1b77302b54ffcee58de9081c4b06d68dfe248ccf
[linux-2.6.git] / fs / ocfs2 / xattr.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * xattr.c
5  *
6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
7  *
8  * CREDITS:
9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public
14  * License version 2 as published by the Free Software Foundation.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  */
21
22 #include <linux/capability.h>
23 #include <linux/fs.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38
39 #define MLOG_MASK_PREFIX ML_XATTR
40 #include <cluster/masklog.h>
41
42 #include "ocfs2.h"
43 #include "alloc.h"
44 #include "dlmglue.h"
45 #include "file.h"
46 #include "symlink.h"
47 #include "sysfile.h"
48 #include "inode.h"
49 #include "journal.h"
50 #include "ocfs2_fs.h"
51 #include "suballoc.h"
52 #include "uptodate.h"
53 #include "buffer_head_io.h"
54 #include "super.h"
55 #include "xattr.h"
56
57
58 struct ocfs2_xattr_def_value_root {
59         struct ocfs2_xattr_value_root   xv;
60         struct ocfs2_extent_rec         er;
61 };
62
63 struct ocfs2_xattr_bucket {
64         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
65 };
66
67 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
68 #define OCFS2_XATTR_INLINE_SIZE 80
69
70 static struct ocfs2_xattr_def_value_root def_xv = {
71         .xv.xr_list.l_count = cpu_to_le16(1),
72 };
73
74 struct xattr_handler *ocfs2_xattr_handlers[] = {
75         &ocfs2_xattr_user_handler,
76         &ocfs2_xattr_trusted_handler,
77         NULL
78 };
79
80 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
81         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
82         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
83 };
84
85 struct ocfs2_xattr_info {
86         int name_index;
87         const char *name;
88         const void *value;
89         size_t value_len;
90 };
91
92 struct ocfs2_xattr_search {
93         struct buffer_head *inode_bh;
94         /*
95          * xattr_bh point to the block buffer head which has extended attribute
96          * when extended attribute in inode, xattr_bh is equal to inode_bh.
97          */
98         struct buffer_head *xattr_bh;
99         struct ocfs2_xattr_header *header;
100         struct ocfs2_xattr_bucket bucket;
101         void *base;
102         void *end;
103         struct ocfs2_xattr_entry *here;
104         int not_found;
105 };
106
107 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
108                                              struct ocfs2_xattr_header *xh,
109                                              int index,
110                                              int *block_off,
111                                              int *new_offset);
112
113 static int ocfs2_xattr_block_find(struct inode *inode,
114                                   int name_index,
115                                   const char *name,
116                                   struct ocfs2_xattr_search *xs);
117 static int ocfs2_xattr_index_block_find(struct inode *inode,
118                                         struct buffer_head *root_bh,
119                                         int name_index,
120                                         const char *name,
121                                         struct ocfs2_xattr_search *xs);
122
123 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
124                                         struct ocfs2_xattr_tree_root *xt,
125                                         char *buffer,
126                                         size_t buffer_size);
127
128 static int ocfs2_xattr_create_index_block(struct inode *inode,
129                                           struct ocfs2_xattr_search *xs);
130
131 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
132                                              struct ocfs2_xattr_info *xi,
133                                              struct ocfs2_xattr_search *xs);
134
135 static int ocfs2_delete_xattr_index_block(struct inode *inode,
136                                           struct buffer_head *xb_bh);
137
138 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
139 {
140         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
141 }
142
143 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
144 {
145         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
146 }
147
148 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
149 {
150         u16 len = sb->s_blocksize -
151                  offsetof(struct ocfs2_xattr_header, xh_entries);
152
153         return len / sizeof(struct ocfs2_xattr_entry);
154 }
155
156 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
157 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
158 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
159
160 static inline const char *ocfs2_xattr_prefix(int name_index)
161 {
162         struct xattr_handler *handler = NULL;
163
164         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
165                 handler = ocfs2_xattr_handler_map[name_index];
166
167         return handler ? handler->prefix : NULL;
168 }
169
170 static u32 ocfs2_xattr_name_hash(struct inode *inode,
171                                  const char *name,
172                                  int name_len)
173 {
174         /* Get hash value of uuid from super block */
175         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
176         int i;
177
178         /* hash extended attribute name */
179         for (i = 0; i < name_len; i++) {
180                 hash = (hash << OCFS2_HASH_SHIFT) ^
181                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
182                        *name++;
183         }
184
185         return hash;
186 }
187
188 /*
189  * ocfs2_xattr_hash_entry()
190  *
191  * Compute the hash of an extended attribute.
192  */
193 static void ocfs2_xattr_hash_entry(struct inode *inode,
194                                    struct ocfs2_xattr_header *header,
195                                    struct ocfs2_xattr_entry *entry)
196 {
197         u32 hash = 0;
198         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
199
200         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
201         entry->xe_name_hash = cpu_to_le32(hash);
202
203         return;
204 }
205
206 static int ocfs2_xattr_extend_allocation(struct inode *inode,
207                                          u32 clusters_to_add,
208                                          struct buffer_head *xattr_bh,
209                                          struct ocfs2_xattr_value_root *xv)
210 {
211         int status = 0;
212         int restart_func = 0;
213         int credits = 0;
214         handle_t *handle = NULL;
215         struct ocfs2_alloc_context *data_ac = NULL;
216         struct ocfs2_alloc_context *meta_ac = NULL;
217         enum ocfs2_alloc_restarted why;
218         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
219         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
220         struct ocfs2_extent_tree et;
221
222         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
223
224         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
225
226 restart_all:
227
228         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
229                                        &data_ac, &meta_ac);
230         if (status) {
231                 mlog_errno(status);
232                 goto leave;
233         }
234
235         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
236                                             clusters_to_add);
237         handle = ocfs2_start_trans(osb, credits);
238         if (IS_ERR(handle)) {
239                 status = PTR_ERR(handle);
240                 handle = NULL;
241                 mlog_errno(status);
242                 goto leave;
243         }
244
245 restarted_transaction:
246         status = ocfs2_journal_access(handle, inode, xattr_bh,
247                                       OCFS2_JOURNAL_ACCESS_WRITE);
248         if (status < 0) {
249                 mlog_errno(status);
250                 goto leave;
251         }
252
253         prev_clusters = le32_to_cpu(xv->xr_clusters);
254         status = ocfs2_add_clusters_in_btree(osb,
255                                              inode,
256                                              &logical_start,
257                                              clusters_to_add,
258                                              0,
259                                              &et,
260                                              handle,
261                                              data_ac,
262                                              meta_ac,
263                                              &why);
264         if ((status < 0) && (status != -EAGAIN)) {
265                 if (status != -ENOSPC)
266                         mlog_errno(status);
267                 goto leave;
268         }
269
270         status = ocfs2_journal_dirty(handle, xattr_bh);
271         if (status < 0) {
272                 mlog_errno(status);
273                 goto leave;
274         }
275
276         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
277
278         if (why != RESTART_NONE && clusters_to_add) {
279                 if (why == RESTART_META) {
280                         mlog(0, "restarting function.\n");
281                         restart_func = 1;
282                 } else {
283                         BUG_ON(why != RESTART_TRANS);
284
285                         mlog(0, "restarting transaction.\n");
286                         /* TODO: This can be more intelligent. */
287                         credits = ocfs2_calc_extend_credits(osb->sb,
288                                                             et.et_root_el,
289                                                             clusters_to_add);
290                         status = ocfs2_extend_trans(handle, credits);
291                         if (status < 0) {
292                                 /* handle still has to be committed at
293                                  * this point. */
294                                 status = -ENOMEM;
295                                 mlog_errno(status);
296                                 goto leave;
297                         }
298                         goto restarted_transaction;
299                 }
300         }
301
302 leave:
303         if (handle) {
304                 ocfs2_commit_trans(osb, handle);
305                 handle = NULL;
306         }
307         if (data_ac) {
308                 ocfs2_free_alloc_context(data_ac);
309                 data_ac = NULL;
310         }
311         if (meta_ac) {
312                 ocfs2_free_alloc_context(meta_ac);
313                 meta_ac = NULL;
314         }
315         if ((!status) && restart_func) {
316                 restart_func = 0;
317                 goto restart_all;
318         }
319
320         return status;
321 }
322
323 static int __ocfs2_remove_xattr_range(struct inode *inode,
324                                       struct buffer_head *root_bh,
325                                       struct ocfs2_xattr_value_root *xv,
326                                       u32 cpos, u32 phys_cpos, u32 len,
327                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
328 {
329         int ret;
330         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
331         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
332         struct inode *tl_inode = osb->osb_tl_inode;
333         handle_t *handle;
334         struct ocfs2_alloc_context *meta_ac = NULL;
335         struct ocfs2_extent_tree et;
336
337         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
338
339         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
340         if (ret) {
341                 mlog_errno(ret);
342                 return ret;
343         }
344
345         mutex_lock(&tl_inode->i_mutex);
346
347         if (ocfs2_truncate_log_needs_flush(osb)) {
348                 ret = __ocfs2_flush_truncate_log(osb);
349                 if (ret < 0) {
350                         mlog_errno(ret);
351                         goto out;
352                 }
353         }
354
355         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
356         if (IS_ERR(handle)) {
357                 ret = PTR_ERR(handle);
358                 mlog_errno(ret);
359                 goto out;
360         }
361
362         ret = ocfs2_journal_access(handle, inode, root_bh,
363                                    OCFS2_JOURNAL_ACCESS_WRITE);
364         if (ret) {
365                 mlog_errno(ret);
366                 goto out_commit;
367         }
368
369         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
370                                   dealloc);
371         if (ret) {
372                 mlog_errno(ret);
373                 goto out_commit;
374         }
375
376         le32_add_cpu(&xv->xr_clusters, -len);
377
378         ret = ocfs2_journal_dirty(handle, root_bh);
379         if (ret) {
380                 mlog_errno(ret);
381                 goto out_commit;
382         }
383
384         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
385         if (ret)
386                 mlog_errno(ret);
387
388 out_commit:
389         ocfs2_commit_trans(osb, handle);
390 out:
391         mutex_unlock(&tl_inode->i_mutex);
392
393         if (meta_ac)
394                 ocfs2_free_alloc_context(meta_ac);
395
396         return ret;
397 }
398
399 static int ocfs2_xattr_shrink_size(struct inode *inode,
400                                    u32 old_clusters,
401                                    u32 new_clusters,
402                                    struct buffer_head *root_bh,
403                                    struct ocfs2_xattr_value_root *xv)
404 {
405         int ret = 0;
406         u32 trunc_len, cpos, phys_cpos, alloc_size;
407         u64 block;
408         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
409         struct ocfs2_cached_dealloc_ctxt dealloc;
410
411         ocfs2_init_dealloc_ctxt(&dealloc);
412
413         if (old_clusters <= new_clusters)
414                 return 0;
415
416         cpos = new_clusters;
417         trunc_len = old_clusters - new_clusters;
418         while (trunc_len) {
419                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
420                                                &alloc_size, &xv->xr_list);
421                 if (ret) {
422                         mlog_errno(ret);
423                         goto out;
424                 }
425
426                 if (alloc_size > trunc_len)
427                         alloc_size = trunc_len;
428
429                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
430                                                  phys_cpos, alloc_size,
431                                                  &dealloc);
432                 if (ret) {
433                         mlog_errno(ret);
434                         goto out;
435                 }
436
437                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
438                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
439                                                        alloc_size);
440                 cpos += alloc_size;
441                 trunc_len -= alloc_size;
442         }
443
444 out:
445         ocfs2_schedule_truncate_log_flush(osb, 1);
446         ocfs2_run_deallocs(osb, &dealloc);
447
448         return ret;
449 }
450
451 static int ocfs2_xattr_value_truncate(struct inode *inode,
452                                       struct buffer_head *root_bh,
453                                       struct ocfs2_xattr_value_root *xv,
454                                       int len)
455 {
456         int ret;
457         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
458         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
459
460         if (new_clusters == old_clusters)
461                 return 0;
462
463         if (new_clusters > old_clusters)
464                 ret = ocfs2_xattr_extend_allocation(inode,
465                                                     new_clusters - old_clusters,
466                                                     root_bh, xv);
467         else
468                 ret = ocfs2_xattr_shrink_size(inode,
469                                               old_clusters, new_clusters,
470                                               root_bh, xv);
471
472         return ret;
473 }
474
475 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
476                                   size_t *result, const char *prefix,
477                                   const char *name, int name_len)
478 {
479         char *p = buffer + *result;
480         int prefix_len = strlen(prefix);
481         int total_len = prefix_len + name_len + 1;
482
483         *result += total_len;
484
485         /* we are just looking for how big our buffer needs to be */
486         if (!size)
487                 return 0;
488
489         if (*result > size)
490                 return -ERANGE;
491
492         memcpy(p, prefix, prefix_len);
493         memcpy(p + prefix_len, name, name_len);
494         p[prefix_len + name_len] = '\0';
495
496         return 0;
497 }
498
499 static int ocfs2_xattr_list_entries(struct inode *inode,
500                                     struct ocfs2_xattr_header *header,
501                                     char *buffer, size_t buffer_size)
502 {
503         size_t result = 0;
504         int i, type, ret;
505         const char *prefix, *name;
506
507         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
508                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
509                 type = ocfs2_xattr_get_type(entry);
510                 prefix = ocfs2_xattr_prefix(type);
511
512                 if (prefix) {
513                         name = (const char *)header +
514                                 le16_to_cpu(entry->xe_name_offset);
515
516                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
517                                                      &result, prefix, name,
518                                                      entry->xe_name_len);
519                         if (ret)
520                                 return ret;
521                 }
522         }
523
524         return result;
525 }
526
527 static int ocfs2_xattr_ibody_list(struct inode *inode,
528                                   struct ocfs2_dinode *di,
529                                   char *buffer,
530                                   size_t buffer_size)
531 {
532         struct ocfs2_xattr_header *header = NULL;
533         struct ocfs2_inode_info *oi = OCFS2_I(inode);
534         int ret = 0;
535
536         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
537                 return ret;
538
539         header = (struct ocfs2_xattr_header *)
540                  ((void *)di + inode->i_sb->s_blocksize -
541                  le16_to_cpu(di->i_xattr_inline_size));
542
543         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
544
545         return ret;
546 }
547
548 static int ocfs2_xattr_block_list(struct inode *inode,
549                                   struct ocfs2_dinode *di,
550                                   char *buffer,
551                                   size_t buffer_size)
552 {
553         struct buffer_head *blk_bh = NULL;
554         struct ocfs2_xattr_block *xb;
555         int ret = 0;
556
557         if (!di->i_xattr_loc)
558                 return ret;
559
560         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
561         if (ret < 0) {
562                 mlog_errno(ret);
563                 return ret;
564         }
565
566         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
567         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
568                 ret = -EIO;
569                 goto cleanup;
570         }
571
572         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
573                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
574                 ret = ocfs2_xattr_list_entries(inode, header,
575                                                buffer, buffer_size);
576         } else {
577                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
578                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
579                                                    buffer, buffer_size);
580         }
581 cleanup:
582         brelse(blk_bh);
583
584         return ret;
585 }
586
587 ssize_t ocfs2_listxattr(struct dentry *dentry,
588                         char *buffer,
589                         size_t size)
590 {
591         int ret = 0, i_ret = 0, b_ret = 0;
592         struct buffer_head *di_bh = NULL;
593         struct ocfs2_dinode *di = NULL;
594         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
595
596         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
597                 return -EOPNOTSUPP;
598
599         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
600                 return ret;
601
602         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
603         if (ret < 0) {
604                 mlog_errno(ret);
605                 return ret;
606         }
607
608         di = (struct ocfs2_dinode *)di_bh->b_data;
609
610         down_read(&oi->ip_xattr_sem);
611         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
612         if (i_ret < 0)
613                 b_ret = 0;
614         else {
615                 if (buffer) {
616                         buffer += i_ret;
617                         size -= i_ret;
618                 }
619                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
620                                                buffer, size);
621                 if (b_ret < 0)
622                         i_ret = 0;
623         }
624         up_read(&oi->ip_xattr_sem);
625         ocfs2_inode_unlock(dentry->d_inode, 0);
626
627         brelse(di_bh);
628
629         return i_ret + b_ret;
630 }
631
632 static int ocfs2_xattr_find_entry(int name_index,
633                                   const char *name,
634                                   struct ocfs2_xattr_search *xs)
635 {
636         struct ocfs2_xattr_entry *entry;
637         size_t name_len;
638         int i, cmp = 1;
639
640         if (name == NULL)
641                 return -EINVAL;
642
643         name_len = strlen(name);
644         entry = xs->here;
645         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
646                 cmp = name_index - ocfs2_xattr_get_type(entry);
647                 if (!cmp)
648                         cmp = name_len - entry->xe_name_len;
649                 if (!cmp)
650                         cmp = memcmp(name, (xs->base +
651                                      le16_to_cpu(entry->xe_name_offset)),
652                                      name_len);
653                 if (cmp == 0)
654                         break;
655                 entry += 1;
656         }
657         xs->here = entry;
658
659         return cmp ? -ENODATA : 0;
660 }
661
662 static int ocfs2_xattr_get_value_outside(struct inode *inode,
663                                          struct ocfs2_xattr_value_root *xv,
664                                          void *buffer,
665                                          size_t len)
666 {
667         u32 cpos, p_cluster, num_clusters, bpc, clusters;
668         u64 blkno;
669         int i, ret = 0;
670         size_t cplen, blocksize;
671         struct buffer_head *bh = NULL;
672         struct ocfs2_extent_list *el;
673
674         el = &xv->xr_list;
675         clusters = le32_to_cpu(xv->xr_clusters);
676         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
677         blocksize = inode->i_sb->s_blocksize;
678
679         cpos = 0;
680         while (cpos < clusters) {
681                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
682                                                &num_clusters, el);
683                 if (ret) {
684                         mlog_errno(ret);
685                         goto out;
686                 }
687
688                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
689                 /* Copy ocfs2_xattr_value */
690                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
691                         ret = ocfs2_read_block(inode, blkno, &bh);
692                         if (ret) {
693                                 mlog_errno(ret);
694                                 goto out;
695                         }
696
697                         cplen = len >= blocksize ? blocksize : len;
698                         memcpy(buffer, bh->b_data, cplen);
699                         len -= cplen;
700                         buffer += cplen;
701
702                         brelse(bh);
703                         bh = NULL;
704                         if (len == 0)
705                                 break;
706                 }
707                 cpos += num_clusters;
708         }
709 out:
710         return ret;
711 }
712
713 static int ocfs2_xattr_ibody_get(struct inode *inode,
714                                  int name_index,
715                                  const char *name,
716                                  void *buffer,
717                                  size_t buffer_size,
718                                  struct ocfs2_xattr_search *xs)
719 {
720         struct ocfs2_inode_info *oi = OCFS2_I(inode);
721         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
722         struct ocfs2_xattr_value_root *xv;
723         size_t size;
724         int ret = 0;
725
726         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
727                 return -ENODATA;
728
729         xs->end = (void *)di + inode->i_sb->s_blocksize;
730         xs->header = (struct ocfs2_xattr_header *)
731                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
732         xs->base = (void *)xs->header;
733         xs->here = xs->header->xh_entries;
734
735         ret = ocfs2_xattr_find_entry(name_index, name, xs);
736         if (ret)
737                 return ret;
738         size = le64_to_cpu(xs->here->xe_value_size);
739         if (buffer) {
740                 if (size > buffer_size)
741                         return -ERANGE;
742                 if (ocfs2_xattr_is_local(xs->here)) {
743                         memcpy(buffer, (void *)xs->base +
744                                le16_to_cpu(xs->here->xe_name_offset) +
745                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
746                 } else {
747                         xv = (struct ocfs2_xattr_value_root *)
748                                 (xs->base + le16_to_cpu(
749                                  xs->here->xe_name_offset) +
750                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
751                         ret = ocfs2_xattr_get_value_outside(inode, xv,
752                                                             buffer, size);
753                         if (ret < 0) {
754                                 mlog_errno(ret);
755                                 return ret;
756                         }
757                 }
758         }
759
760         return size;
761 }
762
763 static int ocfs2_xattr_block_get(struct inode *inode,
764                                  int name_index,
765                                  const char *name,
766                                  void *buffer,
767                                  size_t buffer_size,
768                                  struct ocfs2_xattr_search *xs)
769 {
770         struct ocfs2_xattr_block *xb;
771         struct ocfs2_xattr_value_root *xv;
772         size_t size;
773         int ret = -ENODATA, name_offset, name_len, block_off, i;
774
775         memset(&xs->bucket, 0, sizeof(xs->bucket));
776
777         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
778         if (ret) {
779                 mlog_errno(ret);
780                 goto cleanup;
781         }
782
783         if (xs->not_found) {
784                 ret = -ENODATA;
785                 goto cleanup;
786         }
787
788         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
789         size = le64_to_cpu(xs->here->xe_value_size);
790         if (buffer) {
791                 ret = -ERANGE;
792                 if (size > buffer_size)
793                         goto cleanup;
794
795                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
796                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
797                 i = xs->here - xs->header->xh_entries;
798
799                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
800                         ret = ocfs2_xattr_bucket_get_name_value(inode,
801                                                                 bucket_xh(&xs->bucket),
802                                                                 i,
803                                                                 &block_off,
804                                                                 &name_offset);
805                         xs->base = bucket_block(&xs->bucket, block_off);
806                 }
807                 if (ocfs2_xattr_is_local(xs->here)) {
808                         memcpy(buffer, (void *)xs->base +
809                                name_offset + name_len, size);
810                 } else {
811                         xv = (struct ocfs2_xattr_value_root *)
812                                 (xs->base + name_offset + name_len);
813                         ret = ocfs2_xattr_get_value_outside(inode, xv,
814                                                             buffer, size);
815                         if (ret < 0) {
816                                 mlog_errno(ret);
817                                 goto cleanup;
818                         }
819                 }
820         }
821         ret = size;
822 cleanup:
823         for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
824                 brelse(xs->bucket.bu_bhs[i]);
825         memset(&xs->bucket, 0, sizeof(xs->bucket));
826
827         brelse(xs->xattr_bh);
828         xs->xattr_bh = NULL;
829         return ret;
830 }
831
832 /* ocfs2_xattr_get()
833  *
834  * Copy an extended attribute into the buffer provided.
835  * Buffer is NULL to compute the size of buffer required.
836  */
837 static int ocfs2_xattr_get(struct inode *inode,
838                            int name_index,
839                            const char *name,
840                            void *buffer,
841                            size_t buffer_size)
842 {
843         int ret;
844         struct ocfs2_dinode *di = NULL;
845         struct buffer_head *di_bh = NULL;
846         struct ocfs2_inode_info *oi = OCFS2_I(inode);
847         struct ocfs2_xattr_search xis = {
848                 .not_found = -ENODATA,
849         };
850         struct ocfs2_xattr_search xbs = {
851                 .not_found = -ENODATA,
852         };
853
854         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
855                 return -EOPNOTSUPP;
856
857         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
858                 ret = -ENODATA;
859
860         ret = ocfs2_inode_lock(inode, &di_bh, 0);
861         if (ret < 0) {
862                 mlog_errno(ret);
863                 return ret;
864         }
865         xis.inode_bh = xbs.inode_bh = di_bh;
866         di = (struct ocfs2_dinode *)di_bh->b_data;
867
868         down_read(&oi->ip_xattr_sem);
869         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
870                                     buffer_size, &xis);
871         if (ret == -ENODATA && di->i_xattr_loc)
872                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
873                                             buffer_size, &xbs);
874         up_read(&oi->ip_xattr_sem);
875         ocfs2_inode_unlock(inode, 0);
876
877         brelse(di_bh);
878
879         return ret;
880 }
881
882 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
883                                            struct ocfs2_xattr_value_root *xv,
884                                            const void *value,
885                                            int value_len)
886 {
887         int ret = 0, i, cp_len, credits;
888         u16 blocksize = inode->i_sb->s_blocksize;
889         u32 p_cluster, num_clusters;
890         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
891         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
892         u64 blkno;
893         struct buffer_head *bh = NULL;
894         handle_t *handle;
895
896         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
897
898         credits = clusters * bpc;
899         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
900         if (IS_ERR(handle)) {
901                 ret = PTR_ERR(handle);
902                 mlog_errno(ret);
903                 goto out;
904         }
905
906         while (cpos < clusters) {
907                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
908                                                &num_clusters, &xv->xr_list);
909                 if (ret) {
910                         mlog_errno(ret);
911                         goto out_commit;
912                 }
913
914                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
915
916                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
917                         ret = ocfs2_read_block(inode, blkno, &bh);
918                         if (ret) {
919                                 mlog_errno(ret);
920                                 goto out_commit;
921                         }
922
923                         ret = ocfs2_journal_access(handle,
924                                                    inode,
925                                                    bh,
926                                                    OCFS2_JOURNAL_ACCESS_WRITE);
927                         if (ret < 0) {
928                                 mlog_errno(ret);
929                                 goto out_commit;
930                         }
931
932                         cp_len = value_len > blocksize ? blocksize : value_len;
933                         memcpy(bh->b_data, value, cp_len);
934                         value_len -= cp_len;
935                         value += cp_len;
936                         if (cp_len < blocksize)
937                                 memset(bh->b_data + cp_len, 0,
938                                        blocksize - cp_len);
939
940                         ret = ocfs2_journal_dirty(handle, bh);
941                         if (ret < 0) {
942                                 mlog_errno(ret);
943                                 goto out_commit;
944                         }
945                         brelse(bh);
946                         bh = NULL;
947
948                         /*
949                          * XXX: do we need to empty all the following
950                          * blocks in this cluster?
951                          */
952                         if (!value_len)
953                                 break;
954                 }
955                 cpos += num_clusters;
956         }
957 out_commit:
958         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
959 out:
960         brelse(bh);
961
962         return ret;
963 }
964
965 static int ocfs2_xattr_cleanup(struct inode *inode,
966                                struct ocfs2_xattr_info *xi,
967                                struct ocfs2_xattr_search *xs,
968                                size_t offs)
969 {
970         handle_t *handle = NULL;
971         int ret = 0;
972         size_t name_len = strlen(xi->name);
973         void *val = xs->base + offs;
974         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
975
976         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
977                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
978         if (IS_ERR(handle)) {
979                 ret = PTR_ERR(handle);
980                 mlog_errno(ret);
981                 goto out;
982         }
983         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
984                                    OCFS2_JOURNAL_ACCESS_WRITE);
985         if (ret) {
986                 mlog_errno(ret);
987                 goto out_commit;
988         }
989         /* Decrease xattr count */
990         le16_add_cpu(&xs->header->xh_count, -1);
991         /* Remove the xattr entry and tree root which has already be set*/
992         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
993         memset(val, 0, size);
994
995         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
996         if (ret < 0)
997                 mlog_errno(ret);
998 out_commit:
999         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1000 out:
1001         return ret;
1002 }
1003
1004 static int ocfs2_xattr_update_entry(struct inode *inode,
1005                                     struct ocfs2_xattr_info *xi,
1006                                     struct ocfs2_xattr_search *xs,
1007                                     size_t offs)
1008 {
1009         handle_t *handle = NULL;
1010         int ret = 0;
1011
1012         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1013                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1014         if (IS_ERR(handle)) {
1015                 ret = PTR_ERR(handle);
1016                 mlog_errno(ret);
1017                 goto out;
1018         }
1019         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1020                                    OCFS2_JOURNAL_ACCESS_WRITE);
1021         if (ret) {
1022                 mlog_errno(ret);
1023                 goto out_commit;
1024         }
1025
1026         xs->here->xe_name_offset = cpu_to_le16(offs);
1027         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1028         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1029                 ocfs2_xattr_set_local(xs->here, 1);
1030         else
1031                 ocfs2_xattr_set_local(xs->here, 0);
1032         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1033
1034         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1035         if (ret < 0)
1036                 mlog_errno(ret);
1037 out_commit:
1038         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1039 out:
1040         return ret;
1041 }
1042
1043 /*
1044  * ocfs2_xattr_set_value_outside()
1045  *
1046  * Set large size value in B tree.
1047  */
1048 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1049                                          struct ocfs2_xattr_info *xi,
1050                                          struct ocfs2_xattr_search *xs,
1051                                          size_t offs)
1052 {
1053         size_t name_len = strlen(xi->name);
1054         void *val = xs->base + offs;
1055         struct ocfs2_xattr_value_root *xv = NULL;
1056         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1057         int ret = 0;
1058
1059         memset(val, 0, size);
1060         memcpy(val, xi->name, name_len);
1061         xv = (struct ocfs2_xattr_value_root *)
1062                 (val + OCFS2_XATTR_SIZE(name_len));
1063         xv->xr_clusters = 0;
1064         xv->xr_last_eb_blk = 0;
1065         xv->xr_list.l_tree_depth = 0;
1066         xv->xr_list.l_count = cpu_to_le16(1);
1067         xv->xr_list.l_next_free_rec = 0;
1068
1069         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1070                                          xi->value_len);
1071         if (ret < 0) {
1072                 mlog_errno(ret);
1073                 return ret;
1074         }
1075         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1076                                               xi->value_len);
1077         if (ret < 0) {
1078                 mlog_errno(ret);
1079                 return ret;
1080         }
1081         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1082         if (ret < 0)
1083                 mlog_errno(ret);
1084
1085         return ret;
1086 }
1087
1088 /*
1089  * ocfs2_xattr_set_entry_local()
1090  *
1091  * Set, replace or remove extended attribute in local.
1092  */
1093 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1094                                         struct ocfs2_xattr_info *xi,
1095                                         struct ocfs2_xattr_search *xs,
1096                                         struct ocfs2_xattr_entry *last,
1097                                         size_t min_offs)
1098 {
1099         size_t name_len = strlen(xi->name);
1100         int i;
1101
1102         if (xi->value && xs->not_found) {
1103                 /* Insert the new xattr entry. */
1104                 le16_add_cpu(&xs->header->xh_count, 1);
1105                 ocfs2_xattr_set_type(last, xi->name_index);
1106                 ocfs2_xattr_set_local(last, 1);
1107                 last->xe_name_len = name_len;
1108         } else {
1109                 void *first_val;
1110                 void *val;
1111                 size_t offs, size;
1112
1113                 first_val = xs->base + min_offs;
1114                 offs = le16_to_cpu(xs->here->xe_name_offset);
1115                 val = xs->base + offs;
1116
1117                 if (le64_to_cpu(xs->here->xe_value_size) >
1118                     OCFS2_XATTR_INLINE_SIZE)
1119                         size = OCFS2_XATTR_SIZE(name_len) +
1120                                 OCFS2_XATTR_ROOT_SIZE;
1121                 else
1122                         size = OCFS2_XATTR_SIZE(name_len) +
1123                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1124
1125                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1126                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1127                         /* The old and the new value have the
1128                            same size. Just replace the value. */
1129                         ocfs2_xattr_set_local(xs->here, 1);
1130                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1131                         /* Clear value bytes. */
1132                         memset(val + OCFS2_XATTR_SIZE(name_len),
1133                                0,
1134                                OCFS2_XATTR_SIZE(xi->value_len));
1135                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1136                                xi->value,
1137                                xi->value_len);
1138                         return;
1139                 }
1140                 /* Remove the old name+value. */
1141                 memmove(first_val + size, first_val, val - first_val);
1142                 memset(first_val, 0, size);
1143                 xs->here->xe_name_hash = 0;
1144                 xs->here->xe_name_offset = 0;
1145                 ocfs2_xattr_set_local(xs->here, 1);
1146                 xs->here->xe_value_size = 0;
1147
1148                 min_offs += size;
1149
1150                 /* Adjust all value offsets. */
1151                 last = xs->header->xh_entries;
1152                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1153                         size_t o = le16_to_cpu(last->xe_name_offset);
1154
1155                         if (o < offs)
1156                                 last->xe_name_offset = cpu_to_le16(o + size);
1157                         last += 1;
1158                 }
1159
1160                 if (!xi->value) {
1161                         /* Remove the old entry. */
1162                         last -= 1;
1163                         memmove(xs->here, xs->here + 1,
1164                                 (void *)last - (void *)xs->here);
1165                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1166                         le16_add_cpu(&xs->header->xh_count, -1);
1167                 }
1168         }
1169         if (xi->value) {
1170                 /* Insert the new name+value. */
1171                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1172                                 OCFS2_XATTR_SIZE(xi->value_len);
1173                 void *val = xs->base + min_offs - size;
1174
1175                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1176                 memset(val, 0, size);
1177                 memcpy(val, xi->name, name_len);
1178                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1179                        xi->value,
1180                        xi->value_len);
1181                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1182                 ocfs2_xattr_set_local(xs->here, 1);
1183                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1184         }
1185
1186         return;
1187 }
1188
1189 /*
1190  * ocfs2_xattr_set_entry()
1191  *
1192  * Set extended attribute entry into inode or block.
1193  *
1194  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1195  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1196  * then set value in B tree with set_value_outside().
1197  */
1198 static int ocfs2_xattr_set_entry(struct inode *inode,
1199                                  struct ocfs2_xattr_info *xi,
1200                                  struct ocfs2_xattr_search *xs,
1201                                  int flag)
1202 {
1203         struct ocfs2_xattr_entry *last;
1204         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1205         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1206         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1207         size_t size_l = 0;
1208         handle_t *handle = NULL;
1209         int free, i, ret;
1210         struct ocfs2_xattr_info xi_l = {
1211                 .name_index = xi->name_index,
1212                 .name = xi->name,
1213                 .value = xi->value,
1214                 .value_len = xi->value_len,
1215         };
1216
1217         /* Compute min_offs, last and free space. */
1218         last = xs->header->xh_entries;
1219
1220         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1221                 size_t offs = le16_to_cpu(last->xe_name_offset);
1222                 if (offs < min_offs)
1223                         min_offs = offs;
1224                 last += 1;
1225         }
1226
1227         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1228         if (free < 0)
1229                 return -EIO;
1230
1231         if (!xs->not_found) {
1232                 size_t size = 0;
1233                 if (ocfs2_xattr_is_local(xs->here))
1234                         size = OCFS2_XATTR_SIZE(name_len) +
1235                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1236                 else
1237                         size = OCFS2_XATTR_SIZE(name_len) +
1238                                 OCFS2_XATTR_ROOT_SIZE;
1239                 free += (size + sizeof(struct ocfs2_xattr_entry));
1240         }
1241         /* Check free space in inode or block */
1242         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1243                 if (free < sizeof(struct ocfs2_xattr_entry) +
1244                            OCFS2_XATTR_SIZE(name_len) +
1245                            OCFS2_XATTR_ROOT_SIZE) {
1246                         ret = -ENOSPC;
1247                         goto out;
1248                 }
1249                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1250                 xi_l.value = (void *)&def_xv;
1251                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1252         } else if (xi->value) {
1253                 if (free < sizeof(struct ocfs2_xattr_entry) +
1254                            OCFS2_XATTR_SIZE(name_len) +
1255                            OCFS2_XATTR_SIZE(xi->value_len)) {
1256                         ret = -ENOSPC;
1257                         goto out;
1258                 }
1259         }
1260
1261         if (!xs->not_found) {
1262                 /* For existing extended attribute */
1263                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1264                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1265                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1266                 void *val = xs->base + offs;
1267
1268                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1269                         /* Replace existing local xattr with tree root */
1270                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1271                                                             offs);
1272                         if (ret < 0)
1273                                 mlog_errno(ret);
1274                         goto out;
1275                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1276                         /* For existing xattr which has value outside */
1277                         struct ocfs2_xattr_value_root *xv = NULL;
1278                         xv = (struct ocfs2_xattr_value_root *)(val +
1279                                 OCFS2_XATTR_SIZE(name_len));
1280
1281                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1282                                 /*
1283                                  * If new value need set outside also,
1284                                  * first truncate old value to new value,
1285                                  * then set new value with set_value_outside().
1286                                  */
1287                                 ret = ocfs2_xattr_value_truncate(inode,
1288                                                                  xs->xattr_bh,
1289                                                                  xv,
1290                                                                  xi->value_len);
1291                                 if (ret < 0) {
1292                                         mlog_errno(ret);
1293                                         goto out;
1294                                 }
1295
1296                                 ret = __ocfs2_xattr_set_value_outside(inode,
1297                                                                 xv,
1298                                                                 xi->value,
1299                                                                 xi->value_len);
1300                                 if (ret < 0) {
1301                                         mlog_errno(ret);
1302                                         goto out;
1303                                 }
1304
1305                                 ret = ocfs2_xattr_update_entry(inode,
1306                                                                xi,
1307                                                                xs,
1308                                                                offs);
1309                                 if (ret < 0)
1310                                         mlog_errno(ret);
1311                                 goto out;
1312                         } else {
1313                                 /*
1314                                  * If new value need set in local,
1315                                  * just trucate old value to zero.
1316                                  */
1317                                  ret = ocfs2_xattr_value_truncate(inode,
1318                                                                  xs->xattr_bh,
1319                                                                  xv,
1320                                                                  0);
1321                                 if (ret < 0)
1322                                         mlog_errno(ret);
1323                         }
1324                 }
1325         }
1326
1327         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1328                                    OCFS2_INODE_UPDATE_CREDITS);
1329         if (IS_ERR(handle)) {
1330                 ret = PTR_ERR(handle);
1331                 mlog_errno(ret);
1332                 goto out;
1333         }
1334
1335         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1336                                    OCFS2_JOURNAL_ACCESS_WRITE);
1337         if (ret) {
1338                 mlog_errno(ret);
1339                 goto out_commit;
1340         }
1341
1342         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1343                 /* set extended attribute in external block. */
1344                 ret = ocfs2_extend_trans(handle,
1345                                          OCFS2_INODE_UPDATE_CREDITS +
1346                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1347                 if (ret) {
1348                         mlog_errno(ret);
1349                         goto out_commit;
1350                 }
1351                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1352                                            OCFS2_JOURNAL_ACCESS_WRITE);
1353                 if (ret) {
1354                         mlog_errno(ret);
1355                         goto out_commit;
1356                 }
1357         }
1358
1359         /*
1360          * Set value in local, include set tree root in local.
1361          * This is the first step for value size >INLINE_SIZE.
1362          */
1363         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1364
1365         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1366                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1367                 if (ret < 0) {
1368                         mlog_errno(ret);
1369                         goto out_commit;
1370                 }
1371         }
1372
1373         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1374             (flag & OCFS2_INLINE_XATTR_FL)) {
1375                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1376                 unsigned int xattrsize = osb->s_xattr_inline_size;
1377
1378                 /*
1379                  * Adjust extent record count or inline data size
1380                  * to reserve space for extended attribute.
1381                  */
1382                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1383                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1384                         le16_add_cpu(&idata->id_count, -xattrsize);
1385                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1386                         struct ocfs2_extent_list *el = &di->id2.i_list;
1387                         le16_add_cpu(&el->l_count, -(xattrsize /
1388                                         sizeof(struct ocfs2_extent_rec)));
1389                 }
1390                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1391         }
1392         /* Update xattr flag */
1393         spin_lock(&oi->ip_lock);
1394         oi->ip_dyn_features |= flag;
1395         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1396         spin_unlock(&oi->ip_lock);
1397         /* Update inode ctime */
1398         inode->i_ctime = CURRENT_TIME;
1399         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1400         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1401
1402         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1403         if (ret < 0)
1404                 mlog_errno(ret);
1405
1406 out_commit:
1407         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1408
1409         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1410                 /*
1411                  * Set value outside in B tree.
1412                  * This is the second step for value size > INLINE_SIZE.
1413                  */
1414                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1415                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1416                 if (ret < 0) {
1417                         int ret2;
1418
1419                         mlog_errno(ret);
1420                         /*
1421                          * If set value outside failed, we have to clean
1422                          * the junk tree root we have already set in local.
1423                          */
1424                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1425                         if (ret2 < 0)
1426                                 mlog_errno(ret2);
1427                 }
1428         }
1429 out:
1430         return ret;
1431
1432 }
1433
1434 static int ocfs2_remove_value_outside(struct inode*inode,
1435                                       struct buffer_head *bh,
1436                                       struct ocfs2_xattr_header *header)
1437 {
1438         int ret = 0, i;
1439
1440         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1441                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1442
1443                 if (!ocfs2_xattr_is_local(entry)) {
1444                         struct ocfs2_xattr_value_root *xv;
1445                         void *val;
1446
1447                         val = (void *)header +
1448                                 le16_to_cpu(entry->xe_name_offset);
1449                         xv = (struct ocfs2_xattr_value_root *)
1450                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1451                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1452                         if (ret < 0) {
1453                                 mlog_errno(ret);
1454                                 return ret;
1455                         }
1456                 }
1457         }
1458
1459         return ret;
1460 }
1461
1462 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1463                                     struct buffer_head *di_bh)
1464 {
1465
1466         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1467         struct ocfs2_xattr_header *header;
1468         int ret;
1469
1470         header = (struct ocfs2_xattr_header *)
1471                  ((void *)di + inode->i_sb->s_blocksize -
1472                  le16_to_cpu(di->i_xattr_inline_size));
1473
1474         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1475
1476         return ret;
1477 }
1478
1479 static int ocfs2_xattr_block_remove(struct inode *inode,
1480                                     struct buffer_head *blk_bh)
1481 {
1482         struct ocfs2_xattr_block *xb;
1483         int ret = 0;
1484
1485         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1486         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1487                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1488                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1489         } else
1490                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1491
1492         return ret;
1493 }
1494
1495 static int ocfs2_xattr_free_block(struct inode *inode,
1496                                   u64 block)
1497 {
1498         struct inode *xb_alloc_inode;
1499         struct buffer_head *xb_alloc_bh = NULL;
1500         struct buffer_head *blk_bh = NULL;
1501         struct ocfs2_xattr_block *xb;
1502         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1503         handle_t *handle;
1504         int ret = 0;
1505         u64 blk, bg_blkno;
1506         u16 bit;
1507
1508         ret = ocfs2_read_block(inode, block, &blk_bh);
1509         if (ret < 0) {
1510                 mlog_errno(ret);
1511                 goto out;
1512         }
1513
1514         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1515         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1516                 ret = -EIO;
1517                 goto out;
1518         }
1519
1520         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1521         if (ret < 0) {
1522                 mlog_errno(ret);
1523                 goto out;
1524         }
1525
1526         blk = le64_to_cpu(xb->xb_blkno);
1527         bit = le16_to_cpu(xb->xb_suballoc_bit);
1528         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1529
1530         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1531                                 EXTENT_ALLOC_SYSTEM_INODE,
1532                                 le16_to_cpu(xb->xb_suballoc_slot));
1533         if (!xb_alloc_inode) {
1534                 ret = -ENOMEM;
1535                 mlog_errno(ret);
1536                 goto out;
1537         }
1538         mutex_lock(&xb_alloc_inode->i_mutex);
1539
1540         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1541         if (ret < 0) {
1542                 mlog_errno(ret);
1543                 goto out_mutex;
1544         }
1545
1546         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1547         if (IS_ERR(handle)) {
1548                 ret = PTR_ERR(handle);
1549                 mlog_errno(ret);
1550                 goto out_unlock;
1551         }
1552
1553         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1554                                        bit, bg_blkno, 1);
1555         if (ret < 0)
1556                 mlog_errno(ret);
1557
1558         ocfs2_commit_trans(osb, handle);
1559 out_unlock:
1560         ocfs2_inode_unlock(xb_alloc_inode, 1);
1561         brelse(xb_alloc_bh);
1562 out_mutex:
1563         mutex_unlock(&xb_alloc_inode->i_mutex);
1564         iput(xb_alloc_inode);
1565 out:
1566         brelse(blk_bh);
1567         return ret;
1568 }
1569
1570 /*
1571  * ocfs2_xattr_remove()
1572  *
1573  * Free extended attribute resources associated with this inode.
1574  */
1575 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1576 {
1577         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1578         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1579         handle_t *handle;
1580         int ret;
1581
1582         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1583                 return 0;
1584
1585         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1586                 return 0;
1587
1588         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1589                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1590                 if (ret < 0) {
1591                         mlog_errno(ret);
1592                         goto out;
1593                 }
1594         }
1595
1596         if (di->i_xattr_loc) {
1597                 ret = ocfs2_xattr_free_block(inode,
1598                                              le64_to_cpu(di->i_xattr_loc));
1599                 if (ret < 0) {
1600                         mlog_errno(ret);
1601                         goto out;
1602                 }
1603         }
1604
1605         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1606                                    OCFS2_INODE_UPDATE_CREDITS);
1607         if (IS_ERR(handle)) {
1608                 ret = PTR_ERR(handle);
1609                 mlog_errno(ret);
1610                 goto out;
1611         }
1612         ret = ocfs2_journal_access(handle, inode, di_bh,
1613                                    OCFS2_JOURNAL_ACCESS_WRITE);
1614         if (ret) {
1615                 mlog_errno(ret);
1616                 goto out_commit;
1617         }
1618
1619         di->i_xattr_loc = 0;
1620
1621         spin_lock(&oi->ip_lock);
1622         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1623         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1624         spin_unlock(&oi->ip_lock);
1625
1626         ret = ocfs2_journal_dirty(handle, di_bh);
1627         if (ret < 0)
1628                 mlog_errno(ret);
1629 out_commit:
1630         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1631 out:
1632         return ret;
1633 }
1634
1635 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1636                                         struct ocfs2_dinode *di)
1637 {
1638         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1639         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1640         int free;
1641
1642         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1643                 return 0;
1644
1645         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1646                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1647                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1648         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1649                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1650                         le64_to_cpu(di->i_size);
1651         } else {
1652                 struct ocfs2_extent_list *el = &di->id2.i_list;
1653                 free = (le16_to_cpu(el->l_count) -
1654                         le16_to_cpu(el->l_next_free_rec)) *
1655                         sizeof(struct ocfs2_extent_rec);
1656         }
1657         if (free >= xattrsize)
1658                 return 1;
1659
1660         return 0;
1661 }
1662
1663 /*
1664  * ocfs2_xattr_ibody_find()
1665  *
1666  * Find extended attribute in inode block and
1667  * fill search info into struct ocfs2_xattr_search.
1668  */
1669 static int ocfs2_xattr_ibody_find(struct inode *inode,
1670                                   int name_index,
1671                                   const char *name,
1672                                   struct ocfs2_xattr_search *xs)
1673 {
1674         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1675         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1676         int ret;
1677         int has_space = 0;
1678
1679         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1680                 return 0;
1681
1682         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1683                 down_read(&oi->ip_alloc_sem);
1684                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1685                 up_read(&oi->ip_alloc_sem);
1686                 if (!has_space)
1687                         return 0;
1688         }
1689
1690         xs->xattr_bh = xs->inode_bh;
1691         xs->end = (void *)di + inode->i_sb->s_blocksize;
1692         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1693                 xs->header = (struct ocfs2_xattr_header *)
1694                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1695         else
1696                 xs->header = (struct ocfs2_xattr_header *)
1697                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1698         xs->base = (void *)xs->header;
1699         xs->here = xs->header->xh_entries;
1700
1701         /* Find the named attribute. */
1702         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1703                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1704                 if (ret && ret != -ENODATA)
1705                         return ret;
1706                 xs->not_found = ret;
1707         }
1708
1709         return 0;
1710 }
1711
1712 /*
1713  * ocfs2_xattr_ibody_set()
1714  *
1715  * Set, replace or remove an extended attribute into inode block.
1716  *
1717  */
1718 static int ocfs2_xattr_ibody_set(struct inode *inode,
1719                                  struct ocfs2_xattr_info *xi,
1720                                  struct ocfs2_xattr_search *xs)
1721 {
1722         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1723         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1724         int ret;
1725
1726         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1727                 return -ENOSPC;
1728
1729         down_write(&oi->ip_alloc_sem);
1730         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1731                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1732                         ret = -ENOSPC;
1733                         goto out;
1734                 }
1735         }
1736
1737         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1738                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1739 out:
1740         up_write(&oi->ip_alloc_sem);
1741
1742         return ret;
1743 }
1744
1745 /*
1746  * ocfs2_xattr_block_find()
1747  *
1748  * Find extended attribute in external block and
1749  * fill search info into struct ocfs2_xattr_search.
1750  */
1751 static int ocfs2_xattr_block_find(struct inode *inode,
1752                                   int name_index,
1753                                   const char *name,
1754                                   struct ocfs2_xattr_search *xs)
1755 {
1756         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1757         struct buffer_head *blk_bh = NULL;
1758         struct ocfs2_xattr_block *xb;
1759         int ret = 0;
1760
1761         if (!di->i_xattr_loc)
1762                 return ret;
1763
1764         ret = ocfs2_read_block(inode, le64_to_cpu(di->i_xattr_loc), &blk_bh);
1765         if (ret < 0) {
1766                 mlog_errno(ret);
1767                 return ret;
1768         }
1769
1770         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1771         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
1772                 ret = -EIO;
1773                 goto cleanup;
1774         }
1775
1776         xs->xattr_bh = blk_bh;
1777
1778         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1779                 xs->header = &xb->xb_attrs.xb_header;
1780                 xs->base = (void *)xs->header;
1781                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1782                 xs->here = xs->header->xh_entries;
1783
1784                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1785         } else
1786                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1787                                                    name_index,
1788                                                    name, xs);
1789
1790         if (ret && ret != -ENODATA) {
1791                 xs->xattr_bh = NULL;
1792                 goto cleanup;
1793         }
1794         xs->not_found = ret;
1795         return 0;
1796 cleanup:
1797         brelse(blk_bh);
1798
1799         return ret;
1800 }
1801
1802 /*
1803  * ocfs2_xattr_block_set()
1804  *
1805  * Set, replace or remove an extended attribute into external block.
1806  *
1807  */
1808 static int ocfs2_xattr_block_set(struct inode *inode,
1809                                  struct ocfs2_xattr_info *xi,
1810                                  struct ocfs2_xattr_search *xs)
1811 {
1812         struct buffer_head *new_bh = NULL;
1813         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1814         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1815         struct ocfs2_alloc_context *meta_ac = NULL;
1816         handle_t *handle = NULL;
1817         struct ocfs2_xattr_block *xblk = NULL;
1818         u16 suballoc_bit_start;
1819         u32 num_got;
1820         u64 first_blkno;
1821         int ret;
1822
1823         if (!xs->xattr_bh) {
1824                 /*
1825                  * Alloc one external block for extended attribute
1826                  * outside of inode.
1827                  */
1828                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1829                 if (ret < 0) {
1830                         mlog_errno(ret);
1831                         goto out;
1832                 }
1833                 handle = ocfs2_start_trans(osb,
1834                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1835                 if (IS_ERR(handle)) {
1836                         ret = PTR_ERR(handle);
1837                         mlog_errno(ret);
1838                         goto out;
1839                 }
1840                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1841                                            OCFS2_JOURNAL_ACCESS_CREATE);
1842                 if (ret < 0) {
1843                         mlog_errno(ret);
1844                         goto out_commit;
1845                 }
1846
1847                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1848                                            &suballoc_bit_start, &num_got,
1849                                            &first_blkno);
1850                 if (ret < 0) {
1851                         mlog_errno(ret);
1852                         goto out_commit;
1853                 }
1854
1855                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1856                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1857
1858                 ret = ocfs2_journal_access(handle, inode, new_bh,
1859                                            OCFS2_JOURNAL_ACCESS_CREATE);
1860                 if (ret < 0) {
1861                         mlog_errno(ret);
1862                         goto out_commit;
1863                 }
1864
1865                 /* Initialize ocfs2_xattr_block */
1866                 xs->xattr_bh = new_bh;
1867                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1868                 memset(xblk, 0, inode->i_sb->s_blocksize);
1869                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1870                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1871                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1872                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1873                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1874
1875                 xs->header = &xblk->xb_attrs.xb_header;
1876                 xs->base = (void *)xs->header;
1877                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1878                 xs->here = xs->header->xh_entries;
1879
1880
1881                 ret = ocfs2_journal_dirty(handle, new_bh);
1882                 if (ret < 0) {
1883                         mlog_errno(ret);
1884                         goto out_commit;
1885                 }
1886                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1887                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1888                 if (ret < 0)
1889                         mlog_errno(ret);
1890 out_commit:
1891                 ocfs2_commit_trans(osb, handle);
1892 out:
1893                 if (meta_ac)
1894                         ocfs2_free_alloc_context(meta_ac);
1895                 if (ret < 0)
1896                         return ret;
1897         } else
1898                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1899
1900         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1901                 /* Set extended attribute into external block */
1902                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1903                 if (!ret || ret != -ENOSPC)
1904                         goto end;
1905
1906                 ret = ocfs2_xattr_create_index_block(inode, xs);
1907                 if (ret)
1908                         goto end;
1909         }
1910
1911         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1912
1913 end:
1914
1915         return ret;
1916 }
1917
1918 /*
1919  * ocfs2_xattr_set()
1920  *
1921  * Set, replace or remove an extended attribute for this inode.
1922  * value is NULL to remove an existing extended attribute, else either
1923  * create or replace an extended attribute.
1924  */
1925 int ocfs2_xattr_set(struct inode *inode,
1926                     int name_index,
1927                     const char *name,
1928                     const void *value,
1929                     size_t value_len,
1930                     int flags)
1931 {
1932         struct buffer_head *di_bh = NULL;
1933         struct ocfs2_dinode *di;
1934         int ret;
1935         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1936
1937         struct ocfs2_xattr_info xi = {
1938                 .name_index = name_index,
1939                 .name = name,
1940                 .value = value,
1941                 .value_len = value_len,
1942         };
1943
1944         struct ocfs2_xattr_search xis = {
1945                 .not_found = -ENODATA,
1946         };
1947
1948         struct ocfs2_xattr_search xbs = {
1949                 .not_found = -ENODATA,
1950         };
1951
1952         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1953                 return -EOPNOTSUPP;
1954
1955         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1956         if (ret < 0) {
1957                 mlog_errno(ret);
1958                 return ret;
1959         }
1960         xis.inode_bh = xbs.inode_bh = di_bh;
1961         di = (struct ocfs2_dinode *)di_bh->b_data;
1962
1963         down_write(&OCFS2_I(inode)->ip_xattr_sem);
1964         /*
1965          * Scan inode and external block to find the same name
1966          * extended attribute and collect search infomation.
1967          */
1968         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
1969         if (ret)
1970                 goto cleanup;
1971         if (xis.not_found) {
1972                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
1973                 if (ret)
1974                         goto cleanup;
1975         }
1976
1977         if (xis.not_found && xbs.not_found) {
1978                 ret = -ENODATA;
1979                 if (flags & XATTR_REPLACE)
1980                         goto cleanup;
1981                 ret = 0;
1982                 if (!value)
1983                         goto cleanup;
1984         } else {
1985                 ret = -EEXIST;
1986                 if (flags & XATTR_CREATE)
1987                         goto cleanup;
1988         }
1989
1990         if (!value) {
1991                 /* Remove existing extended attribute */
1992                 if (!xis.not_found)
1993                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
1994                 else if (!xbs.not_found)
1995                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
1996         } else {
1997                 /* We always try to set extended attribute into inode first*/
1998                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
1999                 if (!ret && !xbs.not_found) {
2000                         /*
2001                          * If succeed and that extended attribute existing in
2002                          * external block, then we will remove it.
2003                          */
2004                         xi.value = NULL;
2005                         xi.value_len = 0;
2006                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2007                 } else if (ret == -ENOSPC) {
2008                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2009                                 ret = ocfs2_xattr_block_find(inode, name_index,
2010                                                              name, &xbs);
2011                                 if (ret)
2012                                         goto cleanup;
2013                         }
2014                         /*
2015                          * If no space in inode, we will set extended attribute
2016                          * into external block.
2017                          */
2018                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2019                         if (ret)
2020                                 goto cleanup;
2021                         if (!xis.not_found) {
2022                                 /*
2023                                  * If succeed and that extended attribute
2024                                  * existing in inode, we will remove it.
2025                                  */
2026                                 xi.value = NULL;
2027                                 xi.value_len = 0;
2028                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2029                         }
2030                 }
2031         }
2032 cleanup:
2033         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2034         ocfs2_inode_unlock(inode, 1);
2035         brelse(di_bh);
2036         brelse(xbs.xattr_bh);
2037         for (i = 0; i < blk_per_bucket; i++)
2038                 brelse(xbs.bucket.bu_bhs[i]);
2039
2040         return ret;
2041 }
2042
2043 /*
2044  * Find the xattr extent rec which may contains name_hash.
2045  * e_cpos will be the first name hash of the xattr rec.
2046  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2047  */
2048 static int ocfs2_xattr_get_rec(struct inode *inode,
2049                                u32 name_hash,
2050                                u64 *p_blkno,
2051                                u32 *e_cpos,
2052                                u32 *num_clusters,
2053                                struct ocfs2_extent_list *el)
2054 {
2055         int ret = 0, i;
2056         struct buffer_head *eb_bh = NULL;
2057         struct ocfs2_extent_block *eb;
2058         struct ocfs2_extent_rec *rec = NULL;
2059         u64 e_blkno = 0;
2060
2061         if (el->l_tree_depth) {
2062                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2063                 if (ret) {
2064                         mlog_errno(ret);
2065                         goto out;
2066                 }
2067
2068                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2069                 el = &eb->h_list;
2070
2071                 if (el->l_tree_depth) {
2072                         ocfs2_error(inode->i_sb,
2073                                     "Inode %lu has non zero tree depth in "
2074                                     "xattr tree block %llu\n", inode->i_ino,
2075                                     (unsigned long long)eb_bh->b_blocknr);
2076                         ret = -EROFS;
2077                         goto out;
2078                 }
2079         }
2080
2081         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2082                 rec = &el->l_recs[i];
2083
2084                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2085                         e_blkno = le64_to_cpu(rec->e_blkno);
2086                         break;
2087                 }
2088         }
2089
2090         if (!e_blkno) {
2091                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2092                             "record (%u, %u, 0) in xattr", inode->i_ino,
2093                             le32_to_cpu(rec->e_cpos),
2094                             ocfs2_rec_clusters(el, rec));
2095                 ret = -EROFS;
2096                 goto out;
2097         }
2098
2099         *p_blkno = le64_to_cpu(rec->e_blkno);
2100         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2101         if (e_cpos)
2102                 *e_cpos = le32_to_cpu(rec->e_cpos);
2103 out:
2104         brelse(eb_bh);
2105         return ret;
2106 }
2107
2108 typedef int (xattr_bucket_func)(struct inode *inode,
2109                                 struct ocfs2_xattr_bucket *bucket,
2110                                 void *para);
2111
2112 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2113                                    struct buffer_head *header_bh,
2114                                    int name_index,
2115                                    const char *name,
2116                                    u32 name_hash,
2117                                    u16 *xe_index,
2118                                    int *found)
2119 {
2120         int i, ret = 0, cmp = 1, block_off, new_offset;
2121         struct ocfs2_xattr_header *xh =
2122                         (struct ocfs2_xattr_header *)header_bh->b_data;
2123         size_t name_len = strlen(name);
2124         struct ocfs2_xattr_entry *xe = NULL;
2125         struct buffer_head *name_bh = NULL;
2126         char *xe_name;
2127
2128         /*
2129          * We don't use binary search in the bucket because there
2130          * may be multiple entries with the same name hash.
2131          */
2132         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2133                 xe = &xh->xh_entries[i];
2134
2135                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2136                         continue;
2137                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2138                         break;
2139
2140                 cmp = name_index - ocfs2_xattr_get_type(xe);
2141                 if (!cmp)
2142                         cmp = name_len - xe->xe_name_len;
2143                 if (cmp)
2144                         continue;
2145
2146                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2147                                                         xh,
2148                                                         i,
2149                                                         &block_off,
2150                                                         &new_offset);
2151                 if (ret) {
2152                         mlog_errno(ret);
2153                         break;
2154                 }
2155
2156                 ret = ocfs2_read_block(inode, header_bh->b_blocknr + block_off,
2157                                        &name_bh);
2158                 if (ret) {
2159                         mlog_errno(ret);
2160                         break;
2161                 }
2162                 xe_name = name_bh->b_data + new_offset;
2163
2164                 cmp = memcmp(name, xe_name, name_len);
2165                 brelse(name_bh);
2166                 name_bh = NULL;
2167
2168                 if (cmp == 0) {
2169                         *xe_index = i;
2170                         *found = 1;
2171                         ret = 0;
2172                         break;
2173                 }
2174         }
2175
2176         return ret;
2177 }
2178
2179 /*
2180  * Find the specified xattr entry in a series of buckets.
2181  * This series start from p_blkno and last for num_clusters.
2182  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2183  * the num of the valid buckets.
2184  *
2185  * Return the buffer_head this xattr should reside in. And if the xattr's
2186  * hash is in the gap of 2 buckets, return the lower bucket.
2187  */
2188 static int ocfs2_xattr_bucket_find(struct inode *inode,
2189                                    int name_index,
2190                                    const char *name,
2191                                    u32 name_hash,
2192                                    u64 p_blkno,
2193                                    u32 first_hash,
2194                                    u32 num_clusters,
2195                                    struct ocfs2_xattr_search *xs)
2196 {
2197         int ret, found = 0;
2198         struct buffer_head *bh = NULL;
2199         struct buffer_head *lower_bh = NULL;
2200         struct ocfs2_xattr_header *xh = NULL;
2201         struct ocfs2_xattr_entry *xe = NULL;
2202         u16 index = 0;
2203         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2204         int low_bucket = 0, bucket, high_bucket;
2205         u32 last_hash;
2206         u64 blkno;
2207
2208         ret = ocfs2_read_block(inode, p_blkno, &bh);
2209         if (ret) {
2210                 mlog_errno(ret);
2211                 goto out;
2212         }
2213
2214         xh = (struct ocfs2_xattr_header *)bh->b_data;
2215         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2216
2217         while (low_bucket <= high_bucket) {
2218                 brelse(bh);
2219                 bh = NULL;
2220                 bucket = (low_bucket + high_bucket) / 2;
2221
2222                 blkno = p_blkno + bucket * blk_per_bucket;
2223
2224                 ret = ocfs2_read_block(inode, blkno, &bh);
2225                 if (ret) {
2226                         mlog_errno(ret);
2227                         goto out;
2228                 }
2229
2230                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2231                 xe = &xh->xh_entries[0];
2232                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2233                         high_bucket = bucket - 1;
2234                         continue;
2235                 }
2236
2237                 /*
2238                  * Check whether the hash of the last entry in our
2239                  * bucket is larger than the search one. for an empty
2240                  * bucket, the last one is also the first one.
2241                  */
2242                 if (xh->xh_count)
2243                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2244
2245                 last_hash = le32_to_cpu(xe->xe_name_hash);
2246
2247                 /* record lower_bh which may be the insert place. */
2248                 brelse(lower_bh);
2249                 lower_bh = bh;
2250                 bh = NULL;
2251
2252                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2253                         low_bucket = bucket + 1;
2254                         continue;
2255                 }
2256
2257                 /* the searched xattr should reside in this bucket if exists. */
2258                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2259                                               name_index, name, name_hash,
2260                                               &index, &found);
2261                 if (ret) {
2262                         mlog_errno(ret);
2263                         goto out;
2264                 }
2265                 break;
2266         }
2267
2268         /*
2269          * Record the bucket we have found.
2270          * When the xattr's hash value is in the gap of 2 buckets, we will
2271          * always set it to the previous bucket.
2272          */
2273         if (!lower_bh) {
2274                 /*
2275                  * We can't find any bucket whose first name_hash is less
2276                  * than the find name_hash.
2277                  */
2278                 BUG_ON(bh->b_blocknr != p_blkno);
2279                 lower_bh = bh;
2280                 bh = NULL;
2281         }
2282         xs->bucket.bu_bhs[0] = lower_bh;
2283         lower_bh = NULL;
2284
2285         xs->header = bucket_xh(&xs->bucket);
2286         xs->base = bucket_block(&xs->bucket, 0);
2287         xs->end = xs->base + inode->i_sb->s_blocksize;
2288
2289         if (found) {
2290                 /*
2291                  * If we have found the xattr enty, read all the blocks in
2292                  * this bucket.
2293                  */
2294                 ret = ocfs2_read_blocks(inode, bucket_blkno(&xs->bucket) + 1,
2295                                         blk_per_bucket - 1, &xs->bucket.bu_bhs[1],
2296                                         0);
2297                 if (ret) {
2298                         mlog_errno(ret);
2299                         goto out;
2300                 }
2301
2302                 xs->here = &xs->header->xh_entries[index];
2303                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2304                      (unsigned long long)bucket_blkno(&xs->bucket), index);
2305         } else
2306                 ret = -ENODATA;
2307
2308 out:
2309         brelse(bh);
2310         brelse(lower_bh);
2311         return ret;
2312 }
2313
2314 static int ocfs2_xattr_index_block_find(struct inode *inode,
2315                                         struct buffer_head *root_bh,
2316                                         int name_index,
2317                                         const char *name,
2318                                         struct ocfs2_xattr_search *xs)
2319 {
2320         int ret;
2321         struct ocfs2_xattr_block *xb =
2322                         (struct ocfs2_xattr_block *)root_bh->b_data;
2323         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2324         struct ocfs2_extent_list *el = &xb_root->xt_list;
2325         u64 p_blkno = 0;
2326         u32 first_hash, num_clusters = 0;
2327         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2328
2329         if (le16_to_cpu(el->l_next_free_rec) == 0)
2330                 return -ENODATA;
2331
2332         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2333              name, name_hash, name_index);
2334
2335         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2336                                   &num_clusters, el);
2337         if (ret) {
2338                 mlog_errno(ret);
2339                 goto out;
2340         }
2341
2342         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2343
2344         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2345              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
2346              first_hash);
2347
2348         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2349                                       p_blkno, first_hash, num_clusters, xs);
2350
2351 out:
2352         return ret;
2353 }
2354
2355 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2356                                        u64 blkno,
2357                                        u32 clusters,
2358                                        xattr_bucket_func *func,
2359                                        void *para)
2360 {
2361         int i, j, ret = 0;
2362         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2363         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2364         u32 num_buckets = clusters * bpc;
2365         struct ocfs2_xattr_bucket bucket;
2366
2367         memset(&bucket, 0, sizeof(bucket));
2368
2369         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2370              clusters, (unsigned long long)blkno);
2371
2372         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2373                 ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket,
2374                                         bucket.bu_bhs, 0);
2375                 if (ret) {
2376                         mlog_errno(ret);
2377                         goto out;
2378                 }
2379
2380                 /*
2381                  * The real bucket num in this series of blocks is stored
2382                  * in the 1st bucket.
2383                  */
2384                 if (i == 0)
2385                         num_buckets = le16_to_cpu(bucket_xh(&bucket)->xh_num_buckets);
2386
2387                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
2388                      (unsigned long long)blkno,
2389                      le32_to_cpu(bucket_xh(&bucket)->xh_entries[0].xe_name_hash));
2390                 if (func) {
2391                         ret = func(inode, &bucket, para);
2392                         if (ret) {
2393                                 mlog_errno(ret);
2394                                 break;
2395                         }
2396                 }
2397
2398                 for (j = 0; j < blk_per_bucket; j++)
2399                         brelse(bucket.bu_bhs[j]);
2400                 memset(&bucket, 0, sizeof(bucket));
2401         }
2402
2403 out:
2404         for (j = 0; j < blk_per_bucket; j++)
2405                 brelse(bucket.bu_bhs[j]);
2406
2407         return ret;
2408 }
2409
2410 struct ocfs2_xattr_tree_list {
2411         char *buffer;
2412         size_t buffer_size;
2413         size_t result;
2414 };
2415
2416 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2417                                              struct ocfs2_xattr_header *xh,
2418                                              int index,
2419                                              int *block_off,
2420                                              int *new_offset)
2421 {
2422         u16 name_offset;
2423
2424         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2425                 return -EINVAL;
2426
2427         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2428
2429         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2430         *new_offset = name_offset % inode->i_sb->s_blocksize;
2431
2432         return 0;
2433 }
2434
2435 static int ocfs2_list_xattr_bucket(struct inode *inode,
2436                                    struct ocfs2_xattr_bucket *bucket,
2437                                    void *para)
2438 {
2439         int ret = 0, type;
2440         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2441         int i, block_off, new_offset;
2442         const char *prefix, *name;
2443
2444         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
2445                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
2446                 type = ocfs2_xattr_get_type(entry);
2447                 prefix = ocfs2_xattr_prefix(type);
2448
2449                 if (prefix) {
2450                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2451                                                                 bucket_xh(bucket),
2452                                                                 i,
2453                                                                 &block_off,
2454                                                                 &new_offset);
2455                         if (ret)
2456                                 break;
2457
2458                         name = (const char *)bucket_block(bucket, block_off) +
2459                                 new_offset;
2460                         ret = ocfs2_xattr_list_entry(xl->buffer,
2461                                                      xl->buffer_size,
2462                                                      &xl->result,
2463                                                      prefix, name,
2464                                                      entry->xe_name_len);
2465                         if (ret)
2466                                 break;
2467                 }
2468         }
2469
2470         return ret;
2471 }
2472
2473 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2474                                              struct ocfs2_xattr_tree_root *xt,
2475                                              char *buffer,
2476                                              size_t buffer_size)
2477 {
2478         struct ocfs2_extent_list *el = &xt->xt_list;
2479         int ret = 0;
2480         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2481         u64 p_blkno = 0;
2482         struct ocfs2_xattr_tree_list xl = {
2483                 .buffer = buffer,
2484                 .buffer_size = buffer_size,
2485                 .result = 0,
2486         };
2487
2488         if (le16_to_cpu(el->l_next_free_rec) == 0)
2489                 return 0;
2490
2491         while (name_hash > 0) {
2492                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2493                                           &e_cpos, &num_clusters, el);
2494                 if (ret) {
2495                         mlog_errno(ret);
2496                         goto out;
2497                 }
2498
2499                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2500                                                   ocfs2_list_xattr_bucket,
2501                                                   &xl);
2502                 if (ret) {
2503                         mlog_errno(ret);
2504                         goto out;
2505                 }
2506
2507                 if (e_cpos == 0)
2508                         break;
2509
2510                 name_hash = e_cpos - 1;
2511         }
2512
2513         ret = xl.result;
2514 out:
2515         return ret;
2516 }
2517
2518 static int cmp_xe(const void *a, const void *b)
2519 {
2520         const struct ocfs2_xattr_entry *l = a, *r = b;
2521         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2522         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2523
2524         if (l_hash > r_hash)
2525                 return 1;
2526         if (l_hash < r_hash)
2527                 return -1;
2528         return 0;
2529 }
2530
2531 static void swap_xe(void *a, void *b, int size)
2532 {
2533         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2534
2535         tmp = *l;
2536         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2537         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2538 }
2539
2540 /*
2541  * When the ocfs2_xattr_block is filled up, new bucket will be created
2542  * and all the xattr entries will be moved to the new bucket.
2543  * Note: we need to sort the entries since they are not saved in order
2544  * in the ocfs2_xattr_block.
2545  */
2546 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2547                                            struct buffer_head *xb_bh,
2548                                            struct buffer_head *xh_bh,
2549                                            struct buffer_head *data_bh)
2550 {
2551         int i, blocksize = inode->i_sb->s_blocksize;
2552         u16 offset, size, off_change;
2553         struct ocfs2_xattr_entry *xe;
2554         struct ocfs2_xattr_block *xb =
2555                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2556         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2557         struct ocfs2_xattr_header *xh =
2558                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2559         u16 count = le16_to_cpu(xb_xh->xh_count);
2560         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2561
2562         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2563              (unsigned long long)xb_bh->b_blocknr,
2564              (unsigned long long)xh_bh->b_blocknr);
2565
2566         memset(xh_bh->b_data, 0, blocksize);
2567         if (data_bh)
2568                 memset(data_bh->b_data, 0, blocksize);
2569         /*
2570          * Since the xe_name_offset is based on ocfs2_xattr_header,
2571          * there is a offset change corresponding to the change of
2572          * ocfs2_xattr_header's position.
2573          */
2574         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2575         xe = &xb_xh->xh_entries[count - 1];
2576         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2577         size = blocksize - offset;
2578
2579         /* copy all the names and values. */
2580         if (data_bh)
2581                 target = data_bh->b_data;
2582         memcpy(target + offset, src + offset, size);
2583
2584         /* Init new header now. */
2585         xh->xh_count = xb_xh->xh_count;
2586         xh->xh_num_buckets = cpu_to_le16(1);
2587         xh->xh_name_value_len = cpu_to_le16(size);
2588         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2589
2590         /* copy all the entries. */
2591         target = xh_bh->b_data;
2592         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2593         size = count * sizeof(struct ocfs2_xattr_entry);
2594         memcpy(target + offset, (char *)xb_xh + offset, size);
2595
2596         /* Change the xe offset for all the xe because of the move. */
2597         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2598                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2599         for (i = 0; i < count; i++)
2600                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2601
2602         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2603              offset, size, off_change);
2604
2605         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2606              cmp_xe, swap_xe);
2607 }
2608
2609 /*
2610  * After we move xattr from block to index btree, we have to
2611  * update ocfs2_xattr_search to the new xe and base.
2612  *
2613  * When the entry is in xattr block, xattr_bh indicates the storage place.
2614  * While if the entry is in index b-tree, "bucket" indicates the
2615  * real place of the xattr.
2616  */
2617 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2618                                            struct ocfs2_xattr_search *xs,
2619                                            struct buffer_head *old_bh,
2620                                            struct buffer_head *new_bh)
2621 {
2622         int ret = 0;
2623         char *buf = old_bh->b_data;
2624         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2625         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2626         int i, blocksize = inode->i_sb->s_blocksize;
2627         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2628
2629         xs->bucket.bu_bhs[0] = new_bh;
2630         get_bh(new_bh);
2631         xs->header = bucket_xh(&xs->bucket);
2632
2633         xs->base = new_bh->b_data;
2634         xs->end = xs->base + inode->i_sb->s_blocksize;
2635
2636         if (!xs->not_found) {
2637                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2638                         ret = ocfs2_read_blocks(inode,
2639                                         bucket_blkno(&xs->bucket) + 1,
2640                                         blk_per_bucket - 1, &xs->bucket.bu_bhs[1],
2641                                         0);
2642                         if (ret) {
2643                                 mlog_errno(ret);
2644                                 return ret;
2645                         }
2646
2647                 }
2648                 i = xs->here - old_xh->xh_entries;
2649                 xs->here = &xs->header->xh_entries[i];
2650         }
2651
2652         return ret;
2653 }
2654
2655 static int ocfs2_xattr_create_index_block(struct inode *inode,
2656                                           struct ocfs2_xattr_search *xs)
2657 {
2658         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2659         u32 bit_off, len;
2660         u64 blkno;
2661         handle_t *handle;
2662         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2663         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2664         struct ocfs2_alloc_context *data_ac;
2665         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2666         struct buffer_head *xb_bh = xs->xattr_bh;
2667         struct ocfs2_xattr_block *xb =
2668                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2669         struct ocfs2_xattr_tree_root *xr;
2670         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2671         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2672
2673         mlog(0, "create xattr index block for %llu\n",
2674              (unsigned long long)xb_bh->b_blocknr);
2675
2676         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2677
2678         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2679         if (ret) {
2680                 mlog_errno(ret);
2681                 goto out;
2682         }
2683
2684         /*
2685          * XXX:
2686          * We can use this lock for now, and maybe move to a dedicated mutex
2687          * if performance becomes a problem later.
2688          */
2689         down_write(&oi->ip_alloc_sem);
2690
2691         /*
2692          * 3 more credits, one for xattr block update, one for the 1st block
2693          * of the new xattr bucket and one for the value/data.
2694          */
2695         credits += 3;
2696         handle = ocfs2_start_trans(osb, credits);
2697         if (IS_ERR(handle)) {
2698                 ret = PTR_ERR(handle);
2699                 mlog_errno(ret);
2700                 goto out_sem;
2701         }
2702
2703         ret = ocfs2_journal_access(handle, inode, xb_bh,
2704                                    OCFS2_JOURNAL_ACCESS_WRITE);
2705         if (ret) {
2706                 mlog_errno(ret);
2707                 goto out_commit;
2708         }
2709
2710         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2711         if (ret) {
2712                 mlog_errno(ret);
2713                 goto out_commit;
2714         }
2715
2716         /*
2717          * The bucket may spread in many blocks, and
2718          * we will only touch the 1st block and the last block
2719          * in the whole bucket(one for entry and one for data).
2720          */
2721         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2722
2723         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
2724              (unsigned long long)blkno);
2725
2726         xh_bh = sb_getblk(inode->i_sb, blkno);
2727         if (!xh_bh) {
2728                 ret = -EIO;
2729                 mlog_errno(ret);
2730                 goto out_commit;
2731         }
2732
2733         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2734
2735         ret = ocfs2_journal_access(handle, inode, xh_bh,
2736                                    OCFS2_JOURNAL_ACCESS_CREATE);
2737         if (ret) {
2738                 mlog_errno(ret);
2739                 goto out_commit;
2740         }
2741
2742         if (bpb > 1) {
2743                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2744                 if (!data_bh) {
2745                         ret = -EIO;
2746                         mlog_errno(ret);
2747                         goto out_commit;
2748                 }
2749
2750                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2751
2752                 ret = ocfs2_journal_access(handle, inode, data_bh,
2753                                            OCFS2_JOURNAL_ACCESS_CREATE);
2754                 if (ret) {
2755                         mlog_errno(ret);
2756                         goto out_commit;
2757                 }
2758         }
2759
2760         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2761
2762         ocfs2_journal_dirty(handle, xh_bh);
2763         if (data_bh)
2764                 ocfs2_journal_dirty(handle, data_bh);
2765
2766         ret = ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2767         if (ret) {
2768                 mlog_errno(ret);
2769                 goto out_commit;
2770         }
2771
2772         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2773         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2774                offsetof(struct ocfs2_xattr_block, xb_attrs));
2775
2776         xr = &xb->xb_attrs.xb_root;
2777         xr->xt_clusters = cpu_to_le32(1);
2778         xr->xt_last_eb_blk = 0;
2779         xr->xt_list.l_tree_depth = 0;
2780         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2781         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2782
2783         xr->xt_list.l_recs[0].e_cpos = 0;
2784         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2785         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2786
2787         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2788
2789         ret = ocfs2_journal_dirty(handle, xb_bh);
2790         if (ret) {
2791                 mlog_errno(ret);
2792                 goto out_commit;
2793         }
2794
2795 out_commit:
2796         ocfs2_commit_trans(osb, handle);
2797
2798 out_sem:
2799         up_write(&oi->ip_alloc_sem);
2800
2801 out:
2802         if (data_ac)
2803                 ocfs2_free_alloc_context(data_ac);
2804
2805         brelse(xh_bh);
2806         brelse(data_bh);
2807
2808         return ret;
2809 }
2810
2811 static int cmp_xe_offset(const void *a, const void *b)
2812 {
2813         const struct ocfs2_xattr_entry *l = a, *r = b;
2814         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2815         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2816
2817         if (l_name_offset < r_name_offset)
2818                 return 1;
2819         if (l_name_offset > r_name_offset)
2820                 return -1;
2821         return 0;
2822 }
2823
2824 /*
2825  * defrag a xattr bucket if we find that the bucket has some
2826  * holes beteen name/value pairs.
2827  * We will move all the name/value pairs to the end of the bucket
2828  * so that we can spare some space for insertion.
2829  */
2830 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2831                                      struct ocfs2_xattr_bucket *bucket)
2832 {
2833         int ret, i;
2834         size_t end, offset, len, value_len;
2835         struct ocfs2_xattr_header *xh;
2836         char *entries, *buf, *bucket_buf = NULL;
2837         u64 blkno = bucket_blkno(bucket);
2838         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2839         u16 xh_free_start;
2840         size_t blocksize = inode->i_sb->s_blocksize;
2841         handle_t *handle;
2842         struct buffer_head **bhs;
2843         struct ocfs2_xattr_entry *xe;
2844
2845         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2846                         GFP_NOFS);
2847         if (!bhs)
2848                 return -ENOMEM;
2849
2850         ret = ocfs2_read_blocks(inode, blkno, blk_per_bucket, bhs, 0);
2851         if (ret)
2852                 goto out;
2853
2854         /*
2855          * In order to make the operation more efficient and generic,
2856          * we copy all the blocks into a contiguous memory and do the
2857          * defragment there, so if anything is error, we will not touch
2858          * the real block.
2859          */
2860         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2861         if (!bucket_buf) {
2862                 ret = -EIO;
2863                 goto out;
2864         }
2865
2866         buf = bucket_buf;
2867         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2868                 memcpy(buf, bhs[i]->b_data, blocksize);
2869
2870         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2871         if (IS_ERR(handle)) {
2872                 ret = PTR_ERR(handle);
2873                 handle = NULL;
2874                 mlog_errno(ret);
2875                 goto out;
2876         }
2877
2878         for (i = 0; i < blk_per_bucket; i++) {
2879                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2880                                            OCFS2_JOURNAL_ACCESS_WRITE);
2881                 if (ret < 0) {
2882                         mlog_errno(ret);
2883                         goto commit;
2884                 }
2885         }
2886
2887         xh = (struct ocfs2_xattr_header *)bucket_buf;
2888         entries = (char *)xh->xh_entries;
2889         xh_free_start = le16_to_cpu(xh->xh_free_start);
2890
2891         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2892              "xh_free_start = %u, xh_name_value_len = %u.\n",
2893              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
2894              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
2895
2896         /*
2897          * sort all the entries by their offset.
2898          * the largest will be the first, so that we can
2899          * move them to the end one by one.
2900          */
2901         sort(entries, le16_to_cpu(xh->xh_count),
2902              sizeof(struct ocfs2_xattr_entry),
2903              cmp_xe_offset, swap_xe);
2904
2905         /* Move all name/values to the end of the bucket. */
2906         xe = xh->xh_entries;
2907         end = OCFS2_XATTR_BUCKET_SIZE;
2908         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2909                 offset = le16_to_cpu(xe->xe_name_offset);
2910                 if (ocfs2_xattr_is_local(xe))
2911                         value_len = OCFS2_XATTR_SIZE(
2912                                         le64_to_cpu(xe->xe_value_size));
2913                 else
2914                         value_len = OCFS2_XATTR_ROOT_SIZE;
2915                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2916
2917                 /*
2918                  * We must make sure that the name/value pair
2919                  * exist in the same block. So adjust end to
2920                  * the previous block end if needed.
2921                  */
2922                 if (((end - len) / blocksize !=
2923                         (end - 1) / blocksize))
2924                         end = end - end % blocksize;
2925
2926                 if (end > offset + len) {
2927                         memmove(bucket_buf + end - len,
2928                                 bucket_buf + offset, len);
2929                         xe->xe_name_offset = cpu_to_le16(end - len);
2930                 }
2931
2932                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2933                                 "bucket %llu\n", (unsigned long long)blkno);
2934
2935                 end -= len;
2936         }
2937
2938         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2939                         "bucket %llu\n", (unsigned long long)blkno);
2940
2941         if (xh_free_start == end)
2942                 goto commit;
2943
2944         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2945         xh->xh_free_start = cpu_to_le16(end);
2946
2947         /* sort the entries by their name_hash. */
2948         sort(entries, le16_to_cpu(xh->xh_count),
2949              sizeof(struct ocfs2_xattr_entry),
2950              cmp_xe, swap_xe);
2951
2952         buf = bucket_buf;
2953         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
2954                 memcpy(bhs[i]->b_data, buf, blocksize);
2955                 ocfs2_journal_dirty(handle, bhs[i]);
2956         }
2957
2958 commit:
2959         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2960 out:
2961
2962         if (bhs) {
2963                 for (i = 0; i < blk_per_bucket; i++)
2964                         brelse(bhs[i]);
2965         }
2966         kfree(bhs);
2967
2968         kfree(bucket_buf);
2969         return ret;
2970 }
2971
2972 /*
2973  * Move half nums of the xattr bucket in the previous cluster to this new
2974  * cluster. We only touch the last cluster of the previous extend record.
2975  *
2976  * first_bh is the first buffer_head of a series of bucket in the same
2977  * extent rec and header_bh is the header of one bucket in this cluster.
2978  * They will be updated if we move the data header_bh contains to the new
2979  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
2980  */
2981 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
2982                                                handle_t *handle,
2983                                                struct buffer_head **first_bh,
2984                                                struct buffer_head **header_bh,
2985                                                u64 new_blkno,
2986                                                u64 prev_blkno,
2987                                                u32 num_clusters,
2988                                                u32 *first_hash)
2989 {
2990         int i, ret, credits;
2991         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2992         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
2993         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
2994         int blocksize = inode->i_sb->s_blocksize;
2995         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
2996         struct ocfs2_xattr_header *new_xh;
2997         struct ocfs2_xattr_header *xh =
2998                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
2999
3000         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3001         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3002
3003         prev_bh = *first_bh;
3004         get_bh(prev_bh);
3005         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3006
3007         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3008
3009         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3010              (unsigned long long)prev_blkno, (unsigned long long)new_blkno);
3011
3012         /*
3013          * We need to update the 1st half of the new cluster and
3014          * 1 more for the update of the 1st bucket of the previous
3015          * extent record.
3016          */
3017         credits = bpc / 2 + 1;
3018         ret = ocfs2_extend_trans(handle, credits);
3019         if (ret) {
3020                 mlog_errno(ret);
3021                 goto out;
3022         }
3023
3024         ret = ocfs2_journal_access(handle, inode, prev_bh,
3025                                    OCFS2_JOURNAL_ACCESS_WRITE);
3026         if (ret) {
3027                 mlog_errno(ret);
3028                 goto out;
3029         }
3030
3031         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3032                 old_bh = new_bh = NULL;
3033                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3034                 if (!new_bh) {
3035                         ret = -EIO;
3036                         mlog_errno(ret);
3037                         goto out;
3038                 }
3039
3040                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3041
3042                 ret = ocfs2_journal_access(handle, inode, new_bh,
3043                                            OCFS2_JOURNAL_ACCESS_CREATE);
3044                 if (ret < 0) {
3045                         mlog_errno(ret);
3046                         brelse(new_bh);
3047                         goto out;
3048                 }
3049
3050                 ret = ocfs2_read_block(inode, prev_blkno, &old_bh);
3051                 if (ret < 0) {
3052                         mlog_errno(ret);
3053                         brelse(new_bh);
3054                         goto out;
3055                 }
3056
3057                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3058
3059                 if (i == 0) {
3060                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3061                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3062
3063                         if (first_hash)
3064                                 *first_hash = le32_to_cpu(
3065                                         new_xh->xh_entries[0].xe_name_hash);
3066                         new_first_bh = new_bh;
3067                         get_bh(new_first_bh);
3068                 }
3069
3070                 ocfs2_journal_dirty(handle, new_bh);
3071
3072                 if (*header_bh == old_bh) {
3073                         brelse(*header_bh);
3074                         *header_bh = new_bh;
3075                         get_bh(*header_bh);
3076
3077                         brelse(*first_bh);
3078                         *first_bh = new_first_bh;
3079                         get_bh(*first_bh);
3080                 }
3081                 brelse(new_bh);
3082                 brelse(old_bh);
3083         }
3084
3085         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3086
3087         ocfs2_journal_dirty(handle, prev_bh);
3088 out:
3089         brelse(prev_bh);
3090         brelse(new_first_bh);
3091         return ret;
3092 }
3093
3094 static int ocfs2_read_xattr_bucket(struct inode *inode,
3095                                    u64 blkno,
3096                                    struct buffer_head **bhs,
3097                                    int new)
3098 {
3099         int ret = 0;
3100         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3101
3102         if (!new)
3103                 return ocfs2_read_blocks(inode, blkno,
3104                                          blk_per_bucket, bhs, 0);
3105
3106         for (i = 0; i < blk_per_bucket; i++) {
3107                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3108                 if (bhs[i] == NULL) {
3109                         ret = -EIO;
3110                         mlog_errno(ret);
3111                         break;
3112                 }
3113                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3114         }
3115
3116         return ret;
3117 }
3118
3119 /*
3120  * Find the suitable pos when we divide a bucket into 2.
3121  * We have to make sure the xattrs with the same hash value exist
3122  * in the same bucket.
3123  *
3124  * If this ocfs2_xattr_header covers more than one hash value, find a
3125  * place where the hash value changes.  Try to find the most even split.
3126  * The most common case is that all entries have different hash values,
3127  * and the first check we make will find a place to split.
3128  */
3129 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3130 {
3131         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3132         int count = le16_to_cpu(xh->xh_count);
3133         int delta, middle = count / 2;
3134
3135         /*
3136          * We start at the middle.  Each step gets farther away in both
3137          * directions.  We therefore hit the change in hash value
3138          * nearest to the middle.  Note that this loop does not execute for
3139          * count < 2.
3140          */
3141         for (delta = 0; delta < middle; delta++) {
3142                 /* Let's check delta earlier than middle */
3143                 if (cmp_xe(&entries[middle - delta - 1],
3144                            &entries[middle - delta]))
3145                         return middle - delta;
3146
3147                 /* For even counts, don't walk off the end */
3148                 if ((middle + delta + 1) == count)
3149                         continue;
3150
3151                 /* Now try delta past middle */
3152                 if (cmp_xe(&entries[middle + delta],
3153                            &entries[middle + delta + 1]))
3154                         return middle + delta + 1;
3155         }
3156
3157         /* Every entry had the same hash */
3158         return count;
3159 }
3160
3161 /*
3162  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3163  * first_hash will record the 1st hash of the new bucket.
3164  *
3165  * Normally half of the xattrs will be moved.  But we have to make
3166  * sure that the xattrs with the same hash value are stored in the
3167  * same bucket. If all the xattrs in this bucket have the same hash
3168  * value, the new bucket will be initialized as an empty one and the
3169  * first_hash will be initialized as (hash_value+1).
3170  */
3171 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3172                                     handle_t *handle,
3173                                     u64 blk,
3174                                     u64 new_blk,
3175                                     u32 *first_hash,
3176                                     int new_bucket_head)
3177 {
3178         int ret, i;
3179         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3180         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3181         struct buffer_head **s_bhs, **t_bhs = NULL;
3182         struct ocfs2_xattr_header *xh;
3183         struct ocfs2_xattr_entry *xe;
3184         int blocksize = inode->i_sb->s_blocksize;
3185
3186         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3187              (unsigned long long)blk, (unsigned long long)new_blk);
3188
3189         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3190         if (!s_bhs)
3191                 return -ENOMEM;
3192
3193         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3194         if (ret) {
3195                 mlog_errno(ret);
3196                 goto out;
3197         }
3198
3199         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3200                                    OCFS2_JOURNAL_ACCESS_WRITE);
3201         if (ret) {
3202                 mlog_errno(ret);
3203                 goto out;
3204         }
3205
3206         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3207         if (!t_bhs) {
3208                 ret = -ENOMEM;
3209                 goto out;
3210         }
3211
3212         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3213         if (ret) {
3214                 mlog_errno(ret);
3215                 goto out;
3216         }
3217
3218         for (i = 0; i < blk_per_bucket; i++) {
3219                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3220                                            new_bucket_head ?
3221                                            OCFS2_JOURNAL_ACCESS_CREATE :
3222                                            OCFS2_JOURNAL_ACCESS_WRITE);
3223                 if (ret) {
3224                         mlog_errno(ret);
3225                         goto out;
3226                 }
3227         }
3228
3229         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3230         count = le16_to_cpu(xh->xh_count);
3231         start = ocfs2_xattr_find_divide_pos(xh);
3232
3233         if (start == count) {
3234                 xe = &xh->xh_entries[start-1];
3235
3236                 /*
3237                  * initialized a new empty bucket here.
3238                  * The hash value is set as one larger than
3239                  * that of the last entry in the previous bucket.
3240                  */
3241                 for (i = 0; i < blk_per_bucket; i++)
3242                         memset(t_bhs[i]->b_data, 0, blocksize);
3243
3244                 xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3245                 xh->xh_free_start = cpu_to_le16(blocksize);
3246                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3247                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3248
3249                 goto set_num_buckets;
3250         }
3251
3252         /* copy the whole bucket to the new first. */
3253         for (i = 0; i < blk_per_bucket; i++)
3254                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3255
3256         /* update the new bucket. */
3257         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3258
3259         /*
3260          * Calculate the total name/value len and xh_free_start for
3261          * the old bucket first.
3262          */
3263         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3264         name_value_len = 0;
3265         for (i = 0; i < start; i++) {
3266                 xe = &xh->xh_entries[i];
3267                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3268                 if (ocfs2_xattr_is_local(xe))
3269                         xe_len +=
3270                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3271                 else
3272                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3273                 name_value_len += xe_len;
3274                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3275                         name_offset = le16_to_cpu(xe->xe_name_offset);
3276         }
3277
3278         /*
3279          * Now begin the modification to the new bucket.
3280          *
3281          * In the new bucket, We just move the xattr entry to the beginning
3282          * and don't touch the name/value. So there will be some holes in the
3283          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3284          * called.
3285          */
3286         xe = &xh->xh_entries[start];
3287         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3288         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3289              (int)((char *)xe - (char *)xh),
3290              (int)((char *)xh->xh_entries - (char *)xh));
3291         memmove((char *)xh->xh_entries, (char *)xe, len);
3292         xe = &xh->xh_entries[count - start];
3293         len = sizeof(struct ocfs2_xattr_entry) * start;
3294         memset((char *)xe, 0, len);
3295
3296         le16_add_cpu(&xh->xh_count, -start);
3297         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3298
3299         /* Calculate xh_free_start for the new bucket. */
3300         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3301         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3302                 xe = &xh->xh_entries[i];
3303                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3304                 if (ocfs2_xattr_is_local(xe))
3305                         xe_len +=
3306                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3307                 else
3308                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3309                 if (le16_to_cpu(xe->xe_name_offset) <
3310                     le16_to_cpu(xh->xh_free_start))
3311                         xh->xh_free_start = xe->xe_name_offset;
3312         }
3313
3314 set_num_buckets:
3315         /* set xh->xh_num_buckets for the new xh. */
3316         if (new_bucket_head)
3317                 xh->xh_num_buckets = cpu_to_le16(1);
3318         else
3319                 xh->xh_num_buckets = 0;
3320
3321         for (i = 0; i < blk_per_bucket; i++) {
3322                 ocfs2_journal_dirty(handle, t_bhs[i]);
3323                 if (ret)
3324                         mlog_errno(ret);
3325         }
3326
3327         /* store the first_hash of the new bucket. */
3328         if (first_hash)
3329                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3330
3331         /*
3332          * Now only update the 1st block of the old bucket.  If we
3333          * just added a new empty bucket, there is no need to modify
3334          * it.
3335          */
3336         if (start == count)
3337                 goto out;
3338
3339         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3340         memset(&xh->xh_entries[start], 0,
3341                sizeof(struct ocfs2_xattr_entry) * (count - start));
3342         xh->xh_count = cpu_to_le16(start);
3343         xh->xh_free_start = cpu_to_le16(name_offset);
3344         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3345
3346         ocfs2_journal_dirty(handle, s_bhs[0]);
3347         if (ret)
3348                 mlog_errno(ret);
3349
3350 out:
3351         if (s_bhs) {
3352                 for (i = 0; i < blk_per_bucket; i++)
3353                         brelse(s_bhs[i]);
3354         }
3355         kfree(s_bhs);
3356
3357         if (t_bhs) {
3358                 for (i = 0; i < blk_per_bucket; i++)
3359                         brelse(t_bhs[i]);
3360         }
3361         kfree(t_bhs);
3362
3363         return ret;
3364 }
3365
3366 /*
3367  * Copy xattr from one bucket to another bucket.
3368  *
3369  * The caller must make sure that the journal transaction
3370  * has enough space for journaling.
3371  */
3372 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3373                                  handle_t *handle,
3374                                  u64 s_blkno,
3375                                  u64 t_blkno,
3376                                  int t_is_new)
3377 {
3378         int ret, i;
3379         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3380         int blocksize = inode->i_sb->s_blocksize;
3381         struct buffer_head **s_bhs, **t_bhs = NULL;
3382
3383         BUG_ON(s_blkno == t_blkno);
3384
3385         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3386              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3387              t_is_new);
3388
3389         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3390                         GFP_NOFS);
3391         if (!s_bhs)
3392                 return -ENOMEM;
3393
3394         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3395         if (ret)
3396                 goto out;
3397
3398         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3399                         GFP_NOFS);
3400         if (!t_bhs) {
3401                 ret = -ENOMEM;
3402                 goto out;
3403         }
3404
3405         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3406         if (ret)
3407                 goto out;
3408
3409         for (i = 0; i < blk_per_bucket; i++) {
3410                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3411                                            t_is_new ?
3412                                            OCFS2_JOURNAL_ACCESS_CREATE :
3413                                            OCFS2_JOURNAL_ACCESS_WRITE);
3414                 if (ret)
3415                         goto out;
3416         }
3417
3418         for (i = 0; i < blk_per_bucket; i++) {
3419                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3420                 ocfs2_journal_dirty(handle, t_bhs[i]);
3421         }
3422
3423 out:
3424         if (s_bhs) {
3425                 for (i = 0; i < blk_per_bucket; i++)
3426                         brelse(s_bhs[i]);
3427         }
3428         kfree(s_bhs);
3429
3430         if (t_bhs) {
3431                 for (i = 0; i < blk_per_bucket; i++)
3432                         brelse(t_bhs[i]);
3433         }
3434         kfree(t_bhs);
3435
3436         return ret;
3437 }
3438
3439 /*
3440  * Copy one xattr cluster from src_blk to to_blk.
3441  * The to_blk will become the first bucket header of the cluster, so its
3442  * xh_num_buckets will be initialized as the bucket num in the cluster.
3443  */
3444 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3445                                   handle_t *handle,
3446                                   struct buffer_head *first_bh,
3447                                   u64 src_blk,
3448                                   u64 to_blk,
3449                                   u32 *first_hash)
3450 {
3451         int i, ret, credits;
3452         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3453         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3454         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3455         struct buffer_head *bh = NULL;
3456         struct ocfs2_xattr_header *xh;
3457         u64 to_blk_start = to_blk;
3458
3459         mlog(0, "cp xattrs from cluster %llu to %llu\n",
3460              (unsigned long long)src_blk, (unsigned long long)to_blk);
3461
3462         /*
3463          * We need to update the new cluster and 1 more for the update of
3464          * the 1st bucket of the previous extent rec.
3465          */
3466         credits = bpc + 1;
3467         ret = ocfs2_extend_trans(handle, credits);
3468         if (ret) {
3469                 mlog_errno(ret);
3470                 goto out;
3471         }
3472
3473         ret = ocfs2_journal_access(handle, inode, first_bh,
3474                                    OCFS2_JOURNAL_ACCESS_WRITE);
3475         if (ret) {
3476                 mlog_errno(ret);
3477                 goto out;
3478         }
3479
3480         for (i = 0; i < num_buckets; i++) {
3481                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3482                                             src_blk, to_blk, 1);
3483                 if (ret) {
3484                         mlog_errno(ret);
3485                         goto out;
3486                 }
3487
3488                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3489                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3490         }
3491
3492         /* update the old bucket header. */
3493         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3494         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3495
3496         ocfs2_journal_dirty(handle, first_bh);
3497
3498         /* update the new bucket header. */
3499         ret = ocfs2_read_block(inode, to_blk_start, &bh);
3500         if (ret < 0) {
3501                 mlog_errno(ret);
3502                 goto out;
3503         }
3504
3505         ret = ocfs2_journal_access(handle, inode, bh,
3506                                    OCFS2_JOURNAL_ACCESS_WRITE);
3507         if (ret) {
3508                 mlog_errno(ret);
3509                 goto out;
3510         }
3511
3512         xh = (struct ocfs2_xattr_header *)bh->b_data;
3513         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3514
3515         ocfs2_journal_dirty(handle, bh);
3516
3517         if (first_hash)
3518                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3519 out:
3520         brelse(bh);
3521         return ret;
3522 }
3523
3524 /*
3525  * Move some xattrs in this cluster to the new cluster.
3526  * This function should only be called when bucket size == cluster size.
3527  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3528  */
3529 static int ocfs2_divide_xattr_cluster(struct inode *inode,
3530                                       handle_t *handle,
3531                                       u64 prev_blk,
3532                                       u64 new_blk,
3533                                       u32 *first_hash)
3534 {
3535         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3536         int ret, credits = 2 * blk_per_bucket;
3537
3538         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3539
3540         ret = ocfs2_extend_trans(handle, credits);
3541         if (ret) {
3542                 mlog_errno(ret);
3543                 return ret;
3544         }
3545
3546         /* Move half of the xattr in start_blk to the next bucket. */
3547         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
3548                                           new_blk, first_hash, 1);
3549 }
3550
3551 /*
3552  * Move some xattrs from the old cluster to the new one since they are not
3553  * contiguous in ocfs2 xattr tree.
3554  *
3555  * new_blk starts a new separate cluster, and we will move some xattrs from
3556  * prev_blk to it. v_start will be set as the first name hash value in this
3557  * new cluster so that it can be used as e_cpos during tree insertion and
3558  * don't collide with our original b-tree operations. first_bh and header_bh
3559  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3560  * to extend the insert bucket.
3561  *
3562  * The problem is how much xattr should we move to the new one and when should
3563  * we update first_bh and header_bh?
3564  * 1. If cluster size > bucket size, that means the previous cluster has more
3565  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3566  *    update the first_bh and header_bh if the insert bucket has been moved
3567  *    to the new cluster.
3568  * 2. If cluster_size == bucket_size:
3569  *    a) If the previous extent rec has more than one cluster and the insert
3570  *       place isn't in the last cluster, copy the entire last cluster to the
3571  *       new one. This time, we don't need to upate the first_bh and header_bh
3572  *       since they will not be moved into the new cluster.
3573  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3574  *       the new one. And we set the extend flag to zero if the insert place is
3575  *       moved into the new allocated cluster since no extend is needed.
3576  */
3577 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3578                                             handle_t *handle,
3579                                             struct buffer_head **first_bh,
3580                                             struct buffer_head **header_bh,
3581                                             u64 new_blk,
3582                                             u64 prev_blk,
3583                                             u32 prev_clusters,
3584                                             u32 *v_start,
3585                                             int *extend)
3586 {
3587         int ret = 0;
3588         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3589
3590         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3591              (unsigned long long)prev_blk, prev_clusters,
3592              (unsigned long long)new_blk);
3593
3594         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3595                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3596                                                           handle,
3597                                                           first_bh,
3598                                                           header_bh,
3599                                                           new_blk,
3600                                                           prev_blk,
3601                                                           prev_clusters,
3602                                                           v_start);
3603         else {
3604                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3605
3606                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3607                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3608                                                      last_blk, new_blk,
3609                                                      v_start);
3610                 else {
3611                         ret = ocfs2_divide_xattr_cluster(inode, handle,
3612                                                          last_blk, new_blk,
3613                                                          v_start);
3614
3615                         if ((*header_bh)->b_blocknr == last_blk && extend)
3616                                 *extend = 0;
3617                 }
3618         }
3619
3620         return ret;
3621 }
3622
3623 /*
3624  * Add a new cluster for xattr storage.
3625  *
3626  * If the new cluster is contiguous with the previous one, it will be
3627  * appended to the same extent record, and num_clusters will be updated.
3628  * If not, we will insert a new extent for it and move some xattrs in
3629  * the last cluster into the new allocated one.
3630  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3631  * lose the benefits of hashing because we'll have to search large leaves.
3632  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3633  * if it's bigger).
3634  *
3635  * first_bh is the first block of the previous extent rec and header_bh
3636  * indicates the bucket we will insert the new xattrs. They will be updated
3637  * when the header_bh is moved into the new cluster.
3638  */
3639 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3640                                        struct buffer_head *root_bh,
3641                                        struct buffer_head **first_bh,
3642                                        struct buffer_head **header_bh,
3643                                        u32 *num_clusters,
3644                                        u32 prev_cpos,
3645                                        u64 prev_blkno,
3646                                        int *extend)
3647 {
3648         int ret, credits;
3649         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3650         u32 prev_clusters = *num_clusters;
3651         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3652         u64 block;
3653         handle_t *handle = NULL;
3654         struct ocfs2_alloc_context *data_ac = NULL;
3655         struct ocfs2_alloc_context *meta_ac = NULL;
3656         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3657         struct ocfs2_extent_tree et;
3658
3659         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3660              "previous xattr blkno = %llu\n",
3661              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3662              prev_cpos, (unsigned long long)prev_blkno);
3663
3664         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3665
3666         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3667                                     &data_ac, &meta_ac);
3668         if (ret) {
3669                 mlog_errno(ret);
3670                 goto leave;
3671         }
3672
3673         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3674                                             clusters_to_add);
3675         handle = ocfs2_start_trans(osb, credits);
3676         if (IS_ERR(handle)) {
3677                 ret = PTR_ERR(handle);
3678                 handle = NULL;
3679                 mlog_errno(ret);
3680                 goto leave;
3681         }
3682
3683         ret = ocfs2_journal_access(handle, inode, root_bh,
3684                                    OCFS2_JOURNAL_ACCESS_WRITE);
3685         if (ret < 0) {
3686                 mlog_errno(ret);
3687                 goto leave;
3688         }
3689
3690         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3691                                      clusters_to_add, &bit_off, &num_bits);
3692         if (ret < 0) {
3693                 if (ret != -ENOSPC)
3694                         mlog_errno(ret);
3695                 goto leave;
3696         }
3697
3698         BUG_ON(num_bits > clusters_to_add);
3699
3700         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3701         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3702              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3703
3704         if (prev_blkno + prev_clusters * bpc == block &&
3705             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3706              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3707                 /*
3708                  * If this cluster is contiguous with the old one and
3709                  * adding this new cluster, we don't surpass the limit of
3710                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3711                  * initialized and used like other buckets in the previous
3712                  * cluster.
3713                  * So add it as a contiguous one. The caller will handle
3714                  * its init process.
3715                  */
3716                 v_start = prev_cpos + prev_clusters;
3717                 *num_clusters = prev_clusters + num_bits;
3718                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3719                      num_bits);
3720         } else {
3721                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3722                                                        handle,
3723                                                        first_bh,
3724                                                        header_bh,
3725                                                        block,
3726                                                        prev_blkno,
3727                                                        prev_clusters,
3728                                                        &v_start,
3729                                                        extend);
3730                 if (ret) {
3731                         mlog_errno(ret);
3732                         goto leave;
3733                 }
3734         }
3735
3736         if (handle->h_buffer_credits < credits) {
3737                 /*
3738                  * The journal has been restarted before, and don't
3739                  * have enough space for the insertion, so extend it
3740                  * here.
3741                  */
3742                 ret = ocfs2_extend_trans(handle, credits);
3743                 if (ret) {
3744                         mlog_errno(ret);
3745                         goto leave;
3746                 }
3747         }
3748         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3749              num_bits, (unsigned long long)block, v_start);
3750         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3751                                   num_bits, 0, meta_ac);
3752         if (ret < 0) {
3753                 mlog_errno(ret);
3754                 goto leave;
3755         }
3756
3757         ret = ocfs2_journal_dirty(handle, root_bh);
3758         if (ret < 0) {
3759                 mlog_errno(ret);
3760                 goto leave;
3761         }
3762
3763 leave:
3764         if (handle)
3765                 ocfs2_commit_trans(osb, handle);
3766         if (data_ac)
3767                 ocfs2_free_alloc_context(data_ac);
3768         if (meta_ac)
3769                 ocfs2_free_alloc_context(meta_ac);
3770
3771         return ret;
3772 }
3773
3774 /*
3775  * Extend a new xattr bucket and move xattrs to the end one by one until
3776  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3777  */
3778 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3779                                      struct buffer_head *first_bh,
3780                                      struct buffer_head *start_bh,
3781                                      u32 num_clusters)
3782 {
3783         int ret, credits;
3784         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3785         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3786         u64 start_blk = start_bh->b_blocknr, end_blk;
3787         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3788         handle_t *handle;
3789         struct ocfs2_xattr_header *first_xh =
3790                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3791         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3792
3793         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3794              "from %llu, len = %u\n", (unsigned long long)start_blk,
3795              (unsigned long long)first_bh->b_blocknr, num_clusters);
3796
3797         BUG_ON(bucket >= num_buckets);
3798
3799         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3800
3801         /*
3802          * We will touch all the buckets after the start_bh(include it).
3803          * Add one more bucket and modify the first_bh.
3804          */
3805         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3806         handle = ocfs2_start_trans(osb, credits);
3807         if (IS_ERR(handle)) {
3808                 ret = PTR_ERR(handle);
3809                 handle = NULL;
3810                 mlog_errno(ret);
3811                 goto out;
3812         }
3813
3814         ret = ocfs2_journal_access(handle, inode, first_bh,
3815                                    OCFS2_JOURNAL_ACCESS_WRITE);
3816         if (ret) {
3817                 mlog_errno(ret);
3818                 goto commit;
3819         }
3820
3821         while (end_blk != start_blk) {
3822                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3823                                             end_blk + blk_per_bucket, 0);
3824                 if (ret)
3825                         goto commit;
3826                 end_blk -= blk_per_bucket;
3827         }
3828
3829         /* Move half of the xattr in start_blk to the next bucket. */
3830         ret = ocfs2_divide_xattr_bucket(inode, handle, start_blk,
3831                                         start_blk + blk_per_bucket, NULL, 0);
3832
3833         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3834         ocfs2_journal_dirty(handle, first_bh);
3835
3836 commit:
3837         ocfs2_commit_trans(osb, handle);
3838 out:
3839         return ret;
3840 }
3841
3842 /*
3843  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3844  * xb_bh is the ocfs2_xattr_block.
3845  * We will move all the buckets starting from header_bh to the next place. As
3846  * for this one, half num of its xattrs will be moved to the next one.
3847  *
3848  * We will allocate a new cluster if current cluster is full and adjust
3849  * header_bh and first_bh if the insert place is moved to the new cluster.
3850  */
3851 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3852                                       struct buffer_head *xb_bh,
3853                                       struct buffer_head *header_bh)
3854 {
3855         struct ocfs2_xattr_header *first_xh = NULL;
3856         struct buffer_head *first_bh = NULL;
3857         struct ocfs2_xattr_block *xb =
3858                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3859         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3860         struct ocfs2_extent_list *el = &xb_root->xt_list;
3861         struct ocfs2_xattr_header *xh =
3862                         (struct ocfs2_xattr_header *)header_bh->b_data;
3863         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3864         struct super_block *sb = inode->i_sb;
3865         struct ocfs2_super *osb = OCFS2_SB(sb);
3866         int ret, num_buckets, extend = 1;
3867         u64 p_blkno;
3868         u32 e_cpos, num_clusters;
3869
3870         mlog(0, "Add new xattr bucket starting form %llu\n",
3871              (unsigned long long)header_bh->b_blocknr);
3872
3873         /*
3874          * Add refrence for header_bh here because it may be
3875          * changed in ocfs2_add_new_xattr_cluster and we need
3876          * to free it in the end.
3877          */
3878         get_bh(header_bh);
3879
3880         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3881                                   &num_clusters, el);
3882         if (ret) {
3883                 mlog_errno(ret);
3884                 goto out;
3885         }
3886
3887         ret = ocfs2_read_block(inode, p_blkno, &first_bh);
3888         if (ret) {
3889                 mlog_errno(ret);
3890                 goto out;
3891         }
3892
3893         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3894         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3895
3896         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3897                 ret = ocfs2_add_new_xattr_cluster(inode,
3898                                                   xb_bh,
3899                                                   &first_bh,
3900                                                   &header_bh,
3901                                                   &num_clusters,
3902                                                   e_cpos,
3903                                                   p_blkno,
3904                                                   &extend);
3905                 if (ret) {
3906                         mlog_errno(ret);
3907                         goto out;
3908                 }
3909         }
3910
3911         if (extend)
3912                 ret = ocfs2_extend_xattr_bucket(inode,
3913                                                 first_bh,
3914                                                 header_bh,
3915                                                 num_clusters);
3916         if (ret)
3917                 mlog_errno(ret);
3918 out:
3919         brelse(first_bh);
3920         brelse(header_bh);
3921         return ret;
3922 }
3923
3924 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3925                                         struct ocfs2_xattr_bucket *bucket,
3926                                         int offs)
3927 {
3928         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3929
3930         offs = offs % inode->i_sb->s_blocksize;
3931         return bucket_block(bucket, block_off) + offs;
3932 }
3933
3934 /*
3935  * Handle the normal xattr set, including replace, delete and new.
3936  *
3937  * Note: "local" indicates the real data's locality. So we can't
3938  * just its bucket locality by its length.
3939  */
3940 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3941                                          struct ocfs2_xattr_info *xi,
3942                                          struct ocfs2_xattr_search *xs,
3943                                          u32 name_hash,
3944                                          int local)
3945 {
3946         struct ocfs2_xattr_entry *last, *xe;
3947         int name_len = strlen(xi->name);
3948         struct ocfs2_xattr_header *xh = xs->header;
3949         u16 count = le16_to_cpu(xh->xh_count), start;
3950         size_t blocksize = inode->i_sb->s_blocksize;
3951         char *val;
3952         size_t offs, size, new_size;
3953
3954         last = &xh->xh_entries[count];
3955         if (!xs->not_found) {
3956                 xe = xs->here;
3957                 offs = le16_to_cpu(xe->xe_name_offset);
3958                 if (ocfs2_xattr_is_local(xe))
3959                         size = OCFS2_XATTR_SIZE(name_len) +
3960                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3961                 else
3962                         size = OCFS2_XATTR_SIZE(name_len) +
3963                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3964
3965                 /*
3966                  * If the new value will be stored outside, xi->value has been
3967                  * initalized as an empty ocfs2_xattr_value_root, and the same
3968                  * goes with xi->value_len, so we can set new_size safely here.
3969                  * See ocfs2_xattr_set_in_bucket.
3970                  */
3971                 new_size = OCFS2_XATTR_SIZE(name_len) +
3972                            OCFS2_XATTR_SIZE(xi->value_len);
3973
3974                 le16_add_cpu(&xh->xh_name_value_len, -size);
3975                 if (xi->value) {
3976                         if (new_size > size)
3977                                 goto set_new_name_value;
3978
3979                         /* Now replace the old value with new one. */
3980                         if (local)
3981                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3982                         else
3983                                 xe->xe_value_size = 0;
3984
3985                         val = ocfs2_xattr_bucket_get_val(inode,
3986                                                          &xs->bucket, offs);
3987                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3988                                size - OCFS2_XATTR_SIZE(name_len));
3989                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3990                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3991                                        xi->value, xi->value_len);
3992
3993                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3994                         ocfs2_xattr_set_local(xe, local);
3995                         return;
3996                 } else {
3997                         /*
3998                          * Remove the old entry if there is more than one.
3999                          * We don't remove the last entry so that we can
4000                          * use it to indicate the hash value of the empty
4001                          * bucket.
4002                          */
4003                         last -= 1;
4004                         le16_add_cpu(&xh->xh_count, -1);
4005                         if (xh->xh_count) {
4006                                 memmove(xe, xe + 1,
4007                                         (void *)last - (void *)xe);
4008                                 memset(last, 0,
4009                                        sizeof(struct ocfs2_xattr_entry));
4010                         } else
4011                                 xh->xh_free_start =
4012                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4013
4014                         return;
4015                 }
4016         } else {
4017                 /* find a new entry for insert. */
4018                 int low = 0, high = count - 1, tmp;
4019                 struct ocfs2_xattr_entry *tmp_xe;
4020
4021                 while (low <= high && count) {
4022                         tmp = (low + high) / 2;
4023                         tmp_xe = &xh->xh_entries[tmp];
4024
4025                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4026                                 low = tmp + 1;
4027                         else if (name_hash <
4028                                  le32_to_cpu(tmp_xe->xe_name_hash))
4029                                 high = tmp - 1;
4030                         else {
4031                                 low = tmp;
4032                                 break;
4033                         }
4034                 }
4035
4036                 xe = &xh->xh_entries[low];
4037                 if (low != count)
4038                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4039
4040                 le16_add_cpu(&xh->xh_count, 1);
4041                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4042                 xe->xe_name_hash = cpu_to_le32(name_hash);
4043                 xe->xe_name_len = name_len;
4044                 ocfs2_xattr_set_type(xe, xi->name_index);
4045         }
4046
4047 set_new_name_value:
4048         /* Insert the new name+value. */
4049         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4050
4051         /*
4052          * We must make sure that the name/value pair
4053          * exists in the same block.
4054          */
4055         offs = le16_to_cpu(xh->xh_free_start);
4056         start = offs - size;
4057
4058         if (start >> inode->i_sb->s_blocksize_bits !=
4059             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4060                 offs = offs - offs % blocksize;
4061                 xh->xh_free_start = cpu_to_le16(offs);
4062         }
4063
4064         val = ocfs2_xattr_bucket_get_val(inode,
4065                                          &xs->bucket, offs - size);
4066         xe->xe_name_offset = cpu_to_le16(offs - size);
4067
4068         memset(val, 0, size);
4069         memcpy(val, xi->name, name_len);
4070         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4071
4072         xe->xe_value_size = cpu_to_le64(xi->value_len);
4073         ocfs2_xattr_set_local(xe, local);
4074         xs->here = xe;
4075         le16_add_cpu(&xh->xh_free_start, -size);
4076         le16_add_cpu(&xh->xh_name_value_len, size);
4077
4078         return;
4079 }
4080
4081 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4082                                              handle_t *handle,
4083                                              struct ocfs2_xattr_search *xs,
4084                                              struct buffer_head **bhs,
4085                                              u16 bh_num)
4086 {
4087         int ret = 0, off, block_off;
4088         struct ocfs2_xattr_entry *xe = xs->here;
4089
4090         /*
4091          * First calculate all the blocks we should journal_access
4092          * and journal_dirty. The first block should always be touched.
4093          */
4094         ret = ocfs2_journal_dirty(handle, bhs[0]);
4095         if (ret)
4096                 mlog_errno(ret);
4097
4098         /* calc the data. */
4099         off = le16_to_cpu(xe->xe_name_offset);
4100         block_off = off >> inode->i_sb->s_blocksize_bits;
4101         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4102         if (ret)
4103                 mlog_errno(ret);
4104
4105         return ret;
4106 }
4107
4108 /*
4109  * Set the xattr entry in the specified bucket.
4110  * The bucket is indicated by xs->bucket and it should have the enough
4111  * space for the xattr insertion.
4112  */
4113 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4114                                            struct ocfs2_xattr_info *xi,
4115                                            struct ocfs2_xattr_search *xs,
4116                                            u32 name_hash,
4117                                            int local)
4118 {
4119         int i, ret;
4120         handle_t *handle = NULL;
4121         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4122         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4123
4124         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4125              (unsigned long)xi->value_len, xi->name_index,
4126              (unsigned long long)bucket_blkno(&xs->bucket));
4127
4128         if (!xs->bucket.bu_bhs[1]) {
4129                 ret = ocfs2_read_blocks(inode,
4130                                         bucket_blkno(&xs->bucket) + 1,
4131                                         blk_per_bucket - 1, &xs->bucket.bu_bhs[1],
4132                                         0);
4133                 if (ret) {
4134                         mlog_errno(ret);
4135                         goto out;
4136                 }
4137         }
4138
4139         handle = ocfs2_start_trans(osb, blk_per_bucket);
4140         if (IS_ERR(handle)) {
4141                 ret = PTR_ERR(handle);
4142                 handle = NULL;
4143                 mlog_errno(ret);
4144                 goto out;
4145         }
4146
4147         for (i = 0; i < blk_per_bucket; i++) {
4148                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bu_bhs[i],
4149                                            OCFS2_JOURNAL_ACCESS_WRITE);
4150                 if (ret < 0) {
4151                         mlog_errno(ret);
4152                         goto out;
4153                 }
4154         }
4155
4156         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4157
4158         /*Only dirty the blocks we have touched in set xattr. */
4159         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4160                                                 xs->bucket.bu_bhs, blk_per_bucket);
4161         if (ret)
4162                 mlog_errno(ret);
4163 out:
4164         ocfs2_commit_trans(osb, handle);
4165
4166         return ret;
4167 }
4168
4169 static int ocfs2_xattr_value_update_size(struct inode *inode,
4170                                          struct buffer_head *xe_bh,
4171                                          struct ocfs2_xattr_entry *xe,
4172                                          u64 new_size)
4173 {
4174         int ret;
4175         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4176         handle_t *handle = NULL;
4177
4178         handle = ocfs2_start_trans(osb, 1);
4179         if (IS_ERR(handle)) {
4180                 ret = -ENOMEM;
4181                 mlog_errno(ret);
4182                 goto out;
4183         }
4184
4185         ret = ocfs2_journal_access(handle, inode, xe_bh,
4186                                    OCFS2_JOURNAL_ACCESS_WRITE);
4187         if (ret < 0) {
4188                 mlog_errno(ret);
4189                 goto out_commit;
4190         }
4191
4192         xe->xe_value_size = cpu_to_le64(new_size);
4193
4194         ret = ocfs2_journal_dirty(handle, xe_bh);
4195         if (ret < 0)
4196                 mlog_errno(ret);
4197
4198 out_commit:
4199         ocfs2_commit_trans(osb, handle);
4200 out:
4201         return ret;
4202 }
4203
4204 /*
4205  * Truncate the specified xe_off entry in xattr bucket.
4206  * bucket is indicated by header_bh and len is the new length.
4207  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4208  *
4209  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4210  */
4211 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4212                                              struct buffer_head *header_bh,
4213                                              int xe_off,
4214                                              int len)
4215 {
4216         int ret, offset;
4217         u64 value_blk;
4218         struct buffer_head *value_bh = NULL;
4219         struct ocfs2_xattr_value_root *xv;
4220         struct ocfs2_xattr_entry *xe;
4221         struct ocfs2_xattr_header *xh =
4222                         (struct ocfs2_xattr_header *)header_bh->b_data;
4223         size_t blocksize = inode->i_sb->s_blocksize;
4224
4225         xe = &xh->xh_entries[xe_off];
4226
4227         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4228
4229         offset = le16_to_cpu(xe->xe_name_offset) +
4230                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4231
4232         value_blk = offset / blocksize;
4233
4234         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4235         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4236         value_blk += header_bh->b_blocknr;
4237
4238         ret = ocfs2_read_block(inode, value_blk, &value_bh);
4239         if (ret) {
4240                 mlog_errno(ret);
4241                 goto out;
4242         }
4243
4244         xv = (struct ocfs2_xattr_value_root *)
4245                 (value_bh->b_data + offset % blocksize);
4246
4247         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4248              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4249         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4250         if (ret) {
4251                 mlog_errno(ret);
4252                 goto out;
4253         }
4254
4255         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4256         if (ret) {
4257                 mlog_errno(ret);
4258                 goto out;
4259         }
4260
4261 out:
4262         brelse(value_bh);
4263         return ret;
4264 }
4265
4266 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4267                                                 struct ocfs2_xattr_search *xs,
4268                                                 int len)
4269 {
4270         int ret, offset;
4271         struct ocfs2_xattr_entry *xe = xs->here;
4272         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4273
4274         BUG_ON(!xs->bucket.bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4275
4276         offset = xe - xh->xh_entries;
4277         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bu_bhs[0],
4278                                                 offset, len);
4279         if (ret)
4280                 mlog_errno(ret);
4281
4282         return ret;
4283 }
4284
4285 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4286                                                 struct ocfs2_xattr_search *xs,
4287                                                 char *val,
4288                                                 int value_len)
4289 {
4290         int offset;
4291         struct ocfs2_xattr_value_root *xv;
4292         struct ocfs2_xattr_entry *xe = xs->here;
4293
4294         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4295
4296         offset = le16_to_cpu(xe->xe_name_offset) +
4297                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4298
4299         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4300
4301         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4302 }
4303
4304 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4305                                   struct buffer_head *root_bh,
4306                                   u64 blkno,
4307                                   u32 cpos,
4308                                   u32 len)
4309 {
4310         int ret;
4311         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4312         struct inode *tl_inode = osb->osb_tl_inode;
4313         handle_t *handle;
4314         struct ocfs2_xattr_block *xb =
4315                         (struct ocfs2_xattr_block *)root_bh->b_data;
4316         struct ocfs2_alloc_context *meta_ac = NULL;
4317         struct ocfs2_cached_dealloc_ctxt dealloc;
4318         struct ocfs2_extent_tree et;
4319
4320         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4321
4322         ocfs2_init_dealloc_ctxt(&dealloc);
4323
4324         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4325              cpos, len, (unsigned long long)blkno);
4326
4327         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4328
4329         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4330         if (ret) {
4331                 mlog_errno(ret);
4332                 return ret;
4333         }
4334
4335         mutex_lock(&tl_inode->i_mutex);
4336
4337         if (ocfs2_truncate_log_needs_flush(osb)) {
4338                 ret = __ocfs2_flush_truncate_log(osb);
4339                 if (ret < 0) {
4340                         mlog_errno(ret);
4341                         goto out;
4342                 }
4343         }
4344
4345         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4346         if (IS_ERR(handle)) {
4347                 ret = -ENOMEM;
4348                 mlog_errno(ret);
4349                 goto out;
4350         }
4351
4352         ret = ocfs2_journal_access(handle, inode, root_bh,
4353                                    OCFS2_JOURNAL_ACCESS_WRITE);
4354         if (ret) {
4355                 mlog_errno(ret);
4356                 goto out_commit;
4357         }
4358
4359         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4360                                   &dealloc);
4361         if (ret) {
4362                 mlog_errno(ret);
4363                 goto out_commit;
4364         }
4365
4366         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4367
4368         ret = ocfs2_journal_dirty(handle, root_bh);
4369         if (ret) {
4370                 mlog_errno(ret);
4371                 goto out_commit;
4372         }
4373
4374         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4375         if (ret)
4376                 mlog_errno(ret);
4377
4378 out_commit:
4379         ocfs2_commit_trans(osb, handle);
4380 out:
4381         ocfs2_schedule_truncate_log_flush(osb, 1);
4382
4383         mutex_unlock(&tl_inode->i_mutex);
4384
4385         if (meta_ac)
4386                 ocfs2_free_alloc_context(meta_ac);
4387
4388         ocfs2_run_deallocs(osb, &dealloc);
4389
4390         return ret;
4391 }
4392
4393 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4394                                          struct ocfs2_xattr_search *xs)
4395 {
4396         handle_t *handle = NULL;
4397         struct ocfs2_xattr_header *xh = bucket_xh(&xs->bucket);
4398         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4399                                                 le16_to_cpu(xh->xh_count) - 1];
4400         int ret = 0;
4401
4402         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4403         if (IS_ERR(handle)) {
4404                 ret = PTR_ERR(handle);
4405                 mlog_errno(ret);
4406                 return;
4407         }
4408
4409         ret = ocfs2_journal_access(handle, inode, xs->bucket.bu_bhs[0],
4410                                    OCFS2_JOURNAL_ACCESS_WRITE);
4411         if (ret) {
4412                 mlog_errno(ret);
4413                 goto out_commit;
4414         }
4415
4416         /* Remove the old entry. */
4417         memmove(xs->here, xs->here + 1,
4418                 (void *)last - (void *)xs->here);
4419         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4420         le16_add_cpu(&xh->xh_count, -1);
4421
4422         ret = ocfs2_journal_dirty(handle, xs->bucket.bu_bhs[0]);
4423         if (ret < 0)
4424                 mlog_errno(ret);
4425 out_commit:
4426         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4427 }
4428
4429 /*
4430  * Set the xattr name/value in the bucket specified in xs.
4431  *
4432  * As the new value in xi may be stored in the bucket or in an outside cluster,
4433  * we divide the whole process into 3 steps:
4434  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4435  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4436  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4437  * 4. If the clusters for the new outside value can't be allocated, we need
4438  *    to free the xattr we allocated in set.
4439  */
4440 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4441                                      struct ocfs2_xattr_info *xi,
4442                                      struct ocfs2_xattr_search *xs)
4443 {
4444         int ret, local = 1;
4445         size_t value_len;
4446         char *val = (char *)xi->value;
4447         struct ocfs2_xattr_entry *xe = xs->here;
4448         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4449                                               strlen(xi->name));
4450
4451         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4452                 /*
4453                  * We need to truncate the xattr storage first.
4454                  *
4455                  * If both the old and new value are stored to
4456                  * outside block, we only need to truncate
4457                  * the storage and then set the value outside.
4458                  *
4459                  * If the new value should be stored within block,
4460                  * we should free all the outside block first and
4461                  * the modification to the xattr block will be done
4462                  * by following steps.
4463                  */
4464                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4465                         value_len = xi->value_len;
4466                 else
4467                         value_len = 0;
4468
4469                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4470                                                            value_len);
4471                 if (ret)
4472                         goto out;
4473
4474                 if (value_len)
4475                         goto set_value_outside;
4476         }
4477
4478         value_len = xi->value_len;
4479         /* So we have to handle the inside block change now. */
4480         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4481                 /*
4482                  * If the new value will be stored outside of block,
4483                  * initalize a new empty value root and insert it first.
4484                  */
4485                 local = 0;
4486                 xi->value = &def_xv;
4487                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4488         }
4489
4490         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4491         if (ret) {
4492                 mlog_errno(ret);
4493                 goto out;
4494         }
4495
4496         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4497                 goto out;
4498
4499         /* allocate the space now for the outside block storage. */
4500         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4501                                                    value_len);
4502         if (ret) {
4503                 mlog_errno(ret);
4504
4505                 if (xs->not_found) {
4506                         /*
4507                          * We can't allocate enough clusters for outside
4508                          * storage and we have allocated xattr already,
4509                          * so need to remove it.
4510                          */
4511                         ocfs2_xattr_bucket_remove_xs(inode, xs);
4512                 }
4513                 goto out;
4514         }
4515
4516 set_value_outside:
4517         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4518 out:
4519         return ret;
4520 }
4521
4522 /*
4523  * check whether the xattr bucket is filled up with the same hash value.
4524  * If we want to insert the xattr with the same hash, return -ENOSPC.
4525  * If we want to insert a xattr with different hash value, go ahead
4526  * and ocfs2_divide_xattr_bucket will handle this.
4527  */
4528 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4529                                               struct ocfs2_xattr_bucket *bucket,
4530                                               const char *name)
4531 {
4532         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4533         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4534
4535         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4536                 return 0;
4537
4538         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4539             xh->xh_entries[0].xe_name_hash) {
4540                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4541                      "hash = %u\n",
4542                      (unsigned long long)bucket_blkno(bucket),
4543                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4544                 return -ENOSPC;
4545         }
4546
4547         return 0;
4548 }
4549
4550 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4551                                              struct ocfs2_xattr_info *xi,
4552                                              struct ocfs2_xattr_search *xs)
4553 {
4554         struct ocfs2_xattr_header *xh;
4555         struct ocfs2_xattr_entry *xe;
4556         u16 count, header_size, xh_free_start;
4557         int i, free, max_free, need, old;
4558         size_t value_size = 0, name_len = strlen(xi->name);
4559         size_t blocksize = inode->i_sb->s_blocksize;
4560         int ret, allocation = 0;
4561         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4562
4563         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4564
4565 try_again:
4566         xh = xs->header;
4567         count = le16_to_cpu(xh->xh_count);
4568         xh_free_start = le16_to_cpu(xh->xh_free_start);
4569         header_size = sizeof(struct ocfs2_xattr_header) +
4570                         count * sizeof(struct ocfs2_xattr_entry);
4571         max_free = OCFS2_XATTR_BUCKET_SIZE -
4572                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4573
4574         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4575                         "of %u which exceed block size\n",
4576                         (unsigned long long)bucket_blkno(&xs->bucket),
4577                         header_size);
4578
4579         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4580                 value_size = OCFS2_XATTR_ROOT_SIZE;
4581         else if (xi->value)
4582                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4583
4584         if (xs->not_found)
4585                 need = sizeof(struct ocfs2_xattr_entry) +
4586                         OCFS2_XATTR_SIZE(name_len) + value_size;
4587         else {
4588                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4589
4590                 /*
4591                  * We only replace the old value if the new length is smaller
4592                  * than the old one. Otherwise we will allocate new space in the
4593                  * bucket to store it.
4594                  */
4595                 xe = xs->here;
4596                 if (ocfs2_xattr_is_local(xe))
4597                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4598                 else
4599                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4600
4601                 if (old >= value_size)
4602                         need = 0;
4603         }
4604
4605         free = xh_free_start - header_size;
4606         /*
4607          * We need to make sure the new name/value pair
4608          * can exist in the same block.
4609          */
4610         if (xh_free_start % blocksize < need)
4611                 free -= xh_free_start % blocksize;
4612
4613         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4614              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4615              " %u\n", xs->not_found,
4616              (unsigned long long)bucket_blkno(&xs->bucket),
4617              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4618              le16_to_cpu(xh->xh_name_value_len));
4619
4620         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4621                 if (need <= max_free &&
4622                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4623                         /*
4624                          * We can create the space by defragment. Since only the
4625                          * name/value will be moved, the xe shouldn't be changed
4626                          * in xs.
4627                          */
4628                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4629                         if (ret) {
4630                                 mlog_errno(ret);
4631                                 goto out;
4632                         }
4633
4634                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4635                         free = xh_free_start - header_size;
4636                         if (xh_free_start % blocksize < need)
4637                                 free -= xh_free_start % blocksize;
4638
4639                         if (free >= need)
4640                                 goto xattr_set;
4641
4642                         mlog(0, "Can't get enough space for xattr insert by "
4643                              "defragment. Need %u bytes, but we have %d, so "
4644                              "allocate new bucket for it.\n", need, free);
4645                 }
4646
4647                 /*
4648                  * We have to add new buckets or clusters and one
4649                  * allocation should leave us enough space for insert.
4650                  */
4651                 BUG_ON(allocation);
4652
4653                 /*
4654                  * We do not allow for overlapping ranges between buckets. And
4655                  * the maximum number of collisions we will allow for then is
4656                  * one bucket's worth, so check it here whether we need to
4657                  * add a new bucket for the insert.
4658                  */
4659                 ret = ocfs2_check_xattr_bucket_collision(inode,
4660                                                          &xs->bucket,
4661                                                          xi->name);
4662                 if (ret) {
4663                         mlog_errno(ret);
4664                         goto out;
4665                 }
4666
4667                 ret = ocfs2_add_new_xattr_bucket(inode,
4668                                                  xs->xattr_bh,
4669                                                  xs->bucket.bu_bhs[0]);
4670                 if (ret) {
4671                         mlog_errno(ret);
4672                         goto out;
4673                 }
4674
4675                 for (i = 0; i < blk_per_bucket; i++)
4676                         brelse(xs->bucket.bu_bhs[i]);
4677
4678                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4679
4680                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4681                                                    xi->name_index,
4682                                                    xi->name, xs);
4683                 if (ret && ret != -ENODATA)
4684                         goto out;
4685                 xs->not_found = ret;
4686                 allocation = 1;
4687                 goto try_again;
4688         }
4689
4690 xattr_set:
4691         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4692 out:
4693         mlog_exit(ret);
4694         return ret;
4695 }
4696
4697 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4698                                         struct ocfs2_xattr_bucket *bucket,
4699                                         void *para)
4700 {
4701         int ret = 0;
4702         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4703         u16 i;
4704         struct ocfs2_xattr_entry *xe;
4705
4706         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4707                 xe = &xh->xh_entries[i];
4708                 if (ocfs2_xattr_is_local(xe))
4709                         continue;
4710
4711                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4712                                                         bucket->bu_bhs[0],
4713                                                         i, 0);
4714                 if (ret) {
4715                         mlog_errno(ret);
4716                         break;
4717                 }
4718         }
4719
4720         return ret;
4721 }
4722
4723 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4724                                           struct buffer_head *xb_bh)
4725 {
4726         struct ocfs2_xattr_block *xb =
4727                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4728         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4729         int ret = 0;
4730         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4731         u64 p_blkno;
4732
4733         if (le16_to_cpu(el->l_next_free_rec) == 0)
4734                 return 0;
4735
4736         while (name_hash > 0) {
4737                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4738                                           &e_cpos, &num_clusters, el);
4739                 if (ret) {
4740                         mlog_errno(ret);
4741                         goto out;
4742                 }
4743
4744                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4745                                                   ocfs2_delete_xattr_in_bucket,
4746                                                   NULL);
4747                 if (ret) {
4748                         mlog_errno(ret);
4749                         goto out;
4750                 }
4751
4752                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4753                                              p_blkno, e_cpos, num_clusters);
4754                 if (ret) {
4755                         mlog_errno(ret);
4756                         break;
4757                 }
4758
4759                 if (e_cpos == 0)
4760                         break;
4761
4762                 name_hash = e_cpos - 1;
4763         }
4764
4765 out:
4766         return ret;
4767 }
4768
4769 /*
4770  * 'trusted' attributes support
4771  */
4772 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4773                                        size_t list_size, const char *name,
4774                                        size_t name_len)
4775 {
4776         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
4777         const size_t total_len = prefix_len + name_len + 1;
4778
4779         if (list && total_len <= list_size) {
4780                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4781                 memcpy(list + prefix_len, name, name_len);
4782                 list[prefix_len + name_len] = '\0';
4783         }
4784         return total_len;
4785 }
4786
4787 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4788                                    void *buffer, size_t size)
4789 {
4790         if (strcmp(name, "") == 0)
4791                 return -EINVAL;
4792         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4793                                buffer, size);
4794 }
4795
4796 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4797                                    const void *value, size_t size, int flags)
4798 {
4799         if (strcmp(name, "") == 0)
4800                 return -EINVAL;
4801
4802         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4803                                size, flags);
4804 }
4805
4806 struct xattr_handler ocfs2_xattr_trusted_handler = {
4807         .prefix = XATTR_TRUSTED_PREFIX,
4808         .list   = ocfs2_xattr_trusted_list,
4809         .get    = ocfs2_xattr_trusted_get,
4810         .set    = ocfs2_xattr_trusted_set,
4811 };
4812
4813 /*
4814  * 'user' attributes support
4815  */
4816 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4817                                     size_t list_size, const char *name,
4818                                     size_t name_len)
4819 {
4820         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
4821         const size_t total_len = prefix_len + name_len + 1;
4822         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4823
4824         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4825                 return 0;
4826
4827         if (list && total_len <= list_size) {
4828                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4829                 memcpy(list + prefix_len, name, name_len);
4830                 list[prefix_len + name_len] = '\0';
4831         }
4832         return total_len;
4833 }
4834
4835 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4836                                 void *buffer, size_t size)
4837 {
4838         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4839
4840         if (strcmp(name, "") == 0)
4841                 return -EINVAL;
4842         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4843                 return -EOPNOTSUPP;
4844         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4845                                buffer, size);
4846 }
4847
4848 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4849                                 const void *value, size_t size, int flags)
4850 {
4851         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4852
4853         if (strcmp(name, "") == 0)
4854                 return -EINVAL;
4855         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4856                 return -EOPNOTSUPP;
4857
4858         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4859                                size, flags);
4860 }
4861
4862 struct xattr_handler ocfs2_xattr_user_handler = {
4863         .prefix = XATTR_USER_PREFIX,
4864         .list   = ocfs2_xattr_user_list,
4865         .get    = ocfs2_xattr_user_get,
4866         .set    = ocfs2_xattr_user_set,
4867 };