ocfs2: Add a name indexed b-tree to directory inodes
[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 "blockcheck.h"
46 #include "dlmglue.h"
47 #include "file.h"
48 #include "symlink.h"
49 #include "sysfile.h"
50 #include "inode.h"
51 #include "journal.h"
52 #include "ocfs2_fs.h"
53 #include "suballoc.h"
54 #include "uptodate.h"
55 #include "buffer_head_io.h"
56 #include "super.h"
57 #include "xattr.h"
58
59
60 struct ocfs2_xattr_def_value_root {
61         struct ocfs2_xattr_value_root   xv;
62         struct ocfs2_extent_rec         er;
63 };
64
65 struct ocfs2_xattr_bucket {
66         /* The inode these xattrs are associated with */
67         struct inode *bu_inode;
68
69         /* The actual buffers that make up the bucket */
70         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
71
72         /* How many blocks make up one bucket for this filesystem */
73         int bu_blocks;
74 };
75
76 struct ocfs2_xattr_set_ctxt {
77         handle_t *handle;
78         struct ocfs2_alloc_context *meta_ac;
79         struct ocfs2_alloc_context *data_ac;
80         struct ocfs2_cached_dealloc_ctxt dealloc;
81 };
82
83 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
84 #define OCFS2_XATTR_INLINE_SIZE 80
85 #define OCFS2_XATTR_HEADER_GAP  4
86 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
87                                          - sizeof(struct ocfs2_xattr_header) \
88                                          - OCFS2_XATTR_HEADER_GAP)
89 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
90                                          - sizeof(struct ocfs2_xattr_block) \
91                                          - sizeof(struct ocfs2_xattr_header) \
92                                          - OCFS2_XATTR_HEADER_GAP)
93
94 static struct ocfs2_xattr_def_value_root def_xv = {
95         .xv.xr_list.l_count = cpu_to_le16(1),
96 };
97
98 struct xattr_handler *ocfs2_xattr_handlers[] = {
99         &ocfs2_xattr_user_handler,
100 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
101         &ocfs2_xattr_acl_access_handler,
102         &ocfs2_xattr_acl_default_handler,
103 #endif
104         &ocfs2_xattr_trusted_handler,
105         &ocfs2_xattr_security_handler,
106         NULL
107 };
108
109 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
111 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
112         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
113                                         = &ocfs2_xattr_acl_access_handler,
114         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
115                                         = &ocfs2_xattr_acl_default_handler,
116 #endif
117         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
118         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
119 };
120
121 struct ocfs2_xattr_info {
122         int name_index;
123         const char *name;
124         const void *value;
125         size_t value_len;
126 };
127
128 struct ocfs2_xattr_search {
129         struct buffer_head *inode_bh;
130         /*
131          * xattr_bh point to the block buffer head which has extended attribute
132          * when extended attribute in inode, xattr_bh is equal to inode_bh.
133          */
134         struct buffer_head *xattr_bh;
135         struct ocfs2_xattr_header *header;
136         struct ocfs2_xattr_bucket *bucket;
137         void *base;
138         void *end;
139         struct ocfs2_xattr_entry *here;
140         int not_found;
141 };
142
143 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
144                                              struct ocfs2_xattr_header *xh,
145                                              int index,
146                                              int *block_off,
147                                              int *new_offset);
148
149 static int ocfs2_xattr_block_find(struct inode *inode,
150                                   int name_index,
151                                   const char *name,
152                                   struct ocfs2_xattr_search *xs);
153 static int ocfs2_xattr_index_block_find(struct inode *inode,
154                                         struct buffer_head *root_bh,
155                                         int name_index,
156                                         const char *name,
157                                         struct ocfs2_xattr_search *xs);
158
159 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
160                                         struct ocfs2_xattr_tree_root *xt,
161                                         char *buffer,
162                                         size_t buffer_size);
163
164 static int ocfs2_xattr_create_index_block(struct inode *inode,
165                                           struct ocfs2_xattr_search *xs,
166                                           struct ocfs2_xattr_set_ctxt *ctxt);
167
168 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
169                                              struct ocfs2_xattr_info *xi,
170                                              struct ocfs2_xattr_search *xs,
171                                              struct ocfs2_xattr_set_ctxt *ctxt);
172
173 static int ocfs2_delete_xattr_index_block(struct inode *inode,
174                                           struct buffer_head *xb_bh);
175 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
176                                   u64 src_blk, u64 last_blk, u64 to_blk,
177                                   unsigned int start_bucket,
178                                   u32 *first_hash);
179
180 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
181 {
182         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
183 }
184
185 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
186 {
187         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
188 }
189
190 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
191 {
192         u16 len = sb->s_blocksize -
193                  offsetof(struct ocfs2_xattr_header, xh_entries);
194
195         return len / sizeof(struct ocfs2_xattr_entry);
196 }
197
198 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
199 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
200 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
201
202 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
203 {
204         struct ocfs2_xattr_bucket *bucket;
205         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
206
207         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
208
209         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
210         if (bucket) {
211                 bucket->bu_inode = inode;
212                 bucket->bu_blocks = blks;
213         }
214
215         return bucket;
216 }
217
218 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
219 {
220         int i;
221
222         for (i = 0; i < bucket->bu_blocks; i++) {
223                 brelse(bucket->bu_bhs[i]);
224                 bucket->bu_bhs[i] = NULL;
225         }
226 }
227
228 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
229 {
230         if (bucket) {
231                 ocfs2_xattr_bucket_relse(bucket);
232                 bucket->bu_inode = NULL;
233                 kfree(bucket);
234         }
235 }
236
237 /*
238  * A bucket that has never been written to disk doesn't need to be
239  * read.  We just need the buffer_heads.  Don't call this for
240  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
241  * them fully.
242  */
243 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
244                                    u64 xb_blkno)
245 {
246         int i, rc = 0;
247
248         for (i = 0; i < bucket->bu_blocks; i++) {
249                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
250                                               xb_blkno + i);
251                 if (!bucket->bu_bhs[i]) {
252                         rc = -EIO;
253                         mlog_errno(rc);
254                         break;
255                 }
256
257                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
258                                            bucket->bu_bhs[i]))
259                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
260                                                       bucket->bu_bhs[i]);
261         }
262
263         if (rc)
264                 ocfs2_xattr_bucket_relse(bucket);
265         return rc;
266 }
267
268 /* Read the xattr bucket at xb_blkno */
269 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
270                                    u64 xb_blkno)
271 {
272         int rc;
273
274         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
275                                bucket->bu_blocks, bucket->bu_bhs, 0,
276                                NULL);
277         if (!rc) {
278                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
279                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
280                                                  bucket->bu_bhs,
281                                                  bucket->bu_blocks,
282                                                  &bucket_xh(bucket)->xh_check);
283                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
284                 if (rc)
285                         mlog_errno(rc);
286         }
287
288         if (rc)
289                 ocfs2_xattr_bucket_relse(bucket);
290         return rc;
291 }
292
293 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
294                                              struct ocfs2_xattr_bucket *bucket,
295                                              int type)
296 {
297         int i, rc = 0;
298
299         for (i = 0; i < bucket->bu_blocks; i++) {
300                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
301                                           bucket->bu_bhs[i], type);
302                 if (rc) {
303                         mlog_errno(rc);
304                         break;
305                 }
306         }
307
308         return rc;
309 }
310
311 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
312                                              struct ocfs2_xattr_bucket *bucket)
313 {
314         int i;
315
316         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
317         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
318                                    bucket->bu_bhs, bucket->bu_blocks,
319                                    &bucket_xh(bucket)->xh_check);
320         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
321
322         for (i = 0; i < bucket->bu_blocks; i++)
323                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
324 }
325
326 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
327                                          struct ocfs2_xattr_bucket *src)
328 {
329         int i;
330         int blocksize = src->bu_inode->i_sb->s_blocksize;
331
332         BUG_ON(dest->bu_blocks != src->bu_blocks);
333         BUG_ON(dest->bu_inode != src->bu_inode);
334
335         for (i = 0; i < src->bu_blocks; i++) {
336                 memcpy(bucket_block(dest, i), bucket_block(src, i),
337                        blocksize);
338         }
339 }
340
341 static int ocfs2_validate_xattr_block(struct super_block *sb,
342                                       struct buffer_head *bh)
343 {
344         int rc;
345         struct ocfs2_xattr_block *xb =
346                 (struct ocfs2_xattr_block *)bh->b_data;
347
348         mlog(0, "Validating xattr block %llu\n",
349              (unsigned long long)bh->b_blocknr);
350
351         BUG_ON(!buffer_uptodate(bh));
352
353         /*
354          * If the ecc fails, we return the error but otherwise
355          * leave the filesystem running.  We know any error is
356          * local to this block.
357          */
358         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
359         if (rc)
360                 return rc;
361
362         /*
363          * Errors after here are fatal
364          */
365
366         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
367                 ocfs2_error(sb,
368                             "Extended attribute block #%llu has bad "
369                             "signature %.*s",
370                             (unsigned long long)bh->b_blocknr, 7,
371                             xb->xb_signature);
372                 return -EINVAL;
373         }
374
375         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
376                 ocfs2_error(sb,
377                             "Extended attribute block #%llu has an "
378                             "invalid xb_blkno of %llu",
379                             (unsigned long long)bh->b_blocknr,
380                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
381                 return -EINVAL;
382         }
383
384         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
385                 ocfs2_error(sb,
386                             "Extended attribute block #%llu has an invalid "
387                             "xb_fs_generation of #%u",
388                             (unsigned long long)bh->b_blocknr,
389                             le32_to_cpu(xb->xb_fs_generation));
390                 return -EINVAL;
391         }
392
393         return 0;
394 }
395
396 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
397                                   struct buffer_head **bh)
398 {
399         int rc;
400         struct buffer_head *tmp = *bh;
401
402         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
403                               ocfs2_validate_xattr_block);
404
405         /* If ocfs2_read_block() got us a new bh, pass it up. */
406         if (!rc && !*bh)
407                 *bh = tmp;
408
409         return rc;
410 }
411
412 static inline const char *ocfs2_xattr_prefix(int name_index)
413 {
414         struct xattr_handler *handler = NULL;
415
416         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
417                 handler = ocfs2_xattr_handler_map[name_index];
418
419         return handler ? handler->prefix : NULL;
420 }
421
422 static u32 ocfs2_xattr_name_hash(struct inode *inode,
423                                  const char *name,
424                                  int name_len)
425 {
426         /* Get hash value of uuid from super block */
427         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
428         int i;
429
430         /* hash extended attribute name */
431         for (i = 0; i < name_len; i++) {
432                 hash = (hash << OCFS2_HASH_SHIFT) ^
433                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
434                        *name++;
435         }
436
437         return hash;
438 }
439
440 /*
441  * ocfs2_xattr_hash_entry()
442  *
443  * Compute the hash of an extended attribute.
444  */
445 static void ocfs2_xattr_hash_entry(struct inode *inode,
446                                    struct ocfs2_xattr_header *header,
447                                    struct ocfs2_xattr_entry *entry)
448 {
449         u32 hash = 0;
450         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
451
452         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
453         entry->xe_name_hash = cpu_to_le32(hash);
454
455         return;
456 }
457
458 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
459 {
460         int size = 0;
461
462         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
463                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
464         else
465                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
466         size += sizeof(struct ocfs2_xattr_entry);
467
468         return size;
469 }
470
471 int ocfs2_calc_security_init(struct inode *dir,
472                              struct ocfs2_security_xattr_info *si,
473                              int *want_clusters,
474                              int *xattr_credits,
475                              struct ocfs2_alloc_context **xattr_ac)
476 {
477         int ret = 0;
478         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
479         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
480                                                  si->value_len);
481
482         /*
483          * The max space of security xattr taken inline is
484          * 256(name) + 80(value) + 16(entry) = 352 bytes,
485          * So reserve one metadata block for it is ok.
486          */
487         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
488             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
489                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
490                 if (ret) {
491                         mlog_errno(ret);
492                         return ret;
493                 }
494                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
495         }
496
497         /* reserve clusters for xattr value which will be set in B tree*/
498         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
499                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
500                                                             si->value_len);
501
502                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
503                                                            new_clusters);
504                 *want_clusters += new_clusters;
505         }
506         return ret;
507 }
508
509 int ocfs2_calc_xattr_init(struct inode *dir,
510                           struct buffer_head *dir_bh,
511                           int mode,
512                           struct ocfs2_security_xattr_info *si,
513                           int *want_clusters,
514                           int *xattr_credits,
515                           int *want_meta)
516 {
517         int ret = 0;
518         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
519         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
520
521         if (si->enable)
522                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
523                                                      si->value_len);
524
525         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
526                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
527                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
528                                         "", NULL, 0);
529                 if (acl_len > 0) {
530                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
531                         if (S_ISDIR(mode))
532                                 a_size <<= 1;
533                 } else if (acl_len != 0 && acl_len != -ENODATA) {
534                         mlog_errno(ret);
535                         return ret;
536                 }
537         }
538
539         if (!(s_size + a_size))
540                 return ret;
541
542         /*
543          * The max space of security xattr taken inline is
544          * 256(name) + 80(value) + 16(entry) = 352 bytes,
545          * The max space of acl xattr taken inline is
546          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
547          * when blocksize = 512, may reserve one more cluser for
548          * xattr bucket, otherwise reserve one metadata block
549          * for them is ok.
550          * If this is a new directory with inline data,
551          * we choose to reserve the entire inline area for
552          * directory contents and force an external xattr block.
553          */
554         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
555             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
556             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
557                 *want_meta = *want_meta + 1;
558                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
559         }
560
561         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
562             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
563                 *want_clusters += 1;
564                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
565         }
566
567         /*
568          * reserve credits and clusters for xattrs which has large value
569          * and have to be set outside
570          */
571         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
572                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
573                                                         si->value_len);
574                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
575                                                            new_clusters);
576                 *want_clusters += new_clusters;
577         }
578         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
579             acl_len > OCFS2_XATTR_INLINE_SIZE) {
580                 /* for directory, it has DEFAULT and ACCESS two types of acls */
581                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
582                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
583                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
584                                                            new_clusters);
585                 *want_clusters += new_clusters;
586         }
587
588         return ret;
589 }
590
591 static int ocfs2_xattr_extend_allocation(struct inode *inode,
592                                          u32 clusters_to_add,
593                                          struct ocfs2_xattr_value_buf *vb,
594                                          struct ocfs2_xattr_set_ctxt *ctxt)
595 {
596         int status = 0;
597         handle_t *handle = ctxt->handle;
598         enum ocfs2_alloc_restarted why;
599         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
600         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
601         struct ocfs2_extent_tree et;
602
603         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
604
605         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
606
607         status = vb->vb_access(handle, inode, vb->vb_bh,
608                               OCFS2_JOURNAL_ACCESS_WRITE);
609         if (status < 0) {
610                 mlog_errno(status);
611                 goto leave;
612         }
613
614         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
615         status = ocfs2_add_clusters_in_btree(osb,
616                                              inode,
617                                              &logical_start,
618                                              clusters_to_add,
619                                              0,
620                                              &et,
621                                              handle,
622                                              ctxt->data_ac,
623                                              ctxt->meta_ac,
624                                              &why);
625         if (status < 0) {
626                 mlog_errno(status);
627                 goto leave;
628         }
629
630         status = ocfs2_journal_dirty(handle, vb->vb_bh);
631         if (status < 0) {
632                 mlog_errno(status);
633                 goto leave;
634         }
635
636         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
637
638         /*
639          * We should have already allocated enough space before the transaction,
640          * so no need to restart.
641          */
642         BUG_ON(why != RESTART_NONE || clusters_to_add);
643
644 leave:
645
646         return status;
647 }
648
649 static int __ocfs2_remove_xattr_range(struct inode *inode,
650                                       struct ocfs2_xattr_value_buf *vb,
651                                       u32 cpos, u32 phys_cpos, u32 len,
652                                       struct ocfs2_xattr_set_ctxt *ctxt)
653 {
654         int ret;
655         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
656         handle_t *handle = ctxt->handle;
657         struct ocfs2_extent_tree et;
658
659         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
660
661         ret = vb->vb_access(handle, inode, vb->vb_bh,
662                             OCFS2_JOURNAL_ACCESS_WRITE);
663         if (ret) {
664                 mlog_errno(ret);
665                 goto out;
666         }
667
668         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
669                                   &ctxt->dealloc);
670         if (ret) {
671                 mlog_errno(ret);
672                 goto out;
673         }
674
675         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
676
677         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
678         if (ret) {
679                 mlog_errno(ret);
680                 goto out;
681         }
682
683         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
684         if (ret)
685                 mlog_errno(ret);
686
687 out:
688         return ret;
689 }
690
691 static int ocfs2_xattr_shrink_size(struct inode *inode,
692                                    u32 old_clusters,
693                                    u32 new_clusters,
694                                    struct ocfs2_xattr_value_buf *vb,
695                                    struct ocfs2_xattr_set_ctxt *ctxt)
696 {
697         int ret = 0;
698         u32 trunc_len, cpos, phys_cpos, alloc_size;
699         u64 block;
700
701         if (old_clusters <= new_clusters)
702                 return 0;
703
704         cpos = new_clusters;
705         trunc_len = old_clusters - new_clusters;
706         while (trunc_len) {
707                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
708                                                &alloc_size,
709                                                &vb->vb_xv->xr_list);
710                 if (ret) {
711                         mlog_errno(ret);
712                         goto out;
713                 }
714
715                 if (alloc_size > trunc_len)
716                         alloc_size = trunc_len;
717
718                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
719                                                  phys_cpos, alloc_size,
720                                                  ctxt);
721                 if (ret) {
722                         mlog_errno(ret);
723                         goto out;
724                 }
725
726                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
727                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
728                                                        alloc_size);
729                 cpos += alloc_size;
730                 trunc_len -= alloc_size;
731         }
732
733 out:
734         return ret;
735 }
736
737 static int ocfs2_xattr_value_truncate(struct inode *inode,
738                                       struct ocfs2_xattr_value_buf *vb,
739                                       int len,
740                                       struct ocfs2_xattr_set_ctxt *ctxt)
741 {
742         int ret;
743         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
744         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
745
746         if (new_clusters == old_clusters)
747                 return 0;
748
749         if (new_clusters > old_clusters)
750                 ret = ocfs2_xattr_extend_allocation(inode,
751                                                     new_clusters - old_clusters,
752                                                     vb, ctxt);
753         else
754                 ret = ocfs2_xattr_shrink_size(inode,
755                                               old_clusters, new_clusters,
756                                               vb, ctxt);
757
758         return ret;
759 }
760
761 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
762                                   size_t *result, const char *prefix,
763                                   const char *name, int name_len)
764 {
765         char *p = buffer + *result;
766         int prefix_len = strlen(prefix);
767         int total_len = prefix_len + name_len + 1;
768
769         *result += total_len;
770
771         /* we are just looking for how big our buffer needs to be */
772         if (!size)
773                 return 0;
774
775         if (*result > size)
776                 return -ERANGE;
777
778         memcpy(p, prefix, prefix_len);
779         memcpy(p + prefix_len, name, name_len);
780         p[prefix_len + name_len] = '\0';
781
782         return 0;
783 }
784
785 static int ocfs2_xattr_list_entries(struct inode *inode,
786                                     struct ocfs2_xattr_header *header,
787                                     char *buffer, size_t buffer_size)
788 {
789         size_t result = 0;
790         int i, type, ret;
791         const char *prefix, *name;
792
793         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
794                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
795                 type = ocfs2_xattr_get_type(entry);
796                 prefix = ocfs2_xattr_prefix(type);
797
798                 if (prefix) {
799                         name = (const char *)header +
800                                 le16_to_cpu(entry->xe_name_offset);
801
802                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
803                                                      &result, prefix, name,
804                                                      entry->xe_name_len);
805                         if (ret)
806                                 return ret;
807                 }
808         }
809
810         return result;
811 }
812
813 static int ocfs2_xattr_ibody_list(struct inode *inode,
814                                   struct ocfs2_dinode *di,
815                                   char *buffer,
816                                   size_t buffer_size)
817 {
818         struct ocfs2_xattr_header *header = NULL;
819         struct ocfs2_inode_info *oi = OCFS2_I(inode);
820         int ret = 0;
821
822         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
823                 return ret;
824
825         header = (struct ocfs2_xattr_header *)
826                  ((void *)di + inode->i_sb->s_blocksize -
827                  le16_to_cpu(di->i_xattr_inline_size));
828
829         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
830
831         return ret;
832 }
833
834 static int ocfs2_xattr_block_list(struct inode *inode,
835                                   struct ocfs2_dinode *di,
836                                   char *buffer,
837                                   size_t buffer_size)
838 {
839         struct buffer_head *blk_bh = NULL;
840         struct ocfs2_xattr_block *xb;
841         int ret = 0;
842
843         if (!di->i_xattr_loc)
844                 return ret;
845
846         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
847                                      &blk_bh);
848         if (ret < 0) {
849                 mlog_errno(ret);
850                 return ret;
851         }
852
853         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
854         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
855                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
856                 ret = ocfs2_xattr_list_entries(inode, header,
857                                                buffer, buffer_size);
858         } else {
859                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
860                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
861                                                    buffer, buffer_size);
862         }
863
864         brelse(blk_bh);
865
866         return ret;
867 }
868
869 ssize_t ocfs2_listxattr(struct dentry *dentry,
870                         char *buffer,
871                         size_t size)
872 {
873         int ret = 0, i_ret = 0, b_ret = 0;
874         struct buffer_head *di_bh = NULL;
875         struct ocfs2_dinode *di = NULL;
876         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
877
878         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
879                 return -EOPNOTSUPP;
880
881         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
882                 return ret;
883
884         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
885         if (ret < 0) {
886                 mlog_errno(ret);
887                 return ret;
888         }
889
890         di = (struct ocfs2_dinode *)di_bh->b_data;
891
892         down_read(&oi->ip_xattr_sem);
893         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
894         if (i_ret < 0)
895                 b_ret = 0;
896         else {
897                 if (buffer) {
898                         buffer += i_ret;
899                         size -= i_ret;
900                 }
901                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
902                                                buffer, size);
903                 if (b_ret < 0)
904                         i_ret = 0;
905         }
906         up_read(&oi->ip_xattr_sem);
907         ocfs2_inode_unlock(dentry->d_inode, 0);
908
909         brelse(di_bh);
910
911         return i_ret + b_ret;
912 }
913
914 static int ocfs2_xattr_find_entry(int name_index,
915                                   const char *name,
916                                   struct ocfs2_xattr_search *xs)
917 {
918         struct ocfs2_xattr_entry *entry;
919         size_t name_len;
920         int i, cmp = 1;
921
922         if (name == NULL)
923                 return -EINVAL;
924
925         name_len = strlen(name);
926         entry = xs->here;
927         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
928                 cmp = name_index - ocfs2_xattr_get_type(entry);
929                 if (!cmp)
930                         cmp = name_len - entry->xe_name_len;
931                 if (!cmp)
932                         cmp = memcmp(name, (xs->base +
933                                      le16_to_cpu(entry->xe_name_offset)),
934                                      name_len);
935                 if (cmp == 0)
936                         break;
937                 entry += 1;
938         }
939         xs->here = entry;
940
941         return cmp ? -ENODATA : 0;
942 }
943
944 static int ocfs2_xattr_get_value_outside(struct inode *inode,
945                                          struct ocfs2_xattr_value_root *xv,
946                                          void *buffer,
947                                          size_t len)
948 {
949         u32 cpos, p_cluster, num_clusters, bpc, clusters;
950         u64 blkno;
951         int i, ret = 0;
952         size_t cplen, blocksize;
953         struct buffer_head *bh = NULL;
954         struct ocfs2_extent_list *el;
955
956         el = &xv->xr_list;
957         clusters = le32_to_cpu(xv->xr_clusters);
958         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
959         blocksize = inode->i_sb->s_blocksize;
960
961         cpos = 0;
962         while (cpos < clusters) {
963                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
964                                                &num_clusters, el);
965                 if (ret) {
966                         mlog_errno(ret);
967                         goto out;
968                 }
969
970                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
971                 /* Copy ocfs2_xattr_value */
972                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
973                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
974                         if (ret) {
975                                 mlog_errno(ret);
976                                 goto out;
977                         }
978
979                         cplen = len >= blocksize ? blocksize : len;
980                         memcpy(buffer, bh->b_data, cplen);
981                         len -= cplen;
982                         buffer += cplen;
983
984                         brelse(bh);
985                         bh = NULL;
986                         if (len == 0)
987                                 break;
988                 }
989                 cpos += num_clusters;
990         }
991 out:
992         return ret;
993 }
994
995 static int ocfs2_xattr_ibody_get(struct inode *inode,
996                                  int name_index,
997                                  const char *name,
998                                  void *buffer,
999                                  size_t buffer_size,
1000                                  struct ocfs2_xattr_search *xs)
1001 {
1002         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1003         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1004         struct ocfs2_xattr_value_root *xv;
1005         size_t size;
1006         int ret = 0;
1007
1008         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1009                 return -ENODATA;
1010
1011         xs->end = (void *)di + inode->i_sb->s_blocksize;
1012         xs->header = (struct ocfs2_xattr_header *)
1013                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1014         xs->base = (void *)xs->header;
1015         xs->here = xs->header->xh_entries;
1016
1017         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1018         if (ret)
1019                 return ret;
1020         size = le64_to_cpu(xs->here->xe_value_size);
1021         if (buffer) {
1022                 if (size > buffer_size)
1023                         return -ERANGE;
1024                 if (ocfs2_xattr_is_local(xs->here)) {
1025                         memcpy(buffer, (void *)xs->base +
1026                                le16_to_cpu(xs->here->xe_name_offset) +
1027                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1028                 } else {
1029                         xv = (struct ocfs2_xattr_value_root *)
1030                                 (xs->base + le16_to_cpu(
1031                                  xs->here->xe_name_offset) +
1032                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1033                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1034                                                             buffer, size);
1035                         if (ret < 0) {
1036                                 mlog_errno(ret);
1037                                 return ret;
1038                         }
1039                 }
1040         }
1041
1042         return size;
1043 }
1044
1045 static int ocfs2_xattr_block_get(struct inode *inode,
1046                                  int name_index,
1047                                  const char *name,
1048                                  void *buffer,
1049                                  size_t buffer_size,
1050                                  struct ocfs2_xattr_search *xs)
1051 {
1052         struct ocfs2_xattr_block *xb;
1053         struct ocfs2_xattr_value_root *xv;
1054         size_t size;
1055         int ret = -ENODATA, name_offset, name_len, block_off, i;
1056
1057         xs->bucket = ocfs2_xattr_bucket_new(inode);
1058         if (!xs->bucket) {
1059                 ret = -ENOMEM;
1060                 mlog_errno(ret);
1061                 goto cleanup;
1062         }
1063
1064         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1065         if (ret) {
1066                 mlog_errno(ret);
1067                 goto cleanup;
1068         }
1069
1070         if (xs->not_found) {
1071                 ret = -ENODATA;
1072                 goto cleanup;
1073         }
1074
1075         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1076         size = le64_to_cpu(xs->here->xe_value_size);
1077         if (buffer) {
1078                 ret = -ERANGE;
1079                 if (size > buffer_size)
1080                         goto cleanup;
1081
1082                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1083                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1084                 i = xs->here - xs->header->xh_entries;
1085
1086                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1087                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1088                                                                 bucket_xh(xs->bucket),
1089                                                                 i,
1090                                                                 &block_off,
1091                                                                 &name_offset);
1092                         xs->base = bucket_block(xs->bucket, block_off);
1093                 }
1094                 if (ocfs2_xattr_is_local(xs->here)) {
1095                         memcpy(buffer, (void *)xs->base +
1096                                name_offset + name_len, size);
1097                 } else {
1098                         xv = (struct ocfs2_xattr_value_root *)
1099                                 (xs->base + name_offset + name_len);
1100                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1101                                                             buffer, size);
1102                         if (ret < 0) {
1103                                 mlog_errno(ret);
1104                                 goto cleanup;
1105                         }
1106                 }
1107         }
1108         ret = size;
1109 cleanup:
1110         ocfs2_xattr_bucket_free(xs->bucket);
1111
1112         brelse(xs->xattr_bh);
1113         xs->xattr_bh = NULL;
1114         return ret;
1115 }
1116
1117 int ocfs2_xattr_get_nolock(struct inode *inode,
1118                            struct buffer_head *di_bh,
1119                            int name_index,
1120                            const char *name,
1121                            void *buffer,
1122                            size_t buffer_size)
1123 {
1124         int ret;
1125         struct ocfs2_dinode *di = NULL;
1126         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1127         struct ocfs2_xattr_search xis = {
1128                 .not_found = -ENODATA,
1129         };
1130         struct ocfs2_xattr_search xbs = {
1131                 .not_found = -ENODATA,
1132         };
1133
1134         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1135                 return -EOPNOTSUPP;
1136
1137         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1138                 ret = -ENODATA;
1139
1140         xis.inode_bh = xbs.inode_bh = di_bh;
1141         di = (struct ocfs2_dinode *)di_bh->b_data;
1142
1143         down_read(&oi->ip_xattr_sem);
1144         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1145                                     buffer_size, &xis);
1146         if (ret == -ENODATA && di->i_xattr_loc)
1147                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1148                                             buffer_size, &xbs);
1149         up_read(&oi->ip_xattr_sem);
1150
1151         return ret;
1152 }
1153
1154 /* ocfs2_xattr_get()
1155  *
1156  * Copy an extended attribute into the buffer provided.
1157  * Buffer is NULL to compute the size of buffer required.
1158  */
1159 static int ocfs2_xattr_get(struct inode *inode,
1160                            int name_index,
1161                            const char *name,
1162                            void *buffer,
1163                            size_t buffer_size)
1164 {
1165         int ret;
1166         struct buffer_head *di_bh = NULL;
1167
1168         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1169         if (ret < 0) {
1170                 mlog_errno(ret);
1171                 return ret;
1172         }
1173         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1174                                      name, buffer, buffer_size);
1175
1176         ocfs2_inode_unlock(inode, 0);
1177
1178         brelse(di_bh);
1179
1180         return ret;
1181 }
1182
1183 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1184                                            handle_t *handle,
1185                                            struct ocfs2_xattr_value_root *xv,
1186                                            const void *value,
1187                                            int value_len)
1188 {
1189         int ret = 0, i, cp_len;
1190         u16 blocksize = inode->i_sb->s_blocksize;
1191         u32 p_cluster, num_clusters;
1192         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1193         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1194         u64 blkno;
1195         struct buffer_head *bh = NULL;
1196
1197         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1198
1199         while (cpos < clusters) {
1200                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1201                                                &num_clusters, &xv->xr_list);
1202                 if (ret) {
1203                         mlog_errno(ret);
1204                         goto out;
1205                 }
1206
1207                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1208
1209                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1210                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1211                         if (ret) {
1212                                 mlog_errno(ret);
1213                                 goto out;
1214                         }
1215
1216                         ret = ocfs2_journal_access(handle,
1217                                                    inode,
1218                                                    bh,
1219                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1220                         if (ret < 0) {
1221                                 mlog_errno(ret);
1222                                 goto out;
1223                         }
1224
1225                         cp_len = value_len > blocksize ? blocksize : value_len;
1226                         memcpy(bh->b_data, value, cp_len);
1227                         value_len -= cp_len;
1228                         value += cp_len;
1229                         if (cp_len < blocksize)
1230                                 memset(bh->b_data + cp_len, 0,
1231                                        blocksize - cp_len);
1232
1233                         ret = ocfs2_journal_dirty(handle, bh);
1234                         if (ret < 0) {
1235                                 mlog_errno(ret);
1236                                 goto out;
1237                         }
1238                         brelse(bh);
1239                         bh = NULL;
1240
1241                         /*
1242                          * XXX: do we need to empty all the following
1243                          * blocks in this cluster?
1244                          */
1245                         if (!value_len)
1246                                 break;
1247                 }
1248                 cpos += num_clusters;
1249         }
1250 out:
1251         brelse(bh);
1252
1253         return ret;
1254 }
1255
1256 static int ocfs2_xattr_cleanup(struct inode *inode,
1257                                handle_t *handle,
1258                                struct ocfs2_xattr_info *xi,
1259                                struct ocfs2_xattr_search *xs,
1260                                struct ocfs2_xattr_value_buf *vb,
1261                                size_t offs)
1262 {
1263         int ret = 0;
1264         size_t name_len = strlen(xi->name);
1265         void *val = xs->base + offs;
1266         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1267
1268         ret = vb->vb_access(handle, inode, vb->vb_bh,
1269                             OCFS2_JOURNAL_ACCESS_WRITE);
1270         if (ret) {
1271                 mlog_errno(ret);
1272                 goto out;
1273         }
1274         /* Decrease xattr count */
1275         le16_add_cpu(&xs->header->xh_count, -1);
1276         /* Remove the xattr entry and tree root which has already be set*/
1277         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1278         memset(val, 0, size);
1279
1280         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1281         if (ret < 0)
1282                 mlog_errno(ret);
1283 out:
1284         return ret;
1285 }
1286
1287 static int ocfs2_xattr_update_entry(struct inode *inode,
1288                                     handle_t *handle,
1289                                     struct ocfs2_xattr_info *xi,
1290                                     struct ocfs2_xattr_search *xs,
1291                                     struct ocfs2_xattr_value_buf *vb,
1292                                     size_t offs)
1293 {
1294         int ret;
1295
1296         ret = vb->vb_access(handle, inode, vb->vb_bh,
1297                             OCFS2_JOURNAL_ACCESS_WRITE);
1298         if (ret) {
1299                 mlog_errno(ret);
1300                 goto out;
1301         }
1302
1303         xs->here->xe_name_offset = cpu_to_le16(offs);
1304         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1305         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1306                 ocfs2_xattr_set_local(xs->here, 1);
1307         else
1308                 ocfs2_xattr_set_local(xs->here, 0);
1309         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1310
1311         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1312         if (ret < 0)
1313                 mlog_errno(ret);
1314 out:
1315         return ret;
1316 }
1317
1318 /*
1319  * ocfs2_xattr_set_value_outside()
1320  *
1321  * Set large size value in B tree.
1322  */
1323 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1324                                          struct ocfs2_xattr_info *xi,
1325                                          struct ocfs2_xattr_search *xs,
1326                                          struct ocfs2_xattr_set_ctxt *ctxt,
1327                                          struct ocfs2_xattr_value_buf *vb,
1328                                          size_t offs)
1329 {
1330         size_t name_len = strlen(xi->name);
1331         void *val = xs->base + offs;
1332         struct ocfs2_xattr_value_root *xv = NULL;
1333         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1334         int ret = 0;
1335
1336         memset(val, 0, size);
1337         memcpy(val, xi->name, name_len);
1338         xv = (struct ocfs2_xattr_value_root *)
1339                 (val + OCFS2_XATTR_SIZE(name_len));
1340         xv->xr_clusters = 0;
1341         xv->xr_last_eb_blk = 0;
1342         xv->xr_list.l_tree_depth = 0;
1343         xv->xr_list.l_count = cpu_to_le16(1);
1344         xv->xr_list.l_next_free_rec = 0;
1345         vb->vb_xv = xv;
1346
1347         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1348         if (ret < 0) {
1349                 mlog_errno(ret);
1350                 return ret;
1351         }
1352         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1353         if (ret < 0) {
1354                 mlog_errno(ret);
1355                 return ret;
1356         }
1357         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1358                                               xi->value, xi->value_len);
1359         if (ret < 0)
1360                 mlog_errno(ret);
1361
1362         return ret;
1363 }
1364
1365 /*
1366  * ocfs2_xattr_set_entry_local()
1367  *
1368  * Set, replace or remove extended attribute in local.
1369  */
1370 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1371                                         struct ocfs2_xattr_info *xi,
1372                                         struct ocfs2_xattr_search *xs,
1373                                         struct ocfs2_xattr_entry *last,
1374                                         size_t min_offs)
1375 {
1376         size_t name_len = strlen(xi->name);
1377         int i;
1378
1379         if (xi->value && xs->not_found) {
1380                 /* Insert the new xattr entry. */
1381                 le16_add_cpu(&xs->header->xh_count, 1);
1382                 ocfs2_xattr_set_type(last, xi->name_index);
1383                 ocfs2_xattr_set_local(last, 1);
1384                 last->xe_name_len = name_len;
1385         } else {
1386                 void *first_val;
1387                 void *val;
1388                 size_t offs, size;
1389
1390                 first_val = xs->base + min_offs;
1391                 offs = le16_to_cpu(xs->here->xe_name_offset);
1392                 val = xs->base + offs;
1393
1394                 if (le64_to_cpu(xs->here->xe_value_size) >
1395                     OCFS2_XATTR_INLINE_SIZE)
1396                         size = OCFS2_XATTR_SIZE(name_len) +
1397                                 OCFS2_XATTR_ROOT_SIZE;
1398                 else
1399                         size = OCFS2_XATTR_SIZE(name_len) +
1400                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1401
1402                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1403                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1404                         /* The old and the new value have the
1405                            same size. Just replace the value. */
1406                         ocfs2_xattr_set_local(xs->here, 1);
1407                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1408                         /* Clear value bytes. */
1409                         memset(val + OCFS2_XATTR_SIZE(name_len),
1410                                0,
1411                                OCFS2_XATTR_SIZE(xi->value_len));
1412                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1413                                xi->value,
1414                                xi->value_len);
1415                         return;
1416                 }
1417                 /* Remove the old name+value. */
1418                 memmove(first_val + size, first_val, val - first_val);
1419                 memset(first_val, 0, size);
1420                 xs->here->xe_name_hash = 0;
1421                 xs->here->xe_name_offset = 0;
1422                 ocfs2_xattr_set_local(xs->here, 1);
1423                 xs->here->xe_value_size = 0;
1424
1425                 min_offs += size;
1426
1427                 /* Adjust all value offsets. */
1428                 last = xs->header->xh_entries;
1429                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1430                         size_t o = le16_to_cpu(last->xe_name_offset);
1431
1432                         if (o < offs)
1433                                 last->xe_name_offset = cpu_to_le16(o + size);
1434                         last += 1;
1435                 }
1436
1437                 if (!xi->value) {
1438                         /* Remove the old entry. */
1439                         last -= 1;
1440                         memmove(xs->here, xs->here + 1,
1441                                 (void *)last - (void *)xs->here);
1442                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1443                         le16_add_cpu(&xs->header->xh_count, -1);
1444                 }
1445         }
1446         if (xi->value) {
1447                 /* Insert the new name+value. */
1448                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1449                                 OCFS2_XATTR_SIZE(xi->value_len);
1450                 void *val = xs->base + min_offs - size;
1451
1452                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1453                 memset(val, 0, size);
1454                 memcpy(val, xi->name, name_len);
1455                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1456                        xi->value,
1457                        xi->value_len);
1458                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1459                 ocfs2_xattr_set_local(xs->here, 1);
1460                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1461         }
1462
1463         return;
1464 }
1465
1466 /*
1467  * ocfs2_xattr_set_entry()
1468  *
1469  * Set extended attribute entry into inode or block.
1470  *
1471  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1472  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1473  * then set value in B tree with set_value_outside().
1474  */
1475 static int ocfs2_xattr_set_entry(struct inode *inode,
1476                                  struct ocfs2_xattr_info *xi,
1477                                  struct ocfs2_xattr_search *xs,
1478                                  struct ocfs2_xattr_set_ctxt *ctxt,
1479                                  int flag)
1480 {
1481         struct ocfs2_xattr_entry *last;
1482         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1483         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1484         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1485         size_t size_l = 0;
1486         handle_t *handle = ctxt->handle;
1487         int free, i, ret;
1488         struct ocfs2_xattr_info xi_l = {
1489                 .name_index = xi->name_index,
1490                 .name = xi->name,
1491                 .value = xi->value,
1492                 .value_len = xi->value_len,
1493         };
1494         struct ocfs2_xattr_value_buf vb = {
1495                 .vb_bh = xs->xattr_bh,
1496                 .vb_access = ocfs2_journal_access_di,
1497         };
1498
1499         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1500                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1501                 vb.vb_access = ocfs2_journal_access_xb;
1502         } else
1503                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1504
1505         /* Compute min_offs, last and free space. */
1506         last = xs->header->xh_entries;
1507
1508         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1509                 size_t offs = le16_to_cpu(last->xe_name_offset);
1510                 if (offs < min_offs)
1511                         min_offs = offs;
1512                 last += 1;
1513         }
1514
1515         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1516         if (free < 0)
1517                 return -EIO;
1518
1519         if (!xs->not_found) {
1520                 size_t size = 0;
1521                 if (ocfs2_xattr_is_local(xs->here))
1522                         size = OCFS2_XATTR_SIZE(name_len) +
1523                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1524                 else
1525                         size = OCFS2_XATTR_SIZE(name_len) +
1526                                 OCFS2_XATTR_ROOT_SIZE;
1527                 free += (size + sizeof(struct ocfs2_xattr_entry));
1528         }
1529         /* Check free space in inode or block */
1530         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1531                 if (free < sizeof(struct ocfs2_xattr_entry) +
1532                            OCFS2_XATTR_SIZE(name_len) +
1533                            OCFS2_XATTR_ROOT_SIZE) {
1534                         ret = -ENOSPC;
1535                         goto out;
1536                 }
1537                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1538                 xi_l.value = (void *)&def_xv;
1539                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1540         } else if (xi->value) {
1541                 if (free < sizeof(struct ocfs2_xattr_entry) +
1542                            OCFS2_XATTR_SIZE(name_len) +
1543                            OCFS2_XATTR_SIZE(xi->value_len)) {
1544                         ret = -ENOSPC;
1545                         goto out;
1546                 }
1547         }
1548
1549         if (!xs->not_found) {
1550                 /* For existing extended attribute */
1551                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1552                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1553                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1554                 void *val = xs->base + offs;
1555
1556                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1557                         /* Replace existing local xattr with tree root */
1558                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1559                                                             ctxt, &vb, offs);
1560                         if (ret < 0)
1561                                 mlog_errno(ret);
1562                         goto out;
1563                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1564                         /* For existing xattr which has value outside */
1565                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1566                                 (val + OCFS2_XATTR_SIZE(name_len));
1567
1568                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1569                                 /*
1570                                  * If new value need set outside also,
1571                                  * first truncate old value to new value,
1572                                  * then set new value with set_value_outside().
1573                                  */
1574                                 ret = ocfs2_xattr_value_truncate(inode,
1575                                                                  &vb,
1576                                                                  xi->value_len,
1577                                                                  ctxt);
1578                                 if (ret < 0) {
1579                                         mlog_errno(ret);
1580                                         goto out;
1581                                 }
1582
1583                                 ret = ocfs2_xattr_update_entry(inode,
1584                                                                handle,
1585                                                                xi,
1586                                                                xs,
1587                                                                &vb,
1588                                                                offs);
1589                                 if (ret < 0) {
1590                                         mlog_errno(ret);
1591                                         goto out;
1592                                 }
1593
1594                                 ret = __ocfs2_xattr_set_value_outside(inode,
1595                                                                 handle,
1596                                                                 vb.vb_xv,
1597                                                                 xi->value,
1598                                                                 xi->value_len);
1599                                 if (ret < 0)
1600                                         mlog_errno(ret);
1601                                 goto out;
1602                         } else {
1603                                 /*
1604                                  * If new value need set in local,
1605                                  * just trucate old value to zero.
1606                                  */
1607                                  ret = ocfs2_xattr_value_truncate(inode,
1608                                                                   &vb,
1609                                                                   0,
1610                                                                   ctxt);
1611                                 if (ret < 0)
1612                                         mlog_errno(ret);
1613                         }
1614                 }
1615         }
1616
1617         ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1618                                       OCFS2_JOURNAL_ACCESS_WRITE);
1619         if (ret) {
1620                 mlog_errno(ret);
1621                 goto out;
1622         }
1623
1624         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1625                 ret = vb.vb_access(handle, inode, vb.vb_bh,
1626                                    OCFS2_JOURNAL_ACCESS_WRITE);
1627                 if (ret) {
1628                         mlog_errno(ret);
1629                         goto out;
1630                 }
1631         }
1632
1633         /*
1634          * Set value in local, include set tree root in local.
1635          * This is the first step for value size >INLINE_SIZE.
1636          */
1637         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1638
1639         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1640                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1641                 if (ret < 0) {
1642                         mlog_errno(ret);
1643                         goto out;
1644                 }
1645         }
1646
1647         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1648             (flag & OCFS2_INLINE_XATTR_FL)) {
1649                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1650                 unsigned int xattrsize = osb->s_xattr_inline_size;
1651
1652                 /*
1653                  * Adjust extent record count or inline data size
1654                  * to reserve space for extended attribute.
1655                  */
1656                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1657                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1658                         le16_add_cpu(&idata->id_count, -xattrsize);
1659                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1660                         struct ocfs2_extent_list *el = &di->id2.i_list;
1661                         le16_add_cpu(&el->l_count, -(xattrsize /
1662                                         sizeof(struct ocfs2_extent_rec)));
1663                 }
1664                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1665         }
1666         /* Update xattr flag */
1667         spin_lock(&oi->ip_lock);
1668         oi->ip_dyn_features |= flag;
1669         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1670         spin_unlock(&oi->ip_lock);
1671
1672         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1673         if (ret < 0)
1674                 mlog_errno(ret);
1675
1676         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1677                 /*
1678                  * Set value outside in B tree.
1679                  * This is the second step for value size > INLINE_SIZE.
1680                  */
1681                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1682                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1683                                                     &vb, offs);
1684                 if (ret < 0) {
1685                         int ret2;
1686
1687                         mlog_errno(ret);
1688                         /*
1689                          * If set value outside failed, we have to clean
1690                          * the junk tree root we have already set in local.
1691                          */
1692                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1693                                                    xi, xs, &vb, offs);
1694                         if (ret2 < 0)
1695                                 mlog_errno(ret2);
1696                 }
1697         }
1698 out:
1699         return ret;
1700 }
1701
1702 static int ocfs2_remove_value_outside(struct inode*inode,
1703                                       struct ocfs2_xattr_value_buf *vb,
1704                                       struct ocfs2_xattr_header *header)
1705 {
1706         int ret = 0, i;
1707         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1708         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1709
1710         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1711
1712         ctxt.handle = ocfs2_start_trans(osb,
1713                                         ocfs2_remove_extent_credits(osb->sb));
1714         if (IS_ERR(ctxt.handle)) {
1715                 ret = PTR_ERR(ctxt.handle);
1716                 mlog_errno(ret);
1717                 goto out;
1718         }
1719
1720         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1721                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1722
1723                 if (!ocfs2_xattr_is_local(entry)) {
1724                         void *val;
1725
1726                         val = (void *)header +
1727                                 le16_to_cpu(entry->xe_name_offset);
1728                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1729                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1730                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1731                         if (ret < 0) {
1732                                 mlog_errno(ret);
1733                                 break;
1734                         }
1735                 }
1736         }
1737
1738         ocfs2_commit_trans(osb, ctxt.handle);
1739         ocfs2_schedule_truncate_log_flush(osb, 1);
1740         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1741 out:
1742         return ret;
1743 }
1744
1745 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1746                                     struct buffer_head *di_bh)
1747 {
1748
1749         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1750         struct ocfs2_xattr_header *header;
1751         int ret;
1752         struct ocfs2_xattr_value_buf vb = {
1753                 .vb_bh = di_bh,
1754                 .vb_access = ocfs2_journal_access_di,
1755         };
1756
1757         header = (struct ocfs2_xattr_header *)
1758                  ((void *)di + inode->i_sb->s_blocksize -
1759                  le16_to_cpu(di->i_xattr_inline_size));
1760
1761         ret = ocfs2_remove_value_outside(inode, &vb, header);
1762
1763         return ret;
1764 }
1765
1766 static int ocfs2_xattr_block_remove(struct inode *inode,
1767                                     struct buffer_head *blk_bh)
1768 {
1769         struct ocfs2_xattr_block *xb;
1770         int ret = 0;
1771         struct ocfs2_xattr_value_buf vb = {
1772                 .vb_bh = blk_bh,
1773                 .vb_access = ocfs2_journal_access_xb,
1774         };
1775
1776         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1777         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1778                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1779                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1780         } else
1781                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1782
1783         return ret;
1784 }
1785
1786 static int ocfs2_xattr_free_block(struct inode *inode,
1787                                   u64 block)
1788 {
1789         struct inode *xb_alloc_inode;
1790         struct buffer_head *xb_alloc_bh = NULL;
1791         struct buffer_head *blk_bh = NULL;
1792         struct ocfs2_xattr_block *xb;
1793         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1794         handle_t *handle;
1795         int ret = 0;
1796         u64 blk, bg_blkno;
1797         u16 bit;
1798
1799         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1800         if (ret < 0) {
1801                 mlog_errno(ret);
1802                 goto out;
1803         }
1804
1805         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1806         if (ret < 0) {
1807                 mlog_errno(ret);
1808                 goto out;
1809         }
1810
1811         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1812         blk = le64_to_cpu(xb->xb_blkno);
1813         bit = le16_to_cpu(xb->xb_suballoc_bit);
1814         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1815
1816         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1817                                 EXTENT_ALLOC_SYSTEM_INODE,
1818                                 le16_to_cpu(xb->xb_suballoc_slot));
1819         if (!xb_alloc_inode) {
1820                 ret = -ENOMEM;
1821                 mlog_errno(ret);
1822                 goto out;
1823         }
1824         mutex_lock(&xb_alloc_inode->i_mutex);
1825
1826         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1827         if (ret < 0) {
1828                 mlog_errno(ret);
1829                 goto out_mutex;
1830         }
1831
1832         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1833         if (IS_ERR(handle)) {
1834                 ret = PTR_ERR(handle);
1835                 mlog_errno(ret);
1836                 goto out_unlock;
1837         }
1838
1839         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1840                                        bit, bg_blkno, 1);
1841         if (ret < 0)
1842                 mlog_errno(ret);
1843
1844         ocfs2_commit_trans(osb, handle);
1845 out_unlock:
1846         ocfs2_inode_unlock(xb_alloc_inode, 1);
1847         brelse(xb_alloc_bh);
1848 out_mutex:
1849         mutex_unlock(&xb_alloc_inode->i_mutex);
1850         iput(xb_alloc_inode);
1851 out:
1852         brelse(blk_bh);
1853         return ret;
1854 }
1855
1856 /*
1857  * ocfs2_xattr_remove()
1858  *
1859  * Free extended attribute resources associated with this inode.
1860  */
1861 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1862 {
1863         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1864         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1865         handle_t *handle;
1866         int ret;
1867
1868         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1869                 return 0;
1870
1871         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1872                 return 0;
1873
1874         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1875                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1876                 if (ret < 0) {
1877                         mlog_errno(ret);
1878                         goto out;
1879                 }
1880         }
1881
1882         if (di->i_xattr_loc) {
1883                 ret = ocfs2_xattr_free_block(inode,
1884                                              le64_to_cpu(di->i_xattr_loc));
1885                 if (ret < 0) {
1886                         mlog_errno(ret);
1887                         goto out;
1888                 }
1889         }
1890
1891         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1892                                    OCFS2_INODE_UPDATE_CREDITS);
1893         if (IS_ERR(handle)) {
1894                 ret = PTR_ERR(handle);
1895                 mlog_errno(ret);
1896                 goto out;
1897         }
1898         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1899                                       OCFS2_JOURNAL_ACCESS_WRITE);
1900         if (ret) {
1901                 mlog_errno(ret);
1902                 goto out_commit;
1903         }
1904
1905         di->i_xattr_loc = 0;
1906
1907         spin_lock(&oi->ip_lock);
1908         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1909         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1910         spin_unlock(&oi->ip_lock);
1911
1912         ret = ocfs2_journal_dirty(handle, di_bh);
1913         if (ret < 0)
1914                 mlog_errno(ret);
1915 out_commit:
1916         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1917 out:
1918         return ret;
1919 }
1920
1921 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1922                                         struct ocfs2_dinode *di)
1923 {
1924         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1925         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1926         int free;
1927
1928         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1929                 return 0;
1930
1931         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1932                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1933                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1934         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1935                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1936                         le64_to_cpu(di->i_size);
1937         } else {
1938                 struct ocfs2_extent_list *el = &di->id2.i_list;
1939                 free = (le16_to_cpu(el->l_count) -
1940                         le16_to_cpu(el->l_next_free_rec)) *
1941                         sizeof(struct ocfs2_extent_rec);
1942         }
1943         if (free >= xattrsize)
1944                 return 1;
1945
1946         return 0;
1947 }
1948
1949 /*
1950  * ocfs2_xattr_ibody_find()
1951  *
1952  * Find extended attribute in inode block and
1953  * fill search info into struct ocfs2_xattr_search.
1954  */
1955 static int ocfs2_xattr_ibody_find(struct inode *inode,
1956                                   int name_index,
1957                                   const char *name,
1958                                   struct ocfs2_xattr_search *xs)
1959 {
1960         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1961         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1962         int ret;
1963         int has_space = 0;
1964
1965         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1966                 return 0;
1967
1968         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1969                 down_read(&oi->ip_alloc_sem);
1970                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1971                 up_read(&oi->ip_alloc_sem);
1972                 if (!has_space)
1973                         return 0;
1974         }
1975
1976         xs->xattr_bh = xs->inode_bh;
1977         xs->end = (void *)di + inode->i_sb->s_blocksize;
1978         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1979                 xs->header = (struct ocfs2_xattr_header *)
1980                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1981         else
1982                 xs->header = (struct ocfs2_xattr_header *)
1983                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1984         xs->base = (void *)xs->header;
1985         xs->here = xs->header->xh_entries;
1986
1987         /* Find the named attribute. */
1988         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1989                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1990                 if (ret && ret != -ENODATA)
1991                         return ret;
1992                 xs->not_found = ret;
1993         }
1994
1995         return 0;
1996 }
1997
1998 /*
1999  * ocfs2_xattr_ibody_set()
2000  *
2001  * Set, replace or remove an extended attribute into inode block.
2002  *
2003  */
2004 static int ocfs2_xattr_ibody_set(struct inode *inode,
2005                                  struct ocfs2_xattr_info *xi,
2006                                  struct ocfs2_xattr_search *xs,
2007                                  struct ocfs2_xattr_set_ctxt *ctxt)
2008 {
2009         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2010         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2011         int ret;
2012
2013         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2014                 return -ENOSPC;
2015
2016         down_write(&oi->ip_alloc_sem);
2017         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2018                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2019                         ret = -ENOSPC;
2020                         goto out;
2021                 }
2022         }
2023
2024         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2025                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2026 out:
2027         up_write(&oi->ip_alloc_sem);
2028
2029         return ret;
2030 }
2031
2032 /*
2033  * ocfs2_xattr_block_find()
2034  *
2035  * Find extended attribute in external block and
2036  * fill search info into struct ocfs2_xattr_search.
2037  */
2038 static int ocfs2_xattr_block_find(struct inode *inode,
2039                                   int name_index,
2040                                   const char *name,
2041                                   struct ocfs2_xattr_search *xs)
2042 {
2043         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2044         struct buffer_head *blk_bh = NULL;
2045         struct ocfs2_xattr_block *xb;
2046         int ret = 0;
2047
2048         if (!di->i_xattr_loc)
2049                 return ret;
2050
2051         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2052                                      &blk_bh);
2053         if (ret < 0) {
2054                 mlog_errno(ret);
2055                 return ret;
2056         }
2057
2058         xs->xattr_bh = blk_bh;
2059         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2060
2061         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2062                 xs->header = &xb->xb_attrs.xb_header;
2063                 xs->base = (void *)xs->header;
2064                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2065                 xs->here = xs->header->xh_entries;
2066
2067                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2068         } else
2069                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2070                                                    name_index,
2071                                                    name, xs);
2072
2073         if (ret && ret != -ENODATA) {
2074                 xs->xattr_bh = NULL;
2075                 goto cleanup;
2076         }
2077         xs->not_found = ret;
2078         return 0;
2079 cleanup:
2080         brelse(blk_bh);
2081
2082         return ret;
2083 }
2084
2085 /*
2086  * ocfs2_xattr_block_set()
2087  *
2088  * Set, replace or remove an extended attribute into external block.
2089  *
2090  */
2091 static int ocfs2_xattr_block_set(struct inode *inode,
2092                                  struct ocfs2_xattr_info *xi,
2093                                  struct ocfs2_xattr_search *xs,
2094                                  struct ocfs2_xattr_set_ctxt *ctxt)
2095 {
2096         struct buffer_head *new_bh = NULL;
2097         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2098         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2099         handle_t *handle = ctxt->handle;
2100         struct ocfs2_xattr_block *xblk = NULL;
2101         u16 suballoc_bit_start;
2102         u32 num_got;
2103         u64 first_blkno;
2104         int ret;
2105
2106         if (!xs->xattr_bh) {
2107                 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2108                                               OCFS2_JOURNAL_ACCESS_CREATE);
2109                 if (ret < 0) {
2110                         mlog_errno(ret);
2111                         goto end;
2112                 }
2113
2114                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2115                                            &suballoc_bit_start, &num_got,
2116                                            &first_blkno);
2117                 if (ret < 0) {
2118                         mlog_errno(ret);
2119                         goto end;
2120                 }
2121
2122                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2123                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2124
2125                 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2126                                               OCFS2_JOURNAL_ACCESS_CREATE);
2127                 if (ret < 0) {
2128                         mlog_errno(ret);
2129                         goto end;
2130                 }
2131
2132                 /* Initialize ocfs2_xattr_block */
2133                 xs->xattr_bh = new_bh;
2134                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2135                 memset(xblk, 0, inode->i_sb->s_blocksize);
2136                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2137                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2138                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2139                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2140                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2141
2142                 xs->header = &xblk->xb_attrs.xb_header;
2143                 xs->base = (void *)xs->header;
2144                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2145                 xs->here = xs->header->xh_entries;
2146
2147                 ret = ocfs2_journal_dirty(handle, new_bh);
2148                 if (ret < 0) {
2149                         mlog_errno(ret);
2150                         goto end;
2151                 }
2152                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2153                 ocfs2_journal_dirty(handle, xs->inode_bh);
2154         } else
2155                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2156
2157         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2158                 /* Set extended attribute into external block */
2159                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2160                                             OCFS2_HAS_XATTR_FL);
2161                 if (!ret || ret != -ENOSPC)
2162                         goto end;
2163
2164                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2165                 if (ret)
2166                         goto end;
2167         }
2168
2169         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2170
2171 end:
2172
2173         return ret;
2174 }
2175
2176 /* Check whether the new xattr can be inserted into the inode. */
2177 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2178                                        struct ocfs2_xattr_info *xi,
2179                                        struct ocfs2_xattr_search *xs)
2180 {
2181         u64 value_size;
2182         struct ocfs2_xattr_entry *last;
2183         int free, i;
2184         size_t min_offs = xs->end - xs->base;
2185
2186         if (!xs->header)
2187                 return 0;
2188
2189         last = xs->header->xh_entries;
2190
2191         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2192                 size_t offs = le16_to_cpu(last->xe_name_offset);
2193                 if (offs < min_offs)
2194                         min_offs = offs;
2195                 last += 1;
2196         }
2197
2198         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2199         if (free < 0)
2200                 return 0;
2201
2202         BUG_ON(!xs->not_found);
2203
2204         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2205                 value_size = OCFS2_XATTR_ROOT_SIZE;
2206         else
2207                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2208
2209         if (free >= sizeof(struct ocfs2_xattr_entry) +
2210                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2211                 return 1;
2212
2213         return 0;
2214 }
2215
2216 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2217                                      struct ocfs2_dinode *di,
2218                                      struct ocfs2_xattr_info *xi,
2219                                      struct ocfs2_xattr_search *xis,
2220                                      struct ocfs2_xattr_search *xbs,
2221                                      int *clusters_need,
2222                                      int *meta_need,
2223                                      int *credits_need)
2224 {
2225         int ret = 0, old_in_xb = 0;
2226         int clusters_add = 0, meta_add = 0, credits = 0;
2227         struct buffer_head *bh = NULL;
2228         struct ocfs2_xattr_block *xb = NULL;
2229         struct ocfs2_xattr_entry *xe = NULL;
2230         struct ocfs2_xattr_value_root *xv = NULL;
2231         char *base = NULL;
2232         int name_offset, name_len = 0;
2233         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2234                                                     xi->value_len);
2235         u64 value_size;
2236
2237         /*
2238          * Calculate the clusters we need to write.
2239          * No matter whether we replace an old one or add a new one,
2240          * we need this for writing.
2241          */
2242         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2243                 credits += new_clusters *
2244                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2245
2246         if (xis->not_found && xbs->not_found) {
2247                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2248
2249                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2250                         clusters_add += new_clusters;
2251                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2252                                                         &def_xv.xv.xr_list,
2253                                                         new_clusters);
2254                 }
2255
2256                 goto meta_guess;
2257         }
2258
2259         if (!xis->not_found) {
2260                 xe = xis->here;
2261                 name_offset = le16_to_cpu(xe->xe_name_offset);
2262                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2263                 base = xis->base;
2264                 credits += OCFS2_INODE_UPDATE_CREDITS;
2265         } else {
2266                 int i, block_off = 0;
2267                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2268                 xe = xbs->here;
2269                 name_offset = le16_to_cpu(xe->xe_name_offset);
2270                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2271                 i = xbs->here - xbs->header->xh_entries;
2272                 old_in_xb = 1;
2273
2274                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2275                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2276                                                         bucket_xh(xbs->bucket),
2277                                                         i, &block_off,
2278                                                         &name_offset);
2279                         base = bucket_block(xbs->bucket, block_off);
2280                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2281                 } else {
2282                         base = xbs->base;
2283                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2284                 }
2285         }
2286
2287         /*
2288          * delete a xattr doesn't need metadata and cluster allocation.
2289          * so just calculate the credits and return.
2290          *
2291          * The credits for removing the value tree will be extended
2292          * by ocfs2_remove_extent itself.
2293          */
2294         if (!xi->value) {
2295                 if (!ocfs2_xattr_is_local(xe))
2296                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2297
2298                 goto out;
2299         }
2300
2301         /* do cluster allocation guess first. */
2302         value_size = le64_to_cpu(xe->xe_value_size);
2303
2304         if (old_in_xb) {
2305                 /*
2306                  * In xattr set, we always try to set the xe in inode first,
2307                  * so if it can be inserted into inode successfully, the old
2308                  * one will be removed from the xattr block, and this xattr
2309                  * will be inserted into inode as a new xattr in inode.
2310                  */
2311                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2312                         clusters_add += new_clusters;
2313                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2314                                     OCFS2_INODE_UPDATE_CREDITS;
2315                         if (!ocfs2_xattr_is_local(xe))
2316                                 credits += ocfs2_calc_extend_credits(
2317                                                         inode->i_sb,
2318                                                         &def_xv.xv.xr_list,
2319                                                         new_clusters);
2320                         goto out;
2321                 }
2322         }
2323
2324         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2325                 /* the new values will be stored outside. */
2326                 u32 old_clusters = 0;
2327
2328                 if (!ocfs2_xattr_is_local(xe)) {
2329                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2330                                                                  value_size);
2331                         xv = (struct ocfs2_xattr_value_root *)
2332                              (base + name_offset + name_len);
2333                         value_size = OCFS2_XATTR_ROOT_SIZE;
2334                 } else
2335                         xv = &def_xv.xv;
2336
2337                 if (old_clusters >= new_clusters) {
2338                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2339                         goto out;
2340                 } else {
2341                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2342                         clusters_add += new_clusters - old_clusters;
2343                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2344                                                              &xv->xr_list,
2345                                                              new_clusters -
2346                                                              old_clusters);
2347                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2348                                 goto out;
2349                 }
2350         } else {
2351                 /*
2352                  * Now the new value will be stored inside. So if the new
2353                  * value is smaller than the size of value root or the old
2354                  * value, we don't need any allocation, otherwise we have
2355                  * to guess metadata allocation.
2356                  */
2357                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2358                     (!ocfs2_xattr_is_local(xe) &&
2359                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2360                         goto out;
2361         }
2362
2363 meta_guess:
2364         /* calculate metadata allocation. */
2365         if (di->i_xattr_loc) {
2366                 if (!xbs->xattr_bh) {
2367                         ret = ocfs2_read_xattr_block(inode,
2368                                                      le64_to_cpu(di->i_xattr_loc),
2369                                                      &bh);
2370                         if (ret) {
2371                                 mlog_errno(ret);
2372                                 goto out;
2373                         }
2374
2375                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2376                 } else
2377                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2378
2379                 /*
2380                  * If there is already an xattr tree, good, we can calculate
2381                  * like other b-trees. Otherwise we may have the chance of
2382                  * create a tree, the credit calculation is borrowed from
2383                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2384                  * new tree will be cluster based, so no meta is needed.
2385                  */
2386                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2387                         struct ocfs2_extent_list *el =
2388                                  &xb->xb_attrs.xb_root.xt_list;
2389                         meta_add += ocfs2_extend_meta_needed(el);
2390                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2391                                                              el, 1);
2392                 } else
2393                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2394
2395                 /*
2396                  * This cluster will be used either for new bucket or for
2397                  * new xattr block.
2398                  * If the cluster size is the same as the bucket size, one
2399                  * more is needed since we may need to extend the bucket
2400                  * also.
2401                  */
2402                 clusters_add += 1;
2403                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2404                 if (OCFS2_XATTR_BUCKET_SIZE ==
2405                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2406                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2407                         clusters_add += 1;
2408                 }
2409         } else {
2410                 meta_add += 1;
2411                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2412         }
2413 out:
2414         if (clusters_need)
2415                 *clusters_need = clusters_add;
2416         if (meta_need)
2417                 *meta_need = meta_add;
2418         if (credits_need)
2419                 *credits_need = credits;
2420         brelse(bh);
2421         return ret;
2422 }
2423
2424 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2425                                      struct ocfs2_dinode *di,
2426                                      struct ocfs2_xattr_info *xi,
2427                                      struct ocfs2_xattr_search *xis,
2428                                      struct ocfs2_xattr_search *xbs,
2429                                      struct ocfs2_xattr_set_ctxt *ctxt,
2430                                      int *credits)
2431 {
2432         int clusters_add, meta_add, ret;
2433         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2434
2435         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2436
2437         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2438
2439         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2440                                         &clusters_add, &meta_add, credits);
2441         if (ret) {
2442                 mlog_errno(ret);
2443                 return ret;
2444         }
2445
2446         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2447              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2448
2449         if (meta_add) {
2450                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2451                                                         &ctxt->meta_ac);
2452                 if (ret) {
2453                         mlog_errno(ret);
2454                         goto out;
2455                 }
2456         }
2457
2458         if (clusters_add) {
2459                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2460                 if (ret)
2461                         mlog_errno(ret);
2462         }
2463 out:
2464         if (ret) {
2465                 if (ctxt->meta_ac) {
2466                         ocfs2_free_alloc_context(ctxt->meta_ac);
2467                         ctxt->meta_ac = NULL;
2468                 }
2469
2470                 /*
2471                  * We cannot have an error and a non null ctxt->data_ac.
2472                  */
2473         }
2474
2475         return ret;
2476 }
2477
2478 static int __ocfs2_xattr_set_handle(struct inode *inode,
2479                                     struct ocfs2_dinode *di,
2480                                     struct ocfs2_xattr_info *xi,
2481                                     struct ocfs2_xattr_search *xis,
2482                                     struct ocfs2_xattr_search *xbs,
2483                                     struct ocfs2_xattr_set_ctxt *ctxt)
2484 {
2485         int ret = 0, credits, old_found;
2486
2487         if (!xi->value) {
2488                 /* Remove existing extended attribute */
2489                 if (!xis->not_found)
2490                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2491                 else if (!xbs->not_found)
2492                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2493         } else {
2494                 /* We always try to set extended attribute into inode first*/
2495                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2496                 if (!ret && !xbs->not_found) {
2497                         /*
2498                          * If succeed and that extended attribute existing in
2499                          * external block, then we will remove it.
2500                          */
2501                         xi->value = NULL;
2502                         xi->value_len = 0;
2503
2504                         old_found = xis->not_found;
2505                         xis->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                         xis->not_found = old_found;
2515                         if (ret) {
2516                                 mlog_errno(ret);
2517                                 goto out;
2518                         }
2519
2520                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2521                                         ctxt->handle->h_buffer_credits);
2522                         if (ret) {
2523                                 mlog_errno(ret);
2524                                 goto out;
2525                         }
2526                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2527                 } else if (ret == -ENOSPC) {
2528                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2529                                 ret = ocfs2_xattr_block_find(inode,
2530                                                              xi->name_index,
2531                                                              xi->name, xbs);
2532                                 if (ret)
2533                                         goto out;
2534
2535                                 old_found = xis->not_found;
2536                                 xis->not_found = -ENODATA;
2537                                 ret = ocfs2_calc_xattr_set_need(inode,
2538                                                                 di,
2539                                                                 xi,
2540                                                                 xis,
2541                                                                 xbs,
2542                                                                 NULL,
2543                                                                 NULL,
2544                                                                 &credits);
2545                                 xis->not_found = old_found;
2546                                 if (ret) {
2547                                         mlog_errno(ret);
2548                                         goto out;
2549                                 }
2550
2551                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2552                                         ctxt->handle->h_buffer_credits);
2553                                 if (ret) {
2554                                         mlog_errno(ret);
2555                                         goto out;
2556                                 }
2557                         }
2558                         /*
2559                          * If no space in inode, we will set extended attribute
2560                          * into external block.
2561                          */
2562                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2563                         if (ret)
2564                                 goto out;
2565                         if (!xis->not_found) {
2566                                 /*
2567                                  * If succeed and that extended attribute
2568                                  * existing in inode, we will remove it.
2569                                  */
2570                                 xi->value = NULL;
2571                                 xi->value_len = 0;
2572                                 xbs->not_found = -ENODATA;
2573                                 ret = ocfs2_calc_xattr_set_need(inode,
2574                                                                 di,
2575                                                                 xi,
2576                                                                 xis,
2577                                                                 xbs,
2578                                                                 NULL,
2579                                                                 NULL,
2580                                                                 &credits);
2581                                 if (ret) {
2582                                         mlog_errno(ret);
2583                                         goto out;
2584                                 }
2585
2586                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2587                                                 ctxt->handle->h_buffer_credits);
2588                                 if (ret) {
2589                                         mlog_errno(ret);
2590                                         goto out;
2591                                 }
2592                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2593                                                             xis, ctxt);
2594                         }
2595                 }
2596         }
2597
2598         if (!ret) {
2599                 /* Update inode ctime. */
2600                 ret = ocfs2_journal_access_di(ctxt->handle, inode,
2601                                               xis->inode_bh,
2602                                               OCFS2_JOURNAL_ACCESS_WRITE);
2603                 if (ret) {
2604                         mlog_errno(ret);
2605                         goto out;
2606                 }
2607
2608                 inode->i_ctime = CURRENT_TIME;
2609                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2610                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2611                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2612         }
2613 out:
2614         return ret;
2615 }
2616
2617 /*
2618  * This function only called duing creating inode
2619  * for init security/acl xattrs of the new inode.
2620  * All transanction credits have been reserved in mknod.
2621  */
2622 int ocfs2_xattr_set_handle(handle_t *handle,
2623                            struct inode *inode,
2624                            struct buffer_head *di_bh,
2625                            int name_index,
2626                            const char *name,
2627                            const void *value,
2628                            size_t value_len,
2629                            int flags,
2630                            struct ocfs2_alloc_context *meta_ac,
2631                            struct ocfs2_alloc_context *data_ac)
2632 {
2633         struct ocfs2_dinode *di;
2634         int ret;
2635
2636         struct ocfs2_xattr_info xi = {
2637                 .name_index = name_index,
2638                 .name = name,
2639                 .value = value,
2640                 .value_len = value_len,
2641         };
2642
2643         struct ocfs2_xattr_search xis = {
2644                 .not_found = -ENODATA,
2645         };
2646
2647         struct ocfs2_xattr_search xbs = {
2648                 .not_found = -ENODATA,
2649         };
2650
2651         struct ocfs2_xattr_set_ctxt ctxt = {
2652                 .handle = handle,
2653                 .meta_ac = meta_ac,
2654                 .data_ac = data_ac,
2655         };
2656
2657         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2658                 return -EOPNOTSUPP;
2659
2660         /*
2661          * In extreme situation, may need xattr bucket when
2662          * block size is too small. And we have already reserved
2663          * the credits for bucket in mknod.
2664          */
2665         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2666                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2667                 if (!xbs.bucket) {
2668                         mlog_errno(-ENOMEM);
2669                         return -ENOMEM;
2670                 }
2671         }
2672
2673         xis.inode_bh = xbs.inode_bh = di_bh;
2674         di = (struct ocfs2_dinode *)di_bh->b_data;
2675
2676         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2677
2678         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2679         if (ret)
2680                 goto cleanup;
2681         if (xis.not_found) {
2682                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2683                 if (ret)
2684                         goto cleanup;
2685         }
2686
2687         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2688
2689 cleanup:
2690         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2691         brelse(xbs.xattr_bh);
2692         ocfs2_xattr_bucket_free(xbs.bucket);
2693
2694         return ret;
2695 }
2696
2697 /*
2698  * ocfs2_xattr_set()
2699  *
2700  * Set, replace or remove an extended attribute for this inode.
2701  * value is NULL to remove an existing extended attribute, else either
2702  * create or replace an extended attribute.
2703  */
2704 int ocfs2_xattr_set(struct inode *inode,
2705                     int name_index,
2706                     const char *name,
2707                     const void *value,
2708                     size_t value_len,
2709                     int flags)
2710 {
2711         struct buffer_head *di_bh = NULL;
2712         struct ocfs2_dinode *di;
2713         int ret, credits;
2714         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2715         struct inode *tl_inode = osb->osb_tl_inode;
2716         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2717
2718         struct ocfs2_xattr_info xi = {
2719                 .name_index = name_index,
2720                 .name = name,
2721                 .value = value,
2722                 .value_len = value_len,
2723         };
2724
2725         struct ocfs2_xattr_search xis = {
2726                 .not_found = -ENODATA,
2727         };
2728
2729         struct ocfs2_xattr_search xbs = {
2730                 .not_found = -ENODATA,
2731         };
2732
2733         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2734                 return -EOPNOTSUPP;
2735
2736         /*
2737          * Only xbs will be used on indexed trees.  xis doesn't need a
2738          * bucket.
2739          */
2740         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2741         if (!xbs.bucket) {
2742                 mlog_errno(-ENOMEM);
2743                 return -ENOMEM;
2744         }
2745
2746         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2747         if (ret < 0) {
2748                 mlog_errno(ret);
2749                 goto cleanup_nolock;
2750         }
2751         xis.inode_bh = xbs.inode_bh = di_bh;
2752         di = (struct ocfs2_dinode *)di_bh->b_data;
2753
2754         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2755         /*
2756          * Scan inode and external block to find the same name
2757          * extended attribute and collect search infomation.
2758          */
2759         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2760         if (ret)
2761                 goto cleanup;
2762         if (xis.not_found) {
2763                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2764                 if (ret)
2765                         goto cleanup;
2766         }
2767
2768         if (xis.not_found && xbs.not_found) {
2769                 ret = -ENODATA;
2770                 if (flags & XATTR_REPLACE)
2771                         goto cleanup;
2772                 ret = 0;
2773                 if (!value)
2774                         goto cleanup;
2775         } else {
2776                 ret = -EEXIST;
2777                 if (flags & XATTR_CREATE)
2778                         goto cleanup;
2779         }
2780
2781
2782         mutex_lock(&tl_inode->i_mutex);
2783
2784         if (ocfs2_truncate_log_needs_flush(osb)) {
2785                 ret = __ocfs2_flush_truncate_log(osb);
2786                 if (ret < 0) {
2787                         mutex_unlock(&tl_inode->i_mutex);
2788                         mlog_errno(ret);
2789                         goto cleanup;
2790                 }
2791         }
2792         mutex_unlock(&tl_inode->i_mutex);
2793
2794         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2795                                         &xbs, &ctxt, &credits);
2796         if (ret) {
2797                 mlog_errno(ret);
2798                 goto cleanup;
2799         }
2800
2801         /* we need to update inode's ctime field, so add credit for it. */
2802         credits += OCFS2_INODE_UPDATE_CREDITS;
2803         ctxt.handle = ocfs2_start_trans(osb, credits);
2804         if (IS_ERR(ctxt.handle)) {
2805                 ret = PTR_ERR(ctxt.handle);
2806                 mlog_errno(ret);
2807                 goto cleanup;
2808         }
2809
2810         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2811
2812         ocfs2_commit_trans(osb, ctxt.handle);
2813
2814         if (ctxt.data_ac)
2815                 ocfs2_free_alloc_context(ctxt.data_ac);
2816         if (ctxt.meta_ac)
2817                 ocfs2_free_alloc_context(ctxt.meta_ac);
2818         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2819                 ocfs2_schedule_truncate_log_flush(osb, 1);
2820         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2821 cleanup:
2822         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2823         ocfs2_inode_unlock(inode, 1);
2824 cleanup_nolock:
2825         brelse(di_bh);
2826         brelse(xbs.xattr_bh);
2827         ocfs2_xattr_bucket_free(xbs.bucket);
2828
2829         return ret;
2830 }
2831
2832 /*
2833  * Find the xattr extent rec which may contains name_hash.
2834  * e_cpos will be the first name hash of the xattr rec.
2835  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2836  */
2837 static int ocfs2_xattr_get_rec(struct inode *inode,
2838                                u32 name_hash,
2839                                u64 *p_blkno,
2840                                u32 *e_cpos,
2841                                u32 *num_clusters,
2842                                struct ocfs2_extent_list *el)
2843 {
2844         int ret = 0, i;
2845         struct buffer_head *eb_bh = NULL;
2846         struct ocfs2_extent_block *eb;
2847         struct ocfs2_extent_rec *rec = NULL;
2848         u64 e_blkno = 0;
2849
2850         if (el->l_tree_depth) {
2851                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2852                 if (ret) {
2853                         mlog_errno(ret);
2854                         goto out;
2855                 }
2856
2857                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2858                 el = &eb->h_list;
2859
2860                 if (el->l_tree_depth) {
2861                         ocfs2_error(inode->i_sb,
2862                                     "Inode %lu has non zero tree depth in "
2863                                     "xattr tree block %llu\n", inode->i_ino,
2864                                     (unsigned long long)eb_bh->b_blocknr);
2865                         ret = -EROFS;
2866                         goto out;
2867                 }
2868         }
2869
2870         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2871                 rec = &el->l_recs[i];
2872
2873                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2874                         e_blkno = le64_to_cpu(rec->e_blkno);
2875                         break;
2876                 }
2877         }
2878
2879         if (!e_blkno) {
2880                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2881                             "record (%u, %u, 0) in xattr", inode->i_ino,
2882                             le32_to_cpu(rec->e_cpos),
2883                             ocfs2_rec_clusters(el, rec));
2884                 ret = -EROFS;
2885                 goto out;
2886         }
2887
2888         *p_blkno = le64_to_cpu(rec->e_blkno);
2889         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2890         if (e_cpos)
2891                 *e_cpos = le32_to_cpu(rec->e_cpos);
2892 out:
2893         brelse(eb_bh);
2894         return ret;
2895 }
2896
2897 typedef int (xattr_bucket_func)(struct inode *inode,
2898                                 struct ocfs2_xattr_bucket *bucket,
2899                                 void *para);
2900
2901 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2902                                    struct ocfs2_xattr_bucket *bucket,
2903                                    int name_index,
2904                                    const char *name,
2905                                    u32 name_hash,
2906                                    u16 *xe_index,
2907                                    int *found)
2908 {
2909         int i, ret = 0, cmp = 1, block_off, new_offset;
2910         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2911         size_t name_len = strlen(name);
2912         struct ocfs2_xattr_entry *xe = NULL;
2913         char *xe_name;
2914
2915         /*
2916          * We don't use binary search in the bucket because there
2917          * may be multiple entries with the same name hash.
2918          */
2919         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2920                 xe = &xh->xh_entries[i];
2921
2922                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2923                         continue;
2924                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2925                         break;
2926
2927                 cmp = name_index - ocfs2_xattr_get_type(xe);
2928                 if (!cmp)
2929                         cmp = name_len - xe->xe_name_len;
2930                 if (cmp)
2931                         continue;
2932
2933                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2934                                                         xh,
2935                                                         i,
2936                                                         &block_off,
2937                                                         &new_offset);
2938                 if (ret) {
2939                         mlog_errno(ret);
2940                         break;
2941                 }
2942
2943
2944                 xe_name = bucket_block(bucket, block_off) + new_offset;
2945                 if (!memcmp(name, xe_name, name_len)) {
2946                         *xe_index = i;
2947                         *found = 1;
2948                         ret = 0;
2949                         break;
2950                 }
2951         }
2952
2953         return ret;
2954 }
2955
2956 /*
2957  * Find the specified xattr entry in a series of buckets.
2958  * This series start from p_blkno and last for num_clusters.
2959  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2960  * the num of the valid buckets.
2961  *
2962  * Return the buffer_head this xattr should reside in. And if the xattr's
2963  * hash is in the gap of 2 buckets, return the lower bucket.
2964  */
2965 static int ocfs2_xattr_bucket_find(struct inode *inode,
2966                                    int name_index,
2967                                    const char *name,
2968                                    u32 name_hash,
2969                                    u64 p_blkno,
2970                                    u32 first_hash,
2971                                    u32 num_clusters,
2972                                    struct ocfs2_xattr_search *xs)
2973 {
2974         int ret, found = 0;
2975         struct ocfs2_xattr_header *xh = NULL;
2976         struct ocfs2_xattr_entry *xe = NULL;
2977         u16 index = 0;
2978         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2979         int low_bucket = 0, bucket, high_bucket;
2980         struct ocfs2_xattr_bucket *search;
2981         u32 last_hash;
2982         u64 blkno, lower_blkno = 0;
2983
2984         search = ocfs2_xattr_bucket_new(inode);
2985         if (!search) {
2986                 ret = -ENOMEM;
2987                 mlog_errno(ret);
2988                 goto out;
2989         }
2990
2991         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2992         if (ret) {
2993                 mlog_errno(ret);
2994                 goto out;
2995         }
2996
2997         xh = bucket_xh(search);
2998         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2999         while (low_bucket <= high_bucket) {
3000                 ocfs2_xattr_bucket_relse(search);
3001
3002                 bucket = (low_bucket + high_bucket) / 2;
3003                 blkno = p_blkno + bucket * blk_per_bucket;
3004                 ret = ocfs2_read_xattr_bucket(search, blkno);
3005                 if (ret) {
3006                         mlog_errno(ret);
3007                         goto out;
3008                 }
3009
3010                 xh = bucket_xh(search);
3011                 xe = &xh->xh_entries[0];
3012                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3013                         high_bucket = bucket - 1;
3014                         continue;
3015                 }
3016
3017                 /*
3018                  * Check whether the hash of the last entry in our
3019                  * bucket is larger than the search one. for an empty
3020                  * bucket, the last one is also the first one.
3021                  */
3022                 if (xh->xh_count)
3023                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3024
3025                 last_hash = le32_to_cpu(xe->xe_name_hash);
3026
3027                 /* record lower_blkno which may be the insert place. */
3028                 lower_blkno = blkno;
3029
3030                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3031                         low_bucket = bucket + 1;
3032                         continue;
3033                 }
3034
3035                 /* the searched xattr should reside in this bucket if exists. */
3036                 ret = ocfs2_find_xe_in_bucket(inode, search,
3037                                               name_index, name, name_hash,
3038                                               &index, &found);
3039                 if (ret) {
3040                         mlog_errno(ret);
3041                         goto out;
3042                 }
3043                 break;
3044         }
3045
3046         /*
3047          * Record the bucket we have found.
3048          * When the xattr's hash value is in the gap of 2 buckets, we will
3049          * always set it to the previous bucket.
3050          */
3051         if (!lower_blkno)
3052                 lower_blkno = p_blkno;
3053
3054         /* This should be in cache - we just read it during the search */
3055         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3056         if (ret) {
3057                 mlog_errno(ret);
3058                 goto out;
3059         }
3060
3061         xs->header = bucket_xh(xs->bucket);
3062         xs->base = bucket_block(xs->bucket, 0);
3063         xs->end = xs->base + inode->i_sb->s_blocksize;
3064
3065         if (found) {
3066                 xs->here = &xs->header->xh_entries[index];
3067                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3068                      (unsigned long long)bucket_blkno(xs->bucket), index);
3069         } else
3070                 ret = -ENODATA;
3071
3072 out:
3073         ocfs2_xattr_bucket_free(search);
3074         return ret;
3075 }
3076
3077 static int ocfs2_xattr_index_block_find(struct inode *inode,
3078                                         struct buffer_head *root_bh,
3079                                         int name_index,
3080                                         const char *name,
3081                                         struct ocfs2_xattr_search *xs)
3082 {
3083         int ret;
3084         struct ocfs2_xattr_block *xb =
3085                         (struct ocfs2_xattr_block *)root_bh->b_data;
3086         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3087         struct ocfs2_extent_list *el = &xb_root->xt_list;
3088         u64 p_blkno = 0;
3089         u32 first_hash, num_clusters = 0;
3090         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3091
3092         if (le16_to_cpu(el->l_next_free_rec) == 0)
3093                 return -ENODATA;
3094
3095         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3096              name, name_hash, name_index);
3097
3098         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3099                                   &num_clusters, el);
3100         if (ret) {
3101                 mlog_errno(ret);
3102                 goto out;
3103         }
3104
3105         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3106
3107         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3108              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3109              first_hash);
3110
3111         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3112                                       p_blkno, first_hash, num_clusters, xs);
3113
3114 out:
3115         return ret;
3116 }
3117
3118 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3119                                        u64 blkno,
3120                                        u32 clusters,
3121                                        xattr_bucket_func *func,
3122                                        void *para)
3123 {
3124         int i, ret = 0;
3125         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3126         u32 num_buckets = clusters * bpc;
3127         struct ocfs2_xattr_bucket *bucket;
3128
3129         bucket = ocfs2_xattr_bucket_new(inode);
3130         if (!bucket) {
3131                 mlog_errno(-ENOMEM);
3132                 return -ENOMEM;
3133         }
3134
3135         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3136              clusters, (unsigned long long)blkno);
3137
3138         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3139                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3140                 if (ret) {
3141                         mlog_errno(ret);
3142                         break;
3143                 }
3144
3145                 /*
3146                  * The real bucket num in this series of blocks is stored
3147                  * in the 1st bucket.
3148                  */
3149                 if (i == 0)
3150                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3151
3152                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3153                      (unsigned long long)blkno,
3154                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3155                 if (func) {
3156                         ret = func(inode, bucket, para);
3157                         if (ret)
3158                                 mlog_errno(ret);
3159                         /* Fall through to bucket_relse() */
3160                 }
3161
3162                 ocfs2_xattr_bucket_relse(bucket);
3163                 if (ret)
3164                         break;
3165         }
3166
3167         ocfs2_xattr_bucket_free(bucket);
3168         return ret;
3169 }
3170
3171 struct ocfs2_xattr_tree_list {
3172         char *buffer;
3173         size_t buffer_size;
3174         size_t result;
3175 };
3176
3177 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3178                                              struct ocfs2_xattr_header *xh,
3179                                              int index,
3180                                              int *block_off,
3181                                              int *new_offset)
3182 {
3183         u16 name_offset;
3184
3185         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3186                 return -EINVAL;
3187
3188         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3189
3190         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3191         *new_offset = name_offset % inode->i_sb->s_blocksize;
3192
3193         return 0;
3194 }
3195
3196 static int ocfs2_list_xattr_bucket(struct inode *inode,
3197                                    struct ocfs2_xattr_bucket *bucket,
3198                                    void *para)
3199 {
3200         int ret = 0, type;
3201         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3202         int i, block_off, new_offset;
3203         const char *prefix, *name;
3204
3205         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3206                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3207                 type = ocfs2_xattr_get_type(entry);
3208                 prefix = ocfs2_xattr_prefix(type);
3209
3210                 if (prefix) {
3211                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3212                                                                 bucket_xh(bucket),
3213                                                                 i,
3214                                                                 &block_off,
3215                                                                 &new_offset);
3216                         if (ret)
3217                                 break;
3218
3219                         name = (const char *)bucket_block(bucket, block_off) +
3220                                 new_offset;
3221                         ret = ocfs2_xattr_list_entry(xl->buffer,
3222                                                      xl->buffer_size,
3223                                                      &xl->result,
3224                                                      prefix, name,
3225                                                      entry->xe_name_len);
3226                         if (ret)
3227                                 break;
3228                 }
3229         }
3230
3231         return ret;
3232 }
3233
3234 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3235                                              struct ocfs2_xattr_tree_root *xt,
3236                                              char *buffer,
3237                                              size_t buffer_size)
3238 {
3239         struct ocfs2_extent_list *el = &xt->xt_list;
3240         int ret = 0;
3241         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3242         u64 p_blkno = 0;
3243         struct ocfs2_xattr_tree_list xl = {
3244                 .buffer = buffer,
3245                 .buffer_size = buffer_size,
3246                 .result = 0,
3247         };
3248
3249         if (le16_to_cpu(el->l_next_free_rec) == 0)
3250                 return 0;
3251
3252         while (name_hash > 0) {
3253                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3254                                           &e_cpos, &num_clusters, el);
3255                 if (ret) {
3256                         mlog_errno(ret);
3257                         goto out;
3258                 }
3259
3260                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3261                                                   ocfs2_list_xattr_bucket,
3262                                                   &xl);
3263                 if (ret) {
3264                         mlog_errno(ret);
3265                         goto out;
3266                 }
3267
3268                 if (e_cpos == 0)
3269                         break;
3270
3271                 name_hash = e_cpos - 1;
3272         }
3273
3274         ret = xl.result;
3275 out:
3276         return ret;
3277 }
3278
3279 static int cmp_xe(const void *a, const void *b)
3280 {
3281         const struct ocfs2_xattr_entry *l = a, *r = b;
3282         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3283         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3284
3285         if (l_hash > r_hash)
3286                 return 1;
3287         if (l_hash < r_hash)
3288                 return -1;
3289         return 0;
3290 }
3291
3292 static void swap_xe(void *a, void *b, int size)
3293 {
3294         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3295
3296         tmp = *l;
3297         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3298         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3299 }
3300
3301 /*
3302  * When the ocfs2_xattr_block is filled up, new bucket will be created
3303  * and all the xattr entries will be moved to the new bucket.
3304  * The header goes at the start of the bucket, and the names+values are
3305  * filled from the end.  This is why *target starts as the last buffer.
3306  * Note: we need to sort the entries since they are not saved in order
3307  * in the ocfs2_xattr_block.
3308  */
3309 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3310                                            struct buffer_head *xb_bh,
3311                                            struct ocfs2_xattr_bucket *bucket)
3312 {
3313         int i, blocksize = inode->i_sb->s_blocksize;
3314         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3315         u16 offset, size, off_change;
3316         struct ocfs2_xattr_entry *xe;
3317         struct ocfs2_xattr_block *xb =
3318                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3319         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3320         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3321         u16 count = le16_to_cpu(xb_xh->xh_count);
3322         char *src = xb_bh->b_data;
3323         char *target = bucket_block(bucket, blks - 1);
3324
3325         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3326              (unsigned long long)xb_bh->b_blocknr,
3327              (unsigned long long)bucket_blkno(bucket));
3328
3329         for (i = 0; i < blks; i++)
3330                 memset(bucket_block(bucket, i), 0, blocksize);
3331
3332         /*
3333          * Since the xe_name_offset is based on ocfs2_xattr_header,
3334          * there is a offset change corresponding to the change of
3335          * ocfs2_xattr_header's position.
3336          */
3337         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3338         xe = &xb_xh->xh_entries[count - 1];
3339         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3340         size = blocksize - offset;
3341
3342         /* copy all the names and values. */
3343         memcpy(target + offset, src + offset, size);
3344
3345         /* Init new header now. */
3346         xh->xh_count = xb_xh->xh_count;
3347         xh->xh_num_buckets = cpu_to_le16(1);
3348         xh->xh_name_value_len = cpu_to_le16(size);
3349         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3350
3351         /* copy all the entries. */
3352         target = bucket_block(bucket, 0);
3353         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3354         size = count * sizeof(struct ocfs2_xattr_entry);
3355         memcpy(target + offset, (char *)xb_xh + offset, size);
3356
3357         /* Change the xe offset for all the xe because of the move. */
3358         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3359                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3360         for (i = 0; i < count; i++)
3361                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3362
3363         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3364              offset, size, off_change);
3365
3366         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3367              cmp_xe, swap_xe);
3368 }
3369
3370 /*
3371  * After we move xattr from block to index btree, we have to
3372  * update ocfs2_xattr_search to the new xe and base.
3373  *
3374  * When the entry is in xattr block, xattr_bh indicates the storage place.
3375  * While if the entry is in index b-tree, "bucket" indicates the
3376  * real place of the xattr.
3377  */
3378 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3379                                             struct ocfs2_xattr_search *xs,
3380                                             struct buffer_head *old_bh)
3381 {
3382         char *buf = old_bh->b_data;
3383         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3384         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3385         int i;
3386
3387         xs->header = bucket_xh(xs->bucket);
3388         xs->base = bucket_block(xs->bucket, 0);
3389         xs->end = xs->base + inode->i_sb->s_blocksize;
3390
3391         if (xs->not_found)
3392                 return;
3393
3394         i = xs->here - old_xh->xh_entries;
3395         xs->here = &xs->header->xh_entries[i];
3396 }
3397
3398 static int ocfs2_xattr_create_index_block(struct inode *inode,
3399                                           struct ocfs2_xattr_search *xs,
3400                                           struct ocfs2_xattr_set_ctxt *ctxt)
3401 {
3402         int ret;
3403         u32 bit_off, len;
3404         u64 blkno;
3405         handle_t *handle = ctxt->handle;
3406         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3407         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3408         struct buffer_head *xb_bh = xs->xattr_bh;
3409         struct ocfs2_xattr_block *xb =
3410                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3411         struct ocfs2_xattr_tree_root *xr;
3412         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3413
3414         mlog(0, "create xattr index block for %llu\n",
3415              (unsigned long long)xb_bh->b_blocknr);
3416
3417         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3418         BUG_ON(!xs->bucket);
3419
3420         /*
3421          * XXX:
3422          * We can use this lock for now, and maybe move to a dedicated mutex
3423          * if performance becomes a problem later.
3424          */
3425         down_write(&oi->ip_alloc_sem);
3426
3427         ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3428                                       OCFS2_JOURNAL_ACCESS_WRITE);
3429         if (ret) {
3430                 mlog_errno(ret);
3431                 goto out;
3432         }
3433
3434         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3435                                      1, 1, &bit_off, &len);
3436         if (ret) {
3437                 mlog_errno(ret);
3438                 goto out;
3439         }
3440
3441         /*
3442          * The bucket may spread in many blocks, and
3443          * we will only touch the 1st block and the last block
3444          * in the whole bucket(one for entry and one for data).
3445          */
3446         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3447
3448         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3449              (unsigned long long)blkno);
3450
3451         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3452         if (ret) {
3453                 mlog_errno(ret);
3454                 goto out;
3455         }
3456
3457         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3458                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3459         if (ret) {
3460                 mlog_errno(ret);
3461                 goto out;
3462         }
3463
3464         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3465         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3466
3467         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3468
3469         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3470         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3471                offsetof(struct ocfs2_xattr_block, xb_attrs));
3472
3473         xr = &xb->xb_attrs.xb_root;
3474         xr->xt_clusters = cpu_to_le32(1);
3475         xr->xt_last_eb_blk = 0;
3476         xr->xt_list.l_tree_depth = 0;
3477         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3478         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3479
3480         xr->xt_list.l_recs[0].e_cpos = 0;
3481         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3482         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3483
3484         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3485
3486         ocfs2_journal_dirty(handle, xb_bh);
3487
3488 out:
3489         up_write(&oi->ip_alloc_sem);
3490
3491         return ret;
3492 }
3493
3494 static int cmp_xe_offset(const void *a, const void *b)
3495 {
3496         const struct ocfs2_xattr_entry *l = a, *r = b;
3497         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3498         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3499
3500         if (l_name_offset < r_name_offset)
3501                 return 1;
3502         if (l_name_offset > r_name_offset)
3503                 return -1;
3504         return 0;
3505 }
3506
3507 /*
3508  * defrag a xattr bucket if we find that the bucket has some
3509  * holes beteen name/value pairs.
3510  * We will move all the name/value pairs to the end of the bucket
3511  * so that we can spare some space for insertion.
3512  */
3513 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3514                                      handle_t *handle,
3515                                      struct ocfs2_xattr_bucket *bucket)
3516 {
3517         int ret, i;
3518         size_t end, offset, len, value_len;
3519         struct ocfs2_xattr_header *xh;
3520         char *entries, *buf, *bucket_buf = NULL;
3521         u64 blkno = bucket_blkno(bucket);
3522         u16 xh_free_start;
3523         size_t blocksize = inode->i_sb->s_blocksize;
3524         struct ocfs2_xattr_entry *xe;
3525
3526         /*
3527          * In order to make the operation more efficient and generic,
3528          * we copy all the blocks into a contiguous memory and do the
3529          * defragment there, so if anything is error, we will not touch
3530          * the real block.
3531          */
3532         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3533         if (!bucket_buf) {
3534                 ret = -EIO;
3535                 goto out;
3536         }
3537
3538         buf = bucket_buf;
3539         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3540                 memcpy(buf, bucket_block(bucket, i), blocksize);
3541
3542         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3543                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3544         if (ret < 0) {
3545                 mlog_errno(ret);
3546                 goto out;
3547         }
3548
3549         xh = (struct ocfs2_xattr_header *)bucket_buf;
3550         entries = (char *)xh->xh_entries;
3551         xh_free_start = le16_to_cpu(xh->xh_free_start);
3552
3553         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3554              "xh_free_start = %u, xh_name_value_len = %u.\n",
3555              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3556              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3557
3558         /*
3559          * sort all the entries by their offset.
3560          * the largest will be the first, so that we can
3561          * move them to the end one by one.
3562          */
3563         sort(entries, le16_to_cpu(xh->xh_count),
3564              sizeof(struct ocfs2_xattr_entry),
3565              cmp_xe_offset, swap_xe);
3566
3567         /* Move all name/values to the end of the bucket. */
3568         xe = xh->xh_entries;
3569         end = OCFS2_XATTR_BUCKET_SIZE;
3570         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3571                 offset = le16_to_cpu(xe->xe_name_offset);
3572                 if (ocfs2_xattr_is_local(xe))
3573                         value_len = OCFS2_XATTR_SIZE(
3574                                         le64_to_cpu(xe->xe_value_size));
3575                 else
3576                         value_len = OCFS2_XATTR_ROOT_SIZE;
3577                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3578
3579                 /*
3580                  * We must make sure that the name/value pair
3581                  * exist in the same block. So adjust end to
3582                  * the previous block end if needed.
3583                  */
3584                 if (((end - len) / blocksize !=
3585                         (end - 1) / blocksize))
3586                         end = end - end % blocksize;
3587
3588                 if (end > offset + len) {
3589                         memmove(bucket_buf + end - len,
3590                                 bucket_buf + offset, len);
3591                         xe->xe_name_offset = cpu_to_le16(end - len);
3592                 }
3593
3594                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3595                                 "bucket %llu\n", (unsigned long long)blkno);
3596
3597                 end -= len;
3598         }
3599
3600         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3601                         "bucket %llu\n", (unsigned long long)blkno);
3602
3603         if (xh_free_start == end)
3604                 goto out;
3605
3606         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3607         xh->xh_free_start = cpu_to_le16(end);
3608
3609         /* sort the entries by their name_hash. */
3610         sort(entries, le16_to_cpu(xh->xh_count),
3611              sizeof(struct ocfs2_xattr_entry),
3612              cmp_xe, swap_xe);
3613
3614         buf = bucket_buf;
3615         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3616                 memcpy(bucket_block(bucket, i), buf, blocksize);
3617         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3618
3619 out:
3620         kfree(bucket_buf);
3621         return ret;
3622 }
3623
3624 /*
3625  * prev_blkno points to the start of an existing extent.  new_blkno
3626  * points to a newly allocated extent.  Because we know each of our
3627  * clusters contains more than bucket, we can easily split one cluster
3628  * at a bucket boundary.  So we take the last cluster of the existing
3629  * extent and split it down the middle.  We move the last half of the
3630  * buckets in the last cluster of the existing extent over to the new
3631  * extent.
3632  *
3633  * first_bh is the buffer at prev_blkno so we can update the existing
3634  * extent's bucket count.  header_bh is the bucket were we were hoping
3635  * to insert our xattr.  If the bucket move places the target in the new
3636  * extent, we'll update first_bh and header_bh after modifying the old
3637  * extent.
3638  *
3639  * first_hash will be set as the 1st xe's name_hash in the new extent.
3640  */
3641 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3642                                                handle_t *handle,
3643                                                struct ocfs2_xattr_bucket *first,
3644                                                struct ocfs2_xattr_bucket *target,
3645                                                u64 new_blkno,
3646                                                u32 num_clusters,
3647                                                u32 *first_hash)
3648 {
3649         int ret;
3650         struct super_block *sb = inode->i_sb;
3651         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3652         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3653         int to_move = num_buckets / 2;
3654         u64 src_blkno;
3655         u64 last_cluster_blkno = bucket_blkno(first) +
3656                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3657
3658         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3659         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3660
3661         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3662              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3663
3664         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3665                                      last_cluster_blkno, new_blkno,
3666                                      to_move, first_hash);
3667         if (ret) {
3668                 mlog_errno(ret);
3669                 goto out;
3670         }
3671
3672         /* This is the first bucket that got moved */
3673         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3674
3675         /*
3676          * If the target bucket was part of the moved buckets, we need to
3677          * update first and target.
3678          */
3679         if (bucket_blkno(target) >= src_blkno) {
3680                 /* Find the block for the new target bucket */
3681                 src_blkno = new_blkno +
3682                         (bucket_blkno(target) - src_blkno);
3683
3684                 ocfs2_xattr_bucket_relse(first);
3685                 ocfs2_xattr_bucket_relse(target);
3686
3687                 /*
3688                  * These shouldn't fail - the buffers are in the
3689                  * journal from ocfs2_cp_xattr_bucket().
3690                  */
3691                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3692                 if (ret) {
3693                         mlog_errno(ret);
3694                         goto out;
3695                 }
3696                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3697                 if (ret)
3698                         mlog_errno(ret);
3699
3700         }
3701
3702 out:
3703         return ret;
3704 }
3705
3706 /*
3707  * Find the suitable pos when we divide a bucket into 2.
3708  * We have to make sure the xattrs with the same hash value exist
3709  * in the same bucket.
3710  *
3711  * If this ocfs2_xattr_header covers more than one hash value, find a
3712  * place where the hash value changes.  Try to find the most even split.
3713  * The most common case is that all entries have different hash values,
3714  * and the first check we make will find a place to split.
3715  */
3716 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3717 {
3718         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3719         int count = le16_to_cpu(xh->xh_count);
3720         int delta, middle = count / 2;
3721
3722         /*
3723          * We start at the middle.  Each step gets farther away in both
3724          * directions.  We therefore hit the change in hash value
3725          * nearest to the middle.  Note that this loop does not execute for
3726          * count < 2.
3727          */
3728         for (delta = 0; delta < middle; delta++) {
3729                 /* Let's check delta earlier than middle */
3730                 if (cmp_xe(&entries[middle - delta - 1],
3731                            &entries[middle - delta]))
3732                         return middle - delta;
3733
3734                 /* For even counts, don't walk off the end */
3735                 if ((middle + delta + 1) == count)
3736                         continue;
3737
3738                 /* Now try delta past middle */
3739                 if (cmp_xe(&entries[middle + delta],
3740                            &entries[middle + delta + 1]))
3741                         return middle + delta + 1;
3742         }
3743
3744         /* Every entry had the same hash */
3745         return count;
3746 }
3747
3748 /*
3749  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3750  * first_hash will record the 1st hash of the new bucket.
3751  *
3752  * Normally half of the xattrs will be moved.  But we have to make
3753  * sure that the xattrs with the same hash value are stored in the
3754  * same bucket. If all the xattrs in this bucket have the same hash
3755  * value, the new bucket will be initialized as an empty one and the
3756  * first_hash will be initialized as (hash_value+1).
3757  */
3758 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3759                                     handle_t *handle,
3760                                     u64 blk,
3761                                     u64 new_blk,
3762                                     u32 *first_hash,
3763                                     int new_bucket_head)
3764 {
3765         int ret, i;
3766         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3767         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3768         struct ocfs2_xattr_header *xh;
3769         struct ocfs2_xattr_entry *xe;
3770         int blocksize = inode->i_sb->s_blocksize;
3771
3772         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3773              (unsigned long long)blk, (unsigned long long)new_blk);
3774
3775         s_bucket = ocfs2_xattr_bucket_new(inode);
3776         t_bucket = ocfs2_xattr_bucket_new(inode);
3777         if (!s_bucket || !t_bucket) {
3778                 ret = -ENOMEM;
3779                 mlog_errno(ret);
3780                 goto out;
3781         }
3782
3783         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3784         if (ret) {
3785                 mlog_errno(ret);
3786                 goto out;
3787         }
3788
3789         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3790                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3791         if (ret) {
3792                 mlog_errno(ret);
3793                 goto out;
3794         }
3795
3796         /*
3797          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3798          * there's no need to read it.
3799          */
3800         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3801         if (ret) {
3802                 mlog_errno(ret);
3803                 goto out;
3804         }
3805
3806         /*
3807          * Hey, if we're overwriting t_bucket, what difference does
3808          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3809          * same part of ocfs2_cp_xattr_bucket().
3810          */
3811         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3812                                                 new_bucket_head ?
3813                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3814                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3815         if (ret) {
3816                 mlog_errno(ret);
3817                 goto out;
3818         }
3819
3820         xh = bucket_xh(s_bucket);
3821         count = le16_to_cpu(xh->xh_count);
3822         start = ocfs2_xattr_find_divide_pos(xh);
3823
3824         if (start == count) {
3825                 xe = &xh->xh_entries[start-1];
3826
3827                 /*
3828                  * initialized a new empty bucket here.
3829                  * The hash value is set as one larger than
3830                  * that of the last entry in the previous bucket.
3831                  */
3832                 for (i = 0; i < t_bucket->bu_blocks; i++)
3833                         memset(bucket_block(t_bucket, i), 0, blocksize);
3834
3835                 xh = bucket_xh(t_bucket);
3836                 xh->xh_free_start = cpu_to_le16(blocksize);
3837                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3838                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3839
3840                 goto set_num_buckets;
3841         }
3842
3843         /* copy the whole bucket to the new first. */
3844         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3845
3846         /* update the new bucket. */
3847         xh = bucket_xh(t_bucket);
3848
3849         /*
3850          * Calculate the total name/value len and xh_free_start for
3851          * the old bucket first.
3852          */
3853         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3854         name_value_len = 0;
3855         for (i = 0; i < start; i++) {
3856                 xe = &xh->xh_entries[i];
3857                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3858                 if (ocfs2_xattr_is_local(xe))
3859                         xe_len +=
3860                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3861                 else
3862                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3863                 name_value_len += xe_len;
3864                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3865                         name_offset = le16_to_cpu(xe->xe_name_offset);
3866         }
3867
3868         /*
3869          * Now begin the modification to the new bucket.
3870          *
3871          * In the new bucket, We just move the xattr entry to the beginning
3872          * and don't touch the name/value. So there will be some holes in the
3873          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3874          * called.
3875          */
3876         xe = &xh->xh_entries[start];
3877         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3878         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3879              (int)((char *)xe - (char *)xh),
3880              (int)((char *)xh->xh_entries - (char *)xh));
3881         memmove((char *)xh->xh_entries, (char *)xe, len);
3882         xe = &xh->xh_entries[count - start];
3883         len = sizeof(struct ocfs2_xattr_entry) * start;
3884         memset((char *)xe, 0, len);
3885
3886         le16_add_cpu(&xh->xh_count, -start);
3887         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3888
3889         /* Calculate xh_free_start for the new bucket. */
3890         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3891         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3892                 xe = &xh->xh_entries[i];
3893                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3894                 if (ocfs2_xattr_is_local(xe))
3895                         xe_len +=
3896                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3897                 else
3898                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3899                 if (le16_to_cpu(xe->xe_name_offset) <
3900                     le16_to_cpu(xh->xh_free_start))
3901                         xh->xh_free_start = xe->xe_name_offset;
3902         }
3903
3904 set_num_buckets:
3905         /* set xh->xh_num_buckets for the new xh. */
3906         if (new_bucket_head)
3907                 xh->xh_num_buckets = cpu_to_le16(1);
3908         else
3909                 xh->xh_num_buckets = 0;
3910
3911         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3912
3913         /* store the first_hash of the new bucket. */
3914         if (first_hash)
3915                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3916
3917         /*
3918          * Now only update the 1st block of the old bucket.  If we
3919          * just added a new empty bucket, there is no need to modify
3920          * it.
3921          */
3922         if (start == count)
3923                 goto out;
3924
3925         xh = bucket_xh(s_bucket);
3926         memset(&xh->xh_entries[start], 0,
3927                sizeof(struct ocfs2_xattr_entry) * (count - start));
3928         xh->xh_count = cpu_to_le16(start);
3929         xh->xh_free_start = cpu_to_le16(name_offset);
3930         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3931
3932         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3933
3934 out:
3935         ocfs2_xattr_bucket_free(s_bucket);
3936         ocfs2_xattr_bucket_free(t_bucket);
3937
3938         return ret;
3939 }
3940
3941 /*
3942  * Copy xattr from one bucket to another bucket.
3943  *
3944  * The caller must make sure that the journal transaction
3945  * has enough space for journaling.
3946  */
3947 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3948                                  handle_t *handle,
3949                                  u64 s_blkno,
3950                                  u64 t_blkno,
3951                                  int t_is_new)
3952 {
3953         int ret;
3954         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3955
3956         BUG_ON(s_blkno == t_blkno);
3957
3958         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3959              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3960              t_is_new);
3961
3962         s_bucket = ocfs2_xattr_bucket_new(inode);
3963         t_bucket = ocfs2_xattr_bucket_new(inode);
3964         if (!s_bucket || !t_bucket) {
3965                 ret = -ENOMEM;
3966                 mlog_errno(ret);
3967                 goto out;
3968         }
3969
3970         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3971         if (ret)
3972                 goto out;
3973
3974         /*
3975          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3976          * there's no need to read it.
3977          */
3978         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3979         if (ret)
3980                 goto out;
3981
3982         /*
3983          * Hey, if we're overwriting t_bucket, what difference does
3984          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3985          * cluster to fill, we came here from
3986          * ocfs2_mv_xattr_buckets(), and it is really new -
3987          * ACCESS_CREATE is required.  But we also might have moved data
3988          * out of t_bucket before extending back into it.
3989          * ocfs2_add_new_xattr_bucket() can do this - its call to
3990          * ocfs2_add_new_xattr_cluster() may have created a new extent
3991          * and copied out the end of the old extent.  Then it re-extends
3992          * the old extent back to create space for new xattrs.  That's
3993          * how we get here, and the bucket isn't really new.
3994          */
3995         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3996                                                 t_is_new ?
3997                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3998                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3999         if (ret)
4000                 goto out;
4001
4002         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4003         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4004
4005 out:
4006         ocfs2_xattr_bucket_free(t_bucket);
4007         ocfs2_xattr_bucket_free(s_bucket);
4008
4009         return ret;
4010 }
4011
4012 /*
4013  * src_blk points to the start of an existing extent.  last_blk points to
4014  * last cluster in that extent.  to_blk points to a newly allocated
4015  * extent.  We copy the buckets from the cluster at last_blk to the new
4016  * extent.  If start_bucket is non-zero, we skip that many buckets before
4017  * we start copying.  The new extent's xh_num_buckets gets set to the
4018  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4019  * by the same amount.
4020  */
4021 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4022                                   u64 src_blk, u64 last_blk, u64 to_blk,
4023                                   unsigned int start_bucket,
4024                                   u32 *first_hash)
4025 {
4026         int i, ret, credits;
4027         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4028         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4029         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4030         struct ocfs2_xattr_bucket *old_first, *new_first;
4031
4032         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4033              (unsigned long long)last_blk, (unsigned long long)to_blk);
4034
4035         BUG_ON(start_bucket >= num_buckets);
4036         if (start_bucket) {
4037                 num_buckets -= start_bucket;
4038                 last_blk += (start_bucket * blks_per_bucket);
4039         }
4040
4041         /* The first bucket of the original extent */
4042         old_first = ocfs2_xattr_bucket_new(inode);
4043         /* The first bucket of the new extent */
4044         new_first = ocfs2_xattr_bucket_new(inode);
4045         if (!old_first || !new_first) {
4046                 ret = -ENOMEM;
4047                 mlog_errno(ret);
4048                 goto out;
4049         }
4050
4051         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4052         if (ret) {
4053                 mlog_errno(ret);
4054                 goto out;
4055         }
4056
4057         /*
4058          * We need to update the first bucket of the old extent and all
4059          * the buckets going to the new extent.
4060          */
4061         credits = ((num_buckets + 1) * blks_per_bucket) +
4062                 handle->h_buffer_credits;
4063         ret = ocfs2_extend_trans(handle, credits);
4064         if (ret) {
4065                 mlog_errno(ret);
4066                 goto out;
4067         }
4068
4069         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4070                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4071         if (ret) {
4072                 mlog_errno(ret);
4073                 goto out;
4074         }
4075
4076         for (i = 0; i < num_buckets; i++) {
4077                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4078                                             last_blk + (i * blks_per_bucket),
4079                                             to_blk + (i * blks_per_bucket),
4080                                             1);
4081                 if (ret) {
4082                         mlog_errno(ret);
4083                         goto out;
4084                 }
4085         }
4086
4087         /*
4088          * Get the new bucket ready before we dirty anything
4089          * (This actually shouldn't fail, because we already dirtied
4090          * it once in ocfs2_cp_xattr_bucket()).
4091          */
4092         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4093         if (ret) {
4094                 mlog_errno(ret);
4095                 goto out;
4096         }
4097         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4098                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4099         if (ret) {
4100                 mlog_errno(ret);
4101                 goto out;
4102         }
4103
4104         /* Now update the headers */
4105         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4106         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4107
4108         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4109         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4110
4111         if (first_hash)
4112                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4113
4114 out:
4115         ocfs2_xattr_bucket_free(new_first);
4116         ocfs2_xattr_bucket_free(old_first);
4117         return ret;
4118 }
4119
4120 /*
4121  * Move some xattrs in this cluster to the new cluster.
4122  * This function should only be called when bucket size == cluster size.
4123  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4124  */
4125 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4126                                       handle_t *handle,
4127                                       u64 prev_blk,
4128                                       u64 new_blk,
4129                                       u32 *first_hash)
4130 {
4131         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4132         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4133
4134         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4135
4136         ret = ocfs2_extend_trans(handle, credits);
4137         if (ret) {
4138                 mlog_errno(ret);
4139                 return ret;
4140         }
4141
4142         /* Move half of the xattr in start_blk to the next bucket. */
4143         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4144                                           new_blk, first_hash, 1);
4145 }
4146
4147 /*
4148  * Move some xattrs from the old cluster to the new one since they are not
4149  * contiguous in ocfs2 xattr tree.
4150  *
4151  * new_blk starts a new separate cluster, and we will move some xattrs from
4152  * prev_blk to it. v_start will be set as the first name hash value in this
4153  * new cluster so that it can be used as e_cpos during tree insertion and
4154  * don't collide with our original b-tree operations. first_bh and header_bh
4155  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4156  * to extend the insert bucket.
4157  *
4158  * The problem is how much xattr should we move to the new one and when should
4159  * we update first_bh and header_bh?
4160  * 1. If cluster size > bucket size, that means the previous cluster has more
4161  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4162  *    update the first_bh and header_bh if the insert bucket has been moved
4163  *    to the new cluster.
4164  * 2. If cluster_size == bucket_size:
4165  *    a) If the previous extent rec has more than one cluster and the insert
4166  *       place isn't in the last cluster, copy the entire last cluster to the
4167  *       new one. This time, we don't need to upate the first_bh and header_bh
4168  *       since they will not be moved into the new cluster.
4169  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4170  *       the new one. And we set the extend flag to zero if the insert place is
4171  *       moved into the new allocated cluster since no extend is needed.
4172  */
4173 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4174                                             handle_t *handle,
4175                                             struct ocfs2_xattr_bucket *first,
4176                                             struct ocfs2_xattr_bucket *target,
4177                                             u64 new_blk,
4178                                             u32 prev_clusters,
4179                                             u32 *v_start,
4180                                             int *extend)
4181 {
4182         int ret;
4183
4184         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4185              (unsigned long long)bucket_blkno(first), prev_clusters,
4186              (unsigned long long)new_blk);
4187
4188         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4189                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4190                                                           handle,
4191                                                           first, target,
4192                                                           new_blk,
4193                                                           prev_clusters,
4194                                                           v_start);
4195                 if (ret)
4196                         mlog_errno(ret);
4197         } else {
4198                 /* The start of the last cluster in the first extent */
4199                 u64 last_blk = bucket_blkno(first) +
4200                         ((prev_clusters - 1) *
4201                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4202
4203                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4204                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4205                                                      bucket_blkno(first),
4206                                                      last_blk, new_blk, 0,
4207                                                      v_start);
4208                         if (ret)
4209                                 mlog_errno(ret);
4210                 } else {
4211                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4212                                                          last_blk, new_blk,
4213                                                          v_start);
4214                         if (ret)
4215                                 mlog_errno(ret);
4216
4217                         if ((bucket_blkno(target) == last_blk) && extend)
4218                                 *extend = 0;
4219                 }
4220         }
4221
4222         return ret;
4223 }
4224
4225 /*
4226  * Add a new cluster for xattr storage.
4227  *
4228  * If the new cluster is contiguous with the previous one, it will be
4229  * appended to the same extent record, and num_clusters will be updated.
4230  * If not, we will insert a new extent for it and move some xattrs in
4231  * the last cluster into the new allocated one.
4232  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4233  * lose the benefits of hashing because we'll have to search large leaves.
4234  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4235  * if it's bigger).
4236  *
4237  * first_bh is the first block of the previous extent rec and header_bh
4238  * indicates the bucket we will insert the new xattrs. They will be updated
4239  * when the header_bh is moved into the new cluster.
4240  */
4241 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4242                                        struct buffer_head *root_bh,
4243                                        struct ocfs2_xattr_bucket *first,
4244                                        struct ocfs2_xattr_bucket *target,
4245                                        u32 *num_clusters,
4246                                        u32 prev_cpos,
4247                                        int *extend,
4248                                        struct ocfs2_xattr_set_ctxt *ctxt)
4249 {
4250         int ret;
4251         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4252         u32 prev_clusters = *num_clusters;
4253         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4254         u64 block;
4255         handle_t *handle = ctxt->handle;
4256         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4257         struct ocfs2_extent_tree et;
4258
4259         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4260              "previous xattr blkno = %llu\n",
4261              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4262              prev_cpos, (unsigned long long)bucket_blkno(first));
4263
4264         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4265
4266         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4267                                       OCFS2_JOURNAL_ACCESS_WRITE);
4268         if (ret < 0) {
4269                 mlog_errno(ret);
4270                 goto leave;
4271         }
4272
4273         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4274                                      clusters_to_add, &bit_off, &num_bits);
4275         if (ret < 0) {
4276                 if (ret != -ENOSPC)
4277                         mlog_errno(ret);
4278                 goto leave;
4279         }
4280
4281         BUG_ON(num_bits > clusters_to_add);
4282
4283         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4284         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4285              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4286
4287         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4288             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4289              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4290                 /*
4291                  * If this cluster is contiguous with the old one and
4292                  * adding this new cluster, we don't surpass the limit of
4293                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4294                  * initialized and used like other buckets in the previous
4295                  * cluster.
4296                  * So add it as a contiguous one. The caller will handle
4297                  * its init process.
4298                  */
4299                 v_start = prev_cpos + prev_clusters;
4300                 *num_clusters = prev_clusters + num_bits;
4301                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4302                      num_bits);
4303         } else {
4304                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4305                                                        handle,
4306                                                        first,
4307                                                        target,
4308                                                        block,
4309                                                        prev_clusters,
4310                                                        &v_start,
4311                                                        extend);
4312                 if (ret) {
4313                         mlog_errno(ret);
4314                         goto leave;
4315                 }
4316         }
4317
4318         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4319              num_bits, (unsigned long long)block, v_start);
4320         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4321                                   num_bits, 0, ctxt->meta_ac);
4322         if (ret < 0) {
4323                 mlog_errno(ret);
4324                 goto leave;
4325         }
4326
4327         ret = ocfs2_journal_dirty(handle, root_bh);
4328         if (ret < 0)
4329                 mlog_errno(ret);
4330
4331 leave:
4332         return ret;
4333 }
4334
4335 /*
4336  * We are given an extent.  'first' is the bucket at the very front of
4337  * the extent.  The extent has space for an additional bucket past
4338  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4339  * of the target bucket.  We wish to shift every bucket past the target
4340  * down one, filling in that additional space.  When we get back to the
4341  * target, we split the target between itself and the now-empty bucket
4342  * at target+1 (aka, target_blkno + blks_per_bucket).
4343  */
4344 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4345                                      handle_t *handle,
4346                                      struct ocfs2_xattr_bucket *first,
4347                                      u64 target_blk,
4348                                      u32 num_clusters)
4349 {
4350         int ret, credits;
4351         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4352         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4353         u64 end_blk;
4354         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4355
4356         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4357              "from %llu, len = %u\n", (unsigned long long)target_blk,
4358              (unsigned long long)bucket_blkno(first), num_clusters);
4359
4360         /* The extent must have room for an additional bucket */
4361         BUG_ON(new_bucket >=
4362                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4363
4364         /* end_blk points to the last existing bucket */
4365         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4366
4367         /*
4368          * end_blk is the start of the last existing bucket.
4369          * Thus, (end_blk - target_blk) covers the target bucket and
4370          * every bucket after it up to, but not including, the last
4371          * existing bucket.  Then we add the last existing bucket, the
4372          * new bucket, and the first bucket (3 * blk_per_bucket).
4373          */
4374         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4375                   handle->h_buffer_credits;
4376         ret = ocfs2_extend_trans(handle, credits);
4377         if (ret) {
4378                 mlog_errno(ret);
4379                 goto out;
4380         }
4381
4382         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4383                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4384         if (ret) {
4385                 mlog_errno(ret);
4386                 goto out;
4387         }
4388
4389         while (end_blk != target_blk) {
4390                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4391                                             end_blk + blk_per_bucket, 0);
4392                 if (ret)
4393                         goto out;
4394                 end_blk -= blk_per_bucket;
4395         }
4396
4397         /* Move half of the xattr in target_blkno to the next bucket. */
4398         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4399                                         target_blk + blk_per_bucket, NULL, 0);
4400
4401         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4402         ocfs2_xattr_bucket_journal_dirty(handle, first);
4403
4404 out:
4405         return ret;
4406 }
4407
4408 /*
4409  * Add new xattr bucket in an extent record and adjust the buckets
4410  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4411  * bucket we want to insert into.
4412  *
4413  * In the easy case, we will move all the buckets after target down by
4414  * one. Half of target's xattrs will be moved to the next bucket.
4415  *
4416  * If current cluster is full, we'll allocate a new one.  This may not
4417  * be contiguous.  The underlying calls will make sure that there is
4418  * space for the insert, shifting buckets around if necessary.
4419  * 'target' may be moved by those calls.
4420  */
4421 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4422                                       struct buffer_head *xb_bh,
4423                                       struct ocfs2_xattr_bucket *target,
4424                                       struct ocfs2_xattr_set_ctxt *ctxt)
4425 {
4426         struct ocfs2_xattr_block *xb =
4427                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4428         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4429         struct ocfs2_extent_list *el = &xb_root->xt_list;
4430         u32 name_hash =
4431                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4432         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4433         int ret, num_buckets, extend = 1;
4434         u64 p_blkno;
4435         u32 e_cpos, num_clusters;
4436         /* The bucket at the front of the extent */
4437         struct ocfs2_xattr_bucket *first;
4438
4439         mlog(0, "Add new xattr bucket starting from %llu\n",
4440              (unsigned long long)bucket_blkno(target));
4441
4442         /* The first bucket of the original extent */
4443         first = ocfs2_xattr_bucket_new(inode);
4444         if (!first) {
4445                 ret = -ENOMEM;
4446                 mlog_errno(ret);
4447                 goto out;
4448         }
4449
4450         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4451                                   &num_clusters, el);
4452         if (ret) {
4453                 mlog_errno(ret);
4454                 goto out;
4455         }
4456
4457         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4458         if (ret) {
4459                 mlog_errno(ret);
4460                 goto out;
4461         }
4462
4463         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4464         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4465                 /*
4466                  * This can move first+target if the target bucket moves
4467                  * to the new extent.
4468                  */
4469                 ret = ocfs2_add_new_xattr_cluster(inode,
4470                                                   xb_bh,
4471                                                   first,
4472                                                   target,
4473                                                   &num_clusters,
4474                                                   e_cpos,
4475                                                   &extend,
4476                                                   ctxt);
4477                 if (ret) {
4478                         mlog_errno(ret);
4479                         goto out;
4480                 }
4481         }
4482
4483         if (extend) {
4484                 ret = ocfs2_extend_xattr_bucket(inode,
4485                                                 ctxt->handle,
4486                                                 first,
4487                                                 bucket_blkno(target),
4488                                                 num_clusters);
4489                 if (ret)
4490                         mlog_errno(ret);
4491         }
4492
4493 out:
4494         ocfs2_xattr_bucket_free(first);
4495
4496         return ret;
4497 }
4498
4499 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4500                                         struct ocfs2_xattr_bucket *bucket,
4501                                         int offs)
4502 {
4503         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4504
4505         offs = offs % inode->i_sb->s_blocksize;
4506         return bucket_block(bucket, block_off) + offs;
4507 }
4508
4509 /*
4510  * Handle the normal xattr set, including replace, delete and new.
4511  *
4512  * Note: "local" indicates the real data's locality. So we can't
4513  * just its bucket locality by its length.
4514  */
4515 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4516                                          struct ocfs2_xattr_info *xi,
4517                                          struct ocfs2_xattr_search *xs,
4518                                          u32 name_hash,
4519                                          int local)
4520 {
4521         struct ocfs2_xattr_entry *last, *xe;
4522         int name_len = strlen(xi->name);
4523         struct ocfs2_xattr_header *xh = xs->header;
4524         u16 count = le16_to_cpu(xh->xh_count), start;
4525         size_t blocksize = inode->i_sb->s_blocksize;
4526         char *val;
4527         size_t offs, size, new_size;
4528
4529         last = &xh->xh_entries[count];
4530         if (!xs->not_found) {
4531                 xe = xs->here;
4532                 offs = le16_to_cpu(xe->xe_name_offset);
4533                 if (ocfs2_xattr_is_local(xe))
4534                         size = OCFS2_XATTR_SIZE(name_len) +
4535                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4536                 else
4537                         size = OCFS2_XATTR_SIZE(name_len) +
4538                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4539
4540                 /*
4541                  * If the new value will be stored outside, xi->value has been
4542                  * initalized as an empty ocfs2_xattr_value_root, and the same
4543                  * goes with xi->value_len, so we can set new_size safely here.
4544                  * See ocfs2_xattr_set_in_bucket.
4545                  */
4546                 new_size = OCFS2_XATTR_SIZE(name_len) +
4547                            OCFS2_XATTR_SIZE(xi->value_len);
4548
4549                 le16_add_cpu(&xh->xh_name_value_len, -size);
4550                 if (xi->value) {
4551                         if (new_size > size)
4552                                 goto set_new_name_value;
4553
4554                         /* Now replace the old value with new one. */
4555                         if (local)
4556                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4557                         else
4558                                 xe->xe_value_size = 0;
4559
4560                         val = ocfs2_xattr_bucket_get_val(inode,
4561                                                          xs->bucket, offs);
4562                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4563                                size - OCFS2_XATTR_SIZE(name_len));
4564                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4565                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4566                                        xi->value, xi->value_len);
4567
4568                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4569                         ocfs2_xattr_set_local(xe, local);
4570                         return;
4571                 } else {
4572                         /*
4573                          * Remove the old entry if there is more than one.
4574                          * We don't remove the last entry so that we can
4575                          * use it to indicate the hash value of the empty
4576                          * bucket.
4577                          */
4578                         last -= 1;
4579                         le16_add_cpu(&xh->xh_count, -1);
4580                         if (xh->xh_count) {
4581                                 memmove(xe, xe + 1,
4582                                         (void *)last - (void *)xe);
4583                                 memset(last, 0,
4584                                        sizeof(struct ocfs2_xattr_entry));
4585                         } else
4586                                 xh->xh_free_start =
4587                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4588
4589                         return;
4590                 }
4591         } else {
4592                 /* find a new entry for insert. */
4593                 int low = 0, high = count - 1, tmp;
4594                 struct ocfs2_xattr_entry *tmp_xe;
4595
4596                 while (low <= high && count) {
4597                         tmp = (low + high) / 2;
4598                         tmp_xe = &xh->xh_entries[tmp];
4599
4600                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4601                                 low = tmp + 1;
4602                         else if (name_hash <
4603                                  le32_to_cpu(tmp_xe->xe_name_hash))
4604                                 high = tmp - 1;
4605                         else {
4606                                 low = tmp;
4607                                 break;
4608                         }
4609                 }
4610
4611                 xe = &xh->xh_entries[low];
4612                 if (low != count)
4613                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4614
4615                 le16_add_cpu(&xh->xh_count, 1);
4616                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4617                 xe->xe_name_hash = cpu_to_le32(name_hash);
4618                 xe->xe_name_len = name_len;
4619                 ocfs2_xattr_set_type(xe, xi->name_index);
4620         }
4621
4622 set_new_name_value:
4623         /* Insert the new name+value. */
4624         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4625
4626         /*
4627          * We must make sure that the name/value pair
4628          * exists in the same block.
4629          */
4630         offs = le16_to_cpu(xh->xh_free_start);
4631         start = offs - size;
4632
4633         if (start >> inode->i_sb->s_blocksize_bits !=
4634             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4635                 offs = offs - offs % blocksize;
4636                 xh->xh_free_start = cpu_to_le16(offs);
4637         }
4638
4639         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4640         xe->xe_name_offset = cpu_to_le16(offs - size);
4641
4642         memset(val, 0, size);
4643         memcpy(val, xi->name, name_len);
4644         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4645
4646         xe->xe_value_size = cpu_to_le64(xi->value_len);
4647         ocfs2_xattr_set_local(xe, local);
4648         xs->here = xe;
4649         le16_add_cpu(&xh->xh_free_start, -size);
4650         le16_add_cpu(&xh->xh_name_value_len, size);
4651
4652         return;
4653 }
4654
4655 /*
4656  * Set the xattr entry in the specified bucket.
4657  * The bucket is indicated by xs->bucket and it should have the enough
4658  * space for the xattr insertion.
4659  */
4660 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4661                                            handle_t *handle,
4662                                            struct ocfs2_xattr_info *xi,
4663                                            struct ocfs2_xattr_search *xs,
4664                                            u32 name_hash,
4665                                            int local)
4666 {
4667         int ret;
4668         u64 blkno;
4669
4670         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4671              (unsigned long)xi->value_len, xi->name_index,
4672              (unsigned long long)bucket_blkno(xs->bucket));
4673
4674         if (!xs->bucket->bu_bhs[1]) {
4675                 blkno = bucket_blkno(xs->bucket);
4676                 ocfs2_xattr_bucket_relse(xs->bucket);
4677                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4678                 if (ret) {
4679                         mlog_errno(ret);
4680                         goto out;
4681                 }
4682         }
4683
4684         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4685                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4686         if (ret < 0) {
4687                 mlog_errno(ret);
4688                 goto out;
4689         }
4690
4691         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4692         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4693
4694 out:
4695         return ret;
4696 }
4697
4698 /*
4699  * Truncate the specified xe_off entry in xattr bucket.
4700  * bucket is indicated by header_bh and len is the new length.
4701  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4702  *
4703  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4704  */
4705 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4706                                              struct ocfs2_xattr_bucket *bucket,
4707                                              int xe_off,
4708                                              int len,
4709                                              struct ocfs2_xattr_set_ctxt *ctxt)
4710 {
4711         int ret, offset;
4712         u64 value_blk;
4713         struct ocfs2_xattr_entry *xe;
4714         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4715         size_t blocksize = inode->i_sb->s_blocksize;
4716         struct ocfs2_xattr_value_buf vb = {
4717                 .vb_access = ocfs2_journal_access,
4718         };
4719
4720         xe = &xh->xh_entries[xe_off];
4721
4722         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4723
4724         offset = le16_to_cpu(xe->xe_name_offset) +
4725                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4726
4727         value_blk = offset / blocksize;
4728
4729         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4730         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4731
4732         vb.vb_bh = bucket->bu_bhs[value_blk];
4733         BUG_ON(!vb.vb_bh);
4734
4735         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4736                 (vb.vb_bh->b_data + offset % blocksize);
4737
4738         /*
4739          * From here on out we have to dirty the bucket.  The generic
4740          * value calls only modify one of the bucket's bhs, but we need
4741          * to send the bucket at once.  So if they error, they *could* have
4742          * modified something.  We have to assume they did, and dirty
4743          * the whole bucket.  This leaves us in a consistent state.
4744          */
4745         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4746              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4747         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4748         if (ret) {
4749                 mlog_errno(ret);
4750                 goto out;
4751         }
4752
4753         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4754                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4755         if (ret) {
4756                 mlog_errno(ret);
4757                 goto out;
4758         }
4759
4760         xe->xe_value_size = cpu_to_le64(len);
4761
4762         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4763
4764 out:
4765         return ret;
4766 }
4767
4768 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4769                                         struct ocfs2_xattr_search *xs,
4770                                         int len,
4771                                         struct ocfs2_xattr_set_ctxt *ctxt)
4772 {
4773         int ret, offset;
4774         struct ocfs2_xattr_entry *xe = xs->here;
4775         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4776
4777         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4778
4779         offset = xe - xh->xh_entries;
4780         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4781                                                 offset, len, ctxt);
4782         if (ret)
4783                 mlog_errno(ret);
4784
4785         return ret;
4786 }
4787
4788 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4789                                                 handle_t *handle,
4790                                                 struct ocfs2_xattr_search *xs,
4791                                                 char *val,
4792                                                 int value_len)
4793 {
4794         int ret, offset, block_off;
4795         struct ocfs2_xattr_value_root *xv;
4796         struct ocfs2_xattr_entry *xe = xs->here;
4797         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4798         void *base;
4799
4800         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4801
4802         ret = ocfs2_xattr_bucket_get_name_value(inode, xh,
4803                                                 xe - xh->xh_entries,
4804                                                 &block_off,
4805                                                 &offset);
4806         if (ret) {
4807                 mlog_errno(ret);
4808                 goto out;
4809         }
4810
4811         base = bucket_block(xs->bucket, block_off);
4812         xv = (struct ocfs2_xattr_value_root *)(base + offset +
4813                  OCFS2_XATTR_SIZE(xe->xe_name_len));
4814
4815         ret = __ocfs2_xattr_set_value_outside(inode, handle,
4816                                               xv, val, value_len);
4817         if (ret)
4818                 mlog_errno(ret);
4819 out:
4820         return ret;
4821 }
4822
4823 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4824                                   struct buffer_head *root_bh,
4825                                   u64 blkno,
4826                                   u32 cpos,
4827                                   u32 len)
4828 {
4829         int ret;
4830         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4831         struct inode *tl_inode = osb->osb_tl_inode;
4832         handle_t *handle;
4833         struct ocfs2_xattr_block *xb =
4834                         (struct ocfs2_xattr_block *)root_bh->b_data;
4835         struct ocfs2_alloc_context *meta_ac = NULL;
4836         struct ocfs2_cached_dealloc_ctxt dealloc;
4837         struct ocfs2_extent_tree et;
4838
4839         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4840
4841         ocfs2_init_dealloc_ctxt(&dealloc);
4842
4843         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4844              cpos, len, (unsigned long long)blkno);
4845
4846         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4847
4848         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4849         if (ret) {
4850                 mlog_errno(ret);
4851                 return ret;
4852         }
4853
4854         mutex_lock(&tl_inode->i_mutex);
4855
4856         if (ocfs2_truncate_log_needs_flush(osb)) {
4857                 ret = __ocfs2_flush_truncate_log(osb);
4858                 if (ret < 0) {
4859                         mlog_errno(ret);
4860                         goto out;
4861                 }
4862         }
4863
4864         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4865         if (IS_ERR(handle)) {
4866                 ret = -ENOMEM;
4867                 mlog_errno(ret);
4868                 goto out;
4869         }
4870
4871         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4872                                       OCFS2_JOURNAL_ACCESS_WRITE);
4873         if (ret) {
4874                 mlog_errno(ret);
4875                 goto out_commit;
4876         }
4877
4878         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4879                                   &dealloc);
4880         if (ret) {
4881                 mlog_errno(ret);
4882                 goto out_commit;
4883         }
4884
4885         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4886
4887         ret = ocfs2_journal_dirty(handle, root_bh);
4888         if (ret) {
4889                 mlog_errno(ret);
4890                 goto out_commit;
4891         }
4892
4893         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4894         if (ret)
4895                 mlog_errno(ret);
4896
4897 out_commit:
4898         ocfs2_commit_trans(osb, handle);
4899 out:
4900         ocfs2_schedule_truncate_log_flush(osb, 1);
4901
4902         mutex_unlock(&tl_inode->i_mutex);
4903
4904         if (meta_ac)
4905                 ocfs2_free_alloc_context(meta_ac);
4906
4907         ocfs2_run_deallocs(osb, &dealloc);
4908
4909         return ret;
4910 }
4911
4912 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4913                                          handle_t *handle,
4914                                          struct ocfs2_xattr_search *xs)
4915 {
4916         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4917         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4918                                                 le16_to_cpu(xh->xh_count) - 1];
4919         int ret = 0;
4920
4921         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4922                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4923         if (ret) {
4924                 mlog_errno(ret);
4925                 return;
4926         }
4927
4928         /* Remove the old entry. */
4929         memmove(xs->here, xs->here + 1,
4930                 (void *)last - (void *)xs->here);
4931         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4932         le16_add_cpu(&xh->xh_count, -1);
4933
4934         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4935 }
4936
4937 /*
4938  * Set the xattr name/value in the bucket specified in xs.
4939  *
4940  * As the new value in xi may be stored in the bucket or in an outside cluster,
4941  * we divide the whole process into 3 steps:
4942  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4943  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4944  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4945  * 4. If the clusters for the new outside value can't be allocated, we need
4946  *    to free the xattr we allocated in set.
4947  */
4948 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4949                                      struct ocfs2_xattr_info *xi,
4950                                      struct ocfs2_xattr_search *xs,
4951                                      struct ocfs2_xattr_set_ctxt *ctxt)
4952 {
4953         int ret, local = 1;
4954         size_t value_len;
4955         char *val = (char *)xi->value;
4956         struct ocfs2_xattr_entry *xe = xs->here;
4957         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4958                                               strlen(xi->name));
4959
4960         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4961                 /*
4962                  * We need to truncate the xattr storage first.
4963                  *
4964                  * If both the old and new value are stored to
4965                  * outside block, we only need to truncate
4966                  * the storage and then set the value outside.
4967                  *
4968                  * If the new value should be stored within block,
4969                  * we should free all the outside block first and
4970                  * the modification to the xattr block will be done
4971                  * by following steps.
4972                  */
4973                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4974                         value_len = xi->value_len;
4975                 else
4976                         value_len = 0;
4977
4978                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4979                                                            value_len,
4980                                                            ctxt);
4981                 if (ret)
4982                         goto out;
4983
4984                 if (value_len)
4985                         goto set_value_outside;
4986         }
4987
4988         value_len = xi->value_len;
4989         /* So we have to handle the inside block change now. */
4990         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4991                 /*
4992                  * If the new value will be stored outside of block,
4993                  * initalize a new empty value root and insert it first.
4994                  */
4995                 local = 0;
4996                 xi->value = &def_xv;
4997                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4998         }
4999
5000         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5001                                               name_hash, local);
5002         if (ret) {
5003                 mlog_errno(ret);
5004                 goto out;
5005         }
5006
5007         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5008                 goto out;
5009
5010         /* allocate the space now for the outside block storage. */
5011         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5012                                                    value_len, ctxt);
5013         if (ret) {
5014                 mlog_errno(ret);
5015
5016                 if (xs->not_found) {
5017                         /*
5018                          * We can't allocate enough clusters for outside
5019                          * storage and we have allocated xattr already,
5020                          * so need to remove it.
5021                          */
5022                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5023                 }
5024                 goto out;
5025         }
5026
5027 set_value_outside:
5028         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5029                                                    xs, val, value_len);
5030 out:
5031         return ret;
5032 }
5033
5034 /*
5035  * check whether the xattr bucket is filled up with the same hash value.
5036  * If we want to insert the xattr with the same hash, return -ENOSPC.
5037  * If we want to insert a xattr with different hash value, go ahead
5038  * and ocfs2_divide_xattr_bucket will handle this.
5039  */
5040 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5041                                               struct ocfs2_xattr_bucket *bucket,
5042                                               const char *name)
5043 {
5044         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5045         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5046
5047         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5048                 return 0;
5049
5050         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5051             xh->xh_entries[0].xe_name_hash) {
5052                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5053                      "hash = %u\n",
5054                      (unsigned long long)bucket_blkno(bucket),
5055                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5056                 return -ENOSPC;
5057         }
5058
5059         return 0;
5060 }
5061
5062 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5063                                              struct ocfs2_xattr_info *xi,
5064                                              struct ocfs2_xattr_search *xs,
5065                                              struct ocfs2_xattr_set_ctxt *ctxt)
5066 {
5067         struct ocfs2_xattr_header *xh;
5068         struct ocfs2_xattr_entry *xe;
5069         u16 count, header_size, xh_free_start;
5070         int free, max_free, need, old;
5071         size_t value_size = 0, name_len = strlen(xi->name);
5072         size_t blocksize = inode->i_sb->s_blocksize;
5073         int ret, allocation = 0;
5074
5075         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5076
5077 try_again:
5078         xh = xs->header;
5079         count = le16_to_cpu(xh->xh_count);
5080         xh_free_start = le16_to_cpu(xh->xh_free_start);
5081         header_size = sizeof(struct ocfs2_xattr_header) +
5082                         count * sizeof(struct ocfs2_xattr_entry);
5083         max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5084                 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5085
5086         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5087                         "of %u which exceed block size\n",
5088                         (unsigned long long)bucket_blkno(xs->bucket),
5089                         header_size);
5090
5091         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5092                 value_size = OCFS2_XATTR_ROOT_SIZE;
5093         else if (xi->value)
5094                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5095
5096         if (xs->not_found)
5097                 need = sizeof(struct ocfs2_xattr_entry) +
5098                         OCFS2_XATTR_SIZE(name_len) + value_size;
5099         else {
5100                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5101
5102                 /*
5103                  * We only replace the old value if the new length is smaller
5104                  * than the old one. Otherwise we will allocate new space in the
5105                  * bucket to store it.
5106                  */
5107                 xe = xs->here;
5108                 if (ocfs2_xattr_is_local(xe))
5109                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5110                 else
5111                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5112
5113                 if (old >= value_size)
5114                         need = 0;
5115         }
5116
5117         free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5118         /*
5119          * We need to make sure the new name/value pair
5120          * can exist in the same block.
5121          */
5122         if (xh_free_start % blocksize < need)
5123                 free -= xh_free_start % blocksize;
5124
5125         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5126              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5127              " %u\n", xs->not_found,
5128              (unsigned long long)bucket_blkno(xs->bucket),
5129              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5130              le16_to_cpu(xh->xh_name_value_len));
5131
5132         if (free < need ||
5133             (xs->not_found &&
5134              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5135                 if (need <= max_free &&
5136                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5137                         /*
5138                          * We can create the space by defragment. Since only the
5139                          * name/value will be moved, the xe shouldn't be changed
5140                          * in xs.
5141                          */
5142                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5143                                                         xs->bucket);
5144                         if (ret) {
5145                                 mlog_errno(ret);
5146                                 goto out;
5147                         }
5148
5149                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5150                         free = xh_free_start - header_size
5151                                 - OCFS2_XATTR_HEADER_GAP;
5152                         if (xh_free_start % blocksize < need)
5153                                 free -= xh_free_start % blocksize;
5154
5155                         if (free >= need)
5156                                 goto xattr_set;
5157
5158                         mlog(0, "Can't get enough space for xattr insert by "
5159                              "defragment. Need %u bytes, but we have %d, so "
5160                              "allocate new bucket for it.\n", need, free);
5161                 }
5162
5163                 /*
5164                  * We have to add new buckets or clusters and one
5165                  * allocation should leave us enough space for insert.
5166                  */
5167                 BUG_ON(allocation);
5168
5169                 /*
5170                  * We do not allow for overlapping ranges between buckets. And
5171                  * the maximum number of collisions we will allow for then is
5172                  * one bucket's worth, so check it here whether we need to
5173                  * add a new bucket for the insert.
5174                  */
5175                 ret = ocfs2_check_xattr_bucket_collision(inode,
5176                                                          xs->bucket,
5177                                                          xi->name);
5178                 if (ret) {
5179                         mlog_errno(ret);
5180                         goto out;
5181                 }
5182
5183                 ret = ocfs2_add_new_xattr_bucket(inode,
5184                                                  xs->xattr_bh,
5185                                                  xs->bucket,
5186                                                  ctxt);
5187                 if (ret) {
5188                         mlog_errno(ret);
5189                         goto out;
5190                 }
5191
5192                 /*
5193                  * ocfs2_add_new_xattr_bucket() will have updated
5194                  * xs->bucket if it moved, but it will not have updated
5195                  * any of the other search fields.  Thus, we drop it and
5196                  * re-search.  Everything should be cached, so it'll be
5197                  * quick.
5198                  */
5199                 ocfs2_xattr_bucket_relse(xs->bucket);
5200                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5201                                                    xi->name_index,
5202                                                    xi->name, xs);
5203                 if (ret && ret != -ENODATA)
5204                         goto out;
5205                 xs->not_found = ret;
5206                 allocation = 1;
5207                 goto try_again;
5208         }
5209
5210 xattr_set:
5211         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5212 out:
5213         mlog_exit(ret);
5214         return ret;
5215 }
5216
5217 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5218                                         struct ocfs2_xattr_bucket *bucket,
5219                                         void *para)
5220 {
5221         int ret = 0;
5222         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5223         u16 i;
5224         struct ocfs2_xattr_entry *xe;
5225         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5226         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5227         int credits = ocfs2_remove_extent_credits(osb->sb) +
5228                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5229
5230
5231         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5232
5233         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5234                 xe = &xh->xh_entries[i];
5235                 if (ocfs2_xattr_is_local(xe))
5236                         continue;
5237
5238                 ctxt.handle = ocfs2_start_trans(osb, credits);
5239                 if (IS_ERR(ctxt.handle)) {
5240                         ret = PTR_ERR(ctxt.handle);
5241                         mlog_errno(ret);
5242                         break;
5243                 }
5244
5245                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5246                                                         i, 0, &ctxt);
5247
5248                 ocfs2_commit_trans(osb, ctxt.handle);
5249                 if (ret) {
5250                         mlog_errno(ret);
5251                         break;
5252                 }
5253         }
5254
5255         ocfs2_schedule_truncate_log_flush(osb, 1);
5256         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5257         return ret;
5258 }
5259
5260 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5261                                           struct buffer_head *xb_bh)
5262 {
5263         struct ocfs2_xattr_block *xb =
5264                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5265         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5266         int ret = 0;
5267         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5268         u64 p_blkno;
5269
5270         if (le16_to_cpu(el->l_next_free_rec) == 0)
5271                 return 0;
5272
5273         while (name_hash > 0) {
5274                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5275                                           &e_cpos, &num_clusters, el);
5276                 if (ret) {
5277                         mlog_errno(ret);
5278                         goto out;
5279                 }
5280
5281                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5282                                                   ocfs2_delete_xattr_in_bucket,
5283                                                   NULL);
5284                 if (ret) {
5285                         mlog_errno(ret);
5286                         goto out;
5287                 }
5288
5289                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5290                                              p_blkno, e_cpos, num_clusters);
5291                 if (ret) {
5292                         mlog_errno(ret);
5293                         break;
5294                 }
5295
5296                 if (e_cpos == 0)
5297                         break;
5298
5299                 name_hash = e_cpos - 1;
5300         }
5301
5302 out:
5303         return ret;
5304 }
5305
5306 /*
5307  * 'security' attributes support
5308  */
5309 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5310                                         size_t list_size, const char *name,
5311                                         size_t name_len)
5312 {
5313         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5314         const size_t total_len = prefix_len + name_len + 1;
5315
5316         if (list && total_len <= list_size) {
5317                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5318                 memcpy(list + prefix_len, name, name_len);
5319                 list[prefix_len + name_len] = '\0';
5320         }
5321         return total_len;
5322 }
5323
5324 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5325                                     void *buffer, size_t size)
5326 {
5327         if (strcmp(name, "") == 0)
5328                 return -EINVAL;
5329         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5330                                buffer, size);
5331 }
5332
5333 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5334                                     const void *value, size_t size, int flags)
5335 {
5336         if (strcmp(name, "") == 0)
5337                 return -EINVAL;
5338
5339         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5340                                size, flags);
5341 }
5342
5343 int ocfs2_init_security_get(struct inode *inode,
5344                             struct inode *dir,
5345                             struct ocfs2_security_xattr_info *si)
5346 {
5347         /* check whether ocfs2 support feature xattr */
5348         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5349                 return -EOPNOTSUPP;
5350         return security_inode_init_security(inode, dir, &si->name, &si->value,
5351                                             &si->value_len);
5352 }
5353
5354 int ocfs2_init_security_set(handle_t *handle,
5355                             struct inode *inode,
5356                             struct buffer_head *di_bh,
5357                             struct ocfs2_security_xattr_info *si,
5358                             struct ocfs2_alloc_context *xattr_ac,
5359                             struct ocfs2_alloc_context *data_ac)
5360 {
5361         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5362                                      OCFS2_XATTR_INDEX_SECURITY,
5363                                      si->name, si->value, si->value_len, 0,
5364                                      xattr_ac, data_ac);
5365 }
5366
5367 struct xattr_handler ocfs2_xattr_security_handler = {
5368         .prefix = XATTR_SECURITY_PREFIX,
5369         .list   = ocfs2_xattr_security_list,
5370         .get    = ocfs2_xattr_security_get,
5371         .set    = ocfs2_xattr_security_set,
5372 };
5373
5374 /*
5375  * 'trusted' attributes support
5376  */
5377 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5378                                        size_t list_size, const char *name,
5379                                        size_t name_len)
5380 {
5381         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5382         const size_t total_len = prefix_len + name_len + 1;
5383
5384         if (list && total_len <= list_size) {
5385                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5386                 memcpy(list + prefix_len, name, name_len);
5387                 list[prefix_len + name_len] = '\0';
5388         }
5389         return total_len;
5390 }
5391
5392 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5393                                    void *buffer, size_t size)
5394 {
5395         if (strcmp(name, "") == 0)
5396                 return -EINVAL;
5397         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5398                                buffer, size);
5399 }
5400
5401 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5402                                    const void *value, size_t size, int flags)
5403 {
5404         if (strcmp(name, "") == 0)
5405                 return -EINVAL;
5406
5407         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5408                                size, flags);
5409 }
5410
5411 struct xattr_handler ocfs2_xattr_trusted_handler = {
5412         .prefix = XATTR_TRUSTED_PREFIX,
5413         .list   = ocfs2_xattr_trusted_list,
5414         .get    = ocfs2_xattr_trusted_get,
5415         .set    = ocfs2_xattr_trusted_set,
5416 };
5417
5418 /*
5419  * 'user' attributes support
5420  */
5421 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5422                                     size_t list_size, const char *name,
5423                                     size_t name_len)
5424 {
5425         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5426         const size_t total_len = prefix_len + name_len + 1;
5427         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5428
5429         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5430                 return 0;
5431
5432         if (list && total_len <= list_size) {
5433                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5434                 memcpy(list + prefix_len, name, name_len);
5435                 list[prefix_len + name_len] = '\0';
5436         }
5437         return total_len;
5438 }
5439
5440 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5441                                 void *buffer, size_t size)
5442 {
5443         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5444
5445         if (strcmp(name, "") == 0)
5446                 return -EINVAL;
5447         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5448                 return -EOPNOTSUPP;
5449         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5450                                buffer, size);
5451 }
5452
5453 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5454                                 const void *value, size_t size, int flags)
5455 {
5456         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5457
5458         if (strcmp(name, "") == 0)
5459                 return -EINVAL;
5460         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5461                 return -EOPNOTSUPP;
5462
5463         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5464                                size, flags);
5465 }
5466
5467 struct xattr_handler ocfs2_xattr_user_handler = {
5468         .prefix = XATTR_USER_PREFIX,
5469         .list   = ocfs2_xattr_user_list,
5470         .get    = ocfs2_xattr_user_get,
5471         .set    = ocfs2_xattr_user_set,
5472 };