ocfs2: calculate and reserve credits for xattr value in mknod
[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_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
86                                          - sizeof(struct ocfs2_xattr_header) \
87                                          - sizeof(__u32))
88 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
89                                          - sizeof(struct ocfs2_xattr_block) \
90                                          - sizeof(struct ocfs2_xattr_header) \
91                                          - sizeof(__u32))
92
93 static struct ocfs2_xattr_def_value_root def_xv = {
94         .xv.xr_list.l_count = cpu_to_le16(1),
95 };
96
97 struct xattr_handler *ocfs2_xattr_handlers[] = {
98         &ocfs2_xattr_user_handler,
99 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
100         &ocfs2_xattr_acl_access_handler,
101         &ocfs2_xattr_acl_default_handler,
102 #endif
103         &ocfs2_xattr_trusted_handler,
104         &ocfs2_xattr_security_handler,
105         NULL
106 };
107
108 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
109         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
110 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
112                                         = &ocfs2_xattr_acl_access_handler,
113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
114                                         = &ocfs2_xattr_acl_default_handler,
115 #endif
116         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
117         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
118 };
119
120 struct ocfs2_xattr_info {
121         int name_index;
122         const char *name;
123         const void *value;
124         size_t value_len;
125 };
126
127 struct ocfs2_xattr_search {
128         struct buffer_head *inode_bh;
129         /*
130          * xattr_bh point to the block buffer head which has extended attribute
131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
132          */
133         struct buffer_head *xattr_bh;
134         struct ocfs2_xattr_header *header;
135         struct ocfs2_xattr_bucket *bucket;
136         void *base;
137         void *end;
138         struct ocfs2_xattr_entry *here;
139         int not_found;
140 };
141
142 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
143                                              struct ocfs2_xattr_header *xh,
144                                              int index,
145                                              int *block_off,
146                                              int *new_offset);
147
148 static int ocfs2_xattr_block_find(struct inode *inode,
149                                   int name_index,
150                                   const char *name,
151                                   struct ocfs2_xattr_search *xs);
152 static int ocfs2_xattr_index_block_find(struct inode *inode,
153                                         struct buffer_head *root_bh,
154                                         int name_index,
155                                         const char *name,
156                                         struct ocfs2_xattr_search *xs);
157
158 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
159                                         struct ocfs2_xattr_tree_root *xt,
160                                         char *buffer,
161                                         size_t buffer_size);
162
163 static int ocfs2_xattr_create_index_block(struct inode *inode,
164                                           struct ocfs2_xattr_search *xs,
165                                           struct ocfs2_xattr_set_ctxt *ctxt);
166
167 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
168                                              struct ocfs2_xattr_info *xi,
169                                              struct ocfs2_xattr_search *xs,
170                                              struct ocfs2_xattr_set_ctxt *ctxt);
171
172 static int ocfs2_delete_xattr_index_block(struct inode *inode,
173                                           struct buffer_head *xb_bh);
174 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
175                                   u64 src_blk, u64 last_blk, u64 to_blk,
176                                   unsigned int start_bucket,
177                                   u32 *first_hash);
178
179 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
180 {
181         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
182 }
183
184 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
185 {
186         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
187 }
188
189 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
190 {
191         u16 len = sb->s_blocksize -
192                  offsetof(struct ocfs2_xattr_header, xh_entries);
193
194         return len / sizeof(struct ocfs2_xattr_entry);
195 }
196
197 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
198 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
199 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
200
201 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
202 {
203         struct ocfs2_xattr_bucket *bucket;
204         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
205
206         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
207
208         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
209         if (bucket) {
210                 bucket->bu_inode = inode;
211                 bucket->bu_blocks = blks;
212         }
213
214         return bucket;
215 }
216
217 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
218 {
219         int i;
220
221         for (i = 0; i < bucket->bu_blocks; i++) {
222                 brelse(bucket->bu_bhs[i]);
223                 bucket->bu_bhs[i] = NULL;
224         }
225 }
226
227 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
228 {
229         if (bucket) {
230                 ocfs2_xattr_bucket_relse(bucket);
231                 bucket->bu_inode = NULL;
232                 kfree(bucket);
233         }
234 }
235
236 /*
237  * A bucket that has never been written to disk doesn't need to be
238  * read.  We just need the buffer_heads.  Don't call this for
239  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
240  * them fully.
241  */
242 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
243                                    u64 xb_blkno)
244 {
245         int i, rc = 0;
246
247         for (i = 0; i < bucket->bu_blocks; i++) {
248                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
249                                               xb_blkno + i);
250                 if (!bucket->bu_bhs[i]) {
251                         rc = -EIO;
252                         mlog_errno(rc);
253                         break;
254                 }
255
256                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
257                                            bucket->bu_bhs[i]))
258                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
259                                                       bucket->bu_bhs[i]);
260         }
261
262         if (rc)
263                 ocfs2_xattr_bucket_relse(bucket);
264         return rc;
265 }
266
267 /* Read the xattr bucket at xb_blkno */
268 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
269                                    u64 xb_blkno)
270 {
271         int rc;
272
273         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
274                                bucket->bu_blocks, bucket->bu_bhs, 0,
275                                NULL);
276         if (!rc) {
277                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
278                                                  bucket->bu_bhs,
279                                                  bucket->bu_blocks,
280                                                  &bucket_xh(bucket)->xh_check);
281                 if (rc)
282                         mlog_errno(rc);
283         }
284
285         if (rc)
286                 ocfs2_xattr_bucket_relse(bucket);
287         return rc;
288 }
289
290 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
291                                              struct ocfs2_xattr_bucket *bucket,
292                                              int type)
293 {
294         int i, rc = 0;
295
296         for (i = 0; i < bucket->bu_blocks; i++) {
297                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
298                                           bucket->bu_bhs[i], type);
299                 if (rc) {
300                         mlog_errno(rc);
301                         break;
302                 }
303         }
304
305         return rc;
306 }
307
308 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
309                                              struct ocfs2_xattr_bucket *bucket)
310 {
311         int i;
312
313         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
314                                    bucket->bu_bhs, bucket->bu_blocks,
315                                    &bucket_xh(bucket)->xh_check);
316
317         for (i = 0; i < bucket->bu_blocks; i++)
318                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
319 }
320
321 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
322                                          struct ocfs2_xattr_bucket *src)
323 {
324         int i;
325         int blocksize = src->bu_inode->i_sb->s_blocksize;
326
327         BUG_ON(dest->bu_blocks != src->bu_blocks);
328         BUG_ON(dest->bu_inode != src->bu_inode);
329
330         for (i = 0; i < src->bu_blocks; i++) {
331                 memcpy(bucket_block(dest, i), bucket_block(src, i),
332                        blocksize);
333         }
334 }
335
336 static int ocfs2_validate_xattr_block(struct super_block *sb,
337                                       struct buffer_head *bh)
338 {
339         int rc;
340         struct ocfs2_xattr_block *xb =
341                 (struct ocfs2_xattr_block *)bh->b_data;
342
343         mlog(0, "Validating xattr block %llu\n",
344              (unsigned long long)bh->b_blocknr);
345
346         BUG_ON(!buffer_uptodate(bh));
347
348         /*
349          * If the ecc fails, we return the error but otherwise
350          * leave the filesystem running.  We know any error is
351          * local to this block.
352          */
353         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
354         if (rc)
355                 return rc;
356
357         /*
358          * Errors after here are fatal
359          */
360
361         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
362                 ocfs2_error(sb,
363                             "Extended attribute block #%llu has bad "
364                             "signature %.*s",
365                             (unsigned long long)bh->b_blocknr, 7,
366                             xb->xb_signature);
367                 return -EINVAL;
368         }
369
370         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
371                 ocfs2_error(sb,
372                             "Extended attribute block #%llu has an "
373                             "invalid xb_blkno of %llu",
374                             (unsigned long long)bh->b_blocknr,
375                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
376                 return -EINVAL;
377         }
378
379         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
380                 ocfs2_error(sb,
381                             "Extended attribute block #%llu has an invalid "
382                             "xb_fs_generation of #%u",
383                             (unsigned long long)bh->b_blocknr,
384                             le32_to_cpu(xb->xb_fs_generation));
385                 return -EINVAL;
386         }
387
388         return 0;
389 }
390
391 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
392                                   struct buffer_head **bh)
393 {
394         int rc;
395         struct buffer_head *tmp = *bh;
396
397         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
398                               ocfs2_validate_xattr_block);
399
400         /* If ocfs2_read_block() got us a new bh, pass it up. */
401         if (!rc && !*bh)
402                 *bh = tmp;
403
404         return rc;
405 }
406
407 static inline const char *ocfs2_xattr_prefix(int name_index)
408 {
409         struct xattr_handler *handler = NULL;
410
411         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
412                 handler = ocfs2_xattr_handler_map[name_index];
413
414         return handler ? handler->prefix : NULL;
415 }
416
417 static u32 ocfs2_xattr_name_hash(struct inode *inode,
418                                  const char *name,
419                                  int name_len)
420 {
421         /* Get hash value of uuid from super block */
422         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
423         int i;
424
425         /* hash extended attribute name */
426         for (i = 0; i < name_len; i++) {
427                 hash = (hash << OCFS2_HASH_SHIFT) ^
428                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
429                        *name++;
430         }
431
432         return hash;
433 }
434
435 /*
436  * ocfs2_xattr_hash_entry()
437  *
438  * Compute the hash of an extended attribute.
439  */
440 static void ocfs2_xattr_hash_entry(struct inode *inode,
441                                    struct ocfs2_xattr_header *header,
442                                    struct ocfs2_xattr_entry *entry)
443 {
444         u32 hash = 0;
445         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
446
447         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
448         entry->xe_name_hash = cpu_to_le32(hash);
449
450         return;
451 }
452
453 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
454 {
455         int size = 0;
456
457         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
458                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
459         else
460                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
461         size += sizeof(struct ocfs2_xattr_entry);
462
463         return size;
464 }
465
466 int ocfs2_calc_security_init(struct inode *dir,
467                              struct ocfs2_security_xattr_info *si,
468                              int *want_clusters,
469                              int *xattr_credits,
470                              struct ocfs2_alloc_context **xattr_ac)
471 {
472         int ret = 0;
473         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
474         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
475                                                  si->value_len);
476
477         /*
478          * The max space of security xattr taken inline is
479          * 256(name) + 80(value) + 16(entry) = 352 bytes,
480          * So reserve one metadata block for it is ok.
481          */
482         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
483             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
484                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
485                 if (ret) {
486                         mlog_errno(ret);
487                         return ret;
488                 }
489                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
490         }
491
492         /* reserve clusters for xattr value which will be set in B tree*/
493         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
494                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
495                                                             si->value_len);
496
497                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
498                                                            new_clusters);
499                 *want_clusters += new_clusters;
500         }
501         return ret;
502 }
503
504 int ocfs2_calc_xattr_init(struct inode *dir,
505                           struct buffer_head *dir_bh,
506                           int mode,
507                           struct ocfs2_security_xattr_info *si,
508                           int *want_clusters,
509                           int *xattr_credits,
510                           struct ocfs2_alloc_context **xattr_ac)
511 {
512         int ret = 0;
513         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
514         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
515
516         if (si->enable)
517                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
518                                                      si->value_len);
519
520         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
521                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
522                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
523                                         "", NULL, 0);
524                 if (acl_len > 0) {
525                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
526                         if (S_ISDIR(mode))
527                                 a_size <<= 1;
528                 } else if (acl_len != 0 && acl_len != -ENODATA) {
529                         mlog_errno(ret);
530                         return ret;
531                 }
532         }
533
534         if (!(s_size + a_size))
535                 return ret;
536
537         /*
538          * The max space of security xattr taken inline is
539          * 256(name) + 80(value) + 16(entry) = 352 bytes,
540          * The max space of acl xattr taken inline is
541          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
542          * when blocksize = 512, may reserve one more cluser for
543          * xattr bucket, otherwise reserve one metadata block
544          * for them is ok.
545          */
546         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
547             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
548                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
549                 if (ret) {
550                         mlog_errno(ret);
551                         return ret;
552                 }
553                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
554         }
555
556         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
557             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
558                 *want_clusters += 1;
559                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
560         }
561
562         /*
563          * reserve credits and clusters for xattrs which has large value
564          * and have to be set outside
565          */
566         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
567                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
568                                                         si->value_len);
569                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
570                                                            new_clusters);
571                 *want_clusters += new_clusters;
572         }
573         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
574             acl_len > OCFS2_XATTR_INLINE_SIZE) {
575                 /* for directory, it has DEFAULT and ACCESS two types of acls */
576                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
577                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
578                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
579                                                            new_clusters);
580                 *want_clusters += new_clusters;
581         }
582
583         return ret;
584 }
585
586 static int ocfs2_xattr_extend_allocation(struct inode *inode,
587                                          u32 clusters_to_add,
588                                          struct ocfs2_xattr_value_buf *vb,
589                                          struct ocfs2_xattr_set_ctxt *ctxt)
590 {
591         int status = 0;
592         handle_t *handle = ctxt->handle;
593         enum ocfs2_alloc_restarted why;
594         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
595         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
596         struct ocfs2_extent_tree et;
597
598         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
599
600         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
601
602         status = vb->vb_access(handle, inode, vb->vb_bh,
603                               OCFS2_JOURNAL_ACCESS_WRITE);
604         if (status < 0) {
605                 mlog_errno(status);
606                 goto leave;
607         }
608
609         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
610         status = ocfs2_add_clusters_in_btree(osb,
611                                              inode,
612                                              &logical_start,
613                                              clusters_to_add,
614                                              0,
615                                              &et,
616                                              handle,
617                                              ctxt->data_ac,
618                                              ctxt->meta_ac,
619                                              &why);
620         if (status < 0) {
621                 mlog_errno(status);
622                 goto leave;
623         }
624
625         status = ocfs2_journal_dirty(handle, vb->vb_bh);
626         if (status < 0) {
627                 mlog_errno(status);
628                 goto leave;
629         }
630
631         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
632
633         /*
634          * We should have already allocated enough space before the transaction,
635          * so no need to restart.
636          */
637         BUG_ON(why != RESTART_NONE || clusters_to_add);
638
639 leave:
640
641         return status;
642 }
643
644 static int __ocfs2_remove_xattr_range(struct inode *inode,
645                                       struct ocfs2_xattr_value_buf *vb,
646                                       u32 cpos, u32 phys_cpos, u32 len,
647                                       struct ocfs2_xattr_set_ctxt *ctxt)
648 {
649         int ret;
650         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
651         handle_t *handle = ctxt->handle;
652         struct ocfs2_extent_tree et;
653
654         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
655
656         ret = vb->vb_access(handle, inode, vb->vb_bh,
657                             OCFS2_JOURNAL_ACCESS_WRITE);
658         if (ret) {
659                 mlog_errno(ret);
660                 goto out;
661         }
662
663         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
664                                   &ctxt->dealloc);
665         if (ret) {
666                 mlog_errno(ret);
667                 goto out;
668         }
669
670         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
671
672         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
673         if (ret) {
674                 mlog_errno(ret);
675                 goto out;
676         }
677
678         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
679         if (ret)
680                 mlog_errno(ret);
681
682 out:
683         return ret;
684 }
685
686 static int ocfs2_xattr_shrink_size(struct inode *inode,
687                                    u32 old_clusters,
688                                    u32 new_clusters,
689                                    struct ocfs2_xattr_value_buf *vb,
690                                    struct ocfs2_xattr_set_ctxt *ctxt)
691 {
692         int ret = 0;
693         u32 trunc_len, cpos, phys_cpos, alloc_size;
694         u64 block;
695
696         if (old_clusters <= new_clusters)
697                 return 0;
698
699         cpos = new_clusters;
700         trunc_len = old_clusters - new_clusters;
701         while (trunc_len) {
702                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
703                                                &alloc_size,
704                                                &vb->vb_xv->xr_list);
705                 if (ret) {
706                         mlog_errno(ret);
707                         goto out;
708                 }
709
710                 if (alloc_size > trunc_len)
711                         alloc_size = trunc_len;
712
713                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
714                                                  phys_cpos, alloc_size,
715                                                  ctxt);
716                 if (ret) {
717                         mlog_errno(ret);
718                         goto out;
719                 }
720
721                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
722                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
723                                                        alloc_size);
724                 cpos += alloc_size;
725                 trunc_len -= alloc_size;
726         }
727
728 out:
729         return ret;
730 }
731
732 static int ocfs2_xattr_value_truncate(struct inode *inode,
733                                       struct ocfs2_xattr_value_buf *vb,
734                                       int len,
735                                       struct ocfs2_xattr_set_ctxt *ctxt)
736 {
737         int ret;
738         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
739         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
740
741         if (new_clusters == old_clusters)
742                 return 0;
743
744         if (new_clusters > old_clusters)
745                 ret = ocfs2_xattr_extend_allocation(inode,
746                                                     new_clusters - old_clusters,
747                                                     vb, ctxt);
748         else
749                 ret = ocfs2_xattr_shrink_size(inode,
750                                               old_clusters, new_clusters,
751                                               vb, ctxt);
752
753         return ret;
754 }
755
756 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
757                                   size_t *result, const char *prefix,
758                                   const char *name, int name_len)
759 {
760         char *p = buffer + *result;
761         int prefix_len = strlen(prefix);
762         int total_len = prefix_len + name_len + 1;
763
764         *result += total_len;
765
766         /* we are just looking for how big our buffer needs to be */
767         if (!size)
768                 return 0;
769
770         if (*result > size)
771                 return -ERANGE;
772
773         memcpy(p, prefix, prefix_len);
774         memcpy(p + prefix_len, name, name_len);
775         p[prefix_len + name_len] = '\0';
776
777         return 0;
778 }
779
780 static int ocfs2_xattr_list_entries(struct inode *inode,
781                                     struct ocfs2_xattr_header *header,
782                                     char *buffer, size_t buffer_size)
783 {
784         size_t result = 0;
785         int i, type, ret;
786         const char *prefix, *name;
787
788         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
789                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
790                 type = ocfs2_xattr_get_type(entry);
791                 prefix = ocfs2_xattr_prefix(type);
792
793                 if (prefix) {
794                         name = (const char *)header +
795                                 le16_to_cpu(entry->xe_name_offset);
796
797                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
798                                                      &result, prefix, name,
799                                                      entry->xe_name_len);
800                         if (ret)
801                                 return ret;
802                 }
803         }
804
805         return result;
806 }
807
808 static int ocfs2_xattr_ibody_list(struct inode *inode,
809                                   struct ocfs2_dinode *di,
810                                   char *buffer,
811                                   size_t buffer_size)
812 {
813         struct ocfs2_xattr_header *header = NULL;
814         struct ocfs2_inode_info *oi = OCFS2_I(inode);
815         int ret = 0;
816
817         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
818                 return ret;
819
820         header = (struct ocfs2_xattr_header *)
821                  ((void *)di + inode->i_sb->s_blocksize -
822                  le16_to_cpu(di->i_xattr_inline_size));
823
824         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
825
826         return ret;
827 }
828
829 static int ocfs2_xattr_block_list(struct inode *inode,
830                                   struct ocfs2_dinode *di,
831                                   char *buffer,
832                                   size_t buffer_size)
833 {
834         struct buffer_head *blk_bh = NULL;
835         struct ocfs2_xattr_block *xb;
836         int ret = 0;
837
838         if (!di->i_xattr_loc)
839                 return ret;
840
841         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
842                                      &blk_bh);
843         if (ret < 0) {
844                 mlog_errno(ret);
845                 return ret;
846         }
847
848         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
849         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
850                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
851                 ret = ocfs2_xattr_list_entries(inode, header,
852                                                buffer, buffer_size);
853         } else {
854                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
855                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
856                                                    buffer, buffer_size);
857         }
858
859         brelse(blk_bh);
860
861         return ret;
862 }
863
864 ssize_t ocfs2_listxattr(struct dentry *dentry,
865                         char *buffer,
866                         size_t size)
867 {
868         int ret = 0, i_ret = 0, b_ret = 0;
869         struct buffer_head *di_bh = NULL;
870         struct ocfs2_dinode *di = NULL;
871         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
872
873         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
874                 return -EOPNOTSUPP;
875
876         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
877                 return ret;
878
879         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
880         if (ret < 0) {
881                 mlog_errno(ret);
882                 return ret;
883         }
884
885         di = (struct ocfs2_dinode *)di_bh->b_data;
886
887         down_read(&oi->ip_xattr_sem);
888         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
889         if (i_ret < 0)
890                 b_ret = 0;
891         else {
892                 if (buffer) {
893                         buffer += i_ret;
894                         size -= i_ret;
895                 }
896                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
897                                                buffer, size);
898                 if (b_ret < 0)
899                         i_ret = 0;
900         }
901         up_read(&oi->ip_xattr_sem);
902         ocfs2_inode_unlock(dentry->d_inode, 0);
903
904         brelse(di_bh);
905
906         return i_ret + b_ret;
907 }
908
909 static int ocfs2_xattr_find_entry(int name_index,
910                                   const char *name,
911                                   struct ocfs2_xattr_search *xs)
912 {
913         struct ocfs2_xattr_entry *entry;
914         size_t name_len;
915         int i, cmp = 1;
916
917         if (name == NULL)
918                 return -EINVAL;
919
920         name_len = strlen(name);
921         entry = xs->here;
922         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
923                 cmp = name_index - ocfs2_xattr_get_type(entry);
924                 if (!cmp)
925                         cmp = name_len - entry->xe_name_len;
926                 if (!cmp)
927                         cmp = memcmp(name, (xs->base +
928                                      le16_to_cpu(entry->xe_name_offset)),
929                                      name_len);
930                 if (cmp == 0)
931                         break;
932                 entry += 1;
933         }
934         xs->here = entry;
935
936         return cmp ? -ENODATA : 0;
937 }
938
939 static int ocfs2_xattr_get_value_outside(struct inode *inode,
940                                          struct ocfs2_xattr_value_root *xv,
941                                          void *buffer,
942                                          size_t len)
943 {
944         u32 cpos, p_cluster, num_clusters, bpc, clusters;
945         u64 blkno;
946         int i, ret = 0;
947         size_t cplen, blocksize;
948         struct buffer_head *bh = NULL;
949         struct ocfs2_extent_list *el;
950
951         el = &xv->xr_list;
952         clusters = le32_to_cpu(xv->xr_clusters);
953         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
954         blocksize = inode->i_sb->s_blocksize;
955
956         cpos = 0;
957         while (cpos < clusters) {
958                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
959                                                &num_clusters, el);
960                 if (ret) {
961                         mlog_errno(ret);
962                         goto out;
963                 }
964
965                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
966                 /* Copy ocfs2_xattr_value */
967                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
968                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
969                         if (ret) {
970                                 mlog_errno(ret);
971                                 goto out;
972                         }
973
974                         cplen = len >= blocksize ? blocksize : len;
975                         memcpy(buffer, bh->b_data, cplen);
976                         len -= cplen;
977                         buffer += cplen;
978
979                         brelse(bh);
980                         bh = NULL;
981                         if (len == 0)
982                                 break;
983                 }
984                 cpos += num_clusters;
985         }
986 out:
987         return ret;
988 }
989
990 static int ocfs2_xattr_ibody_get(struct inode *inode,
991                                  int name_index,
992                                  const char *name,
993                                  void *buffer,
994                                  size_t buffer_size,
995                                  struct ocfs2_xattr_search *xs)
996 {
997         struct ocfs2_inode_info *oi = OCFS2_I(inode);
998         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
999         struct ocfs2_xattr_value_root *xv;
1000         size_t size;
1001         int ret = 0;
1002
1003         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1004                 return -ENODATA;
1005
1006         xs->end = (void *)di + inode->i_sb->s_blocksize;
1007         xs->header = (struct ocfs2_xattr_header *)
1008                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1009         xs->base = (void *)xs->header;
1010         xs->here = xs->header->xh_entries;
1011
1012         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1013         if (ret)
1014                 return ret;
1015         size = le64_to_cpu(xs->here->xe_value_size);
1016         if (buffer) {
1017                 if (size > buffer_size)
1018                         return -ERANGE;
1019                 if (ocfs2_xattr_is_local(xs->here)) {
1020                         memcpy(buffer, (void *)xs->base +
1021                                le16_to_cpu(xs->here->xe_name_offset) +
1022                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1023                 } else {
1024                         xv = (struct ocfs2_xattr_value_root *)
1025                                 (xs->base + le16_to_cpu(
1026                                  xs->here->xe_name_offset) +
1027                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1028                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1029                                                             buffer, size);
1030                         if (ret < 0) {
1031                                 mlog_errno(ret);
1032                                 return ret;
1033                         }
1034                 }
1035         }
1036
1037         return size;
1038 }
1039
1040 static int ocfs2_xattr_block_get(struct inode *inode,
1041                                  int name_index,
1042                                  const char *name,
1043                                  void *buffer,
1044                                  size_t buffer_size,
1045                                  struct ocfs2_xattr_search *xs)
1046 {
1047         struct ocfs2_xattr_block *xb;
1048         struct ocfs2_xattr_value_root *xv;
1049         size_t size;
1050         int ret = -ENODATA, name_offset, name_len, block_off, i;
1051
1052         xs->bucket = ocfs2_xattr_bucket_new(inode);
1053         if (!xs->bucket) {
1054                 ret = -ENOMEM;
1055                 mlog_errno(ret);
1056                 goto cleanup;
1057         }
1058
1059         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1060         if (ret) {
1061                 mlog_errno(ret);
1062                 goto cleanup;
1063         }
1064
1065         if (xs->not_found) {
1066                 ret = -ENODATA;
1067                 goto cleanup;
1068         }
1069
1070         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1071         size = le64_to_cpu(xs->here->xe_value_size);
1072         if (buffer) {
1073                 ret = -ERANGE;
1074                 if (size > buffer_size)
1075                         goto cleanup;
1076
1077                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1078                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1079                 i = xs->here - xs->header->xh_entries;
1080
1081                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1082                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1083                                                                 bucket_xh(xs->bucket),
1084                                                                 i,
1085                                                                 &block_off,
1086                                                                 &name_offset);
1087                         xs->base = bucket_block(xs->bucket, block_off);
1088                 }
1089                 if (ocfs2_xattr_is_local(xs->here)) {
1090                         memcpy(buffer, (void *)xs->base +
1091                                name_offset + name_len, size);
1092                 } else {
1093                         xv = (struct ocfs2_xattr_value_root *)
1094                                 (xs->base + name_offset + name_len);
1095                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1096                                                             buffer, size);
1097                         if (ret < 0) {
1098                                 mlog_errno(ret);
1099                                 goto cleanup;
1100                         }
1101                 }
1102         }
1103         ret = size;
1104 cleanup:
1105         ocfs2_xattr_bucket_free(xs->bucket);
1106
1107         brelse(xs->xattr_bh);
1108         xs->xattr_bh = NULL;
1109         return ret;
1110 }
1111
1112 int ocfs2_xattr_get_nolock(struct inode *inode,
1113                            struct buffer_head *di_bh,
1114                            int name_index,
1115                            const char *name,
1116                            void *buffer,
1117                            size_t buffer_size)
1118 {
1119         int ret;
1120         struct ocfs2_dinode *di = NULL;
1121         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1122         struct ocfs2_xattr_search xis = {
1123                 .not_found = -ENODATA,
1124         };
1125         struct ocfs2_xattr_search xbs = {
1126                 .not_found = -ENODATA,
1127         };
1128
1129         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1130                 return -EOPNOTSUPP;
1131
1132         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1133                 ret = -ENODATA;
1134
1135         xis.inode_bh = xbs.inode_bh = di_bh;
1136         di = (struct ocfs2_dinode *)di_bh->b_data;
1137
1138         down_read(&oi->ip_xattr_sem);
1139         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1140                                     buffer_size, &xis);
1141         if (ret == -ENODATA && di->i_xattr_loc)
1142                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1143                                             buffer_size, &xbs);
1144         up_read(&oi->ip_xattr_sem);
1145
1146         return ret;
1147 }
1148
1149 /* ocfs2_xattr_get()
1150  *
1151  * Copy an extended attribute into the buffer provided.
1152  * Buffer is NULL to compute the size of buffer required.
1153  */
1154 static int ocfs2_xattr_get(struct inode *inode,
1155                            int name_index,
1156                            const char *name,
1157                            void *buffer,
1158                            size_t buffer_size)
1159 {
1160         int ret;
1161         struct buffer_head *di_bh = NULL;
1162
1163         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1164         if (ret < 0) {
1165                 mlog_errno(ret);
1166                 return ret;
1167         }
1168         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1169                                      name, buffer, buffer_size);
1170
1171         ocfs2_inode_unlock(inode, 0);
1172
1173         brelse(di_bh);
1174
1175         return ret;
1176 }
1177
1178 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1179                                            handle_t *handle,
1180                                            struct ocfs2_xattr_value_root *xv,
1181                                            const void *value,
1182                                            int value_len)
1183 {
1184         int ret = 0, i, cp_len;
1185         u16 blocksize = inode->i_sb->s_blocksize;
1186         u32 p_cluster, num_clusters;
1187         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1188         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1189         u64 blkno;
1190         struct buffer_head *bh = NULL;
1191
1192         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1193
1194         while (cpos < clusters) {
1195                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1196                                                &num_clusters, &xv->xr_list);
1197                 if (ret) {
1198                         mlog_errno(ret);
1199                         goto out;
1200                 }
1201
1202                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1203
1204                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1205                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1206                         if (ret) {
1207                                 mlog_errno(ret);
1208                                 goto out;
1209                         }
1210
1211                         ret = ocfs2_journal_access(handle,
1212                                                    inode,
1213                                                    bh,
1214                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1215                         if (ret < 0) {
1216                                 mlog_errno(ret);
1217                                 goto out;
1218                         }
1219
1220                         cp_len = value_len > blocksize ? blocksize : value_len;
1221                         memcpy(bh->b_data, value, cp_len);
1222                         value_len -= cp_len;
1223                         value += cp_len;
1224                         if (cp_len < blocksize)
1225                                 memset(bh->b_data + cp_len, 0,
1226                                        blocksize - cp_len);
1227
1228                         ret = ocfs2_journal_dirty(handle, bh);
1229                         if (ret < 0) {
1230                                 mlog_errno(ret);
1231                                 goto out;
1232                         }
1233                         brelse(bh);
1234                         bh = NULL;
1235
1236                         /*
1237                          * XXX: do we need to empty all the following
1238                          * blocks in this cluster?
1239                          */
1240                         if (!value_len)
1241                                 break;
1242                 }
1243                 cpos += num_clusters;
1244         }
1245 out:
1246         brelse(bh);
1247
1248         return ret;
1249 }
1250
1251 static int ocfs2_xattr_cleanup(struct inode *inode,
1252                                handle_t *handle,
1253                                struct ocfs2_xattr_info *xi,
1254                                struct ocfs2_xattr_search *xs,
1255                                struct ocfs2_xattr_value_buf *vb,
1256                                size_t offs)
1257 {
1258         int ret = 0;
1259         size_t name_len = strlen(xi->name);
1260         void *val = xs->base + offs;
1261         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1262
1263         ret = vb->vb_access(handle, inode, vb->vb_bh,
1264                             OCFS2_JOURNAL_ACCESS_WRITE);
1265         if (ret) {
1266                 mlog_errno(ret);
1267                 goto out;
1268         }
1269         /* Decrease xattr count */
1270         le16_add_cpu(&xs->header->xh_count, -1);
1271         /* Remove the xattr entry and tree root which has already be set*/
1272         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1273         memset(val, 0, size);
1274
1275         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1276         if (ret < 0)
1277                 mlog_errno(ret);
1278 out:
1279         return ret;
1280 }
1281
1282 static int ocfs2_xattr_update_entry(struct inode *inode,
1283                                     handle_t *handle,
1284                                     struct ocfs2_xattr_info *xi,
1285                                     struct ocfs2_xattr_search *xs,
1286                                     struct ocfs2_xattr_value_buf *vb,
1287                                     size_t offs)
1288 {
1289         int ret;
1290
1291         ret = vb->vb_access(handle, inode, vb->vb_bh,
1292                             OCFS2_JOURNAL_ACCESS_WRITE);
1293         if (ret) {
1294                 mlog_errno(ret);
1295                 goto out;
1296         }
1297
1298         xs->here->xe_name_offset = cpu_to_le16(offs);
1299         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1300         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1301                 ocfs2_xattr_set_local(xs->here, 1);
1302         else
1303                 ocfs2_xattr_set_local(xs->here, 0);
1304         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1305
1306         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1307         if (ret < 0)
1308                 mlog_errno(ret);
1309 out:
1310         return ret;
1311 }
1312
1313 /*
1314  * ocfs2_xattr_set_value_outside()
1315  *
1316  * Set large size value in B tree.
1317  */
1318 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1319                                          struct ocfs2_xattr_info *xi,
1320                                          struct ocfs2_xattr_search *xs,
1321                                          struct ocfs2_xattr_set_ctxt *ctxt,
1322                                          struct ocfs2_xattr_value_buf *vb,
1323                                          size_t offs)
1324 {
1325         size_t name_len = strlen(xi->name);
1326         void *val = xs->base + offs;
1327         struct ocfs2_xattr_value_root *xv = NULL;
1328         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1329         int ret = 0;
1330
1331         memset(val, 0, size);
1332         memcpy(val, xi->name, name_len);
1333         xv = (struct ocfs2_xattr_value_root *)
1334                 (val + OCFS2_XATTR_SIZE(name_len));
1335         xv->xr_clusters = 0;
1336         xv->xr_last_eb_blk = 0;
1337         xv->xr_list.l_tree_depth = 0;
1338         xv->xr_list.l_count = cpu_to_le16(1);
1339         xv->xr_list.l_next_free_rec = 0;
1340         vb->vb_xv = xv;
1341
1342         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1343         if (ret < 0) {
1344                 mlog_errno(ret);
1345                 return ret;
1346         }
1347         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1348         if (ret < 0) {
1349                 mlog_errno(ret);
1350                 return ret;
1351         }
1352         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1353                                               xi->value, xi->value_len);
1354         if (ret < 0)
1355                 mlog_errno(ret);
1356
1357         return ret;
1358 }
1359
1360 /*
1361  * ocfs2_xattr_set_entry_local()
1362  *
1363  * Set, replace or remove extended attribute in local.
1364  */
1365 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1366                                         struct ocfs2_xattr_info *xi,
1367                                         struct ocfs2_xattr_search *xs,
1368                                         struct ocfs2_xattr_entry *last,
1369                                         size_t min_offs)
1370 {
1371         size_t name_len = strlen(xi->name);
1372         int i;
1373
1374         if (xi->value && xs->not_found) {
1375                 /* Insert the new xattr entry. */
1376                 le16_add_cpu(&xs->header->xh_count, 1);
1377                 ocfs2_xattr_set_type(last, xi->name_index);
1378                 ocfs2_xattr_set_local(last, 1);
1379                 last->xe_name_len = name_len;
1380         } else {
1381                 void *first_val;
1382                 void *val;
1383                 size_t offs, size;
1384
1385                 first_val = xs->base + min_offs;
1386                 offs = le16_to_cpu(xs->here->xe_name_offset);
1387                 val = xs->base + offs;
1388
1389                 if (le64_to_cpu(xs->here->xe_value_size) >
1390                     OCFS2_XATTR_INLINE_SIZE)
1391                         size = OCFS2_XATTR_SIZE(name_len) +
1392                                 OCFS2_XATTR_ROOT_SIZE;
1393                 else
1394                         size = OCFS2_XATTR_SIZE(name_len) +
1395                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1396
1397                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1398                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1399                         /* The old and the new value have the
1400                            same size. Just replace the value. */
1401                         ocfs2_xattr_set_local(xs->here, 1);
1402                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1403                         /* Clear value bytes. */
1404                         memset(val + OCFS2_XATTR_SIZE(name_len),
1405                                0,
1406                                OCFS2_XATTR_SIZE(xi->value_len));
1407                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1408                                xi->value,
1409                                xi->value_len);
1410                         return;
1411                 }
1412                 /* Remove the old name+value. */
1413                 memmove(first_val + size, first_val, val - first_val);
1414                 memset(first_val, 0, size);
1415                 xs->here->xe_name_hash = 0;
1416                 xs->here->xe_name_offset = 0;
1417                 ocfs2_xattr_set_local(xs->here, 1);
1418                 xs->here->xe_value_size = 0;
1419
1420                 min_offs += size;
1421
1422                 /* Adjust all value offsets. */
1423                 last = xs->header->xh_entries;
1424                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1425                         size_t o = le16_to_cpu(last->xe_name_offset);
1426
1427                         if (o < offs)
1428                                 last->xe_name_offset = cpu_to_le16(o + size);
1429                         last += 1;
1430                 }
1431
1432                 if (!xi->value) {
1433                         /* Remove the old entry. */
1434                         last -= 1;
1435                         memmove(xs->here, xs->here + 1,
1436                                 (void *)last - (void *)xs->here);
1437                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1438                         le16_add_cpu(&xs->header->xh_count, -1);
1439                 }
1440         }
1441         if (xi->value) {
1442                 /* Insert the new name+value. */
1443                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1444                                 OCFS2_XATTR_SIZE(xi->value_len);
1445                 void *val = xs->base + min_offs - size;
1446
1447                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1448                 memset(val, 0, size);
1449                 memcpy(val, xi->name, name_len);
1450                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1451                        xi->value,
1452                        xi->value_len);
1453                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1454                 ocfs2_xattr_set_local(xs->here, 1);
1455                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1456         }
1457
1458         return;
1459 }
1460
1461 /*
1462  * ocfs2_xattr_set_entry()
1463  *
1464  * Set extended attribute entry into inode or block.
1465  *
1466  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1467  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1468  * then set value in B tree with set_value_outside().
1469  */
1470 static int ocfs2_xattr_set_entry(struct inode *inode,
1471                                  struct ocfs2_xattr_info *xi,
1472                                  struct ocfs2_xattr_search *xs,
1473                                  struct ocfs2_xattr_set_ctxt *ctxt,
1474                                  int flag)
1475 {
1476         struct ocfs2_xattr_entry *last;
1477         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1478         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1479         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1480         size_t size_l = 0;
1481         handle_t *handle = ctxt->handle;
1482         int free, i, ret;
1483         struct ocfs2_xattr_info xi_l = {
1484                 .name_index = xi->name_index,
1485                 .name = xi->name,
1486                 .value = xi->value,
1487                 .value_len = xi->value_len,
1488         };
1489         struct ocfs2_xattr_value_buf vb = {
1490                 .vb_bh = xs->xattr_bh,
1491                 .vb_access = ocfs2_journal_access_di,
1492         };
1493
1494         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1495                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1496                 vb.vb_access = ocfs2_journal_access_xb;
1497         } else
1498                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1499
1500         /* Compute min_offs, last and free space. */
1501         last = xs->header->xh_entries;
1502
1503         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1504                 size_t offs = le16_to_cpu(last->xe_name_offset);
1505                 if (offs < min_offs)
1506                         min_offs = offs;
1507                 last += 1;
1508         }
1509
1510         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1511         if (free < 0)
1512                 return -EIO;
1513
1514         if (!xs->not_found) {
1515                 size_t size = 0;
1516                 if (ocfs2_xattr_is_local(xs->here))
1517                         size = OCFS2_XATTR_SIZE(name_len) +
1518                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1519                 else
1520                         size = OCFS2_XATTR_SIZE(name_len) +
1521                                 OCFS2_XATTR_ROOT_SIZE;
1522                 free += (size + sizeof(struct ocfs2_xattr_entry));
1523         }
1524         /* Check free space in inode or block */
1525         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1526                 if (free < sizeof(struct ocfs2_xattr_entry) +
1527                            OCFS2_XATTR_SIZE(name_len) +
1528                            OCFS2_XATTR_ROOT_SIZE) {
1529                         ret = -ENOSPC;
1530                         goto out;
1531                 }
1532                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1533                 xi_l.value = (void *)&def_xv;
1534                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1535         } else if (xi->value) {
1536                 if (free < sizeof(struct ocfs2_xattr_entry) +
1537                            OCFS2_XATTR_SIZE(name_len) +
1538                            OCFS2_XATTR_SIZE(xi->value_len)) {
1539                         ret = -ENOSPC;
1540                         goto out;
1541                 }
1542         }
1543
1544         if (!xs->not_found) {
1545                 /* For existing extended attribute */
1546                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1547                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1548                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1549                 void *val = xs->base + offs;
1550
1551                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1552                         /* Replace existing local xattr with tree root */
1553                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1554                                                             ctxt, &vb, offs);
1555                         if (ret < 0)
1556                                 mlog_errno(ret);
1557                         goto out;
1558                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1559                         /* For existing xattr which has value outside */
1560                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1561                                 (val + OCFS2_XATTR_SIZE(name_len));
1562
1563                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1564                                 /*
1565                                  * If new value need set outside also,
1566                                  * first truncate old value to new value,
1567                                  * then set new value with set_value_outside().
1568                                  */
1569                                 ret = ocfs2_xattr_value_truncate(inode,
1570                                                                  &vb,
1571                                                                  xi->value_len,
1572                                                                  ctxt);
1573                                 if (ret < 0) {
1574                                         mlog_errno(ret);
1575                                         goto out;
1576                                 }
1577
1578                                 ret = ocfs2_xattr_update_entry(inode,
1579                                                                handle,
1580                                                                xi,
1581                                                                xs,
1582                                                                &vb,
1583                                                                offs);
1584                                 if (ret < 0) {
1585                                         mlog_errno(ret);
1586                                         goto out;
1587                                 }
1588
1589                                 ret = __ocfs2_xattr_set_value_outside(inode,
1590                                                                 handle,
1591                                                                 vb.vb_xv,
1592                                                                 xi->value,
1593                                                                 xi->value_len);
1594                                 if (ret < 0)
1595                                         mlog_errno(ret);
1596                                 goto out;
1597                         } else {
1598                                 /*
1599                                  * If new value need set in local,
1600                                  * just trucate old value to zero.
1601                                  */
1602                                  ret = ocfs2_xattr_value_truncate(inode,
1603                                                                   &vb,
1604                                                                   0,
1605                                                                   ctxt);
1606                                 if (ret < 0)
1607                                         mlog_errno(ret);
1608                         }
1609                 }
1610         }
1611
1612         ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1613                                       OCFS2_JOURNAL_ACCESS_WRITE);
1614         if (ret) {
1615                 mlog_errno(ret);
1616                 goto out;
1617         }
1618
1619         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1620                 ret = vb.vb_access(handle, inode, vb.vb_bh,
1621                                    OCFS2_JOURNAL_ACCESS_WRITE);
1622                 if (ret) {
1623                         mlog_errno(ret);
1624                         goto out;
1625                 }
1626         }
1627
1628         /*
1629          * Set value in local, include set tree root in local.
1630          * This is the first step for value size >INLINE_SIZE.
1631          */
1632         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1633
1634         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1635                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1636                 if (ret < 0) {
1637                         mlog_errno(ret);
1638                         goto out;
1639                 }
1640         }
1641
1642         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1643             (flag & OCFS2_INLINE_XATTR_FL)) {
1644                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1645                 unsigned int xattrsize = osb->s_xattr_inline_size;
1646
1647                 /*
1648                  * Adjust extent record count or inline data size
1649                  * to reserve space for extended attribute.
1650                  */
1651                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1652                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1653                         le16_add_cpu(&idata->id_count, -xattrsize);
1654                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1655                         struct ocfs2_extent_list *el = &di->id2.i_list;
1656                         le16_add_cpu(&el->l_count, -(xattrsize /
1657                                         sizeof(struct ocfs2_extent_rec)));
1658                 }
1659                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1660         }
1661         /* Update xattr flag */
1662         spin_lock(&oi->ip_lock);
1663         oi->ip_dyn_features |= flag;
1664         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1665         spin_unlock(&oi->ip_lock);
1666
1667         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1668         if (ret < 0)
1669                 mlog_errno(ret);
1670
1671         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1672                 /*
1673                  * Set value outside in B tree.
1674                  * This is the second step for value size > INLINE_SIZE.
1675                  */
1676                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1677                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1678                                                     &vb, offs);
1679                 if (ret < 0) {
1680                         int ret2;
1681
1682                         mlog_errno(ret);
1683                         /*
1684                          * If set value outside failed, we have to clean
1685                          * the junk tree root we have already set in local.
1686                          */
1687                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1688                                                    xi, xs, &vb, offs);
1689                         if (ret2 < 0)
1690                                 mlog_errno(ret2);
1691                 }
1692         }
1693 out:
1694         return ret;
1695 }
1696
1697 static int ocfs2_remove_value_outside(struct inode*inode,
1698                                       struct ocfs2_xattr_value_buf *vb,
1699                                       struct ocfs2_xattr_header *header)
1700 {
1701         int ret = 0, i;
1702         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1703         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1704
1705         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1706
1707         ctxt.handle = ocfs2_start_trans(osb,
1708                                         ocfs2_remove_extent_credits(osb->sb));
1709         if (IS_ERR(ctxt.handle)) {
1710                 ret = PTR_ERR(ctxt.handle);
1711                 mlog_errno(ret);
1712                 goto out;
1713         }
1714
1715         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1716                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1717
1718                 if (!ocfs2_xattr_is_local(entry)) {
1719                         void *val;
1720
1721                         val = (void *)header +
1722                                 le16_to_cpu(entry->xe_name_offset);
1723                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1724                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1725                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1726                         if (ret < 0) {
1727                                 mlog_errno(ret);
1728                                 break;
1729                         }
1730                 }
1731         }
1732
1733         ocfs2_commit_trans(osb, ctxt.handle);
1734         ocfs2_schedule_truncate_log_flush(osb, 1);
1735         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1736 out:
1737         return ret;
1738 }
1739
1740 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1741                                     struct buffer_head *di_bh)
1742 {
1743
1744         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1745         struct ocfs2_xattr_header *header;
1746         int ret;
1747         struct ocfs2_xattr_value_buf vb = {
1748                 .vb_bh = di_bh,
1749                 .vb_access = ocfs2_journal_access_di,
1750         };
1751
1752         header = (struct ocfs2_xattr_header *)
1753                  ((void *)di + inode->i_sb->s_blocksize -
1754                  le16_to_cpu(di->i_xattr_inline_size));
1755
1756         ret = ocfs2_remove_value_outside(inode, &vb, header);
1757
1758         return ret;
1759 }
1760
1761 static int ocfs2_xattr_block_remove(struct inode *inode,
1762                                     struct buffer_head *blk_bh)
1763 {
1764         struct ocfs2_xattr_block *xb;
1765         int ret = 0;
1766         struct ocfs2_xattr_value_buf vb = {
1767                 .vb_bh = blk_bh,
1768                 .vb_access = ocfs2_journal_access_xb,
1769         };
1770
1771         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1772         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1773                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1774                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1775         } else
1776                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1777
1778         return ret;
1779 }
1780
1781 static int ocfs2_xattr_free_block(struct inode *inode,
1782                                   u64 block)
1783 {
1784         struct inode *xb_alloc_inode;
1785         struct buffer_head *xb_alloc_bh = NULL;
1786         struct buffer_head *blk_bh = NULL;
1787         struct ocfs2_xattr_block *xb;
1788         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1789         handle_t *handle;
1790         int ret = 0;
1791         u64 blk, bg_blkno;
1792         u16 bit;
1793
1794         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1795         if (ret < 0) {
1796                 mlog_errno(ret);
1797                 goto out;
1798         }
1799
1800         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1801         if (ret < 0) {
1802                 mlog_errno(ret);
1803                 goto out;
1804         }
1805
1806         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1807         blk = le64_to_cpu(xb->xb_blkno);
1808         bit = le16_to_cpu(xb->xb_suballoc_bit);
1809         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1810
1811         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1812                                 EXTENT_ALLOC_SYSTEM_INODE,
1813                                 le16_to_cpu(xb->xb_suballoc_slot));
1814         if (!xb_alloc_inode) {
1815                 ret = -ENOMEM;
1816                 mlog_errno(ret);
1817                 goto out;
1818         }
1819         mutex_lock(&xb_alloc_inode->i_mutex);
1820
1821         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1822         if (ret < 0) {
1823                 mlog_errno(ret);
1824                 goto out_mutex;
1825         }
1826
1827         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1828         if (IS_ERR(handle)) {
1829                 ret = PTR_ERR(handle);
1830                 mlog_errno(ret);
1831                 goto out_unlock;
1832         }
1833
1834         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1835                                        bit, bg_blkno, 1);
1836         if (ret < 0)
1837                 mlog_errno(ret);
1838
1839         ocfs2_commit_trans(osb, handle);
1840 out_unlock:
1841         ocfs2_inode_unlock(xb_alloc_inode, 1);
1842         brelse(xb_alloc_bh);
1843 out_mutex:
1844         mutex_unlock(&xb_alloc_inode->i_mutex);
1845         iput(xb_alloc_inode);
1846 out:
1847         brelse(blk_bh);
1848         return ret;
1849 }
1850
1851 /*
1852  * ocfs2_xattr_remove()
1853  *
1854  * Free extended attribute resources associated with this inode.
1855  */
1856 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1857 {
1858         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1859         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1860         handle_t *handle;
1861         int ret;
1862
1863         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1864                 return 0;
1865
1866         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1867                 return 0;
1868
1869         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1870                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1871                 if (ret < 0) {
1872                         mlog_errno(ret);
1873                         goto out;
1874                 }
1875         }
1876
1877         if (di->i_xattr_loc) {
1878                 ret = ocfs2_xattr_free_block(inode,
1879                                              le64_to_cpu(di->i_xattr_loc));
1880                 if (ret < 0) {
1881                         mlog_errno(ret);
1882                         goto out;
1883                 }
1884         }
1885
1886         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1887                                    OCFS2_INODE_UPDATE_CREDITS);
1888         if (IS_ERR(handle)) {
1889                 ret = PTR_ERR(handle);
1890                 mlog_errno(ret);
1891                 goto out;
1892         }
1893         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1894                                       OCFS2_JOURNAL_ACCESS_WRITE);
1895         if (ret) {
1896                 mlog_errno(ret);
1897                 goto out_commit;
1898         }
1899
1900         di->i_xattr_loc = 0;
1901
1902         spin_lock(&oi->ip_lock);
1903         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1904         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1905         spin_unlock(&oi->ip_lock);
1906
1907         ret = ocfs2_journal_dirty(handle, di_bh);
1908         if (ret < 0)
1909                 mlog_errno(ret);
1910 out_commit:
1911         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1912 out:
1913         return ret;
1914 }
1915
1916 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1917                                         struct ocfs2_dinode *di)
1918 {
1919         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1920         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1921         int free;
1922
1923         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1924                 return 0;
1925
1926         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1927                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1928                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1929         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1930                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1931                         le64_to_cpu(di->i_size);
1932         } else {
1933                 struct ocfs2_extent_list *el = &di->id2.i_list;
1934                 free = (le16_to_cpu(el->l_count) -
1935                         le16_to_cpu(el->l_next_free_rec)) *
1936                         sizeof(struct ocfs2_extent_rec);
1937         }
1938         if (free >= xattrsize)
1939                 return 1;
1940
1941         return 0;
1942 }
1943
1944 /*
1945  * ocfs2_xattr_ibody_find()
1946  *
1947  * Find extended attribute in inode block and
1948  * fill search info into struct ocfs2_xattr_search.
1949  */
1950 static int ocfs2_xattr_ibody_find(struct inode *inode,
1951                                   int name_index,
1952                                   const char *name,
1953                                   struct ocfs2_xattr_search *xs)
1954 {
1955         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1956         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1957         int ret;
1958         int has_space = 0;
1959
1960         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1961                 return 0;
1962
1963         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1964                 down_read(&oi->ip_alloc_sem);
1965                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1966                 up_read(&oi->ip_alloc_sem);
1967                 if (!has_space)
1968                         return 0;
1969         }
1970
1971         xs->xattr_bh = xs->inode_bh;
1972         xs->end = (void *)di + inode->i_sb->s_blocksize;
1973         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1974                 xs->header = (struct ocfs2_xattr_header *)
1975                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1976         else
1977                 xs->header = (struct ocfs2_xattr_header *)
1978                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1979         xs->base = (void *)xs->header;
1980         xs->here = xs->header->xh_entries;
1981
1982         /* Find the named attribute. */
1983         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1984                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1985                 if (ret && ret != -ENODATA)
1986                         return ret;
1987                 xs->not_found = ret;
1988         }
1989
1990         return 0;
1991 }
1992
1993 /*
1994  * ocfs2_xattr_ibody_set()
1995  *
1996  * Set, replace or remove an extended attribute into inode block.
1997  *
1998  */
1999 static int ocfs2_xattr_ibody_set(struct inode *inode,
2000                                  struct ocfs2_xattr_info *xi,
2001                                  struct ocfs2_xattr_search *xs,
2002                                  struct ocfs2_xattr_set_ctxt *ctxt)
2003 {
2004         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2005         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2006         int ret;
2007
2008         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2009                 return -ENOSPC;
2010
2011         down_write(&oi->ip_alloc_sem);
2012         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2013                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2014                         ret = -ENOSPC;
2015                         goto out;
2016                 }
2017         }
2018
2019         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2020                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2021 out:
2022         up_write(&oi->ip_alloc_sem);
2023
2024         return ret;
2025 }
2026
2027 /*
2028  * ocfs2_xattr_block_find()
2029  *
2030  * Find extended attribute in external block and
2031  * fill search info into struct ocfs2_xattr_search.
2032  */
2033 static int ocfs2_xattr_block_find(struct inode *inode,
2034                                   int name_index,
2035                                   const char *name,
2036                                   struct ocfs2_xattr_search *xs)
2037 {
2038         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2039         struct buffer_head *blk_bh = NULL;
2040         struct ocfs2_xattr_block *xb;
2041         int ret = 0;
2042
2043         if (!di->i_xattr_loc)
2044                 return ret;
2045
2046         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2047                                      &blk_bh);
2048         if (ret < 0) {
2049                 mlog_errno(ret);
2050                 return ret;
2051         }
2052
2053         xs->xattr_bh = blk_bh;
2054         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2055
2056         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2057                 xs->header = &xb->xb_attrs.xb_header;
2058                 xs->base = (void *)xs->header;
2059                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2060                 xs->here = xs->header->xh_entries;
2061
2062                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2063         } else
2064                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2065                                                    name_index,
2066                                                    name, xs);
2067
2068         if (ret && ret != -ENODATA) {
2069                 xs->xattr_bh = NULL;
2070                 goto cleanup;
2071         }
2072         xs->not_found = ret;
2073         return 0;
2074 cleanup:
2075         brelse(blk_bh);
2076
2077         return ret;
2078 }
2079
2080 /*
2081  * ocfs2_xattr_block_set()
2082  *
2083  * Set, replace or remove an extended attribute into external block.
2084  *
2085  */
2086 static int ocfs2_xattr_block_set(struct inode *inode,
2087                                  struct ocfs2_xattr_info *xi,
2088                                  struct ocfs2_xattr_search *xs,
2089                                  struct ocfs2_xattr_set_ctxt *ctxt)
2090 {
2091         struct buffer_head *new_bh = NULL;
2092         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2093         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2094         handle_t *handle = ctxt->handle;
2095         struct ocfs2_xattr_block *xblk = NULL;
2096         u16 suballoc_bit_start;
2097         u32 num_got;
2098         u64 first_blkno;
2099         int ret;
2100
2101         if (!xs->xattr_bh) {
2102                 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2103                                               OCFS2_JOURNAL_ACCESS_CREATE);
2104                 if (ret < 0) {
2105                         mlog_errno(ret);
2106                         goto end;
2107                 }
2108
2109                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2110                                            &suballoc_bit_start, &num_got,
2111                                            &first_blkno);
2112                 if (ret < 0) {
2113                         mlog_errno(ret);
2114                         goto end;
2115                 }
2116
2117                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2118                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2119
2120                 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2121                                               OCFS2_JOURNAL_ACCESS_CREATE);
2122                 if (ret < 0) {
2123                         mlog_errno(ret);
2124                         goto end;
2125                 }
2126
2127                 /* Initialize ocfs2_xattr_block */
2128                 xs->xattr_bh = new_bh;
2129                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2130                 memset(xblk, 0, inode->i_sb->s_blocksize);
2131                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2132                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2133                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2134                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2135                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2136
2137                 xs->header = &xblk->xb_attrs.xb_header;
2138                 xs->base = (void *)xs->header;
2139                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2140                 xs->here = xs->header->xh_entries;
2141
2142                 ret = ocfs2_journal_dirty(handle, new_bh);
2143                 if (ret < 0) {
2144                         mlog_errno(ret);
2145                         goto end;
2146                 }
2147                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2148                 ocfs2_journal_dirty(handle, xs->inode_bh);
2149         } else
2150                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2151
2152         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2153                 /* Set extended attribute into external block */
2154                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2155                                             OCFS2_HAS_XATTR_FL);
2156                 if (!ret || ret != -ENOSPC)
2157                         goto end;
2158
2159                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2160                 if (ret)
2161                         goto end;
2162         }
2163
2164         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2165
2166 end:
2167
2168         return ret;
2169 }
2170
2171 /* Check whether the new xattr can be inserted into the inode. */
2172 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2173                                        struct ocfs2_xattr_info *xi,
2174                                        struct ocfs2_xattr_search *xs)
2175 {
2176         u64 value_size;
2177         struct ocfs2_xattr_entry *last;
2178         int free, i;
2179         size_t min_offs = xs->end - xs->base;
2180
2181         if (!xs->header)
2182                 return 0;
2183
2184         last = xs->header->xh_entries;
2185
2186         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2187                 size_t offs = le16_to_cpu(last->xe_name_offset);
2188                 if (offs < min_offs)
2189                         min_offs = offs;
2190                 last += 1;
2191         }
2192
2193         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2194         if (free < 0)
2195                 return 0;
2196
2197         BUG_ON(!xs->not_found);
2198
2199         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2200                 value_size = OCFS2_XATTR_ROOT_SIZE;
2201         else
2202                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2203
2204         if (free >= sizeof(struct ocfs2_xattr_entry) +
2205                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2206                 return 1;
2207
2208         return 0;
2209 }
2210
2211 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2212                                      struct ocfs2_dinode *di,
2213                                      struct ocfs2_xattr_info *xi,
2214                                      struct ocfs2_xattr_search *xis,
2215                                      struct ocfs2_xattr_search *xbs,
2216                                      int *clusters_need,
2217                                      int *meta_need,
2218                                      int *credits_need)
2219 {
2220         int ret = 0, old_in_xb = 0;
2221         int clusters_add = 0, meta_add = 0, credits = 0;
2222         struct buffer_head *bh = NULL;
2223         struct ocfs2_xattr_block *xb = NULL;
2224         struct ocfs2_xattr_entry *xe = NULL;
2225         struct ocfs2_xattr_value_root *xv = NULL;
2226         char *base = NULL;
2227         int name_offset, name_len = 0;
2228         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2229                                                     xi->value_len);
2230         u64 value_size;
2231
2232         /*
2233          * Calculate the clusters we need to write.
2234          * No matter whether we replace an old one or add a new one,
2235          * we need this for writing.
2236          */
2237         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2238                 credits += new_clusters *
2239                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2240
2241         if (xis->not_found && xbs->not_found) {
2242                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2243
2244                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2245                         clusters_add += new_clusters;
2246                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2247                                                         &def_xv.xv.xr_list,
2248                                                         new_clusters);
2249                 }
2250
2251                 goto meta_guess;
2252         }
2253
2254         if (!xis->not_found) {
2255                 xe = xis->here;
2256                 name_offset = le16_to_cpu(xe->xe_name_offset);
2257                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2258                 base = xis->base;
2259                 credits += OCFS2_INODE_UPDATE_CREDITS;
2260         } else {
2261                 int i, block_off = 0;
2262                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2263                 xe = xbs->here;
2264                 name_offset = le16_to_cpu(xe->xe_name_offset);
2265                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2266                 i = xbs->here - xbs->header->xh_entries;
2267                 old_in_xb = 1;
2268
2269                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2270                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2271                                                         bucket_xh(xbs->bucket),
2272                                                         i, &block_off,
2273                                                         &name_offset);
2274                         base = bucket_block(xbs->bucket, block_off);
2275                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2276                 } else {
2277                         base = xbs->base;
2278                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2279                 }
2280         }
2281
2282         /*
2283          * delete a xattr doesn't need metadata and cluster allocation.
2284          * so just calculate the credits and return.
2285          *
2286          * The credits for removing the value tree will be extended
2287          * by ocfs2_remove_extent itself.
2288          */
2289         if (!xi->value) {
2290                 if (!ocfs2_xattr_is_local(xe))
2291                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2292
2293                 goto out;
2294         }
2295
2296         /* do cluster allocation guess first. */
2297         value_size = le64_to_cpu(xe->xe_value_size);
2298
2299         if (old_in_xb) {
2300                 /*
2301                  * In xattr set, we always try to set the xe in inode first,
2302                  * so if it can be inserted into inode successfully, the old
2303                  * one will be removed from the xattr block, and this xattr
2304                  * will be inserted into inode as a new xattr in inode.
2305                  */
2306                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2307                         clusters_add += new_clusters;
2308                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2309                                     OCFS2_INODE_UPDATE_CREDITS;
2310                         if (!ocfs2_xattr_is_local(xe))
2311                                 credits += ocfs2_calc_extend_credits(
2312                                                         inode->i_sb,
2313                                                         &def_xv.xv.xr_list,
2314                                                         new_clusters);
2315                         goto out;
2316                 }
2317         }
2318
2319         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2320                 /* the new values will be stored outside. */
2321                 u32 old_clusters = 0;
2322
2323                 if (!ocfs2_xattr_is_local(xe)) {
2324                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2325                                                                  value_size);
2326                         xv = (struct ocfs2_xattr_value_root *)
2327                              (base + name_offset + name_len);
2328                         value_size = OCFS2_XATTR_ROOT_SIZE;
2329                 } else
2330                         xv = &def_xv.xv;
2331
2332                 if (old_clusters >= new_clusters) {
2333                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2334                         goto out;
2335                 } else {
2336                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2337                         clusters_add += new_clusters - old_clusters;
2338                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2339                                                              &xv->xr_list,
2340                                                              new_clusters -
2341                                                              old_clusters);
2342                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2343                                 goto out;
2344                 }
2345         } else {
2346                 /*
2347                  * Now the new value will be stored inside. So if the new
2348                  * value is smaller than the size of value root or the old
2349                  * value, we don't need any allocation, otherwise we have
2350                  * to guess metadata allocation.
2351                  */
2352                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2353                     (!ocfs2_xattr_is_local(xe) &&
2354                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2355                         goto out;
2356         }
2357
2358 meta_guess:
2359         /* calculate metadata allocation. */
2360         if (di->i_xattr_loc) {
2361                 if (!xbs->xattr_bh) {
2362                         ret = ocfs2_read_xattr_block(inode,
2363                                                      le64_to_cpu(di->i_xattr_loc),
2364                                                      &bh);
2365                         if (ret) {
2366                                 mlog_errno(ret);
2367                                 goto out;
2368                         }
2369
2370                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2371                 } else
2372                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2373
2374                 /*
2375                  * If there is already an xattr tree, good, we can calculate
2376                  * like other b-trees. Otherwise we may have the chance of
2377                  * create a tree, the credit calculation is borrowed from
2378                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2379                  * new tree will be cluster based, so no meta is needed.
2380                  */
2381                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2382                         struct ocfs2_extent_list *el =
2383                                  &xb->xb_attrs.xb_root.xt_list;
2384                         meta_add += ocfs2_extend_meta_needed(el);
2385                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2386                                                              el, 1);
2387                 } else
2388                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2389
2390                 /*
2391                  * This cluster will be used either for new bucket or for
2392                  * new xattr block.
2393                  * If the cluster size is the same as the bucket size, one
2394                  * more is needed since we may need to extend the bucket
2395                  * also.
2396                  */
2397                 clusters_add += 1;
2398                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2399                 if (OCFS2_XATTR_BUCKET_SIZE ==
2400                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2401                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2402                         clusters_add += 1;
2403                 }
2404         } else {
2405                 meta_add += 1;
2406                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2407         }
2408 out:
2409         if (clusters_need)
2410                 *clusters_need = clusters_add;
2411         if (meta_need)
2412                 *meta_need = meta_add;
2413         if (credits_need)
2414                 *credits_need = credits;
2415         brelse(bh);
2416         return ret;
2417 }
2418
2419 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2420                                      struct ocfs2_dinode *di,
2421                                      struct ocfs2_xattr_info *xi,
2422                                      struct ocfs2_xattr_search *xis,
2423                                      struct ocfs2_xattr_search *xbs,
2424                                      struct ocfs2_xattr_set_ctxt *ctxt,
2425                                      int *credits)
2426 {
2427         int clusters_add, meta_add, ret;
2428         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2429
2430         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2431
2432         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2433
2434         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2435                                         &clusters_add, &meta_add, credits);
2436         if (ret) {
2437                 mlog_errno(ret);
2438                 return ret;
2439         }
2440
2441         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2442              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2443
2444         if (meta_add) {
2445                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2446                                                         &ctxt->meta_ac);
2447                 if (ret) {
2448                         mlog_errno(ret);
2449                         goto out;
2450                 }
2451         }
2452
2453         if (clusters_add) {
2454                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2455                 if (ret)
2456                         mlog_errno(ret);
2457         }
2458 out:
2459         if (ret) {
2460                 if (ctxt->meta_ac) {
2461                         ocfs2_free_alloc_context(ctxt->meta_ac);
2462                         ctxt->meta_ac = NULL;
2463                 }
2464
2465                 /*
2466                  * We cannot have an error and a non null ctxt->data_ac.
2467                  */
2468         }
2469
2470         return ret;
2471 }
2472
2473 static int __ocfs2_xattr_set_handle(struct inode *inode,
2474                                     struct ocfs2_dinode *di,
2475                                     struct ocfs2_xattr_info *xi,
2476                                     struct ocfs2_xattr_search *xis,
2477                                     struct ocfs2_xattr_search *xbs,
2478                                     struct ocfs2_xattr_set_ctxt *ctxt)
2479 {
2480         int ret = 0, credits, old_found;
2481
2482         if (!xi->value) {
2483                 /* Remove existing extended attribute */
2484                 if (!xis->not_found)
2485                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2486                 else if (!xbs->not_found)
2487                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2488         } else {
2489                 /* We always try to set extended attribute into inode first*/
2490                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2491                 if (!ret && !xbs->not_found) {
2492                         /*
2493                          * If succeed and that extended attribute existing in
2494                          * external block, then we will remove it.
2495                          */
2496                         xi->value = NULL;
2497                         xi->value_len = 0;
2498
2499                         old_found = xis->not_found;
2500                         xis->not_found = -ENODATA;
2501                         ret = ocfs2_calc_xattr_set_need(inode,
2502                                                         di,
2503                                                         xi,
2504                                                         xis,
2505                                                         xbs,
2506                                                         NULL,
2507                                                         NULL,
2508                                                         &credits);
2509                         xis->not_found = old_found;
2510                         if (ret) {
2511                                 mlog_errno(ret);
2512                                 goto out;
2513                         }
2514
2515                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2516                                         ctxt->handle->h_buffer_credits);
2517                         if (ret) {
2518                                 mlog_errno(ret);
2519                                 goto out;
2520                         }
2521                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2522                 } else if (ret == -ENOSPC) {
2523                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2524                                 ret = ocfs2_xattr_block_find(inode,
2525                                                              xi->name_index,
2526                                                              xi->name, xbs);
2527                                 if (ret)
2528                                         goto out;
2529
2530                                 old_found = xis->not_found;
2531                                 xis->not_found = -ENODATA;
2532                                 ret = ocfs2_calc_xattr_set_need(inode,
2533                                                                 di,
2534                                                                 xi,
2535                                                                 xis,
2536                                                                 xbs,
2537                                                                 NULL,
2538                                                                 NULL,
2539                                                                 &credits);
2540                                 xis->not_found = old_found;
2541                                 if (ret) {
2542                                         mlog_errno(ret);
2543                                         goto out;
2544                                 }
2545
2546                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2547                                         ctxt->handle->h_buffer_credits);
2548                                 if (ret) {
2549                                         mlog_errno(ret);
2550                                         goto out;
2551                                 }
2552                         }
2553                         /*
2554                          * If no space in inode, we will set extended attribute
2555                          * into external block.
2556                          */
2557                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2558                         if (ret)
2559                                 goto out;
2560                         if (!xis->not_found) {
2561                                 /*
2562                                  * If succeed and that extended attribute
2563                                  * existing in inode, we will remove it.
2564                                  */
2565                                 xi->value = NULL;
2566                                 xi->value_len = 0;
2567                                 xbs->not_found = -ENODATA;
2568                                 ret = ocfs2_calc_xattr_set_need(inode,
2569                                                                 di,
2570                                                                 xi,
2571                                                                 xis,
2572                                                                 xbs,
2573                                                                 NULL,
2574                                                                 NULL,
2575                                                                 &credits);
2576                                 if (ret) {
2577                                         mlog_errno(ret);
2578                                         goto out;
2579                                 }
2580
2581                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2582                                                 ctxt->handle->h_buffer_credits);
2583                                 if (ret) {
2584                                         mlog_errno(ret);
2585                                         goto out;
2586                                 }
2587                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2588                                                             xis, ctxt);
2589                         }
2590                 }
2591         }
2592
2593         if (!ret) {
2594                 /* Update inode ctime. */
2595                 ret = ocfs2_journal_access(ctxt->handle, inode, xis->inode_bh,
2596                                            OCFS2_JOURNAL_ACCESS_WRITE);
2597                 if (ret) {
2598                         mlog_errno(ret);
2599                         goto out;
2600                 }
2601
2602                 inode->i_ctime = CURRENT_TIME;
2603                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2604                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2605                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2606         }
2607 out:
2608         return ret;
2609 }
2610
2611 /*
2612  * This function only called duing creating inode
2613  * for init security/acl xattrs of the new inode.
2614  * The xattrs could be put into ibody or extent block,
2615  * xattr bucket would not be use in this case.
2616  * transanction credits also be reserved in here.
2617  */
2618 int ocfs2_xattr_set_handle(handle_t *handle,
2619                            struct inode *inode,
2620                            struct buffer_head *di_bh,
2621                            int name_index,
2622                            const char *name,
2623                            const void *value,
2624                            size_t value_len,
2625                            int flags,
2626                            struct ocfs2_alloc_context *meta_ac,
2627                            struct ocfs2_alloc_context *data_ac)
2628 {
2629         struct ocfs2_dinode *di;
2630         int ret;
2631
2632         struct ocfs2_xattr_info xi = {
2633                 .name_index = name_index,
2634                 .name = name,
2635                 .value = value,
2636                 .value_len = value_len,
2637         };
2638
2639         struct ocfs2_xattr_search xis = {
2640                 .not_found = -ENODATA,
2641         };
2642
2643         struct ocfs2_xattr_search xbs = {
2644                 .not_found = -ENODATA,
2645         };
2646
2647         struct ocfs2_xattr_set_ctxt ctxt = {
2648                 .handle = handle,
2649                 .meta_ac = meta_ac,
2650                 .data_ac = data_ac,
2651         };
2652
2653         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2654                 return -EOPNOTSUPP;
2655
2656         xis.inode_bh = xbs.inode_bh = di_bh;
2657         di = (struct ocfs2_dinode *)di_bh->b_data;
2658
2659         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2660
2661         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2662         if (ret)
2663                 goto cleanup;
2664         if (xis.not_found) {
2665                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2666                 if (ret)
2667                         goto cleanup;
2668         }
2669
2670         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2671
2672 cleanup:
2673         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2674         brelse(xbs.xattr_bh);
2675
2676         return ret;
2677 }
2678
2679 /*
2680  * ocfs2_xattr_set()
2681  *
2682  * Set, replace or remove an extended attribute for this inode.
2683  * value is NULL to remove an existing extended attribute, else either
2684  * create or replace an extended attribute.
2685  */
2686 int ocfs2_xattr_set(struct inode *inode,
2687                     int name_index,
2688                     const char *name,
2689                     const void *value,
2690                     size_t value_len,
2691                     int flags)
2692 {
2693         struct buffer_head *di_bh = NULL;
2694         struct ocfs2_dinode *di;
2695         int ret, credits;
2696         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2697         struct inode *tl_inode = osb->osb_tl_inode;
2698         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2699
2700         struct ocfs2_xattr_info xi = {
2701                 .name_index = name_index,
2702                 .name = name,
2703                 .value = value,
2704                 .value_len = value_len,
2705         };
2706
2707         struct ocfs2_xattr_search xis = {
2708                 .not_found = -ENODATA,
2709         };
2710
2711         struct ocfs2_xattr_search xbs = {
2712                 .not_found = -ENODATA,
2713         };
2714
2715         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2716                 return -EOPNOTSUPP;
2717
2718         /*
2719          * Only xbs will be used on indexed trees.  xis doesn't need a
2720          * bucket.
2721          */
2722         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2723         if (!xbs.bucket) {
2724                 mlog_errno(-ENOMEM);
2725                 return -ENOMEM;
2726         }
2727
2728         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2729         if (ret < 0) {
2730                 mlog_errno(ret);
2731                 goto cleanup_nolock;
2732         }
2733         xis.inode_bh = xbs.inode_bh = di_bh;
2734         di = (struct ocfs2_dinode *)di_bh->b_data;
2735
2736         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2737         /*
2738          * Scan inode and external block to find the same name
2739          * extended attribute and collect search infomation.
2740          */
2741         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2742         if (ret)
2743                 goto cleanup;
2744         if (xis.not_found) {
2745                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2746                 if (ret)
2747                         goto cleanup;
2748         }
2749
2750         if (xis.not_found && xbs.not_found) {
2751                 ret = -ENODATA;
2752                 if (flags & XATTR_REPLACE)
2753                         goto cleanup;
2754                 ret = 0;
2755                 if (!value)
2756                         goto cleanup;
2757         } else {
2758                 ret = -EEXIST;
2759                 if (flags & XATTR_CREATE)
2760                         goto cleanup;
2761         }
2762
2763
2764         mutex_lock(&tl_inode->i_mutex);
2765
2766         if (ocfs2_truncate_log_needs_flush(osb)) {
2767                 ret = __ocfs2_flush_truncate_log(osb);
2768                 if (ret < 0) {
2769                         mutex_unlock(&tl_inode->i_mutex);
2770                         mlog_errno(ret);
2771                         goto cleanup;
2772                 }
2773         }
2774         mutex_unlock(&tl_inode->i_mutex);
2775
2776         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2777                                         &xbs, &ctxt, &credits);
2778         if (ret) {
2779                 mlog_errno(ret);
2780                 goto cleanup;
2781         }
2782
2783         /* we need to update inode's ctime field, so add credit for it. */
2784         credits += OCFS2_INODE_UPDATE_CREDITS;
2785         ctxt.handle = ocfs2_start_trans(osb, credits);
2786         if (IS_ERR(ctxt.handle)) {
2787                 ret = PTR_ERR(ctxt.handle);
2788                 mlog_errno(ret);
2789                 goto cleanup;
2790         }
2791
2792         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2793
2794         ocfs2_commit_trans(osb, ctxt.handle);
2795
2796         if (ctxt.data_ac)
2797                 ocfs2_free_alloc_context(ctxt.data_ac);
2798         if (ctxt.meta_ac)
2799                 ocfs2_free_alloc_context(ctxt.meta_ac);
2800         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2801                 ocfs2_schedule_truncate_log_flush(osb, 1);
2802         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2803 cleanup:
2804         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2805         ocfs2_inode_unlock(inode, 1);
2806 cleanup_nolock:
2807         brelse(di_bh);
2808         brelse(xbs.xattr_bh);
2809         ocfs2_xattr_bucket_free(xbs.bucket);
2810
2811         return ret;
2812 }
2813
2814 /*
2815  * Find the xattr extent rec which may contains name_hash.
2816  * e_cpos will be the first name hash of the xattr rec.
2817  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2818  */
2819 static int ocfs2_xattr_get_rec(struct inode *inode,
2820                                u32 name_hash,
2821                                u64 *p_blkno,
2822                                u32 *e_cpos,
2823                                u32 *num_clusters,
2824                                struct ocfs2_extent_list *el)
2825 {
2826         int ret = 0, i;
2827         struct buffer_head *eb_bh = NULL;
2828         struct ocfs2_extent_block *eb;
2829         struct ocfs2_extent_rec *rec = NULL;
2830         u64 e_blkno = 0;
2831
2832         if (el->l_tree_depth) {
2833                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2834                 if (ret) {
2835                         mlog_errno(ret);
2836                         goto out;
2837                 }
2838
2839                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2840                 el = &eb->h_list;
2841
2842                 if (el->l_tree_depth) {
2843                         ocfs2_error(inode->i_sb,
2844                                     "Inode %lu has non zero tree depth in "
2845                                     "xattr tree block %llu\n", inode->i_ino,
2846                                     (unsigned long long)eb_bh->b_blocknr);
2847                         ret = -EROFS;
2848                         goto out;
2849                 }
2850         }
2851
2852         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2853                 rec = &el->l_recs[i];
2854
2855                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2856                         e_blkno = le64_to_cpu(rec->e_blkno);
2857                         break;
2858                 }
2859         }
2860
2861         if (!e_blkno) {
2862                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2863                             "record (%u, %u, 0) in xattr", inode->i_ino,
2864                             le32_to_cpu(rec->e_cpos),
2865                             ocfs2_rec_clusters(el, rec));
2866                 ret = -EROFS;
2867                 goto out;
2868         }
2869
2870         *p_blkno = le64_to_cpu(rec->e_blkno);
2871         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2872         if (e_cpos)
2873                 *e_cpos = le32_to_cpu(rec->e_cpos);
2874 out:
2875         brelse(eb_bh);
2876         return ret;
2877 }
2878
2879 typedef int (xattr_bucket_func)(struct inode *inode,
2880                                 struct ocfs2_xattr_bucket *bucket,
2881                                 void *para);
2882
2883 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2884                                    struct ocfs2_xattr_bucket *bucket,
2885                                    int name_index,
2886                                    const char *name,
2887                                    u32 name_hash,
2888                                    u16 *xe_index,
2889                                    int *found)
2890 {
2891         int i, ret = 0, cmp = 1, block_off, new_offset;
2892         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2893         size_t name_len = strlen(name);
2894         struct ocfs2_xattr_entry *xe = NULL;
2895         char *xe_name;
2896
2897         /*
2898          * We don't use binary search in the bucket because there
2899          * may be multiple entries with the same name hash.
2900          */
2901         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2902                 xe = &xh->xh_entries[i];
2903
2904                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2905                         continue;
2906                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2907                         break;
2908
2909                 cmp = name_index - ocfs2_xattr_get_type(xe);
2910                 if (!cmp)
2911                         cmp = name_len - xe->xe_name_len;
2912                 if (cmp)
2913                         continue;
2914
2915                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2916                                                         xh,
2917                                                         i,
2918                                                         &block_off,
2919                                                         &new_offset);
2920                 if (ret) {
2921                         mlog_errno(ret);
2922                         break;
2923                 }
2924
2925
2926                 xe_name = bucket_block(bucket, block_off) + new_offset;
2927                 if (!memcmp(name, xe_name, name_len)) {
2928                         *xe_index = i;
2929                         *found = 1;
2930                         ret = 0;
2931                         break;
2932                 }
2933         }
2934
2935         return ret;
2936 }
2937
2938 /*
2939  * Find the specified xattr entry in a series of buckets.
2940  * This series start from p_blkno and last for num_clusters.
2941  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2942  * the num of the valid buckets.
2943  *
2944  * Return the buffer_head this xattr should reside in. And if the xattr's
2945  * hash is in the gap of 2 buckets, return the lower bucket.
2946  */
2947 static int ocfs2_xattr_bucket_find(struct inode *inode,
2948                                    int name_index,
2949                                    const char *name,
2950                                    u32 name_hash,
2951                                    u64 p_blkno,
2952                                    u32 first_hash,
2953                                    u32 num_clusters,
2954                                    struct ocfs2_xattr_search *xs)
2955 {
2956         int ret, found = 0;
2957         struct ocfs2_xattr_header *xh = NULL;
2958         struct ocfs2_xattr_entry *xe = NULL;
2959         u16 index = 0;
2960         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2961         int low_bucket = 0, bucket, high_bucket;
2962         struct ocfs2_xattr_bucket *search;
2963         u32 last_hash;
2964         u64 blkno, lower_blkno = 0;
2965
2966         search = ocfs2_xattr_bucket_new(inode);
2967         if (!search) {
2968                 ret = -ENOMEM;
2969                 mlog_errno(ret);
2970                 goto out;
2971         }
2972
2973         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2974         if (ret) {
2975                 mlog_errno(ret);
2976                 goto out;
2977         }
2978
2979         xh = bucket_xh(search);
2980         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2981         while (low_bucket <= high_bucket) {
2982                 ocfs2_xattr_bucket_relse(search);
2983
2984                 bucket = (low_bucket + high_bucket) / 2;
2985                 blkno = p_blkno + bucket * blk_per_bucket;
2986                 ret = ocfs2_read_xattr_bucket(search, blkno);
2987                 if (ret) {
2988                         mlog_errno(ret);
2989                         goto out;
2990                 }
2991
2992                 xh = bucket_xh(search);
2993                 xe = &xh->xh_entries[0];
2994                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2995                         high_bucket = bucket - 1;
2996                         continue;
2997                 }
2998
2999                 /*
3000                  * Check whether the hash of the last entry in our
3001                  * bucket is larger than the search one. for an empty
3002                  * bucket, the last one is also the first one.
3003                  */
3004                 if (xh->xh_count)
3005                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3006
3007                 last_hash = le32_to_cpu(xe->xe_name_hash);
3008
3009                 /* record lower_blkno which may be the insert place. */
3010                 lower_blkno = blkno;
3011
3012                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3013                         low_bucket = bucket + 1;
3014                         continue;
3015                 }
3016
3017                 /* the searched xattr should reside in this bucket if exists. */
3018                 ret = ocfs2_find_xe_in_bucket(inode, search,
3019                                               name_index, name, name_hash,
3020                                               &index, &found);
3021                 if (ret) {
3022                         mlog_errno(ret);
3023                         goto out;
3024                 }
3025                 break;
3026         }
3027
3028         /*
3029          * Record the bucket we have found.
3030          * When the xattr's hash value is in the gap of 2 buckets, we will
3031          * always set it to the previous bucket.
3032          */
3033         if (!lower_blkno)
3034                 lower_blkno = p_blkno;
3035
3036         /* This should be in cache - we just read it during the search */
3037         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3038         if (ret) {
3039                 mlog_errno(ret);
3040                 goto out;
3041         }
3042
3043         xs->header = bucket_xh(xs->bucket);
3044         xs->base = bucket_block(xs->bucket, 0);
3045         xs->end = xs->base + inode->i_sb->s_blocksize;
3046
3047         if (found) {
3048                 xs->here = &xs->header->xh_entries[index];
3049                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3050                      (unsigned long long)bucket_blkno(xs->bucket), index);
3051         } else
3052                 ret = -ENODATA;
3053
3054 out:
3055         ocfs2_xattr_bucket_free(search);
3056         return ret;
3057 }
3058
3059 static int ocfs2_xattr_index_block_find(struct inode *inode,
3060                                         struct buffer_head *root_bh,
3061                                         int name_index,
3062                                         const char *name,
3063                                         struct ocfs2_xattr_search *xs)
3064 {
3065         int ret;
3066         struct ocfs2_xattr_block *xb =
3067                         (struct ocfs2_xattr_block *)root_bh->b_data;
3068         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3069         struct ocfs2_extent_list *el = &xb_root->xt_list;
3070         u64 p_blkno = 0;
3071         u32 first_hash, num_clusters = 0;
3072         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3073
3074         if (le16_to_cpu(el->l_next_free_rec) == 0)
3075                 return -ENODATA;
3076
3077         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3078              name, name_hash, name_index);
3079
3080         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3081                                   &num_clusters, el);
3082         if (ret) {
3083                 mlog_errno(ret);
3084                 goto out;
3085         }
3086
3087         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3088
3089         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3090              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3091              first_hash);
3092
3093         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3094                                       p_blkno, first_hash, num_clusters, xs);
3095
3096 out:
3097         return ret;
3098 }
3099
3100 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3101                                        u64 blkno,
3102                                        u32 clusters,
3103                                        xattr_bucket_func *func,
3104                                        void *para)
3105 {
3106         int i, ret = 0;
3107         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3108         u32 num_buckets = clusters * bpc;
3109         struct ocfs2_xattr_bucket *bucket;
3110
3111         bucket = ocfs2_xattr_bucket_new(inode);
3112         if (!bucket) {
3113                 mlog_errno(-ENOMEM);
3114                 return -ENOMEM;
3115         }
3116
3117         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3118              clusters, (unsigned long long)blkno);
3119
3120         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3121                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3122                 if (ret) {
3123                         mlog_errno(ret);
3124                         break;
3125                 }
3126
3127                 /*
3128                  * The real bucket num in this series of blocks is stored
3129                  * in the 1st bucket.
3130                  */
3131                 if (i == 0)
3132                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3133
3134                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3135                      (unsigned long long)blkno,
3136                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3137                 if (func) {
3138                         ret = func(inode, bucket, para);
3139                         if (ret)
3140                                 mlog_errno(ret);
3141                         /* Fall through to bucket_relse() */
3142                 }
3143
3144                 ocfs2_xattr_bucket_relse(bucket);
3145                 if (ret)
3146                         break;
3147         }
3148
3149         ocfs2_xattr_bucket_free(bucket);
3150         return ret;
3151 }
3152
3153 struct ocfs2_xattr_tree_list {
3154         char *buffer;
3155         size_t buffer_size;
3156         size_t result;
3157 };
3158
3159 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3160                                              struct ocfs2_xattr_header *xh,
3161                                              int index,
3162                                              int *block_off,
3163                                              int *new_offset)
3164 {
3165         u16 name_offset;
3166
3167         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3168                 return -EINVAL;
3169
3170         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3171
3172         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3173         *new_offset = name_offset % inode->i_sb->s_blocksize;
3174
3175         return 0;
3176 }
3177
3178 static int ocfs2_list_xattr_bucket(struct inode *inode,
3179                                    struct ocfs2_xattr_bucket *bucket,
3180                                    void *para)
3181 {
3182         int ret = 0, type;
3183         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3184         int i, block_off, new_offset;
3185         const char *prefix, *name;
3186
3187         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3188                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3189                 type = ocfs2_xattr_get_type(entry);
3190                 prefix = ocfs2_xattr_prefix(type);
3191
3192                 if (prefix) {
3193                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3194                                                                 bucket_xh(bucket),
3195                                                                 i,
3196                                                                 &block_off,
3197                                                                 &new_offset);
3198                         if (ret)
3199                                 break;
3200
3201                         name = (const char *)bucket_block(bucket, block_off) +
3202                                 new_offset;
3203                         ret = ocfs2_xattr_list_entry(xl->buffer,
3204                                                      xl->buffer_size,
3205                                                      &xl->result,
3206                                                      prefix, name,
3207                                                      entry->xe_name_len);
3208                         if (ret)
3209                                 break;
3210                 }
3211         }
3212
3213         return ret;
3214 }
3215
3216 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3217                                              struct ocfs2_xattr_tree_root *xt,
3218                                              char *buffer,
3219                                              size_t buffer_size)
3220 {
3221         struct ocfs2_extent_list *el = &xt->xt_list;
3222         int ret = 0;
3223         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3224         u64 p_blkno = 0;
3225         struct ocfs2_xattr_tree_list xl = {
3226                 .buffer = buffer,
3227                 .buffer_size = buffer_size,
3228                 .result = 0,
3229         };
3230
3231         if (le16_to_cpu(el->l_next_free_rec) == 0)
3232                 return 0;
3233
3234         while (name_hash > 0) {
3235                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3236                                           &e_cpos, &num_clusters, el);
3237                 if (ret) {
3238                         mlog_errno(ret);
3239                         goto out;
3240                 }
3241
3242                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3243                                                   ocfs2_list_xattr_bucket,
3244                                                   &xl);
3245                 if (ret) {
3246                         mlog_errno(ret);
3247                         goto out;
3248                 }
3249
3250                 if (e_cpos == 0)
3251                         break;
3252
3253                 name_hash = e_cpos - 1;
3254         }
3255
3256         ret = xl.result;
3257 out:
3258         return ret;
3259 }
3260
3261 static int cmp_xe(const void *a, const void *b)
3262 {
3263         const struct ocfs2_xattr_entry *l = a, *r = b;
3264         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3265         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3266
3267         if (l_hash > r_hash)
3268                 return 1;
3269         if (l_hash < r_hash)
3270                 return -1;
3271         return 0;
3272 }
3273
3274 static void swap_xe(void *a, void *b, int size)
3275 {
3276         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3277
3278         tmp = *l;
3279         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3280         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3281 }
3282
3283 /*
3284  * When the ocfs2_xattr_block is filled up, new bucket will be created
3285  * and all the xattr entries will be moved to the new bucket.
3286  * The header goes at the start of the bucket, and the names+values are
3287  * filled from the end.  This is why *target starts as the last buffer.
3288  * Note: we need to sort the entries since they are not saved in order
3289  * in the ocfs2_xattr_block.
3290  */
3291 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3292                                            struct buffer_head *xb_bh,
3293                                            struct ocfs2_xattr_bucket *bucket)
3294 {
3295         int i, blocksize = inode->i_sb->s_blocksize;
3296         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3297         u16 offset, size, off_change;
3298         struct ocfs2_xattr_entry *xe;
3299         struct ocfs2_xattr_block *xb =
3300                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3301         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3302         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3303         u16 count = le16_to_cpu(xb_xh->xh_count);
3304         char *src = xb_bh->b_data;
3305         char *target = bucket_block(bucket, blks - 1);
3306
3307         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3308              (unsigned long long)xb_bh->b_blocknr,
3309              (unsigned long long)bucket_blkno(bucket));
3310
3311         for (i = 0; i < blks; i++)
3312                 memset(bucket_block(bucket, i), 0, blocksize);
3313
3314         /*
3315          * Since the xe_name_offset is based on ocfs2_xattr_header,
3316          * there is a offset change corresponding to the change of
3317          * ocfs2_xattr_header's position.
3318          */
3319         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3320         xe = &xb_xh->xh_entries[count - 1];
3321         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3322         size = blocksize - offset;
3323
3324         /* copy all the names and values. */
3325         memcpy(target + offset, src + offset, size);
3326
3327         /* Init new header now. */
3328         xh->xh_count = xb_xh->xh_count;
3329         xh->xh_num_buckets = cpu_to_le16(1);
3330         xh->xh_name_value_len = cpu_to_le16(size);
3331         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3332
3333         /* copy all the entries. */
3334         target = bucket_block(bucket, 0);
3335         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3336         size = count * sizeof(struct ocfs2_xattr_entry);
3337         memcpy(target + offset, (char *)xb_xh + offset, size);
3338
3339         /* Change the xe offset for all the xe because of the move. */
3340         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3341                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3342         for (i = 0; i < count; i++)
3343                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3344
3345         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3346              offset, size, off_change);
3347
3348         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3349              cmp_xe, swap_xe);
3350 }
3351
3352 /*
3353  * After we move xattr from block to index btree, we have to
3354  * update ocfs2_xattr_search to the new xe and base.
3355  *
3356  * When the entry is in xattr block, xattr_bh indicates the storage place.
3357  * While if the entry is in index b-tree, "bucket" indicates the
3358  * real place of the xattr.
3359  */
3360 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3361                                             struct ocfs2_xattr_search *xs,
3362                                             struct buffer_head *old_bh)
3363 {
3364         char *buf = old_bh->b_data;
3365         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3366         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3367         int i;
3368
3369         xs->header = bucket_xh(xs->bucket);
3370         xs->base = bucket_block(xs->bucket, 0);
3371         xs->end = xs->base + inode->i_sb->s_blocksize;
3372
3373         if (xs->not_found)
3374                 return;
3375
3376         i = xs->here - old_xh->xh_entries;
3377         xs->here = &xs->header->xh_entries[i];
3378 }
3379
3380 static int ocfs2_xattr_create_index_block(struct inode *inode,
3381                                           struct ocfs2_xattr_search *xs,
3382                                           struct ocfs2_xattr_set_ctxt *ctxt)
3383 {
3384         int ret;
3385         u32 bit_off, len;
3386         u64 blkno;
3387         handle_t *handle = ctxt->handle;
3388         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3389         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3390         struct buffer_head *xb_bh = xs->xattr_bh;
3391         struct ocfs2_xattr_block *xb =
3392                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3393         struct ocfs2_xattr_tree_root *xr;
3394         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3395
3396         mlog(0, "create xattr index block for %llu\n",
3397              (unsigned long long)xb_bh->b_blocknr);
3398
3399         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3400         BUG_ON(!xs->bucket);
3401
3402         /*
3403          * XXX:
3404          * We can use this lock for now, and maybe move to a dedicated mutex
3405          * if performance becomes a problem later.
3406          */
3407         down_write(&oi->ip_alloc_sem);
3408
3409         ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3410                                       OCFS2_JOURNAL_ACCESS_WRITE);
3411         if (ret) {
3412                 mlog_errno(ret);
3413                 goto out;
3414         }
3415
3416         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3417                                      1, 1, &bit_off, &len);
3418         if (ret) {
3419                 mlog_errno(ret);
3420                 goto out;
3421         }
3422
3423         /*
3424          * The bucket may spread in many blocks, and
3425          * we will only touch the 1st block and the last block
3426          * in the whole bucket(one for entry and one for data).
3427          */
3428         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3429
3430         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3431              (unsigned long long)blkno);
3432
3433         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3434         if (ret) {
3435                 mlog_errno(ret);
3436                 goto out;
3437         }
3438
3439         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3440                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3441         if (ret) {
3442                 mlog_errno(ret);
3443                 goto out;
3444         }
3445
3446         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3447         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3448
3449         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3450
3451         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3452         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3453                offsetof(struct ocfs2_xattr_block, xb_attrs));
3454
3455         xr = &xb->xb_attrs.xb_root;
3456         xr->xt_clusters = cpu_to_le32(1);
3457         xr->xt_last_eb_blk = 0;
3458         xr->xt_list.l_tree_depth = 0;
3459         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3460         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3461
3462         xr->xt_list.l_recs[0].e_cpos = 0;
3463         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3464         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3465
3466         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3467
3468         ocfs2_journal_dirty(handle, xb_bh);
3469
3470 out:
3471         up_write(&oi->ip_alloc_sem);
3472
3473         return ret;
3474 }
3475
3476 static int cmp_xe_offset(const void *a, const void *b)
3477 {
3478         const struct ocfs2_xattr_entry *l = a, *r = b;
3479         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3480         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3481
3482         if (l_name_offset < r_name_offset)
3483                 return 1;
3484         if (l_name_offset > r_name_offset)
3485                 return -1;
3486         return 0;
3487 }
3488
3489 /*
3490  * defrag a xattr bucket if we find that the bucket has some
3491  * holes beteen name/value pairs.
3492  * We will move all the name/value pairs to the end of the bucket
3493  * so that we can spare some space for insertion.
3494  */
3495 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3496                                      handle_t *handle,
3497                                      struct ocfs2_xattr_bucket *bucket)
3498 {
3499         int ret, i;
3500         size_t end, offset, len, value_len;
3501         struct ocfs2_xattr_header *xh;
3502         char *entries, *buf, *bucket_buf = NULL;
3503         u64 blkno = bucket_blkno(bucket);
3504         u16 xh_free_start;
3505         size_t blocksize = inode->i_sb->s_blocksize;
3506         struct ocfs2_xattr_entry *xe;
3507
3508         /*
3509          * In order to make the operation more efficient and generic,
3510          * we copy all the blocks into a contiguous memory and do the
3511          * defragment there, so if anything is error, we will not touch
3512          * the real block.
3513          */
3514         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3515         if (!bucket_buf) {
3516                 ret = -EIO;
3517                 goto out;
3518         }
3519
3520         buf = bucket_buf;
3521         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3522                 memcpy(buf, bucket_block(bucket, i), blocksize);
3523
3524         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3525                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3526         if (ret < 0) {
3527                 mlog_errno(ret);
3528                 goto out;
3529         }
3530
3531         xh = (struct ocfs2_xattr_header *)bucket_buf;
3532         entries = (char *)xh->xh_entries;
3533         xh_free_start = le16_to_cpu(xh->xh_free_start);
3534
3535         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3536              "xh_free_start = %u, xh_name_value_len = %u.\n",
3537              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3538              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3539
3540         /*
3541          * sort all the entries by their offset.
3542          * the largest will be the first, so that we can
3543          * move them to the end one by one.
3544          */
3545         sort(entries, le16_to_cpu(xh->xh_count),
3546              sizeof(struct ocfs2_xattr_entry),
3547              cmp_xe_offset, swap_xe);
3548
3549         /* Move all name/values to the end of the bucket. */
3550         xe = xh->xh_entries;
3551         end = OCFS2_XATTR_BUCKET_SIZE;
3552         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3553                 offset = le16_to_cpu(xe->xe_name_offset);
3554                 if (ocfs2_xattr_is_local(xe))
3555                         value_len = OCFS2_XATTR_SIZE(
3556                                         le64_to_cpu(xe->xe_value_size));
3557                 else
3558                         value_len = OCFS2_XATTR_ROOT_SIZE;
3559                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3560
3561                 /*
3562                  * We must make sure that the name/value pair
3563                  * exist in the same block. So adjust end to
3564                  * the previous block end if needed.
3565                  */
3566                 if (((end - len) / blocksize !=
3567                         (end - 1) / blocksize))
3568                         end = end - end % blocksize;
3569
3570                 if (end > offset + len) {
3571                         memmove(bucket_buf + end - len,
3572                                 bucket_buf + offset, len);
3573                         xe->xe_name_offset = cpu_to_le16(end - len);
3574                 }
3575
3576                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3577                                 "bucket %llu\n", (unsigned long long)blkno);
3578
3579                 end -= len;
3580         }
3581
3582         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3583                         "bucket %llu\n", (unsigned long long)blkno);
3584
3585         if (xh_free_start == end)
3586                 goto out;
3587
3588         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3589         xh->xh_free_start = cpu_to_le16(end);
3590
3591         /* sort the entries by their name_hash. */
3592         sort(entries, le16_to_cpu(xh->xh_count),
3593              sizeof(struct ocfs2_xattr_entry),
3594              cmp_xe, swap_xe);
3595
3596         buf = bucket_buf;
3597         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3598                 memcpy(bucket_block(bucket, i), buf, blocksize);
3599         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3600
3601 out:
3602         kfree(bucket_buf);
3603         return ret;
3604 }
3605
3606 /*
3607  * prev_blkno points to the start of an existing extent.  new_blkno
3608  * points to a newly allocated extent.  Because we know each of our
3609  * clusters contains more than bucket, we can easily split one cluster
3610  * at a bucket boundary.  So we take the last cluster of the existing
3611  * extent and split it down the middle.  We move the last half of the
3612  * buckets in the last cluster of the existing extent over to the new
3613  * extent.
3614  *
3615  * first_bh is the buffer at prev_blkno so we can update the existing
3616  * extent's bucket count.  header_bh is the bucket were we were hoping
3617  * to insert our xattr.  If the bucket move places the target in the new
3618  * extent, we'll update first_bh and header_bh after modifying the old
3619  * extent.
3620  *
3621  * first_hash will be set as the 1st xe's name_hash in the new extent.
3622  */
3623 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3624                                                handle_t *handle,
3625                                                struct ocfs2_xattr_bucket *first,
3626                                                struct ocfs2_xattr_bucket *target,
3627                                                u64 new_blkno,
3628                                                u32 num_clusters,
3629                                                u32 *first_hash)
3630 {
3631         int ret;
3632         struct super_block *sb = inode->i_sb;
3633         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3634         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3635         int to_move = num_buckets / 2;
3636         u64 src_blkno;
3637         u64 last_cluster_blkno = bucket_blkno(first) +
3638                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3639
3640         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3641         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3642
3643         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3644              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3645
3646         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3647                                      last_cluster_blkno, new_blkno,
3648                                      to_move, first_hash);
3649         if (ret) {
3650                 mlog_errno(ret);
3651                 goto out;
3652         }
3653
3654         /* This is the first bucket that got moved */
3655         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3656
3657         /*
3658          * If the target bucket was part of the moved buckets, we need to
3659          * update first and target.
3660          */
3661         if (bucket_blkno(target) >= src_blkno) {
3662                 /* Find the block for the new target bucket */
3663                 src_blkno = new_blkno +
3664                         (bucket_blkno(target) - src_blkno);
3665
3666                 ocfs2_xattr_bucket_relse(first);
3667                 ocfs2_xattr_bucket_relse(target);
3668
3669                 /*
3670                  * These shouldn't fail - the buffers are in the
3671                  * journal from ocfs2_cp_xattr_bucket().
3672                  */
3673                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3674                 if (ret) {
3675                         mlog_errno(ret);
3676                         goto out;
3677                 }
3678                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3679                 if (ret)
3680                         mlog_errno(ret);
3681
3682         }
3683
3684 out:
3685         return ret;
3686 }
3687
3688 /*
3689  * Find the suitable pos when we divide a bucket into 2.
3690  * We have to make sure the xattrs with the same hash value exist
3691  * in the same bucket.
3692  *
3693  * If this ocfs2_xattr_header covers more than one hash value, find a
3694  * place where the hash value changes.  Try to find the most even split.
3695  * The most common case is that all entries have different hash values,
3696  * and the first check we make will find a place to split.
3697  */
3698 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3699 {
3700         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3701         int count = le16_to_cpu(xh->xh_count);
3702         int delta, middle = count / 2;
3703
3704         /*
3705          * We start at the middle.  Each step gets farther away in both
3706          * directions.  We therefore hit the change in hash value
3707          * nearest to the middle.  Note that this loop does not execute for
3708          * count < 2.
3709          */
3710         for (delta = 0; delta < middle; delta++) {
3711                 /* Let's check delta earlier than middle */
3712                 if (cmp_xe(&entries[middle - delta - 1],
3713                            &entries[middle - delta]))
3714                         return middle - delta;
3715
3716                 /* For even counts, don't walk off the end */
3717                 if ((middle + delta + 1) == count)
3718                         continue;
3719
3720                 /* Now try delta past middle */
3721                 if (cmp_xe(&entries[middle + delta],
3722                            &entries[middle + delta + 1]))
3723                         return middle + delta + 1;
3724         }
3725
3726         /* Every entry had the same hash */
3727         return count;
3728 }
3729
3730 /*
3731  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3732  * first_hash will record the 1st hash of the new bucket.
3733  *
3734  * Normally half of the xattrs will be moved.  But we have to make
3735  * sure that the xattrs with the same hash value are stored in the
3736  * same bucket. If all the xattrs in this bucket have the same hash
3737  * value, the new bucket will be initialized as an empty one and the
3738  * first_hash will be initialized as (hash_value+1).
3739  */
3740 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3741                                     handle_t *handle,
3742                                     u64 blk,
3743                                     u64 new_blk,
3744                                     u32 *first_hash,
3745                                     int new_bucket_head)
3746 {
3747         int ret, i;
3748         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3749         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3750         struct ocfs2_xattr_header *xh;
3751         struct ocfs2_xattr_entry *xe;
3752         int blocksize = inode->i_sb->s_blocksize;
3753
3754         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3755              (unsigned long long)blk, (unsigned long long)new_blk);
3756
3757         s_bucket = ocfs2_xattr_bucket_new(inode);
3758         t_bucket = ocfs2_xattr_bucket_new(inode);
3759         if (!s_bucket || !t_bucket) {
3760                 ret = -ENOMEM;
3761                 mlog_errno(ret);
3762                 goto out;
3763         }
3764
3765         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3766         if (ret) {
3767                 mlog_errno(ret);
3768                 goto out;
3769         }
3770
3771         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3772                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3773         if (ret) {
3774                 mlog_errno(ret);
3775                 goto out;
3776         }
3777
3778         /*
3779          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3780          * there's no need to read it.
3781          */
3782         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3783         if (ret) {
3784                 mlog_errno(ret);
3785                 goto out;
3786         }
3787
3788         /*
3789          * Hey, if we're overwriting t_bucket, what difference does
3790          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3791          * same part of ocfs2_cp_xattr_bucket().
3792          */
3793         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3794                                                 new_bucket_head ?
3795                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3796                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3797         if (ret) {
3798                 mlog_errno(ret);
3799                 goto out;
3800         }
3801
3802         xh = bucket_xh(s_bucket);
3803         count = le16_to_cpu(xh->xh_count);
3804         start = ocfs2_xattr_find_divide_pos(xh);
3805
3806         if (start == count) {
3807                 xe = &xh->xh_entries[start-1];
3808
3809                 /*
3810                  * initialized a new empty bucket here.
3811                  * The hash value is set as one larger than
3812                  * that of the last entry in the previous bucket.
3813                  */
3814                 for (i = 0; i < t_bucket->bu_blocks; i++)
3815                         memset(bucket_block(t_bucket, i), 0, blocksize);
3816
3817                 xh = bucket_xh(t_bucket);
3818                 xh->xh_free_start = cpu_to_le16(blocksize);
3819                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3820                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3821
3822                 goto set_num_buckets;
3823         }
3824
3825         /* copy the whole bucket to the new first. */
3826         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3827
3828         /* update the new bucket. */
3829         xh = bucket_xh(t_bucket);
3830
3831         /*
3832          * Calculate the total name/value len and xh_free_start for
3833          * the old bucket first.
3834          */
3835         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3836         name_value_len = 0;
3837         for (i = 0; i < start; i++) {
3838                 xe = &xh->xh_entries[i];
3839                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3840                 if (ocfs2_xattr_is_local(xe))
3841                         xe_len +=
3842                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3843                 else
3844                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3845                 name_value_len += xe_len;
3846                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3847                         name_offset = le16_to_cpu(xe->xe_name_offset);
3848         }
3849
3850         /*
3851          * Now begin the modification to the new bucket.
3852          *
3853          * In the new bucket, We just move the xattr entry to the beginning
3854          * and don't touch the name/value. So there will be some holes in the
3855          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3856          * called.
3857          */
3858         xe = &xh->xh_entries[start];
3859         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3860         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3861              (int)((char *)xe - (char *)xh),
3862              (int)((char *)xh->xh_entries - (char *)xh));
3863         memmove((char *)xh->xh_entries, (char *)xe, len);
3864         xe = &xh->xh_entries[count - start];
3865         len = sizeof(struct ocfs2_xattr_entry) * start;
3866         memset((char *)xe, 0, len);
3867
3868         le16_add_cpu(&xh->xh_count, -start);
3869         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3870
3871         /* Calculate xh_free_start for the new bucket. */
3872         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3873         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3874                 xe = &xh->xh_entries[i];
3875                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3876                 if (ocfs2_xattr_is_local(xe))
3877                         xe_len +=
3878                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3879                 else
3880                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3881                 if (le16_to_cpu(xe->xe_name_offset) <
3882                     le16_to_cpu(xh->xh_free_start))
3883                         xh->xh_free_start = xe->xe_name_offset;
3884         }
3885
3886 set_num_buckets:
3887         /* set xh->xh_num_buckets for the new xh. */
3888         if (new_bucket_head)
3889                 xh->xh_num_buckets = cpu_to_le16(1);
3890         else
3891                 xh->xh_num_buckets = 0;
3892
3893         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3894
3895         /* store the first_hash of the new bucket. */
3896         if (first_hash)
3897                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3898
3899         /*
3900          * Now only update the 1st block of the old bucket.  If we
3901          * just added a new empty bucket, there is no need to modify
3902          * it.
3903          */
3904         if (start == count)
3905                 goto out;
3906
3907         xh = bucket_xh(s_bucket);
3908         memset(&xh->xh_entries[start], 0,
3909                sizeof(struct ocfs2_xattr_entry) * (count - start));
3910         xh->xh_count = cpu_to_le16(start);
3911         xh->xh_free_start = cpu_to_le16(name_offset);
3912         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3913
3914         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3915
3916 out:
3917         ocfs2_xattr_bucket_free(s_bucket);
3918         ocfs2_xattr_bucket_free(t_bucket);
3919
3920         return ret;
3921 }
3922
3923 /*
3924  * Copy xattr from one bucket to another bucket.
3925  *
3926  * The caller must make sure that the journal transaction
3927  * has enough space for journaling.
3928  */
3929 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3930                                  handle_t *handle,
3931                                  u64 s_blkno,
3932                                  u64 t_blkno,
3933                                  int t_is_new)
3934 {
3935         int ret;
3936         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3937
3938         BUG_ON(s_blkno == t_blkno);
3939
3940         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3941              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3942              t_is_new);
3943
3944         s_bucket = ocfs2_xattr_bucket_new(inode);
3945         t_bucket = ocfs2_xattr_bucket_new(inode);
3946         if (!s_bucket || !t_bucket) {
3947                 ret = -ENOMEM;
3948                 mlog_errno(ret);
3949                 goto out;
3950         }
3951
3952         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3953         if (ret)
3954                 goto out;
3955
3956         /*
3957          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3958          * there's no need to read it.
3959          */
3960         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3961         if (ret)
3962                 goto out;
3963
3964         /*
3965          * Hey, if we're overwriting t_bucket, what difference does
3966          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3967          * cluster to fill, we came here from
3968          * ocfs2_mv_xattr_buckets(), and it is really new -
3969          * ACCESS_CREATE is required.  But we also might have moved data
3970          * out of t_bucket before extending back into it.
3971          * ocfs2_add_new_xattr_bucket() can do this - its call to
3972          * ocfs2_add_new_xattr_cluster() may have created a new extent
3973          * and copied out the end of the old extent.  Then it re-extends
3974          * the old extent back to create space for new xattrs.  That's
3975          * how we get here, and the bucket isn't really new.
3976          */
3977         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3978                                                 t_is_new ?
3979                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3980                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3981         if (ret)
3982                 goto out;
3983
3984         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3985         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3986
3987 out:
3988         ocfs2_xattr_bucket_free(t_bucket);
3989         ocfs2_xattr_bucket_free(s_bucket);
3990
3991         return ret;
3992 }
3993
3994 /*
3995  * src_blk points to the start of an existing extent.  last_blk points to
3996  * last cluster in that extent.  to_blk points to a newly allocated
3997  * extent.  We copy the buckets from the cluster at last_blk to the new
3998  * extent.  If start_bucket is non-zero, we skip that many buckets before
3999  * we start copying.  The new extent's xh_num_buckets gets set to the
4000  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4001  * by the same amount.
4002  */
4003 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4004                                   u64 src_blk, u64 last_blk, u64 to_blk,
4005                                   unsigned int start_bucket,
4006                                   u32 *first_hash)
4007 {
4008         int i, ret, credits;
4009         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4010         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4011         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4012         struct ocfs2_xattr_bucket *old_first, *new_first;
4013
4014         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4015              (unsigned long long)last_blk, (unsigned long long)to_blk);
4016
4017         BUG_ON(start_bucket >= num_buckets);
4018         if (start_bucket) {
4019                 num_buckets -= start_bucket;
4020                 last_blk += (start_bucket * blks_per_bucket);
4021         }
4022
4023         /* The first bucket of the original extent */
4024         old_first = ocfs2_xattr_bucket_new(inode);
4025         /* The first bucket of the new extent */
4026         new_first = ocfs2_xattr_bucket_new(inode);
4027         if (!old_first || !new_first) {
4028                 ret = -ENOMEM;
4029                 mlog_errno(ret);
4030                 goto out;
4031         }
4032
4033         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4034         if (ret) {
4035                 mlog_errno(ret);
4036                 goto out;
4037         }
4038
4039         /*
4040          * We need to update the first bucket of the old extent and all
4041          * the buckets going to the new extent.
4042          */
4043         credits = ((num_buckets + 1) * blks_per_bucket) +
4044                 handle->h_buffer_credits;
4045         ret = ocfs2_extend_trans(handle, credits);
4046         if (ret) {
4047                 mlog_errno(ret);
4048                 goto out;
4049         }
4050
4051         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4052                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4053         if (ret) {
4054                 mlog_errno(ret);
4055                 goto out;
4056         }
4057
4058         for (i = 0; i < num_buckets; i++) {
4059                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4060                                             last_blk + (i * blks_per_bucket),
4061                                             to_blk + (i * blks_per_bucket),
4062                                             1);
4063                 if (ret) {
4064                         mlog_errno(ret);
4065                         goto out;
4066                 }
4067         }
4068
4069         /*
4070          * Get the new bucket ready before we dirty anything
4071          * (This actually shouldn't fail, because we already dirtied
4072          * it once in ocfs2_cp_xattr_bucket()).
4073          */
4074         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4075         if (ret) {
4076                 mlog_errno(ret);
4077                 goto out;
4078         }
4079         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4080                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4081         if (ret) {
4082                 mlog_errno(ret);
4083                 goto out;
4084         }
4085
4086         /* Now update the headers */
4087         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4088         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4089
4090         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4091         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4092
4093         if (first_hash)
4094                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4095
4096 out:
4097         ocfs2_xattr_bucket_free(new_first);
4098         ocfs2_xattr_bucket_free(old_first);
4099         return ret;
4100 }
4101
4102 /*
4103  * Move some xattrs in this cluster to the new cluster.
4104  * This function should only be called when bucket size == cluster size.
4105  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4106  */
4107 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4108                                       handle_t *handle,
4109                                       u64 prev_blk,
4110                                       u64 new_blk,
4111                                       u32 *first_hash)
4112 {
4113         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4114         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4115
4116         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4117
4118         ret = ocfs2_extend_trans(handle, credits);
4119         if (ret) {
4120                 mlog_errno(ret);
4121                 return ret;
4122         }
4123
4124         /* Move half of the xattr in start_blk to the next bucket. */
4125         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4126                                           new_blk, first_hash, 1);
4127 }
4128
4129 /*
4130  * Move some xattrs from the old cluster to the new one since they are not
4131  * contiguous in ocfs2 xattr tree.
4132  *
4133  * new_blk starts a new separate cluster, and we will move some xattrs from
4134  * prev_blk to it. v_start will be set as the first name hash value in this
4135  * new cluster so that it can be used as e_cpos during tree insertion and
4136  * don't collide with our original b-tree operations. first_bh and header_bh
4137  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4138  * to extend the insert bucket.
4139  *
4140  * The problem is how much xattr should we move to the new one and when should
4141  * we update first_bh and header_bh?
4142  * 1. If cluster size > bucket size, that means the previous cluster has more
4143  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4144  *    update the first_bh and header_bh if the insert bucket has been moved
4145  *    to the new cluster.
4146  * 2. If cluster_size == bucket_size:
4147  *    a) If the previous extent rec has more than one cluster and the insert
4148  *       place isn't in the last cluster, copy the entire last cluster to the
4149  *       new one. This time, we don't need to upate the first_bh and header_bh
4150  *       since they will not be moved into the new cluster.
4151  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4152  *       the new one. And we set the extend flag to zero if the insert place is
4153  *       moved into the new allocated cluster since no extend is needed.
4154  */
4155 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4156                                             handle_t *handle,
4157                                             struct ocfs2_xattr_bucket *first,
4158                                             struct ocfs2_xattr_bucket *target,
4159                                             u64 new_blk,
4160                                             u32 prev_clusters,
4161                                             u32 *v_start,
4162                                             int *extend)
4163 {
4164         int ret;
4165
4166         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4167              (unsigned long long)bucket_blkno(first), prev_clusters,
4168              (unsigned long long)new_blk);
4169
4170         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4171                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4172                                                           handle,
4173                                                           first, target,
4174                                                           new_blk,
4175                                                           prev_clusters,
4176                                                           v_start);
4177                 if (ret)
4178                         mlog_errno(ret);
4179         } else {
4180                 /* The start of the last cluster in the first extent */
4181                 u64 last_blk = bucket_blkno(first) +
4182                         ((prev_clusters - 1) *
4183                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4184
4185                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4186                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4187                                                      bucket_blkno(first),
4188                                                      last_blk, new_blk, 0,
4189                                                      v_start);
4190                         if (ret)
4191                                 mlog_errno(ret);
4192                 } else {
4193                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4194                                                          last_blk, new_blk,
4195                                                          v_start);
4196                         if (ret)
4197                                 mlog_errno(ret);
4198
4199                         if ((bucket_blkno(target) == last_blk) && extend)
4200                                 *extend = 0;
4201                 }
4202         }
4203
4204         return ret;
4205 }
4206
4207 /*
4208  * Add a new cluster for xattr storage.
4209  *
4210  * If the new cluster is contiguous with the previous one, it will be
4211  * appended to the same extent record, and num_clusters will be updated.
4212  * If not, we will insert a new extent for it and move some xattrs in
4213  * the last cluster into the new allocated one.
4214  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4215  * lose the benefits of hashing because we'll have to search large leaves.
4216  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4217  * if it's bigger).
4218  *
4219  * first_bh is the first block of the previous extent rec and header_bh
4220  * indicates the bucket we will insert the new xattrs. They will be updated
4221  * when the header_bh is moved into the new cluster.
4222  */
4223 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4224                                        struct buffer_head *root_bh,
4225                                        struct ocfs2_xattr_bucket *first,
4226                                        struct ocfs2_xattr_bucket *target,
4227                                        u32 *num_clusters,
4228                                        u32 prev_cpos,
4229                                        int *extend,
4230                                        struct ocfs2_xattr_set_ctxt *ctxt)
4231 {
4232         int ret;
4233         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4234         u32 prev_clusters = *num_clusters;
4235         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4236         u64 block;
4237         handle_t *handle = ctxt->handle;
4238         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4239         struct ocfs2_extent_tree et;
4240
4241         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4242              "previous xattr blkno = %llu\n",
4243              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4244              prev_cpos, (unsigned long long)bucket_blkno(first));
4245
4246         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4247
4248         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4249                                       OCFS2_JOURNAL_ACCESS_WRITE);
4250         if (ret < 0) {
4251                 mlog_errno(ret);
4252                 goto leave;
4253         }
4254
4255         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4256                                      clusters_to_add, &bit_off, &num_bits);
4257         if (ret < 0) {
4258                 if (ret != -ENOSPC)
4259                         mlog_errno(ret);
4260                 goto leave;
4261         }
4262
4263         BUG_ON(num_bits > clusters_to_add);
4264
4265         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4266         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4267              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4268
4269         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4270             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4271              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4272                 /*
4273                  * If this cluster is contiguous with the old one and
4274                  * adding this new cluster, we don't surpass the limit of
4275                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4276                  * initialized and used like other buckets in the previous
4277                  * cluster.
4278                  * So add it as a contiguous one. The caller will handle
4279                  * its init process.
4280                  */
4281                 v_start = prev_cpos + prev_clusters;
4282                 *num_clusters = prev_clusters + num_bits;
4283                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4284                      num_bits);
4285         } else {
4286                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4287                                                        handle,
4288                                                        first,
4289                                                        target,
4290                                                        block,
4291                                                        prev_clusters,
4292                                                        &v_start,
4293                                                        extend);
4294                 if (ret) {
4295                         mlog_errno(ret);
4296                         goto leave;
4297                 }
4298         }
4299
4300         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4301              num_bits, (unsigned long long)block, v_start);
4302         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4303                                   num_bits, 0, ctxt->meta_ac);
4304         if (ret < 0) {
4305                 mlog_errno(ret);
4306                 goto leave;
4307         }
4308
4309         ret = ocfs2_journal_dirty(handle, root_bh);
4310         if (ret < 0)
4311                 mlog_errno(ret);
4312
4313 leave:
4314         return ret;
4315 }
4316
4317 /*
4318  * We are given an extent.  'first' is the bucket at the very front of
4319  * the extent.  The extent has space for an additional bucket past
4320  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4321  * of the target bucket.  We wish to shift every bucket past the target
4322  * down one, filling in that additional space.  When we get back to the
4323  * target, we split the target between itself and the now-empty bucket
4324  * at target+1 (aka, target_blkno + blks_per_bucket).
4325  */
4326 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4327                                      handle_t *handle,
4328                                      struct ocfs2_xattr_bucket *first,
4329                                      u64 target_blk,
4330                                      u32 num_clusters)
4331 {
4332         int ret, credits;
4333         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4334         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4335         u64 end_blk;
4336         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4337
4338         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4339              "from %llu, len = %u\n", (unsigned long long)target_blk,
4340              (unsigned long long)bucket_blkno(first), num_clusters);
4341
4342         /* The extent must have room for an additional bucket */
4343         BUG_ON(new_bucket >=
4344                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4345
4346         /* end_blk points to the last existing bucket */
4347         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4348
4349         /*
4350          * end_blk is the start of the last existing bucket.
4351          * Thus, (end_blk - target_blk) covers the target bucket and
4352          * every bucket after it up to, but not including, the last
4353          * existing bucket.  Then we add the last existing bucket, the
4354          * new bucket, and the first bucket (3 * blk_per_bucket).
4355          */
4356         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4357                   handle->h_buffer_credits;
4358         ret = ocfs2_extend_trans(handle, credits);
4359         if (ret) {
4360                 mlog_errno(ret);
4361                 goto out;
4362         }
4363
4364         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4365                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4366         if (ret) {
4367                 mlog_errno(ret);
4368                 goto out;
4369         }
4370
4371         while (end_blk != target_blk) {
4372                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4373                                             end_blk + blk_per_bucket, 0);
4374                 if (ret)
4375                         goto out;
4376                 end_blk -= blk_per_bucket;
4377         }
4378
4379         /* Move half of the xattr in target_blkno to the next bucket. */
4380         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4381                                         target_blk + blk_per_bucket, NULL, 0);
4382
4383         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4384         ocfs2_xattr_bucket_journal_dirty(handle, first);
4385
4386 out:
4387         return ret;
4388 }
4389
4390 /*
4391  * Add new xattr bucket in an extent record and adjust the buckets
4392  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4393  * bucket we want to insert into.
4394  *
4395  * In the easy case, we will move all the buckets after target down by
4396  * one. Half of target's xattrs will be moved to the next bucket.
4397  *
4398  * If current cluster is full, we'll allocate a new one.  This may not
4399  * be contiguous.  The underlying calls will make sure that there is
4400  * space for the insert, shifting buckets around if necessary.
4401  * 'target' may be moved by those calls.
4402  */
4403 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4404                                       struct buffer_head *xb_bh,
4405                                       struct ocfs2_xattr_bucket *target,
4406                                       struct ocfs2_xattr_set_ctxt *ctxt)
4407 {
4408         struct ocfs2_xattr_block *xb =
4409                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4410         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4411         struct ocfs2_extent_list *el = &xb_root->xt_list;
4412         u32 name_hash =
4413                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4414         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4415         int ret, num_buckets, extend = 1;
4416         u64 p_blkno;
4417         u32 e_cpos, num_clusters;
4418         /* The bucket at the front of the extent */
4419         struct ocfs2_xattr_bucket *first;
4420
4421         mlog(0, "Add new xattr bucket starting from %llu\n",
4422              (unsigned long long)bucket_blkno(target));
4423
4424         /* The first bucket of the original extent */
4425         first = ocfs2_xattr_bucket_new(inode);
4426         if (!first) {
4427                 ret = -ENOMEM;
4428                 mlog_errno(ret);
4429                 goto out;
4430         }
4431
4432         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4433                                   &num_clusters, el);
4434         if (ret) {
4435                 mlog_errno(ret);
4436                 goto out;
4437         }
4438
4439         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4440         if (ret) {
4441                 mlog_errno(ret);
4442                 goto out;
4443         }
4444
4445         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4446         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4447                 /*
4448                  * This can move first+target if the target bucket moves
4449                  * to the new extent.
4450                  */
4451                 ret = ocfs2_add_new_xattr_cluster(inode,
4452                                                   xb_bh,
4453                                                   first,
4454                                                   target,
4455                                                   &num_clusters,
4456                                                   e_cpos,
4457                                                   &extend,
4458                                                   ctxt);
4459                 if (ret) {
4460                         mlog_errno(ret);
4461                         goto out;
4462                 }
4463         }
4464
4465         if (extend) {
4466                 ret = ocfs2_extend_xattr_bucket(inode,
4467                                                 ctxt->handle,
4468                                                 first,
4469                                                 bucket_blkno(target),
4470                                                 num_clusters);
4471                 if (ret)
4472                         mlog_errno(ret);
4473         }
4474
4475 out:
4476         ocfs2_xattr_bucket_free(first);
4477
4478         return ret;
4479 }
4480
4481 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4482                                         struct ocfs2_xattr_bucket *bucket,
4483                                         int offs)
4484 {
4485         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4486
4487         offs = offs % inode->i_sb->s_blocksize;
4488         return bucket_block(bucket, block_off) + offs;
4489 }
4490
4491 /*
4492  * Handle the normal xattr set, including replace, delete and new.
4493  *
4494  * Note: "local" indicates the real data's locality. So we can't
4495  * just its bucket locality by its length.
4496  */
4497 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4498                                          struct ocfs2_xattr_info *xi,
4499                                          struct ocfs2_xattr_search *xs,
4500                                          u32 name_hash,
4501                                          int local)
4502 {
4503         struct ocfs2_xattr_entry *last, *xe;
4504         int name_len = strlen(xi->name);
4505         struct ocfs2_xattr_header *xh = xs->header;
4506         u16 count = le16_to_cpu(xh->xh_count), start;
4507         size_t blocksize = inode->i_sb->s_blocksize;
4508         char *val;
4509         size_t offs, size, new_size;
4510
4511         last = &xh->xh_entries[count];
4512         if (!xs->not_found) {
4513                 xe = xs->here;
4514                 offs = le16_to_cpu(xe->xe_name_offset);
4515                 if (ocfs2_xattr_is_local(xe))
4516                         size = OCFS2_XATTR_SIZE(name_len) +
4517                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4518                 else
4519                         size = OCFS2_XATTR_SIZE(name_len) +
4520                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4521
4522                 /*
4523                  * If the new value will be stored outside, xi->value has been
4524                  * initalized as an empty ocfs2_xattr_value_root, and the same
4525                  * goes with xi->value_len, so we can set new_size safely here.
4526                  * See ocfs2_xattr_set_in_bucket.
4527                  */
4528                 new_size = OCFS2_XATTR_SIZE(name_len) +
4529                            OCFS2_XATTR_SIZE(xi->value_len);
4530
4531                 le16_add_cpu(&xh->xh_name_value_len, -size);
4532                 if (xi->value) {
4533                         if (new_size > size)
4534                                 goto set_new_name_value;
4535
4536                         /* Now replace the old value with new one. */
4537                         if (local)
4538                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4539                         else
4540                                 xe->xe_value_size = 0;
4541
4542                         val = ocfs2_xattr_bucket_get_val(inode,
4543                                                          xs->bucket, offs);
4544                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4545                                size - OCFS2_XATTR_SIZE(name_len));
4546                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4547                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4548                                        xi->value, xi->value_len);
4549
4550                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4551                         ocfs2_xattr_set_local(xe, local);
4552                         return;
4553                 } else {
4554                         /*
4555                          * Remove the old entry if there is more than one.
4556                          * We don't remove the last entry so that we can
4557                          * use it to indicate the hash value of the empty
4558                          * bucket.
4559                          */
4560                         last -= 1;
4561                         le16_add_cpu(&xh->xh_count, -1);
4562                         if (xh->xh_count) {
4563                                 memmove(xe, xe + 1,
4564                                         (void *)last - (void *)xe);
4565                                 memset(last, 0,
4566                                        sizeof(struct ocfs2_xattr_entry));
4567                         } else
4568                                 xh->xh_free_start =
4569                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4570
4571                         return;
4572                 }
4573         } else {
4574                 /* find a new entry for insert. */
4575                 int low = 0, high = count - 1, tmp;
4576                 struct ocfs2_xattr_entry *tmp_xe;
4577
4578                 while (low <= high && count) {
4579                         tmp = (low + high) / 2;
4580                         tmp_xe = &xh->xh_entries[tmp];
4581
4582                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4583                                 low = tmp + 1;
4584                         else if (name_hash <
4585                                  le32_to_cpu(tmp_xe->xe_name_hash))
4586                                 high = tmp - 1;
4587                         else {
4588                                 low = tmp;
4589                                 break;
4590                         }
4591                 }
4592
4593                 xe = &xh->xh_entries[low];
4594                 if (low != count)
4595                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4596
4597                 le16_add_cpu(&xh->xh_count, 1);
4598                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4599                 xe->xe_name_hash = cpu_to_le32(name_hash);
4600                 xe->xe_name_len = name_len;
4601                 ocfs2_xattr_set_type(xe, xi->name_index);
4602         }
4603
4604 set_new_name_value:
4605         /* Insert the new name+value. */
4606         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4607
4608         /*
4609          * We must make sure that the name/value pair
4610          * exists in the same block.
4611          */
4612         offs = le16_to_cpu(xh->xh_free_start);
4613         start = offs - size;
4614
4615         if (start >> inode->i_sb->s_blocksize_bits !=
4616             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4617                 offs = offs - offs % blocksize;
4618                 xh->xh_free_start = cpu_to_le16(offs);
4619         }
4620
4621         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4622         xe->xe_name_offset = cpu_to_le16(offs - size);
4623
4624         memset(val, 0, size);
4625         memcpy(val, xi->name, name_len);
4626         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4627
4628         xe->xe_value_size = cpu_to_le64(xi->value_len);
4629         ocfs2_xattr_set_local(xe, local);
4630         xs->here = xe;
4631         le16_add_cpu(&xh->xh_free_start, -size);
4632         le16_add_cpu(&xh->xh_name_value_len, size);
4633
4634         return;
4635 }
4636
4637 /*
4638  * Set the xattr entry in the specified bucket.
4639  * The bucket is indicated by xs->bucket and it should have the enough
4640  * space for the xattr insertion.
4641  */
4642 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4643                                            handle_t *handle,
4644                                            struct ocfs2_xattr_info *xi,
4645                                            struct ocfs2_xattr_search *xs,
4646                                            u32 name_hash,
4647                                            int local)
4648 {
4649         int ret;
4650         u64 blkno;
4651
4652         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4653              (unsigned long)xi->value_len, xi->name_index,
4654              (unsigned long long)bucket_blkno(xs->bucket));
4655
4656         if (!xs->bucket->bu_bhs[1]) {
4657                 blkno = bucket_blkno(xs->bucket);
4658                 ocfs2_xattr_bucket_relse(xs->bucket);
4659                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4660                 if (ret) {
4661                         mlog_errno(ret);
4662                         goto out;
4663                 }
4664         }
4665
4666         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4667                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4668         if (ret < 0) {
4669                 mlog_errno(ret);
4670                 goto out;
4671         }
4672
4673         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4674         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4675
4676 out:
4677         return ret;
4678 }
4679
4680 /*
4681  * Truncate the specified xe_off entry in xattr bucket.
4682  * bucket is indicated by header_bh and len is the new length.
4683  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4684  *
4685  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4686  */
4687 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4688                                              struct ocfs2_xattr_bucket *bucket,
4689                                              int xe_off,
4690                                              int len,
4691                                              struct ocfs2_xattr_set_ctxt *ctxt)
4692 {
4693         int ret, offset;
4694         u64 value_blk;
4695         struct ocfs2_xattr_entry *xe;
4696         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4697         size_t blocksize = inode->i_sb->s_blocksize;
4698         struct ocfs2_xattr_value_buf vb = {
4699                 .vb_access = ocfs2_journal_access,
4700         };
4701
4702         xe = &xh->xh_entries[xe_off];
4703
4704         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4705
4706         offset = le16_to_cpu(xe->xe_name_offset) +
4707                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4708
4709         value_blk = offset / blocksize;
4710
4711         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4712         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4713
4714         vb.vb_bh = bucket->bu_bhs[value_blk];
4715         BUG_ON(!vb.vb_bh);
4716
4717         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4718                 (vb.vb_bh->b_data + offset % blocksize);
4719
4720         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4721                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4722         if (ret) {
4723                 mlog_errno(ret);
4724                 goto out;
4725         }
4726
4727         /*
4728          * From here on out we have to dirty the bucket.  The generic
4729          * value calls only modify one of the bucket's bhs, but we need
4730          * to send the bucket at once.  So if they error, they *could* have
4731          * modified something.  We have to assume they did, and dirty
4732          * the whole bucket.  This leaves us in a consistent state.
4733          */
4734         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4735              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4736         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4737         if (ret) {
4738                 mlog_errno(ret);
4739                 goto out_dirty;
4740         }
4741
4742         xe->xe_value_size = cpu_to_le64(len);
4743
4744 out_dirty:
4745         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4746
4747 out:
4748         return ret;
4749 }
4750
4751 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4752                                         struct ocfs2_xattr_search *xs,
4753                                         int len,
4754                                         struct ocfs2_xattr_set_ctxt *ctxt)
4755 {
4756         int ret, offset;
4757         struct ocfs2_xattr_entry *xe = xs->here;
4758         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4759
4760         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4761
4762         offset = xe - xh->xh_entries;
4763         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4764                                                 offset, len, ctxt);
4765         if (ret)
4766                 mlog_errno(ret);
4767
4768         return ret;
4769 }
4770
4771 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4772                                                 handle_t *handle,
4773                                                 struct ocfs2_xattr_search *xs,
4774                                                 char *val,
4775                                                 int value_len)
4776 {
4777         int offset;
4778         struct ocfs2_xattr_value_root *xv;
4779         struct ocfs2_xattr_entry *xe = xs->here;
4780
4781         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4782
4783         offset = le16_to_cpu(xe->xe_name_offset) +
4784                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4785
4786         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4787
4788         return __ocfs2_xattr_set_value_outside(inode, handle,
4789                                                xv, val, value_len);
4790 }
4791
4792 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4793                                   struct buffer_head *root_bh,
4794                                   u64 blkno,
4795                                   u32 cpos,
4796                                   u32 len)
4797 {
4798         int ret;
4799         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4800         struct inode *tl_inode = osb->osb_tl_inode;
4801         handle_t *handle;
4802         struct ocfs2_xattr_block *xb =
4803                         (struct ocfs2_xattr_block *)root_bh->b_data;
4804         struct ocfs2_alloc_context *meta_ac = NULL;
4805         struct ocfs2_cached_dealloc_ctxt dealloc;
4806         struct ocfs2_extent_tree et;
4807
4808         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4809
4810         ocfs2_init_dealloc_ctxt(&dealloc);
4811
4812         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4813              cpos, len, (unsigned long long)blkno);
4814
4815         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4816
4817         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4818         if (ret) {
4819                 mlog_errno(ret);
4820                 return ret;
4821         }
4822
4823         mutex_lock(&tl_inode->i_mutex);
4824
4825         if (ocfs2_truncate_log_needs_flush(osb)) {
4826                 ret = __ocfs2_flush_truncate_log(osb);
4827                 if (ret < 0) {
4828                         mlog_errno(ret);
4829                         goto out;
4830                 }
4831         }
4832
4833         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4834         if (IS_ERR(handle)) {
4835                 ret = -ENOMEM;
4836                 mlog_errno(ret);
4837                 goto out;
4838         }
4839
4840         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4841                                       OCFS2_JOURNAL_ACCESS_WRITE);
4842         if (ret) {
4843                 mlog_errno(ret);
4844                 goto out_commit;
4845         }
4846
4847         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4848                                   &dealloc);
4849         if (ret) {
4850                 mlog_errno(ret);
4851                 goto out_commit;
4852         }
4853
4854         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4855
4856         ret = ocfs2_journal_dirty(handle, root_bh);
4857         if (ret) {
4858                 mlog_errno(ret);
4859                 goto out_commit;
4860         }
4861
4862         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4863         if (ret)
4864                 mlog_errno(ret);
4865
4866 out_commit:
4867         ocfs2_commit_trans(osb, handle);
4868 out:
4869         ocfs2_schedule_truncate_log_flush(osb, 1);
4870
4871         mutex_unlock(&tl_inode->i_mutex);
4872
4873         if (meta_ac)
4874                 ocfs2_free_alloc_context(meta_ac);
4875
4876         ocfs2_run_deallocs(osb, &dealloc);
4877
4878         return ret;
4879 }
4880
4881 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4882                                          handle_t *handle,
4883                                          struct ocfs2_xattr_search *xs)
4884 {
4885         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4886         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4887                                                 le16_to_cpu(xh->xh_count) - 1];
4888         int ret = 0;
4889
4890         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4891                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4892         if (ret) {
4893                 mlog_errno(ret);
4894                 return;
4895         }
4896
4897         /* Remove the old entry. */
4898         memmove(xs->here, xs->here + 1,
4899                 (void *)last - (void *)xs->here);
4900         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4901         le16_add_cpu(&xh->xh_count, -1);
4902
4903         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4904 }
4905
4906 /*
4907  * Set the xattr name/value in the bucket specified in xs.
4908  *
4909  * As the new value in xi may be stored in the bucket or in an outside cluster,
4910  * we divide the whole process into 3 steps:
4911  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4912  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4913  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4914  * 4. If the clusters for the new outside value can't be allocated, we need
4915  *    to free the xattr we allocated in set.
4916  */
4917 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4918                                      struct ocfs2_xattr_info *xi,
4919                                      struct ocfs2_xattr_search *xs,
4920                                      struct ocfs2_xattr_set_ctxt *ctxt)
4921 {
4922         int ret, local = 1;
4923         size_t value_len;
4924         char *val = (char *)xi->value;
4925         struct ocfs2_xattr_entry *xe = xs->here;
4926         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4927                                               strlen(xi->name));
4928
4929         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4930                 /*
4931                  * We need to truncate the xattr storage first.
4932                  *
4933                  * If both the old and new value are stored to
4934                  * outside block, we only need to truncate
4935                  * the storage and then set the value outside.
4936                  *
4937                  * If the new value should be stored within block,
4938                  * we should free all the outside block first and
4939                  * the modification to the xattr block will be done
4940                  * by following steps.
4941                  */
4942                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4943                         value_len = xi->value_len;
4944                 else
4945                         value_len = 0;
4946
4947                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4948                                                            value_len,
4949                                                            ctxt);
4950                 if (ret)
4951                         goto out;
4952
4953                 if (value_len)
4954                         goto set_value_outside;
4955         }
4956
4957         value_len = xi->value_len;
4958         /* So we have to handle the inside block change now. */
4959         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4960                 /*
4961                  * If the new value will be stored outside of block,
4962                  * initalize a new empty value root and insert it first.
4963                  */
4964                 local = 0;
4965                 xi->value = &def_xv;
4966                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4967         }
4968
4969         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4970                                               name_hash, local);
4971         if (ret) {
4972                 mlog_errno(ret);
4973                 goto out;
4974         }
4975
4976         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4977                 goto out;
4978
4979         /* allocate the space now for the outside block storage. */
4980         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4981                                                    value_len, ctxt);
4982         if (ret) {
4983                 mlog_errno(ret);
4984
4985                 if (xs->not_found) {
4986                         /*
4987                          * We can't allocate enough clusters for outside
4988                          * storage and we have allocated xattr already,
4989                          * so need to remove it.
4990                          */
4991                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4992                 }
4993                 goto out;
4994         }
4995
4996 set_value_outside:
4997         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4998                                                    xs, val, value_len);
4999 out:
5000         return ret;
5001 }
5002
5003 /*
5004  * check whether the xattr bucket is filled up with the same hash value.
5005  * If we want to insert the xattr with the same hash, return -ENOSPC.
5006  * If we want to insert a xattr with different hash value, go ahead
5007  * and ocfs2_divide_xattr_bucket will handle this.
5008  */
5009 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5010                                               struct ocfs2_xattr_bucket *bucket,
5011                                               const char *name)
5012 {
5013         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5014         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5015
5016         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5017                 return 0;
5018
5019         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5020             xh->xh_entries[0].xe_name_hash) {
5021                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5022                      "hash = %u\n",
5023                      (unsigned long long)bucket_blkno(bucket),
5024                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5025                 return -ENOSPC;
5026         }
5027
5028         return 0;
5029 }
5030
5031 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5032                                              struct ocfs2_xattr_info *xi,
5033                                              struct ocfs2_xattr_search *xs,
5034                                              struct ocfs2_xattr_set_ctxt *ctxt)
5035 {
5036         struct ocfs2_xattr_header *xh;
5037         struct ocfs2_xattr_entry *xe;
5038         u16 count, header_size, xh_free_start;
5039         int free, max_free, need, old;
5040         size_t value_size = 0, name_len = strlen(xi->name);
5041         size_t blocksize = inode->i_sb->s_blocksize;
5042         int ret, allocation = 0;
5043
5044         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5045
5046 try_again:
5047         xh = xs->header;
5048         count = le16_to_cpu(xh->xh_count);
5049         xh_free_start = le16_to_cpu(xh->xh_free_start);
5050         header_size = sizeof(struct ocfs2_xattr_header) +
5051                         count * sizeof(struct ocfs2_xattr_entry);
5052         max_free = OCFS2_XATTR_BUCKET_SIZE -
5053                 le16_to_cpu(xh->xh_name_value_len) - header_size;
5054
5055         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5056                         "of %u which exceed block size\n",
5057                         (unsigned long long)bucket_blkno(xs->bucket),
5058                         header_size);
5059
5060         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5061                 value_size = OCFS2_XATTR_ROOT_SIZE;
5062         else if (xi->value)
5063                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5064
5065         if (xs->not_found)
5066                 need = sizeof(struct ocfs2_xattr_entry) +
5067                         OCFS2_XATTR_SIZE(name_len) + value_size;
5068         else {
5069                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5070
5071                 /*
5072                  * We only replace the old value if the new length is smaller
5073                  * than the old one. Otherwise we will allocate new space in the
5074                  * bucket to store it.
5075                  */
5076                 xe = xs->here;
5077                 if (ocfs2_xattr_is_local(xe))
5078                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5079                 else
5080                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5081
5082                 if (old >= value_size)
5083                         need = 0;
5084         }
5085
5086         free = xh_free_start - header_size;
5087         /*
5088          * We need to make sure the new name/value pair
5089          * can exist in the same block.
5090          */
5091         if (xh_free_start % blocksize < need)
5092                 free -= xh_free_start % blocksize;
5093
5094         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5095              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5096              " %u\n", xs->not_found,
5097              (unsigned long long)bucket_blkno(xs->bucket),
5098              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5099              le16_to_cpu(xh->xh_name_value_len));
5100
5101         if (free < need ||
5102             (xs->not_found &&
5103              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5104                 if (need <= max_free &&
5105                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5106                         /*
5107                          * We can create the space by defragment. Since only the
5108                          * name/value will be moved, the xe shouldn't be changed
5109                          * in xs.
5110                          */
5111                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5112                                                         xs->bucket);
5113                         if (ret) {
5114                                 mlog_errno(ret);
5115                                 goto out;
5116                         }
5117
5118                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5119                         free = xh_free_start - header_size;
5120                         if (xh_free_start % blocksize < need)
5121                                 free -= xh_free_start % blocksize;
5122
5123                         if (free >= need)
5124                                 goto xattr_set;
5125
5126                         mlog(0, "Can't get enough space for xattr insert by "
5127                              "defragment. Need %u bytes, but we have %d, so "
5128                              "allocate new bucket for it.\n", need, free);
5129                 }
5130
5131                 /*
5132                  * We have to add new buckets or clusters and one
5133                  * allocation should leave us enough space for insert.
5134                  */
5135                 BUG_ON(allocation);
5136
5137                 /*
5138                  * We do not allow for overlapping ranges between buckets. And
5139                  * the maximum number of collisions we will allow for then is
5140                  * one bucket's worth, so check it here whether we need to
5141                  * add a new bucket for the insert.
5142                  */
5143                 ret = ocfs2_check_xattr_bucket_collision(inode,
5144                                                          xs->bucket,
5145                                                          xi->name);
5146                 if (ret) {
5147                         mlog_errno(ret);
5148                         goto out;
5149                 }
5150
5151                 ret = ocfs2_add_new_xattr_bucket(inode,
5152                                                  xs->xattr_bh,
5153                                                  xs->bucket,
5154                                                  ctxt);
5155                 if (ret) {
5156                         mlog_errno(ret);
5157                         goto out;
5158                 }
5159
5160                 /*
5161                  * ocfs2_add_new_xattr_bucket() will have updated
5162                  * xs->bucket if it moved, but it will not have updated
5163                  * any of the other search fields.  Thus, we drop it and
5164                  * re-search.  Everything should be cached, so it'll be
5165                  * quick.
5166                  */
5167                 ocfs2_xattr_bucket_relse(xs->bucket);
5168                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5169                                                    xi->name_index,
5170                                                    xi->name, xs);
5171                 if (ret && ret != -ENODATA)
5172                         goto out;
5173                 xs->not_found = ret;
5174                 allocation = 1;
5175                 goto try_again;
5176         }
5177
5178 xattr_set:
5179         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5180 out:
5181         mlog_exit(ret);
5182         return ret;
5183 }
5184
5185 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5186                                         struct ocfs2_xattr_bucket *bucket,
5187                                         void *para)
5188 {
5189         int ret = 0;
5190         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5191         u16 i;
5192         struct ocfs2_xattr_entry *xe;
5193         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5194         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5195         int credits = ocfs2_remove_extent_credits(osb->sb) +
5196                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5197
5198
5199         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5200
5201         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5202                 xe = &xh->xh_entries[i];
5203                 if (ocfs2_xattr_is_local(xe))
5204                         continue;
5205
5206                 ctxt.handle = ocfs2_start_trans(osb, credits);
5207                 if (IS_ERR(ctxt.handle)) {
5208                         ret = PTR_ERR(ctxt.handle);
5209                         mlog_errno(ret);
5210                         break;
5211                 }
5212
5213                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5214                                                         i, 0, &ctxt);
5215
5216                 ocfs2_commit_trans(osb, ctxt.handle);
5217                 if (ret) {
5218                         mlog_errno(ret);
5219                         break;
5220                 }
5221         }
5222
5223         ocfs2_schedule_truncate_log_flush(osb, 1);
5224         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5225         return ret;
5226 }
5227
5228 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5229                                           struct buffer_head *xb_bh)
5230 {
5231         struct ocfs2_xattr_block *xb =
5232                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5233         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5234         int ret = 0;
5235         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5236         u64 p_blkno;
5237
5238         if (le16_to_cpu(el->l_next_free_rec) == 0)
5239                 return 0;
5240
5241         while (name_hash > 0) {
5242                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5243                                           &e_cpos, &num_clusters, el);
5244                 if (ret) {
5245                         mlog_errno(ret);
5246                         goto out;
5247                 }
5248
5249                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5250                                                   ocfs2_delete_xattr_in_bucket,
5251                                                   NULL);
5252                 if (ret) {
5253                         mlog_errno(ret);
5254                         goto out;
5255                 }
5256
5257                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5258                                              p_blkno, e_cpos, num_clusters);
5259                 if (ret) {
5260                         mlog_errno(ret);
5261                         break;
5262                 }
5263
5264                 if (e_cpos == 0)
5265                         break;
5266
5267                 name_hash = e_cpos - 1;
5268         }
5269
5270 out:
5271         return ret;
5272 }
5273
5274 /*
5275  * 'security' attributes support
5276  */
5277 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5278                                         size_t list_size, const char *name,
5279                                         size_t name_len)
5280 {
5281         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5282         const size_t total_len = prefix_len + name_len + 1;
5283
5284         if (list && total_len <= list_size) {
5285                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5286                 memcpy(list + prefix_len, name, name_len);
5287                 list[prefix_len + name_len] = '\0';
5288         }
5289         return total_len;
5290 }
5291
5292 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5293                                     void *buffer, size_t size)
5294 {
5295         if (strcmp(name, "") == 0)
5296                 return -EINVAL;
5297         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5298                                buffer, size);
5299 }
5300
5301 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5302                                     const void *value, size_t size, int flags)
5303 {
5304         if (strcmp(name, "") == 0)
5305                 return -EINVAL;
5306
5307         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5308                                size, flags);
5309 }
5310
5311 int ocfs2_init_security_get(struct inode *inode,
5312                             struct inode *dir,
5313                             struct ocfs2_security_xattr_info *si)
5314 {
5315         return security_inode_init_security(inode, dir, &si->name, &si->value,
5316                                             &si->value_len);
5317 }
5318
5319 int ocfs2_init_security_set(handle_t *handle,
5320                             struct inode *inode,
5321                             struct buffer_head *di_bh,
5322                             struct ocfs2_security_xattr_info *si,
5323                             struct ocfs2_alloc_context *xattr_ac,
5324                             struct ocfs2_alloc_context *data_ac)
5325 {
5326         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5327                                      OCFS2_XATTR_INDEX_SECURITY,
5328                                      si->name, si->value, si->value_len, 0,
5329                                      xattr_ac, data_ac);
5330 }
5331
5332 struct xattr_handler ocfs2_xattr_security_handler = {
5333         .prefix = XATTR_SECURITY_PREFIX,
5334         .list   = ocfs2_xattr_security_list,
5335         .get    = ocfs2_xattr_security_get,
5336         .set    = ocfs2_xattr_security_set,
5337 };
5338
5339 /*
5340  * 'trusted' attributes support
5341  */
5342 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5343                                        size_t list_size, const char *name,
5344                                        size_t name_len)
5345 {
5346         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5347         const size_t total_len = prefix_len + name_len + 1;
5348
5349         if (list && total_len <= list_size) {
5350                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5351                 memcpy(list + prefix_len, name, name_len);
5352                 list[prefix_len + name_len] = '\0';
5353         }
5354         return total_len;
5355 }
5356
5357 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5358                                    void *buffer, size_t size)
5359 {
5360         if (strcmp(name, "") == 0)
5361                 return -EINVAL;
5362         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5363                                buffer, size);
5364 }
5365
5366 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5367                                    const void *value, size_t size, int flags)
5368 {
5369         if (strcmp(name, "") == 0)
5370                 return -EINVAL;
5371
5372         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5373                                size, flags);
5374 }
5375
5376 struct xattr_handler ocfs2_xattr_trusted_handler = {
5377         .prefix = XATTR_TRUSTED_PREFIX,
5378         .list   = ocfs2_xattr_trusted_list,
5379         .get    = ocfs2_xattr_trusted_get,
5380         .set    = ocfs2_xattr_trusted_set,
5381 };
5382
5383 /*
5384  * 'user' attributes support
5385  */
5386 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5387                                     size_t list_size, const char *name,
5388                                     size_t name_len)
5389 {
5390         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5391         const size_t total_len = prefix_len + name_len + 1;
5392         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5393
5394         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5395                 return 0;
5396
5397         if (list && total_len <= list_size) {
5398                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5399                 memcpy(list + prefix_len, name, name_len);
5400                 list[prefix_len + name_len] = '\0';
5401         }
5402         return total_len;
5403 }
5404
5405 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5406                                 void *buffer, size_t size)
5407 {
5408         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5409
5410         if (strcmp(name, "") == 0)
5411                 return -EINVAL;
5412         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5413                 return -EOPNOTSUPP;
5414         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5415                                buffer, size);
5416 }
5417
5418 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5419                                 const void *value, size_t size, int flags)
5420 {
5421         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5422
5423         if (strcmp(name, "") == 0)
5424                 return -EINVAL;
5425         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5426                 return -EOPNOTSUPP;
5427
5428         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5429                                size, flags);
5430 }
5431
5432 struct xattr_handler ocfs2_xattr_user_handler = {
5433         .prefix = XATTR_USER_PREFIX,
5434         .list   = ocfs2_xattr_user_list,
5435         .get    = ocfs2_xattr_user_get,
5436         .set    = ocfs2_xattr_user_set,
5437 };