2483                         goto out;
2484                 }
2485
2486                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2487                 el = &eb->h_list;
2488
2489                 if (el->l_tree_depth) {
2490                         ocfs2_error(inode->i_sb,
2491                                     "Inode %lu has non zero tree depth in "
2492                                     "xattr tree block %llu\n", inode->i_ino,
2493                                     (unsigned long long)eb_bh->b_blocknr);
2494                         ret = -EROFS;
2495                         goto out;
2496                 }
2497         }
2498
2499         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2500                 rec = &el->l_recs[i];
2501
2502                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2503                         e_blkno = le64_to_cpu(rec->e_blkno);
2504                         break;
2505                 }
2506         }
2507
2508         if (!e_blkno) {
2509                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2510                             "record (%u, %u, 0) in xattr", inode->i_ino,
2511                             le32_to_cpu(rec->e_cpos),
2512                             ocfs2_rec_clusters(el, rec));
2513                 ret = -EROFS;
2514                 goto out;
2515         }
2516
2517         *p_blkno = le64_to_cpu(rec->e_blkno);
2518         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2519         if (e_cpos)
2520                 *e_cpos = le32_to_cpu(rec->e_cpos);
2521 out:
2522         brelse(eb_bh);
2523         return ret;
2524 }
2525
2526 typedef int (xattr_bucket_func)(struct inode *inode,
2527                                 struct ocfs2_xattr_bucket *bucket,
2528                                 void *para);
2529
2530 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2531                                    struct ocfs2_xattr_bucket *bucket,
2532                                    int name_index,
2533                                    const char *name,
2534                                    u32 name_hash,
2535                                    u16 *xe_index,
2536                                    int *found)
2537 {
2538         int i, ret = 0, cmp = 1, block_off, new_offset;
2539         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2540         size_t name_len = strlen(name);
2541         struct ocfs2_xattr_entry *xe = NULL;
2542         char *xe_name;
2543
2544         /*
2545          * We don't use binary search in the bucket because there
2546          * may be multiple entries with the same name hash.
2547          */
2548         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2549                 xe = &xh->xh_entries[i];
2550
2551                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2552                         continue;
2553                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2554                         break;
2555
2556                 cmp = name_index - ocfs2_xattr_get_type(xe);
2557                 if (!cmp)
2558                         cmp = name_len - xe->xe_name_len;
2559                 if (cmp)
2560                         continue;
2561
2562                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2563                                                         xh,
2564                                                         i,
2565                                                         &block_off,
2566                                                         &new_offset);
2567                 if (ret) {
2568                         mlog_errno(ret);
2569                         break;
2570                 }
2571
2572                 xe_name = bucket_block(bucket, block_off) + new_offset;
2573                 if (!memcmp(name, xe_name, name_len)) {
2574                         *xe_index = i;
2575                         *found = 1;
2576                         ret = 0;
2577                         break;
2578                 }
2579         }
2580
2581         return ret;
2582 }
2583
2584 /*
2585  * Find the specified xattr entry in a series of buckets.
2586  * This series start from p_blkno and last for num_clusters.
2587  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2588  * the num of the valid buckets.
2589  *
2590  * Return the buffer_head this xattr should reside in. And if the xattr's
2591  * hash is in the gap of 2 buckets, return the lower bucket.
2592  */
2593 static int ocfs2_xattr_bucket_find(struct inode *inode,
2594                                    int name_index,
2595                                    const char *name,
2596                                    u32 name_hash,
2597                                    u64 p_blkno,
2598                                    u32 first_hash,
2599                                    u32 num_clusters,
2600                                    struct ocfs2_xattr_search *xs)
2601 {
2602         int ret, found = 0;
2603         struct ocfs2_xattr_header *xh = NULL;
2604         struct ocfs2_xattr_entry *xe = NULL;
2605         u16 index = 0;
2606         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2607         int low_bucket = 0, bucket, high_bucket;
2608         struct ocfs2_xattr_bucket *search;
2609         u32 last_hash;
2610         u64 blkno, lower_blkno = 0;
2611
2612         search = ocfs2_xattr_bucket_new(inode);
2613         if (!search) {
2614                 ret = -ENOMEM;
2615                 mlog_errno(ret);
2616                 goto out;
2617         }
2618
2619         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2620         if (ret) {
2621                 mlog_errno(ret);
2622                 goto out;
2623         }
2624
2625         xh = bucket_xh(search);
2626         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2627         while (low_bucket <= high_bucket) {
2628                 ocfs2_xattr_bucket_relse(search);
2629
2630                 bucket = (low_bucket + high_bucket) / 2;
2631                 blkno = p_blkno + bucket * blk_per_bucket;
2632                 ret = ocfs2_read_xattr_bucket(search, blkno);
2633                 if (ret) {
2634                         mlog_errno(ret);
2635                         goto out;
2636                 }
2637
2638                 xh = bucket_xh(search);
2639                 xe = &xh->xh_entries[0];
2640                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2641                         high_bucket = bucket - 1;
2642                         continue;
2643                 }
2644
2645                 /*
2646                  * Check whether the hash of the last entry in our
2647                  * bucket is larger than the search one. for an empty
2648                  * bucket, the last one is also the first one.
2649                  */
2650                 if (xh->xh_count)
2651                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2652
2653                 last_hash = le32_to_cpu(xe->xe_name_hash);
2654
2655                 /* record lower_blkno which may be the insert place. */
2656                 lower_blkno = blkno;
2657
2658                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2659                         low_bucket = bucket + 1;
2660                         continue;
2661                 }
2662
2663                 /* the searched xattr should reside in this bucket if exists. */
2664                 ret = ocfs2_find_xe_in_bucket(inode, search,
2665                                               name_index, name, name_hash,
2666                                               &index, &found);
2667                 if (ret) {
2668                         mlog_errno(ret);
2669                         goto out;
2670                 }
2671                 break;
2672         }
2673
2674         /*
2675          * Record the bucket we have found.
2676          * When the xattr's hash value is in the gap of 2 buckets, we will
2677          * always set it to the previous bucket.
2678          */
2679         if (!lower_blkno)
2680                 lower_blkno = p_blkno;
2681
2682         /* This should be in cache - we just read it during the search */
2683         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
2684         if (ret) {
2685                 mlog_errno(ret);
2686                 goto out;
2687         }
2688
2689         xs->header = bucket_xh(xs->bucket);
2690         xs->base = bucket_block(xs->bucket, 0);
2691         xs->end = xs->base + inode->i_sb->s_blocksize;
2692
2693         if (found) {
2694                 xs->here = &xs->header->xh_entries[index];
2695                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2696                      (unsigned long long)bucket_blkno(xs->bucket), index);
2697         } else
2698                 ret = -ENODATA;
2699
2700 out:
2701         ocfs2_xattr_bucket_free(search);
2702         return ret;
2703 }
2704
2705 static int ocfs2_xattr_index_block_find(struct inode *inode,
2706                                         struct buffer_head *root_bh,
2707                                         int name_index,
2708                                         const char *name,
2709                                         struct ocfs2_xattr_search *xs)
2710 {
2711         int ret;
2712         struct ocfs2_xattr_block *xb =
2713                         (struct ocfs2_xattr_block *)root_bh->b_data;
2714         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2715         struct ocfs2_extent_list *el = &xb_root->xt_list;
2716         u64 p_blkno = 0;
2717         u32 first_hash, num_clusters = 0;
2718         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2719
2720         if (le16_to_cpu(el->l_next_free_rec) == 0)
2721                 return -ENODATA;
2722
2723         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2724              name, name_hash, name_index);
2725
2726         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2727                                   &num_clusters, el);
2728         if (ret) {
2729                 mlog_errno(ret);
2730                 goto out;
2731         }
2732
2733         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2734
2735         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2736              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
2737              first_hash);
2738
2739         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2740                                       p_blkno, first_hash, num_clusters, xs);
2741
2742 out:
2743         return ret;
2744 }
2745
2746 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2747                                        u64 blkno,
2748                                        u32 clusters,
2749                                        xattr_bucket_func *func,
2750                                        void *para)
2751 {
2752         int i, ret = 0;
2753         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2754         u32 num_buckets = clusters * bpc;
2755         struct ocfs2_xattr_bucket *bucket;
2756
2757         bucket = ocfs2_xattr_bucket_new(inode);
2758         if (!bucket) {
2759                 mlog_errno(-ENOMEM);
2760                 return -ENOMEM;
2761         }
2762
2763         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2764              clusters, (unsigned long long)blkno);
2765
2766         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
2767                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
2768                 if (ret) {
2769                         mlog_errno(ret);
2770                         break;
2771                 }
2772
2773                 /*
2774                  * The real bucket num in this series of blocks is stored
2775                  * in the 1st bucket.
2776                  */
2777                 if (i == 0)
2778                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
2779
2780                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
2781                      (unsigned long long)blkno,
2782                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
2783                 if (func) {
2784                         ret = func(inode, bucket, para);
2785                         if (ret)
2786                                 mlog_errno(ret);
2787                         /* Fall through to bucket_relse() */
2788                 }
2789
2790                 ocfs2_xattr_bucket_relse(bucket);
2791                 if (ret)
2792                         break;
2793         }
2794
2795         ocfs2_xattr_bucket_free(bucket);
2796         return ret;
2797 }
2798
2799 struct ocfs2_xattr_tree_list {
2800         char *buffer;
2801         size_t buffer_size;
2802         size_t result;
2803 };
2804
2805 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2806                                              struct ocfs2_xattr_header *xh,
2807                                              int index,
2808                                              int *block_off,
2809                                              int *new_offset)
2810 {
2811         u16 name_offset;
2812
2813         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2814                 return -EINVAL;
2815
2816         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2817
2818         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2819         *new_offset = name_offset % inode->i_sb->s_blocksize;
2820
2821         return 0;
2822 }
2823
2824 static int ocfs2_list_xattr_bucket(struct inode *inode,
2825                                    struct ocfs2_xattr_bucket *bucket,
2826                                    void *para)
2827 {
2828         int ret = 0, type;
2829         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2830         int i, block_off, new_offset;
2831         const char *prefix, *name;
2832
2833         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
2834                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
2835                 type = ocfs2_xattr_get_type(entry);
2836                 prefix = ocfs2_xattr_prefix(type);
2837
2838                 if (prefix) {
2839                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2840                                                                 bucket_xh(bucket),
2841                                                                 i,
2842                                                                 &block_off,
2843                                                                 &new_offset);
2844                         if (ret)
2845                                 break;
2846
2847                         name = (const char *)bucket_block(bucket, block_off) +
2848                                 new_offset;
2849                         ret = ocfs2_xattr_list_entry(xl->buffer,
2850                                                      xl->buffer_size,
2851                                                      &xl->result,
2852                                                      prefix, name,
2853                                                      entry->xe_name_len);
2854                         if (ret)
2855                                 break;
2856                 }
2857         }
2858
2859         return ret;
2860 }
2861
2862 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2863                                              struct ocfs2_xattr_tree_root *xt,
2864                                              char *buffer,
2865                                              size_t buffer_size)
2866 {
2867         struct ocfs2_extent_list *el = &xt->xt_list;
2868         int ret = 0;
2869         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2870         u64 p_blkno = 0;
2871         struct ocfs2_xattr_tree_list xl = {
2872                 .buffer = buffer,
2873                 .buffer_size = buffer_size,
2874                 .result = 0,
2875         };
2876
2877         if (le16_to_cpu(el->l_next_free_rec) == 0)
2878                 return 0;
2879
2880         while (name_hash > 0) {
2881                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2882                                           &e_cpos, &num_clusters, el);
2883                 if (ret) {
2884                         mlog_errno(ret);
2885                         goto out;
2886                 }
2887
2888                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2889                                                   ocfs2_list_xattr_bucket,
2890                                                   &xl);
2891                 if (ret) {
2892                         mlog_errno(ret);
2893                         goto out;
2894                 }
2895
2896                 if (e_cpos == 0)
2897                         break;
2898
2899                 name_hash = e_cpos - 1;
2900         }
2901
2902         ret = xl.result;
2903 out:
2904         return ret;
2905 }
2906
2907 static int cmp_xe(const void *a, const void *b)
2908 {
2909         const struct ocfs2_xattr_entry *l = a, *r = b;
2910         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2911         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2912
2913         if (l_hash > r_hash)
2914                 return 1;
2915         if (l_hash < r_hash)
2916                 return -1;
2917         return 0;
2918 }
2919
2920 static void swap_xe(void *a, void *b, int size)
2921 {
2922         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2923
2924         tmp = *l;
2925         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2926         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2927 }
2928
2929 /*
2930  * When the ocfs2_xattr_block is filled up, new bucket will be created
2931  * and all the xattr entries will be moved to the new bucket.
2932  * The header goes at the start of the bucket, and the names+values are
2933  * filled from the end.  This is why *target starts as the last buffer.
2934  * Note: we need to sort the entries since they are not saved in order
2935  * in the ocfs2_xattr_block.
2936  */
2937 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2938                                            struct buffer_head *xb_bh,
2939                                            struct ocfs2_xattr_bucket *bucket)
2940 {
2941         int i, blocksize = inode->i_sb->s_blocksize;
2942         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2943         u16 offset, size, off_change;
2944         struct ocfs2_xattr_entry *xe;
2945         struct ocfs2_xattr_block *xb =
2946                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2947         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2948         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2949         u16 count = le16_to_cpu(xb_xh->xh_count);
2950         char *src = xb_bh->b_data;
2951         char *target = bucket_block(bucket, blks - 1);
2952
2953         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2954              (unsigned long long)xb_bh->b_blocknr,
2955              (unsigned long long)bucket_blkno(bucket));
2956
2957         for (i = 0; i < blks; i++)
2958                 memset(bucket_block(bucket, i), 0, blocksize);
2959
2960         /*
2961          * Since the xe_name_offset is based on ocfs2_xattr_header,
2962          * there is a offset change corresponding to the change of
2963          * ocfs2_xattr_header's position.
2964          */
2965         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2966         xe = &xb_xh->xh_entries[count - 1];
2967         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2968         size = blocksize - offset;
2969
2970         /* copy all the names and values. */
2971         memcpy(target + offset, src + offset, size);
2972
2973         /* Init new header now. */
2974         xh->xh_count = xb_xh->xh_count;
2975         xh->xh_num_buckets = cpu_to_le16(1);
2976         xh->xh_name_value_len = cpu_to_le16(size);
2977         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2978
2979         /* copy all the entries. */
2980         target = bucket_block(bucket, 0);
2981         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2982         size = count * sizeof(struct ocfs2_xattr_entry);
2983         memcpy(target + offset, (char *)xb_xh + offset, size);
2984
2985         /* Change the xe offset for all the xe because of the move. */
2986         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2987                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2988         for (i = 0; i < count; i++)
2989                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2990
2991         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2992              offset, size, off_change);
2993
2994         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2995              cmp_xe, swap_xe);
2996 }
2997
2998 /*
2999  * After we move xattr from block to index btree, we have to
3000  * update ocfs2_xattr_search to the new xe and base.
3001  *
3002  * When the entry is in xattr block, xattr_bh indicates the storage place.
3003  * While if the entry is in index b-tree, "bucket" indicates the
3004  * real place of the xattr.
3005  */
3006 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3007                                             struct ocfs2_xattr_search *xs,
3008                                             struct buffer_head *old_bh)
3009 {
3010         char *buf = old_bh->b_data;
3011         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3012         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3013         int i;
3014
3015         xs->header = bucket_xh(xs->bucket);
3016         xs->base = bucket_block(xs->bucket, 0);
3017         xs->end = xs->base + inode->i_sb->s_blocksize;
3018
3019         if (xs->not_found)
3020                 return;
3021
3022         i = xs->here - old_xh->xh_entries;
3023         xs->here = &xs->header->xh_entries[i];
3024 }
3025
3026 static int ocfs2_xattr_create_index_block(struct inode *inode,
3027                                           struct ocfs2_xattr_search *xs,
3028                                           struct ocfs2_xattr_set_ctxt *ctxt)
3029 {
3030         int ret;
3031         u32 bit_off, len;
3032         u64 blkno;
3033         handle_t *handle = ctxt->handle;
3034         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3035         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3036         struct buffer_head *xb_bh = xs->xattr_bh;
3037         struct ocfs2_xattr_block *xb =
3038                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3039         struct ocfs2_xattr_tree_root *xr;
3040         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3041
3042         mlog(0, "create xattr index block for %llu\n",
3043              (unsigned long long)xb_bh->b_blocknr);
3044
3045         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3046         BUG_ON(!xs->bucket);
3047
3048         /*
3049          * XXX:
3050          * We can use this lock for now, and maybe move to a dedicated mutex
3051          * if performance becomes a problem later.
3052          */
3053         down_write(&oi->ip_alloc_sem);
3054
3055         ret = ocfs2_journal_access(handle, inode, xb_bh,
3056                                    OCFS2_JOURNAL_ACCESS_WRITE);
3057         if (ret) {
3058                 mlog_errno(ret);
3059                 goto out;
3060         }
3061
3062         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3063                                      1, 1, &bit_off, &len);
3064         if (ret) {
3065                 mlog_errno(ret);
3066                 goto out;
3067         }
3068
3069         /*
3070          * The bucket may spread in many blocks, and
3071          * we will only touch the 1st block and the last block
3072          * in the whole bucket(one for entry and one for data).
3073          */
3074         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3075
3076         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3077              (unsigned long long)blkno);
3078
3079         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3080         if (ret) {
3081                 mlog_errno(ret);
3082                 goto out;
3083         }
3084
3085         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3086                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3087         if (ret) {
3088                 mlog_errno(ret);
3089                 goto out;
3090         }
3091
3092         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3093         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3094
3095         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3096
3097         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3098         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3099                offsetof(struct ocfs2_xattr_block, xb_attrs));
3100
3101         xr = &xb->xb_attrs.xb_root;
3102         xr->xt_clusters = cpu_to_le32(1);
3103         xr->xt_last_eb_blk = 0;
3104         xr->xt_list.l_tree_depth = 0;
3105         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3106         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3107
3108         xr->xt_list.l_recs[0].e_cpos = 0;
3109         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3110         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3111
3112         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3113
3114         ocfs2_journal_dirty(handle, xb_bh);
3115
3116 out:
3117         up_write(&oi->ip_alloc_sem);
3118
3119         return ret;
3120 }
3121
3122 static int cmp_xe_offset(const void *a, const void *b)
3123 {
3124         const struct ocfs2_xattr_entry *l = a, *r = b;
3125         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3126         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3127
3128         if (l_name_offset < r_name_offset)
3129                 return 1;
3130         if (l_name_offset > r_name_offset)
3131                 return -1;
3132         return 0;
3133 }
3134
3135 /*
3136  * defrag a xattr bucket if we find that the bucket has some
3137  * holes beteen name/value pairs.
3138  * We will move all the name/value pairs to the end of the bucket
3139  * so that we can spare some space for insertion.
3140  */
3141 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3142                                      handle_t *handle,
3143                                      struct ocfs2_xattr_bucket *bucket)
3144 {
3145         int ret, i;
3146         size_t end, offset, len, value_len;
3147         struct ocfs2_xattr_header *xh;
3148         char *entries, *buf, *bucket_buf = NULL;
3149         u64 blkno = bucket_blkno(bucket);
3150         u16 xh_free_start;
3151         size_t blocksize = inode->i_sb->s_blocksize;
3152         struct ocfs2_xattr_entry *xe;
3153
3154         /*
3155          * In order to make the operation more efficient and generic,
3156          * we copy all the blocks into a contiguous memory and do the
3157          * defragment there, so if anything is error, we will not touch
3158          * the real block.
3159          */
3160         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3161         if (!bucket_buf) {
3162                 ret = -EIO;
3163                 goto out;
3164         }
3165
3166         buf = bucket_buf;
3167         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3168                 memcpy(buf, bucket_block(bucket, i), blocksize);
3169
3170         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3171                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3172         if (ret < 0) {
3173                 mlog_errno(ret);
3174                 goto out;
3175         }
3176
3177         xh = (struct ocfs2_xattr_header *)bucket_buf;
3178         entries = (char *)xh->xh_entries;
3179         xh_free_start = le16_to_cpu(xh->xh_free_start);
3180
3181         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3182              "xh_free_start = %u, xh_name_value_len = %u.\n",
3183              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3184              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3185
3186         /*
3187          * sort all the entries by their offset.
3188          * the largest will be the first, so that we can
3189          * move them to the end one by one.
3190          */
3191         sort(entries, le16_to_cpu(xh->xh_count),
3192              sizeof(struct ocfs2_xattr_entry),
3193              cmp_xe_offset, swap_xe);
3194
3195         /* Move all name/values to the end of the bucket. */
3196         xe = xh->xh_entries;
3197         end = OCFS2_XATTR_BUCKET_SIZE;
3198         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3199                 offset = le16_to_cpu(xe->xe_name_offset);
3200                 if (ocfs2_xattr_is_local(xe))
3201                         value_len = OCFS2_XATTR_SIZE(
3202                                         le64_to_cpu(xe->xe_value_size));
3203                 else
3204                         value_len = OCFS2_XATTR_ROOT_SIZE;
3205                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3206
3207                 /*
3208                  * We must make sure that the name/value pair
3209                  * exist in the same block. So adjust end to
3210                  * the previous block end if needed.
3211                  */
3212                 if (((end - len) / blocksize !=
3213                         (end - 1) / blocksize))
3214                         end = end - end % blocksize;
3215
3216                 if (end > offset + len) {
3217                         memmove(bucket_buf + end - len,
3218                                 bucket_buf + offset, len);
3219                         xe->xe_name_offset = cpu_to_le16(end - len);
3220                 }
3221
3222                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3223                                 "bucket %llu\n", (unsigned long long)blkno);
3224
3225                 end -= len;
3226         }
3227
3228         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3229                         "bucket %llu\n", (unsigned long long)blkno);
3230
3231         if (xh_free_start == end)
3232                 goto out;
3233
3234         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3235         xh->xh_free_start = cpu_to_le16(end);
3236
3237         /* sort the entries by their name_hash. */
3238         sort(entries, le16_to_cpu(xh->xh_count),
3239              sizeof(struct ocfs2_xattr_entry),
3240              cmp_xe, swap_xe);
3241
3242         buf = bucket_buf;
3243         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3244                 memcpy(bucket_block(bucket, i), buf, blocksize);
3245         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3246
3247 out:
3248         kfree(bucket_buf);
3249         return ret;
3250 }
3251
3252 /*
3253  * Move half nums of the xattr bucket in the previous cluster to this new
3254  * cluster. We only touch the last cluster of the previous extend record.
3255  *
3256  * first_bh is the first buffer_head of a series of bucket in the same
3257  * extent rec and header_bh is the header of one bucket in this cluster.
3258  * They will be updated if we move the data header_bh contains to the new
3259  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3260  */
3261 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3262                                                handle_t *handle,
3263                                                struct buffer_head **first_bh,
3264                                                struct buffer_head **header_bh,
3265                                                u64 new_blkno,
3266                                                u64 prev_blkno,
3267                                                u32 num_clusters,
3268                                                u32 *first_hash)
3269 {
3270         int i, ret, credits;
3271         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3272         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3273         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3274         int blocksize = inode->i_sb->s_blocksize;
3275         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3276         struct ocfs2_xattr_header *new_xh;
3277         struct ocfs2_xattr_header *xh =
3278                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3279
3280         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3281         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3282
3283         prev_bh = *first_bh;
3284         get_bh(prev_bh);
3285         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3286
3287         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3288
3289         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3290              (unsigned long long)prev_blkno, (unsigned long long)new_blkno);
3291
3292         /*
3293          * We need to update the 1st half of the new cluster and
3294          * 1 more for the update of the 1st bucket of the previous
3295          * extent record.
3296          */
3297         credits = bpc / 2 + 1 + handle->h_buffer_credits;
3298         ret = ocfs2_extend_trans(handle, credits);
3299         if (ret) {
3300                 mlog_errno(ret);
3301                 goto out;
3302         }
3303
3304         ret = ocfs2_journal_access(handle, inode, prev_bh,
3305                                    OCFS2_JOURNAL_ACCESS_WRITE);
3306         if (ret) {
3307                 mlog_errno(ret);
3308                 goto out;
3309         }
3310
3311         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3312                 old_bh = new_bh = NULL;
3313                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3314                 if (!new_bh) {
3315                         ret = -EIO;
3316                         mlog_errno(ret);
3317                         goto out;
3318                 }
3319
3320                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3321
3322                 ret = ocfs2_journal_access(handle, inode, new_bh,
3323                                            OCFS2_JOURNAL_ACCESS_CREATE);
3324                 if (ret < 0) {
3325                         mlog_errno(ret);
3326                         brelse(new_bh);
3327                         goto out;
3328                 }
3329
3330                 ret = ocfs2_read_block(inode, prev_blkno, &old_bh);
3331                 if (ret < 0) {
3332                         mlog_errno(ret);
3333                         brelse(new_bh);
3334                         goto out;
3335                 }
3336
3337                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3338
3339                 if (i == 0) {
3340                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3341                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3342
3343                         if (first_hash)
3344                                 *first_hash = le32_to_cpu(
3345                                         new_xh->xh_entries[0].xe_name_hash);
3346                         new_first_bh = new_bh;
3347                         get_bh(new_first_bh);
3348                 }
3349
3350                 ocfs2_journal_dirty(handle, new_bh);
3351
3352                 if (*header_bh == old_bh) {
3353                         brelse(*header_bh);
3354                         *header_bh = new_bh;
3355                         get_bh(*header_bh);
3356
3357                         brelse(*first_bh);
3358                         *first_bh = new_first_bh;
3359                         get_bh(*first_bh);
3360                 }
3361                 brelse(new_bh);
3362                 brelse(old_bh);
3363         }
3364
3365         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3366
3367         ocfs2_journal_dirty(handle, prev_bh);
3368 out:
3369         brelse(prev_bh);
3370         brelse(new_first_bh);
3371         return ret;
3372 }
3373
3374 /*
3375  * Find the suitable pos when we divide a bucket into 2.
3376  * We have to make sure the xattrs with the same hash value exist
3377  * in the same bucket.
3378  *
3379  * If this ocfs2_xattr_header covers more than one hash value, find a
3380  * place where the hash value changes.  Try to find the most even split.
3381  * The most common case is that all entries have different hash values,
3382  * and the first check we make will find a place to split.
3383  */
3384 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3385 {
3386         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3387         int count = le16_to_cpu(xh->xh_count);
3388         int delta, middle = count / 2;
3389
3390         /*
3391          * We start at the middle.  Each step gets farther away in both
3392          * directions.  We therefore hit the change in hash value
3393          * nearest to the middle.  Note that this loop does not execute for
3394          * count < 2.
3395          */
3396         for (delta = 0; delta < middle; delta++) {
3397                 /* Let's check delta earlier than middle */
3398                 if (cmp_xe(&entries[middle - delta - 1],
3399                            &entries[middle - delta]))
3400                         return middle - delta;
3401
3402                 /* For even counts, don't walk off the end */
3403                 if ((middle + delta + 1) == count)
3404                         continue;
3405
3406                 /* Now try delta past middle */
3407                 if (cmp_xe(&entries[middle + delta],
3408                            &entries[middle + delta + 1]))
3409                         return middle + delta + 1;
3410         }
3411
3412         /* Every entry had the same hash */
3413         return count;
3414 }
3415
3416 /*
3417  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3418  * first_hash will record the 1st hash of the new bucket.
3419  *
3420  * Normally half of the xattrs will be moved.  But we have to make
3421  * sure that the xattrs with the same hash value are stored in the
3422  * same bucket. If all the xattrs in this bucket have the same hash
3423  * value, the new bucket will be initialized as an empty one and the
3424  * first_hash will be initialized as (hash_value+1).
3425  */
3426 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3427                                     handle_t *handle,
3428                                     u64 blk,
3429                                     u64 new_blk,
3430                                     u32 *first_hash,
3431                                     int new_bucket_head)
3432 {
3433         int ret, i;
3434         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3435         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3436         struct ocfs2_xattr_header *xh;
3437         struct ocfs2_xattr_entry *xe;
3438         int blocksize = inode->i_sb->s_blocksize;
3439
3440         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3441              (unsigned long long)blk, (unsigned long long)new_blk);
3442
3443         s_bucket = ocfs2_xattr_bucket_new(inode);
3444         t_bucket = ocfs2_xattr_bucket_new(inode);
3445         if (!s_bucket || !t_bucket) {
3446                 ret = -ENOMEM;
3447                 mlog_errno(ret);
3448                 goto out;
3449         }
3450
3451         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3452         if (ret) {
3453                 mlog_errno(ret);
3454                 goto out;
3455         }
3456
3457         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3458                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3459         if (ret) {
3460                 mlog_errno(ret);
3461                 goto out;
3462         }
3463
3464         /*
3465          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3466          * there's no need to read it.
3467          */
3468         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3469         if (ret) {
3470                 mlog_errno(ret);
3471                 goto out;
3472         }
3473
3474         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3475                                                 new_bucket_head ?
3476                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3477                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3478         if (ret) {
3479                 mlog_errno(ret);
3480                 goto out;
3481         }
3482
3483         xh = bucket_xh(s_bucket);
3484         count = le16_to_cpu(xh->xh_count);
3485         start = ocfs2_xattr_find_divide_pos(xh);
3486
3487         if (start == count) {
3488                 xe = &xh->xh_entries[start-1];
3489
3490                 /*
3491                  * initialized a new empty bucket here.
3492                  * The hash value is set as one larger than
3493                  * that of the last entry in the previous bucket.
3494                  */
3495                 for (i = 0; i < t_bucket->bu_blocks; i++)
3496                         memset(bucket_block(t_bucket, i), 0, blocksize);
3497
3498                 xh = bucket_xh(t_bucket);
3499                 xh->xh_free_start = cpu_to_le16(blocksize);
3500                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3501                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3502
3503                 goto set_num_buckets;
3504         }
3505
3506         /* copy the whole bucket to the new first. */
3507         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3508
3509         /* update the new bucket. */
3510         xh = bucket_xh(t_bucket);
3511
3512         /*
3513          * Calculate the total name/value len and xh_free_start for
3514          * the old bucket first.
3515          */
3516         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3517         name_value_len = 0;
3518         for (i = 0; i < start; i++) {
3519                 xe = &xh->xh_entries[i];
3520                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3521                 if (ocfs2_xattr_is_local(xe))
3522                         xe_len +=
3523                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3524                 else
3525                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3526                 name_value_len += xe_len;
3527                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3528                         name_offset = le16_to_cpu(xe->xe_name_offset);
3529         }
3530
3531         /*
3532          * Now begin the modification to the new bucket.
3533          *
3534          * In the new bucket, We just move the xattr entry to the beginning
3535          * and don't touch the name/value. So there will be some holes in the
3536          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3537          * called.
3538          */
3539         xe = &xh->xh_entries[start];
3540         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3541         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3542              (int)((char *)xe - (char *)xh),
3543              (int)((char *)xh->xh_entries - (char *)xh));
3544         memmove((char *)xh->xh_entries, (char *)xe, len);
3545         xe = &xh->xh_entries[count - start];
3546         len = sizeof(struct ocfs2_xattr_entry) * start;
3547         memset((char *)xe, 0, len);
3548
3549         le16_add_cpu(&xh->xh_count, -start);
3550         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3551
3552         /* Calculate xh_free_start for the new bucket. */
3553         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3554         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3555                 xe = &xh->xh_entries[i];
3556                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3557                 if (ocfs2_xattr_is_local(xe))
3558                         xe_len +=
3559                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3560                 else
3561                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3562                 if (le16_to_cpu(xe->xe_name_offset) <
3563                     le16_to_cpu(xh->xh_free_start))
3564                         xh->xh_free_start = xe->xe_name_offset;
3565         }
3566
3567 set_num_buckets:
3568         /* set xh->xh_num_buckets for the new xh. */
3569         if (new_bucket_head)
3570                 xh->xh_num_buckets = cpu_to_le16(1);
3571         else
3572                 xh->xh_num_buckets = 0;
3573
3574         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3575
3576         /* store the first_hash of the new bucket. */
3577         if (first_hash)
3578                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3579
3580         /*
3581          * Now only update the 1st block of the old bucket.  If we
3582          * just added a new empty bucket, there is no need to modify
3583          * it.
3584          */
3585         if (start == count)
3586                 goto out;
3587
3588         xh = bucket_xh(s_bucket);
3589         memset(&xh->xh_entries[start], 0,
3590                sizeof(struct ocfs2_xattr_entry) * (count - start));
3591         xh->xh_count = cpu_to_le16(start);
3592         xh->xh_free_start = cpu_to_le16(name_offset);
3593         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3594
3595         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3596
3597 out:
3598         ocfs2_xattr_bucket_free(s_bucket);
3599         ocfs2_xattr_bucket_free(t_bucket);
3600
3601         return ret;
3602 }
3603
3604 /*
3605  * Copy xattr from one bucket to another bucket.
3606  *
3607  * The caller must make sure that the journal transaction
3608  * has enough space for journaling.
3609  */
3610 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3611                                  handle_t *handle,
3612                                  u64 s_blkno,
3613                                  u64 t_blkno,
3614                                  int t_is_new)
3615 {
3616         int ret;
3617         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3618
3619         BUG_ON(s_blkno == t_blkno);
3620
3621         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3622              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3623              t_is_new);
3624
3625         s_bucket = ocfs2_xattr_bucket_new(inode);
3626         t_bucket = ocfs2_xattr_bucket_new(inode);
3627         if (!s_bucket || !t_bucket) {
3628                 ret = -ENOMEM;
3629                 mlog_errno(ret);
3630                 goto out;
3631         }
3632   
3633         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3634         if (ret)
3635                 goto out;
3636
3637         /*
3638          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3639          * there's no need to read it.
3640          */
3641         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3642         if (ret)
3643                 goto out;
3644
3645         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3646                                                 t_is_new ?
3647                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3648                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3649         if (ret)
3650                 goto out;
3651
3652         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3653         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3654
3655 out:
3656         ocfs2_xattr_bucket_free(t_bucket);
3657         ocfs2_xattr_bucket_free(s_bucket);
3658
3659         return ret;
3660 }
3661
3662 /*
3663  * Copy one xattr cluster from src_blk to to_blk.
3664  * The to_blk will become the first bucket header of the cluster, so its
3665  * xh_num_buckets will be initialized as the bucket num in the cluster.
3666  */
3667 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3668                                   handle_t *handle,
3669                                   struct buffer_head *first_bh,
3670                                   u64 src_blk,
3671                                   u64 to_blk,
3672                                   u32 *first_hash)
3673 {
3674         int i, ret, credits;
3675         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3676         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3677         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3678         struct buffer_head *bh = NULL;
3679         struct ocfs2_xattr_header *xh;
3680         u64 to_blk_start = to_blk;
3681
3682         mlog(0, "cp xattrs from cluster %llu to %llu\n",
3683              (unsigned long long)src_blk, (unsigned long long)to_blk);
3684
3685         /*
3686          * We need to update the new cluster and 1 more for the update of
3687          * the 1st bucket of the previous extent rec.
3688          */
3689         credits = bpc + 1 + handle->h_buffer_credits;
3690         ret = ocfs2_extend_trans(handle, credits);
3691         if (ret) {
3692                 mlog_errno(ret);
3693                 goto out;
3694         }
3695
3696         ret = ocfs2_journal_access(handle, inode, first_bh,
3697                                    OCFS2_JOURNAL_ACCESS_WRITE);
3698         if (ret) {
3699                 mlog_errno(ret);
3700                 goto out;
3701         }
3702
3703         for (i = 0; i < num_buckets; i++) {
3704                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3705                                             src_blk, to_blk, 1);
3706                 if (ret) {
3707                         mlog_errno(ret);
3708                         goto out;
3709                 }
3710
3711                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3712                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3713         }
3714
3715         /* update the old bucket header. */
3716         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3717         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3718
3719         ocfs2_journal_dirty(handle, first_bh);
3720
3721         /* update the new bucket header. */
3722         ret = ocfs2_read_block(inode, to_blk_start, &bh);
3723         if (ret < 0) {
3724                 mlog_errno(ret);
3725                 goto out;
3726         }
3727
3728         ret = ocfs2_journal_access(handle, inode, bh,
3729                                    OCFS2_JOURNAL_ACCESS_WRITE);
3730         if (ret) {
3731                 mlog_errno(ret);
3732                 goto out;
3733         }
3734
3735         xh = (struct ocfs2_xattr_header *)bh->b_data;
3736         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3737
3738         ocfs2_journal_dirty(handle, bh);
3739
3740         if (first_hash)
3741                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3742 out:
3743         brelse(bh);
3744         return ret;
3745 }
3746
3747 /*
3748  * Move some xattrs in this cluster to the new cluster.
3749  * This function should only be called when bucket size == cluster size.
3750  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3751  */
3752 static int ocfs2_divide_xattr_cluster(struct inode *inode,
3753                                       handle_t *handle,
3754                                       u64 prev_blk,
3755                                       u64 new_blk,
3756                                       u32 *first_hash)
3757 {
3758         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3759         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
3760
3761         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3762
3763         ret = ocfs2_extend_trans(handle, credits);
3764         if (ret) {
3765                 mlog_errno(ret);
3766                 return ret;
3767         }
3768
3769         /* Move half of the xattr in start_blk to the next bucket. */
3770         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
3771                                           new_blk, first_hash, 1);
3772 }
3773
3774 /*
3775  * Move some xattrs from the old cluster to the new one since they are not
3776  * contiguous in ocfs2 xattr tree.
3777  *
3778  * new_blk starts a new separate cluster, and we will move some xattrs from
3779  * prev_blk to it. v_start will be set as the first name hash value in this
3780  * new cluster so that it can be used as e_cpos during tree insertion and
3781  * don't collide with our original b-tree operations. first_bh and header_bh
3782  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3783  * to extend the insert bucket.
3784  *
3785  * The problem is how much xattr should we move to the new one and when should
3786  * we update first_bh and header_bh?
3787  * 1. If cluster size > bucket size, that means the previous cluster has more
3788  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3789  *    update the first_bh and header_bh if the insert bucket has been moved
3790  *    to the new cluster.
3791  * 2. If cluster_size == bucket_size:
3792  *    a) If the previous extent rec has more than one cluster and the insert
3793  *       place isn't in the last cluster, copy the entire last cluster to the
3794  *       new one. This time, we don't need to upate the first_bh and header_bh
3795  *       since they will not be moved into the new cluster.
3796  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3797  *       the new one. And we set the extend flag to zero if the insert place is
3798  *       moved into the new allocated cluster since no extend is needed.
3799  */
3800 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3801                                             handle_t *handle,
3802                                             struct buffer_head **first_bh,
3803                                             struct buffer_head **header_bh,
3804                                             u64 new_blk,
3805                                             u64 prev_blk,
3806                                             u32 prev_clusters,
3807                                             u32 *v_start,
3808                                             int *extend)
3809 {
3810         int ret = 0;
3811         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3812
3813         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3814              (unsigned long long)prev_blk, prev_clusters,
3815              (unsigned long long)new_blk);
3816
3817         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3818                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3819                                                           handle,
3820                                                           first_bh,
3821                                                           header_bh,
3822                                                           new_blk,
3823                                                           prev_blk,
3824                                                           prev_clusters,
3825                                                           v_start);
3826         else {
3827                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3828
3829                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3830                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3831                                                      last_blk, new_blk,
3832                                                      v_start);
3833                 else {
3834                         ret = ocfs2_divide_xattr_cluster(inode, handle,
3835                                                          last_blk, new_blk,
3836                                                          v_start);
3837
3838                         if ((*header_bh)->b_blocknr == last_blk && extend)
3839                                 *extend = 0;
3840                 }
3841         }
3842
3843         return ret;
3844 }
3845
3846 /*
3847  * Add a new cluster for xattr storage.
3848  *
3849  * If the new cluster is contiguous with the previous one, it will be
3850  * appended to the same extent record, and num_clusters will be updated.
3851  * If not, we will insert a new extent for it and move some xattrs in
3852  * the last cluster into the new allocated one.
3853  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3854  * lose the benefits of hashing because we'll have to search large leaves.
3855  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3856  * if it's bigger).
3857  *
3858  * first_bh is the first block of the previous extent rec and header_bh
3859  * indicates the bucket we will insert the new xattrs. They will be updated
3860  * when the header_bh is moved into the new cluster.
3861  */
3862 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3863                                        struct buffer_head *root_bh,
3864                                        struct buffer_head **first_bh,
3865                                        struct buffer_head **header_bh,
3866                                        u32 *num_clusters,
3867                                        u32 prev_cpos,
3868                                        u64 prev_blkno,
3869                                        int *extend,
3870                                        struct ocfs2_xattr_set_ctxt *ctxt)
3871 {
3872         int ret;
3873         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3874         u32 prev_clusters = *num_clusters;
3875         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3876         u64 block;
3877         handle_t *handle = ctxt->handle;
3878         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3879         struct ocfs2_extent_tree et;
3880
3881         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3882              "previous xattr blkno = %llu\n",
3883              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3884              prev_cpos, (unsigned long long)prev_blkno);
3885
3886         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3887
3888         ret = ocfs2_journal_access(handle, inode, root_bh,
3889                                    OCFS2_JOURNAL_ACCESS_WRITE);
3890         if (ret < 0) {
3891                 mlog_errno(ret);
3892                 goto leave;
3893         }
3894
3895         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
3896                                      clusters_to_add, &bit_off, &num_bits);
3897         if (ret < 0) {
3898                 if (ret != -ENOSPC)
3899                         mlog_errno(ret);
3900                 goto leave;
3901         }
3902
3903         BUG_ON(num_bits > clusters_to_add);
3904
3905         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3906         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3907              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3908
3909         if (prev_blkno + prev_clusters * bpc == block &&
3910             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3911              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3912                 /*
3913                  * If this cluster is contiguous with the old one and
3914                  * adding this new cluster, we don't surpass the limit of
3915                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3916                  * initialized and used like other buckets in the previous
3917                  * cluster.
3918                  * So add it as a contiguous one. The caller will handle
3919                  * its init process.
3920                  */
3921                 v_start = prev_cpos + prev_clusters;
3922                 *num_clusters = prev_clusters + num_bits;
3923                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3924                      num_bits);
3925         } else {
3926                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3927                                                        handle,
3928                                                        first_bh,
3929                                                        header_bh,
3930                                                        block,
3931                                                        prev_blkno,
3932                                                        prev_clusters,
3933                                                        &v_start,
3934                                                        extend);
3935                 if (ret) {
3936                         mlog_errno(ret);
3937                         goto leave;
3938                 }
3939         }
3940
3941         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3942              num_bits, (unsigned long long)block, v_start);
3943         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3944                                   num_bits, 0, ctxt->meta_ac);
3945         if (ret < 0) {
3946                 mlog_errno(ret);
3947                 goto leave;
3948         }
3949
3950         ret = ocfs2_journal_dirty(handle, root_bh);
3951         if (ret < 0)
3952                 mlog_errno(ret);
3953
3954 leave:
3955         return ret;
3956 }
3957
3958 /*
3959  * Extend a new xattr bucket and move xattrs to the end one by one until
3960  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3961  */
3962 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3963                                      handle_t *handle,
3964                                      struct buffer_head *first_bh,
3965                                      struct buffer_head *start_bh,
3966                                      u32 num_clusters)
3967 {
3968         int ret, credits;
3969         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3970         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3971         u64 start_blk = start_bh->b_blocknr, end_blk;
3972         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3973         struct ocfs2_xattr_header *first_xh =
3974                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3975         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3976
3977         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3978              "from %llu, len = %u\n", (unsigned long long)start_blk,
3979              (unsigned long long)first_bh->b_blocknr, num_clusters);
3980
3981         BUG_ON(bucket >= num_buckets);
3982
3983         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3984
3985         /*
3986          * We will touch all the buckets after the start_bh(include it).
3987          * Then we add one more bucket.
3988          */
3989         credits = end_blk - start_blk + 3 * blk_per_bucket + 1 +
3990                   handle->h_buffer_credits;
3991         ret = ocfs2_extend_trans(handle, credits);
3992         if (ret) {
3993                 mlog_errno(ret);
3994                 goto out;
3995         }
3996
3997         ret = ocfs2_journal_access(handle, inode, first_bh,
3998                                    OCFS2_JOURNAL_ACCESS_WRITE);
3999         if (ret) {
4000                 mlog_errno(ret);
4001                 goto out;
4002         }
4003
4004         while (end_blk != start_blk) {
4005                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4006                                             end_blk + blk_per_bucket, 0);
4007                 if (ret)
4008                         goto out;
4009                 end_blk -= blk_per_bucket;
4010         }
4011
4012         /* Move half of the xattr in start_blk to the next bucket. */
4013         ret = ocfs2_divide_xattr_bucket(inode, handle, start_blk,
4014                                         start_blk + blk_per_bucket, NULL, 0);
4015
4016         le16_add_cpu(&first_xh->xh_num_buckets, 1);
4017         ocfs2_journal_dirty(handle, first_bh);
4018
4019 out:
4020         return ret;
4021 }
4022
4023 /*
4024  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
4025  * xb_bh is the ocfs2_xattr_block.
4026  * We will move all the buckets starting from header_bh to the next place. As
4027  * for this one, half num of its xattrs will be moved to the next one.
4028  *
4029  * We will allocate a new cluster if current cluster is full and adjust
4030  * header_bh and first_bh if the insert place is moved to the new cluster.
4031  */
4032 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4033                                       struct buffer_head *xb_bh,
4034                                       struct buffer_head *header_bh,
4035                                       struct ocfs2_xattr_set_ctxt *ctxt)
4036 {
4037         struct ocfs2_xattr_header *first_xh = NULL;
4038         struct buffer_head *first_bh = NULL;
4039         struct ocfs2_xattr_block *xb =
4040                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4041         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4042         struct ocfs2_extent_list *el = &xb_root->xt_list;
4043         struct ocfs2_xattr_header *xh =
4044                         (struct ocfs2_xattr_header *)header_bh->b_data;
4045         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4046         struct super_block *sb = inode->i_sb;
4047         struct ocfs2_super *osb = OCFS2_SB(sb);
4048         int ret, num_buckets, extend = 1;
4049         u64 p_blkno;
4050         u32 e_cpos, num_clusters;
4051
4052         mlog(0, "Add new xattr bucket starting form %llu\n",
4053              (unsigned long long)header_bh->b_blocknr);
4054
4055         /*
4056          * Add refrence for header_bh here because it may be
4057          * changed in ocfs2_add_new_xattr_cluster and we need
4058          * to free it in the end.
4059          */
4060         get_bh(header_bh);
4061
4062         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4063                                   &num_clusters, el);
4064         if (ret) {
4065                 mlog_errno(ret);
4066                 goto out;
4067         }
4068
4069         ret = ocfs2_read_block(inode, p_blkno, &first_bh);
4070         if (ret) {
4071                 mlog_errno(ret);
4072                 goto out;
4073         }
4074
4075         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4076         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4077
4078         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
4079                 ret = ocfs2_add_new_xattr_cluster(inode,
4080                                                   xb_bh,
4081                                                   &first_bh,
4082                                                   &header_bh,
4083                                                   &num_clusters,
4084                                                   e_cpos,
4085                                                   p_blkno,
4086                                                   &extend,
4087                                                   ctxt);
4088                 if (ret) {
4089                         mlog_errno(ret);
4090                         goto out;
4091                 }
4092         }
4093
4094         if (extend)
4095                 ret = ocfs2_extend_xattr_bucket(inode,
4096                                                 ctxt->handle,
4097                                                 first_bh,
4098                                                 header_bh,
4099                                                 num_clusters);
4100         if (ret)
4101                 mlog_errno(ret);
4102 out:
4103         brelse(first_bh);
4104         brelse(header_bh);
4105         return ret;
4106 }
4107
4108 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4109                                         struct ocfs2_xattr_bucket *bucket,
4110                                         int offs)
4111 {
4112         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4113
4114         offs = offs % inode->i_sb->s_blocksize;
4115         return bucket_block(bucket, block_off) + offs;
4116 }
4117
4118 /*
4119  * Handle the normal xattr set, including replace, delete and new.
4120  *
4121  * Note: "local" indicates the real data's locality. So we can't
4122  * just its bucket locality by its length.
4123  */
4124 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4125                                          struct ocfs2_xattr_info *xi,
4126                                          struct ocfs2_xattr_search *xs,
4127                                          u32 name_hash,
4128                                          int local)
4129 {
4130         struct ocfs2_xattr_entry *last, *xe;
4131         int name_len = strlen(xi->name);
4132         struct ocfs2_xattr_header *xh = xs->header;
4133         u16 count = le16_to_cpu(xh->xh_count), start;
4134         size_t blocksize = inode->i_sb->s_blocksize;
4135         char *val;
4136         size_t offs, size, new_size;
4137
4138         last = &xh->xh_entries[count];
4139         if (!xs->not_found) {
4140                 xe = xs->here;
4141                 offs = le16_to_cpu(xe->xe_name_offset);
4142                 if (ocfs2_xattr_is_local(xe))
4143                         size = OCFS2_XATTR_SIZE(name_len) +
4144                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4145                 else
4146                         size = OCFS2_XATTR_SIZE(name_len) +
4147                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4148
4149                 /*
4150                  * If the new value will be stored outside, xi->value has been
4151                  * initalized as an empty ocfs2_xattr_value_root, and the same
4152                  * goes with xi->value_len, so we can set new_size safely here.
4153                  * See ocfs2_xattr_set_in_bucket.
4154                  */
4155                 new_size = OCFS2_XATTR_SIZE(name_len) +
4156                            OCFS2_XATTR_SIZE(xi->value_len);
4157
4158                 le16_add_cpu(&xh->xh_name_value_len, -size);
4159                 if (xi->value) {
4160                         if (new_size > size)
4161                                 goto set_new_name_value;
4162
4163                         /* Now replace the old value with new one. */
4164                         if (local)
4165                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4166                         else
4167                                 xe->xe_value_size = 0;
4168
4169                         val = ocfs2_xattr_bucket_get_val(inode,
4170                                                          xs->bucket, offs);
4171                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4172                                size - OCFS2_XATTR_SIZE(name_len));
4173                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4174                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4175                                        xi->value, xi->value_len);
4176
4177                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4178                         ocfs2_xattr_set_local(xe, local);
4179                         return;
4180                 } else {
4181                         /*
4182                          * Remove the old entry if there is more than one.
4183                          * We don't remove the last entry so that we can
4184                          * use it to indicate the hash value of the empty
4185                          * bucket.
4186                          */
4187                         last -= 1;
4188                         le16_add_cpu(&xh->xh_count, -1);
4189                         if (xh->xh_count) {
4190                                 memmove(xe, xe + 1,
4191                                         (void *)last - (void *)xe);
4192                                 memset(last, 0,
4193                                        sizeof(struct ocfs2_xattr_entry));
4194                         } else
4195                                 xh->xh_free_start =
4196                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4197
4198                         return;
4199                 }
4200         } else {
4201                 /* find a new entry for insert. */
4202                 int low = 0, high = count - 1, tmp;
4203                 struct ocfs2_xattr_entry *tmp_xe;
4204
4205                 while (low <= high && count) {
4206                         tmp = (low + high) / 2;
4207                         tmp_xe = &xh->xh_entries[tmp];
4208
4209                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4210                                 low = tmp + 1;
4211                         else if (name_hash <
4212                                  le32_to_cpu(tmp_xe->xe_name_hash))
4213                                 high = tmp - 1;
4214                         else {
4215                                 low = tmp;
4216                                 break;
4217                         }
4218                 }
4219
4220                 xe = &xh->xh_entries[low];
4221                 if (low != count)
4222                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4223
4224                 le16_add_cpu(&xh->xh_count, 1);
4225                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4226                 xe->xe_name_hash = cpu_to_le32(name_hash);
4227                 xe->xe_name_len = name_len;
4228                 ocfs2_xattr_set_type(xe, xi->name_index);
4229         }
4230
4231 set_new_name_value:
4232         /* Insert the new name+value. */
4233         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4234
4235         /*
4236          * We must make sure that the name/value pair
4237          * exists in the same block.
4238          */
4239         offs = le16_to_cpu(xh->xh_free_start);
4240         start = offs - size;
4241
4242         if (start >> inode->i_sb->s_blocksize_bits !=
4243             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4244                 offs = offs - offs % blocksize;
4245                 xh->xh_free_start = cpu_to_le16(offs);
4246         }
4247
4248         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4249         xe->xe_name_offset = cpu_to_le16(offs - size);
4250
4251         memset(val, 0, size);
4252         memcpy(val, xi->name, name_len);
4253         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4254
4255         xe->xe_value_size = cpu_to_le64(xi->value_len);
4256         ocfs2_xattr_set_local(xe, local);
4257         xs->here = xe;
4258         le16_add_cpu(&xh->xh_free_start, -size);
4259         le16_add_cpu(&xh->xh_name_value_len, size);
4260
4261         return;
4262 }
4263
4264 /*
4265  * Set the xattr entry in the specified bucket.
4266  * The bucket is indicated by xs->bucket and it should have the enough
4267  * space for the xattr insertion.
4268  */
4269 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4270                                            handle_t *handle,
4271                                            struct ocfs2_xattr_info *xi,
4272                                            struct ocfs2_xattr_search *xs,
4273                                            u32 name_hash,
4274                                            int local)
4275 {
4276         int ret;
4277         u64 blkno;
4278
4279         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4280              (unsigned long)xi->value_len, xi->name_index,
4281              (unsigned long long)bucket_blkno(xs->bucket));
4282
4283         if (!xs->bucket->bu_bhs[1]) {
4284                 blkno = bucket_blkno(xs->bucket);
4285                 ocfs2_xattr_bucket_relse(xs->bucket);
4286                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4287                 if (ret) {
4288                         mlog_errno(ret);
4289                         goto out;
4290                 }
4291         }
4292
4293         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4294                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4295         if (ret < 0) {
4296                 mlog_errno(ret);
4297                 goto out;
4298         }
4299
4300         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4301         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4302
4303 out:
4304         return ret;
4305 }
4306
4307 static int ocfs2_xattr_value_update_size(struct inode *inode,
4308                                          handle_t *handle,
4309                                          struct buffer_head *xe_bh,
4310                                          struct ocfs2_xattr_entry *xe,
4311                                          u64 new_size)
4312 {
4313         int ret;
4314
4315         ret = ocfs2_journal_access(handle, inode, xe_bh,
4316                                    OCFS2_JOURNAL_ACCESS_WRITE);
4317         if (ret < 0) {
4318                 mlog_errno(ret);
4319                 goto out;
4320         }
4321
4322         xe->xe_value_size = cpu_to_le64(new_size);
4323
4324         ret = ocfs2_journal_dirty(handle, xe_bh);
4325         if (ret < 0)
4326                 mlog_errno(ret);
4327
4328 out:
4329         return ret;
4330 }
4331
4332 /*
4333  * Truncate the specified xe_off entry in xattr bucket.
4334  * bucket is indicated by header_bh and len is the new length.
4335  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4336  *
4337  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4338  */
4339 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4340                                              struct buffer_head *header_bh,
4341                                              int xe_off,
4342                                              int len,
4343                                              struct ocfs2_xattr_set_ctxt *ctxt)
4344 {
4345         int ret, offset;
4346         u64 value_blk;
4347         struct buffer_head *value_bh = NULL;
4348         struct ocfs2_xattr_value_root *xv;
4349         struct ocfs2_xattr_entry *xe;
4350         struct ocfs2_xattr_header *xh =
4351                         (struct ocfs2_xattr_header *)header_bh->b_data;
4352         size_t blocksize = inode->i_sb->s_blocksize;
4353
4354         xe = &xh->xh_entries[xe_off];
4355
4356         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4357
4358         offset = le16_to_cpu(xe->xe_name_offset) +
4359                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4360
4361         value_blk = offset / blocksize;
4362
4363         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4364         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4365         value_blk += header_bh->b_blocknr;
4366
4367         ret = ocfs2_read_block(inode, value_blk, &value_bh);
4368         if (ret) {
4369                 mlog_errno(ret);
4370                 goto out;
4371         }
4372
4373         xv = (struct ocfs2_xattr_value_root *)
4374                 (value_bh->b_data + offset % blocksize);
4375
4376         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4377              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4378         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len, ctxt);
4379         if (ret) {
4380                 mlog_errno(ret);
4381                 goto out;
4382         }
4383
4384         ret = ocfs2_xattr_value_update_size(inode, ctxt->handle,
4385                                             header_bh, xe, len);
4386         if (ret) {
4387                 mlog_errno(ret);
4388                 goto out;
4389         }
4390
4391 out:
4392         brelse(value_bh);
4393         return ret;
4394 }
4395
4396 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4397                                         struct ocfs2_xattr_search *xs,
4398                                         int len,
4399                                         struct ocfs2_xattr_set_ctxt *ctxt)
4400 {
4401         int ret, offset;
4402         struct ocfs2_xattr_entry *xe = xs->here;
4403         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4404
4405         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4406
4407         offset = xe - xh->xh_entries;
4408         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket->bu_bhs[0],
4409                                                 offset, len, ctxt);
4410         if (ret)
4411                 mlog_errno(ret);
4412
4413         return ret;
4414 }
4415
4416 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4417                                                 handle_t *handle,
4418                                                 struct ocfs2_xattr_search *xs,
4419                                                 char *val,
4420                                                 int value_len)
4421 {
4422         int offset;
4423         struct ocfs2_xattr_value_root *xv;
4424         struct ocfs2_xattr_entry *xe = xs->here;
4425
4426         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4427
4428         offset = le16_to_cpu(xe->xe_name_offset) +
4429                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4430
4431         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4432
4433         return __ocfs2_xattr_set_value_outside(inode, handle,
4434                                                xv, val, value_len);
4435 }
4436
4437 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4438                                   struct buffer_head *root_bh,
4439                                   u64 blkno,
4440                                   u32 cpos,
4441                                   u32 len)
4442 {
4443         int ret;
4444         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4445         struct inode *tl_inode = osb->osb_tl_inode;
4446         handle_t *handle;
4447         struct ocfs2_xattr_block *xb =
4448                         (struct ocfs2_xattr_block *)root_bh->b_data;
4449         struct ocfs2_alloc_context *meta_ac = NULL;
4450         struct ocfs2_cached_dealloc_ctxt dealloc;
4451         struct ocfs2_extent_tree et;
4452
4453         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4454
4455         ocfs2_init_dealloc_ctxt(&dealloc);
4456
4457         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4458              cpos, len, (unsigned long long)blkno);
4459
4460         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4461
4462         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4463         if (ret) {
4464                 mlog_errno(ret);
4465                 return ret;
4466         }
4467
4468         mutex_lock(&tl_inode->i_mutex);
4469
4470         if (ocfs2_truncate_log_needs_flush(osb)) {
4471                 ret = __ocfs2_flush_truncate_log(osb);
4472                 if (ret < 0) {
4473                         mlog_errno(ret);
4474                         goto out;
4475                 }
4476         }
4477
4478         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4479         if (IS_ERR(handle)) {
4480                 ret = -ENOMEM;
4481                 mlog_errno(ret);
4482                 goto out;
4483         }
4484
4485         ret = ocfs2_journal_access(handle, inode, root_bh,
4486                                    OCFS2_JOURNAL_ACCESS_WRITE);
4487         if (ret) {
4488                 mlog_errno(ret);
4489                 goto out_commit;
4490         }
4491
4492         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4493                                   &dealloc);
4494         if (ret) {
4495                 mlog_errno(ret);
4496                 goto out_commit;
4497         }
4498
4499         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4500
4501         ret = ocfs2_journal_dirty(handle, root_bh);
4502         if (ret) {
4503                 mlog_errno(ret);
4504                 goto out_commit;
4505         }
4506
4507         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4508         if (ret)
4509                 mlog_errno(ret);
4510
4511 out_commit:
4512         ocfs2_commit_trans(osb, handle);
4513 out:
4514         ocfs2_schedule_truncate_log_flush(osb, 1);
4515
4516         mutex_unlock(&tl_inode->i_mutex);
4517
4518         if (meta_ac)
4519                 ocfs2_free_alloc_context(meta_ac);
4520
4521         ocfs2_run_deallocs(osb, &dealloc);
4522
4523         return ret;
4524 }
4525
4526 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4527                                          handle_t *handle,
4528                                          struct ocfs2_xattr_search *xs)
4529 {
4530         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4531         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4532                                                 le16_to_cpu(xh->xh_count) - 1];
4533         int ret = 0;
4534
4535         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4536                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4537         if (ret) {
4538                 mlog_errno(ret);
4539                 return;
4540         }
4541
4542         /* Remove the old entry. */
4543         memmove(xs->here, xs->here + 1,
4544                 (void *)last - (void *)xs->here);
4545         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4546         le16_add_cpu(&xh->xh_count, -1);
4547
4548         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4549 }
4550
4551 /*
4552  * Set the xattr name/value in the bucket specified in xs.
4553  *
4554  * As the new value in xi may be stored in the bucket or in an outside cluster,
4555  * we divide the whole process into 3 steps:
4556  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4557  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4558  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4559  * 4. If the clusters for the new outside value can't be allocated, we need
4560  *    to free the xattr we allocated in set.
4561  */
4562 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4563                                      struct ocfs2_xattr_info *xi,
4564                                      struct ocfs2_xattr_search *xs,
4565                                      struct ocfs2_xattr_set_ctxt *ctxt)
4566 {
4567         int ret, local = 1;
4568         size_t value_len;
4569         char *val = (char *)xi->value;
4570         struct ocfs2_xattr_entry *xe = xs->here;
4571         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4572                                               strlen(xi->name));
4573
4574         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4575                 /*
4576                  * We need to truncate the xattr storage first.
4577                  *
4578                  * If both the old and new value are stored to
4579                  * outside block, we only need to truncate
4580                  * the storage and then set the value outside.
4581                  *
4582                  * If the new value should be stored within block,
4583                  * we should free all the outside block first and
4584                  * the modification to the xattr block will be done
4585                  * by following steps.
4586                  */
4587                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4588                         value_len = xi->value_len;
4589                 else
4590                         value_len = 0;
4591
4592                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4593                                                            value_len,
4594                                                            ctxt);
4595                 if (ret)
4596                         goto out;
4597
4598                 if (value_len)
4599                         goto set_value_outside;
4600         }
4601
4602         value_len = xi->value_len;
4603         /* So we have to handle the inside block change now. */
4604         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4605                 /*
4606                  * If the new value will be stored outside of block,
4607                  * initalize a new empty value root and insert it first.
4608                  */
4609                 local = 0;
4610                 xi->value = &def_xv;
4611                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4612         }
4613
4614         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4615                                               name_hash, local);
4616         if (ret) {
4617                 mlog_errno(ret);
4618                 goto out;
4619         }
4620
4621         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4622                 goto out;
4623
4624         /* allocate the space now for the outside block storage. */
4625         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4626                                                    value_len, ctxt);
4627         if (ret) {
4628                 mlog_errno(ret);
4629
4630                 if (xs->not_found) {
4631                         /*
4632                          * We can't allocate enough clusters for outside
4633                          * storage and we have allocated xattr already,
4634                          * so need to remove it.
4635                          */
4636                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4637                 }
4638                 goto out;
4639         }
4640
4641 set_value_outside:
4642         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4643                                                    xs, val, value_len);
4644 out:
4645         return ret;
4646 }
4647
4648 /*
4649  * check whether the xattr bucket is filled up with the same hash value.
4650  * If we want to insert the xattr with the same hash, return -ENOSPC.
4651  * If we want to insert a xattr with different hash value, go ahead
4652  * and ocfs2_divide_xattr_bucket will handle this.
4653  */
4654 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4655                                               struct ocfs2_xattr_bucket *bucket,
4656                                               const char *name)
4657 {
4658         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4659         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4660
4661         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4662                 return 0;
4663
4664         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4665             xh->xh_entries[0].xe_name_hash) {
4666                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4667                      "hash = %u\n",
4668                      (unsigned long long)bucket_blkno(bucket),
4669                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4670                 return -ENOSPC;
4671         }
4672
4673         return 0;
4674 }
4675
4676 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4677                                              struct ocfs2_xattr_info *xi,
4678                                              struct ocfs2_xattr_search *xs,
4679                                              struct ocfs2_xattr_set_ctxt *ctxt)
4680 {
4681         struct ocfs2_xattr_header *xh;
4682         struct ocfs2_xattr_entry *xe;
4683         u16 count, header_size, xh_free_start;
4684         int free, max_free, need, old;
4685         size_t value_size = 0, name_len = strlen(xi->name);
4686         size_t blocksize = inode->i_sb->s_blocksize;
4687         int ret, allocation = 0;
4688
4689         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4690
4691 try_again:
4692         xh = xs->header;
4693         count = le16_to_cpu(xh->xh_count);
4694         xh_free_start = le16_to_cpu(xh->xh_free_start);
4695         header_size = sizeof(struct ocfs2_xattr_header) +
4696                         count * sizeof(struct ocfs2_xattr_entry);
4697         max_free = OCFS2_XATTR_BUCKET_SIZE -
4698                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4699
4700         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4701                         "of %u which exceed block size\n",
4702                         (unsigned long long)bucket_blkno(xs->bucket),
4703                         header_size);
4704
4705         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4706                 value_size = OCFS2_XATTR_ROOT_SIZE;
4707         else if (xi->value)
4708                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4709
4710         if (xs->not_found)
4711                 need = sizeof(struct ocfs2_xattr_entry) +
4712                         OCFS2_XATTR_SIZE(name_len) + value_size;
4713         else {
4714                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4715
4716                 /*
4717                  * We only replace the old value if the new length is smaller
4718                  * than the old one. Otherwise we will allocate new space in the
4719                  * bucket to store it.
4720                  */
4721                 xe = xs->here;
4722                 if (ocfs2_xattr_is_local(xe))
4723                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4724                 else
4725                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4726
4727                 if (old >= value_size)
4728                         need = 0;
4729         }
4730
4731         free = xh_free_start - header_size;
4732         /*
4733          * We need to make sure the new name/value pair
4734          * can exist in the same block.
4735          */
4736         if (xh_free_start % blocksize < need)
4737                 free -= xh_free_start % blocksize;
4738
4739         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4740              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4741              " %u\n", xs->not_found,
4742              (unsigned long long)bucket_blkno(xs->bucket),
4743              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4744              le16_to_cpu(xh->xh_name_value_len));
4745
4746         if (free < need ||
4747             (xs->not_found &&
4748              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
4749                 if (need <= max_free &&
4750                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4751                         /*
4752                          * We can create the space by defragment. Since only the
4753                          * name/value will be moved, the xe shouldn't be changed
4754                          * in xs.
4755                          */
4756                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
4757                                                         xs->bucket);
4758                         if (ret) {
4759                                 mlog_errno(ret);
4760                                 goto out;
4761                         }
4762
4763                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4764                         free = xh_free_start - header_size;
4765                         if (xh_free_start % blocksize < need)
4766                                 free -= xh_free_start % blocksize;
4767
4768                         if (free >= need)
4769                                 goto xattr_set;
4770
4771                         mlog(0, "Can't get enough space for xattr insert by "
4772                              "defragment. Need %u bytes, but we have %d, so "
4773                              "allocate new bucket for it.\n", need, free);
4774                 }
4775
4776                 /*
4777                  * We have to add new buckets or clusters and one
4778                  * allocation should leave us enough space for insert.
4779                  */
4780                 BUG_ON(allocation);
4781
4782                 /*
4783                  * We do not allow for overlapping ranges between buckets. And
4784                  * the maximum number of collisions we will allow for then is
4785                  * one bucket's worth, so check it here whether we need to
4786                  * add a new bucket for the insert.
4787                  */
4788                 ret = ocfs2_check_xattr_bucket_collision(inode,
4789                                                          xs->bucket,
4790                                                          xi->name);
4791                 if (ret) {
4792                         mlog_errno(ret);
4793                         goto out;
4794                 }
4795
4796                 ret = ocfs2_add_new_xattr_bucket(inode,
4797                                                  xs->xattr_bh,
4798                                                  xs->bucket->bu_bhs[0],
4799                                                  ctxt);
4800                 if (ret) {
4801                         mlog_errno(ret);
4802                         goto out;
4803                 }
4804
4805                 ocfs2_xattr_bucket_relse(xs->bucket);
4806
4807                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4808                                                    xi->name_index,
4809                                                    xi->name, xs);
4810                 if (ret && ret != -ENODATA)
4811                         goto out;
4812                 xs->not_found = ret;
4813                 allocation = 1;
4814                 goto try_again;
4815         }
4816
4817 xattr_set:
4818         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
4819 out:
4820         mlog_exit(ret);
4821         return ret;
4822 }
4823
4824 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4825                                         struct ocfs2_xattr_bucket *bucket,
4826                                         void *para)
4827 {
4828         int ret = 0;
4829         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4830         u16 i;
4831         struct ocfs2_xattr_entry *xe;
4832         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4833         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
4834
4835         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
4836
4837         ctxt.handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4838         if (IS_ERR(ctxt.handle)) {
4839                 ret = PTR_ERR(ctxt.handle);
4840                 mlog_errno(ret);
4841                 goto out;
4842         }
4843
4844         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4845                 xe = &xh->xh_entries[i];
4846                 if (ocfs2_xattr_is_local(xe))
4847                         continue;
4848
4849                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4850                                                         bucket->bu_bhs[0],
4851                                                         i, 0, &ctxt);
4852                 if (ret) {
4853                         mlog_errno(ret);
4854                         break;
4855                 }
4856         }
4857
4858         ret = ocfs2_commit_trans(osb, ctxt.handle);
4859         ocfs2_schedule_truncate_log_flush(osb, 1);
4860         ocfs2_run_deallocs(osb, &ctxt.dealloc);
4861 out:
4862         return ret;
4863 }
4864
4865 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4866                                           struct buffer_head *xb_bh)
4867 {
4868         struct ocfs2_xattr_block *xb =
4869                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4870         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4871         int ret = 0;
4872         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4873         u64 p_blkno;
4874
4875         if (le16_to_cpu(el->l_next_free_rec) == 0)
4876                 return 0;
4877
4878         while (name_hash > 0) {
4879                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4880                                           &e_cpos, &num_clusters, el);
4881                 if (ret) {
4882                         mlog_errno(ret);
4883                         goto out;
4884                 }
4885
4886                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4887                                                   ocfs2_delete_xattr_in_bucket,
4888                                                   NULL);
4889                 if (ret) {
4890                         mlog_errno(ret);
4891                         goto out;
4892                 }
4893
4894                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4895                                              p_blkno, e_cpos, num_clusters);
4896                 if (ret) {
4897                         mlog_errno(ret);
4898                         break;
4899                 }
4900
4901                 if (e_cpos == 0)
4902                         break;
4903
4904                 name_hash = e_cpos - 1;
4905         }
4906
4907 out:
4908         return ret;
4909 }
4910
4911 /*
4912  * 'trusted' attributes support
4913  */
4914 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4915                                        size_t list_size, const char *name,
4916                                        size_t name_len)
4917 {
4918         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
4919         const size_t total_len = prefix_len + name_len + 1;
4920
4921         if (list && total_len <= list_size) {
4922                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4923                 memcpy(list + prefix_len, name, name_len);
4924                 list[prefix_len + name_len] = '\0';
4925         }
4926         return total_len;
4927 }
4928
4929 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4930                                    void *buffer, size_t size)
4931 {
4932         if (strcmp(name, "") == 0)
4933                 return -EINVAL;
4934         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4935                                buffer, size);
4936 }
4937
4938 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4939                                    const void *value, size_t size, int flags)
4940 {
4941         if (strcmp(name, "") == 0)
4942                 return -EINVAL;
4943
4944         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4945                                size, flags);
4946 }
4947
4948 struct xattr_handler ocfs2_xattr_trusted_handler = {
4949         .prefix = XATTR_TRUSTED_PREFIX,
4950         .list   = ocfs2_xattr_trusted_list,
4951         .get    = ocfs2_xattr_trusted_get,
4952         .set    = ocfs2_xattr_trusted_set,
4953 };
4954
4955 /*
4956  * 'user' attributes support
4957  */
4958 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4959                                     size_t list_size, const char *name,
4960                                     size_t name_len)
4961 {
4962         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
4963         const size_t total_len = prefix_len + name_len + 1;
4964         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4965
4966         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4967                 return 0;
4968
4969         if (list && total_len <= list_size) {
4970                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4971                 memcpy(list + prefix_len, name, name_len);
4972                 list[prefix_len + name_len] = '\0';
4973         }
4974         return total_len;
4975 }
4976
4977 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4978                                 void *buffer, size_t size)
4979 {
4980         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4981
4982         if (strcmp(name, "") == 0)
4983                 return -EINVAL;
4984         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4985                 return -EOPNOTSUPP;
4986         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4987                                buffer, size);
4988 }
4989
4990 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4991                                 const void *value, size_t size, int flags)
4992 {
4993         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4994
4995         if (strcmp(name, "") == 0)
4996                 return -EINVAL;
4997         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4998                 return -EOPNOTSUPP;
4999
5000         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5001                                size, flags);
5002 }
5003
5004 struct xattr_handler ocfs2_xattr_user_handler = {
5005         .prefix = XATTR_USER_PREFIX,
5006         .list   = ocfs2_xattr_user_list,
5007         .get    = ocfs2_xattr_user_get,
5008         .set    = ocfs2_xattr_user_set,
5009 };