ocfs2: Move buckets up into ocfs2_add_new_xattr_cluster().
[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 #include <linux/security.h>
39
40 #define MLOG_MASK_PREFIX ML_XATTR
41 #include <cluster/masklog.h>
42
43 #include "ocfs2.h"
44 #include "alloc.h"
45 #include "dlmglue.h"
46 #include "file.h"
47 #include "symlink.h"
48 #include "sysfile.h"
49 #include "inode.h"
50 #include "journal.h"
51 #include "ocfs2_fs.h"
52 #include "suballoc.h"
53 #include "uptodate.h"
54 #include "buffer_head_io.h"
55 #include "super.h"
56 #include "xattr.h"
57
58
59 struct ocfs2_xattr_def_value_root {
60         struct ocfs2_xattr_value_root   xv;
61         struct ocfs2_extent_rec         er;
62 };
63
64 struct ocfs2_xattr_bucket {
65         /* The inode these xattrs are associated with */
66         struct inode *bu_inode;
67
68         /* The actual buffers that make up the bucket */
69         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
70
71         /* How many blocks make up one bucket for this filesystem */
72         int bu_blocks;
73 };
74
75 struct ocfs2_xattr_set_ctxt {
76         handle_t *handle;
77         struct ocfs2_alloc_context *meta_ac;
78         struct ocfs2_alloc_context *data_ac;
79         struct ocfs2_cached_dealloc_ctxt dealloc;
80 };
81
82 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
83 #define OCFS2_XATTR_INLINE_SIZE 80
84 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
85                                          - sizeof(struct ocfs2_xattr_header) \
86                                          - sizeof(__u32))
87 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
88                                          - sizeof(struct ocfs2_xattr_block) \
89                                          - sizeof(struct ocfs2_xattr_header) \
90                                          - sizeof(__u32))
91
92 static struct ocfs2_xattr_def_value_root def_xv = {
93         .xv.xr_list.l_count = cpu_to_le16(1),
94 };
95
96 struct xattr_handler *ocfs2_xattr_handlers[] = {
97         &ocfs2_xattr_user_handler,
98 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
99         &ocfs2_xattr_acl_access_handler,
100         &ocfs2_xattr_acl_default_handler,
101 #endif
102         &ocfs2_xattr_trusted_handler,
103         &ocfs2_xattr_security_handler,
104         NULL
105 };
106
107 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
108         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
109 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
110         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
111                                         = &ocfs2_xattr_acl_access_handler,
112         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
113                                         = &ocfs2_xattr_acl_default_handler,
114 #endif
115         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
116         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
117 };
118
119 struct ocfs2_xattr_info {
120         int name_index;
121         const char *name;
122         const void *value;
123         size_t value_len;
124 };
125
126 struct ocfs2_xattr_search {
127         struct buffer_head *inode_bh;
128         /*
129          * xattr_bh point to the block buffer head which has extended attribute
130          * when extended attribute in inode, xattr_bh is equal to inode_bh.
131          */
132         struct buffer_head *xattr_bh;
133         struct ocfs2_xattr_header *header;
134         struct ocfs2_xattr_bucket *bucket;
135         void *base;
136         void *end;
137         struct ocfs2_xattr_entry *here;
138         int not_found;
139 };
140
141 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
142                                              struct ocfs2_xattr_header *xh,
143                                              int index,
144                                              int *block_off,
145                                              int *new_offset);
146
147 static int ocfs2_xattr_block_find(struct inode *inode,
148                                   int name_index,
149                                   const char *name,
150                                   struct ocfs2_xattr_search *xs);
151 static int ocfs2_xattr_index_block_find(struct inode *inode,
152                                         struct buffer_head *root_bh,
153                                         int name_index,
154                                         const char *name,
155                                         struct ocfs2_xattr_search *xs);
156
157 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
158                                         struct ocfs2_xattr_tree_root *xt,
159                                         char *buffer,
160                                         size_t buffer_size);
161
162 static int ocfs2_xattr_create_index_block(struct inode *inode,
163                                           struct ocfs2_xattr_search *xs,
164                                           struct ocfs2_xattr_set_ctxt *ctxt);
165
166 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
167                                              struct ocfs2_xattr_info *xi,
168                                              struct ocfs2_xattr_search *xs,
169                                              struct ocfs2_xattr_set_ctxt *ctxt);
170
171 static int ocfs2_delete_xattr_index_block(struct inode *inode,
172                                           struct buffer_head *xb_bh);
173 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
174                                   u64 src_blk, u64 last_blk, u64 to_blk,
175                                   unsigned int start_bucket,
176                                   u32 *first_hash);
177
178 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
179 {
180         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
181 }
182
183 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
184 {
185         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
186 }
187
188 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
189 {
190         u16 len = sb->s_blocksize -
191                  offsetof(struct ocfs2_xattr_header, xh_entries);
192
193         return len / sizeof(struct ocfs2_xattr_entry);
194 }
195
196 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
197 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
198 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
199
200 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
201 {
202         struct ocfs2_xattr_bucket *bucket;
203         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
204
205         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
206
207         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
208         if (bucket) {
209                 bucket->bu_inode = inode;
210                 bucket->bu_blocks = blks;
211         }
212
213         return bucket;
214 }
215
216 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
217 {
218         int i;
219
220         for (i = 0; i < bucket->bu_blocks; i++) {
221                 brelse(bucket->bu_bhs[i]);
222                 bucket->bu_bhs[i] = NULL;
223         }
224 }
225
226 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
227 {
228         if (bucket) {
229                 ocfs2_xattr_bucket_relse(bucket);
230                 bucket->bu_inode = NULL;
231                 kfree(bucket);
232         }
233 }
234
235 /*
236  * A bucket that has never been written to disk doesn't need to be
237  * read.  We just need the buffer_heads.  Don't call this for
238  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
239  * them fully.
240  */
241 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
242                                    u64 xb_blkno)
243 {
244         int i, rc = 0;
245
246         for (i = 0; i < bucket->bu_blocks; i++) {
247                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
248                                               xb_blkno + i);
249                 if (!bucket->bu_bhs[i]) {
250                         rc = -EIO;
251                         mlog_errno(rc);
252                         break;
253                 }
254
255                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
256                                            bucket->bu_bhs[i]))
257                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
258                                                       bucket->bu_bhs[i]);
259         }
260
261         if (rc)
262                 ocfs2_xattr_bucket_relse(bucket);
263         return rc;
264 }
265
266 /* Read the xattr bucket at xb_blkno */
267 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
268                                    u64 xb_blkno)
269 {
270         int rc;
271
272         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
273                                bucket->bu_blocks, bucket->bu_bhs, 0,
274                                NULL);
275         if (rc)
276                 ocfs2_xattr_bucket_relse(bucket);
277         return rc;
278 }
279
280 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
281                                              struct ocfs2_xattr_bucket *bucket,
282                                              int type)
283 {
284         int i, rc = 0;
285
286         for (i = 0; i < bucket->bu_blocks; i++) {
287                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
288                                           bucket->bu_bhs[i], type);
289                 if (rc) {
290                         mlog_errno(rc);
291                         break;
292                 }
293         }
294
295         return rc;
296 }
297
298 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
299                                              struct ocfs2_xattr_bucket *bucket)
300 {
301         int i;
302
303         for (i = 0; i < bucket->bu_blocks; i++)
304                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
305 }
306
307 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
308                                          struct ocfs2_xattr_bucket *src)
309 {
310         int i;
311         int blocksize = src->bu_inode->i_sb->s_blocksize;
312
313         BUG_ON(dest->bu_blocks != src->bu_blocks);
314         BUG_ON(dest->bu_inode != src->bu_inode);
315
316         for (i = 0; i < src->bu_blocks; i++) {
317                 memcpy(bucket_block(dest, i), bucket_block(src, i),
318                        blocksize);
319         }
320 }
321
322 static int ocfs2_validate_xattr_block(struct super_block *sb,
323                                       struct buffer_head *bh)
324 {
325         struct ocfs2_xattr_block *xb =
326                 (struct ocfs2_xattr_block *)bh->b_data;
327
328         mlog(0, "Validating xattr block %llu\n",
329              (unsigned long long)bh->b_blocknr);
330
331         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
332                 ocfs2_error(sb,
333                             "Extended attribute block #%llu has bad "
334                             "signature %.*s",
335                             (unsigned long long)bh->b_blocknr, 7,
336                             xb->xb_signature);
337                 return -EINVAL;
338         }
339
340         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
341                 ocfs2_error(sb,
342                             "Extended attribute block #%llu has an "
343                             "invalid xb_blkno of %llu",
344                             (unsigned long long)bh->b_blocknr,
345                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
346                 return -EINVAL;
347         }
348
349         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
350                 ocfs2_error(sb,
351                             "Extended attribute block #%llu has an invalid "
352                             "xb_fs_generation of #%u",
353                             (unsigned long long)bh->b_blocknr,
354                             le32_to_cpu(xb->xb_fs_generation));
355                 return -EINVAL;
356         }
357
358         return 0;
359 }
360
361 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
362                                   struct buffer_head **bh)
363 {
364         int rc;
365         struct buffer_head *tmp = *bh;
366
367         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
368                               ocfs2_validate_xattr_block);
369
370         /* If ocfs2_read_block() got us a new bh, pass it up. */
371         if (!rc && !*bh)
372                 *bh = tmp;
373
374         return rc;
375 }
376
377 static inline const char *ocfs2_xattr_prefix(int name_index)
378 {
379         struct xattr_handler *handler = NULL;
380
381         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
382                 handler = ocfs2_xattr_handler_map[name_index];
383
384         return handler ? handler->prefix : NULL;
385 }
386
387 static u32 ocfs2_xattr_name_hash(struct inode *inode,
388                                  const char *name,
389                                  int name_len)
390 {
391         /* Get hash value of uuid from super block */
392         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
393         int i;
394
395         /* hash extended attribute name */
396         for (i = 0; i < name_len; i++) {
397                 hash = (hash << OCFS2_HASH_SHIFT) ^
398                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
399                        *name++;
400         }
401
402         return hash;
403 }
404
405 /*
406  * ocfs2_xattr_hash_entry()
407  *
408  * Compute the hash of an extended attribute.
409  */
410 static void ocfs2_xattr_hash_entry(struct inode *inode,
411                                    struct ocfs2_xattr_header *header,
412                                    struct ocfs2_xattr_entry *entry)
413 {
414         u32 hash = 0;
415         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
416
417         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
418         entry->xe_name_hash = cpu_to_le32(hash);
419
420         return;
421 }
422
423 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
424 {
425         int size = 0;
426
427         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
428                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
429         else
430                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
431         size += sizeof(struct ocfs2_xattr_entry);
432
433         return size;
434 }
435
436 int ocfs2_calc_security_init(struct inode *dir,
437                              struct ocfs2_security_xattr_info *si,
438                              int *want_clusters,
439                              int *xattr_credits,
440                              struct ocfs2_alloc_context **xattr_ac)
441 {
442         int ret = 0;
443         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
444         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
445                                                  si->value_len);
446
447         /*
448          * The max space of security xattr taken inline is
449          * 256(name) + 80(value) + 16(entry) = 352 bytes,
450          * So reserve one metadata block for it is ok.
451          */
452         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
453             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
454                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
455                 if (ret) {
456                         mlog_errno(ret);
457                         return ret;
458                 }
459                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
460         }
461
462         /* reserve clusters for xattr value which will be set in B tree*/
463         if (si->value_len > OCFS2_XATTR_INLINE_SIZE)
464                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
465                                                            si->value_len);
466         return ret;
467 }
468
469 int ocfs2_calc_xattr_init(struct inode *dir,
470                           struct buffer_head *dir_bh,
471                           int mode,
472                           struct ocfs2_security_xattr_info *si,
473                           int *want_clusters,
474                           int *xattr_credits,
475                           struct ocfs2_alloc_context **xattr_ac)
476 {
477         int ret = 0;
478         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
479         int s_size = 0;
480         int a_size = 0;
481         int acl_len = 0;
482
483         if (si->enable)
484                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
485                                                      si->value_len);
486
487         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
488                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
489                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
490                                         "", NULL, 0);
491                 if (acl_len > 0) {
492                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
493                         if (S_ISDIR(mode))
494                                 a_size <<= 1;
495                 } else if (acl_len != 0 && acl_len != -ENODATA) {
496                         mlog_errno(ret);
497                         return ret;
498                 }
499         }
500
501         if (!(s_size + a_size))
502                 return ret;
503
504         /*
505          * The max space of security xattr taken inline is
506          * 256(name) + 80(value) + 16(entry) = 352 bytes,
507          * The max space of acl xattr taken inline is
508          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
509          * when blocksize = 512, may reserve one more cluser for
510          * xattr bucket, otherwise reserve one metadata block
511          * for them is ok.
512          */
513         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
514             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
515                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
516                 if (ret) {
517                         mlog_errno(ret);
518                         return ret;
519                 }
520                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
521         }
522
523         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
524             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
525                 *want_clusters += 1;
526                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
527         }
528
529         /* reserve clusters for xattr value which will be set in B tree*/
530         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE)
531                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
532                                                            si->value_len);
533         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
534             acl_len > OCFS2_XATTR_INLINE_SIZE) {
535                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
536                 if (S_ISDIR(mode))
537                         *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
538                                                                    acl_len);
539         }
540
541         return ret;
542 }
543
544 static int ocfs2_xattr_extend_allocation(struct inode *inode,
545                                          u32 clusters_to_add,
546                                          struct buffer_head *xattr_bh,
547                                          struct ocfs2_xattr_value_root *xv,
548                                          struct ocfs2_xattr_set_ctxt *ctxt)
549 {
550         int status = 0;
551         handle_t *handle = ctxt->handle;
552         enum ocfs2_alloc_restarted why;
553         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
554         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
555         struct ocfs2_extent_tree et;
556
557         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
558
559         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
560
561         status = ocfs2_journal_access(handle, inode, xattr_bh,
562                                       OCFS2_JOURNAL_ACCESS_WRITE);
563         if (status < 0) {
564                 mlog_errno(status);
565                 goto leave;
566         }
567
568         prev_clusters = le32_to_cpu(xv->xr_clusters);
569         status = ocfs2_add_clusters_in_btree(osb,
570                                              inode,
571                                              &logical_start,
572                                              clusters_to_add,
573                                              0,
574                                              &et,
575                                              handle,
576                                              ctxt->data_ac,
577                                              ctxt->meta_ac,
578                                              &why);
579         if (status < 0) {
580                 mlog_errno(status);
581                 goto leave;
582         }
583
584         status = ocfs2_journal_dirty(handle, xattr_bh);
585         if (status < 0) {
586                 mlog_errno(status);
587                 goto leave;
588         }
589
590         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
591
592         /*
593          * We should have already allocated enough space before the transaction,
594          * so no need to restart.
595          */
596         BUG_ON(why != RESTART_NONE || clusters_to_add);
597
598 leave:
599
600         return status;
601 }
602
603 static int __ocfs2_remove_xattr_range(struct inode *inode,
604                                       struct buffer_head *root_bh,
605                                       struct ocfs2_xattr_value_root *xv,
606                                       u32 cpos, u32 phys_cpos, u32 len,
607                                       struct ocfs2_xattr_set_ctxt *ctxt)
608 {
609         int ret;
610         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
611         handle_t *handle = ctxt->handle;
612         struct ocfs2_extent_tree et;
613
614         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
615
616         ret = ocfs2_journal_access(handle, inode, root_bh,
617                                    OCFS2_JOURNAL_ACCESS_WRITE);
618         if (ret) {
619                 mlog_errno(ret);
620                 goto out;
621         }
622
623         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
624                                   &ctxt->dealloc);
625         if (ret) {
626                 mlog_errno(ret);
627                 goto out;
628         }
629
630         le32_add_cpu(&xv->xr_clusters, -len);
631
632         ret = ocfs2_journal_dirty(handle, root_bh);
633         if (ret) {
634                 mlog_errno(ret);
635                 goto out;
636         }
637
638         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
639         if (ret)
640                 mlog_errno(ret);
641
642 out:
643         return ret;
644 }
645
646 static int ocfs2_xattr_shrink_size(struct inode *inode,
647                                    u32 old_clusters,
648                                    u32 new_clusters,
649                                    struct buffer_head *root_bh,
650                                    struct ocfs2_xattr_value_root *xv,
651                                    struct ocfs2_xattr_set_ctxt *ctxt)
652 {
653         int ret = 0;
654         u32 trunc_len, cpos, phys_cpos, alloc_size;
655         u64 block;
656
657         if (old_clusters <= new_clusters)
658                 return 0;
659
660         cpos = new_clusters;
661         trunc_len = old_clusters - new_clusters;
662         while (trunc_len) {
663                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
664                                                &alloc_size, &xv->xr_list);
665                 if (ret) {
666                         mlog_errno(ret);
667                         goto out;
668                 }
669
670                 if (alloc_size > trunc_len)
671                         alloc_size = trunc_len;
672
673                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
674                                                  phys_cpos, alloc_size,
675                                                  ctxt);
676                 if (ret) {
677                         mlog_errno(ret);
678                         goto out;
679                 }
680
681                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
682                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
683                                                        alloc_size);
684                 cpos += alloc_size;
685                 trunc_len -= alloc_size;
686         }
687
688 out:
689         return ret;
690 }
691
692 static int ocfs2_xattr_value_truncate(struct inode *inode,
693                                       struct buffer_head *root_bh,
694                                       struct ocfs2_xattr_value_root *xv,
695                                       int len,
696                                       struct ocfs2_xattr_set_ctxt *ctxt)
697 {
698         int ret;
699         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
700         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
701
702         if (new_clusters == old_clusters)
703                 return 0;
704
705         if (new_clusters > old_clusters)
706                 ret = ocfs2_xattr_extend_allocation(inode,
707                                                     new_clusters - old_clusters,
708                                                     root_bh, xv, ctxt);
709         else
710                 ret = ocfs2_xattr_shrink_size(inode,
711                                               old_clusters, new_clusters,
712                                               root_bh, xv, ctxt);
713
714         return ret;
715 }
716
717 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
718                                   size_t *result, const char *prefix,
719                                   const char *name, int name_len)
720 {
721         char *p = buffer + *result;
722         int prefix_len = strlen(prefix);
723         int total_len = prefix_len + name_len + 1;
724
725         *result += total_len;
726
727         /* we are just looking for how big our buffer needs to be */
728         if (!size)
729                 return 0;
730
731         if (*result > size)
732                 return -ERANGE;
733
734         memcpy(p, prefix, prefix_len);
735         memcpy(p + prefix_len, name, name_len);
736         p[prefix_len + name_len] = '\0';
737
738         return 0;
739 }
740
741 static int ocfs2_xattr_list_entries(struct inode *inode,
742                                     struct ocfs2_xattr_header *header,
743                                     char *buffer, size_t buffer_size)
744 {
745         size_t result = 0;
746         int i, type, ret;
747         const char *prefix, *name;
748
749         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
750                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
751                 type = ocfs2_xattr_get_type(entry);
752                 prefix = ocfs2_xattr_prefix(type);
753
754                 if (prefix) {
755                         name = (const char *)header +
756                                 le16_to_cpu(entry->xe_name_offset);
757
758                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
759                                                      &result, prefix, name,
760                                                      entry->xe_name_len);
761                         if (ret)
762                                 return ret;
763                 }
764         }
765
766         return result;
767 }
768
769 static int ocfs2_xattr_ibody_list(struct inode *inode,
770                                   struct ocfs2_dinode *di,
771                                   char *buffer,
772                                   size_t buffer_size)
773 {
774         struct ocfs2_xattr_header *header = NULL;
775         struct ocfs2_inode_info *oi = OCFS2_I(inode);
776         int ret = 0;
777
778         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
779                 return ret;
780
781         header = (struct ocfs2_xattr_header *)
782                  ((void *)di + inode->i_sb->s_blocksize -
783                  le16_to_cpu(di->i_xattr_inline_size));
784
785         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
786
787         return ret;
788 }
789
790 static int ocfs2_xattr_block_list(struct inode *inode,
791                                   struct ocfs2_dinode *di,
792                                   char *buffer,
793                                   size_t buffer_size)
794 {
795         struct buffer_head *blk_bh = NULL;
796         struct ocfs2_xattr_block *xb;
797         int ret = 0;
798
799         if (!di->i_xattr_loc)
800                 return ret;
801
802         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
803                                      &blk_bh);
804         if (ret < 0) {
805                 mlog_errno(ret);
806                 return ret;
807         }
808
809         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
810         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
811                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
812                 ret = ocfs2_xattr_list_entries(inode, header,
813                                                buffer, buffer_size);
814         } else {
815                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
816                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
817                                                    buffer, buffer_size);
818         }
819
820         brelse(blk_bh);
821
822         return ret;
823 }
824
825 ssize_t ocfs2_listxattr(struct dentry *dentry,
826                         char *buffer,
827                         size_t size)
828 {
829         int ret = 0, i_ret = 0, b_ret = 0;
830         struct buffer_head *di_bh = NULL;
831         struct ocfs2_dinode *di = NULL;
832         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
833
834         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
835                 return -EOPNOTSUPP;
836
837         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
838                 return ret;
839
840         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
841         if (ret < 0) {
842                 mlog_errno(ret);
843                 return ret;
844         }
845
846         di = (struct ocfs2_dinode *)di_bh->b_data;
847
848         down_read(&oi->ip_xattr_sem);
849         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
850         if (i_ret < 0)
851                 b_ret = 0;
852         else {
853                 if (buffer) {
854                         buffer += i_ret;
855                         size -= i_ret;
856                 }
857                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
858                                                buffer, size);
859                 if (b_ret < 0)
860                         i_ret = 0;
861         }
862         up_read(&oi->ip_xattr_sem);
863         ocfs2_inode_unlock(dentry->d_inode, 0);
864
865         brelse(di_bh);
866
867         return i_ret + b_ret;
868 }
869
870 static int ocfs2_xattr_find_entry(int name_index,
871                                   const char *name,
872                                   struct ocfs2_xattr_search *xs)
873 {
874         struct ocfs2_xattr_entry *entry;
875         size_t name_len;
876         int i, cmp = 1;
877
878         if (name == NULL)
879                 return -EINVAL;
880
881         name_len = strlen(name);
882         entry = xs->here;
883         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
884                 cmp = name_index - ocfs2_xattr_get_type(entry);
885                 if (!cmp)
886                         cmp = name_len - entry->xe_name_len;
887                 if (!cmp)
888                         cmp = memcmp(name, (xs->base +
889                                      le16_to_cpu(entry->xe_name_offset)),
890                                      name_len);
891                 if (cmp == 0)
892                         break;
893                 entry += 1;
894         }
895         xs->here = entry;
896
897         return cmp ? -ENODATA : 0;
898 }
899
900 static int ocfs2_xattr_get_value_outside(struct inode *inode,
901                                          struct ocfs2_xattr_value_root *xv,
902                                          void *buffer,
903                                          size_t len)
904 {
905         u32 cpos, p_cluster, num_clusters, bpc, clusters;
906         u64 blkno;
907         int i, ret = 0;
908         size_t cplen, blocksize;
909         struct buffer_head *bh = NULL;
910         struct ocfs2_extent_list *el;
911
912         el = &xv->xr_list;
913         clusters = le32_to_cpu(xv->xr_clusters);
914         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
915         blocksize = inode->i_sb->s_blocksize;
916
917         cpos = 0;
918         while (cpos < clusters) {
919                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
920                                                &num_clusters, el);
921                 if (ret) {
922                         mlog_errno(ret);
923                         goto out;
924                 }
925
926                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
927                 /* Copy ocfs2_xattr_value */
928                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
929                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
930                         if (ret) {
931                                 mlog_errno(ret);
932                                 goto out;
933                         }
934
935                         cplen = len >= blocksize ? blocksize : len;
936                         memcpy(buffer, bh->b_data, cplen);
937                         len -= cplen;
938                         buffer += cplen;
939
940                         brelse(bh);
941                         bh = NULL;
942                         if (len == 0)
943                                 break;
944                 }
945                 cpos += num_clusters;
946         }
947 out:
948         return ret;
949 }
950
951 static int ocfs2_xattr_ibody_get(struct inode *inode,
952                                  int name_index,
953                                  const char *name,
954                                  void *buffer,
955                                  size_t buffer_size,
956                                  struct ocfs2_xattr_search *xs)
957 {
958         struct ocfs2_inode_info *oi = OCFS2_I(inode);
959         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
960         struct ocfs2_xattr_value_root *xv;
961         size_t size;
962         int ret = 0;
963
964         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
965                 return -ENODATA;
966
967         xs->end = (void *)di + inode->i_sb->s_blocksize;
968         xs->header = (struct ocfs2_xattr_header *)
969                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
970         xs->base = (void *)xs->header;
971         xs->here = xs->header->xh_entries;
972
973         ret = ocfs2_xattr_find_entry(name_index, name, xs);
974         if (ret)
975                 return ret;
976         size = le64_to_cpu(xs->here->xe_value_size);
977         if (buffer) {
978                 if (size > buffer_size)
979                         return -ERANGE;
980                 if (ocfs2_xattr_is_local(xs->here)) {
981                         memcpy(buffer, (void *)xs->base +
982                                le16_to_cpu(xs->here->xe_name_offset) +
983                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
984                 } else {
985                         xv = (struct ocfs2_xattr_value_root *)
986                                 (xs->base + le16_to_cpu(
987                                  xs->here->xe_name_offset) +
988                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
989                         ret = ocfs2_xattr_get_value_outside(inode, xv,
990                                                             buffer, size);
991                         if (ret < 0) {
992                                 mlog_errno(ret);
993                                 return ret;
994                         }
995                 }
996         }
997
998         return size;
999 }
1000
1001 static int ocfs2_xattr_block_get(struct inode *inode,
1002                                  int name_index,
1003                                  const char *name,
1004                                  void *buffer,
1005                                  size_t buffer_size,
1006                                  struct ocfs2_xattr_search *xs)
1007 {
1008         struct ocfs2_xattr_block *xb;
1009         struct ocfs2_xattr_value_root *xv;
1010         size_t size;
1011         int ret = -ENODATA, name_offset, name_len, block_off, i;
1012
1013         xs->bucket = ocfs2_xattr_bucket_new(inode);
1014         if (!xs->bucket) {
1015                 ret = -ENOMEM;
1016                 mlog_errno(ret);
1017                 goto cleanup;
1018         }
1019
1020         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1021         if (ret) {
1022                 mlog_errno(ret);
1023                 goto cleanup;
1024         }
1025
1026         if (xs->not_found) {
1027                 ret = -ENODATA;
1028                 goto cleanup;
1029         }
1030
1031         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1032         size = le64_to_cpu(xs->here->xe_value_size);
1033         if (buffer) {
1034                 ret = -ERANGE;
1035                 if (size > buffer_size)
1036                         goto cleanup;
1037
1038                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1039                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1040                 i = xs->here - xs->header->xh_entries;
1041
1042                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1043                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1044                                                                 bucket_xh(xs->bucket),
1045                                                                 i,
1046                                                                 &block_off,
1047                                                                 &name_offset);
1048                         xs->base = bucket_block(xs->bucket, block_off);
1049                 }
1050                 if (ocfs2_xattr_is_local(xs->here)) {
1051                         memcpy(buffer, (void *)xs->base +
1052                                name_offset + name_len, size);
1053                 } else {
1054                         xv = (struct ocfs2_xattr_value_root *)
1055                                 (xs->base + name_offset + name_len);
1056                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1057                                                             buffer, size);
1058                         if (ret < 0) {
1059                                 mlog_errno(ret);
1060                                 goto cleanup;
1061                         }
1062                 }
1063         }
1064         ret = size;
1065 cleanup:
1066         ocfs2_xattr_bucket_free(xs->bucket);
1067
1068         brelse(xs->xattr_bh);
1069         xs->xattr_bh = NULL;
1070         return ret;
1071 }
1072
1073 int ocfs2_xattr_get_nolock(struct inode *inode,
1074                            struct buffer_head *di_bh,
1075                            int name_index,
1076                            const char *name,
1077                            void *buffer,
1078                            size_t buffer_size)
1079 {
1080         int ret;
1081         struct ocfs2_dinode *di = NULL;
1082         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1083         struct ocfs2_xattr_search xis = {
1084                 .not_found = -ENODATA,
1085         };
1086         struct ocfs2_xattr_search xbs = {
1087                 .not_found = -ENODATA,
1088         };
1089
1090         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1091                 return -EOPNOTSUPP;
1092
1093         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1094                 ret = -ENODATA;
1095
1096         xis.inode_bh = xbs.inode_bh = di_bh;
1097         di = (struct ocfs2_dinode *)di_bh->b_data;
1098
1099         down_read(&oi->ip_xattr_sem);
1100         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1101                                     buffer_size, &xis);
1102         if (ret == -ENODATA && di->i_xattr_loc)
1103                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1104                                             buffer_size, &xbs);
1105         up_read(&oi->ip_xattr_sem);
1106
1107         return ret;
1108 }
1109
1110 /* ocfs2_xattr_get()
1111  *
1112  * Copy an extended attribute into the buffer provided.
1113  * Buffer is NULL to compute the size of buffer required.
1114  */
1115 static int ocfs2_xattr_get(struct inode *inode,
1116                            int name_index,
1117                            const char *name,
1118                            void *buffer,
1119                            size_t buffer_size)
1120 {
1121         int ret;
1122         struct buffer_head *di_bh = NULL;
1123
1124         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1125         if (ret < 0) {
1126                 mlog_errno(ret);
1127                 return ret;
1128         }
1129         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1130                                      name, buffer, buffer_size);
1131
1132         ocfs2_inode_unlock(inode, 0);
1133
1134         brelse(di_bh);
1135
1136         return ret;
1137 }
1138
1139 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1140                                            handle_t *handle,
1141                                            struct ocfs2_xattr_value_root *xv,
1142                                            const void *value,
1143                                            int value_len)
1144 {
1145         int ret = 0, i, cp_len, credits;
1146         u16 blocksize = inode->i_sb->s_blocksize;
1147         u32 p_cluster, num_clusters;
1148         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1149         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1150         u64 blkno;
1151         struct buffer_head *bh = NULL;
1152
1153         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1154
1155         /*
1156          * In __ocfs2_xattr_set_value_outside has already been dirtied,
1157          * so we don't need to worry about whether ocfs2_extend_trans
1158          * will create a new transactio for us or not.
1159          */
1160         credits = clusters * bpc;
1161         ret = ocfs2_extend_trans(handle, credits);
1162         if (ret) {
1163                 mlog_errno(ret);
1164                 goto out;
1165         }
1166
1167         while (cpos < clusters) {
1168                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1169                                                &num_clusters, &xv->xr_list);
1170                 if (ret) {
1171                         mlog_errno(ret);
1172                         goto out;
1173                 }
1174
1175                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1176
1177                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1178                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1179                         if (ret) {
1180                                 mlog_errno(ret);
1181                                 goto out;
1182                         }
1183
1184                         ret = ocfs2_journal_access(handle,
1185                                                    inode,
1186                                                    bh,
1187                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1188                         if (ret < 0) {
1189                                 mlog_errno(ret);
1190                                 goto out;
1191                         }
1192
1193                         cp_len = value_len > blocksize ? blocksize : value_len;
1194                         memcpy(bh->b_data, value, cp_len);
1195                         value_len -= cp_len;
1196                         value += cp_len;
1197                         if (cp_len < blocksize)
1198                                 memset(bh->b_data + cp_len, 0,
1199                                        blocksize - cp_len);
1200
1201                         ret = ocfs2_journal_dirty(handle, bh);
1202                         if (ret < 0) {
1203                                 mlog_errno(ret);
1204                                 goto out;
1205                         }
1206                         brelse(bh);
1207                         bh = NULL;
1208
1209                         /*
1210                          * XXX: do we need to empty all the following
1211                          * blocks in this cluster?
1212                          */
1213                         if (!value_len)
1214                                 break;
1215                 }
1216                 cpos += num_clusters;
1217         }
1218 out:
1219         brelse(bh);
1220
1221         return ret;
1222 }
1223
1224 static int ocfs2_xattr_cleanup(struct inode *inode,
1225                                handle_t *handle,
1226                                struct ocfs2_xattr_info *xi,
1227                                struct ocfs2_xattr_search *xs,
1228                                size_t offs)
1229 {
1230         int ret = 0;
1231         size_t name_len = strlen(xi->name);
1232         void *val = xs->base + offs;
1233         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1234
1235         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1236                                    OCFS2_JOURNAL_ACCESS_WRITE);
1237         if (ret) {
1238                 mlog_errno(ret);
1239                 goto out;
1240         }
1241         /* Decrease xattr count */
1242         le16_add_cpu(&xs->header->xh_count, -1);
1243         /* Remove the xattr entry and tree root which has already be set*/
1244         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1245         memset(val, 0, size);
1246
1247         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1248         if (ret < 0)
1249                 mlog_errno(ret);
1250 out:
1251         return ret;
1252 }
1253
1254 static int ocfs2_xattr_update_entry(struct inode *inode,
1255                                     handle_t *handle,
1256                                     struct ocfs2_xattr_info *xi,
1257                                     struct ocfs2_xattr_search *xs,
1258                                     size_t offs)
1259 {
1260         int ret;
1261
1262         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1263                                    OCFS2_JOURNAL_ACCESS_WRITE);
1264         if (ret) {
1265                 mlog_errno(ret);
1266                 goto out;
1267         }
1268
1269         xs->here->xe_name_offset = cpu_to_le16(offs);
1270         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1271         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1272                 ocfs2_xattr_set_local(xs->here, 1);
1273         else
1274                 ocfs2_xattr_set_local(xs->here, 0);
1275         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1276
1277         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1278         if (ret < 0)
1279                 mlog_errno(ret);
1280 out:
1281         return ret;
1282 }
1283
1284 /*
1285  * ocfs2_xattr_set_value_outside()
1286  *
1287  * Set large size value in B tree.
1288  */
1289 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1290                                          struct ocfs2_xattr_info *xi,
1291                                          struct ocfs2_xattr_search *xs,
1292                                          struct ocfs2_xattr_set_ctxt *ctxt,
1293                                          size_t offs)
1294 {
1295         size_t name_len = strlen(xi->name);
1296         void *val = xs->base + offs;
1297         struct ocfs2_xattr_value_root *xv = NULL;
1298         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1299         int ret = 0;
1300
1301         memset(val, 0, size);
1302         memcpy(val, xi->name, name_len);
1303         xv = (struct ocfs2_xattr_value_root *)
1304                 (val + OCFS2_XATTR_SIZE(name_len));
1305         xv->xr_clusters = 0;
1306         xv->xr_last_eb_blk = 0;
1307         xv->xr_list.l_tree_depth = 0;
1308         xv->xr_list.l_count = cpu_to_le16(1);
1309         xv->xr_list.l_next_free_rec = 0;
1310
1311         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1312                                          xi->value_len, ctxt);
1313         if (ret < 0) {
1314                 mlog_errno(ret);
1315                 return ret;
1316         }
1317         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, offs);
1318         if (ret < 0) {
1319                 mlog_errno(ret);
1320                 return ret;
1321         }
1322         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, xv,
1323                                               xi->value, xi->value_len);
1324         if (ret < 0)
1325                 mlog_errno(ret);
1326
1327         return ret;
1328 }
1329
1330 /*
1331  * ocfs2_xattr_set_entry_local()
1332  *
1333  * Set, replace or remove extended attribute in local.
1334  */
1335 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1336                                         struct ocfs2_xattr_info *xi,
1337                                         struct ocfs2_xattr_search *xs,
1338                                         struct ocfs2_xattr_entry *last,
1339                                         size_t min_offs)
1340 {
1341         size_t name_len = strlen(xi->name);
1342         int i;
1343
1344         if (xi->value && xs->not_found) {
1345                 /* Insert the new xattr entry. */
1346                 le16_add_cpu(&xs->header->xh_count, 1);
1347                 ocfs2_xattr_set_type(last, xi->name_index);
1348                 ocfs2_xattr_set_local(last, 1);
1349                 last->xe_name_len = name_len;
1350         } else {
1351                 void *first_val;
1352                 void *val;
1353                 size_t offs, size;
1354
1355                 first_val = xs->base + min_offs;
1356                 offs = le16_to_cpu(xs->here->xe_name_offset);
1357                 val = xs->base + offs;
1358
1359                 if (le64_to_cpu(xs->here->xe_value_size) >
1360                     OCFS2_XATTR_INLINE_SIZE)
1361                         size = OCFS2_XATTR_SIZE(name_len) +
1362                                 OCFS2_XATTR_ROOT_SIZE;
1363                 else
1364                         size = OCFS2_XATTR_SIZE(name_len) +
1365                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1366
1367                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1368                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1369                         /* The old and the new value have the
1370                            same size. Just replace the value. */
1371                         ocfs2_xattr_set_local(xs->here, 1);
1372                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1373                         /* Clear value bytes. */
1374                         memset(val + OCFS2_XATTR_SIZE(name_len),
1375                                0,
1376                                OCFS2_XATTR_SIZE(xi->value_len));
1377                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1378                                xi->value,
1379                                xi->value_len);
1380                         return;
1381                 }
1382                 /* Remove the old name+value. */
1383                 memmove(first_val + size, first_val, val - first_val);
1384                 memset(first_val, 0, size);
1385                 xs->here->xe_name_hash = 0;
1386                 xs->here->xe_name_offset = 0;
1387                 ocfs2_xattr_set_local(xs->here, 1);
1388                 xs->here->xe_value_size = 0;
1389
1390                 min_offs += size;
1391
1392                 /* Adjust all value offsets. */
1393                 last = xs->header->xh_entries;
1394                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1395                         size_t o = le16_to_cpu(last->xe_name_offset);
1396
1397                         if (o < offs)
1398                                 last->xe_name_offset = cpu_to_le16(o + size);
1399                         last += 1;
1400                 }
1401
1402                 if (!xi->value) {
1403                         /* Remove the old entry. */
1404                         last -= 1;
1405                         memmove(xs->here, xs->here + 1,
1406                                 (void *)last - (void *)xs->here);
1407                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1408                         le16_add_cpu(&xs->header->xh_count, -1);
1409                 }
1410         }
1411         if (xi->value) {
1412                 /* Insert the new name+value. */
1413                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1414                                 OCFS2_XATTR_SIZE(xi->value_len);
1415                 void *val = xs->base + min_offs - size;
1416
1417                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1418                 memset(val, 0, size);
1419                 memcpy(val, xi->name, name_len);
1420                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1421                        xi->value,
1422                        xi->value_len);
1423                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1424                 ocfs2_xattr_set_local(xs->here, 1);
1425                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1426         }
1427
1428         return;
1429 }
1430
1431 /*
1432  * ocfs2_xattr_set_entry()
1433  *
1434  * Set extended attribute entry into inode or block.
1435  *
1436  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1437  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1438  * then set value in B tree with set_value_outside().
1439  */
1440 static int ocfs2_xattr_set_entry(struct inode *inode,
1441                                  struct ocfs2_xattr_info *xi,
1442                                  struct ocfs2_xattr_search *xs,
1443                                  struct ocfs2_xattr_set_ctxt *ctxt,
1444                                  int flag)
1445 {
1446         struct ocfs2_xattr_entry *last;
1447         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1448         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1449         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1450         size_t size_l = 0;
1451         handle_t *handle = ctxt->handle;
1452         int free, i, ret;
1453         struct ocfs2_xattr_info xi_l = {
1454                 .name_index = xi->name_index,
1455                 .name = xi->name,
1456                 .value = xi->value,
1457                 .value_len = xi->value_len,
1458         };
1459
1460         /* Compute min_offs, last and free space. */
1461         last = xs->header->xh_entries;
1462
1463         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1464                 size_t offs = le16_to_cpu(last->xe_name_offset);
1465                 if (offs < min_offs)
1466                         min_offs = offs;
1467                 last += 1;
1468         }
1469
1470         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1471         if (free < 0)
1472                 return -EIO;
1473
1474         if (!xs->not_found) {
1475                 size_t size = 0;
1476                 if (ocfs2_xattr_is_local(xs->here))
1477                         size = OCFS2_XATTR_SIZE(name_len) +
1478                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1479                 else
1480                         size = OCFS2_XATTR_SIZE(name_len) +
1481                                 OCFS2_XATTR_ROOT_SIZE;
1482                 free += (size + sizeof(struct ocfs2_xattr_entry));
1483         }
1484         /* Check free space in inode or block */
1485         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1486                 if (free < sizeof(struct ocfs2_xattr_entry) +
1487                            OCFS2_XATTR_SIZE(name_len) +
1488                            OCFS2_XATTR_ROOT_SIZE) {
1489                         ret = -ENOSPC;
1490                         goto out;
1491                 }
1492                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1493                 xi_l.value = (void *)&def_xv;
1494                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1495         } else if (xi->value) {
1496                 if (free < sizeof(struct ocfs2_xattr_entry) +
1497                            OCFS2_XATTR_SIZE(name_len) +
1498                            OCFS2_XATTR_SIZE(xi->value_len)) {
1499                         ret = -ENOSPC;
1500                         goto out;
1501                 }
1502         }
1503
1504         if (!xs->not_found) {
1505                 /* For existing extended attribute */
1506                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1507                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1508                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1509                 void *val = xs->base + offs;
1510
1511                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1512                         /* Replace existing local xattr with tree root */
1513                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1514                                                             ctxt, offs);
1515                         if (ret < 0)
1516                                 mlog_errno(ret);
1517                         goto out;
1518                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1519                         /* For existing xattr which has value outside */
1520                         struct ocfs2_xattr_value_root *xv = NULL;
1521                         xv = (struct ocfs2_xattr_value_root *)(val +
1522                                 OCFS2_XATTR_SIZE(name_len));
1523
1524                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1525                                 /*
1526                                  * If new value need set outside also,
1527                                  * first truncate old value to new value,
1528                                  * then set new value with set_value_outside().
1529                                  */
1530                                 ret = ocfs2_xattr_value_truncate(inode,
1531                                                                  xs->xattr_bh,
1532                                                                  xv,
1533                                                                  xi->value_len,
1534                                                                  ctxt);
1535                                 if (ret < 0) {
1536                                         mlog_errno(ret);
1537                                         goto out;
1538                                 }
1539
1540                                 ret = ocfs2_xattr_update_entry(inode,
1541                                                                handle,
1542                                                                xi,
1543                                                                xs,
1544                                                                offs);
1545                                 if (ret < 0) {
1546                                         mlog_errno(ret);
1547                                         goto out;
1548                                 }
1549
1550                                 ret = __ocfs2_xattr_set_value_outside(inode,
1551                                                                 handle,
1552                                                                 xv,
1553                                                                 xi->value,
1554                                                                 xi->value_len);
1555                                 if (ret < 0)
1556                                         mlog_errno(ret);
1557                                 goto out;
1558                         } else {
1559                                 /*
1560                                  * If new value need set in local,
1561                                  * just trucate old value to zero.
1562                                  */
1563                                  ret = ocfs2_xattr_value_truncate(inode,
1564                                                                   xs->xattr_bh,
1565                                                                   xv,
1566                                                                   0,
1567                                                                   ctxt);
1568                                 if (ret < 0)
1569                                         mlog_errno(ret);
1570                         }
1571                 }
1572         }
1573
1574         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1575                                    OCFS2_JOURNAL_ACCESS_WRITE);
1576         if (ret) {
1577                 mlog_errno(ret);
1578                 goto out;
1579         }
1580
1581         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1582                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1583                                            OCFS2_JOURNAL_ACCESS_WRITE);
1584                 if (ret) {
1585                         mlog_errno(ret);
1586                         goto out;
1587                 }
1588         }
1589
1590         /*
1591          * Set value in local, include set tree root in local.
1592          * This is the first step for value size >INLINE_SIZE.
1593          */
1594         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1595
1596         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1597                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1598                 if (ret < 0) {
1599                         mlog_errno(ret);
1600                         goto out;
1601                 }
1602         }
1603
1604         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1605             (flag & OCFS2_INLINE_XATTR_FL)) {
1606                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1607                 unsigned int xattrsize = osb->s_xattr_inline_size;
1608
1609                 /*
1610                  * Adjust extent record count or inline data size
1611                  * to reserve space for extended attribute.
1612                  */
1613                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1614                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1615                         le16_add_cpu(&idata->id_count, -xattrsize);
1616                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1617                         struct ocfs2_extent_list *el = &di->id2.i_list;
1618                         le16_add_cpu(&el->l_count, -(xattrsize /
1619                                         sizeof(struct ocfs2_extent_rec)));
1620                 }
1621                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1622         }
1623         /* Update xattr flag */
1624         spin_lock(&oi->ip_lock);
1625         oi->ip_dyn_features |= flag;
1626         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1627         spin_unlock(&oi->ip_lock);
1628         /* Update inode ctime */
1629         inode->i_ctime = CURRENT_TIME;
1630         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1631         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1632
1633         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1634         if (ret < 0)
1635                 mlog_errno(ret);
1636
1637         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1638                 /*
1639                  * Set value outside in B tree.
1640                  * This is the second step for value size > INLINE_SIZE.
1641                  */
1642                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1643                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt, offs);
1644                 if (ret < 0) {
1645                         int ret2;
1646
1647                         mlog_errno(ret);
1648                         /*
1649                          * If set value outside failed, we have to clean
1650                          * the junk tree root we have already set in local.
1651                          */
1652                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1653                                                    xi, xs, offs);
1654                         if (ret2 < 0)
1655                                 mlog_errno(ret2);
1656                 }
1657         }
1658 out:
1659         return ret;
1660 }
1661
1662 static int ocfs2_remove_value_outside(struct inode*inode,
1663                                       struct buffer_head *bh,
1664                                       struct ocfs2_xattr_header *header)
1665 {
1666         int ret = 0, i;
1667         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1668         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1669
1670         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1671
1672         ctxt.handle = ocfs2_start_trans(osb,
1673                                         ocfs2_remove_extent_credits(osb->sb));
1674         if (IS_ERR(ctxt.handle)) {
1675                 ret = PTR_ERR(ctxt.handle);
1676                 mlog_errno(ret);
1677                 goto out;
1678         }
1679
1680         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1681                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1682
1683                 if (!ocfs2_xattr_is_local(entry)) {
1684                         struct ocfs2_xattr_value_root *xv;
1685                         void *val;
1686
1687                         val = (void *)header +
1688                                 le16_to_cpu(entry->xe_name_offset);
1689                         xv = (struct ocfs2_xattr_value_root *)
1690                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1691                         ret = ocfs2_xattr_value_truncate(inode, bh, xv,
1692                                                          0, &ctxt);
1693                         if (ret < 0) {
1694                                 mlog_errno(ret);
1695                                 break;
1696                         }
1697                 }
1698         }
1699
1700         ocfs2_commit_trans(osb, ctxt.handle);
1701         ocfs2_schedule_truncate_log_flush(osb, 1);
1702         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1703 out:
1704         return ret;
1705 }
1706
1707 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1708                                     struct buffer_head *di_bh)
1709 {
1710
1711         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1712         struct ocfs2_xattr_header *header;
1713         int ret;
1714
1715         header = (struct ocfs2_xattr_header *)
1716                  ((void *)di + inode->i_sb->s_blocksize -
1717                  le16_to_cpu(di->i_xattr_inline_size));
1718
1719         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1720
1721         return ret;
1722 }
1723
1724 static int ocfs2_xattr_block_remove(struct inode *inode,
1725                                     struct buffer_head *blk_bh)
1726 {
1727         struct ocfs2_xattr_block *xb;
1728         int ret = 0;
1729
1730         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1731         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1732                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1733                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1734         } else
1735                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1736
1737         return ret;
1738 }
1739
1740 static int ocfs2_xattr_free_block(struct inode *inode,
1741                                   u64 block)
1742 {
1743         struct inode *xb_alloc_inode;
1744         struct buffer_head *xb_alloc_bh = NULL;
1745         struct buffer_head *blk_bh = NULL;
1746         struct ocfs2_xattr_block *xb;
1747         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1748         handle_t *handle;
1749         int ret = 0;
1750         u64 blk, bg_blkno;
1751         u16 bit;
1752
1753         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1754         if (ret < 0) {
1755                 mlog_errno(ret);
1756                 goto out;
1757         }
1758
1759         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1760         if (ret < 0) {
1761                 mlog_errno(ret);
1762                 goto out;
1763         }
1764
1765         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1766         blk = le64_to_cpu(xb->xb_blkno);
1767         bit = le16_to_cpu(xb->xb_suballoc_bit);
1768         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1769
1770         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1771                                 EXTENT_ALLOC_SYSTEM_INODE,
1772                                 le16_to_cpu(xb->xb_suballoc_slot));
1773         if (!xb_alloc_inode) {
1774                 ret = -ENOMEM;
1775                 mlog_errno(ret);
1776                 goto out;
1777         }
1778         mutex_lock(&xb_alloc_inode->i_mutex);
1779
1780         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1781         if (ret < 0) {
1782                 mlog_errno(ret);
1783                 goto out_mutex;
1784         }
1785
1786         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1787         if (IS_ERR(handle)) {
1788                 ret = PTR_ERR(handle);
1789                 mlog_errno(ret);
1790                 goto out_unlock;
1791         }
1792
1793         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1794                                        bit, bg_blkno, 1);
1795         if (ret < 0)
1796                 mlog_errno(ret);
1797
1798         ocfs2_commit_trans(osb, handle);
1799 out_unlock:
1800         ocfs2_inode_unlock(xb_alloc_inode, 1);
1801         brelse(xb_alloc_bh);
1802 out_mutex:
1803         mutex_unlock(&xb_alloc_inode->i_mutex);
1804         iput(xb_alloc_inode);
1805 out:
1806         brelse(blk_bh);
1807         return ret;
1808 }
1809
1810 /*
1811  * ocfs2_xattr_remove()
1812  *
1813  * Free extended attribute resources associated with this inode.
1814  */
1815 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1816 {
1817         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1818         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1819         handle_t *handle;
1820         int ret;
1821
1822         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1823                 return 0;
1824
1825         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1826                 return 0;
1827
1828         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1829                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1830                 if (ret < 0) {
1831                         mlog_errno(ret);
1832                         goto out;
1833                 }
1834         }
1835
1836         if (di->i_xattr_loc) {
1837                 ret = ocfs2_xattr_free_block(inode,
1838                                              le64_to_cpu(di->i_xattr_loc));
1839                 if (ret < 0) {
1840                         mlog_errno(ret);
1841                         goto out;
1842                 }
1843         }
1844
1845         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1846                                    OCFS2_INODE_UPDATE_CREDITS);
1847         if (IS_ERR(handle)) {
1848                 ret = PTR_ERR(handle);
1849                 mlog_errno(ret);
1850                 goto out;
1851         }
1852         ret = ocfs2_journal_access(handle, inode, di_bh,
1853                                    OCFS2_JOURNAL_ACCESS_WRITE);
1854         if (ret) {
1855                 mlog_errno(ret);
1856                 goto out_commit;
1857         }
1858
1859         di->i_xattr_loc = 0;
1860
1861         spin_lock(&oi->ip_lock);
1862         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1863         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1864         spin_unlock(&oi->ip_lock);
1865
1866         ret = ocfs2_journal_dirty(handle, di_bh);
1867         if (ret < 0)
1868                 mlog_errno(ret);
1869 out_commit:
1870         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1871 out:
1872         return ret;
1873 }
1874
1875 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1876                                         struct ocfs2_dinode *di)
1877 {
1878         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1879         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1880         int free;
1881
1882         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1883                 return 0;
1884
1885         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1886                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1887                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1888         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1889                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1890                         le64_to_cpu(di->i_size);
1891         } else {
1892                 struct ocfs2_extent_list *el = &di->id2.i_list;
1893                 free = (le16_to_cpu(el->l_count) -
1894                         le16_to_cpu(el->l_next_free_rec)) *
1895                         sizeof(struct ocfs2_extent_rec);
1896         }
1897         if (free >= xattrsize)
1898                 return 1;
1899
1900         return 0;
1901 }
1902
1903 /*
1904  * ocfs2_xattr_ibody_find()
1905  *
1906  * Find extended attribute in inode block and
1907  * fill search info into struct ocfs2_xattr_search.
1908  */
1909 static int ocfs2_xattr_ibody_find(struct inode *inode,
1910                                   int name_index,
1911                                   const char *name,
1912                                   struct ocfs2_xattr_search *xs)
1913 {
1914         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1915         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1916         int ret;
1917         int has_space = 0;
1918
1919         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1920                 return 0;
1921
1922         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1923                 down_read(&oi->ip_alloc_sem);
1924                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1925                 up_read(&oi->ip_alloc_sem);
1926                 if (!has_space)
1927                         return 0;
1928         }
1929
1930         xs->xattr_bh = xs->inode_bh;
1931         xs->end = (void *)di + inode->i_sb->s_blocksize;
1932         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1933                 xs->header = (struct ocfs2_xattr_header *)
1934                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1935         else
1936                 xs->header = (struct ocfs2_xattr_header *)
1937                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1938         xs->base = (void *)xs->header;
1939         xs->here = xs->header->xh_entries;
1940
1941         /* Find the named attribute. */
1942         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1943                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1944                 if (ret && ret != -ENODATA)
1945                         return ret;
1946                 xs->not_found = ret;
1947         }
1948
1949         return 0;
1950 }
1951
1952 /*
1953  * ocfs2_xattr_ibody_set()
1954  *
1955  * Set, replace or remove an extended attribute into inode block.
1956  *
1957  */
1958 static int ocfs2_xattr_ibody_set(struct inode *inode,
1959                                  struct ocfs2_xattr_info *xi,
1960                                  struct ocfs2_xattr_search *xs,
1961                                  struct ocfs2_xattr_set_ctxt *ctxt)
1962 {
1963         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1964         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1965         int ret;
1966
1967         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1968                 return -ENOSPC;
1969
1970         down_write(&oi->ip_alloc_sem);
1971         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1972                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1973                         ret = -ENOSPC;
1974                         goto out;
1975                 }
1976         }
1977
1978         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
1979                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1980 out:
1981         up_write(&oi->ip_alloc_sem);
1982
1983         return ret;
1984 }
1985
1986 /*
1987  * ocfs2_xattr_block_find()
1988  *
1989  * Find extended attribute in external block and
1990  * fill search info into struct ocfs2_xattr_search.
1991  */
1992 static int ocfs2_xattr_block_find(struct inode *inode,
1993                                   int name_index,
1994                                   const char *name,
1995                                   struct ocfs2_xattr_search *xs)
1996 {
1997         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1998         struct buffer_head *blk_bh = NULL;
1999         struct ocfs2_xattr_block *xb;
2000         int ret = 0;
2001
2002         if (!di->i_xattr_loc)
2003                 return ret;
2004
2005         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2006                                      &blk_bh);
2007         if (ret < 0) {
2008                 mlog_errno(ret);
2009                 return ret;
2010         }
2011
2012         xs->xattr_bh = blk_bh;
2013         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2014
2015         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2016                 xs->header = &xb->xb_attrs.xb_header;
2017                 xs->base = (void *)xs->header;
2018                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2019                 xs->here = xs->header->xh_entries;
2020
2021                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2022         } else
2023                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2024                                                    name_index,
2025                                                    name, xs);
2026
2027         if (ret && ret != -ENODATA) {
2028                 xs->xattr_bh = NULL;
2029                 goto cleanup;
2030         }
2031         xs->not_found = ret;
2032         return 0;
2033 cleanup:
2034         brelse(blk_bh);
2035
2036         return ret;
2037 }
2038
2039 /*
2040  * ocfs2_xattr_block_set()
2041  *
2042  * Set, replace or remove an extended attribute into external block.
2043  *
2044  */
2045 static int ocfs2_xattr_block_set(struct inode *inode,
2046                                  struct ocfs2_xattr_info *xi,
2047                                  struct ocfs2_xattr_search *xs,
2048                                  struct ocfs2_xattr_set_ctxt *ctxt)
2049 {
2050         struct buffer_head *new_bh = NULL;
2051         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2052         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2053         handle_t *handle = ctxt->handle;
2054         struct ocfs2_xattr_block *xblk = NULL;
2055         u16 suballoc_bit_start;
2056         u32 num_got;
2057         u64 first_blkno;
2058         int ret;
2059
2060         if (!xs->xattr_bh) {
2061                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
2062                                            OCFS2_JOURNAL_ACCESS_CREATE);
2063                 if (ret < 0) {
2064                         mlog_errno(ret);
2065                         goto end;
2066                 }
2067
2068                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2069                                            &suballoc_bit_start, &num_got,
2070                                            &first_blkno);
2071                 if (ret < 0) {
2072                         mlog_errno(ret);
2073                         goto end;
2074                 }
2075
2076                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2077                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2078
2079                 ret = ocfs2_journal_access(handle, inode, new_bh,
2080                                            OCFS2_JOURNAL_ACCESS_CREATE);
2081                 if (ret < 0) {
2082                         mlog_errno(ret);
2083                         goto end;
2084                 }
2085
2086                 /* Initialize ocfs2_xattr_block */
2087                 xs->xattr_bh = new_bh;
2088                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2089                 memset(xblk, 0, inode->i_sb->s_blocksize);
2090                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2091                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2092                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2093                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2094                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2095
2096                 xs->header = &xblk->xb_attrs.xb_header;
2097                 xs->base = (void *)xs->header;
2098                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2099                 xs->here = xs->header->xh_entries;
2100
2101                 ret = ocfs2_journal_dirty(handle, new_bh);
2102                 if (ret < 0) {
2103                         mlog_errno(ret);
2104                         goto end;
2105                 }
2106                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2107                 ocfs2_journal_dirty(handle, xs->inode_bh);
2108         } else
2109                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2110
2111         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2112                 /* Set extended attribute into external block */
2113                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2114                                             OCFS2_HAS_XATTR_FL);
2115                 if (!ret || ret != -ENOSPC)
2116                         goto end;
2117
2118                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2119                 if (ret)
2120                         goto end;
2121         }
2122
2123         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2124
2125 end:
2126
2127         return ret;
2128 }
2129
2130 /* Check whether the new xattr can be inserted into the inode. */
2131 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2132                                        struct ocfs2_xattr_info *xi,
2133                                        struct ocfs2_xattr_search *xs)
2134 {
2135         u64 value_size;
2136         struct ocfs2_xattr_entry *last;
2137         int free, i;
2138         size_t min_offs = xs->end - xs->base;
2139
2140         if (!xs->header)
2141                 return 0;
2142
2143         last = xs->header->xh_entries;
2144
2145         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2146                 size_t offs = le16_to_cpu(last->xe_name_offset);
2147                 if (offs < min_offs)
2148                         min_offs = offs;
2149                 last += 1;
2150         }
2151
2152         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2153         if (free < 0)
2154                 return 0;
2155
2156         BUG_ON(!xs->not_found);
2157
2158         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2159                 value_size = OCFS2_XATTR_ROOT_SIZE;
2160         else
2161                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2162
2163         if (free >= sizeof(struct ocfs2_xattr_entry) +
2164                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2165                 return 1;
2166
2167         return 0;
2168 }
2169
2170 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2171                                      struct ocfs2_dinode *di,
2172                                      struct ocfs2_xattr_info *xi,
2173                                      struct ocfs2_xattr_search *xis,
2174                                      struct ocfs2_xattr_search *xbs,
2175                                      int *clusters_need,
2176                                      int *meta_need,
2177                                      int *credits_need)
2178 {
2179         int ret = 0, old_in_xb = 0;
2180         int clusters_add = 0, meta_add = 0, credits = 0;
2181         struct buffer_head *bh = NULL;
2182         struct ocfs2_xattr_block *xb = NULL;
2183         struct ocfs2_xattr_entry *xe = NULL;
2184         struct ocfs2_xattr_value_root *xv = NULL;
2185         char *base = NULL;
2186         int name_offset, name_len = 0;
2187         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2188                                                     xi->value_len);
2189         u64 value_size;
2190
2191         if (xis->not_found && xbs->not_found) {
2192                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2193
2194                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2195                         clusters_add += new_clusters;
2196                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2197                                                         &def_xv.xv.xr_list,
2198                                                         new_clusters);
2199                 }
2200
2201                 goto meta_guess;
2202         }
2203
2204         if (!xis->not_found) {
2205                 xe = xis->here;
2206                 name_offset = le16_to_cpu(xe->xe_name_offset);
2207                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2208                 base = xis->base;
2209                 credits += OCFS2_INODE_UPDATE_CREDITS;
2210         } else {
2211                 int i, block_off = 0;
2212                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2213                 xe = xbs->here;
2214                 name_offset = le16_to_cpu(xe->xe_name_offset);
2215                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2216                 i = xbs->here - xbs->header->xh_entries;
2217                 old_in_xb = 1;
2218
2219                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2220                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2221                                                         bucket_xh(xbs->bucket),
2222                                                         i, &block_off,
2223                                                         &name_offset);
2224                         base = bucket_block(xbs->bucket, block_off);
2225                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2226                 } else {
2227                         base = xbs->base;
2228                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2229                 }
2230         }
2231
2232         /*
2233          * delete a xattr doesn't need metadata and cluster allocation.
2234          * so just calculate the credits and return.
2235          *
2236          * The credits for removing the value tree will be extended
2237          * by ocfs2_remove_extent itself.
2238          */
2239         if (!xi->value) {
2240                 if (!ocfs2_xattr_is_local(xe))
2241                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2242
2243                 goto out;
2244         }
2245
2246         /* do cluster allocation guess first. */
2247         value_size = le64_to_cpu(xe->xe_value_size);
2248
2249         if (old_in_xb) {
2250                 /*
2251                  * In xattr set, we always try to set the xe in inode first,
2252                  * so if it can be inserted into inode successfully, the old
2253                  * one will be removed from the xattr block, and this xattr
2254                  * will be inserted into inode as a new xattr in inode.
2255                  */
2256                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2257                         clusters_add += new_clusters;
2258                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2259                                     OCFS2_INODE_UPDATE_CREDITS;
2260                         if (!ocfs2_xattr_is_local(xe))
2261                                 credits += ocfs2_calc_extend_credits(
2262                                                         inode->i_sb,
2263                                                         &def_xv.xv.xr_list,
2264                                                         new_clusters);
2265                         goto out;
2266                 }
2267         }
2268
2269         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2270                 /* the new values will be stored outside. */
2271                 u32 old_clusters = 0;
2272
2273                 if (!ocfs2_xattr_is_local(xe)) {
2274                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2275                                                                  value_size);
2276                         xv = (struct ocfs2_xattr_value_root *)
2277                              (base + name_offset + name_len);
2278                         value_size = OCFS2_XATTR_ROOT_SIZE;
2279                 } else
2280                         xv = &def_xv.xv;
2281
2282                 if (old_clusters >= new_clusters) {
2283                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2284                         goto out;
2285                 } else {
2286                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2287                         clusters_add += new_clusters - old_clusters;
2288                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2289                                                              &xv->xr_list,
2290                                                              new_clusters -
2291                                                              old_clusters);
2292                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2293                                 goto out;
2294                 }
2295         } else {
2296                 /*
2297                  * Now the new value will be stored inside. So if the new
2298                  * value is smaller than the size of value root or the old
2299                  * value, we don't need any allocation, otherwise we have
2300                  * to guess metadata allocation.
2301                  */
2302                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2303                     (!ocfs2_xattr_is_local(xe) &&
2304                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2305                         goto out;
2306         }
2307
2308 meta_guess:
2309         /* calculate metadata allocation. */
2310         if (di->i_xattr_loc) {
2311                 if (!xbs->xattr_bh) {
2312                         ret = ocfs2_read_xattr_block(inode,
2313                                                      le64_to_cpu(di->i_xattr_loc),
2314                                                      &bh);
2315                         if (ret) {
2316                                 mlog_errno(ret);
2317                                 goto out;
2318                         }
2319
2320                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2321                 } else
2322                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2323
2324                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2325                         struct ocfs2_extent_list *el =
2326                                  &xb->xb_attrs.xb_root.xt_list;
2327                         meta_add += ocfs2_extend_meta_needed(el);
2328                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2329                                                              el, 1);
2330                 }
2331
2332                 /*
2333                  * This cluster will be used either for new bucket or for
2334                  * new xattr block.
2335                  * If the cluster size is the same as the bucket size, one
2336                  * more is needed since we may need to extend the bucket
2337                  * also.
2338                  */
2339                 clusters_add += 1;
2340                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2341                 if (OCFS2_XATTR_BUCKET_SIZE ==
2342                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2343                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2344                         clusters_add += 1;
2345                 }
2346         } else {
2347                 meta_add += 1;
2348                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2349         }
2350 out:
2351         if (clusters_need)
2352                 *clusters_need = clusters_add;
2353         if (meta_need)
2354                 *meta_need = meta_add;
2355         if (credits_need)
2356                 *credits_need = credits;
2357         brelse(bh);
2358         return ret;
2359 }
2360
2361 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2362                                      struct ocfs2_dinode *di,
2363                                      struct ocfs2_xattr_info *xi,
2364                                      struct ocfs2_xattr_search *xis,
2365                                      struct ocfs2_xattr_search *xbs,
2366                                      struct ocfs2_xattr_set_ctxt *ctxt,
2367                                      int *credits)
2368 {
2369         int clusters_add, meta_add, ret;
2370         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2371
2372         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2373
2374         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2375
2376         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2377                                         &clusters_add, &meta_add, credits);
2378         if (ret) {
2379                 mlog_errno(ret);
2380                 return ret;
2381         }
2382
2383         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2384              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2385
2386         if (meta_add) {
2387                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2388                                                         &ctxt->meta_ac);
2389                 if (ret) {
2390                         mlog_errno(ret);
2391                         goto out;
2392                 }
2393         }
2394
2395         if (clusters_add) {
2396                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2397                 if (ret)
2398                         mlog_errno(ret);
2399         }
2400 out:
2401         if (ret) {
2402                 if (ctxt->meta_ac) {
2403                         ocfs2_free_alloc_context(ctxt->meta_ac);
2404                         ctxt->meta_ac = NULL;
2405                 }
2406
2407                 /*
2408                  * We cannot have an error and a non null ctxt->data_ac.
2409                  */
2410         }
2411
2412         return ret;
2413 }
2414
2415 static int __ocfs2_xattr_set_handle(struct inode *inode,
2416                                     struct ocfs2_dinode *di,
2417                                     struct ocfs2_xattr_info *xi,
2418                                     struct ocfs2_xattr_search *xis,
2419                                     struct ocfs2_xattr_search *xbs,
2420                                     struct ocfs2_xattr_set_ctxt *ctxt)
2421 {
2422         int ret = 0, credits, old_found;
2423
2424         if (!xi->value) {
2425                 /* Remove existing extended attribute */
2426                 if (!xis->not_found)
2427                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2428                 else if (!xbs->not_found)
2429                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2430         } else {
2431                 /* We always try to set extended attribute into inode first*/
2432                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2433                 if (!ret && !xbs->not_found) {
2434                         /*
2435                          * If succeed and that extended attribute existing in
2436                          * external block, then we will remove it.
2437                          */
2438                         xi->value = NULL;
2439                         xi->value_len = 0;
2440
2441                         old_found = xis->not_found;
2442                         xis->not_found = -ENODATA;
2443                         ret = ocfs2_calc_xattr_set_need(inode,
2444                                                         di,
2445                                                         xi,
2446                                                         xis,
2447                                                         xbs,
2448                                                         NULL,
2449                                                         NULL,
2450                                                         &credits);
2451                         xis->not_found = old_found;
2452                         if (ret) {
2453                                 mlog_errno(ret);
2454                                 goto out;
2455                         }
2456
2457                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2458                                         ctxt->handle->h_buffer_credits);
2459                         if (ret) {
2460                                 mlog_errno(ret);
2461                                 goto out;
2462                         }
2463                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2464                 } else if (ret == -ENOSPC) {
2465                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2466                                 ret = ocfs2_xattr_block_find(inode,
2467                                                              xi->name_index,
2468                                                              xi->name, xbs);
2469                                 if (ret)
2470                                         goto out;
2471
2472                                 old_found = xis->not_found;
2473                                 xis->not_found = -ENODATA;
2474                                 ret = ocfs2_calc_xattr_set_need(inode,
2475                                                                 di,
2476                                                                 xi,
2477                                                                 xis,
2478                                                                 xbs,
2479                                                                 NULL,
2480                                                                 NULL,
2481                                                                 &credits);
2482                                 xis->not_found = old_found;
2483                                 if (ret) {
2484                                         mlog_errno(ret);
2485                                         goto out;
2486                                 }
2487
2488                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2489                                         ctxt->handle->h_buffer_credits);
2490                                 if (ret) {
2491                                         mlog_errno(ret);
2492                                         goto out;
2493                                 }
2494                         }
2495                         /*
2496                          * If no space in inode, we will set extended attribute
2497                          * into external block.
2498                          */
2499                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2500                         if (ret)
2501                                 goto out;
2502                         if (!xis->not_found) {
2503                                 /*
2504                                  * If succeed and that extended attribute
2505                                  * existing in inode, we will remove it.
2506                                  */
2507                                 xi->value = NULL;
2508                                 xi->value_len = 0;
2509                                 xbs->not_found = -ENODATA;
2510                                 ret = ocfs2_calc_xattr_set_need(inode,
2511                                                                 di,
2512                                                                 xi,
2513                                                                 xis,
2514                                                                 xbs,
2515                                                                 NULL,
2516                                                                 NULL,
2517                                                                 &credits);
2518                                 if (ret) {
2519                                         mlog_errno(ret);
2520                                         goto out;
2521                                 }
2522
2523                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2524                                                 ctxt->handle->h_buffer_credits);
2525                                 if (ret) {
2526                                         mlog_errno(ret);
2527                                         goto out;
2528                                 }
2529                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2530                                                             xis, ctxt);
2531                         }
2532                 }
2533         }
2534
2535 out:
2536         return ret;
2537 }
2538
2539 /*
2540  * This function only called duing creating inode
2541  * for init security/acl xattrs of the new inode.
2542  * The xattrs could be put into ibody or extent block,
2543  * xattr bucket would not be use in this case.
2544  * transanction credits also be reserved in here.
2545  */
2546 int ocfs2_xattr_set_handle(handle_t *handle,
2547                            struct inode *inode,
2548                            struct buffer_head *di_bh,
2549                            int name_index,
2550                            const char *name,
2551                            const void *value,
2552                            size_t value_len,
2553                            int flags,
2554                            struct ocfs2_alloc_context *meta_ac,
2555                            struct ocfs2_alloc_context *data_ac)
2556 {
2557         struct ocfs2_dinode *di;
2558         int ret;
2559
2560         struct ocfs2_xattr_info xi = {
2561                 .name_index = name_index,
2562                 .name = name,
2563                 .value = value,
2564                 .value_len = value_len,
2565         };
2566
2567         struct ocfs2_xattr_search xis = {
2568                 .not_found = -ENODATA,
2569         };
2570
2571         struct ocfs2_xattr_search xbs = {
2572                 .not_found = -ENODATA,
2573         };
2574
2575         struct ocfs2_xattr_set_ctxt ctxt = {
2576                 .handle = handle,
2577                 .meta_ac = meta_ac,
2578                 .data_ac = data_ac,
2579         };
2580
2581         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2582                 return -EOPNOTSUPP;
2583
2584         xis.inode_bh = xbs.inode_bh = di_bh;
2585         di = (struct ocfs2_dinode *)di_bh->b_data;
2586
2587         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2588
2589         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2590         if (ret)
2591                 goto cleanup;
2592         if (xis.not_found) {
2593                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2594                 if (ret)
2595                         goto cleanup;
2596         }
2597
2598         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2599
2600 cleanup:
2601         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2602         brelse(xbs.xattr_bh);
2603
2604         return ret;
2605 }
2606
2607 /*
2608  * ocfs2_xattr_set()
2609  *
2610  * Set, replace or remove an extended attribute for this inode.
2611  * value is NULL to remove an existing extended attribute, else either
2612  * create or replace an extended attribute.
2613  */
2614 int ocfs2_xattr_set(struct inode *inode,
2615                     int name_index,
2616                     const char *name,
2617                     const void *value,
2618                     size_t value_len,
2619                     int flags)
2620 {
2621         struct buffer_head *di_bh = NULL;
2622         struct ocfs2_dinode *di;
2623         int ret, credits;
2624         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2625         struct inode *tl_inode = osb->osb_tl_inode;
2626         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2627
2628         struct ocfs2_xattr_info xi = {
2629                 .name_index = name_index,
2630                 .name = name,
2631                 .value = value,
2632                 .value_len = value_len,
2633         };
2634
2635         struct ocfs2_xattr_search xis = {
2636                 .not_found = -ENODATA,
2637         };
2638
2639         struct ocfs2_xattr_search xbs = {
2640                 .not_found = -ENODATA,
2641         };
2642
2643         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2644                 return -EOPNOTSUPP;
2645
2646         /*
2647          * Only xbs will be used on indexed trees.  xis doesn't need a
2648          * bucket.
2649          */
2650         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2651         if (!xbs.bucket) {
2652                 mlog_errno(-ENOMEM);
2653                 return -ENOMEM;
2654         }
2655
2656         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2657         if (ret < 0) {
2658                 mlog_errno(ret);
2659                 goto cleanup_nolock;
2660         }
2661         xis.inode_bh = xbs.inode_bh = di_bh;
2662         di = (struct ocfs2_dinode *)di_bh->b_data;
2663
2664         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2665         /*
2666          * Scan inode and external block to find the same name
2667          * extended attribute and collect search infomation.
2668          */
2669         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2670         if (ret)
2671                 goto cleanup;
2672         if (xis.not_found) {
2673                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2674                 if (ret)
2675                         goto cleanup;
2676         }
2677
2678         if (xis.not_found && xbs.not_found) {
2679                 ret = -ENODATA;
2680                 if (flags & XATTR_REPLACE)
2681                         goto cleanup;
2682                 ret = 0;
2683                 if (!value)
2684                         goto cleanup;
2685         } else {
2686                 ret = -EEXIST;
2687                 if (flags & XATTR_CREATE)
2688                         goto cleanup;
2689         }
2690
2691
2692         mutex_lock(&tl_inode->i_mutex);
2693
2694         if (ocfs2_truncate_log_needs_flush(osb)) {
2695                 ret = __ocfs2_flush_truncate_log(osb);
2696                 if (ret < 0) {
2697                         mutex_unlock(&tl_inode->i_mutex);
2698                         mlog_errno(ret);
2699                         goto cleanup;
2700                 }
2701         }
2702         mutex_unlock(&tl_inode->i_mutex);
2703
2704         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2705                                         &xbs, &ctxt, &credits);
2706         if (ret) {
2707                 mlog_errno(ret);
2708                 goto cleanup;
2709         }
2710
2711         ctxt.handle = ocfs2_start_trans(osb, credits);
2712         if (IS_ERR(ctxt.handle)) {
2713                 ret = PTR_ERR(ctxt.handle);
2714                 mlog_errno(ret);
2715                 goto cleanup;
2716         }
2717
2718         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2719
2720         ocfs2_commit_trans(osb, ctxt.handle);
2721
2722         if (ctxt.data_ac)
2723                 ocfs2_free_alloc_context(ctxt.data_ac);
2724         if (ctxt.meta_ac)
2725                 ocfs2_free_alloc_context(ctxt.meta_ac);
2726         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2727                 ocfs2_schedule_truncate_log_flush(osb, 1);
2728         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2729 cleanup:
2730         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2731         ocfs2_inode_unlock(inode, 1);
2732 cleanup_nolock:
2733         brelse(di_bh);
2734         brelse(xbs.xattr_bh);
2735         ocfs2_xattr_bucket_free(xbs.bucket);
2736
2737         return ret;
2738 }
2739
2740 /*
2741  * Find the xattr extent rec which may contains name_hash.
2742  * e_cpos will be the first name hash of the xattr rec.
2743  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2744  */
2745 static int ocfs2_xattr_get_rec(struct inode *inode,
2746                                u32 name_hash,
2747                                u64 *p_blkno,
2748                                u32 *e_cpos,
2749                                u32 *num_clusters,
2750                                struct ocfs2_extent_list *el)
2751 {
2752         int ret = 0, i;
2753         struct buffer_head *eb_bh = NULL;
2754         struct ocfs2_extent_block *eb;
2755         struct ocfs2_extent_rec *rec = NULL;
2756         u64 e_blkno = 0;
2757
2758         if (el->l_tree_depth) {
2759                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2760                 if (ret) {
2761                         mlog_errno(ret);
2762                         goto out;
2763                 }
2764
2765                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2766                 el = &eb->h_list;
2767
2768                 if (el->l_tree_depth) {
2769                         ocfs2_error(inode->i_sb,
2770                                     "Inode %lu has non zero tree depth in "
2771                                     "xattr tree block %llu\n", inode->i_ino,
2772                                     (unsigned long long)eb_bh->b_blocknr);
2773                         ret = -EROFS;
2774                         goto out;
2775                 }
2776         }
2777
2778         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2779                 rec = &el->l_recs[i];
2780
2781                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2782                         e_blkno = le64_to_cpu(rec->e_blkno);
2783                         break;
2784                 }
2785         }
2786
2787         if (!e_blkno) {
2788                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2789                             "record (%u, %u, 0) in xattr", inode->i_ino,
2790                             le32_to_cpu(rec->e_cpos),
2791                             ocfs2_rec_clusters(el, rec));
2792                 ret = -EROFS;
2793                 goto out;
2794         }
2795
2796         *p_blkno = le64_to_cpu(rec->e_blkno);
2797         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2798         if (e_cpos)
2799                 *e_cpos = le32_to_cpu(rec->e_cpos);
2800 out:
2801         brelse(eb_bh);
2802         return ret;
2803 }
2804
2805 typedef int (xattr_bucket_func)(struct inode *inode,
2806                                 struct ocfs2_xattr_bucket *bucket,
2807                                 void *para);
2808
2809 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2810                                    struct ocfs2_xattr_bucket *bucket,
2811                                    int name_index,
2812                                    const char *name,
2813                                    u32 name_hash,
2814                                    u16 *xe_index,
2815                                    int *found)
2816 {
2817         int i, ret = 0, cmp = 1, block_off, new_offset;
2818         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2819         size_t name_len = strlen(name);
2820         struct ocfs2_xattr_entry *xe = NULL;
2821         char *xe_name;
2822
2823         /*
2824          * We don't use binary search in the bucket because there
2825          * may be multiple entries with the same name hash.
2826          */
2827         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2828                 xe = &xh->xh_entries[i];
2829
2830                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2831                         continue;
2832                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2833                         break;
2834
2835                 cmp = name_index - ocfs2_xattr_get_type(xe);
2836                 if (!cmp)
2837                         cmp = name_len - xe->xe_name_len;
2838                 if (cmp)
2839                         continue;
2840
2841                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2842                                                         xh,
2843                                                         i,
2844                                                         &block_off,
2845                                                         &new_offset);
2846                 if (ret) {
2847                         mlog_errno(ret);
2848                         break;
2849                 }
2850
2851
2852                 xe_name = bucket_block(bucket, block_off) + new_offset;
2853                 if (!memcmp(name, xe_name, name_len)) {
2854                         *xe_index = i;
2855                         *found = 1;
2856                         ret = 0;
2857                         break;
2858                 }
2859         }
2860
2861         return ret;
2862 }
2863
2864 /*
2865  * Find the specified xattr entry in a series of buckets.
2866  * This series start from p_blkno and last for num_clusters.
2867  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2868  * the num of the valid buckets.
2869  *
2870  * Return the buffer_head this xattr should reside in. And if the xattr's
2871  * hash is in the gap of 2 buckets, return the lower bucket.
2872  */
2873 static int ocfs2_xattr_bucket_find(struct inode *inode,
2874                                    int name_index,
2875                                    const char *name,
2876                                    u32 name_hash,
2877                                    u64 p_blkno,
2878                                    u32 first_hash,
2879                                    u32 num_clusters,
2880                                    struct ocfs2_xattr_search *xs)
2881 {
2882         int ret, found = 0;
2883         struct ocfs2_xattr_header *xh = NULL;
2884         struct ocfs2_xattr_entry *xe = NULL;
2885         u16 index = 0;
2886         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2887         int low_bucket = 0, bucket, high_bucket;
2888         struct ocfs2_xattr_bucket *search;
2889         u32 last_hash;
2890         u64 blkno, lower_blkno = 0;
2891
2892         search = ocfs2_xattr_bucket_new(inode);
2893         if (!search) {
2894                 ret = -ENOMEM;
2895                 mlog_errno(ret);
2896                 goto out;
2897         }
2898
2899         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2900         if (ret) {
2901                 mlog_errno(ret);
2902                 goto out;
2903         }
2904
2905         xh = bucket_xh(search);
2906         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2907         while (low_bucket <= high_bucket) {
2908                 ocfs2_xattr_bucket_relse(search);
2909
2910                 bucket = (low_bucket + high_bucket) / 2;
2911                 blkno = p_blkno + bucket * blk_per_bucket;
2912                 ret = ocfs2_read_xattr_bucket(search, blkno);
2913                 if (ret) {
2914                         mlog_errno(ret);
2915                         goto out;
2916                 }
2917
2918                 xh = bucket_xh(search);
2919                 xe = &xh->xh_entries[0];
2920                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2921                         high_bucket = bucket - 1;
2922                         continue;
2923                 }
2924
2925                 /*
2926                  * Check whether the hash of the last entry in our
2927                  * bucket is larger than the search one. for an empty
2928                  * bucket, the last one is also the first one.
2929                  */
2930                 if (xh->xh_count)
2931                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2932
2933                 last_hash = le32_to_cpu(xe->xe_name_hash);
2934
2935                 /* record lower_blkno which may be the insert place. */
2936                 lower_blkno = blkno;
2937
2938                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2939                         low_bucket = bucket + 1;
2940                         continue;
2941                 }
2942
2943                 /* the searched xattr should reside in this bucket if exists. */
2944                 ret = ocfs2_find_xe_in_bucket(inode, search,
2945                                               name_index, name, name_hash,
2946                                               &index, &found);
2947                 if (ret) {
2948                         mlog_errno(ret);
2949                         goto out;
2950                 }
2951                 break;
2952         }
2953
2954         /*
2955          * Record the bucket we have found.
2956          * When the xattr's hash value is in the gap of 2 buckets, we will
2957          * always set it to the previous bucket.
2958          */
2959         if (!lower_blkno)
2960                 lower_blkno = p_blkno;
2961
2962         /* This should be in cache - we just read it during the search */
2963         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
2964         if (ret) {
2965                 mlog_errno(ret);
2966                 goto out;
2967         }
2968
2969         xs->header = bucket_xh(xs->bucket);
2970         xs->base = bucket_block(xs->bucket, 0);
2971         xs->end = xs->base + inode->i_sb->s_blocksize;
2972
2973         if (found) {
2974                 xs->here = &xs->header->xh_entries[index];
2975                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2976                      (unsigned long long)bucket_blkno(xs->bucket), index);
2977         } else
2978                 ret = -ENODATA;
2979
2980 out:
2981         ocfs2_xattr_bucket_free(search);
2982         return ret;
2983 }
2984
2985 static int ocfs2_xattr_index_block_find(struct inode *inode,
2986                                         struct buffer_head *root_bh,
2987                                         int name_index,
2988                                         const char *name,
2989                                         struct ocfs2_xattr_search *xs)
2990 {
2991         int ret;
2992         struct ocfs2_xattr_block *xb =
2993                         (struct ocfs2_xattr_block *)root_bh->b_data;
2994         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2995         struct ocfs2_extent_list *el = &xb_root->xt_list;
2996         u64 p_blkno = 0;
2997         u32 first_hash, num_clusters = 0;
2998         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2999
3000         if (le16_to_cpu(el->l_next_free_rec) == 0)
3001                 return -ENODATA;
3002
3003         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3004              name, name_hash, name_index);
3005
3006         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3007                                   &num_clusters, el);
3008         if (ret) {
3009                 mlog_errno(ret);
3010                 goto out;
3011         }
3012
3013         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3014
3015         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3016              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3017              first_hash);
3018
3019         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3020                                       p_blkno, first_hash, num_clusters, xs);
3021
3022 out:
3023         return ret;
3024 }
3025
3026 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3027                                        u64 blkno,
3028                                        u32 clusters,
3029                                        xattr_bucket_func *func,
3030                                        void *para)
3031 {
3032         int i, ret = 0;
3033         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3034         u32 num_buckets = clusters * bpc;
3035         struct ocfs2_xattr_bucket *bucket;
3036
3037         bucket = ocfs2_xattr_bucket_new(inode);
3038         if (!bucket) {
3039                 mlog_errno(-ENOMEM);
3040                 return -ENOMEM;
3041         }
3042
3043         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3044              clusters, (unsigned long long)blkno);
3045
3046         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3047                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3048                 if (ret) {
3049                         mlog_errno(ret);
3050                         break;
3051                 }
3052
3053                 /*
3054                  * The real bucket num in this series of blocks is stored
3055                  * in the 1st bucket.
3056                  */
3057                 if (i == 0)
3058                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3059
3060                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3061                      (unsigned long long)blkno,
3062                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3063                 if (func) {
3064                         ret = func(inode, bucket, para);
3065                         if (ret)
3066                                 mlog_errno(ret);
3067                         /* Fall through to bucket_relse() */
3068                 }
3069
3070                 ocfs2_xattr_bucket_relse(bucket);
3071                 if (ret)
3072                         break;
3073         }
3074
3075         ocfs2_xattr_bucket_free(bucket);
3076         return ret;
3077 }
3078
3079 struct ocfs2_xattr_tree_list {
3080         char *buffer;
3081         size_t buffer_size;
3082         size_t result;
3083 };
3084
3085 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3086                                              struct ocfs2_xattr_header *xh,
3087                                              int index,
3088                                              int *block_off,
3089                                              int *new_offset)
3090 {
3091         u16 name_offset;
3092
3093         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3094                 return -EINVAL;
3095
3096         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3097
3098         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3099         *new_offset = name_offset % inode->i_sb->s_blocksize;
3100
3101         return 0;
3102 }
3103
3104 static int ocfs2_list_xattr_bucket(struct inode *inode,
3105                                    struct ocfs2_xattr_bucket *bucket,
3106                                    void *para)
3107 {
3108         int ret = 0, type;
3109         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3110         int i, block_off, new_offset;
3111         const char *prefix, *name;
3112
3113         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3114                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3115                 type = ocfs2_xattr_get_type(entry);
3116                 prefix = ocfs2_xattr_prefix(type);
3117
3118                 if (prefix) {
3119                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3120                                                                 bucket_xh(bucket),
3121                                                                 i,
3122                                                                 &block_off,
3123                                                                 &new_offset);
3124                         if (ret)
3125                                 break;
3126
3127                         name = (const char *)bucket_block(bucket, block_off) +
3128                                 new_offset;
3129                         ret = ocfs2_xattr_list_entry(xl->buffer,
3130                                                      xl->buffer_size,
3131                                                      &xl->result,
3132                                                      prefix, name,
3133                                                      entry->xe_name_len);
3134                         if (ret)
3135                                 break;
3136                 }
3137         }
3138
3139         return ret;
3140 }
3141
3142 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3143                                              struct ocfs2_xattr_tree_root *xt,
3144                                              char *buffer,
3145                                              size_t buffer_size)
3146 {
3147         struct ocfs2_extent_list *el = &xt->xt_list;
3148         int ret = 0;
3149         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3150         u64 p_blkno = 0;
3151         struct ocfs2_xattr_tree_list xl = {
3152                 .buffer = buffer,
3153                 .buffer_size = buffer_size,
3154                 .result = 0,
3155         };
3156
3157         if (le16_to_cpu(el->l_next_free_rec) == 0)
3158                 return 0;
3159
3160         while (name_hash > 0) {
3161                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3162                                           &e_cpos, &num_clusters, el);
3163                 if (ret) {
3164                         mlog_errno(ret);
3165                         goto out;
3166                 }
3167
3168                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3169                                                   ocfs2_list_xattr_bucket,
3170                                                   &xl);
3171                 if (ret) {
3172                         mlog_errno(ret);
3173                         goto out;
3174                 }
3175
3176                 if (e_cpos == 0)
3177                         break;
3178
3179                 name_hash = e_cpos - 1;
3180         }
3181
3182         ret = xl.result;
3183 out:
3184         return ret;
3185 }
3186
3187 static int cmp_xe(const void *a, const void *b)
3188 {
3189         const struct ocfs2_xattr_entry *l = a, *r = b;
3190         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3191         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3192
3193         if (l_hash > r_hash)
3194                 return 1;
3195         if (l_hash < r_hash)
3196                 return -1;
3197         return 0;
3198 }
3199
3200 static void swap_xe(void *a, void *b, int size)
3201 {
3202         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3203
3204         tmp = *l;
3205         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3206         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3207 }
3208
3209 /*
3210  * When the ocfs2_xattr_block is filled up, new bucket will be created
3211  * and all the xattr entries will be moved to the new bucket.
3212  * The header goes at the start of the bucket, and the names+values are
3213  * filled from the end.  This is why *target starts as the last buffer.
3214  * Note: we need to sort the entries since they are not saved in order
3215  * in the ocfs2_xattr_block.
3216  */
3217 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3218                                            struct buffer_head *xb_bh,
3219                                            struct ocfs2_xattr_bucket *bucket)
3220 {
3221         int i, blocksize = inode->i_sb->s_blocksize;
3222         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3223         u16 offset, size, off_change;
3224         struct ocfs2_xattr_entry *xe;
3225         struct ocfs2_xattr_block *xb =
3226                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3227         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3228         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3229         u16 count = le16_to_cpu(xb_xh->xh_count);
3230         char *src = xb_bh->b_data;
3231         char *target = bucket_block(bucket, blks - 1);
3232
3233         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3234              (unsigned long long)xb_bh->b_blocknr,
3235              (unsigned long long)bucket_blkno(bucket));
3236
3237         for (i = 0; i < blks; i++)
3238                 memset(bucket_block(bucket, i), 0, blocksize);
3239
3240         /*
3241          * Since the xe_name_offset is based on ocfs2_xattr_header,
3242          * there is a offset change corresponding to the change of
3243          * ocfs2_xattr_header's position.
3244          */
3245         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3246         xe = &xb_xh->xh_entries[count - 1];
3247         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3248         size = blocksize - offset;
3249
3250         /* copy all the names and values. */
3251         memcpy(target + offset, src + offset, size);
3252
3253         /* Init new header now. */
3254         xh->xh_count = xb_xh->xh_count;
3255         xh->xh_num_buckets = cpu_to_le16(1);
3256         xh->xh_name_value_len = cpu_to_le16(size);
3257         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3258
3259         /* copy all the entries. */
3260         target = bucket_block(bucket, 0);
3261         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3262         size = count * sizeof(struct ocfs2_xattr_entry);
3263         memcpy(target + offset, (char *)xb_xh + offset, size);
3264
3265         /* Change the xe offset for all the xe because of the move. */
3266         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3267                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3268         for (i = 0; i < count; i++)
3269                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3270
3271         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3272              offset, size, off_change);
3273
3274         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3275              cmp_xe, swap_xe);
3276 }
3277
3278 /*
3279  * After we move xattr from block to index btree, we have to
3280  * update ocfs2_xattr_search to the new xe and base.
3281  *
3282  * When the entry is in xattr block, xattr_bh indicates the storage place.
3283  * While if the entry is in index b-tree, "bucket" indicates the
3284  * real place of the xattr.
3285  */
3286 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3287                                             struct ocfs2_xattr_search *xs,
3288                                             struct buffer_head *old_bh)
3289 {
3290         char *buf = old_bh->b_data;
3291         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3292         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3293         int i;
3294
3295         xs->header = bucket_xh(xs->bucket);
3296         xs->base = bucket_block(xs->bucket, 0);
3297         xs->end = xs->base + inode->i_sb->s_blocksize;
3298
3299         if (xs->not_found)
3300                 return;
3301
3302         i = xs->here - old_xh->xh_entries;
3303         xs->here = &xs->header->xh_entries[i];
3304 }
3305
3306 static int ocfs2_xattr_create_index_block(struct inode *inode,
3307                                           struct ocfs2_xattr_search *xs,
3308                                           struct ocfs2_xattr_set_ctxt *ctxt)
3309 {
3310         int ret;
3311         u32 bit_off, len;
3312         u64 blkno;
3313         handle_t *handle = ctxt->handle;
3314         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3315         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3316         struct buffer_head *xb_bh = xs->xattr_bh;
3317         struct ocfs2_xattr_block *xb =
3318                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3319         struct ocfs2_xattr_tree_root *xr;
3320         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3321
3322         mlog(0, "create xattr index block for %llu\n",
3323              (unsigned long long)xb_bh->b_blocknr);
3324
3325         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3326         BUG_ON(!xs->bucket);
3327
3328         /*
3329          * XXX:
3330          * We can use this lock for now, and maybe move to a dedicated mutex
3331          * if performance becomes a problem later.
3332          */
3333         down_write(&oi->ip_alloc_sem);
3334
3335         ret = ocfs2_journal_access(handle, inode, xb_bh,
3336                                    OCFS2_JOURNAL_ACCESS_WRITE);
3337         if (ret) {
3338                 mlog_errno(ret);
3339                 goto out;
3340         }
3341
3342         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3343                                      1, 1, &bit_off, &len);
3344         if (ret) {
3345                 mlog_errno(ret);
3346                 goto out;
3347         }
3348
3349         /*
3350          * The bucket may spread in many blocks, and
3351          * we will only touch the 1st block and the last block
3352          * in the whole bucket(one for entry and one for data).
3353          */
3354         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3355
3356         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3357              (unsigned long long)blkno);
3358
3359         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3360         if (ret) {
3361                 mlog_errno(ret);
3362                 goto out;
3363         }
3364
3365         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3366                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3367         if (ret) {
3368                 mlog_errno(ret);
3369                 goto out;
3370         }
3371
3372         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3373         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3374
3375         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3376
3377         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3378         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3379                offsetof(struct ocfs2_xattr_block, xb_attrs));
3380
3381         xr = &xb->xb_attrs.xb_root;
3382         xr->xt_clusters = cpu_to_le32(1);
3383         xr->xt_last_eb_blk = 0;
3384         xr->xt_list.l_tree_depth = 0;
3385         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3386         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3387
3388         xr->xt_list.l_recs[0].e_cpos = 0;
3389         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3390         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3391
3392         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3393
3394         ocfs2_journal_dirty(handle, xb_bh);
3395
3396 out:
3397         up_write(&oi->ip_alloc_sem);
3398
3399         return ret;
3400 }
3401
3402 static int cmp_xe_offset(const void *a, const void *b)
3403 {
3404         const struct ocfs2_xattr_entry *l = a, *r = b;
3405         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3406         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3407
3408         if (l_name_offset < r_name_offset)
3409                 return 1;
3410         if (l_name_offset > r_name_offset)
3411                 return -1;
3412         return 0;
3413 }
3414
3415 /*
3416  * defrag a xattr bucket if we find that the bucket has some
3417  * holes beteen name/value pairs.
3418  * We will move all the name/value pairs to the end of the bucket
3419  * so that we can spare some space for insertion.
3420  */
3421 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3422                                      handle_t *handle,
3423                                      struct ocfs2_xattr_bucket *bucket)
3424 {
3425         int ret, i;
3426         size_t end, offset, len, value_len;
3427         struct ocfs2_xattr_header *xh;
3428         char *entries, *buf, *bucket_buf = NULL;
3429         u64 blkno = bucket_blkno(bucket);
3430         u16 xh_free_start;
3431         size_t blocksize = inode->i_sb->s_blocksize;
3432         struct ocfs2_xattr_entry *xe;
3433
3434         /*
3435          * In order to make the operation more efficient and generic,
3436          * we copy all the blocks into a contiguous memory and do the
3437          * defragment there, so if anything is error, we will not touch
3438          * the real block.
3439          */
3440         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3441         if (!bucket_buf) {
3442                 ret = -EIO;
3443                 goto out;
3444         }
3445
3446         buf = bucket_buf;
3447         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3448                 memcpy(buf, bucket_block(bucket, i), blocksize);
3449
3450         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3451                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3452         if (ret < 0) {
3453                 mlog_errno(ret);
3454                 goto out;
3455         }
3456
3457         xh = (struct ocfs2_xattr_header *)bucket_buf;
3458         entries = (char *)xh->xh_entries;
3459         xh_free_start = le16_to_cpu(xh->xh_free_start);
3460
3461         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3462              "xh_free_start = %u, xh_name_value_len = %u.\n",
3463              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3464              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3465
3466         /*
3467          * sort all the entries by their offset.
3468          * the largest will be the first, so that we can
3469          * move them to the end one by one.
3470          */
3471         sort(entries, le16_to_cpu(xh->xh_count),
3472              sizeof(struct ocfs2_xattr_entry),
3473              cmp_xe_offset, swap_xe);
3474
3475         /* Move all name/values to the end of the bucket. */
3476         xe = xh->xh_entries;
3477         end = OCFS2_XATTR_BUCKET_SIZE;
3478         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3479                 offset = le16_to_cpu(xe->xe_name_offset);
3480                 if (ocfs2_xattr_is_local(xe))
3481                         value_len = OCFS2_XATTR_SIZE(
3482                                         le64_to_cpu(xe->xe_value_size));
3483                 else
3484                         value_len = OCFS2_XATTR_ROOT_SIZE;
3485                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3486
3487                 /*
3488                  * We must make sure that the name/value pair
3489                  * exist in the same block. So adjust end to
3490                  * the previous block end if needed.
3491                  */
3492                 if (((end - len) / blocksize !=
3493                         (end - 1) / blocksize))
3494                         end = end - end % blocksize;
3495
3496                 if (end > offset + len) {
3497                         memmove(bucket_buf + end - len,
3498                                 bucket_buf + offset, len);
3499                         xe->xe_name_offset = cpu_to_le16(end - len);
3500                 }
3501
3502                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3503                                 "bucket %llu\n", (unsigned long long)blkno);
3504
3505                 end -= len;
3506         }
3507
3508         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3509                         "bucket %llu\n", (unsigned long long)blkno);
3510
3511         if (xh_free_start == end)
3512                 goto out;
3513
3514         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3515         xh->xh_free_start = cpu_to_le16(end);
3516
3517         /* sort the entries by their name_hash. */
3518         sort(entries, le16_to_cpu(xh->xh_count),
3519              sizeof(struct ocfs2_xattr_entry),
3520              cmp_xe, swap_xe);
3521
3522         buf = bucket_buf;
3523         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3524                 memcpy(bucket_block(bucket, i), buf, blocksize);
3525         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3526
3527 out:
3528         kfree(bucket_buf);
3529         return ret;
3530 }
3531
3532 /*
3533  * prev_blkno points to the start of an existing extent.  new_blkno
3534  * points to a newly allocated extent.  Because we know each of our
3535  * clusters contains more than bucket, we can easily split one cluster
3536  * at a bucket boundary.  So we take the last cluster of the existing
3537  * extent and split it down the middle.  We move the last half of the
3538  * buckets in the last cluster of the existing extent over to the new
3539  * extent.
3540  *
3541  * first_bh is the buffer at prev_blkno so we can update the existing
3542  * extent's bucket count.  header_bh is the bucket were we were hoping
3543  * to insert our xattr.  If the bucket move places the target in the new
3544  * extent, we'll update first_bh and header_bh after modifying the old
3545  * extent.
3546  *
3547  * first_hash will be set as the 1st xe's name_hash in the new extent.
3548  */
3549 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3550                                                handle_t *handle,
3551                                                struct ocfs2_xattr_bucket *first,
3552                                                struct ocfs2_xattr_bucket *target,
3553                                                u64 new_blkno,
3554                                                u32 num_clusters,
3555                                                u32 *first_hash)
3556 {
3557         int ret;
3558         struct super_block *sb = inode->i_sb;
3559         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3560         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3561         int to_move = num_buckets / 2;
3562         u64 src_blkno;
3563         u64 last_cluster_blkno = bucket_blkno(first) +
3564                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3565
3566         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3567         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3568
3569         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3570              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3571
3572         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3573                                      last_cluster_blkno, new_blkno,
3574                                      to_move, first_hash);
3575         if (ret) {
3576                 mlog_errno(ret);
3577                 goto out;
3578         }
3579
3580         /* This is the first bucket that got moved */
3581         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3582
3583         /*
3584          * If the target bucket was part of the moved buckets, we need to
3585          * update first and target.
3586          */
3587         if (bucket_blkno(target) >= src_blkno) {
3588                 /* Find the block for the new target bucket */
3589                 src_blkno = new_blkno +
3590                         (bucket_blkno(target) - src_blkno);
3591
3592                 ocfs2_xattr_bucket_relse(first);
3593                 ocfs2_xattr_bucket_relse(target);
3594
3595                 /*
3596                  * These shouldn't fail - the buffers are in the
3597                  * journal from ocfs2_cp_xattr_bucket().
3598                  */
3599                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3600                 if (ret) {
3601                         mlog_errno(ret);
3602                         goto out;
3603                 }
3604                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3605                 if (ret)
3606                         mlog_errno(ret);
3607
3608         }
3609
3610 out:
3611         return ret;
3612 }
3613
3614 /*
3615  * Find the suitable pos when we divide a bucket into 2.
3616  * We have to make sure the xattrs with the same hash value exist
3617  * in the same bucket.
3618  *
3619  * If this ocfs2_xattr_header covers more than one hash value, find a
3620  * place where the hash value changes.  Try to find the most even split.
3621  * The most common case is that all entries have different hash values,
3622  * and the first check we make will find a place to split.
3623  */
3624 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3625 {
3626         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3627         int count = le16_to_cpu(xh->xh_count);
3628         int delta, middle = count / 2;
3629
3630         /*
3631          * We start at the middle.  Each step gets farther away in both
3632          * directions.  We therefore hit the change in hash value
3633          * nearest to the middle.  Note that this loop does not execute for
3634          * count < 2.
3635          */
3636         for (delta = 0; delta < middle; delta++) {
3637                 /* Let's check delta earlier than middle */
3638                 if (cmp_xe(&entries[middle - delta - 1],
3639                            &entries[middle - delta]))
3640                         return middle - delta;
3641
3642                 /* For even counts, don't walk off the end */
3643                 if ((middle + delta + 1) == count)
3644                         continue;
3645
3646                 /* Now try delta past middle */
3647                 if (cmp_xe(&entries[middle + delta],
3648                            &entries[middle + delta + 1]))
3649                         return middle + delta + 1;
3650         }
3651
3652         /* Every entry had the same hash */
3653         return count;
3654 }
3655
3656 /*
3657  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3658  * first_hash will record the 1st hash of the new bucket.
3659  *
3660  * Normally half of the xattrs will be moved.  But we have to make
3661  * sure that the xattrs with the same hash value are stored in the
3662  * same bucket. If all the xattrs in this bucket have the same hash
3663  * value, the new bucket will be initialized as an empty one and the
3664  * first_hash will be initialized as (hash_value+1).
3665  */
3666 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3667                                     handle_t *handle,
3668                                     u64 blk,
3669                                     u64 new_blk,
3670                                     u32 *first_hash,
3671                                     int new_bucket_head)
3672 {
3673         int ret, i;
3674         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3675         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3676         struct ocfs2_xattr_header *xh;
3677         struct ocfs2_xattr_entry *xe;
3678         int blocksize = inode->i_sb->s_blocksize;
3679
3680         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3681              (unsigned long long)blk, (unsigned long long)new_blk);
3682
3683         s_bucket = ocfs2_xattr_bucket_new(inode);
3684         t_bucket = ocfs2_xattr_bucket_new(inode);
3685         if (!s_bucket || !t_bucket) {
3686                 ret = -ENOMEM;
3687                 mlog_errno(ret);
3688                 goto out;
3689         }
3690
3691         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3692         if (ret) {
3693                 mlog_errno(ret);
3694                 goto out;
3695         }
3696
3697         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3698                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3699         if (ret) {
3700                 mlog_errno(ret);
3701                 goto out;
3702         }
3703
3704         /*
3705          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3706          * there's no need to read it.
3707          */
3708         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3709         if (ret) {
3710                 mlog_errno(ret);
3711                 goto out;
3712         }
3713
3714         /*
3715          * Hey, if we're overwriting t_bucket, what difference does
3716          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3717          * same part of ocfs2_cp_xattr_bucket().
3718          */
3719         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3720                                                 new_bucket_head ?
3721                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3722                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3723         if (ret) {
3724                 mlog_errno(ret);
3725                 goto out;
3726         }
3727
3728         xh = bucket_xh(s_bucket);
3729         count = le16_to_cpu(xh->xh_count);
3730         start = ocfs2_xattr_find_divide_pos(xh);
3731
3732         if (start == count) {
3733                 xe = &xh->xh_entries[start-1];
3734
3735                 /*
3736                  * initialized a new empty bucket here.
3737                  * The hash value is set as one larger than
3738                  * that of the last entry in the previous bucket.
3739                  */
3740                 for (i = 0; i < t_bucket->bu_blocks; i++)
3741                         memset(bucket_block(t_bucket, i), 0, blocksize);
3742
3743                 xh = bucket_xh(t_bucket);
3744                 xh->xh_free_start = cpu_to_le16(blocksize);
3745                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3746                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3747
3748                 goto set_num_buckets;
3749         }
3750
3751         /* copy the whole bucket to the new first. */
3752         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3753
3754         /* update the new bucket. */
3755         xh = bucket_xh(t_bucket);
3756
3757         /*
3758          * Calculate the total name/value len and xh_free_start for
3759          * the old bucket first.
3760          */
3761         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3762         name_value_len = 0;
3763         for (i = 0; i < start; i++) {
3764                 xe = &xh->xh_entries[i];
3765                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3766                 if (ocfs2_xattr_is_local(xe))
3767                         xe_len +=
3768                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3769                 else
3770                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3771                 name_value_len += xe_len;
3772                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3773                         name_offset = le16_to_cpu(xe->xe_name_offset);
3774         }
3775
3776         /*
3777          * Now begin the modification to the new bucket.
3778          *
3779          * In the new bucket, We just move the xattr entry to the beginning
3780          * and don't touch the name/value. So there will be some holes in the
3781          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3782          * called.
3783          */
3784         xe = &xh->xh_entries[start];
3785         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3786         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3787              (int)((char *)xe - (char *)xh),
3788              (int)((char *)xh->xh_entries - (char *)xh));
3789         memmove((char *)xh->xh_entries, (char *)xe, len);
3790         xe = &xh->xh_entries[count - start];
3791         len = sizeof(struct ocfs2_xattr_entry) * start;
3792         memset((char *)xe, 0, len);
3793
3794         le16_add_cpu(&xh->xh_count, -start);
3795         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3796
3797         /* Calculate xh_free_start for the new bucket. */
3798         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3799         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3800                 xe = &xh->xh_entries[i];
3801                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3802                 if (ocfs2_xattr_is_local(xe))
3803                         xe_len +=
3804                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3805                 else
3806                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3807                 if (le16_to_cpu(xe->xe_name_offset) <
3808                     le16_to_cpu(xh->xh_free_start))
3809                         xh->xh_free_start = xe->xe_name_offset;
3810         }
3811
3812 set_num_buckets:
3813         /* set xh->xh_num_buckets for the new xh. */
3814         if (new_bucket_head)
3815                 xh->xh_num_buckets = cpu_to_le16(1);
3816         else
3817                 xh->xh_num_buckets = 0;
3818
3819         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3820
3821         /* store the first_hash of the new bucket. */
3822         if (first_hash)
3823                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3824
3825         /*
3826          * Now only update the 1st block of the old bucket.  If we
3827          * just added a new empty bucket, there is no need to modify
3828          * it.
3829          */
3830         if (start == count)
3831                 goto out;
3832
3833         xh = bucket_xh(s_bucket);
3834         memset(&xh->xh_entries[start], 0,
3835                sizeof(struct ocfs2_xattr_entry) * (count - start));
3836         xh->xh_count = cpu_to_le16(start);
3837         xh->xh_free_start = cpu_to_le16(name_offset);
3838         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3839
3840         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3841
3842 out:
3843         ocfs2_xattr_bucket_free(s_bucket);
3844         ocfs2_xattr_bucket_free(t_bucket);
3845
3846         return ret;
3847 }
3848
3849 /*
3850  * Copy xattr from one bucket to another bucket.
3851  *
3852  * The caller must make sure that the journal transaction
3853  * has enough space for journaling.
3854  */
3855 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3856                                  handle_t *handle,
3857                                  u64 s_blkno,
3858                                  u64 t_blkno,
3859                                  int t_is_new)
3860 {
3861         int ret;
3862         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3863
3864         BUG_ON(s_blkno == t_blkno);
3865
3866         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3867              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3868              t_is_new);
3869
3870         s_bucket = ocfs2_xattr_bucket_new(inode);
3871         t_bucket = ocfs2_xattr_bucket_new(inode);
3872         if (!s_bucket || !t_bucket) {
3873                 ret = -ENOMEM;
3874                 mlog_errno(ret);
3875                 goto out;
3876         }
3877
3878         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3879         if (ret)
3880                 goto out;
3881
3882         /*
3883          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3884          * there's no need to read it.
3885          */
3886         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3887         if (ret)
3888                 goto out;
3889
3890         /*
3891          * Hey, if we're overwriting t_bucket, what difference does
3892          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3893          * cluster to fill, we came here from
3894          * ocfs2_mv_xattr_buckets(), and it is really new -
3895          * ACCESS_CREATE is required.  But we also might have moved data
3896          * out of t_bucket before extending back into it.
3897          * ocfs2_add_new_xattr_bucket() can do this - its call to
3898          * ocfs2_add_new_xattr_cluster() may have created a new extent
3899          * and copied out the end of the old extent.  Then it re-extends
3900          * the old extent back to create space for new xattrs.  That's
3901          * how we get here, and the bucket isn't really new.
3902          */
3903         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3904                                                 t_is_new ?
3905                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3906                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3907         if (ret)
3908                 goto out;
3909
3910         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3911         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3912
3913 out:
3914         ocfs2_xattr_bucket_free(t_bucket);
3915         ocfs2_xattr_bucket_free(s_bucket);
3916
3917         return ret;
3918 }
3919
3920 /*
3921  * src_blk points to the start of an existing extent.  last_blk points to
3922  * last cluster in that extent.  to_blk points to a newly allocated
3923  * extent.  We copy the buckets from the cluster at last_blk to the new
3924  * extent.  If start_bucket is non-zero, we skip that many buckets before
3925  * we start copying.  The new extent's xh_num_buckets gets set to the
3926  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
3927  * by the same amount.
3928  */
3929 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
3930                                   u64 src_blk, u64 last_blk, u64 to_blk,
3931                                   unsigned int start_bucket,
3932                                   u32 *first_hash)
3933 {
3934         int i, ret, credits;
3935         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3936         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3937         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3938         struct ocfs2_xattr_bucket *old_first, *new_first;
3939
3940         mlog(0, "mv xattrs from cluster %llu to %llu\n",
3941              (unsigned long long)last_blk, (unsigned long long)to_blk);
3942
3943         BUG_ON(start_bucket >= num_buckets);
3944         if (start_bucket) {
3945                 num_buckets -= start_bucket;
3946                 last_blk += (start_bucket * blks_per_bucket);
3947         }
3948
3949         /* The first bucket of the original extent */
3950         old_first = ocfs2_xattr_bucket_new(inode);
3951         /* The first bucket of the new extent */
3952         new_first = ocfs2_xattr_bucket_new(inode);
3953         if (!old_first || !new_first) {
3954                 ret = -ENOMEM;
3955                 mlog_errno(ret);
3956                 goto out;
3957         }
3958
3959         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
3960         if (ret) {
3961                 mlog_errno(ret);
3962                 goto out;
3963         }
3964
3965         /*
3966          * We need to update the first bucket of the old extent and all
3967          * the buckets going to the new extent.
3968          */
3969         credits = ((num_buckets + 1) * blks_per_bucket) +
3970                 handle->h_buffer_credits;
3971         ret = ocfs2_extend_trans(handle, credits);
3972         if (ret) {
3973                 mlog_errno(ret);
3974                 goto out;
3975         }
3976
3977         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
3978                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3979         if (ret) {
3980                 mlog_errno(ret);
3981                 goto out;
3982         }
3983
3984         for (i = 0; i < num_buckets; i++) {
3985                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3986                                             last_blk + (i * blks_per_bucket),
3987                                             to_blk + (i * blks_per_bucket),
3988                                             1);
3989                 if (ret) {
3990                         mlog_errno(ret);
3991                         goto out;
3992                 }
3993         }
3994
3995         /*
3996          * Get the new bucket ready before we dirty anything
3997          * (This actually shouldn't fail, because we already dirtied
3998          * it once in ocfs2_cp_xattr_bucket()).
3999          */
4000         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4001         if (ret) {
4002                 mlog_errno(ret);
4003                 goto out;
4004         }
4005         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4006                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4007         if (ret) {
4008                 mlog_errno(ret);
4009                 goto out;
4010         }
4011
4012         /* Now update the headers */
4013         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4014         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4015
4016         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4017         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4018
4019         if (first_hash)
4020                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4021
4022 out:
4023         ocfs2_xattr_bucket_free(new_first);
4024         ocfs2_xattr_bucket_free(old_first);
4025         return ret;
4026 }
4027
4028 /*
4029  * Move some xattrs in this cluster to the new cluster.
4030  * This function should only be called when bucket size == cluster size.
4031  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4032  */
4033 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4034                                       handle_t *handle,
4035                                       u64 prev_blk,
4036                                       u64 new_blk,
4037                                       u32 *first_hash)
4038 {
4039         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4040         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4041
4042         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4043
4044         ret = ocfs2_extend_trans(handle, credits);
4045         if (ret) {
4046                 mlog_errno(ret);
4047                 return ret;
4048         }
4049
4050         /* Move half of the xattr in start_blk to the next bucket. */
4051         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4052                                           new_blk, first_hash, 1);
4053 }
4054
4055 /*
4056  * Move some xattrs from the old cluster to the new one since they are not
4057  * contiguous in ocfs2 xattr tree.
4058  *
4059  * new_blk starts a new separate cluster, and we will move some xattrs from
4060  * prev_blk to it. v_start will be set as the first name hash value in this
4061  * new cluster so that it can be used as e_cpos during tree insertion and
4062  * don't collide with our original b-tree operations. first_bh and header_bh
4063  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4064  * to extend the insert bucket.
4065  *
4066  * The problem is how much xattr should we move to the new one and when should
4067  * we update first_bh and header_bh?
4068  * 1. If cluster size > bucket size, that means the previous cluster has more
4069  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4070  *    update the first_bh and header_bh if the insert bucket has been moved
4071  *    to the new cluster.
4072  * 2. If cluster_size == bucket_size:
4073  *    a) If the previous extent rec has more than one cluster and the insert
4074  *       place isn't in the last cluster, copy the entire last cluster to the
4075  *       new one. This time, we don't need to upate the first_bh and header_bh
4076  *       since they will not be moved into the new cluster.
4077  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4078  *       the new one. And we set the extend flag to zero if the insert place is
4079  *       moved into the new allocated cluster since no extend is needed.
4080  */
4081 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4082                                             handle_t *handle,
4083                                             struct ocfs2_xattr_bucket *first,
4084                                             struct ocfs2_xattr_bucket *target,
4085                                             u64 new_blk,
4086                                             u32 prev_clusters,
4087                                             u32 *v_start,
4088                                             int *extend)
4089 {
4090         int ret;
4091
4092         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4093              (unsigned long long)bucket_blkno(first), prev_clusters,
4094              (unsigned long long)new_blk);
4095
4096         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4097                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4098                                                           handle,
4099                                                           first, target,
4100                                                           new_blk,
4101                                                           prev_clusters,
4102                                                           v_start);
4103                 if (ret)
4104                         mlog_errno(ret);
4105         } else {
4106                 /* The start of the last cluster in the first extent */
4107                 u64 last_blk = bucket_blkno(first) +
4108                         ((prev_clusters - 1) *
4109                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4110
4111                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4112                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4113                                                      bucket_blkno(first),
4114                                                      last_blk, new_blk, 0,
4115                                                      v_start);
4116                         if (ret)
4117                                 mlog_errno(ret);
4118                 } else {
4119                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4120                                                          last_blk, new_blk,
4121                                                          v_start);
4122                         if (ret)
4123                                 mlog_errno(ret);
4124
4125                         if ((bucket_blkno(target) == last_blk) && extend)
4126                                 *extend = 0;
4127                 }
4128         }
4129
4130         return ret;
4131 }
4132
4133 /*
4134  * Add a new cluster for xattr storage.
4135  *
4136  * If the new cluster is contiguous with the previous one, it will be
4137  * appended to the same extent record, and num_clusters will be updated.
4138  * If not, we will insert a new extent for it and move some xattrs in
4139  * the last cluster into the new allocated one.
4140  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4141  * lose the benefits of hashing because we'll have to search large leaves.
4142  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4143  * if it's bigger).
4144  *
4145  * first_bh is the first block of the previous extent rec and header_bh
4146  * indicates the bucket we will insert the new xattrs. They will be updated
4147  * when the header_bh is moved into the new cluster.
4148  */
4149 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4150                                        struct buffer_head *root_bh,
4151                                        struct buffer_head **first_bh,
4152                                        struct buffer_head **header_bh,
4153                                        u32 *num_clusters,
4154                                        u32 prev_cpos,
4155                                        u64 prev_blkno,
4156                                        int *extend,
4157                                        struct ocfs2_xattr_set_ctxt *ctxt)
4158 {
4159         int ret;
4160         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4161         u32 prev_clusters = *num_clusters;
4162         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4163         u64 block;
4164         handle_t *handle = ctxt->handle;
4165         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4166         struct ocfs2_extent_tree et;
4167         struct ocfs2_xattr_bucket *first, *target;
4168
4169         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4170              "previous xattr blkno = %llu\n",
4171              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4172              prev_cpos, (unsigned long long)prev_blkno);
4173
4174         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4175
4176         /* The first bucket of the original extent */
4177         first = ocfs2_xattr_bucket_new(inode);
4178         /* The target bucket for insert */
4179         target = ocfs2_xattr_bucket_new(inode);
4180         if (!first || !target) {
4181                 ret = -ENOMEM;
4182                 mlog_errno(ret);
4183                 goto leave;
4184         }
4185
4186         BUG_ON(prev_blkno != (*first_bh)->b_blocknr);
4187         ret = ocfs2_read_xattr_bucket(first, prev_blkno);
4188         if (ret) {
4189                 mlog_errno(ret);
4190                 goto leave;
4191         }
4192
4193         ret = ocfs2_read_xattr_bucket(target, (*header_bh)->b_blocknr);
4194         if (ret) {
4195                 mlog_errno(ret);
4196                 goto leave;
4197         }
4198
4199         ret = ocfs2_journal_access(handle, inode, root_bh,
4200                                    OCFS2_JOURNAL_ACCESS_WRITE);
4201         if (ret < 0) {
4202                 mlog_errno(ret);
4203                 goto leave;
4204         }
4205
4206         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4207                                      clusters_to_add, &bit_off, &num_bits);
4208         if (ret < 0) {
4209                 if (ret != -ENOSPC)
4210                         mlog_errno(ret);
4211                 goto leave;
4212         }
4213
4214         BUG_ON(num_bits > clusters_to_add);
4215
4216         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4217         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4218              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4219
4220         if (prev_blkno + prev_clusters * bpc == block &&
4221             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4222              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4223                 /*
4224                  * If this cluster is contiguous with the old one and
4225                  * adding this new cluster, we don't surpass the limit of
4226                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4227                  * initialized and used like other buckets in the previous
4228                  * cluster.
4229                  * So add it as a contiguous one. The caller will handle
4230                  * its init process.
4231                  */
4232                 v_start = prev_cpos + prev_clusters;
4233                 *num_clusters = prev_clusters + num_bits;
4234                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4235                      num_bits);
4236         } else {
4237                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4238                                                        handle,
4239                                                        first,
4240                                                        target,
4241                                                        block,
4242                                                        prev_clusters,
4243                                                        &v_start,
4244                                                        extend);
4245                 if (ret) {
4246                         mlog_errno(ret);
4247                         goto leave;
4248                 }
4249
4250                 /* Did first+target get moved? */
4251                 if (prev_blkno != bucket_blkno(first)) {
4252                         brelse(*first_bh);
4253                         *first_bh = first->bu_bhs[0];
4254                         get_bh(*first_bh);
4255
4256                         brelse(*header_bh);
4257                         *header_bh = target->bu_bhs[0];
4258                         get_bh(*header_bh);
4259                 }
4260         }
4261
4262         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4263              num_bits, (unsigned long long)block, v_start);
4264         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4265                                   num_bits, 0, ctxt->meta_ac);
4266         if (ret < 0) {
4267                 mlog_errno(ret);
4268                 goto leave;
4269         }
4270
4271         ret = ocfs2_journal_dirty(handle, root_bh);
4272         if (ret < 0)
4273                 mlog_errno(ret);
4274
4275 leave:
4276         ocfs2_xattr_bucket_free(first);
4277         ocfs2_xattr_bucket_free(target);
4278         return ret;
4279 }
4280
4281 /*
4282  * We are given an extent.  'first' is the bucket at the very front of
4283  * the extent.  The extent has space for an additional bucket past
4284  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4285  * of the target bucket.  We wish to shift every bucket past the target
4286  * down one, filling in that additional space.  When we get back to the
4287  * target, we split the target between itself and the now-empty bucket
4288  * at target+1 (aka, target_blkno + blks_per_bucket).
4289  */
4290 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4291                                      handle_t *handle,
4292                                      struct ocfs2_xattr_bucket *first,
4293                                      u64 target_blk,
4294                                      u32 num_clusters)
4295 {
4296         int ret, credits;
4297         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4298         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4299         u64 end_blk;
4300         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4301
4302         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4303              "from %llu, len = %u\n", (unsigned long long)target_blk,
4304              (unsigned long long)bucket_blkno(first), num_clusters);
4305
4306         /* The extent must have room for an additional bucket */
4307         BUG_ON(new_bucket >=
4308                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4309
4310         /* end_blk points to the last existing bucket */
4311         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4312
4313         /*
4314          * end_blk is the start of the last existing bucket.
4315          * Thus, (end_blk - target_blk) covers the target bucket and
4316          * every bucket after it up to, but not including, the last
4317          * existing bucket.  Then we add the last existing bucket, the
4318          * new bucket, and the first bucket (3 * blk_per_bucket).
4319          */
4320         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4321                   handle->h_buffer_credits;
4322         ret = ocfs2_extend_trans(handle, credits);
4323         if (ret) {
4324                 mlog_errno(ret);
4325                 goto out;
4326         }
4327
4328         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4329                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4330         if (ret) {
4331                 mlog_errno(ret);
4332                 goto out;
4333         }
4334
4335         while (end_blk != target_blk) {
4336                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4337                                             end_blk + blk_per_bucket, 0);
4338                 if (ret)
4339                         goto out;
4340                 end_blk -= blk_per_bucket;
4341         }
4342
4343         /* Move half of the xattr in target_blkno to the next bucket. */
4344         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4345                                         target_blk + blk_per_bucket, NULL, 0);
4346
4347         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4348         ocfs2_xattr_bucket_journal_dirty(handle, first);
4349
4350 out:
4351         return ret;
4352 }
4353
4354 /*
4355  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
4356  * xb_bh is the ocfs2_xattr_block.
4357  * We will move all the buckets starting from header_bh to the next place. As
4358  * for this one, half num of its xattrs will be moved to the next one.
4359  *
4360  * We will allocate a new cluster if current cluster is full and adjust
4361  * header_bh and first_bh if the insert place is moved to the new cluster.
4362  */
4363 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4364                                       struct buffer_head *xb_bh,
4365                                       struct buffer_head *header_bh,
4366                                       struct ocfs2_xattr_set_ctxt *ctxt)
4367 {
4368         struct ocfs2_xattr_header *first_xh = NULL;
4369         struct buffer_head *first_bh = NULL;
4370         struct ocfs2_xattr_block *xb =
4371                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4372         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4373         struct ocfs2_extent_list *el = &xb_root->xt_list;
4374         struct ocfs2_xattr_header *xh =
4375                         (struct ocfs2_xattr_header *)header_bh->b_data;
4376         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4377         struct super_block *sb = inode->i_sb;
4378         struct ocfs2_super *osb = OCFS2_SB(sb);
4379         int ret, num_buckets, extend = 1;
4380         u64 p_blkno;
4381         u32 e_cpos, num_clusters;
4382         /* The bucket at the front of the extent */
4383         struct ocfs2_xattr_bucket *first;
4384
4385         mlog(0, "Add new xattr bucket starting form %llu\n",
4386              (unsigned long long)header_bh->b_blocknr);
4387
4388         first = ocfs2_xattr_bucket_new(inode);
4389         if (!first) {
4390                 ret = -ENOMEM;
4391                 mlog_errno(ret);
4392                 goto out;
4393         }
4394
4395         /*
4396          * Add refrence for header_bh here because it may be
4397          * changed in ocfs2_add_new_xattr_cluster and we need
4398          * to free it in the end.
4399          */
4400         get_bh(header_bh);
4401
4402         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4403                                   &num_clusters, el);
4404         if (ret) {
4405                 mlog_errno(ret);
4406                 goto out;
4407         }
4408
4409         ret = ocfs2_read_block(inode, p_blkno, &first_bh, NULL);
4410         if (ret) {
4411                 mlog_errno(ret);
4412                 goto out;
4413         }
4414
4415         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4416         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4417
4418         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
4419                 ret = ocfs2_add_new_xattr_cluster(inode,
4420                                                   xb_bh,
4421                                                   &first_bh,
4422                                                   &header_bh,
4423                                                   &num_clusters,
4424                                                   e_cpos,
4425                                                   p_blkno,
4426                                                   &extend,
4427                                                   ctxt);
4428                 if (ret) {
4429                         mlog_errno(ret);
4430                         goto out;
4431                 }
4432         }
4433
4434         if (extend) {
4435                 /* These bucket reads should be cached */
4436                 ret = ocfs2_read_xattr_bucket(first, first_bh->b_blocknr);
4437                 if (ret) {
4438                         mlog_errno(ret);
4439                         goto out;
4440                 }
4441                 ret = ocfs2_extend_xattr_bucket(inode,
4442                                                 ctxt->handle,
4443                                                 first, header_bh->b_blocknr,
4444                                                 num_clusters);
4445                 if (ret)
4446                         mlog_errno(ret);
4447         }
4448
4449 out:
4450         brelse(first_bh);
4451         brelse(header_bh);
4452         ocfs2_xattr_bucket_free(first);
4453         return ret;
4454 }
4455
4456 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4457                                         struct ocfs2_xattr_bucket *bucket,
4458                                         int offs)
4459 {
4460         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4461
4462         offs = offs % inode->i_sb->s_blocksize;
4463         return bucket_block(bucket, block_off) + offs;
4464 }
4465
4466 /*
4467  * Handle the normal xattr set, including replace, delete and new.
4468  *
4469  * Note: "local" indicates the real data's locality. So we can't
4470  * just its bucket locality by its length.
4471  */
4472 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4473                                          struct ocfs2_xattr_info *xi,
4474                                          struct ocfs2_xattr_search *xs,
4475                                          u32 name_hash,
4476                                          int local)
4477 {
4478         struct ocfs2_xattr_entry *last, *xe;
4479         int name_len = strlen(xi->name);
4480         struct ocfs2_xattr_header *xh = xs->header;
4481         u16 count = le16_to_cpu(xh->xh_count), start;
4482         size_t blocksize = inode->i_sb->s_blocksize;
4483         char *val;
4484         size_t offs, size, new_size;
4485
4486         last = &xh->xh_entries[count];
4487         if (!xs->not_found) {
4488                 xe = xs->here;
4489                 offs = le16_to_cpu(xe->xe_name_offset);
4490                 if (ocfs2_xattr_is_local(xe))
4491                         size = OCFS2_XATTR_SIZE(name_len) +
4492                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4493                 else
4494                         size = OCFS2_XATTR_SIZE(name_len) +
4495                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4496
4497                 /*
4498                  * If the new value will be stored outside, xi->value has been
4499                  * initalized as an empty ocfs2_xattr_value_root, and the same
4500                  * goes with xi->value_len, so we can set new_size safely here.
4501                  * See ocfs2_xattr_set_in_bucket.
4502                  */
4503                 new_size = OCFS2_XATTR_SIZE(name_len) +
4504                            OCFS2_XATTR_SIZE(xi->value_len);
4505
4506                 le16_add_cpu(&xh->xh_name_value_len, -size);
4507                 if (xi->value) {
4508                         if (new_size > size)
4509                                 goto set_new_name_value;
4510
4511                         /* Now replace the old value with new one. */
4512                         if (local)
4513                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4514                         else
4515                                 xe->xe_value_size = 0;
4516
4517                         val = ocfs2_xattr_bucket_get_val(inode,
4518                                                          xs->bucket, offs);
4519                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4520                                size - OCFS2_XATTR_SIZE(name_len));
4521                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4522                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4523                                        xi->value, xi->value_len);
4524
4525                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4526                         ocfs2_xattr_set_local(xe, local);
4527                         return;
4528                 } else {
4529                         /*
4530                          * Remove the old entry if there is more than one.
4531                          * We don't remove the last entry so that we can
4532                          * use it to indicate the hash value of the empty
4533                          * bucket.
4534                          */
4535                         last -= 1;
4536                         le16_add_cpu(&xh->xh_count, -1);
4537                         if (xh->xh_count) {
4538                                 memmove(xe, xe + 1,
4539                                         (void *)last - (void *)xe);
4540                                 memset(last, 0,
4541                                        sizeof(struct ocfs2_xattr_entry));
4542                         } else
4543                                 xh->xh_free_start =
4544                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4545
4546                         return;
4547                 }
4548         } else {
4549                 /* find a new entry for insert. */
4550                 int low = 0, high = count - 1, tmp;
4551                 struct ocfs2_xattr_entry *tmp_xe;
4552
4553                 while (low <= high && count) {
4554                         tmp = (low + high) / 2;
4555                         tmp_xe = &xh->xh_entries[tmp];
4556
4557                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4558                                 low = tmp + 1;
4559                         else if (name_hash <
4560                                  le32_to_cpu(tmp_xe->xe_name_hash))
4561                                 high = tmp - 1;
4562                         else {
4563                                 low = tmp;
4564                                 break;
4565                         }
4566                 }
4567
4568                 xe = &xh->xh_entries[low];
4569                 if (low != count)
4570                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4571
4572                 le16_add_cpu(&xh->xh_count, 1);
4573                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4574                 xe->xe_name_hash = cpu_to_le32(name_hash);
4575                 xe->xe_name_len = name_len;
4576                 ocfs2_xattr_set_type(xe, xi->name_index);
4577         }
4578
4579 set_new_name_value:
4580         /* Insert the new name+value. */
4581         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4582
4583         /*
4584          * We must make sure that the name/value pair
4585          * exists in the same block.
4586          */
4587         offs = le16_to_cpu(xh->xh_free_start);
4588         start = offs - size;
4589
4590         if (start >> inode->i_sb->s_blocksize_bits !=
4591             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4592                 offs = offs - offs % blocksize;
4593                 xh->xh_free_start = cpu_to_le16(offs);
4594         }
4595
4596         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4597         xe->xe_name_offset = cpu_to_le16(offs - size);
4598
4599         memset(val, 0, size);
4600         memcpy(val, xi->name, name_len);
4601         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4602
4603         xe->xe_value_size = cpu_to_le64(xi->value_len);
4604         ocfs2_xattr_set_local(xe, local);
4605         xs->here = xe;
4606         le16_add_cpu(&xh->xh_free_start, -size);
4607         le16_add_cpu(&xh->xh_name_value_len, size);
4608
4609         return;
4610 }
4611
4612 /*
4613  * Set the xattr entry in the specified bucket.
4614  * The bucket is indicated by xs->bucket and it should have the enough
4615  * space for the xattr insertion.
4616  */
4617 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4618                                            handle_t *handle,
4619                                            struct ocfs2_xattr_info *xi,
4620                                            struct ocfs2_xattr_search *xs,
4621                                            u32 name_hash,
4622                                            int local)
4623 {
4624         int ret;
4625         u64 blkno;
4626
4627         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4628              (unsigned long)xi->value_len, xi->name_index,
4629              (unsigned long long)bucket_blkno(xs->bucket));
4630
4631         if (!xs->bucket->bu_bhs[1]) {
4632                 blkno = bucket_blkno(xs->bucket);
4633                 ocfs2_xattr_bucket_relse(xs->bucket);
4634                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4635                 if (ret) {
4636                         mlog_errno(ret);
4637                         goto out;
4638                 }
4639         }
4640
4641         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4642                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4643         if (ret < 0) {
4644                 mlog_errno(ret);
4645                 goto out;
4646         }
4647
4648         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4649         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4650
4651 out:
4652         return ret;
4653 }
4654
4655 /*
4656  * Truncate the specified xe_off entry in xattr bucket.
4657  * bucket is indicated by header_bh and len is the new length.
4658  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4659  *
4660  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4661  */
4662 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4663                                              struct ocfs2_xattr_bucket *bucket,
4664                                              int xe_off,
4665                                              int len,
4666                                              struct ocfs2_xattr_set_ctxt *ctxt)
4667 {
4668         int ret, offset;
4669         u64 value_blk;
4670         struct buffer_head *value_bh = NULL;
4671         struct ocfs2_xattr_value_root *xv;
4672         struct ocfs2_xattr_entry *xe;
4673         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4674         size_t blocksize = inode->i_sb->s_blocksize;
4675
4676         xe = &xh->xh_entries[xe_off];
4677
4678         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4679
4680         offset = le16_to_cpu(xe->xe_name_offset) +
4681                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4682
4683         value_blk = offset / blocksize;
4684
4685         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4686         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4687
4688         value_bh = bucket->bu_bhs[value_blk];
4689         BUG_ON(!value_bh);
4690
4691         xv = (struct ocfs2_xattr_value_root *)
4692                 (value_bh->b_data + offset % blocksize);
4693
4694         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4695                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4696         if (ret) {
4697                 mlog_errno(ret);
4698                 goto out;
4699         }
4700
4701         /*
4702          * From here on out we have to dirty the bucket.  The generic
4703          * value calls only modify one of the bucket's bhs, but we need
4704          * to send the bucket at once.  So if they error, they *could* have
4705          * modified something.  We have to assume they did, and dirty
4706          * the whole bucket.  This leaves us in a consistent state.
4707          */
4708         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4709              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4710         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len, ctxt);
4711         if (ret) {
4712                 mlog_errno(ret);
4713                 goto out_dirty;
4714         }
4715
4716         xe->xe_value_size = cpu_to_le64(len);
4717
4718 out_dirty:
4719         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4720
4721 out:
4722         return ret;
4723 }
4724
4725 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4726                                         struct ocfs2_xattr_search *xs,
4727                                         int len,
4728                                         struct ocfs2_xattr_set_ctxt *ctxt)
4729 {
4730         int ret, offset;
4731         struct ocfs2_xattr_entry *xe = xs->here;
4732         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4733
4734         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4735
4736         offset = xe - xh->xh_entries;
4737         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4738                                                 offset, len, ctxt);
4739         if (ret)
4740                 mlog_errno(ret);
4741
4742         return ret;
4743 }
4744
4745 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4746                                                 handle_t *handle,
4747                                                 struct ocfs2_xattr_search *xs,
4748                                                 char *val,
4749                                                 int value_len)
4750 {
4751         int offset;
4752         struct ocfs2_xattr_value_root *xv;
4753         struct ocfs2_xattr_entry *xe = xs->here;
4754
4755         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4756
4757         offset = le16_to_cpu(xe->xe_name_offset) +
4758                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4759
4760         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4761
4762         return __ocfs2_xattr_set_value_outside(inode, handle,
4763                                                xv, val, value_len);
4764 }
4765
4766 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4767                                   struct buffer_head *root_bh,
4768                                   u64 blkno,
4769                                   u32 cpos,
4770                                   u32 len)
4771 {
4772         int ret;
4773         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4774         struct inode *tl_inode = osb->osb_tl_inode;
4775         handle_t *handle;
4776         struct ocfs2_xattr_block *xb =
4777                         (struct ocfs2_xattr_block *)root_bh->b_data;
4778         struct ocfs2_alloc_context *meta_ac = NULL;
4779         struct ocfs2_cached_dealloc_ctxt dealloc;
4780         struct ocfs2_extent_tree et;
4781
4782         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4783
4784         ocfs2_init_dealloc_ctxt(&dealloc);
4785
4786         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4787              cpos, len, (unsigned long long)blkno);
4788
4789         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4790
4791         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4792         if (ret) {
4793                 mlog_errno(ret);
4794                 return ret;
4795         }
4796
4797         mutex_lock(&tl_inode->i_mutex);
4798
4799         if (ocfs2_truncate_log_needs_flush(osb)) {
4800                 ret = __ocfs2_flush_truncate_log(osb);
4801                 if (ret < 0) {
4802                         mlog_errno(ret);
4803                         goto out;
4804                 }
4805         }
4806
4807         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4808         if (IS_ERR(handle)) {
4809                 ret = -ENOMEM;
4810                 mlog_errno(ret);
4811                 goto out;
4812         }
4813
4814         ret = ocfs2_journal_access(handle, inode, root_bh,
4815                                    OCFS2_JOURNAL_ACCESS_WRITE);
4816         if (ret) {
4817                 mlog_errno(ret);
4818                 goto out_commit;
4819         }
4820
4821         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4822                                   &dealloc);
4823         if (ret) {
4824                 mlog_errno(ret);
4825                 goto out_commit;
4826         }
4827
4828         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4829
4830         ret = ocfs2_journal_dirty(handle, root_bh);
4831         if (ret) {
4832                 mlog_errno(ret);
4833                 goto out_commit;
4834         }
4835
4836         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4837         if (ret)
4838                 mlog_errno(ret);
4839
4840 out_commit:
4841         ocfs2_commit_trans(osb, handle);
4842 out:
4843         ocfs2_schedule_truncate_log_flush(osb, 1);
4844
4845         mutex_unlock(&tl_inode->i_mutex);
4846
4847         if (meta_ac)
4848                 ocfs2_free_alloc_context(meta_ac);
4849
4850         ocfs2_run_deallocs(osb, &dealloc);
4851
4852         return ret;
4853 }
4854
4855 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4856                                          handle_t *handle,
4857                                          struct ocfs2_xattr_search *xs)
4858 {
4859         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4860         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4861                                                 le16_to_cpu(xh->xh_count) - 1];
4862         int ret = 0;
4863
4864         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4865                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4866         if (ret) {
4867                 mlog_errno(ret);
4868                 return;
4869         }
4870
4871         /* Remove the old entry. */
4872         memmove(xs->here, xs->here + 1,
4873                 (void *)last - (void *)xs->here);
4874         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4875         le16_add_cpu(&xh->xh_count, -1);
4876
4877         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4878 }
4879
4880 /*
4881  * Set the xattr name/value in the bucket specified in xs.
4882  *
4883  * As the new value in xi may be stored in the bucket or in an outside cluster,
4884  * we divide the whole process into 3 steps:
4885  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4886  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4887  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4888  * 4. If the clusters for the new outside value can't be allocated, we need
4889  *    to free the xattr we allocated in set.
4890  */
4891 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4892                                      struct ocfs2_xattr_info *xi,
4893                                      struct ocfs2_xattr_search *xs,
4894                                      struct ocfs2_xattr_set_ctxt *ctxt)
4895 {
4896         int ret, local = 1;
4897         size_t value_len;
4898         char *val = (char *)xi->value;
4899         struct ocfs2_xattr_entry *xe = xs->here;
4900         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4901                                               strlen(xi->name));
4902
4903         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4904                 /*
4905                  * We need to truncate the xattr storage first.
4906                  *
4907                  * If both the old and new value are stored to
4908                  * outside block, we only need to truncate
4909                  * the storage and then set the value outside.
4910                  *
4911                  * If the new value should be stored within block,
4912                  * we should free all the outside block first and
4913                  * the modification to the xattr block will be done
4914                  * by following steps.
4915                  */
4916                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4917                         value_len = xi->value_len;
4918                 else
4919                         value_len = 0;
4920
4921                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4922                                                            value_len,
4923                                                            ctxt);
4924                 if (ret)
4925                         goto out;
4926
4927                 if (value_len)
4928                         goto set_value_outside;
4929         }
4930
4931         value_len = xi->value_len;
4932         /* So we have to handle the inside block change now. */
4933         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4934                 /*
4935                  * If the new value will be stored outside of block,
4936                  * initalize a new empty value root and insert it first.
4937                  */
4938                 local = 0;
4939                 xi->value = &def_xv;
4940                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4941         }
4942
4943         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4944                                               name_hash, local);
4945         if (ret) {
4946                 mlog_errno(ret);
4947                 goto out;
4948         }
4949
4950         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4951                 goto out;
4952
4953         /* allocate the space now for the outside block storage. */
4954         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4955                                                    value_len, ctxt);
4956         if (ret) {
4957                 mlog_errno(ret);
4958
4959                 if (xs->not_found) {
4960                         /*
4961                          * We can't allocate enough clusters for outside
4962                          * storage and we have allocated xattr already,
4963                          * so need to remove it.
4964                          */
4965                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4966                 }
4967                 goto out;
4968         }
4969
4970 set_value_outside:
4971         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4972                                                    xs, val, value_len);
4973 out:
4974         return ret;
4975 }
4976
4977 /*
4978  * check whether the xattr bucket is filled up with the same hash value.
4979  * If we want to insert the xattr with the same hash, return -ENOSPC.
4980  * If we want to insert a xattr with different hash value, go ahead
4981  * and ocfs2_divide_xattr_bucket will handle this.
4982  */
4983 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4984                                               struct ocfs2_xattr_bucket *bucket,
4985                                               const char *name)
4986 {
4987         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4988         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4989
4990         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4991                 return 0;
4992
4993         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4994             xh->xh_entries[0].xe_name_hash) {
4995                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4996                      "hash = %u\n",
4997                      (unsigned long long)bucket_blkno(bucket),
4998                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4999                 return -ENOSPC;
5000         }
5001
5002         return 0;
5003 }
5004
5005 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5006                                              struct ocfs2_xattr_info *xi,
5007                                              struct ocfs2_xattr_search *xs,
5008                                              struct ocfs2_xattr_set_ctxt *ctxt)
5009 {
5010         struct ocfs2_xattr_header *xh;
5011         struct ocfs2_xattr_entry *xe;
5012         u16 count, header_size, xh_free_start;
5013         int free, max_free, need, old;
5014         size_t value_size = 0, name_len = strlen(xi->name);
5015         size_t blocksize = inode->i_sb->s_blocksize;
5016         int ret, allocation = 0;
5017
5018         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5019
5020 try_again:
5021         xh = xs->header;
5022         count = le16_to_cpu(xh->xh_count);
5023         xh_free_start = le16_to_cpu(xh->xh_free_start);
5024         header_size = sizeof(struct ocfs2_xattr_header) +
5025                         count * sizeof(struct ocfs2_xattr_entry);
5026         max_free = OCFS2_XATTR_BUCKET_SIZE -
5027                 le16_to_cpu(xh->xh_name_value_len) - header_size;
5028
5029         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5030                         "of %u which exceed block size\n",
5031                         (unsigned long long)bucket_blkno(xs->bucket),
5032                         header_size);
5033
5034         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5035                 value_size = OCFS2_XATTR_ROOT_SIZE;
5036         else if (xi->value)
5037                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5038
5039         if (xs->not_found)
5040                 need = sizeof(struct ocfs2_xattr_entry) +
5041                         OCFS2_XATTR_SIZE(name_len) + value_size;
5042         else {
5043                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5044
5045                 /*
5046                  * We only replace the old value if the new length is smaller
5047                  * than the old one. Otherwise we will allocate new space in the
5048                  * bucket to store it.
5049                  */
5050                 xe = xs->here;
5051                 if (ocfs2_xattr_is_local(xe))
5052                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5053                 else
5054                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5055
5056                 if (old >= value_size)
5057                         need = 0;
5058         }
5059
5060         free = xh_free_start - header_size;
5061         /*
5062          * We need to make sure the new name/value pair
5063          * can exist in the same block.
5064          */
5065         if (xh_free_start % blocksize < need)
5066                 free -= xh_free_start % blocksize;
5067
5068         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5069              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5070              " %u\n", xs->not_found,
5071              (unsigned long long)bucket_blkno(xs->bucket),
5072              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5073              le16_to_cpu(xh->xh_name_value_len));
5074
5075         if (free < need ||
5076             (xs->not_found &&
5077              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5078                 if (need <= max_free &&
5079                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5080                         /*
5081                          * We can create the space by defragment. Since only the
5082                          * name/value will be moved, the xe shouldn't be changed
5083                          * in xs.
5084                          */
5085                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5086                                                         xs->bucket);
5087                         if (ret) {
5088                                 mlog_errno(ret);
5089                                 goto out;
5090                         }
5091
5092                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5093                         free = xh_free_start - header_size;
5094                         if (xh_free_start % blocksize < need)
5095                                 free -= xh_free_start % blocksize;
5096
5097                         if (free >= need)
5098                                 goto xattr_set;
5099
5100                         mlog(0, "Can't get enough space for xattr insert by "
5101                              "defragment. Need %u bytes, but we have %d, so "
5102                              "allocate new bucket for it.\n", need, free);
5103                 }
5104
5105                 /*
5106                  * We have to add new buckets or clusters and one
5107                  * allocation should leave us enough space for insert.
5108                  */
5109                 BUG_ON(allocation);
5110
5111                 /*
5112                  * We do not allow for overlapping ranges between buckets. And
5113                  * the maximum number of collisions we will allow for then is
5114                  * one bucket's worth, so check it here whether we need to
5115                  * add a new bucket for the insert.
5116                  */
5117                 ret = ocfs2_check_xattr_bucket_collision(inode,
5118                                                          xs->bucket,
5119                                                          xi->name);
5120                 if (ret) {
5121                         mlog_errno(ret);
5122                         goto out;
5123                 }
5124
5125                 ret = ocfs2_add_new_xattr_bucket(inode,
5126                                                  xs->xattr_bh,
5127                                                  xs->bucket->bu_bhs[0],
5128                                                  ctxt);
5129                 if (ret) {
5130                         mlog_errno(ret);
5131                         goto out;
5132                 }
5133
5134                 ocfs2_xattr_bucket_relse(xs->bucket);
5135
5136                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5137                                                    xi->name_index,
5138                                                    xi->name, xs);
5139                 if (ret && ret != -ENODATA)
5140                         goto out;
5141                 xs->not_found = ret;
5142                 allocation = 1;
5143                 goto try_again;
5144         }
5145
5146 xattr_set:
5147         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5148 out:
5149         mlog_exit(ret);
5150         return ret;
5151 }
5152
5153 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5154                                         struct ocfs2_xattr_bucket *bucket,
5155                                         void *para)
5156 {
5157         int ret = 0;
5158         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5159         u16 i;
5160         struct ocfs2_xattr_entry *xe;
5161         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5162         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5163         int credits = ocfs2_remove_extent_credits(osb->sb) +
5164                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5165
5166
5167         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5168
5169         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5170                 xe = &xh->xh_entries[i];
5171                 if (ocfs2_xattr_is_local(xe))
5172                         continue;
5173
5174                 ctxt.handle = ocfs2_start_trans(osb, credits);
5175                 if (IS_ERR(ctxt.handle)) {
5176                         ret = PTR_ERR(ctxt.handle);
5177                         mlog_errno(ret);
5178                         break;
5179                 }
5180
5181                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5182                                                         i, 0, &ctxt);
5183
5184                 ocfs2_commit_trans(osb, ctxt.handle);
5185                 if (ret) {
5186                         mlog_errno(ret);
5187                         break;
5188                 }
5189         }
5190
5191         ocfs2_schedule_truncate_log_flush(osb, 1);
5192         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5193         return ret;
5194 }
5195
5196 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5197                                           struct buffer_head *xb_bh)
5198 {
5199         struct ocfs2_xattr_block *xb =
5200                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5201         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5202         int ret = 0;
5203         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5204         u64 p_blkno;
5205
5206         if (le16_to_cpu(el->l_next_free_rec) == 0)
5207                 return 0;
5208
5209         while (name_hash > 0) {
5210                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5211                                           &e_cpos, &num_clusters, el);
5212                 if (ret) {
5213                         mlog_errno(ret);
5214                         goto out;
5215                 }
5216
5217                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5218                                                   ocfs2_delete_xattr_in_bucket,
5219                                                   NULL);
5220                 if (ret) {
5221                         mlog_errno(ret);
5222                         goto out;
5223                 }
5224
5225                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5226                                              p_blkno, e_cpos, num_clusters);
5227                 if (ret) {
5228                         mlog_errno(ret);
5229                         break;
5230                 }
5231
5232                 if (e_cpos == 0)
5233                         break;
5234
5235                 name_hash = e_cpos - 1;
5236         }
5237
5238 out:
5239         return ret;
5240 }
5241
5242 /*
5243  * 'security' attributes support
5244  */
5245 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5246                                         size_t list_size, const char *name,
5247                                         size_t name_len)
5248 {
5249         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5250         const size_t total_len = prefix_len + name_len + 1;
5251
5252         if (list && total_len <= list_size) {
5253                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5254                 memcpy(list + prefix_len, name, name_len);
5255                 list[prefix_len + name_len] = '\0';
5256         }
5257         return total_len;
5258 }
5259
5260 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5261                                     void *buffer, size_t size)
5262 {
5263         if (strcmp(name, "") == 0)
5264                 return -EINVAL;
5265         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5266                                buffer, size);
5267 }
5268
5269 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5270                                     const void *value, size_t size, int flags)
5271 {
5272         if (strcmp(name, "") == 0)
5273                 return -EINVAL;
5274
5275         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5276                                size, flags);
5277 }
5278
5279 int ocfs2_init_security_get(struct inode *inode,
5280                             struct inode *dir,
5281                             struct ocfs2_security_xattr_info *si)
5282 {
5283         return security_inode_init_security(inode, dir, &si->name, &si->value,
5284                                             &si->value_len);
5285 }
5286
5287 int ocfs2_init_security_set(handle_t *handle,
5288                             struct inode *inode,
5289                             struct buffer_head *di_bh,
5290                             struct ocfs2_security_xattr_info *si,
5291                             struct ocfs2_alloc_context *xattr_ac,
5292                             struct ocfs2_alloc_context *data_ac)
5293 {
5294         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5295                                      OCFS2_XATTR_INDEX_SECURITY,
5296                                      si->name, si->value, si->value_len, 0,
5297                                      xattr_ac, data_ac);
5298 }
5299
5300 struct xattr_handler ocfs2_xattr_security_handler = {
5301         .prefix = XATTR_SECURITY_PREFIX,
5302         .list   = ocfs2_xattr_security_list,
5303         .get    = ocfs2_xattr_security_get,
5304         .set    = ocfs2_xattr_security_set,
5305 };
5306
5307 /*
5308  * 'trusted' attributes support
5309  */
5310 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5311                                        size_t list_size, const char *name,
5312                                        size_t name_len)
5313 {
5314         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5315         const size_t total_len = prefix_len + name_len + 1;
5316
5317         if (list && total_len <= list_size) {
5318                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5319                 memcpy(list + prefix_len, name, name_len);
5320                 list[prefix_len + name_len] = '\0';
5321         }
5322         return total_len;
5323 }
5324
5325 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5326                                    void *buffer, size_t size)
5327 {
5328         if (strcmp(name, "") == 0)
5329                 return -EINVAL;
5330         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5331                                buffer, size);
5332 }
5333
5334 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5335                                    const void *value, size_t size, int flags)
5336 {
5337         if (strcmp(name, "") == 0)
5338                 return -EINVAL;
5339
5340         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5341                                size, flags);
5342 }
5343
5344 struct xattr_handler ocfs2_xattr_trusted_handler = {
5345         .prefix = XATTR_TRUSTED_PREFIX,
5346         .list   = ocfs2_xattr_trusted_list,
5347         .get    = ocfs2_xattr_trusted_get,
5348         .set    = ocfs2_xattr_trusted_set,
5349 };
5350
5351 /*
5352  * 'user' attributes support
5353  */
5354 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5355                                     size_t list_size, const char *name,
5356                                     size_t name_len)
5357 {
5358         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5359         const size_t total_len = prefix_len + name_len + 1;
5360         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5361
5362         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5363                 return 0;
5364
5365         if (list && total_len <= list_size) {
5366                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5367                 memcpy(list + prefix_len, name, name_len);
5368                 list[prefix_len + name_len] = '\0';
5369         }
5370         return total_len;
5371 }
5372
5373 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5374                                 void *buffer, size_t size)
5375 {
5376         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5377
5378         if (strcmp(name, "") == 0)
5379                 return -EINVAL;
5380         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5381                 return -EOPNOTSUPP;
5382         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5383                                buffer, size);
5384 }
5385
5386 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5387                                 const void *value, size_t size, int flags)
5388 {
5389         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5390
5391         if (strcmp(name, "") == 0)
5392                 return -EINVAL;
5393         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5394                 return -EOPNOTSUPP;
5395
5396         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5397                                size, flags);
5398 }
5399
5400 struct xattr_handler ocfs2_xattr_user_handler = {
5401         .prefix = XATTR_USER_PREFIX,
5402         .list   = ocfs2_xattr_user_list,
5403         .get    = ocfs2_xattr_user_get,
5404         .set    = ocfs2_xattr_user_set,
5405 };