ocfs2: Explain t_is_new in ocfs2_cp_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
174 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
175 {
176         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
177 }
178
179 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
180 {
181         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
182 }
183
184 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
185 {
186         u16 len = sb->s_blocksize -
187                  offsetof(struct ocfs2_xattr_header, xh_entries);
188
189         return len / sizeof(struct ocfs2_xattr_entry);
190 }
191
192 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
193 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
194 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
195
196 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
197 {
198         struct ocfs2_xattr_bucket *bucket;
199         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
200
201         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
202
203         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
204         if (bucket) {
205                 bucket->bu_inode = inode;
206                 bucket->bu_blocks = blks;
207         }
208
209         return bucket;
210 }
211
212 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
213 {
214         int i;
215
216         for (i = 0; i < bucket->bu_blocks; i++) {
217                 brelse(bucket->bu_bhs[i]);
218                 bucket->bu_bhs[i] = NULL;
219         }
220 }
221
222 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
223 {
224         if (bucket) {
225                 ocfs2_xattr_bucket_relse(bucket);
226                 bucket->bu_inode = NULL;
227                 kfree(bucket);
228         }
229 }
230
231 /*
232  * A bucket that has never been written to disk doesn't need to be
233  * read.  We just need the buffer_heads.  Don't call this for
234  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
235  * them fully.
236  */
237 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
238                                    u64 xb_blkno)
239 {
240         int i, rc = 0;
241
242         for (i = 0; i < bucket->bu_blocks; i++) {
243                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
244                                               xb_blkno + i);
245                 if (!bucket->bu_bhs[i]) {
246                         rc = -EIO;
247                         mlog_errno(rc);
248                         break;
249                 }
250
251                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
252                                            bucket->bu_bhs[i]))
253                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
254                                                       bucket->bu_bhs[i]);
255         }
256
257         if (rc)
258                 ocfs2_xattr_bucket_relse(bucket);
259         return rc;
260 }
261
262 /* Read the xattr bucket at xb_blkno */
263 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
264                                    u64 xb_blkno)
265 {
266         int rc;
267
268         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
269                                bucket->bu_blocks, bucket->bu_bhs, 0,
270                                NULL);
271         if (rc)
272                 ocfs2_xattr_bucket_relse(bucket);
273         return rc;
274 }
275
276 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
277                                              struct ocfs2_xattr_bucket *bucket,
278                                              int type)
279 {
280         int i, rc = 0;
281
282         for (i = 0; i < bucket->bu_blocks; i++) {
283                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
284                                           bucket->bu_bhs[i], type);
285                 if (rc) {
286                         mlog_errno(rc);
287                         break;
288                 }
289         }
290
291         return rc;
292 }
293
294 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
295                                              struct ocfs2_xattr_bucket *bucket)
296 {
297         int i;
298
299         for (i = 0; i < bucket->bu_blocks; i++)
300                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
301 }
302
303 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
304                                          struct ocfs2_xattr_bucket *src)
305 {
306         int i;
307         int blocksize = src->bu_inode->i_sb->s_blocksize;
308
309         BUG_ON(dest->bu_blocks != src->bu_blocks);
310         BUG_ON(dest->bu_inode != src->bu_inode);
311
312         for (i = 0; i < src->bu_blocks; i++) {
313                 memcpy(bucket_block(dest, i), bucket_block(src, i),
314                        blocksize);
315         }
316 }
317
318 static int ocfs2_validate_xattr_block(struct super_block *sb,
319                                       struct buffer_head *bh)
320 {
321         struct ocfs2_xattr_block *xb =
322                 (struct ocfs2_xattr_block *)bh->b_data;
323
324         mlog(0, "Validating xattr block %llu\n",
325              (unsigned long long)bh->b_blocknr);
326
327         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
328                 ocfs2_error(sb,
329                             "Extended attribute block #%llu has bad "
330                             "signature %.*s",
331                             (unsigned long long)bh->b_blocknr, 7,
332                             xb->xb_signature);
333                 return -EINVAL;
334         }
335
336         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
337                 ocfs2_error(sb,
338                             "Extended attribute block #%llu has an "
339                             "invalid xb_blkno of %llu",
340                             (unsigned long long)bh->b_blocknr,
341                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
342                 return -EINVAL;
343         }
344
345         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
346                 ocfs2_error(sb,
347                             "Extended attribute block #%llu has an invalid "
348                             "xb_fs_generation of #%u",
349                             (unsigned long long)bh->b_blocknr,
350                             le32_to_cpu(xb->xb_fs_generation));
351                 return -EINVAL;
352         }
353
354         return 0;
355 }
356
357 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
358                                   struct buffer_head **bh)
359 {
360         int rc;
361         struct buffer_head *tmp = *bh;
362
363         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
364                               ocfs2_validate_xattr_block);
365
366         /* If ocfs2_read_block() got us a new bh, pass it up. */
367         if (!rc && !*bh)
368                 *bh = tmp;
369
370         return rc;
371 }
372
373 static inline const char *ocfs2_xattr_prefix(int name_index)
374 {
375         struct xattr_handler *handler = NULL;
376
377         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
378                 handler = ocfs2_xattr_handler_map[name_index];
379
380         return handler ? handler->prefix : NULL;
381 }
382
383 static u32 ocfs2_xattr_name_hash(struct inode *inode,
384                                  const char *name,
385                                  int name_len)
386 {
387         /* Get hash value of uuid from super block */
388         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
389         int i;
390
391         /* hash extended attribute name */
392         for (i = 0; i < name_len; i++) {
393                 hash = (hash << OCFS2_HASH_SHIFT) ^
394                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
395                        *name++;
396         }
397
398         return hash;
399 }
400
401 /*
402  * ocfs2_xattr_hash_entry()
403  *
404  * Compute the hash of an extended attribute.
405  */
406 static void ocfs2_xattr_hash_entry(struct inode *inode,
407                                    struct ocfs2_xattr_header *header,
408                                    struct ocfs2_xattr_entry *entry)
409 {
410         u32 hash = 0;
411         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
412
413         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
414         entry->xe_name_hash = cpu_to_le32(hash);
415
416         return;
417 }
418
419 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
420 {
421         int size = 0;
422
423         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
424                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
425         else
426                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
427         size += sizeof(struct ocfs2_xattr_entry);
428
429         return size;
430 }
431
432 int ocfs2_calc_security_init(struct inode *dir,
433                              struct ocfs2_security_xattr_info *si,
434                              int *want_clusters,
435                              int *xattr_credits,
436                              struct ocfs2_alloc_context **xattr_ac)
437 {
438         int ret = 0;
439         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
440         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
441                                                  si->value_len);
442
443         /*
444          * The max space of security xattr taken inline is
445          * 256(name) + 80(value) + 16(entry) = 352 bytes,
446          * So reserve one metadata block for it is ok.
447          */
448         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
449             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
450                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
451                 if (ret) {
452                         mlog_errno(ret);
453                         return ret;
454                 }
455                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
456         }
457
458         /* reserve clusters for xattr value which will be set in B tree*/
459         if (si->value_len > OCFS2_XATTR_INLINE_SIZE)
460                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
461                                                            si->value_len);
462         return ret;
463 }
464
465 int ocfs2_calc_xattr_init(struct inode *dir,
466                           struct buffer_head *dir_bh,
467                           int mode,
468                           struct ocfs2_security_xattr_info *si,
469                           int *want_clusters,
470                           int *xattr_credits,
471                           struct ocfs2_alloc_context **xattr_ac)
472 {
473         int ret = 0;
474         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
475         int s_size = 0;
476         int a_size = 0;
477         int acl_len = 0;
478
479         if (si->enable)
480                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
481                                                      si->value_len);
482
483         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
484                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
485                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
486                                         "", NULL, 0);
487                 if (acl_len > 0) {
488                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
489                         if (S_ISDIR(mode))
490                                 a_size <<= 1;
491                 } else if (acl_len != 0 && acl_len != -ENODATA) {
492                         mlog_errno(ret);
493                         return ret;
494                 }
495         }
496
497         if (!(s_size + a_size))
498                 return ret;
499
500         /*
501          * The max space of security xattr taken inline is
502          * 256(name) + 80(value) + 16(entry) = 352 bytes,
503          * The max space of acl xattr taken inline is
504          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
505          * when blocksize = 512, may reserve one more cluser for
506          * xattr bucket, otherwise reserve one metadata block
507          * for them is ok.
508          */
509         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
510             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
511                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
512                 if (ret) {
513                         mlog_errno(ret);
514                         return ret;
515                 }
516                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
517         }
518
519         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
520             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
521                 *want_clusters += 1;
522                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
523         }
524
525         /* reserve clusters for xattr value which will be set in B tree*/
526         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE)
527                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
528                                                            si->value_len);
529         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
530             acl_len > OCFS2_XATTR_INLINE_SIZE) {
531                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
532                 if (S_ISDIR(mode))
533                         *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
534                                                                    acl_len);
535         }
536
537         return ret;
538 }
539
540 static int ocfs2_xattr_extend_allocation(struct inode *inode,
541                                          u32 clusters_to_add,
542                                          struct buffer_head *xattr_bh,
543                                          struct ocfs2_xattr_value_root *xv,
544                                          struct ocfs2_xattr_set_ctxt *ctxt)
545 {
546         int status = 0;
547         handle_t *handle = ctxt->handle;
548         enum ocfs2_alloc_restarted why;
549         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
550         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
551         struct ocfs2_extent_tree et;
552
553         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
554
555         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
556
557         status = ocfs2_journal_access(handle, inode, xattr_bh,
558                                       OCFS2_JOURNAL_ACCESS_WRITE);
559         if (status < 0) {
560                 mlog_errno(status);
561                 goto leave;
562         }
563
564         prev_clusters = le32_to_cpu(xv->xr_clusters);
565         status = ocfs2_add_clusters_in_btree(osb,
566                                              inode,
567                                              &logical_start,
568                                              clusters_to_add,
569                                              0,
570                                              &et,
571                                              handle,
572                                              ctxt->data_ac,
573                                              ctxt->meta_ac,
574                                              &why);
575         if (status < 0) {
576                 mlog_errno(status);
577                 goto leave;
578         }
579
580         status = ocfs2_journal_dirty(handle, xattr_bh);
581         if (status < 0) {
582                 mlog_errno(status);
583                 goto leave;
584         }
585
586         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
587
588         /*
589          * We should have already allocated enough space before the transaction,
590          * so no need to restart.
591          */
592         BUG_ON(why != RESTART_NONE || clusters_to_add);
593
594 leave:
595
596         return status;
597 }
598
599 static int __ocfs2_remove_xattr_range(struct inode *inode,
600                                       struct buffer_head *root_bh,
601                                       struct ocfs2_xattr_value_root *xv,
602                                       u32 cpos, u32 phys_cpos, u32 len,
603                                       struct ocfs2_xattr_set_ctxt *ctxt)
604 {
605         int ret;
606         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
607         handle_t *handle = ctxt->handle;
608         struct ocfs2_extent_tree et;
609
610         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
611
612         ret = ocfs2_journal_access(handle, inode, root_bh,
613                                    OCFS2_JOURNAL_ACCESS_WRITE);
614         if (ret) {
615                 mlog_errno(ret);
616                 goto out;
617         }
618
619         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
620                                   &ctxt->dealloc);
621         if (ret) {
622                 mlog_errno(ret);
623                 goto out;
624         }
625
626         le32_add_cpu(&xv->xr_clusters, -len);
627
628         ret = ocfs2_journal_dirty(handle, root_bh);
629         if (ret) {
630                 mlog_errno(ret);
631                 goto out;
632         }
633
634         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
635         if (ret)
636                 mlog_errno(ret);
637
638 out:
639         return ret;
640 }
641
642 static int ocfs2_xattr_shrink_size(struct inode *inode,
643                                    u32 old_clusters,
644                                    u32 new_clusters,
645                                    struct buffer_head *root_bh,
646                                    struct ocfs2_xattr_value_root *xv,
647                                    struct ocfs2_xattr_set_ctxt *ctxt)
648 {
649         int ret = 0;
650         u32 trunc_len, cpos, phys_cpos, alloc_size;
651         u64 block;
652
653         if (old_clusters <= new_clusters)
654                 return 0;
655
656         cpos = new_clusters;
657         trunc_len = old_clusters - new_clusters;
658         while (trunc_len) {
659                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
660                                                &alloc_size, &xv->xr_list);
661                 if (ret) {
662                         mlog_errno(ret);
663                         goto out;
664                 }
665
666                 if (alloc_size > trunc_len)
667                         alloc_size = trunc_len;
668
669                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
670                                                  phys_cpos, alloc_size,
671                                                  ctxt);
672                 if (ret) {
673                         mlog_errno(ret);
674                         goto out;
675                 }
676
677                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
678                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
679                                                        alloc_size);
680                 cpos += alloc_size;
681                 trunc_len -= alloc_size;
682         }
683
684 out:
685         return ret;
686 }
687
688 static int ocfs2_xattr_value_truncate(struct inode *inode,
689                                       struct buffer_head *root_bh,
690                                       struct ocfs2_xattr_value_root *xv,
691                                       int len,
692                                       struct ocfs2_xattr_set_ctxt *ctxt)
693 {
694         int ret;
695         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
696         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
697
698         if (new_clusters == old_clusters)
699                 return 0;
700
701         if (new_clusters > old_clusters)
702                 ret = ocfs2_xattr_extend_allocation(inode,
703                                                     new_clusters - old_clusters,
704                                                     root_bh, xv, ctxt);
705         else
706                 ret = ocfs2_xattr_shrink_size(inode,
707                                               old_clusters, new_clusters,
708                                               root_bh, xv, ctxt);
709
710         return ret;
711 }
712
713 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
714                                   size_t *result, const char *prefix,
715                                   const char *name, int name_len)
716 {
717         char *p = buffer + *result;
718         int prefix_len = strlen(prefix);
719         int total_len = prefix_len + name_len + 1;
720
721         *result += total_len;
722
723         /* we are just looking for how big our buffer needs to be */
724         if (!size)
725                 return 0;
726
727         if (*result > size)
728                 return -ERANGE;
729
730         memcpy(p, prefix, prefix_len);
731         memcpy(p + prefix_len, name, name_len);
732         p[prefix_len + name_len] = '\0';
733
734         return 0;
735 }
736
737 static int ocfs2_xattr_list_entries(struct inode *inode,
738                                     struct ocfs2_xattr_header *header,
739                                     char *buffer, size_t buffer_size)
740 {
741         size_t result = 0;
742         int i, type, ret;
743         const char *prefix, *name;
744
745         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
746                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
747                 type = ocfs2_xattr_get_type(entry);
748                 prefix = ocfs2_xattr_prefix(type);
749
750                 if (prefix) {
751                         name = (const char *)header +
752                                 le16_to_cpu(entry->xe_name_offset);
753
754                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
755                                                      &result, prefix, name,
756                                                      entry->xe_name_len);
757                         if (ret)
758                                 return ret;
759                 }
760         }
761
762         return result;
763 }
764
765 static int ocfs2_xattr_ibody_list(struct inode *inode,
766                                   struct ocfs2_dinode *di,
767                                   char *buffer,
768                                   size_t buffer_size)
769 {
770         struct ocfs2_xattr_header *header = NULL;
771         struct ocfs2_inode_info *oi = OCFS2_I(inode);
772         int ret = 0;
773
774         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
775                 return ret;
776
777         header = (struct ocfs2_xattr_header *)
778                  ((void *)di + inode->i_sb->s_blocksize -
779                  le16_to_cpu(di->i_xattr_inline_size));
780
781         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
782
783         return ret;
784 }
785
786 static int ocfs2_xattr_block_list(struct inode *inode,
787                                   struct ocfs2_dinode *di,
788                                   char *buffer,
789                                   size_t buffer_size)
790 {
791         struct buffer_head *blk_bh = NULL;
792         struct ocfs2_xattr_block *xb;
793         int ret = 0;
794
795         if (!di->i_xattr_loc)
796                 return ret;
797
798         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
799                                      &blk_bh);
800         if (ret < 0) {
801                 mlog_errno(ret);
802                 return ret;
803         }
804
805         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
806         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
807                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
808                 ret = ocfs2_xattr_list_entries(inode, header,
809                                                buffer, buffer_size);
810         } else {
811                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
812                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
813                                                    buffer, buffer_size);
814         }
815
816         brelse(blk_bh);
817
818         return ret;
819 }
820
821 ssize_t ocfs2_listxattr(struct dentry *dentry,
822                         char *buffer,
823                         size_t size)
824 {
825         int ret = 0, i_ret = 0, b_ret = 0;
826         struct buffer_head *di_bh = NULL;
827         struct ocfs2_dinode *di = NULL;
828         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
829
830         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
831                 return -EOPNOTSUPP;
832
833         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
834                 return ret;
835
836         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
837         if (ret < 0) {
838                 mlog_errno(ret);
839                 return ret;
840         }
841
842         di = (struct ocfs2_dinode *)di_bh->b_data;
843
844         down_read(&oi->ip_xattr_sem);
845         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
846         if (i_ret < 0)
847                 b_ret = 0;
848         else {
849                 if (buffer) {
850                         buffer += i_ret;
851                         size -= i_ret;
852                 }
853                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
854                                                buffer, size);
855                 if (b_ret < 0)
856                         i_ret = 0;
857         }
858         up_read(&oi->ip_xattr_sem);
859         ocfs2_inode_unlock(dentry->d_inode, 0);
860
861         brelse(di_bh);
862
863         return i_ret + b_ret;
864 }
865
866 static int ocfs2_xattr_find_entry(int name_index,
867                                   const char *name,
868                                   struct ocfs2_xattr_search *xs)
869 {
870         struct ocfs2_xattr_entry *entry;
871         size_t name_len;
872         int i, cmp = 1;
873
874         if (name == NULL)
875                 return -EINVAL;
876
877         name_len = strlen(name);
878         entry = xs->here;
879         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
880                 cmp = name_index - ocfs2_xattr_get_type(entry);
881                 if (!cmp)
882                         cmp = name_len - entry->xe_name_len;
883                 if (!cmp)
884                         cmp = memcmp(name, (xs->base +
885                                      le16_to_cpu(entry->xe_name_offset)),
886                                      name_len);
887                 if (cmp == 0)
888                         break;
889                 entry += 1;
890         }
891         xs->here = entry;
892
893         return cmp ? -ENODATA : 0;
894 }
895
896 static int ocfs2_xattr_get_value_outside(struct inode *inode,
897                                          struct ocfs2_xattr_value_root *xv,
898                                          void *buffer,
899                                          size_t len)
900 {
901         u32 cpos, p_cluster, num_clusters, bpc, clusters;
902         u64 blkno;
903         int i, ret = 0;
904         size_t cplen, blocksize;
905         struct buffer_head *bh = NULL;
906         struct ocfs2_extent_list *el;
907
908         el = &xv->xr_list;
909         clusters = le32_to_cpu(xv->xr_clusters);
910         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
911         blocksize = inode->i_sb->s_blocksize;
912
913         cpos = 0;
914         while (cpos < clusters) {
915                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
916                                                &num_clusters, el);
917                 if (ret) {
918                         mlog_errno(ret);
919                         goto out;
920                 }
921
922                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
923                 /* Copy ocfs2_xattr_value */
924                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
925                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
926                         if (ret) {
927                                 mlog_errno(ret);
928                                 goto out;
929                         }
930
931                         cplen = len >= blocksize ? blocksize : len;
932                         memcpy(buffer, bh->b_data, cplen);
933                         len -= cplen;
934                         buffer += cplen;
935
936                         brelse(bh);
937                         bh = NULL;
938                         if (len == 0)
939                                 break;
940                 }
941                 cpos += num_clusters;
942         }
943 out:
944         return ret;
945 }
946
947 static int ocfs2_xattr_ibody_get(struct inode *inode,
948                                  int name_index,
949                                  const char *name,
950                                  void *buffer,
951                                  size_t buffer_size,
952                                  struct ocfs2_xattr_search *xs)
953 {
954         struct ocfs2_inode_info *oi = OCFS2_I(inode);
955         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
956         struct ocfs2_xattr_value_root *xv;
957         size_t size;
958         int ret = 0;
959
960         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
961                 return -ENODATA;
962
963         xs->end = (void *)di + inode->i_sb->s_blocksize;
964         xs->header = (struct ocfs2_xattr_header *)
965                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
966         xs->base = (void *)xs->header;
967         xs->here = xs->header->xh_entries;
968
969         ret = ocfs2_xattr_find_entry(name_index, name, xs);
970         if (ret)
971                 return ret;
972         size = le64_to_cpu(xs->here->xe_value_size);
973         if (buffer) {
974                 if (size > buffer_size)
975                         return -ERANGE;
976                 if (ocfs2_xattr_is_local(xs->here)) {
977                         memcpy(buffer, (void *)xs->base +
978                                le16_to_cpu(xs->here->xe_name_offset) +
979                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
980                 } else {
981                         xv = (struct ocfs2_xattr_value_root *)
982                                 (xs->base + le16_to_cpu(
983                                  xs->here->xe_name_offset) +
984                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
985                         ret = ocfs2_xattr_get_value_outside(inode, xv,
986                                                             buffer, size);
987                         if (ret < 0) {
988                                 mlog_errno(ret);
989                                 return ret;
990                         }
991                 }
992         }
993
994         return size;
995 }
996
997 static int ocfs2_xattr_block_get(struct inode *inode,
998                                  int name_index,
999                                  const char *name,
1000                                  void *buffer,
1001                                  size_t buffer_size,
1002                                  struct ocfs2_xattr_search *xs)
1003 {
1004         struct ocfs2_xattr_block *xb;
1005         struct ocfs2_xattr_value_root *xv;
1006         size_t size;
1007         int ret = -ENODATA, name_offset, name_len, block_off, i;
1008
1009         xs->bucket = ocfs2_xattr_bucket_new(inode);
1010         if (!xs->bucket) {
1011                 ret = -ENOMEM;
1012                 mlog_errno(ret);
1013                 goto cleanup;
1014         }
1015
1016         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1017         if (ret) {
1018                 mlog_errno(ret);
1019                 goto cleanup;
1020         }
1021
1022         if (xs->not_found) {
1023                 ret = -ENODATA;
1024                 goto cleanup;
1025         }
1026
1027         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1028         size = le64_to_cpu(xs->here->xe_value_size);
1029         if (buffer) {
1030                 ret = -ERANGE;
1031                 if (size > buffer_size)
1032                         goto cleanup;
1033
1034                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1035                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1036                 i = xs->here - xs->header->xh_entries;
1037
1038                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1039                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1040                                                                 bucket_xh(xs->bucket),
1041                                                                 i,
1042                                                                 &block_off,
1043                                                                 &name_offset);
1044                         xs->base = bucket_block(xs->bucket, block_off);
1045                 }
1046                 if (ocfs2_xattr_is_local(xs->here)) {
1047                         memcpy(buffer, (void *)xs->base +
1048                                name_offset + name_len, size);
1049                 } else {
1050                         xv = (struct ocfs2_xattr_value_root *)
1051                                 (xs->base + name_offset + name_len);
1052                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1053                                                             buffer, size);
1054                         if (ret < 0) {
1055                                 mlog_errno(ret);
1056                                 goto cleanup;
1057                         }
1058                 }
1059         }
1060         ret = size;
1061 cleanup:
1062         ocfs2_xattr_bucket_free(xs->bucket);
1063
1064         brelse(xs->xattr_bh);
1065         xs->xattr_bh = NULL;
1066         return ret;
1067 }
1068
1069 int ocfs2_xattr_get_nolock(struct inode *inode,
1070                            struct buffer_head *di_bh,
1071                            int name_index,
1072                            const char *name,
1073                            void *buffer,
1074                            size_t buffer_size)
1075 {
1076         int ret;
1077         struct ocfs2_dinode *di = NULL;
1078         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1079         struct ocfs2_xattr_search xis = {
1080                 .not_found = -ENODATA,
1081         };
1082         struct ocfs2_xattr_search xbs = {
1083                 .not_found = -ENODATA,
1084         };
1085
1086         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1087                 return -EOPNOTSUPP;
1088
1089         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1090                 ret = -ENODATA;
1091
1092         xis.inode_bh = xbs.inode_bh = di_bh;
1093         di = (struct ocfs2_dinode *)di_bh->b_data;
1094
1095         down_read(&oi->ip_xattr_sem);
1096         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1097                                     buffer_size, &xis);
1098         if (ret == -ENODATA && di->i_xattr_loc)
1099                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1100                                             buffer_size, &xbs);
1101         up_read(&oi->ip_xattr_sem);
1102
1103         return ret;
1104 }
1105
1106 /* ocfs2_xattr_get()
1107  *
1108  * Copy an extended attribute into the buffer provided.
1109  * Buffer is NULL to compute the size of buffer required.
1110  */
1111 static int ocfs2_xattr_get(struct inode *inode,
1112                            int name_index,
1113                            const char *name,
1114                            void *buffer,
1115                            size_t buffer_size)
1116 {
1117         int ret;
1118         struct buffer_head *di_bh = NULL;
1119
1120         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1121         if (ret < 0) {
1122                 mlog_errno(ret);
1123                 return ret;
1124         }
1125         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1126                                      name, buffer, buffer_size);
1127
1128         ocfs2_inode_unlock(inode, 0);
1129
1130         brelse(di_bh);
1131
1132         return ret;
1133 }
1134
1135 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1136                                            handle_t *handle,
1137                                            struct ocfs2_xattr_value_root *xv,
1138                                            const void *value,
1139                                            int value_len)
1140 {
1141         int ret = 0, i, cp_len, credits;
1142         u16 blocksize = inode->i_sb->s_blocksize;
1143         u32 p_cluster, num_clusters;
1144         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1145         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1146         u64 blkno;
1147         struct buffer_head *bh = NULL;
1148
1149         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1150
1151         /*
1152          * In __ocfs2_xattr_set_value_outside has already been dirtied,
1153          * so we don't need to worry about whether ocfs2_extend_trans
1154          * will create a new transactio for us or not.
1155          */
1156         credits = clusters * bpc;
1157         ret = ocfs2_extend_trans(handle, credits);
1158         if (ret) {
1159                 mlog_errno(ret);
1160                 goto out;
1161         }
1162
1163         while (cpos < clusters) {
1164                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1165                                                &num_clusters, &xv->xr_list);
1166                 if (ret) {
1167                         mlog_errno(ret);
1168                         goto out;
1169                 }
1170
1171                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1172
1173                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1174                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1175                         if (ret) {
1176                                 mlog_errno(ret);
1177                                 goto out;
1178                         }
1179
1180                         ret = ocfs2_journal_access(handle,
1181                                                    inode,
1182                                                    bh,
1183                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1184                         if (ret < 0) {
1185                                 mlog_errno(ret);
1186                                 goto out;
1187                         }
1188
1189                         cp_len = value_len > blocksize ? blocksize : value_len;
1190                         memcpy(bh->b_data, value, cp_len);
1191                         value_len -= cp_len;
1192                         value += cp_len;
1193                         if (cp_len < blocksize)
1194                                 memset(bh->b_data + cp_len, 0,
1195                                        blocksize - cp_len);
1196
1197                         ret = ocfs2_journal_dirty(handle, bh);
1198                         if (ret < 0) {
1199                                 mlog_errno(ret);
1200                                 goto out;
1201                         }
1202                         brelse(bh);
1203                         bh = NULL;
1204
1205                         /*
1206                          * XXX: do we need to empty all the following
1207                          * blocks in this cluster?
1208                          */
1209                         if (!value_len)
1210                                 break;
1211                 }
1212                 cpos += num_clusters;
1213         }
1214 out:
1215         brelse(bh);
1216
1217         return ret;
1218 }
1219
1220 static int ocfs2_xattr_cleanup(struct inode *inode,
1221                                handle_t *handle,
1222                                struct ocfs2_xattr_info *xi,
1223                                struct ocfs2_xattr_search *xs,
1224                                size_t offs)
1225 {
1226         int ret = 0;
1227         size_t name_len = strlen(xi->name);
1228         void *val = xs->base + offs;
1229         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1230
1231         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1232                                    OCFS2_JOURNAL_ACCESS_WRITE);
1233         if (ret) {
1234                 mlog_errno(ret);
1235                 goto out;
1236         }
1237         /* Decrease xattr count */
1238         le16_add_cpu(&xs->header->xh_count, -1);
1239         /* Remove the xattr entry and tree root which has already be set*/
1240         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1241         memset(val, 0, size);
1242
1243         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1244         if (ret < 0)
1245                 mlog_errno(ret);
1246 out:
1247         return ret;
1248 }
1249
1250 static int ocfs2_xattr_update_entry(struct inode *inode,
1251                                     handle_t *handle,
1252                                     struct ocfs2_xattr_info *xi,
1253                                     struct ocfs2_xattr_search *xs,
1254                                     size_t offs)
1255 {
1256         int ret;
1257
1258         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1259                                    OCFS2_JOURNAL_ACCESS_WRITE);
1260         if (ret) {
1261                 mlog_errno(ret);
1262                 goto out;
1263         }
1264
1265         xs->here->xe_name_offset = cpu_to_le16(offs);
1266         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1267         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1268                 ocfs2_xattr_set_local(xs->here, 1);
1269         else
1270                 ocfs2_xattr_set_local(xs->here, 0);
1271         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1272
1273         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1274         if (ret < 0)
1275                 mlog_errno(ret);
1276 out:
1277         return ret;
1278 }
1279
1280 /*
1281  * ocfs2_xattr_set_value_outside()
1282  *
1283  * Set large size value in B tree.
1284  */
1285 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1286                                          struct ocfs2_xattr_info *xi,
1287                                          struct ocfs2_xattr_search *xs,
1288                                          struct ocfs2_xattr_set_ctxt *ctxt,
1289                                          size_t offs)
1290 {
1291         size_t name_len = strlen(xi->name);
1292         void *val = xs->base + offs;
1293         struct ocfs2_xattr_value_root *xv = NULL;
1294         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1295         int ret = 0;
1296
1297         memset(val, 0, size);
1298         memcpy(val, xi->name, name_len);
1299         xv = (struct ocfs2_xattr_value_root *)
1300                 (val + OCFS2_XATTR_SIZE(name_len));
1301         xv->xr_clusters = 0;
1302         xv->xr_last_eb_blk = 0;
1303         xv->xr_list.l_tree_depth = 0;
1304         xv->xr_list.l_count = cpu_to_le16(1);
1305         xv->xr_list.l_next_free_rec = 0;
1306
1307         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1308                                          xi->value_len, ctxt);
1309         if (ret < 0) {
1310                 mlog_errno(ret);
1311                 return ret;
1312         }
1313         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, offs);
1314         if (ret < 0) {
1315                 mlog_errno(ret);
1316                 return ret;
1317         }
1318         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, xv,
1319                                               xi->value, xi->value_len);
1320         if (ret < 0)
1321                 mlog_errno(ret);
1322
1323         return ret;
1324 }
1325
1326 /*
1327  * ocfs2_xattr_set_entry_local()
1328  *
1329  * Set, replace or remove extended attribute in local.
1330  */
1331 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1332                                         struct ocfs2_xattr_info *xi,
1333                                         struct ocfs2_xattr_search *xs,
1334                                         struct ocfs2_xattr_entry *last,
1335                                         size_t min_offs)
1336 {
1337         size_t name_len = strlen(xi->name);
1338         int i;
1339
1340         if (xi->value && xs->not_found) {
1341                 /* Insert the new xattr entry. */
1342                 le16_add_cpu(&xs->header->xh_count, 1);
1343                 ocfs2_xattr_set_type(last, xi->name_index);
1344                 ocfs2_xattr_set_local(last, 1);
1345                 last->xe_name_len = name_len;
1346         } else {
1347                 void *first_val;
1348                 void *val;
1349                 size_t offs, size;
1350
1351                 first_val = xs->base + min_offs;
1352                 offs = le16_to_cpu(xs->here->xe_name_offset);
1353                 val = xs->base + offs;
1354
1355                 if (le64_to_cpu(xs->here->xe_value_size) >
1356                     OCFS2_XATTR_INLINE_SIZE)
1357                         size = OCFS2_XATTR_SIZE(name_len) +
1358                                 OCFS2_XATTR_ROOT_SIZE;
1359                 else
1360                         size = OCFS2_XATTR_SIZE(name_len) +
1361                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1362
1363                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1364                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1365                         /* The old and the new value have the
1366                            same size. Just replace the value. */
1367                         ocfs2_xattr_set_local(xs->here, 1);
1368                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1369                         /* Clear value bytes. */
1370                         memset(val + OCFS2_XATTR_SIZE(name_len),
1371                                0,
1372                                OCFS2_XATTR_SIZE(xi->value_len));
1373                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1374                                xi->value,
1375                                xi->value_len);
1376                         return;
1377                 }
1378                 /* Remove the old name+value. */
1379                 memmove(first_val + size, first_val, val - first_val);
1380                 memset(first_val, 0, size);
1381                 xs->here->xe_name_hash = 0;
1382                 xs->here->xe_name_offset = 0;
1383                 ocfs2_xattr_set_local(xs->here, 1);
1384                 xs->here->xe_value_size = 0;
1385
1386                 min_offs += size;
1387
1388                 /* Adjust all value offsets. */
1389                 last = xs->header->xh_entries;
1390                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1391                         size_t o = le16_to_cpu(last->xe_name_offset);
1392
1393                         if (o < offs)
1394                                 last->xe_name_offset = cpu_to_le16(o + size);
1395                         last += 1;
1396                 }
1397
1398                 if (!xi->value) {
1399                         /* Remove the old entry. */
1400                         last -= 1;
1401                         memmove(xs->here, xs->here + 1,
1402                                 (void *)last - (void *)xs->here);
1403                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1404                         le16_add_cpu(&xs->header->xh_count, -1);
1405                 }
1406         }
1407         if (xi->value) {
1408                 /* Insert the new name+value. */
1409                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1410                                 OCFS2_XATTR_SIZE(xi->value_len);
1411                 void *val = xs->base + min_offs - size;
1412
1413                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1414                 memset(val, 0, size);
1415                 memcpy(val, xi->name, name_len);
1416                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1417                        xi->value,
1418                        xi->value_len);
1419                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1420                 ocfs2_xattr_set_local(xs->here, 1);
1421                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1422         }
1423
1424         return;
1425 }
1426
1427 /*
1428  * ocfs2_xattr_set_entry()
1429  *
1430  * Set extended attribute entry into inode or block.
1431  *
1432  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1433  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1434  * then set value in B tree with set_value_outside().
1435  */
1436 static int ocfs2_xattr_set_entry(struct inode *inode,
1437                                  struct ocfs2_xattr_info *xi,
1438                                  struct ocfs2_xattr_search *xs,
1439                                  struct ocfs2_xattr_set_ctxt *ctxt,
1440                                  int flag)
1441 {
1442         struct ocfs2_xattr_entry *last;
1443         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1444         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1445         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1446         size_t size_l = 0;
1447         handle_t *handle = ctxt->handle;
1448         int free, i, ret;
1449         struct ocfs2_xattr_info xi_l = {
1450                 .name_index = xi->name_index,
1451                 .name = xi->name,
1452                 .value = xi->value,
1453                 .value_len = xi->value_len,
1454         };
1455
1456         /* Compute min_offs, last and free space. */
1457         last = xs->header->xh_entries;
1458
1459         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1460                 size_t offs = le16_to_cpu(last->xe_name_offset);
1461                 if (offs < min_offs)
1462                         min_offs = offs;
1463                 last += 1;
1464         }
1465
1466         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1467         if (free < 0)
1468                 return -EIO;
1469
1470         if (!xs->not_found) {
1471                 size_t size = 0;
1472                 if (ocfs2_xattr_is_local(xs->here))
1473                         size = OCFS2_XATTR_SIZE(name_len) +
1474                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1475                 else
1476                         size = OCFS2_XATTR_SIZE(name_len) +
1477                                 OCFS2_XATTR_ROOT_SIZE;
1478                 free += (size + sizeof(struct ocfs2_xattr_entry));
1479         }
1480         /* Check free space in inode or block */
1481         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1482                 if (free < sizeof(struct ocfs2_xattr_entry) +
1483                            OCFS2_XATTR_SIZE(name_len) +
1484                            OCFS2_XATTR_ROOT_SIZE) {
1485                         ret = -ENOSPC;
1486                         goto out;
1487                 }
1488                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1489                 xi_l.value = (void *)&def_xv;
1490                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1491         } else if (xi->value) {
1492                 if (free < sizeof(struct ocfs2_xattr_entry) +
1493                            OCFS2_XATTR_SIZE(name_len) +
1494                            OCFS2_XATTR_SIZE(xi->value_len)) {
1495                         ret = -ENOSPC;
1496                         goto out;
1497                 }
1498         }
1499
1500         if (!xs->not_found) {
1501                 /* For existing extended attribute */
1502                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1503                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1504                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1505                 void *val = xs->base + offs;
1506
1507                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1508                         /* Replace existing local xattr with tree root */
1509                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1510                                                             ctxt, offs);
1511                         if (ret < 0)
1512                                 mlog_errno(ret);
1513                         goto out;
1514                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1515                         /* For existing xattr which has value outside */
1516                         struct ocfs2_xattr_value_root *xv = NULL;
1517                         xv = (struct ocfs2_xattr_value_root *)(val +
1518                                 OCFS2_XATTR_SIZE(name_len));
1519
1520                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1521                                 /*
1522                                  * If new value need set outside also,
1523                                  * first truncate old value to new value,
1524                                  * then set new value with set_value_outside().
1525                                  */
1526                                 ret = ocfs2_xattr_value_truncate(inode,
1527                                                                  xs->xattr_bh,
1528                                                                  xv,
1529                                                                  xi->value_len,
1530                                                                  ctxt);
1531                                 if (ret < 0) {
1532                                         mlog_errno(ret);
1533                                         goto out;
1534                                 }
1535
1536                                 ret = ocfs2_xattr_update_entry(inode,
1537                                                                handle,
1538                                                                xi,
1539                                                                xs,
1540                                                                offs);
1541                                 if (ret < 0) {
1542                                         mlog_errno(ret);
1543                                         goto out;
1544                                 }
1545
1546                                 ret = __ocfs2_xattr_set_value_outside(inode,
1547                                                                 handle,
1548                                                                 xv,
1549                                                                 xi->value,
1550                                                                 xi->value_len);
1551                                 if (ret < 0)
1552                                         mlog_errno(ret);
1553                                 goto out;
1554                         } else {
1555                                 /*
1556                                  * If new value need set in local,
1557                                  * just trucate old value to zero.
1558                                  */
1559                                  ret = ocfs2_xattr_value_truncate(inode,
1560                                                                   xs->xattr_bh,
1561                                                                   xv,
1562                                                                   0,
1563                                                                   ctxt);
1564                                 if (ret < 0)
1565                                         mlog_errno(ret);
1566                         }
1567                 }
1568         }
1569
1570         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1571                                    OCFS2_JOURNAL_ACCESS_WRITE);
1572         if (ret) {
1573                 mlog_errno(ret);
1574                 goto out;
1575         }
1576
1577         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1578                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1579                                            OCFS2_JOURNAL_ACCESS_WRITE);
1580                 if (ret) {
1581                         mlog_errno(ret);
1582                         goto out;
1583                 }
1584         }
1585
1586         /*
1587          * Set value in local, include set tree root in local.
1588          * This is the first step for value size >INLINE_SIZE.
1589          */
1590         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1591
1592         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1593                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1594                 if (ret < 0) {
1595                         mlog_errno(ret);
1596                         goto out;
1597                 }
1598         }
1599
1600         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1601             (flag & OCFS2_INLINE_XATTR_FL)) {
1602                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1603                 unsigned int xattrsize = osb->s_xattr_inline_size;
1604
1605                 /*
1606                  * Adjust extent record count or inline data size
1607                  * to reserve space for extended attribute.
1608                  */
1609                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1610                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1611                         le16_add_cpu(&idata->id_count, -xattrsize);
1612                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1613                         struct ocfs2_extent_list *el = &di->id2.i_list;
1614                         le16_add_cpu(&el->l_count, -(xattrsize /
1615                                         sizeof(struct ocfs2_extent_rec)));
1616                 }
1617                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1618         }
1619         /* Update xattr flag */
1620         spin_lock(&oi->ip_lock);
1621         oi->ip_dyn_features |= flag;
1622         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1623         spin_unlock(&oi->ip_lock);
1624         /* Update inode ctime */
1625         inode->i_ctime = CURRENT_TIME;
1626         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1627         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1628
1629         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1630         if (ret < 0)
1631                 mlog_errno(ret);
1632
1633         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1634                 /*
1635                  * Set value outside in B tree.
1636                  * This is the second step for value size > INLINE_SIZE.
1637                  */
1638                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1639                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt, offs);
1640                 if (ret < 0) {
1641                         int ret2;
1642
1643                         mlog_errno(ret);
1644                         /*
1645                          * If set value outside failed, we have to clean
1646                          * the junk tree root we have already set in local.
1647                          */
1648                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1649                                                    xi, xs, offs);
1650                         if (ret2 < 0)
1651                                 mlog_errno(ret2);
1652                 }
1653         }
1654 out:
1655         return ret;
1656 }
1657
1658 static int ocfs2_remove_value_outside(struct inode*inode,
1659                                       struct buffer_head *bh,
1660                                       struct ocfs2_xattr_header *header)
1661 {
1662         int ret = 0, i;
1663         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1664         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1665
1666         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1667
1668         ctxt.handle = ocfs2_start_trans(osb,
1669                                         ocfs2_remove_extent_credits(osb->sb));
1670         if (IS_ERR(ctxt.handle)) {
1671                 ret = PTR_ERR(ctxt.handle);
1672                 mlog_errno(ret);
1673                 goto out;
1674         }
1675
1676         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1677                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1678
1679                 if (!ocfs2_xattr_is_local(entry)) {
1680                         struct ocfs2_xattr_value_root *xv;
1681                         void *val;
1682
1683                         val = (void *)header +
1684                                 le16_to_cpu(entry->xe_name_offset);
1685                         xv = (struct ocfs2_xattr_value_root *)
1686                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1687                         ret = ocfs2_xattr_value_truncate(inode, bh, xv,
1688                                                          0, &ctxt);
1689                         if (ret < 0) {
1690                                 mlog_errno(ret);
1691                                 break;
1692                         }
1693                 }
1694         }
1695
1696         ocfs2_commit_trans(osb, ctxt.handle);
1697         ocfs2_schedule_truncate_log_flush(osb, 1);
1698         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1699 out:
1700         return ret;
1701 }
1702
1703 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1704                                     struct buffer_head *di_bh)
1705 {
1706
1707         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1708         struct ocfs2_xattr_header *header;
1709         int ret;
1710
1711         header = (struct ocfs2_xattr_header *)
1712                  ((void *)di + inode->i_sb->s_blocksize -
1713                  le16_to_cpu(di->i_xattr_inline_size));
1714
1715         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1716
1717         return ret;
1718 }
1719
1720 static int ocfs2_xattr_block_remove(struct inode *inode,
1721                                     struct buffer_head *blk_bh)
1722 {
1723         struct ocfs2_xattr_block *xb;
1724         int ret = 0;
1725
1726         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1727         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1728                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1729                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1730         } else
1731                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1732
1733         return ret;
1734 }
1735
1736 static int ocfs2_xattr_free_block(struct inode *inode,
1737                                   u64 block)
1738 {
1739         struct inode *xb_alloc_inode;
1740         struct buffer_head *xb_alloc_bh = NULL;
1741         struct buffer_head *blk_bh = NULL;
1742         struct ocfs2_xattr_block *xb;
1743         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1744         handle_t *handle;
1745         int ret = 0;
1746         u64 blk, bg_blkno;
1747         u16 bit;
1748
1749         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1750         if (ret < 0) {
1751                 mlog_errno(ret);
1752                 goto out;
1753         }
1754
1755         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1756         if (ret < 0) {
1757                 mlog_errno(ret);
1758                 goto out;
1759         }
1760
1761         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1762         blk = le64_to_cpu(xb->xb_blkno);
1763         bit = le16_to_cpu(xb->xb_suballoc_bit);
1764         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1765
1766         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1767                                 EXTENT_ALLOC_SYSTEM_INODE,
1768                                 le16_to_cpu(xb->xb_suballoc_slot));
1769         if (!xb_alloc_inode) {
1770                 ret = -ENOMEM;
1771                 mlog_errno(ret);
1772                 goto out;
1773         }
1774         mutex_lock(&xb_alloc_inode->i_mutex);
1775
1776         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1777         if (ret < 0) {
1778                 mlog_errno(ret);
1779                 goto out_mutex;
1780         }
1781
1782         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1783         if (IS_ERR(handle)) {
1784                 ret = PTR_ERR(handle);
1785                 mlog_errno(ret);
1786                 goto out_unlock;
1787         }
1788
1789         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1790                                        bit, bg_blkno, 1);
1791         if (ret < 0)
1792                 mlog_errno(ret);
1793
1794         ocfs2_commit_trans(osb, handle);
1795 out_unlock:
1796         ocfs2_inode_unlock(xb_alloc_inode, 1);
1797         brelse(xb_alloc_bh);
1798 out_mutex:
1799         mutex_unlock(&xb_alloc_inode->i_mutex);
1800         iput(xb_alloc_inode);
1801 out:
1802         brelse(blk_bh);
1803         return ret;
1804 }
1805
1806 /*
1807  * ocfs2_xattr_remove()
1808  *
1809  * Free extended attribute resources associated with this inode.
1810  */
1811 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1812 {
1813         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1814         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1815         handle_t *handle;
1816         int ret;
1817
1818         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1819                 return 0;
1820
1821         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1822                 return 0;
1823
1824         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1825                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1826                 if (ret < 0) {
1827                         mlog_errno(ret);
1828                         goto out;
1829                 }
1830         }
1831
1832         if (di->i_xattr_loc) {
1833                 ret = ocfs2_xattr_free_block(inode,
1834                                              le64_to_cpu(di->i_xattr_loc));
1835                 if (ret < 0) {
1836                         mlog_errno(ret);
1837                         goto out;
1838                 }
1839         }
1840
1841         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1842                                    OCFS2_INODE_UPDATE_CREDITS);
1843         if (IS_ERR(handle)) {
1844                 ret = PTR_ERR(handle);
1845                 mlog_errno(ret);
1846                 goto out;
1847         }
1848         ret = ocfs2_journal_access(handle, inode, di_bh,
1849                                    OCFS2_JOURNAL_ACCESS_WRITE);
1850         if (ret) {
1851                 mlog_errno(ret);
1852                 goto out_commit;
1853         }
1854
1855         di->i_xattr_loc = 0;
1856
1857         spin_lock(&oi->ip_lock);
1858         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1859         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1860         spin_unlock(&oi->ip_lock);
1861
1862         ret = ocfs2_journal_dirty(handle, di_bh);
1863         if (ret < 0)
1864                 mlog_errno(ret);
1865 out_commit:
1866         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1867 out:
1868         return ret;
1869 }
1870
1871 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1872                                         struct ocfs2_dinode *di)
1873 {
1874         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1875         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1876         int free;
1877
1878         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1879                 return 0;
1880
1881         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1882                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1883                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1884         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1885                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1886                         le64_to_cpu(di->i_size);
1887         } else {
1888                 struct ocfs2_extent_list *el = &di->id2.i_list;
1889                 free = (le16_to_cpu(el->l_count) -
1890                         le16_to_cpu(el->l_next_free_rec)) *
1891                         sizeof(struct ocfs2_extent_rec);
1892         }
1893         if (free >= xattrsize)
1894                 return 1;
1895
1896         return 0;
1897 }
1898
1899 /*
1900  * ocfs2_xattr_ibody_find()
1901  *
1902  * Find extended attribute in inode block and
1903  * fill search info into struct ocfs2_xattr_search.
1904  */
1905 static int ocfs2_xattr_ibody_find(struct inode *inode,
1906                                   int name_index,
1907                                   const char *name,
1908                                   struct ocfs2_xattr_search *xs)
1909 {
1910         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1911         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1912         int ret;
1913         int has_space = 0;
1914
1915         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1916                 return 0;
1917
1918         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1919                 down_read(&oi->ip_alloc_sem);
1920                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1921                 up_read(&oi->ip_alloc_sem);
1922                 if (!has_space)
1923                         return 0;
1924         }
1925
1926         xs->xattr_bh = xs->inode_bh;
1927         xs->end = (void *)di + inode->i_sb->s_blocksize;
1928         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1929                 xs->header = (struct ocfs2_xattr_header *)
1930                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1931         else
1932                 xs->header = (struct ocfs2_xattr_header *)
1933                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1934         xs->base = (void *)xs->header;
1935         xs->here = xs->header->xh_entries;
1936
1937         /* Find the named attribute. */
1938         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1939                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1940                 if (ret && ret != -ENODATA)
1941                         return ret;
1942                 xs->not_found = ret;
1943         }
1944
1945         return 0;
1946 }
1947
1948 /*
1949  * ocfs2_xattr_ibody_set()
1950  *
1951  * Set, replace or remove an extended attribute into inode block.
1952  *
1953  */
1954 static int ocfs2_xattr_ibody_set(struct inode *inode,
1955                                  struct ocfs2_xattr_info *xi,
1956                                  struct ocfs2_xattr_search *xs,
1957                                  struct ocfs2_xattr_set_ctxt *ctxt)
1958 {
1959         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1960         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1961         int ret;
1962
1963         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1964                 return -ENOSPC;
1965
1966         down_write(&oi->ip_alloc_sem);
1967         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1968                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1969                         ret = -ENOSPC;
1970                         goto out;
1971                 }
1972         }
1973
1974         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
1975                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1976 out:
1977         up_write(&oi->ip_alloc_sem);
1978
1979         return ret;
1980 }
1981
1982 /*
1983  * ocfs2_xattr_block_find()
1984  *
1985  * Find extended attribute in external block and
1986  * fill search info into struct ocfs2_xattr_search.
1987  */
1988 static int ocfs2_xattr_block_find(struct inode *inode,
1989                                   int name_index,
1990                                   const char *name,
1991                                   struct ocfs2_xattr_search *xs)
1992 {
1993         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1994         struct buffer_head *blk_bh = NULL;
1995         struct ocfs2_xattr_block *xb;
1996         int ret = 0;
1997
1998         if (!di->i_xattr_loc)
1999                 return ret;
2000
2001         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2002                                      &blk_bh);
2003         if (ret < 0) {
2004                 mlog_errno(ret);
2005                 return ret;
2006         }
2007
2008         xs->xattr_bh = blk_bh;
2009         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2010
2011         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2012                 xs->header = &xb->xb_attrs.xb_header;
2013                 xs->base = (void *)xs->header;
2014                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2015                 xs->here = xs->header->xh_entries;
2016
2017                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2018         } else
2019                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2020                                                    name_index,
2021                                                    name, xs);
2022
2023         if (ret && ret != -ENODATA) {
2024                 xs->xattr_bh = NULL;
2025                 goto cleanup;
2026         }
2027         xs->not_found = ret;
2028         return 0;
2029 cleanup:
2030         brelse(blk_bh);
2031
2032         return ret;
2033 }
2034
2035 /*
2036  * ocfs2_xattr_block_set()
2037  *
2038  * Set, replace or remove an extended attribute into external block.
2039  *
2040  */
2041 static int ocfs2_xattr_block_set(struct inode *inode,
2042                                  struct ocfs2_xattr_info *xi,
2043                                  struct ocfs2_xattr_search *xs,
2044                                  struct ocfs2_xattr_set_ctxt *ctxt)
2045 {
2046         struct buffer_head *new_bh = NULL;
2047         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2048         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2049         handle_t *handle = ctxt->handle;
2050         struct ocfs2_xattr_block *xblk = NULL;
2051         u16 suballoc_bit_start;
2052         u32 num_got;
2053         u64 first_blkno;
2054         int ret;
2055
2056         if (!xs->xattr_bh) {
2057                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
2058                                            OCFS2_JOURNAL_ACCESS_CREATE);
2059                 if (ret < 0) {
2060                         mlog_errno(ret);
2061                         goto end;
2062                 }
2063
2064                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2065                                            &suballoc_bit_start, &num_got,
2066                                            &first_blkno);
2067                 if (ret < 0) {
2068                         mlog_errno(ret);
2069                         goto end;
2070                 }
2071
2072                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2073                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2074
2075                 ret = ocfs2_journal_access(handle, inode, new_bh,
2076                                            OCFS2_JOURNAL_ACCESS_CREATE);
2077                 if (ret < 0) {
2078                         mlog_errno(ret);
2079                         goto end;
2080                 }
2081
2082                 /* Initialize ocfs2_xattr_block */
2083                 xs->xattr_bh = new_bh;
2084                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2085                 memset(xblk, 0, inode->i_sb->s_blocksize);
2086                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2087                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2088                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2089                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2090                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2091
2092                 xs->header = &xblk->xb_attrs.xb_header;
2093                 xs->base = (void *)xs->header;
2094                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2095                 xs->here = xs->header->xh_entries;
2096
2097                 ret = ocfs2_journal_dirty(handle, new_bh);
2098                 if (ret < 0) {
2099                         mlog_errno(ret);
2100                         goto end;
2101                 }
2102                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2103                 ocfs2_journal_dirty(handle, xs->inode_bh);
2104         } else
2105                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2106
2107         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2108                 /* Set extended attribute into external block */
2109                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2110                                             OCFS2_HAS_XATTR_FL);
2111                 if (!ret || ret != -ENOSPC)
2112                         goto end;
2113
2114                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2115                 if (ret)
2116                         goto end;
2117         }
2118
2119         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2120
2121 end:
2122
2123         return ret;
2124 }
2125
2126 /* Check whether the new xattr can be inserted into the inode. */
2127 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2128                                        struct ocfs2_xattr_info *xi,
2129                                        struct ocfs2_xattr_search *xs)
2130 {
2131         u64 value_size;
2132         struct ocfs2_xattr_entry *last;
2133         int free, i;
2134         size_t min_offs = xs->end - xs->base;
2135
2136         if (!xs->header)
2137                 return 0;
2138
2139         last = xs->header->xh_entries;
2140
2141         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2142                 size_t offs = le16_to_cpu(last->xe_name_offset);
2143                 if (offs < min_offs)
2144                         min_offs = offs;
2145                 last += 1;
2146         }
2147
2148         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2149         if (free < 0)
2150                 return 0;
2151
2152         BUG_ON(!xs->not_found);
2153
2154         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2155                 value_size = OCFS2_XATTR_ROOT_SIZE;
2156         else
2157                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2158
2159         if (free >= sizeof(struct ocfs2_xattr_entry) +
2160                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2161                 return 1;
2162
2163         return 0;
2164 }
2165
2166 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2167                                      struct ocfs2_dinode *di,
2168                                      struct ocfs2_xattr_info *xi,
2169                                      struct ocfs2_xattr_search *xis,
2170                                      struct ocfs2_xattr_search *xbs,
2171                                      int *clusters_need,
2172                                      int *meta_need,
2173                                      int *credits_need)
2174 {
2175         int ret = 0, old_in_xb = 0;
2176         int clusters_add = 0, meta_add = 0, credits = 0;
2177         struct buffer_head *bh = NULL;
2178         struct ocfs2_xattr_block *xb = NULL;
2179         struct ocfs2_xattr_entry *xe = NULL;
2180         struct ocfs2_xattr_value_root *xv = NULL;
2181         char *base = NULL;
2182         int name_offset, name_len = 0;
2183         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2184                                                     xi->value_len);
2185         u64 value_size;
2186
2187         if (xis->not_found && xbs->not_found) {
2188                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2189
2190                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2191                         clusters_add += new_clusters;
2192                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2193                                                         &def_xv.xv.xr_list,
2194                                                         new_clusters);
2195                 }
2196
2197                 goto meta_guess;
2198         }
2199
2200         if (!xis->not_found) {
2201                 xe = xis->here;
2202                 name_offset = le16_to_cpu(xe->xe_name_offset);
2203                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2204                 base = xis->base;
2205                 credits += OCFS2_INODE_UPDATE_CREDITS;
2206         } else {
2207                 int i, block_off = 0;
2208                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2209                 xe = xbs->here;
2210                 name_offset = le16_to_cpu(xe->xe_name_offset);
2211                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2212                 i = xbs->here - xbs->header->xh_entries;
2213                 old_in_xb = 1;
2214
2215                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2216                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2217                                                         bucket_xh(xbs->bucket),
2218                                                         i, &block_off,
2219                                                         &name_offset);
2220                         base = bucket_block(xbs->bucket, block_off);
2221                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2222                 } else {
2223                         base = xbs->base;
2224                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2225                 }
2226         }
2227
2228         /*
2229          * delete a xattr doesn't need metadata and cluster allocation.
2230          * so just calculate the credits and return.
2231          *
2232          * The credits for removing the value tree will be extended
2233          * by ocfs2_remove_extent itself.
2234          */
2235         if (!xi->value) {
2236                 if (!ocfs2_xattr_is_local(xe))
2237                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2238
2239                 goto out;
2240         }
2241
2242         /* do cluster allocation guess first. */
2243         value_size = le64_to_cpu(xe->xe_value_size);
2244
2245         if (old_in_xb) {
2246                 /*
2247                  * In xattr set, we always try to set the xe in inode first,
2248                  * so if it can be inserted into inode successfully, the old
2249                  * one will be removed from the xattr block, and this xattr
2250                  * will be inserted into inode as a new xattr in inode.
2251                  */
2252                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2253                         clusters_add += new_clusters;
2254                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2255                                     OCFS2_INODE_UPDATE_CREDITS;
2256                         if (!ocfs2_xattr_is_local(xe))
2257                                 credits += ocfs2_calc_extend_credits(
2258                                                         inode->i_sb,
2259                                                         &def_xv.xv.xr_list,
2260                                                         new_clusters);
2261                         goto out;
2262                 }
2263         }
2264
2265         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2266                 /* the new values will be stored outside. */
2267                 u32 old_clusters = 0;
2268
2269                 if (!ocfs2_xattr_is_local(xe)) {
2270                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2271                                                                  value_size);
2272                         xv = (struct ocfs2_xattr_value_root *)
2273                              (base + name_offset + name_len);
2274                         value_size = OCFS2_XATTR_ROOT_SIZE;
2275                 } else
2276                         xv = &def_xv.xv;
2277
2278                 if (old_clusters >= new_clusters) {
2279                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2280                         goto out;
2281                 } else {
2282                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2283                         clusters_add += new_clusters - old_clusters;
2284                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2285                                                              &xv->xr_list,
2286                                                              new_clusters -
2287                                                              old_clusters);
2288                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2289                                 goto out;
2290                 }
2291         } else {
2292                 /*
2293                  * Now the new value will be stored inside. So if the new
2294                  * value is smaller than the size of value root or the old
2295                  * value, we don't need any allocation, otherwise we have
2296                  * to guess metadata allocation.
2297                  */
2298                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2299                     (!ocfs2_xattr_is_local(xe) &&
2300                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2301                         goto out;
2302         }
2303
2304 meta_guess:
2305         /* calculate metadata allocation. */
2306         if (di->i_xattr_loc) {
2307                 if (!xbs->xattr_bh) {
2308                         ret = ocfs2_read_xattr_block(inode,
2309                                                      le64_to_cpu(di->i_xattr_loc),
2310                                                      &bh);
2311                         if (ret) {
2312                                 mlog_errno(ret);
2313                                 goto out;
2314                         }
2315
2316                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2317                 } else
2318                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2319
2320                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2321                         struct ocfs2_extent_list *el =
2322                                  &xb->xb_attrs.xb_root.xt_list;
2323                         meta_add += ocfs2_extend_meta_needed(el);
2324                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2325                                                              el, 1);
2326                 }
2327
2328                 /*
2329                  * This cluster will be used either for new bucket or for
2330                  * new xattr block.
2331                  * If the cluster size is the same as the bucket size, one
2332                  * more is needed since we may need to extend the bucket
2333                  * also.
2334                  */
2335                 clusters_add += 1;
2336                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2337                 if (OCFS2_XATTR_BUCKET_SIZE ==
2338                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2339                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2340                         clusters_add += 1;
2341                 }
2342         } else {
2343                 meta_add += 1;
2344                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2345         }
2346 out:
2347         if (clusters_need)
2348                 *clusters_need = clusters_add;
2349         if (meta_need)
2350                 *meta_need = meta_add;
2351         if (credits_need)
2352                 *credits_need = credits;
2353         brelse(bh);
2354         return ret;
2355 }
2356
2357 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2358                                      struct ocfs2_dinode *di,
2359                                      struct ocfs2_xattr_info *xi,
2360                                      struct ocfs2_xattr_search *xis,
2361                                      struct ocfs2_xattr_search *xbs,
2362                                      struct ocfs2_xattr_set_ctxt *ctxt,
2363                                      int *credits)
2364 {
2365         int clusters_add, meta_add, ret;
2366         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2367
2368         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2369
2370         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2371
2372         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2373                                         &clusters_add, &meta_add, credits);
2374         if (ret) {
2375                 mlog_errno(ret);
2376                 return ret;
2377         }
2378
2379         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2380              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2381
2382         if (meta_add) {
2383                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2384                                                         &ctxt->meta_ac);
2385                 if (ret) {
2386                         mlog_errno(ret);
2387                         goto out;
2388                 }
2389         }
2390
2391         if (clusters_add) {
2392                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2393                 if (ret)
2394                         mlog_errno(ret);
2395         }
2396 out:
2397         if (ret) {
2398                 if (ctxt->meta_ac) {
2399                         ocfs2_free_alloc_context(ctxt->meta_ac);
2400                         ctxt->meta_ac = NULL;
2401                 }
2402
2403                 /*
2404                  * We cannot have an error and a non null ctxt->data_ac.
2405                  */
2406         }
2407
2408         return ret;
2409 }
2410
2411 static int __ocfs2_xattr_set_handle(struct inode *inode,
2412                                     struct ocfs2_dinode *di,
2413                                     struct ocfs2_xattr_info *xi,
2414                                     struct ocfs2_xattr_search *xis,
2415                                     struct ocfs2_xattr_search *xbs,
2416                                     struct ocfs2_xattr_set_ctxt *ctxt)
2417 {
2418         int ret = 0, credits, old_found;
2419
2420         if (!xi->value) {
2421                 /* Remove existing extended attribute */
2422                 if (!xis->not_found)
2423                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2424                 else if (!xbs->not_found)
2425                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2426         } else {
2427                 /* We always try to set extended attribute into inode first*/
2428                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2429                 if (!ret && !xbs->not_found) {
2430                         /*
2431                          * If succeed and that extended attribute existing in
2432                          * external block, then we will remove it.
2433                          */
2434                         xi->value = NULL;
2435                         xi->value_len = 0;
2436
2437                         old_found = xis->not_found;
2438                         xis->not_found = -ENODATA;
2439                         ret = ocfs2_calc_xattr_set_need(inode,
2440                                                         di,
2441                                                         xi,
2442                                                         xis,
2443                                                         xbs,
2444                                                         NULL,
2445                                                         NULL,
2446                                                         &credits);
2447                         xis->not_found = old_found;
2448                         if (ret) {
2449                                 mlog_errno(ret);
2450                                 goto out;
2451                         }
2452
2453                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2454                                         ctxt->handle->h_buffer_credits);
2455                         if (ret) {
2456                                 mlog_errno(ret);
2457                                 goto out;
2458                         }
2459                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2460                 } else if (ret == -ENOSPC) {
2461                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2462                                 ret = ocfs2_xattr_block_find(inode,
2463                                                              xi->name_index,
2464                                                              xi->name, xbs);
2465                                 if (ret)
2466                                         goto out;
2467
2468                                 old_found = xis->not_found;
2469                                 xis->not_found = -ENODATA;
2470                                 ret = ocfs2_calc_xattr_set_need(inode,
2471                                                                 di,
2472                                                                 xi,
2473                                                                 xis,
2474                                                                 xbs,
2475                                                                 NULL,
2476                                                                 NULL,
2477                                                                 &credits);
2478                                 xis->not_found = old_found;
2479                                 if (ret) {
2480                                         mlog_errno(ret);
2481                                         goto out;
2482                                 }
2483
2484                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2485                                         ctxt->handle->h_buffer_credits);
2486                                 if (ret) {
2487                                         mlog_errno(ret);
2488                                         goto out;
2489                                 }
2490                         }
2491                         /*
2492                          * If no space in inode, we will set extended attribute
2493                          * into external block.
2494                          */
2495                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2496                         if (ret)
2497                                 goto out;
2498                         if (!xis->not_found) {
2499                                 /*
2500                                  * If succeed and that extended attribute
2501                                  * existing in inode, we will remove it.
2502                                  */
2503                                 xi->value = NULL;
2504                                 xi->value_len = 0;
2505                                 xbs->not_found = -ENODATA;
2506                                 ret = ocfs2_calc_xattr_set_need(inode,
2507                                                                 di,
2508                                                                 xi,
2509                                                                 xis,
2510                                                                 xbs,
2511                                                                 NULL,
2512                                                                 NULL,
2513                                                                 &credits);
2514                                 if (ret) {
2515                                         mlog_errno(ret);
2516                                         goto out;
2517                                 }
2518
2519                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2520                                                 ctxt->handle->h_buffer_credits);
2521                                 if (ret) {
2522                                         mlog_errno(ret);
2523                                         goto out;
2524                                 }
2525                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2526                                                             xis, ctxt);
2527                         }
2528                 }
2529         }
2530
2531 out:
2532         return ret;
2533 }
2534
2535 /*
2536  * This function only called duing creating inode
2537  * for init security/acl xattrs of the new inode.
2538  * The xattrs could be put into ibody or extent block,
2539  * xattr bucket would not be use in this case.
2540  * transanction credits also be reserved in here.
2541  */
2542 int ocfs2_xattr_set_handle(handle_t *handle,
2543                            struct inode *inode,
2544                            struct buffer_head *di_bh,
2545                            int name_index,
2546                            const char *name,
2547                            const void *value,
2548                            size_t value_len,
2549                            int flags,
2550                            struct ocfs2_alloc_context *meta_ac,
2551                            struct ocfs2_alloc_context *data_ac)
2552 {
2553         struct ocfs2_dinode *di;
2554         int ret;
2555
2556         struct ocfs2_xattr_info xi = {
2557                 .name_index = name_index,
2558                 .name = name,
2559                 .value = value,
2560                 .value_len = value_len,
2561         };
2562
2563         struct ocfs2_xattr_search xis = {
2564                 .not_found = -ENODATA,
2565         };
2566
2567         struct ocfs2_xattr_search xbs = {
2568                 .not_found = -ENODATA,
2569         };
2570
2571         struct ocfs2_xattr_set_ctxt ctxt = {
2572                 .handle = handle,
2573                 .meta_ac = meta_ac,
2574                 .data_ac = data_ac,
2575         };
2576
2577         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2578                 return -EOPNOTSUPP;
2579
2580         xis.inode_bh = xbs.inode_bh = di_bh;
2581         di = (struct ocfs2_dinode *)di_bh->b_data;
2582
2583         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2584
2585         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2586         if (ret)
2587                 goto cleanup;
2588         if (xis.not_found) {
2589                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2590                 if (ret)
2591                         goto cleanup;
2592         }
2593
2594         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2595
2596 cleanup:
2597         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2598         brelse(xbs.xattr_bh);
2599
2600         return ret;
2601 }
2602
2603 /*
2604  * ocfs2_xattr_set()
2605  *
2606  * Set, replace or remove an extended attribute for this inode.
2607  * value is NULL to remove an existing extended attribute, else either
2608  * create or replace an extended attribute.
2609  */
2610 int ocfs2_xattr_set(struct inode *inode,
2611                     int name_index,
2612                     const char *name,
2613                     const void *value,
2614                     size_t value_len,
2615                     int flags)
2616 {
2617         struct buffer_head *di_bh = NULL;
2618         struct ocfs2_dinode *di;
2619         int ret, credits;
2620         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2621         struct inode *tl_inode = osb->osb_tl_inode;
2622         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2623
2624         struct ocfs2_xattr_info xi = {
2625                 .name_index = name_index,
2626                 .name = name,
2627                 .value = value,
2628                 .value_len = value_len,
2629         };
2630
2631         struct ocfs2_xattr_search xis = {
2632                 .not_found = -ENODATA,
2633         };
2634
2635         struct ocfs2_xattr_search xbs = {
2636                 .not_found = -ENODATA,
2637         };
2638
2639         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2640                 return -EOPNOTSUPP;
2641
2642         /*
2643          * Only xbs will be used on indexed trees.  xis doesn't need a
2644          * bucket.
2645          */
2646         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2647         if (!xbs.bucket) {
2648                 mlog_errno(-ENOMEM);
2649                 return -ENOMEM;
2650         }
2651
2652         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2653         if (ret < 0) {
2654                 mlog_errno(ret);
2655                 goto cleanup_nolock;
2656         }
2657         xis.inode_bh = xbs.inode_bh = di_bh;
2658         di = (struct ocfs2_dinode *)di_bh->b_data;
2659
2660         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2661         /*
2662          * Scan inode and external block to find the same name
2663          * extended attribute and collect search infomation.
2664          */
2665         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2666         if (ret)
2667                 goto cleanup;
2668         if (xis.not_found) {
2669                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2670                 if (ret)
2671                         goto cleanup;
2672         }
2673
2674         if (xis.not_found && xbs.not_found) {
2675                 ret = -ENODATA;
2676                 if (flags & XATTR_REPLACE)
2677                         goto cleanup;
2678                 ret = 0;
2679                 if (!value)
2680                         goto cleanup;
2681         } else {
2682                 ret = -EEXIST;
2683                 if (flags & XATTR_CREATE)
2684                         goto cleanup;
2685         }
2686
2687
2688         mutex_lock(&tl_inode->i_mutex);
2689
2690         if (ocfs2_truncate_log_needs_flush(osb)) {
2691                 ret = __ocfs2_flush_truncate_log(osb);
2692                 if (ret < 0) {
2693                         mutex_unlock(&tl_inode->i_mutex);
2694                         mlog_errno(ret);
2695                         goto cleanup;
2696                 }
2697         }
2698         mutex_unlock(&tl_inode->i_mutex);
2699
2700         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2701                                         &xbs, &ctxt, &credits);
2702         if (ret) {
2703                 mlog_errno(ret);
2704                 goto cleanup;
2705         }
2706
2707         ctxt.handle = ocfs2_start_trans(osb, credits);
2708         if (IS_ERR(ctxt.handle)) {
2709                 ret = PTR_ERR(ctxt.handle);
2710                 mlog_errno(ret);
2711                 goto cleanup;
2712         }
2713
2714         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2715
2716         ocfs2_commit_trans(osb, ctxt.handle);
2717
2718         if (ctxt.data_ac)
2719                 ocfs2_free_alloc_context(ctxt.data_ac);
2720         if (ctxt.meta_ac)
2721                 ocfs2_free_alloc_context(ctxt.meta_ac);
2722         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2723                 ocfs2_schedule_truncate_log_flush(osb, 1);
2724         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2725 cleanup:
2726         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2727         ocfs2_inode_unlock(inode, 1);
2728 cleanup_nolock:
2729         brelse(di_bh);
2730         brelse(xbs.xattr_bh);
2731         ocfs2_xattr_bucket_free(xbs.bucket);
2732
2733         return ret;
2734 }
2735
2736 /*
2737  * Find the xattr extent rec which may contains name_hash.
2738  * e_cpos will be the first name hash of the xattr rec.
2739  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2740  */
2741 static int ocfs2_xattr_get_rec(struct inode *inode,
2742                                u32 name_hash,
2743                                u64 *p_blkno,
2744                                u32 *e_cpos,
2745                                u32 *num_clusters,
2746                                struct ocfs2_extent_list *el)
2747 {
2748         int ret = 0, i;
2749         struct buffer_head *eb_bh = NULL;
2750         struct ocfs2_extent_block *eb;
2751         struct ocfs2_extent_rec *rec = NULL;
2752         u64 e_blkno = 0;
2753
2754         if (el->l_tree_depth) {
2755                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2756                 if (ret) {
2757                         mlog_errno(ret);
2758                         goto out;
2759                 }
2760
2761                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2762                 el = &eb->h_list;
2763
2764                 if (el->l_tree_depth) {
2765                         ocfs2_error(inode->i_sb,
2766                                     "Inode %lu has non zero tree depth in "
2767                                     "xattr tree block %llu\n", inode->i_ino,
2768                                     (unsigned long long)eb_bh->b_blocknr);
2769                         ret = -EROFS;
2770                         goto out;
2771                 }
2772         }
2773
2774         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2775                 rec = &el->l_recs[i];
2776
2777                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2778                         e_blkno = le64_to_cpu(rec->e_blkno);
2779                         break;
2780                 }
2781         }
2782
2783         if (!e_blkno) {
2784                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2785                             "record (%u, %u, 0) in xattr", inode->i_ino,
2786                             le32_to_cpu(rec->e_cpos),
2787                             ocfs2_rec_clusters(el, rec));
2788                 ret = -EROFS;
2789                 goto out;
2790         }
2791
2792         *p_blkno = le64_to_cpu(rec->e_blkno);
2793         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2794         if (e_cpos)
2795                 *e_cpos = le32_to_cpu(rec->e_cpos);
2796 out:
2797         brelse(eb_bh);
2798         return ret;
2799 }
2800
2801 typedef int (xattr_bucket_func)(struct inode *inode,
2802                                 struct ocfs2_xattr_bucket *bucket,
2803                                 void *para);
2804
2805 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2806                                    struct ocfs2_xattr_bucket *bucket,
2807                                    int name_index,
2808                                    const char *name,
2809                                    u32 name_hash,
2810                                    u16 *xe_index,
2811                                    int *found)
2812 {
2813         int i, ret = 0, cmp = 1, block_off, new_offset;
2814         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2815         size_t name_len = strlen(name);
2816         struct ocfs2_xattr_entry *xe = NULL;
2817         char *xe_name;
2818
2819         /*
2820          * We don't use binary search in the bucket because there
2821          * may be multiple entries with the same name hash.
2822          */
2823         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2824                 xe = &xh->xh_entries[i];
2825
2826                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2827                         continue;
2828                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2829                         break;
2830
2831                 cmp = name_index - ocfs2_xattr_get_type(xe);
2832                 if (!cmp)
2833                         cmp = name_len - xe->xe_name_len;
2834                 if (cmp)
2835                         continue;
2836
2837                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2838                                                         xh,
2839                                                         i,
2840                                                         &block_off,
2841                                                         &new_offset);
2842                 if (ret) {
2843                         mlog_errno(ret);
2844                         break;
2845                 }
2846
2847
2848                 xe_name = bucket_block(bucket, block_off) + new_offset;
2849                 if (!memcmp(name, xe_name, name_len)) {
2850                         *xe_index = i;
2851                         *found = 1;
2852                         ret = 0;
2853                         break;
2854                 }
2855         }
2856
2857         return ret;
2858 }
2859
2860 /*
2861  * Find the specified xattr entry in a series of buckets.
2862  * This series start from p_blkno and last for num_clusters.
2863  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2864  * the num of the valid buckets.
2865  *
2866  * Return the buffer_head this xattr should reside in. And if the xattr's
2867  * hash is in the gap of 2 buckets, return the lower bucket.
2868  */
2869 static int ocfs2_xattr_bucket_find(struct inode *inode,
2870                                    int name_index,
2871                                    const char *name,
2872                                    u32 name_hash,
2873                                    u64 p_blkno,
2874                                    u32 first_hash,
2875                                    u32 num_clusters,
2876                                    struct ocfs2_xattr_search *xs)
2877 {
2878         int ret, found = 0;
2879         struct ocfs2_xattr_header *xh = NULL;
2880         struct ocfs2_xattr_entry *xe = NULL;
2881         u16 index = 0;
2882         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2883         int low_bucket = 0, bucket, high_bucket;
2884         struct ocfs2_xattr_bucket *search;
2885         u32 last_hash;
2886         u64 blkno, lower_blkno = 0;
2887
2888         search = ocfs2_xattr_bucket_new(inode);
2889         if (!search) {
2890                 ret = -ENOMEM;
2891                 mlog_errno(ret);
2892                 goto out;
2893         }
2894
2895         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2896         if (ret) {
2897                 mlog_errno(ret);
2898                 goto out;
2899         }
2900
2901         xh = bucket_xh(search);
2902         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2903         while (low_bucket <= high_bucket) {
2904                 ocfs2_xattr_bucket_relse(search);
2905
2906                 bucket = (low_bucket + high_bucket) / 2;
2907                 blkno = p_blkno + bucket * blk_per_bucket;
2908                 ret = ocfs2_read_xattr_bucket(search, blkno);
2909                 if (ret) {
2910                         mlog_errno(ret);
2911                         goto out;
2912                 }
2913
2914                 xh = bucket_xh(search);
2915                 xe = &xh->xh_entries[0];
2916                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2917                         high_bucket = bucket - 1;
2918                         continue;
2919                 }
2920
2921                 /*
2922                  * Check whether the hash of the last entry in our
2923                  * bucket is larger than the search one. for an empty
2924                  * bucket, the last one is also the first one.
2925                  */
2926                 if (xh->xh_count)
2927                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2928
2929                 last_hash = le32_to_cpu(xe->xe_name_hash);
2930
2931                 /* record lower_blkno which may be the insert place. */
2932                 lower_blkno = blkno;
2933
2934                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2935                         low_bucket = bucket + 1;
2936                         continue;
2937                 }
2938
2939                 /* the searched xattr should reside in this bucket if exists. */
2940                 ret = ocfs2_find_xe_in_bucket(inode, search,
2941                                               name_index, name, name_hash,
2942                                               &index, &found);
2943                 if (ret) {
2944                         mlog_errno(ret);
2945                         goto out;
2946                 }
2947                 break;
2948         }
2949
2950         /*
2951          * Record the bucket we have found.
2952          * When the xattr's hash value is in the gap of 2 buckets, we will
2953          * always set it to the previous bucket.
2954          */
2955         if (!lower_blkno)
2956                 lower_blkno = p_blkno;
2957
2958         /* This should be in cache - we just read it during the search */
2959         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
2960         if (ret) {
2961                 mlog_errno(ret);
2962                 goto out;
2963         }
2964
2965         xs->header = bucket_xh(xs->bucket);
2966         xs->base = bucket_block(xs->bucket, 0);
2967         xs->end = xs->base + inode->i_sb->s_blocksize;
2968
2969         if (found) {
2970                 xs->here = &xs->header->xh_entries[index];
2971                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2972                      (unsigned long long)bucket_blkno(xs->bucket), index);
2973         } else
2974                 ret = -ENODATA;
2975
2976 out:
2977         ocfs2_xattr_bucket_free(search);
2978         return ret;
2979 }
2980
2981 static int ocfs2_xattr_index_block_find(struct inode *inode,
2982                                         struct buffer_head *root_bh,
2983                                         int name_index,
2984                                         const char *name,
2985                                         struct ocfs2_xattr_search *xs)
2986 {
2987         int ret;
2988         struct ocfs2_xattr_block *xb =
2989                         (struct ocfs2_xattr_block *)root_bh->b_data;
2990         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2991         struct ocfs2_extent_list *el = &xb_root->xt_list;
2992         u64 p_blkno = 0;
2993         u32 first_hash, num_clusters = 0;
2994         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2995
2996         if (le16_to_cpu(el->l_next_free_rec) == 0)
2997                 return -ENODATA;
2998
2999         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3000              name, name_hash, name_index);
3001
3002         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3003                                   &num_clusters, el);
3004         if (ret) {
3005                 mlog_errno(ret);
3006                 goto out;
3007         }
3008
3009         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3010
3011         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3012              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3013              first_hash);
3014
3015         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3016                                       p_blkno, first_hash, num_clusters, xs);
3017
3018 out:
3019         return ret;
3020 }
3021
3022 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3023                                        u64 blkno,
3024                                        u32 clusters,
3025                                        xattr_bucket_func *func,
3026                                        void *para)
3027 {
3028         int i, ret = 0;
3029         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3030         u32 num_buckets = clusters * bpc;
3031         struct ocfs2_xattr_bucket *bucket;
3032
3033         bucket = ocfs2_xattr_bucket_new(inode);
3034         if (!bucket) {
3035                 mlog_errno(-ENOMEM);
3036                 return -ENOMEM;
3037         }
3038
3039         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3040              clusters, (unsigned long long)blkno);
3041
3042         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3043                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3044                 if (ret) {
3045                         mlog_errno(ret);
3046                         break;
3047                 }
3048
3049                 /*
3050                  * The real bucket num in this series of blocks is stored
3051                  * in the 1st bucket.
3052                  */
3053                 if (i == 0)
3054                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3055
3056                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3057                      (unsigned long long)blkno,
3058                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3059                 if (func) {
3060                         ret = func(inode, bucket, para);
3061                         if (ret)
3062                                 mlog_errno(ret);
3063                         /* Fall through to bucket_relse() */
3064                 }
3065
3066                 ocfs2_xattr_bucket_relse(bucket);
3067                 if (ret)
3068                         break;
3069         }
3070
3071         ocfs2_xattr_bucket_free(bucket);
3072         return ret;
3073 }
3074
3075 struct ocfs2_xattr_tree_list {
3076         char *buffer;
3077         size_t buffer_size;
3078         size_t result;
3079 };
3080
3081 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3082                                              struct ocfs2_xattr_header *xh,
3083                                              int index,
3084                                              int *block_off,
3085                                              int *new_offset)
3086 {
3087         u16 name_offset;
3088
3089         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3090                 return -EINVAL;
3091
3092         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3093
3094         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3095         *new_offset = name_offset % inode->i_sb->s_blocksize;
3096
3097         return 0;
3098 }
3099
3100 static int ocfs2_list_xattr_bucket(struct inode *inode,
3101                                    struct ocfs2_xattr_bucket *bucket,
3102                                    void *para)
3103 {
3104         int ret = 0, type;
3105         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3106         int i, block_off, new_offset;
3107         const char *prefix, *name;
3108
3109         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3110                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3111                 type = ocfs2_xattr_get_type(entry);
3112                 prefix = ocfs2_xattr_prefix(type);
3113
3114                 if (prefix) {
3115                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3116                                                                 bucket_xh(bucket),
3117                                                                 i,
3118                                                                 &block_off,
3119                                                                 &new_offset);
3120                         if (ret)
3121                                 break;
3122
3123                         name = (const char *)bucket_block(bucket, block_off) +
3124                                 new_offset;
3125                         ret = ocfs2_xattr_list_entry(xl->buffer,
3126                                                      xl->buffer_size,
3127                                                      &xl->result,
3128                                                      prefix, name,
3129                                                      entry->xe_name_len);
3130                         if (ret)
3131                                 break;
3132                 }
3133         }
3134
3135         return ret;
3136 }
3137
3138 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3139                                              struct ocfs2_xattr_tree_root *xt,
3140                                              char *buffer,
3141                                              size_t buffer_size)
3142 {
3143         struct ocfs2_extent_list *el = &xt->xt_list;
3144         int ret = 0;
3145         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3146         u64 p_blkno = 0;
3147         struct ocfs2_xattr_tree_list xl = {
3148                 .buffer = buffer,
3149                 .buffer_size = buffer_size,
3150                 .result = 0,
3151         };
3152
3153         if (le16_to_cpu(el->l_next_free_rec) == 0)
3154                 return 0;
3155
3156         while (name_hash > 0) {
3157                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3158                                           &e_cpos, &num_clusters, el);
3159                 if (ret) {
3160                         mlog_errno(ret);
3161                         goto out;
3162                 }
3163
3164                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3165                                                   ocfs2_list_xattr_bucket,
3166                                                   &xl);
3167                 if (ret) {
3168                         mlog_errno(ret);
3169                         goto out;
3170                 }
3171
3172                 if (e_cpos == 0)
3173                         break;
3174
3175                 name_hash = e_cpos - 1;
3176         }
3177
3178         ret = xl.result;
3179 out:
3180         return ret;
3181 }
3182
3183 static int cmp_xe(const void *a, const void *b)
3184 {
3185         const struct ocfs2_xattr_entry *l = a, *r = b;
3186         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3187         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3188
3189         if (l_hash > r_hash)
3190                 return 1;
3191         if (l_hash < r_hash)
3192                 return -1;
3193         return 0;
3194 }
3195
3196 static void swap_xe(void *a, void *b, int size)
3197 {
3198         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3199
3200         tmp = *l;
3201         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3202         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3203 }
3204
3205 /*
3206  * When the ocfs2_xattr_block is filled up, new bucket will be created
3207  * and all the xattr entries will be moved to the new bucket.
3208  * The header goes at the start of the bucket, and the names+values are
3209  * filled from the end.  This is why *target starts as the last buffer.
3210  * Note: we need to sort the entries since they are not saved in order
3211  * in the ocfs2_xattr_block.
3212  */
3213 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3214                                            struct buffer_head *xb_bh,
3215                                            struct ocfs2_xattr_bucket *bucket)
3216 {
3217         int i, blocksize = inode->i_sb->s_blocksize;
3218         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3219         u16 offset, size, off_change;
3220         struct ocfs2_xattr_entry *xe;
3221         struct ocfs2_xattr_block *xb =
3222                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3223         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3224         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3225         u16 count = le16_to_cpu(xb_xh->xh_count);
3226         char *src = xb_bh->b_data;
3227         char *target = bucket_block(bucket, blks - 1);
3228
3229         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3230              (unsigned long long)xb_bh->b_blocknr,
3231              (unsigned long long)bucket_blkno(bucket));
3232
3233         for (i = 0; i < blks; i++)
3234                 memset(bucket_block(bucket, i), 0, blocksize);
3235
3236         /*
3237          * Since the xe_name_offset is based on ocfs2_xattr_header,
3238          * there is a offset change corresponding to the change of
3239          * ocfs2_xattr_header's position.
3240          */
3241         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3242         xe = &xb_xh->xh_entries[count - 1];
3243         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3244         size = blocksize - offset;
3245
3246         /* copy all the names and values. */
3247         memcpy(target + offset, src + offset, size);
3248
3249         /* Init new header now. */
3250         xh->xh_count = xb_xh->xh_count;
3251         xh->xh_num_buckets = cpu_to_le16(1);
3252         xh->xh_name_value_len = cpu_to_le16(size);
3253         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3254
3255         /* copy all the entries. */
3256         target = bucket_block(bucket, 0);
3257         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3258         size = count * sizeof(struct ocfs2_xattr_entry);
3259         memcpy(target + offset, (char *)xb_xh + offset, size);
3260
3261         /* Change the xe offset for all the xe because of the move. */
3262         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3263                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3264         for (i = 0; i < count; i++)
3265                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3266
3267         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3268              offset, size, off_change);
3269
3270         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3271              cmp_xe, swap_xe);
3272 }
3273
3274 /*
3275  * After we move xattr from block to index btree, we have to
3276  * update ocfs2_xattr_search to the new xe and base.
3277  *
3278  * When the entry is in xattr block, xattr_bh indicates the storage place.
3279  * While if the entry is in index b-tree, "bucket" indicates the
3280  * real place of the xattr.
3281  */
3282 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3283                                             struct ocfs2_xattr_search *xs,
3284                                             struct buffer_head *old_bh)
3285 {
3286         char *buf = old_bh->b_data;
3287         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3288         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3289         int i;
3290
3291         xs->header = bucket_xh(xs->bucket);
3292         xs->base = bucket_block(xs->bucket, 0);
3293         xs->end = xs->base + inode->i_sb->s_blocksize;
3294
3295         if (xs->not_found)
3296                 return;
3297
3298         i = xs->here - old_xh->xh_entries;
3299         xs->here = &xs->header->xh_entries[i];
3300 }
3301
3302 static int ocfs2_xattr_create_index_block(struct inode *inode,
3303                                           struct ocfs2_xattr_search *xs,
3304                                           struct ocfs2_xattr_set_ctxt *ctxt)
3305 {
3306         int ret;
3307         u32 bit_off, len;
3308         u64 blkno;
3309         handle_t *handle = ctxt->handle;
3310         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3311         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3312         struct buffer_head *xb_bh = xs->xattr_bh;
3313         struct ocfs2_xattr_block *xb =
3314                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3315         struct ocfs2_xattr_tree_root *xr;
3316         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3317
3318         mlog(0, "create xattr index block for %llu\n",
3319              (unsigned long long)xb_bh->b_blocknr);
3320
3321         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3322         BUG_ON(!xs->bucket);
3323
3324         /*
3325          * XXX:
3326          * We can use this lock for now, and maybe move to a dedicated mutex
3327          * if performance becomes a problem later.
3328          */
3329         down_write(&oi->ip_alloc_sem);
3330
3331         ret = ocfs2_journal_access(handle, inode, xb_bh,
3332                                    OCFS2_JOURNAL_ACCESS_WRITE);
3333         if (ret) {
3334                 mlog_errno(ret);
3335                 goto out;
3336         }
3337
3338         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3339                                      1, 1, &bit_off, &len);
3340         if (ret) {
3341                 mlog_errno(ret);
3342                 goto out;
3343         }
3344
3345         /*
3346          * The bucket may spread in many blocks, and
3347          * we will only touch the 1st block and the last block
3348          * in the whole bucket(one for entry and one for data).
3349          */
3350         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3351
3352         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3353              (unsigned long long)blkno);
3354
3355         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3356         if (ret) {
3357                 mlog_errno(ret);
3358                 goto out;
3359         }
3360
3361         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3362                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3363         if (ret) {
3364                 mlog_errno(ret);
3365                 goto out;
3366         }
3367
3368         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3369         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3370
3371         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3372
3373         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3374         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3375                offsetof(struct ocfs2_xattr_block, xb_attrs));
3376
3377         xr = &xb->xb_attrs.xb_root;
3378         xr->xt_clusters = cpu_to_le32(1);
3379         xr->xt_last_eb_blk = 0;
3380         xr->xt_list.l_tree_depth = 0;
3381         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3382         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3383
3384         xr->xt_list.l_recs[0].e_cpos = 0;
3385         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3386         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3387
3388         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3389
3390         ocfs2_journal_dirty(handle, xb_bh);
3391
3392 out:
3393         up_write(&oi->ip_alloc_sem);
3394
3395         return ret;
3396 }
3397
3398 static int cmp_xe_offset(const void *a, const void *b)
3399 {
3400         const struct ocfs2_xattr_entry *l = a, *r = b;
3401         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3402         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3403
3404         if (l_name_offset < r_name_offset)
3405                 return 1;
3406         if (l_name_offset > r_name_offset)
3407                 return -1;
3408         return 0;
3409 }
3410
3411 /*
3412  * defrag a xattr bucket if we find that the bucket has some
3413  * holes beteen name/value pairs.
3414  * We will move all the name/value pairs to the end of the bucket
3415  * so that we can spare some space for insertion.
3416  */
3417 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3418                                      handle_t *handle,
3419                                      struct ocfs2_xattr_bucket *bucket)
3420 {
3421         int ret, i;
3422         size_t end, offset, len, value_len;
3423         struct ocfs2_xattr_header *xh;
3424         char *entries, *buf, *bucket_buf = NULL;
3425         u64 blkno = bucket_blkno(bucket);
3426         u16 xh_free_start;
3427         size_t blocksize = inode->i_sb->s_blocksize;
3428         struct ocfs2_xattr_entry *xe;
3429
3430         /*
3431          * In order to make the operation more efficient and generic,
3432          * we copy all the blocks into a contiguous memory and do the
3433          * defragment there, so if anything is error, we will not touch
3434          * the real block.
3435          */
3436         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3437         if (!bucket_buf) {
3438                 ret = -EIO;
3439                 goto out;
3440         }
3441
3442         buf = bucket_buf;
3443         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3444                 memcpy(buf, bucket_block(bucket, i), blocksize);
3445
3446         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3447                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3448         if (ret < 0) {
3449                 mlog_errno(ret);
3450                 goto out;
3451         }
3452
3453         xh = (struct ocfs2_xattr_header *)bucket_buf;
3454         entries = (char *)xh->xh_entries;
3455         xh_free_start = le16_to_cpu(xh->xh_free_start);
3456
3457         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3458              "xh_free_start = %u, xh_name_value_len = %u.\n",
3459              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3460              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3461
3462         /*
3463          * sort all the entries by their offset.
3464          * the largest will be the first, so that we can
3465          * move them to the end one by one.
3466          */
3467         sort(entries, le16_to_cpu(xh->xh_count),
3468              sizeof(struct ocfs2_xattr_entry),
3469              cmp_xe_offset, swap_xe);
3470
3471         /* Move all name/values to the end of the bucket. */
3472         xe = xh->xh_entries;
3473         end = OCFS2_XATTR_BUCKET_SIZE;
3474         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3475                 offset = le16_to_cpu(xe->xe_name_offset);
3476                 if (ocfs2_xattr_is_local(xe))
3477                         value_len = OCFS2_XATTR_SIZE(
3478                                         le64_to_cpu(xe->xe_value_size));
3479                 else
3480                         value_len = OCFS2_XATTR_ROOT_SIZE;
3481                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3482
3483                 /*
3484                  * We must make sure that the name/value pair
3485                  * exist in the same block. So adjust end to
3486                  * the previous block end if needed.
3487                  */
3488                 if (((end - len) / blocksize !=
3489                         (end - 1) / blocksize))
3490                         end = end - end % blocksize;
3491
3492                 if (end > offset + len) {
3493                         memmove(bucket_buf + end - len,
3494                                 bucket_buf + offset, len);
3495                         xe->xe_name_offset = cpu_to_le16(end - len);
3496                 }
3497
3498                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3499                                 "bucket %llu\n", (unsigned long long)blkno);
3500
3501                 end -= len;
3502         }
3503
3504         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3505                         "bucket %llu\n", (unsigned long long)blkno);
3506
3507         if (xh_free_start == end)
3508                 goto out;
3509
3510         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3511         xh->xh_free_start = cpu_to_le16(end);
3512
3513         /* sort the entries by their name_hash. */
3514         sort(entries, le16_to_cpu(xh->xh_count),
3515              sizeof(struct ocfs2_xattr_entry),
3516              cmp_xe, swap_xe);
3517
3518         buf = bucket_buf;
3519         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3520                 memcpy(bucket_block(bucket, i), buf, blocksize);
3521         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3522
3523 out:
3524         kfree(bucket_buf);
3525         return ret;
3526 }
3527
3528 /*
3529  * Move half nums of the xattr bucket in the previous cluster to this new
3530  * cluster. We only touch the last cluster of the previous extend record.
3531  *
3532  * first_bh is the first buffer_head of a series of bucket in the same
3533  * extent rec and header_bh is the header of one bucket in this cluster.
3534  * They will be updated if we move the data header_bh contains to the new
3535  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3536  */
3537 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3538                                                handle_t *handle,
3539                                                struct buffer_head **first_bh,
3540                                                struct buffer_head **header_bh,
3541                                                u64 new_blkno,
3542                                                u64 prev_blkno,
3543                                                u32 num_clusters,
3544                                                u32 *first_hash)
3545 {
3546         int i, ret, credits;
3547         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3548         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3549         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3550         int blocksize = inode->i_sb->s_blocksize;
3551         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3552         struct ocfs2_xattr_header *new_xh;
3553         struct ocfs2_xattr_header *xh =
3554                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3555
3556         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3557         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3558
3559         prev_bh = *first_bh;
3560         get_bh(prev_bh);
3561         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3562
3563         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3564
3565         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3566              (unsigned long long)prev_blkno, (unsigned long long)new_blkno);
3567
3568         /*
3569          * We need to update the 1st half of the new cluster and
3570          * 1 more for the update of the 1st bucket of the previous
3571          * extent record.
3572          */
3573         credits = bpc / 2 + 1 + handle->h_buffer_credits;
3574         ret = ocfs2_extend_trans(handle, credits);
3575         if (ret) {
3576                 mlog_errno(ret);
3577                 goto out;
3578         }
3579
3580         ret = ocfs2_journal_access(handle, inode, prev_bh,
3581                                    OCFS2_JOURNAL_ACCESS_WRITE);
3582         if (ret) {
3583                 mlog_errno(ret);
3584                 goto out;
3585         }
3586
3587         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3588                 old_bh = new_bh = NULL;
3589                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3590                 if (!new_bh) {
3591                         ret = -EIO;
3592                         mlog_errno(ret);
3593                         goto out;
3594                 }
3595
3596                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3597
3598                 ret = ocfs2_journal_access(handle, inode, new_bh,
3599                                            OCFS2_JOURNAL_ACCESS_CREATE);
3600                 if (ret < 0) {
3601                         mlog_errno(ret);
3602                         brelse(new_bh);
3603                         goto out;
3604                 }
3605
3606                 ret = ocfs2_read_block(inode, prev_blkno, &old_bh, NULL);
3607                 if (ret < 0) {
3608                         mlog_errno(ret);
3609                         brelse(new_bh);
3610                         goto out;
3611                 }
3612
3613                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3614
3615                 if (i == 0) {
3616                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3617                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3618
3619                         if (first_hash)
3620                                 *first_hash = le32_to_cpu(
3621                                         new_xh->xh_entries[0].xe_name_hash);
3622                         new_first_bh = new_bh;
3623                         get_bh(new_first_bh);
3624                 }
3625
3626                 ocfs2_journal_dirty(handle, new_bh);
3627
3628                 if (*header_bh == old_bh) {
3629                         brelse(*header_bh);
3630                         *header_bh = new_bh;
3631                         get_bh(*header_bh);
3632
3633                         brelse(*first_bh);
3634                         *first_bh = new_first_bh;
3635                         get_bh(*first_bh);
3636                 }
3637                 brelse(new_bh);
3638                 brelse(old_bh);
3639         }
3640
3641         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3642
3643         ocfs2_journal_dirty(handle, prev_bh);
3644 out:
3645         brelse(prev_bh);
3646         brelse(new_first_bh);
3647         return ret;
3648 }
3649
3650 /*
3651  * Find the suitable pos when we divide a bucket into 2.
3652  * We have to make sure the xattrs with the same hash value exist
3653  * in the same bucket.
3654  *
3655  * If this ocfs2_xattr_header covers more than one hash value, find a
3656  * place where the hash value changes.  Try to find the most even split.
3657  * The most common case is that all entries have different hash values,
3658  * and the first check we make will find a place to split.
3659  */
3660 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3661 {
3662         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3663         int count = le16_to_cpu(xh->xh_count);
3664         int delta, middle = count / 2;
3665
3666         /*
3667          * We start at the middle.  Each step gets farther away in both
3668          * directions.  We therefore hit the change in hash value
3669          * nearest to the middle.  Note that this loop does not execute for
3670          * count < 2.
3671          */
3672         for (delta = 0; delta < middle; delta++) {
3673                 /* Let's check delta earlier than middle */
3674                 if (cmp_xe(&entries[middle - delta - 1],
3675                            &entries[middle - delta]))
3676                         return middle - delta;
3677
3678                 /* For even counts, don't walk off the end */
3679                 if ((middle + delta + 1) == count)
3680                         continue;
3681
3682                 /* Now try delta past middle */
3683                 if (cmp_xe(&entries[middle + delta],
3684                            &entries[middle + delta + 1]))
3685                         return middle + delta + 1;
3686         }
3687
3688         /* Every entry had the same hash */
3689         return count;
3690 }
3691
3692 /*
3693  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3694  * first_hash will record the 1st hash of the new bucket.
3695  *
3696  * Normally half of the xattrs will be moved.  But we have to make
3697  * sure that the xattrs with the same hash value are stored in the
3698  * same bucket. If all the xattrs in this bucket have the same hash
3699  * value, the new bucket will be initialized as an empty one and the
3700  * first_hash will be initialized as (hash_value+1).
3701  */
3702 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3703                                     handle_t *handle,
3704                                     u64 blk,
3705                                     u64 new_blk,
3706                                     u32 *first_hash,
3707                                     int new_bucket_head)
3708 {
3709         int ret, i;
3710         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3711         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3712         struct ocfs2_xattr_header *xh;
3713         struct ocfs2_xattr_entry *xe;
3714         int blocksize = inode->i_sb->s_blocksize;
3715
3716         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3717              (unsigned long long)blk, (unsigned long long)new_blk);
3718
3719         s_bucket = ocfs2_xattr_bucket_new(inode);
3720         t_bucket = ocfs2_xattr_bucket_new(inode);
3721         if (!s_bucket || !t_bucket) {
3722                 ret = -ENOMEM;
3723                 mlog_errno(ret);
3724                 goto out;
3725         }
3726
3727         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3728         if (ret) {
3729                 mlog_errno(ret);
3730                 goto out;
3731         }
3732
3733         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3734                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3735         if (ret) {
3736                 mlog_errno(ret);
3737                 goto out;
3738         }
3739
3740         /*
3741          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3742          * there's no need to read it.
3743          */
3744         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3745         if (ret) {
3746                 mlog_errno(ret);
3747                 goto out;
3748         }
3749
3750         /*
3751          * Hey, if we're overwriting t_bucket, what difference does
3752          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3753          * same part of ocfs2_cp_xattr_bucket().
3754          */
3755         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3756                                                 new_bucket_head ?
3757                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3758                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3759         if (ret) {
3760                 mlog_errno(ret);
3761                 goto out;
3762         }
3763
3764         xh = bucket_xh(s_bucket);
3765         count = le16_to_cpu(xh->xh_count);
3766         start = ocfs2_xattr_find_divide_pos(xh);
3767
3768         if (start == count) {
3769                 xe = &xh->xh_entries[start-1];
3770
3771                 /*
3772                  * initialized a new empty bucket here.
3773                  * The hash value is set as one larger than
3774                  * that of the last entry in the previous bucket.
3775                  */
3776                 for (i = 0; i < t_bucket->bu_blocks; i++)
3777                         memset(bucket_block(t_bucket, i), 0, blocksize);
3778
3779                 xh = bucket_xh(t_bucket);
3780                 xh->xh_free_start = cpu_to_le16(blocksize);
3781                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3782                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3783
3784                 goto set_num_buckets;
3785         }
3786
3787         /* copy the whole bucket to the new first. */
3788         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3789
3790         /* update the new bucket. */
3791         xh = bucket_xh(t_bucket);
3792
3793         /*
3794          * Calculate the total name/value len and xh_free_start for
3795          * the old bucket first.
3796          */
3797         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3798         name_value_len = 0;
3799         for (i = 0; i < start; 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                 name_value_len += xe_len;
3808                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3809                         name_offset = le16_to_cpu(xe->xe_name_offset);
3810         }
3811
3812         /*
3813          * Now begin the modification to the new bucket.
3814          *
3815          * In the new bucket, We just move the xattr entry to the beginning
3816          * and don't touch the name/value. So there will be some holes in the
3817          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3818          * called.
3819          */
3820         xe = &xh->xh_entries[start];
3821         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3822         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3823              (int)((char *)xe - (char *)xh),
3824              (int)((char *)xh->xh_entries - (char *)xh));
3825         memmove((char *)xh->xh_entries, (char *)xe, len);
3826         xe = &xh->xh_entries[count - start];
3827         len = sizeof(struct ocfs2_xattr_entry) * start;
3828         memset((char *)xe, 0, len);
3829
3830         le16_add_cpu(&xh->xh_count, -start);
3831         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3832
3833         /* Calculate xh_free_start for the new bucket. */
3834         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3835         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3836                 xe = &xh->xh_entries[i];
3837                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3838                 if (ocfs2_xattr_is_local(xe))
3839                         xe_len +=
3840                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3841                 else
3842                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3843                 if (le16_to_cpu(xe->xe_name_offset) <
3844                     le16_to_cpu(xh->xh_free_start))
3845                         xh->xh_free_start = xe->xe_name_offset;
3846         }
3847
3848 set_num_buckets:
3849         /* set xh->xh_num_buckets for the new xh. */
3850         if (new_bucket_head)
3851                 xh->xh_num_buckets = cpu_to_le16(1);
3852         else
3853                 xh->xh_num_buckets = 0;
3854
3855         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3856
3857         /* store the first_hash of the new bucket. */
3858         if (first_hash)
3859                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3860
3861         /*
3862          * Now only update the 1st block of the old bucket.  If we
3863          * just added a new empty bucket, there is no need to modify
3864          * it.
3865          */
3866         if (start == count)
3867                 goto out;
3868
3869         xh = bucket_xh(s_bucket);
3870         memset(&xh->xh_entries[start], 0,
3871                sizeof(struct ocfs2_xattr_entry) * (count - start));
3872         xh->xh_count = cpu_to_le16(start);
3873         xh->xh_free_start = cpu_to_le16(name_offset);
3874         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3875
3876         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3877
3878 out:
3879         ocfs2_xattr_bucket_free(s_bucket);
3880         ocfs2_xattr_bucket_free(t_bucket);
3881
3882         return ret;
3883 }
3884
3885 /*
3886  * Copy xattr from one bucket to another bucket.
3887  *
3888  * The caller must make sure that the journal transaction
3889  * has enough space for journaling.
3890  */
3891 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3892                                  handle_t *handle,
3893                                  u64 s_blkno,
3894                                  u64 t_blkno,
3895                                  int t_is_new)
3896 {
3897         int ret;
3898         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3899
3900         BUG_ON(s_blkno == t_blkno);
3901
3902         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3903              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3904              t_is_new);
3905
3906         s_bucket = ocfs2_xattr_bucket_new(inode);
3907         t_bucket = ocfs2_xattr_bucket_new(inode);
3908         if (!s_bucket || !t_bucket) {
3909                 ret = -ENOMEM;
3910                 mlog_errno(ret);
3911                 goto out;
3912         }
3913
3914         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3915         if (ret)
3916                 goto out;
3917
3918         /*
3919          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3920          * there's no need to read it.
3921          */
3922         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3923         if (ret)
3924                 goto out;
3925
3926         /*
3927          * Hey, if we're overwriting t_bucket, what difference does
3928          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3929          * cluster to fill, we came here from ocfs2_cp_xattr_cluster(), and
3930          * it is really new - ACCESS_CREATE is required.  But we also
3931          * might have moved data out of t_bucket before extending back
3932          * into it.  ocfs2_add_new_xattr_bucket() can do this - its call
3933          * to ocfs2_add_new_xattr_cluster() may have created a new extent
3934          * and copied out the end of the old extent.  Then it re-extends
3935          * the old extent back to create space for new xattrs.  That's
3936          * how we get here, and the bucket isn't really new.
3937          */
3938         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3939                                                 t_is_new ?
3940                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3941                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3942         if (ret)
3943                 goto out;
3944
3945         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3946         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3947
3948 out:
3949         ocfs2_xattr_bucket_free(t_bucket);
3950         ocfs2_xattr_bucket_free(s_bucket);
3951
3952         return ret;
3953 }
3954
3955 /*
3956  * src_blk points to the last cluster of an existing extent.  to_blk
3957  * points to a newly allocated extent.  We copy the cluster over to the
3958  * new extent, initializing its xh_num_buckets.  The old extent's
3959  * xh_num_buckets shrinks by the same amount.
3960  */
3961 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3962                                   handle_t *handle,
3963                                   struct buffer_head *first_bh,
3964                                   u64 src_blk,
3965                                   u64 to_blk,
3966                                   u32 *first_hash)
3967 {
3968         int i, ret, credits;
3969         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3970         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3971         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3972         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3973         struct ocfs2_xattr_bucket *old_first, *new_first;
3974
3975         mlog(0, "cp xattrs from cluster %llu to %llu\n",
3976              (unsigned long long)src_blk, (unsigned long long)to_blk);
3977
3978         /* The first bucket of the original extent */
3979         old_first = ocfs2_xattr_bucket_new(inode);
3980         /* The first bucket of the new extent */
3981         new_first = ocfs2_xattr_bucket_new(inode);
3982         if (!old_first || !new_first) {
3983                 ret = -ENOMEM;
3984                 mlog_errno(ret);
3985                 goto out;
3986         }
3987
3988         ret = ocfs2_read_xattr_bucket(old_first, first_bh->b_blocknr);
3989         if (ret) {
3990                 mlog_errno(ret);
3991                 goto out;
3992         }
3993
3994         /*
3995          * We need to update the first bucket of the old extent and the
3996          * entire first cluster of the new extent.
3997          */
3998         credits = blks_per_bucket + bpc + handle->h_buffer_credits;
3999         ret = ocfs2_extend_trans(handle, credits);
4000         if (ret) {
4001                 mlog_errno(ret);
4002                 goto out;
4003         }
4004
4005         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4006                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4007         if (ret) {
4008                 mlog_errno(ret);
4009                 goto out;
4010         }
4011
4012         for (i = 0; i < num_buckets; i++) {
4013                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4014                                             src_blk + (i * blks_per_bucket),
4015                                             to_blk + (i * blks_per_bucket),
4016                                             1);
4017                 if (ret) {
4018                         mlog_errno(ret);
4019                         goto out;
4020                 }
4021         }
4022
4023         /*
4024          * Get the new bucket ready before we dirty anything
4025          * (This actually shouldn't fail, because we already dirtied
4026          * it once in ocfs2_cp_xattr_bucket()).
4027          */
4028         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4029         if (ret) {
4030                 mlog_errno(ret);
4031                 goto out;
4032         }
4033         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4034                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4035         if (ret) {
4036                 mlog_errno(ret);
4037                 goto out;
4038         }
4039
4040         /* Now update the headers */
4041         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4042         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4043
4044         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4045         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4046
4047         if (first_hash)
4048                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4049
4050 out:
4051         ocfs2_xattr_bucket_free(new_first);
4052         ocfs2_xattr_bucket_free(old_first);
4053         return ret;
4054 }
4055
4056 /*
4057  * Move some xattrs in this cluster to the new cluster.
4058  * This function should only be called when bucket size == cluster size.
4059  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4060  */
4061 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4062                                       handle_t *handle,
4063                                       u64 prev_blk,
4064                                       u64 new_blk,
4065                                       u32 *first_hash)
4066 {
4067         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4068         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4069
4070         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4071
4072         ret = ocfs2_extend_trans(handle, credits);
4073         if (ret) {
4074                 mlog_errno(ret);
4075                 return ret;
4076         }
4077
4078         /* Move half of the xattr in start_blk to the next bucket. */
4079         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4080                                           new_blk, first_hash, 1);
4081 }
4082
4083 /*
4084  * Move some xattrs from the old cluster to the new one since they are not
4085  * contiguous in ocfs2 xattr tree.
4086  *
4087  * new_blk starts a new separate cluster, and we will move some xattrs from
4088  * prev_blk to it. v_start will be set as the first name hash value in this
4089  * new cluster so that it can be used as e_cpos during tree insertion and
4090  * don't collide with our original b-tree operations. first_bh and header_bh
4091  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4092  * to extend the insert bucket.
4093  *
4094  * The problem is how much xattr should we move to the new one and when should
4095  * we update first_bh and header_bh?
4096  * 1. If cluster size > bucket size, that means the previous cluster has more
4097  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4098  *    update the first_bh and header_bh if the insert bucket has been moved
4099  *    to the new cluster.
4100  * 2. If cluster_size == bucket_size:
4101  *    a) If the previous extent rec has more than one cluster and the insert
4102  *       place isn't in the last cluster, copy the entire last cluster to the
4103  *       new one. This time, we don't need to upate the first_bh and header_bh
4104  *       since they will not be moved into the new cluster.
4105  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4106  *       the new one. And we set the extend flag to zero if the insert place is
4107  *       moved into the new allocated cluster since no extend is needed.
4108  */
4109 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4110                                             handle_t *handle,
4111                                             struct buffer_head **first_bh,
4112                                             struct buffer_head **header_bh,
4113                                             u64 new_blk,
4114                                             u64 prev_blk,
4115                                             u32 prev_clusters,
4116                                             u32 *v_start,
4117                                             int *extend)
4118 {
4119         int ret = 0;
4120         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4121
4122         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4123              (unsigned long long)prev_blk, prev_clusters,
4124              (unsigned long long)new_blk);
4125
4126         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
4127                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4128                                                           handle,
4129                                                           first_bh,
4130                                                           header_bh,
4131                                                           new_blk,
4132                                                           prev_blk,
4133                                                           prev_clusters,
4134                                                           v_start);
4135         else {
4136                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
4137
4138                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
4139                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
4140                                                      last_blk, new_blk,
4141                                                      v_start);
4142                 else {
4143                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4144                                                          last_blk, new_blk,
4145                                                          v_start);
4146
4147                         if ((*header_bh)->b_blocknr == last_blk && extend)
4148                                 *extend = 0;
4149                 }
4150         }
4151
4152         return ret;
4153 }
4154
4155 /*
4156  * Add a new cluster for xattr storage.
4157  *
4158  * If the new cluster is contiguous with the previous one, it will be
4159  * appended to the same extent record, and num_clusters will be updated.
4160  * If not, we will insert a new extent for it and move some xattrs in
4161  * the last cluster into the new allocated one.
4162  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4163  * lose the benefits of hashing because we'll have to search large leaves.
4164  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4165  * if it's bigger).
4166  *
4167  * first_bh is the first block of the previous extent rec and header_bh
4168  * indicates the bucket we will insert the new xattrs. They will be updated
4169  * when the header_bh is moved into the new cluster.
4170  */
4171 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4172                                        struct buffer_head *root_bh,
4173                                        struct buffer_head **first_bh,
4174                                        struct buffer_head **header_bh,
4175                                        u32 *num_clusters,
4176                                        u32 prev_cpos,
4177                                        u64 prev_blkno,
4178                                        int *extend,
4179                                        struct ocfs2_xattr_set_ctxt *ctxt)
4180 {
4181         int ret;
4182         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4183         u32 prev_clusters = *num_clusters;
4184         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4185         u64 block;
4186         handle_t *handle = ctxt->handle;
4187         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4188         struct ocfs2_extent_tree et;
4189
4190         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4191              "previous xattr blkno = %llu\n",
4192              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4193              prev_cpos, (unsigned long long)prev_blkno);
4194
4195         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4196
4197         ret = ocfs2_journal_access(handle, inode, root_bh,
4198                                    OCFS2_JOURNAL_ACCESS_WRITE);
4199         if (ret < 0) {
4200                 mlog_errno(ret);
4201                 goto leave;
4202         }
4203
4204         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4205                                      clusters_to_add, &bit_off, &num_bits);
4206         if (ret < 0) {
4207                 if (ret != -ENOSPC)
4208                         mlog_errno(ret);
4209                 goto leave;
4210         }
4211
4212         BUG_ON(num_bits > clusters_to_add);
4213
4214         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4215         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4216              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4217
4218         if (prev_blkno + prev_clusters * bpc == block &&
4219             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4220              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4221                 /*
4222                  * If this cluster is contiguous with the old one and
4223                  * adding this new cluster, we don't surpass the limit of
4224                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4225                  * initialized and used like other buckets in the previous
4226                  * cluster.
4227                  * So add it as a contiguous one. The caller will handle
4228                  * its init process.
4229                  */
4230                 v_start = prev_cpos + prev_clusters;
4231                 *num_clusters = prev_clusters + num_bits;
4232                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4233                      num_bits);
4234         } else {
4235                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4236                                                        handle,
4237                                                        first_bh,
4238                                                        header_bh,
4239                                                        block,
4240                                                        prev_blkno,
4241                                                        prev_clusters,
4242                                                        &v_start,
4243                                                        extend);
4244                 if (ret) {
4245                         mlog_errno(ret);
4246                         goto leave;
4247                 }
4248         }
4249
4250         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4251              num_bits, (unsigned long long)block, v_start);
4252         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4253                                   num_bits, 0, ctxt->meta_ac);
4254         if (ret < 0) {
4255                 mlog_errno(ret);
4256                 goto leave;
4257         }
4258
4259         ret = ocfs2_journal_dirty(handle, root_bh);
4260         if (ret < 0)
4261                 mlog_errno(ret);
4262
4263 leave:
4264         return ret;
4265 }
4266
4267 /*
4268  * We are given an extent.  'first' is the bucket at the very front of
4269  * the extent.  The extent has space for an additional bucket past
4270  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4271  * of the target bucket.  We wish to shift every bucket past the target
4272  * down one, filling in that additional space.  When we get back to the
4273  * target, we split the target between itself and the now-empty bucket
4274  * at target+1 (aka, target_blkno + blks_per_bucket).
4275  */
4276 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4277                                      handle_t *handle,
4278                                      struct ocfs2_xattr_bucket *first,
4279                                      u64 target_blk,
4280                                      u32 num_clusters)
4281 {
4282         int ret, credits;
4283         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4284         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4285         u64 end_blk;
4286         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4287
4288         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4289              "from %llu, len = %u\n", (unsigned long long)target_blk,
4290              (unsigned long long)bucket_blkno(first), num_clusters);
4291
4292         /* The extent must have room for an additional bucket */
4293         BUG_ON(new_bucket >=
4294                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4295
4296         /* end_blk points to the last existing bucket */
4297         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4298
4299         /*
4300          * end_blk is the start of the last existing bucket.
4301          * Thus, (end_blk - target_blk) covers the target bucket and
4302          * every bucket after it up to, but not including, the last
4303          * existing bucket.  Then we add the last existing bucket, the
4304          * new bucket, and the first bucket (3 * blk_per_bucket).
4305          */
4306         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4307                   handle->h_buffer_credits;
4308         ret = ocfs2_extend_trans(handle, credits);
4309         if (ret) {
4310                 mlog_errno(ret);
4311                 goto out;
4312         }
4313
4314         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4315                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4316         if (ret) {
4317                 mlog_errno(ret);
4318                 goto out;
4319         }
4320
4321         while (end_blk != target_blk) {
4322                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4323                                             end_blk + blk_per_bucket, 0);
4324                 if (ret)
4325                         goto out;
4326                 end_blk -= blk_per_bucket;
4327         }
4328
4329         /* Move half of the xattr in target_blkno to the next bucket. */
4330         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4331                                         target_blk + blk_per_bucket, NULL, 0);
4332
4333         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4334         ocfs2_xattr_bucket_journal_dirty(handle, first);
4335
4336 out:
4337         return ret;
4338 }
4339
4340 /*
4341  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
4342  * xb_bh is the ocfs2_xattr_block.
4343  * We will move all the buckets starting from header_bh to the next place. As
4344  * for this one, half num of its xattrs will be moved to the next one.
4345  *
4346  * We will allocate a new cluster if current cluster is full and adjust
4347  * header_bh and first_bh if the insert place is moved to the new cluster.
4348  */
4349 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4350                                       struct buffer_head *xb_bh,
4351                                       struct buffer_head *header_bh,
4352                                       struct ocfs2_xattr_set_ctxt *ctxt)
4353 {
4354         struct ocfs2_xattr_header *first_xh = NULL;
4355         struct buffer_head *first_bh = NULL;
4356         struct ocfs2_xattr_block *xb =
4357                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4358         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4359         struct ocfs2_extent_list *el = &xb_root->xt_list;
4360         struct ocfs2_xattr_header *xh =
4361                         (struct ocfs2_xattr_header *)header_bh->b_data;
4362         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4363         struct super_block *sb = inode->i_sb;
4364         struct ocfs2_super *osb = OCFS2_SB(sb);
4365         int ret, num_buckets, extend = 1;
4366         u64 p_blkno;
4367         u32 e_cpos, num_clusters;
4368         /* The bucket at the front of the extent */
4369         struct ocfs2_xattr_bucket *first;
4370
4371         mlog(0, "Add new xattr bucket starting form %llu\n",
4372              (unsigned long long)header_bh->b_blocknr);
4373
4374         first = ocfs2_xattr_bucket_new(inode);
4375         if (!first) {
4376                 ret = -ENOMEM;
4377                 mlog_errno(ret);
4378                 goto out;
4379         }
4380
4381         /*
4382          * Add refrence for header_bh here because it may be
4383          * changed in ocfs2_add_new_xattr_cluster and we need
4384          * to free it in the end.
4385          */
4386         get_bh(header_bh);
4387
4388         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4389                                   &num_clusters, el);
4390         if (ret) {
4391                 mlog_errno(ret);
4392                 goto out;
4393         }
4394
4395         ret = ocfs2_read_block(inode, p_blkno, &first_bh, NULL);
4396         if (ret) {
4397                 mlog_errno(ret);
4398                 goto out;
4399         }
4400
4401         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4402         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4403
4404         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
4405                 ret = ocfs2_add_new_xattr_cluster(inode,
4406                                                   xb_bh,
4407                                                   &first_bh,
4408                                                   &header_bh,
4409                                                   &num_clusters,
4410                                                   e_cpos,
4411                                                   p_blkno,
4412                                                   &extend,
4413                                                   ctxt);
4414                 if (ret) {
4415                         mlog_errno(ret);
4416                         goto out;
4417                 }
4418         }
4419
4420         if (extend) {
4421                 /* These bucket reads should be cached */
4422                 ret = ocfs2_read_xattr_bucket(first, first_bh->b_blocknr);
4423                 if (ret) {
4424                         mlog_errno(ret);
4425                         goto out;
4426                 }
4427                 ret = ocfs2_extend_xattr_bucket(inode,
4428                                                 ctxt->handle,
4429                                                 first, header_bh->b_blocknr,
4430                                                 num_clusters);
4431                 if (ret)
4432                         mlog_errno(ret);
4433         }
4434
4435 out:
4436         brelse(first_bh);
4437         brelse(header_bh);
4438         ocfs2_xattr_bucket_free(first);
4439         return ret;
4440 }
4441
4442 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4443                                         struct ocfs2_xattr_bucket *bucket,
4444                                         int offs)
4445 {
4446         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4447
4448         offs = offs % inode->i_sb->s_blocksize;
4449         return bucket_block(bucket, block_off) + offs;
4450 }
4451
4452 /*
4453  * Handle the normal xattr set, including replace, delete and new.
4454  *
4455  * Note: "local" indicates the real data's locality. So we can't
4456  * just its bucket locality by its length.
4457  */
4458 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4459                                          struct ocfs2_xattr_info *xi,
4460                                          struct ocfs2_xattr_search *xs,
4461                                          u32 name_hash,
4462                                          int local)
4463 {
4464         struct ocfs2_xattr_entry *last, *xe;
4465         int name_len = strlen(xi->name);
4466         struct ocfs2_xattr_header *xh = xs->header;
4467         u16 count = le16_to_cpu(xh->xh_count), start;
4468         size_t blocksize = inode->i_sb->s_blocksize;
4469         char *val;
4470         size_t offs, size, new_size;
4471
4472         last = &xh->xh_entries[count];
4473         if (!xs->not_found) {
4474                 xe = xs->here;
4475                 offs = le16_to_cpu(xe->xe_name_offset);
4476                 if (ocfs2_xattr_is_local(xe))
4477                         size = OCFS2_XATTR_SIZE(name_len) +
4478                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4479                 else
4480                         size = OCFS2_XATTR_SIZE(name_len) +
4481                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4482
4483                 /*
4484                  * If the new value will be stored outside, xi->value has been
4485                  * initalized as an empty ocfs2_xattr_value_root, and the same
4486                  * goes with xi->value_len, so we can set new_size safely here.
4487                  * See ocfs2_xattr_set_in_bucket.
4488                  */
4489                 new_size = OCFS2_XATTR_SIZE(name_len) +
4490                            OCFS2_XATTR_SIZE(xi->value_len);
4491
4492                 le16_add_cpu(&xh->xh_name_value_len, -size);
4493                 if (xi->value) {
4494                         if (new_size > size)
4495                                 goto set_new_name_value;
4496
4497                         /* Now replace the old value with new one. */
4498                         if (local)
4499                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4500                         else
4501                                 xe->xe_value_size = 0;
4502
4503                         val = ocfs2_xattr_bucket_get_val(inode,
4504                                                          xs->bucket, offs);
4505                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4506                                size - OCFS2_XATTR_SIZE(name_len));
4507                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4508                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4509                                        xi->value, xi->value_len);
4510
4511                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4512                         ocfs2_xattr_set_local(xe, local);
4513                         return;
4514                 } else {
4515                         /*
4516                          * Remove the old entry if there is more than one.
4517                          * We don't remove the last entry so that we can
4518                          * use it to indicate the hash value of the empty
4519                          * bucket.
4520                          */
4521                         last -= 1;
4522                         le16_add_cpu(&xh->xh_count, -1);
4523                         if (xh->xh_count) {
4524                                 memmove(xe, xe + 1,
4525                                         (void *)last - (void *)xe);
4526                                 memset(last, 0,
4527                                        sizeof(struct ocfs2_xattr_entry));
4528                         } else
4529                                 xh->xh_free_start =
4530                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4531
4532                         return;
4533                 }
4534         } else {
4535                 /* find a new entry for insert. */
4536                 int low = 0, high = count - 1, tmp;
4537                 struct ocfs2_xattr_entry *tmp_xe;
4538
4539                 while (low <= high && count) {
4540                         tmp = (low + high) / 2;
4541                         tmp_xe = &xh->xh_entries[tmp];
4542
4543                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4544                                 low = tmp + 1;
4545                         else if (name_hash <
4546                                  le32_to_cpu(tmp_xe->xe_name_hash))
4547                                 high = tmp - 1;
4548                         else {
4549                                 low = tmp;
4550                                 break;
4551                         }
4552                 }
4553
4554                 xe = &xh->xh_entries[low];
4555                 if (low != count)
4556                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4557
4558                 le16_add_cpu(&xh->xh_count, 1);
4559                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4560                 xe->xe_name_hash = cpu_to_le32(name_hash);
4561                 xe->xe_name_len = name_len;
4562                 ocfs2_xattr_set_type(xe, xi->name_index);
4563         }
4564
4565 set_new_name_value:
4566         /* Insert the new name+value. */
4567         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4568
4569         /*
4570          * We must make sure that the name/value pair
4571          * exists in the same block.
4572          */
4573         offs = le16_to_cpu(xh->xh_free_start);
4574         start = offs - size;
4575
4576         if (start >> inode->i_sb->s_blocksize_bits !=
4577             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4578                 offs = offs - offs % blocksize;
4579                 xh->xh_free_start = cpu_to_le16(offs);
4580         }
4581
4582         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4583         xe->xe_name_offset = cpu_to_le16(offs - size);
4584
4585         memset(val, 0, size);
4586         memcpy(val, xi->name, name_len);
4587         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4588
4589         xe->xe_value_size = cpu_to_le64(xi->value_len);
4590         ocfs2_xattr_set_local(xe, local);
4591         xs->here = xe;
4592         le16_add_cpu(&xh->xh_free_start, -size);
4593         le16_add_cpu(&xh->xh_name_value_len, size);
4594
4595         return;
4596 }
4597
4598 /*
4599  * Set the xattr entry in the specified bucket.
4600  * The bucket is indicated by xs->bucket and it should have the enough
4601  * space for the xattr insertion.
4602  */
4603 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4604                                            handle_t *handle,
4605                                            struct ocfs2_xattr_info *xi,
4606                                            struct ocfs2_xattr_search *xs,
4607                                            u32 name_hash,
4608                                            int local)
4609 {
4610         int ret;
4611         u64 blkno;
4612
4613         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4614              (unsigned long)xi->value_len, xi->name_index,
4615              (unsigned long long)bucket_blkno(xs->bucket));
4616
4617         if (!xs->bucket->bu_bhs[1]) {
4618                 blkno = bucket_blkno(xs->bucket);
4619                 ocfs2_xattr_bucket_relse(xs->bucket);
4620                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4621                 if (ret) {
4622                         mlog_errno(ret);
4623                         goto out;
4624                 }
4625         }
4626
4627         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4628                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4629         if (ret < 0) {
4630                 mlog_errno(ret);
4631                 goto out;
4632         }
4633
4634         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4635         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4636
4637 out:
4638         return ret;
4639 }
4640
4641 /*
4642  * Truncate the specified xe_off entry in xattr bucket.
4643  * bucket is indicated by header_bh and len is the new length.
4644  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4645  *
4646  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4647  */
4648 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4649                                              struct ocfs2_xattr_bucket *bucket,
4650                                              int xe_off,
4651                                              int len,
4652                                              struct ocfs2_xattr_set_ctxt *ctxt)
4653 {
4654         int ret, offset;
4655         u64 value_blk;
4656         struct buffer_head *value_bh = NULL;
4657         struct ocfs2_xattr_value_root *xv;
4658         struct ocfs2_xattr_entry *xe;
4659         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4660         size_t blocksize = inode->i_sb->s_blocksize;
4661
4662         xe = &xh->xh_entries[xe_off];
4663
4664         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4665
4666         offset = le16_to_cpu(xe->xe_name_offset) +
4667                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4668
4669         value_blk = offset / blocksize;
4670
4671         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4672         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4673
4674         value_bh = bucket->bu_bhs[value_blk];
4675         BUG_ON(!value_bh);
4676
4677         xv = (struct ocfs2_xattr_value_root *)
4678                 (value_bh->b_data + offset % blocksize);
4679
4680         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4681                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4682         if (ret) {
4683                 mlog_errno(ret);
4684                 goto out;
4685         }
4686
4687         /*
4688          * From here on out we have to dirty the bucket.  The generic
4689          * value calls only modify one of the bucket's bhs, but we need
4690          * to send the bucket at once.  So if they error, they *could* have
4691          * modified something.  We have to assume they did, and dirty
4692          * the whole bucket.  This leaves us in a consistent state.
4693          */
4694         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4695              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4696         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len, ctxt);
4697         if (ret) {
4698                 mlog_errno(ret);
4699                 goto out_dirty;
4700         }
4701
4702         xe->xe_value_size = cpu_to_le64(len);
4703
4704 out_dirty:
4705         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4706
4707 out:
4708         return ret;
4709 }
4710
4711 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4712                                         struct ocfs2_xattr_search *xs,
4713                                         int len,
4714                                         struct ocfs2_xattr_set_ctxt *ctxt)
4715 {
4716         int ret, offset;
4717         struct ocfs2_xattr_entry *xe = xs->here;
4718         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4719
4720         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4721
4722         offset = xe - xh->xh_entries;
4723         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4724                                                 offset, len, ctxt);
4725         if (ret)
4726                 mlog_errno(ret);
4727
4728         return ret;
4729 }
4730
4731 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4732                                                 handle_t *handle,
4733                                                 struct ocfs2_xattr_search *xs,
4734                                                 char *val,
4735                                                 int value_len)
4736 {
4737         int offset;
4738         struct ocfs2_xattr_value_root *xv;
4739         struct ocfs2_xattr_entry *xe = xs->here;
4740
4741         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4742
4743         offset = le16_to_cpu(xe->xe_name_offset) +
4744                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4745
4746         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4747
4748         return __ocfs2_xattr_set_value_outside(inode, handle,
4749                                                xv, val, value_len);
4750 }
4751
4752 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4753                                   struct buffer_head *root_bh,
4754                                   u64 blkno,
4755                                   u32 cpos,
4756                                   u32 len)
4757 {
4758         int ret;
4759         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4760         struct inode *tl_inode = osb->osb_tl_inode;
4761         handle_t *handle;
4762         struct ocfs2_xattr_block *xb =
4763                         (struct ocfs2_xattr_block *)root_bh->b_data;
4764         struct ocfs2_alloc_context *meta_ac = NULL;
4765         struct ocfs2_cached_dealloc_ctxt dealloc;
4766         struct ocfs2_extent_tree et;
4767
4768         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4769
4770         ocfs2_init_dealloc_ctxt(&dealloc);
4771
4772         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4773              cpos, len, (unsigned long long)blkno);
4774
4775         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4776
4777         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4778         if (ret) {
4779                 mlog_errno(ret);
4780                 return ret;
4781         }
4782
4783         mutex_lock(&tl_inode->i_mutex);
4784
4785         if (ocfs2_truncate_log_needs_flush(osb)) {
4786                 ret = __ocfs2_flush_truncate_log(osb);
4787                 if (ret < 0) {
4788                         mlog_errno(ret);
4789                         goto out;
4790                 }
4791         }
4792
4793         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4794         if (IS_ERR(handle)) {
4795                 ret = -ENOMEM;
4796                 mlog_errno(ret);
4797                 goto out;
4798         }
4799
4800         ret = ocfs2_journal_access(handle, inode, root_bh,
4801                                    OCFS2_JOURNAL_ACCESS_WRITE);
4802         if (ret) {
4803                 mlog_errno(ret);
4804                 goto out_commit;
4805         }
4806
4807         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4808                                   &dealloc);
4809         if (ret) {
4810                 mlog_errno(ret);
4811                 goto out_commit;
4812         }
4813
4814         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4815
4816         ret = ocfs2_journal_dirty(handle, root_bh);
4817         if (ret) {
4818                 mlog_errno(ret);
4819                 goto out_commit;
4820         }
4821
4822         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4823         if (ret)
4824                 mlog_errno(ret);
4825
4826 out_commit:
4827         ocfs2_commit_trans(osb, handle);
4828 out:
4829         ocfs2_schedule_truncate_log_flush(osb, 1);
4830
4831         mutex_unlock(&tl_inode->i_mutex);
4832
4833         if (meta_ac)
4834                 ocfs2_free_alloc_context(meta_ac);
4835
4836         ocfs2_run_deallocs(osb, &dealloc);
4837
4838         return ret;
4839 }
4840
4841 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4842                                          handle_t *handle,
4843                                          struct ocfs2_xattr_search *xs)
4844 {
4845         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4846         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4847                                                 le16_to_cpu(xh->xh_count) - 1];
4848         int ret = 0;
4849
4850         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4851                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4852         if (ret) {
4853                 mlog_errno(ret);
4854                 return;
4855         }
4856
4857         /* Remove the old entry. */
4858         memmove(xs->here, xs->here + 1,
4859                 (void *)last - (void *)xs->here);
4860         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4861         le16_add_cpu(&xh->xh_count, -1);
4862
4863         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4864 }
4865
4866 /*
4867  * Set the xattr name/value in the bucket specified in xs.
4868  *
4869  * As the new value in xi may be stored in the bucket or in an outside cluster,
4870  * we divide the whole process into 3 steps:
4871  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4872  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4873  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4874  * 4. If the clusters for the new outside value can't be allocated, we need
4875  *    to free the xattr we allocated in set.
4876  */
4877 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4878                                      struct ocfs2_xattr_info *xi,
4879                                      struct ocfs2_xattr_search *xs,
4880                                      struct ocfs2_xattr_set_ctxt *ctxt)
4881 {
4882         int ret, local = 1;
4883         size_t value_len;
4884         char *val = (char *)xi->value;
4885         struct ocfs2_xattr_entry *xe = xs->here;
4886         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4887                                               strlen(xi->name));
4888
4889         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4890                 /*
4891                  * We need to truncate the xattr storage first.
4892                  *
4893                  * If both the old and new value are stored to
4894                  * outside block, we only need to truncate
4895                  * the storage and then set the value outside.
4896                  *
4897                  * If the new value should be stored within block,
4898                  * we should free all the outside block first and
4899                  * the modification to the xattr block will be done
4900                  * by following steps.
4901                  */
4902                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4903                         value_len = xi->value_len;
4904                 else
4905                         value_len = 0;
4906
4907                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4908                                                            value_len,
4909                                                            ctxt);
4910                 if (ret)
4911                         goto out;
4912
4913                 if (value_len)
4914                         goto set_value_outside;
4915         }
4916
4917         value_len = xi->value_len;
4918         /* So we have to handle the inside block change now. */
4919         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4920                 /*
4921                  * If the new value will be stored outside of block,
4922                  * initalize a new empty value root and insert it first.
4923                  */
4924                 local = 0;
4925                 xi->value = &def_xv;
4926                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4927         }
4928
4929         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4930                                               name_hash, local);
4931         if (ret) {
4932                 mlog_errno(ret);
4933                 goto out;
4934         }
4935
4936         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4937                 goto out;
4938
4939         /* allocate the space now for the outside block storage. */
4940         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4941                                                    value_len, ctxt);
4942         if (ret) {
4943                 mlog_errno(ret);
4944
4945                 if (xs->not_found) {
4946                         /*
4947                          * We can't allocate enough clusters for outside
4948                          * storage and we have allocated xattr already,
4949                          * so need to remove it.
4950                          */
4951                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4952                 }
4953                 goto out;
4954         }
4955
4956 set_value_outside:
4957         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4958                                                    xs, val, value_len);
4959 out:
4960         return ret;
4961 }
4962
4963 /*
4964  * check whether the xattr bucket is filled up with the same hash value.
4965  * If we want to insert the xattr with the same hash, return -ENOSPC.
4966  * If we want to insert a xattr with different hash value, go ahead
4967  * and ocfs2_divide_xattr_bucket will handle this.
4968  */
4969 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4970                                               struct ocfs2_xattr_bucket *bucket,
4971                                               const char *name)
4972 {
4973         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4974         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4975
4976         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4977                 return 0;
4978
4979         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4980             xh->xh_entries[0].xe_name_hash) {
4981                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4982                      "hash = %u\n",
4983                      (unsigned long long)bucket_blkno(bucket),
4984                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4985                 return -ENOSPC;
4986         }
4987
4988         return 0;
4989 }
4990
4991 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4992                                              struct ocfs2_xattr_info *xi,
4993                                              struct ocfs2_xattr_search *xs,
4994                                              struct ocfs2_xattr_set_ctxt *ctxt)
4995 {
4996         struct ocfs2_xattr_header *xh;
4997         struct ocfs2_xattr_entry *xe;
4998         u16 count, header_size, xh_free_start;
4999         int free, max_free, need, old;
5000         size_t value_size = 0, name_len = strlen(xi->name);
5001         size_t blocksize = inode->i_sb->s_blocksize;
5002         int ret, allocation = 0;
5003
5004         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5005
5006 try_again:
5007         xh = xs->header;
5008         count = le16_to_cpu(xh->xh_count);
5009         xh_free_start = le16_to_cpu(xh->xh_free_start);
5010         header_size = sizeof(struct ocfs2_xattr_header) +
5011                         count * sizeof(struct ocfs2_xattr_entry);
5012         max_free = OCFS2_XATTR_BUCKET_SIZE -
5013                 le16_to_cpu(xh->xh_name_value_len) - header_size;
5014
5015         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5016                         "of %u which exceed block size\n",
5017                         (unsigned long long)bucket_blkno(xs->bucket),
5018                         header_size);
5019
5020         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5021                 value_size = OCFS2_XATTR_ROOT_SIZE;
5022         else if (xi->value)
5023                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5024
5025         if (xs->not_found)
5026                 need = sizeof(struct ocfs2_xattr_entry) +
5027                         OCFS2_XATTR_SIZE(name_len) + value_size;
5028         else {
5029                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5030
5031                 /*
5032                  * We only replace the old value if the new length is smaller
5033                  * than the old one. Otherwise we will allocate new space in the
5034                  * bucket to store it.
5035                  */
5036                 xe = xs->here;
5037                 if (ocfs2_xattr_is_local(xe))
5038                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5039                 else
5040                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5041
5042                 if (old >= value_size)
5043                         need = 0;
5044         }
5045
5046         free = xh_free_start - header_size;
5047         /*
5048          * We need to make sure the new name/value pair
5049          * can exist in the same block.
5050          */
5051         if (xh_free_start % blocksize < need)
5052                 free -= xh_free_start % blocksize;
5053
5054         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5055              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5056              " %u\n", xs->not_found,
5057              (unsigned long long)bucket_blkno(xs->bucket),
5058              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5059              le16_to_cpu(xh->xh_name_value_len));
5060
5061         if (free < need ||
5062             (xs->not_found &&
5063              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5064                 if (need <= max_free &&
5065                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5066                         /*
5067                          * We can create the space by defragment. Since only the
5068                          * name/value will be moved, the xe shouldn't be changed
5069                          * in xs.
5070                          */
5071                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5072                                                         xs->bucket);
5073                         if (ret) {
5074                                 mlog_errno(ret);
5075                                 goto out;
5076                         }
5077
5078                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5079                         free = xh_free_start - header_size;
5080                         if (xh_free_start % blocksize < need)
5081                                 free -= xh_free_start % blocksize;
5082
5083                         if (free >= need)
5084                                 goto xattr_set;
5085
5086                         mlog(0, "Can't get enough space for xattr insert by "
5087                              "defragment. Need %u bytes, but we have %d, so "
5088                              "allocate new bucket for it.\n", need, free);
5089                 }
5090
5091                 /*
5092                  * We have to add new buckets or clusters and one
5093                  * allocation should leave us enough space for insert.
5094                  */
5095                 BUG_ON(allocation);
5096
5097                 /*
5098                  * We do not allow for overlapping ranges between buckets. And
5099                  * the maximum number of collisions we will allow for then is
5100                  * one bucket's worth, so check it here whether we need to
5101                  * add a new bucket for the insert.
5102                  */
5103                 ret = ocfs2_check_xattr_bucket_collision(inode,
5104                                                          xs->bucket,
5105                                                          xi->name);
5106                 if (ret) {
5107                         mlog_errno(ret);
5108                         goto out;
5109                 }
5110
5111                 ret = ocfs2_add_new_xattr_bucket(inode,
5112                                                  xs->xattr_bh,
5113                                                  xs->bucket->bu_bhs[0],
5114                                                  ctxt);
5115                 if (ret) {
5116                         mlog_errno(ret);
5117                         goto out;
5118                 }
5119
5120                 ocfs2_xattr_bucket_relse(xs->bucket);
5121
5122                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5123                                                    xi->name_index,
5124                                                    xi->name, xs);
5125                 if (ret && ret != -ENODATA)
5126                         goto out;
5127                 xs->not_found = ret;
5128                 allocation = 1;
5129                 goto try_again;
5130         }
5131
5132 xattr_set:
5133         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5134 out:
5135         mlog_exit(ret);
5136         return ret;
5137 }
5138
5139 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5140                                         struct ocfs2_xattr_bucket *bucket,
5141                                         void *para)
5142 {
5143         int ret = 0;
5144         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5145         u16 i;
5146         struct ocfs2_xattr_entry *xe;
5147         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5148         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5149         int credits = ocfs2_remove_extent_credits(osb->sb) +
5150                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5151
5152
5153         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5154
5155         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5156                 xe = &xh->xh_entries[i];
5157                 if (ocfs2_xattr_is_local(xe))
5158                         continue;
5159
5160                 ctxt.handle = ocfs2_start_trans(osb, credits);
5161                 if (IS_ERR(ctxt.handle)) {
5162                         ret = PTR_ERR(ctxt.handle);
5163                         mlog_errno(ret);
5164                         break;
5165                 }
5166
5167                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5168                                                         i, 0, &ctxt);
5169
5170                 ocfs2_commit_trans(osb, ctxt.handle);
5171                 if (ret) {
5172                         mlog_errno(ret);
5173                         break;
5174                 }
5175         }
5176
5177         ocfs2_schedule_truncate_log_flush(osb, 1);
5178         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5179         return ret;
5180 }
5181
5182 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5183                                           struct buffer_head *xb_bh)
5184 {
5185         struct ocfs2_xattr_block *xb =
5186                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5187         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5188         int ret = 0;
5189         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5190         u64 p_blkno;
5191
5192         if (le16_to_cpu(el->l_next_free_rec) == 0)
5193                 return 0;
5194
5195         while (name_hash > 0) {
5196                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5197                                           &e_cpos, &num_clusters, el);
5198                 if (ret) {
5199                         mlog_errno(ret);
5200                         goto out;
5201                 }
5202
5203                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5204                                                   ocfs2_delete_xattr_in_bucket,
5205                                                   NULL);
5206                 if (ret) {
5207                         mlog_errno(ret);
5208                         goto out;
5209                 }
5210
5211                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5212                                              p_blkno, e_cpos, num_clusters);
5213                 if (ret) {
5214                         mlog_errno(ret);
5215                         break;
5216                 }
5217
5218                 if (e_cpos == 0)
5219                         break;
5220
5221                 name_hash = e_cpos - 1;
5222         }
5223
5224 out:
5225         return ret;
5226 }
5227
5228 /*
5229  * 'security' attributes support
5230  */
5231 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5232                                         size_t list_size, const char *name,
5233                                         size_t name_len)
5234 {
5235         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5236         const size_t total_len = prefix_len + name_len + 1;
5237
5238         if (list && total_len <= list_size) {
5239                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5240                 memcpy(list + prefix_len, name, name_len);
5241                 list[prefix_len + name_len] = '\0';
5242         }
5243         return total_len;
5244 }
5245
5246 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5247                                     void *buffer, size_t size)
5248 {
5249         if (strcmp(name, "") == 0)
5250                 return -EINVAL;
5251         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5252                                buffer, size);
5253 }
5254
5255 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5256                                     const void *value, size_t size, int flags)
5257 {
5258         if (strcmp(name, "") == 0)
5259                 return -EINVAL;
5260
5261         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5262                                size, flags);
5263 }
5264
5265 int ocfs2_init_security_get(struct inode *inode,
5266                             struct inode *dir,
5267                             struct ocfs2_security_xattr_info *si)
5268 {
5269         return security_inode_init_security(inode, dir, &si->name, &si->value,
5270                                             &si->value_len);
5271 }
5272
5273 int ocfs2_init_security_set(handle_t *handle,
5274                             struct inode *inode,
5275                             struct buffer_head *di_bh,
5276                             struct ocfs2_security_xattr_info *si,
5277                             struct ocfs2_alloc_context *xattr_ac,
5278                             struct ocfs2_alloc_context *data_ac)
5279 {
5280         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5281                                      OCFS2_XATTR_INDEX_SECURITY,
5282                                      si->name, si->value, si->value_len, 0,
5283                                      xattr_ac, data_ac);
5284 }
5285
5286 struct xattr_handler ocfs2_xattr_security_handler = {
5287         .prefix = XATTR_SECURITY_PREFIX,
5288         .list   = ocfs2_xattr_security_list,
5289         .get    = ocfs2_xattr_security_get,
5290         .set    = ocfs2_xattr_security_set,
5291 };
5292
5293 /*
5294  * 'trusted' attributes support
5295  */
5296 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5297                                        size_t list_size, const char *name,
5298                                        size_t name_len)
5299 {
5300         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5301         const size_t total_len = prefix_len + name_len + 1;
5302
5303         if (list && total_len <= list_size) {
5304                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5305                 memcpy(list + prefix_len, name, name_len);
5306                 list[prefix_len + name_len] = '\0';
5307         }
5308         return total_len;
5309 }
5310
5311 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5312                                    void *buffer, size_t size)
5313 {
5314         if (strcmp(name, "") == 0)
5315                 return -EINVAL;
5316         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5317                                buffer, size);
5318 }
5319
5320 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5321                                    const void *value, size_t size, int flags)
5322 {
5323         if (strcmp(name, "") == 0)
5324                 return -EINVAL;
5325
5326         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5327                                size, flags);
5328 }
5329
5330 struct xattr_handler ocfs2_xattr_trusted_handler = {
5331         .prefix = XATTR_TRUSTED_PREFIX,
5332         .list   = ocfs2_xattr_trusted_list,
5333         .get    = ocfs2_xattr_trusted_get,
5334         .set    = ocfs2_xattr_trusted_set,
5335 };
5336
5337 /*
5338  * 'user' attributes support
5339  */
5340 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5341                                     size_t list_size, const char *name,
5342                                     size_t name_len)
5343 {
5344         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5345         const size_t total_len = prefix_len + name_len + 1;
5346         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5347
5348         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5349                 return 0;
5350
5351         if (list && total_len <= list_size) {
5352                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5353                 memcpy(list + prefix_len, name, name_len);
5354                 list[prefix_len + name_len] = '\0';
5355         }
5356         return total_len;
5357 }
5358
5359 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5360                                 void *buffer, size_t size)
5361 {
5362         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5363
5364         if (strcmp(name, "") == 0)
5365                 return -EINVAL;
5366         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5367                 return -EOPNOTSUPP;
5368         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5369                                buffer, size);
5370 }
5371
5372 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5373                                 const void *value, size_t size, int flags)
5374 {
5375         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5376
5377         if (strcmp(name, "") == 0)
5378                 return -EINVAL;
5379         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5380                 return -EOPNOTSUPP;
5381
5382         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5383                                size, flags);
5384 }
5385
5386 struct xattr_handler ocfs2_xattr_user_handler = {
5387         .prefix = XATTR_USER_PREFIX,
5388         .list   = ocfs2_xattr_user_list,
5389         .get    = ocfs2_xattr_user_get,
5390         .set    = ocfs2_xattr_user_set,
5391 };