1bf12c453f998b454531bb6f494c241246039d45
[linux-2.6.git] / fs / ocfs2 / xattr.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * xattr.c
5  *
6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
7  *
8  * CREDITS:
9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
11  *
12  * This program is free software; you can redistribute it and/or
13  * modify it under the terms of the GNU General Public
14  * License version 2 as published by the Free Software Foundation.
15  *
16  * This program is distributed in the hope that it will be useful,
17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
19  * General Public License for more details.
20  */
21
22 #include <linux/capability.h>
23 #include <linux/fs.h>
24 #include <linux/types.h>
25 #include <linux/slab.h>
26 #include <linux/highmem.h>
27 #include <linux/pagemap.h>
28 #include <linux/uio.h>
29 #include <linux/sched.h>
30 #include <linux/splice.h>
31 #include <linux/mount.h>
32 #include <linux/writeback.h>
33 #include <linux/falloc.h>
34 #include <linux/sort.h>
35 #include <linux/init.h>
36 #include <linux/module.h>
37 #include <linux/string.h>
38 #include <linux/security.h>
39
40 #define MLOG_MASK_PREFIX ML_XATTR
41 #include <cluster/masklog.h>
42
43 #include "ocfs2.h"
44 #include "alloc.h"
45 #include "blockcheck.h"
46 #include "dlmglue.h"
47 #include "file.h"
48 #include "symlink.h"
49 #include "sysfile.h"
50 #include "inode.h"
51 #include "journal.h"
52 #include "ocfs2_fs.h"
53 #include "suballoc.h"
54 #include "uptodate.h"
55 #include "buffer_head_io.h"
56 #include "super.h"
57 #include "xattr.h"
58
59
60 struct ocfs2_xattr_def_value_root {
61         struct ocfs2_xattr_value_root   xv;
62         struct ocfs2_extent_rec         er;
63 };
64
65 struct ocfs2_xattr_bucket {
66         /* The inode these xattrs are associated with */
67         struct inode *bu_inode;
68
69         /* The actual buffers that make up the bucket */
70         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
71
72         /* How many blocks make up one bucket for this filesystem */
73         int bu_blocks;
74 };
75
76 struct ocfs2_xattr_set_ctxt {
77         handle_t *handle;
78         struct ocfs2_alloc_context *meta_ac;
79         struct ocfs2_alloc_context *data_ac;
80         struct ocfs2_cached_dealloc_ctxt dealloc;
81 };
82
83 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
84 #define OCFS2_XATTR_INLINE_SIZE 80
85 #define OCFS2_XATTR_HEADER_GAP  4
86 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
87                                          - sizeof(struct ocfs2_xattr_header) \
88                                          - OCFS2_XATTR_HEADER_GAP)
89 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
90                                          - sizeof(struct ocfs2_xattr_block) \
91                                          - sizeof(struct ocfs2_xattr_header) \
92                                          - OCFS2_XATTR_HEADER_GAP)
93
94 static struct ocfs2_xattr_def_value_root def_xv = {
95         .xv.xr_list.l_count = cpu_to_le16(1),
96 };
97
98 struct xattr_handler *ocfs2_xattr_handlers[] = {
99         &ocfs2_xattr_user_handler,
100 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
101         &ocfs2_xattr_acl_access_handler,
102         &ocfs2_xattr_acl_default_handler,
103 #endif
104         &ocfs2_xattr_trusted_handler,
105         &ocfs2_xattr_security_handler,
106         NULL
107 };
108
109 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
111 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
112         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
113                                         = &ocfs2_xattr_acl_access_handler,
114         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
115                                         = &ocfs2_xattr_acl_default_handler,
116 #endif
117         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
118         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
119 };
120
121 struct ocfs2_xattr_info {
122         int name_index;
123         const char *name;
124         const void *value;
125         size_t value_len;
126 };
127
128 struct ocfs2_xattr_search {
129         struct buffer_head *inode_bh;
130         /*
131          * xattr_bh point to the block buffer head which has extended attribute
132          * when extended attribute in inode, xattr_bh is equal to inode_bh.
133          */
134         struct buffer_head *xattr_bh;
135         struct ocfs2_xattr_header *header;
136         struct ocfs2_xattr_bucket *bucket;
137         void *base;
138         void *end;
139         struct ocfs2_xattr_entry *here;
140         int not_found;
141 };
142
143 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
144                                              struct ocfs2_xattr_header *xh,
145                                              int index,
146                                              int *block_off,
147                                              int *new_offset);
148
149 static int ocfs2_xattr_block_find(struct inode *inode,
150                                   int name_index,
151                                   const char *name,
152                                   struct ocfs2_xattr_search *xs);
153 static int ocfs2_xattr_index_block_find(struct inode *inode,
154                                         struct buffer_head *root_bh,
155                                         int name_index,
156                                         const char *name,
157                                         struct ocfs2_xattr_search *xs);
158
159 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
160                                         struct ocfs2_xattr_tree_root *xt,
161                                         char *buffer,
162                                         size_t buffer_size);
163
164 static int ocfs2_xattr_create_index_block(struct inode *inode,
165                                           struct ocfs2_xattr_search *xs,
166                                           struct ocfs2_xattr_set_ctxt *ctxt);
167
168 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
169                                              struct ocfs2_xattr_info *xi,
170                                              struct ocfs2_xattr_search *xs,
171                                              struct ocfs2_xattr_set_ctxt *ctxt);
172
173 static int ocfs2_delete_xattr_index_block(struct inode *inode,
174                                           struct buffer_head *xb_bh);
175 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
176                                   u64 src_blk, u64 last_blk, u64 to_blk,
177                                   unsigned int start_bucket,
178                                   u32 *first_hash);
179
180 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
181 {
182         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
183 }
184
185 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
186 {
187         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
188 }
189
190 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
191 {
192         u16 len = sb->s_blocksize -
193                  offsetof(struct ocfs2_xattr_header, xh_entries);
194
195         return len / sizeof(struct ocfs2_xattr_entry);
196 }
197
198 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
199 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
200 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
201
202 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
203 {
204         struct ocfs2_xattr_bucket *bucket;
205         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
206
207         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
208
209         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
210         if (bucket) {
211                 bucket->bu_inode = inode;
212                 bucket->bu_blocks = blks;
213         }
214
215         return bucket;
216 }
217
218 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
219 {
220         int i;
221
222         for (i = 0; i < bucket->bu_blocks; i++) {
223                 brelse(bucket->bu_bhs[i]);
224                 bucket->bu_bhs[i] = NULL;
225         }
226 }
227
228 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
229 {
230         if (bucket) {
231                 ocfs2_xattr_bucket_relse(bucket);
232                 bucket->bu_inode = NULL;
233                 kfree(bucket);
234         }
235 }
236
237 /*
238  * A bucket that has never been written to disk doesn't need to be
239  * read.  We just need the buffer_heads.  Don't call this for
240  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
241  * them fully.
242  */
243 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
244                                    u64 xb_blkno)
245 {
246         int i, rc = 0;
247
248         for (i = 0; i < bucket->bu_blocks; i++) {
249                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
250                                               xb_blkno + i);
251                 if (!bucket->bu_bhs[i]) {
252                         rc = -EIO;
253                         mlog_errno(rc);
254                         break;
255                 }
256
257                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
258                                            bucket->bu_bhs[i]))
259                         ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
260                                                       bucket->bu_bhs[i]);
261         }
262
263         if (rc)
264                 ocfs2_xattr_bucket_relse(bucket);
265         return rc;
266 }
267
268 /* Read the xattr bucket at xb_blkno */
269 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
270                                    u64 xb_blkno)
271 {
272         int rc;
273
274         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
275                                bucket->bu_blocks, bucket->bu_bhs, 0,
276                                NULL);
277         if (!rc) {
278                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
279                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
280                                                  bucket->bu_bhs,
281                                                  bucket->bu_blocks,
282                                                  &bucket_xh(bucket)->xh_check);
283                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
284                 if (rc)
285                         mlog_errno(rc);
286         }
287
288         if (rc)
289                 ocfs2_xattr_bucket_relse(bucket);
290         return rc;
291 }
292
293 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
294                                              struct ocfs2_xattr_bucket *bucket,
295                                              int type)
296 {
297         int i, rc = 0;
298
299         for (i = 0; i < bucket->bu_blocks; i++) {
300                 rc = ocfs2_journal_access(handle,
301                                           INODE_CACHE(bucket->bu_inode),
302                                           bucket->bu_bhs[i], type);
303                 if (rc) {
304                         mlog_errno(rc);
305                         break;
306                 }
307         }
308
309         return rc;
310 }
311
312 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
313                                              struct ocfs2_xattr_bucket *bucket)
314 {
315         int i;
316
317         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
318         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
319                                    bucket->bu_bhs, bucket->bu_blocks,
320                                    &bucket_xh(bucket)->xh_check);
321         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
322
323         for (i = 0; i < bucket->bu_blocks; i++)
324                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
325 }
326
327 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
328                                          struct ocfs2_xattr_bucket *src)
329 {
330         int i;
331         int blocksize = src->bu_inode->i_sb->s_blocksize;
332
333         BUG_ON(dest->bu_blocks != src->bu_blocks);
334         BUG_ON(dest->bu_inode != src->bu_inode);
335
336         for (i = 0; i < src->bu_blocks; i++) {
337                 memcpy(bucket_block(dest, i), bucket_block(src, i),
338                        blocksize);
339         }
340 }
341
342 static int ocfs2_validate_xattr_block(struct super_block *sb,
343                                       struct buffer_head *bh)
344 {
345         int rc;
346         struct ocfs2_xattr_block *xb =
347                 (struct ocfs2_xattr_block *)bh->b_data;
348
349         mlog(0, "Validating xattr block %llu\n",
350              (unsigned long long)bh->b_blocknr);
351
352         BUG_ON(!buffer_uptodate(bh));
353
354         /*
355          * If the ecc fails, we return the error but otherwise
356          * leave the filesystem running.  We know any error is
357          * local to this block.
358          */
359         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
360         if (rc)
361                 return rc;
362
363         /*
364          * Errors after here are fatal
365          */
366
367         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
368                 ocfs2_error(sb,
369                             "Extended attribute block #%llu has bad "
370                             "signature %.*s",
371                             (unsigned long long)bh->b_blocknr, 7,
372                             xb->xb_signature);
373                 return -EINVAL;
374         }
375
376         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
377                 ocfs2_error(sb,
378                             "Extended attribute block #%llu has an "
379                             "invalid xb_blkno of %llu",
380                             (unsigned long long)bh->b_blocknr,
381                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
382                 return -EINVAL;
383         }
384
385         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
386                 ocfs2_error(sb,
387                             "Extended attribute block #%llu has an invalid "
388                             "xb_fs_generation of #%u",
389                             (unsigned long long)bh->b_blocknr,
390                             le32_to_cpu(xb->xb_fs_generation));
391                 return -EINVAL;
392         }
393
394         return 0;
395 }
396
397 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
398                                   struct buffer_head **bh)
399 {
400         int rc;
401         struct buffer_head *tmp = *bh;
402
403         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
404                               ocfs2_validate_xattr_block);
405
406         /* If ocfs2_read_block() got us a new bh, pass it up. */
407         if (!rc && !*bh)
408                 *bh = tmp;
409
410         return rc;
411 }
412
413 static inline const char *ocfs2_xattr_prefix(int name_index)
414 {
415         struct xattr_handler *handler = NULL;
416
417         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
418                 handler = ocfs2_xattr_handler_map[name_index];
419
420         return handler ? handler->prefix : NULL;
421 }
422
423 static u32 ocfs2_xattr_name_hash(struct inode *inode,
424                                  const char *name,
425                                  int name_len)
426 {
427         /* Get hash value of uuid from super block */
428         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
429         int i;
430
431         /* hash extended attribute name */
432         for (i = 0; i < name_len; i++) {
433                 hash = (hash << OCFS2_HASH_SHIFT) ^
434                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
435                        *name++;
436         }
437
438         return hash;
439 }
440
441 /*
442  * ocfs2_xattr_hash_entry()
443  *
444  * Compute the hash of an extended attribute.
445  */
446 static void ocfs2_xattr_hash_entry(struct inode *inode,
447                                    struct ocfs2_xattr_header *header,
448                                    struct ocfs2_xattr_entry *entry)
449 {
450         u32 hash = 0;
451         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
452
453         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
454         entry->xe_name_hash = cpu_to_le32(hash);
455
456         return;
457 }
458
459 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
460 {
461         int size = 0;
462
463         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
464                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
465         else
466                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
467         size += sizeof(struct ocfs2_xattr_entry);
468
469         return size;
470 }
471
472 int ocfs2_calc_security_init(struct inode *dir,
473                              struct ocfs2_security_xattr_info *si,
474                              int *want_clusters,
475                              int *xattr_credits,
476                              struct ocfs2_alloc_context **xattr_ac)
477 {
478         int ret = 0;
479         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
480         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
481                                                  si->value_len);
482
483         /*
484          * The max space of security xattr taken inline is
485          * 256(name) + 80(value) + 16(entry) = 352 bytes,
486          * So reserve one metadata block for it is ok.
487          */
488         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
489             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
490                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
491                 if (ret) {
492                         mlog_errno(ret);
493                         return ret;
494                 }
495                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
496         }
497
498         /* reserve clusters for xattr value which will be set in B tree*/
499         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
500                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
501                                                             si->value_len);
502
503                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
504                                                            new_clusters);
505                 *want_clusters += new_clusters;
506         }
507         return ret;
508 }
509
510 int ocfs2_calc_xattr_init(struct inode *dir,
511                           struct buffer_head *dir_bh,
512                           int mode,
513                           struct ocfs2_security_xattr_info *si,
514                           int *want_clusters,
515                           int *xattr_credits,
516                           int *want_meta)
517 {
518         int ret = 0;
519         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
520         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
521
522         if (si->enable)
523                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
524                                                      si->value_len);
525
526         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
527                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
528                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
529                                         "", NULL, 0);
530                 if (acl_len > 0) {
531                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
532                         if (S_ISDIR(mode))
533                                 a_size <<= 1;
534                 } else if (acl_len != 0 && acl_len != -ENODATA) {
535                         mlog_errno(ret);
536                         return ret;
537                 }
538         }
539
540         if (!(s_size + a_size))
541                 return ret;
542
543         /*
544          * The max space of security xattr taken inline is
545          * 256(name) + 80(value) + 16(entry) = 352 bytes,
546          * The max space of acl xattr taken inline is
547          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
548          * when blocksize = 512, may reserve one more cluser for
549          * xattr bucket, otherwise reserve one metadata block
550          * for them is ok.
551          * If this is a new directory with inline data,
552          * we choose to reserve the entire inline area for
553          * directory contents and force an external xattr block.
554          */
555         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
556             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
557             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
558                 *want_meta = *want_meta + 1;
559                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
560         }
561
562         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
563             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
564                 *want_clusters += 1;
565                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
566         }
567
568         /*
569          * reserve credits and clusters for xattrs which has large value
570          * and have to be set outside
571          */
572         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
573                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
574                                                         si->value_len);
575                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
576                                                            new_clusters);
577                 *want_clusters += new_clusters;
578         }
579         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
580             acl_len > OCFS2_XATTR_INLINE_SIZE) {
581                 /* for directory, it has DEFAULT and ACCESS two types of acls */
582                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
583                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
584                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
585                                                            new_clusters);
586                 *want_clusters += new_clusters;
587         }
588
589         return ret;
590 }
591
592 static int ocfs2_xattr_extend_allocation(struct inode *inode,
593                                          u32 clusters_to_add,
594                                          struct ocfs2_xattr_value_buf *vb,
595                                          struct ocfs2_xattr_set_ctxt *ctxt)
596 {
597         int status = 0;
598         handle_t *handle = ctxt->handle;
599         enum ocfs2_alloc_restarted why;
600         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
601         struct ocfs2_extent_tree et;
602
603         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
604
605         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
606
607         status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
608                               OCFS2_JOURNAL_ACCESS_WRITE);
609         if (status < 0) {
610                 mlog_errno(status);
611                 goto leave;
612         }
613
614         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
615         status = ocfs2_add_clusters_in_btree(handle,
616                                              &et,
617                                              &logical_start,
618                                              clusters_to_add,
619                                              0,
620                                              ctxt->data_ac,
621                                              ctxt->meta_ac,
622                                              &why);
623         if (status < 0) {
624                 mlog_errno(status);
625                 goto leave;
626         }
627
628         status = ocfs2_journal_dirty(handle, vb->vb_bh);
629         if (status < 0) {
630                 mlog_errno(status);
631                 goto leave;
632         }
633
634         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
635
636         /*
637          * We should have already allocated enough space before the transaction,
638          * so no need to restart.
639          */
640         BUG_ON(why != RESTART_NONE || clusters_to_add);
641
642 leave:
643
644         return status;
645 }
646
647 static int __ocfs2_remove_xattr_range(struct inode *inode,
648                                       struct ocfs2_xattr_value_buf *vb,
649                                       u32 cpos, u32 phys_cpos, u32 len,
650                                       struct ocfs2_xattr_set_ctxt *ctxt)
651 {
652         int ret;
653         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
654         handle_t *handle = ctxt->handle;
655         struct ocfs2_extent_tree et;
656
657         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
658
659         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
660                             OCFS2_JOURNAL_ACCESS_WRITE);
661         if (ret) {
662                 mlog_errno(ret);
663                 goto out;
664         }
665
666         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
667                                   &ctxt->dealloc);
668         if (ret) {
669                 mlog_errno(ret);
670                 goto out;
671         }
672
673         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
674
675         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
676         if (ret) {
677                 mlog_errno(ret);
678                 goto out;
679         }
680
681         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
682         if (ret)
683                 mlog_errno(ret);
684
685 out:
686         return ret;
687 }
688
689 static int ocfs2_xattr_shrink_size(struct inode *inode,
690                                    u32 old_clusters,
691                                    u32 new_clusters,
692                                    struct ocfs2_xattr_value_buf *vb,
693                                    struct ocfs2_xattr_set_ctxt *ctxt)
694 {
695         int ret = 0;
696         u32 trunc_len, cpos, phys_cpos, alloc_size;
697         u64 block;
698
699         if (old_clusters <= new_clusters)
700                 return 0;
701
702         cpos = new_clusters;
703         trunc_len = old_clusters - new_clusters;
704         while (trunc_len) {
705                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
706                                                &alloc_size,
707                                                &vb->vb_xv->xr_list);
708                 if (ret) {
709                         mlog_errno(ret);
710                         goto out;
711                 }
712
713                 if (alloc_size > trunc_len)
714                         alloc_size = trunc_len;
715
716                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
717                                                  phys_cpos, alloc_size,
718                                                  ctxt);
719                 if (ret) {
720                         mlog_errno(ret);
721                         goto out;
722                 }
723
724                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
725                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
726                                                        block, alloc_size);
727                 cpos += alloc_size;
728                 trunc_len -= alloc_size;
729         }
730
731 out:
732         return ret;
733 }
734
735 static int ocfs2_xattr_value_truncate(struct inode *inode,
736                                       struct ocfs2_xattr_value_buf *vb,
737                                       int len,
738                                       struct ocfs2_xattr_set_ctxt *ctxt)
739 {
740         int ret;
741         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
742         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
743
744         if (new_clusters == old_clusters)
745                 return 0;
746
747         if (new_clusters > old_clusters)
748                 ret = ocfs2_xattr_extend_allocation(inode,
749                                                     new_clusters - old_clusters,
750                                                     vb, ctxt);
751         else
752                 ret = ocfs2_xattr_shrink_size(inode,
753                                               old_clusters, new_clusters,
754                                               vb, ctxt);
755
756         return ret;
757 }
758
759 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
760                                   size_t *result, const char *prefix,
761                                   const char *name, int name_len)
762 {
763         char *p = buffer + *result;
764         int prefix_len = strlen(prefix);
765         int total_len = prefix_len + name_len + 1;
766
767         *result += total_len;
768
769         /* we are just looking for how big our buffer needs to be */
770         if (!size)
771                 return 0;
772
773         if (*result > size)
774                 return -ERANGE;
775
776         memcpy(p, prefix, prefix_len);
777         memcpy(p + prefix_len, name, name_len);
778         p[prefix_len + name_len] = '\0';
779
780         return 0;
781 }
782
783 static int ocfs2_xattr_list_entries(struct inode *inode,
784                                     struct ocfs2_xattr_header *header,
785                                     char *buffer, size_t buffer_size)
786 {
787         size_t result = 0;
788         int i, type, ret;
789         const char *prefix, *name;
790
791         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
792                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
793                 type = ocfs2_xattr_get_type(entry);
794                 prefix = ocfs2_xattr_prefix(type);
795
796                 if (prefix) {
797                         name = (const char *)header +
798                                 le16_to_cpu(entry->xe_name_offset);
799
800                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
801                                                      &result, prefix, name,
802                                                      entry->xe_name_len);
803                         if (ret)
804                                 return ret;
805                 }
806         }
807
808         return result;
809 }
810
811 static int ocfs2_xattr_ibody_list(struct inode *inode,
812                                   struct ocfs2_dinode *di,
813                                   char *buffer,
814                                   size_t buffer_size)
815 {
816         struct ocfs2_xattr_header *header = NULL;
817         struct ocfs2_inode_info *oi = OCFS2_I(inode);
818         int ret = 0;
819
820         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
821                 return ret;
822
823         header = (struct ocfs2_xattr_header *)
824                  ((void *)di + inode->i_sb->s_blocksize -
825                  le16_to_cpu(di->i_xattr_inline_size));
826
827         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
828
829         return ret;
830 }
831
832 static int ocfs2_xattr_block_list(struct inode *inode,
833                                   struct ocfs2_dinode *di,
834                                   char *buffer,
835                                   size_t buffer_size)
836 {
837         struct buffer_head *blk_bh = NULL;
838         struct ocfs2_xattr_block *xb;
839         int ret = 0;
840
841         if (!di->i_xattr_loc)
842                 return ret;
843
844         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
845                                      &blk_bh);
846         if (ret < 0) {
847                 mlog_errno(ret);
848                 return ret;
849         }
850
851         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
852         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
853                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
854                 ret = ocfs2_xattr_list_entries(inode, header,
855                                                buffer, buffer_size);
856         } else {
857                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
858                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
859                                                    buffer, buffer_size);
860         }
861
862         brelse(blk_bh);
863
864         return ret;
865 }
866
867 ssize_t ocfs2_listxattr(struct dentry *dentry,
868                         char *buffer,
869                         size_t size)
870 {
871         int ret = 0, i_ret = 0, b_ret = 0;
872         struct buffer_head *di_bh = NULL;
873         struct ocfs2_dinode *di = NULL;
874         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
875
876         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
877                 return -EOPNOTSUPP;
878
879         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
880                 return ret;
881
882         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
883         if (ret < 0) {
884                 mlog_errno(ret);
885                 return ret;
886         }
887
888         di = (struct ocfs2_dinode *)di_bh->b_data;
889
890         down_read(&oi->ip_xattr_sem);
891         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
892         if (i_ret < 0)
893                 b_ret = 0;
894         else {
895                 if (buffer) {
896                         buffer += i_ret;
897                         size -= i_ret;
898                 }
899                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
900                                                buffer, size);
901                 if (b_ret < 0)
902                         i_ret = 0;
903         }
904         up_read(&oi->ip_xattr_sem);
905         ocfs2_inode_unlock(dentry->d_inode, 0);
906
907         brelse(di_bh);
908
909         return i_ret + b_ret;
910 }
911
912 static int ocfs2_xattr_find_entry(int name_index,
913                                   const char *name,
914                                   struct ocfs2_xattr_search *xs)
915 {
916         struct ocfs2_xattr_entry *entry;
917         size_t name_len;
918         int i, cmp = 1;
919
920         if (name == NULL)
921                 return -EINVAL;
922
923         name_len = strlen(name);
924         entry = xs->here;
925         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
926                 cmp = name_index - ocfs2_xattr_get_type(entry);
927                 if (!cmp)
928                         cmp = name_len - entry->xe_name_len;
929                 if (!cmp)
930                         cmp = memcmp(name, (xs->base +
931                                      le16_to_cpu(entry->xe_name_offset)),
932                                      name_len);
933                 if (cmp == 0)
934                         break;
935                 entry += 1;
936         }
937         xs->here = entry;
938
939         return cmp ? -ENODATA : 0;
940 }
941
942 static int ocfs2_xattr_get_value_outside(struct inode *inode,
943                                          struct ocfs2_xattr_value_root *xv,
944                                          void *buffer,
945                                          size_t len)
946 {
947         u32 cpos, p_cluster, num_clusters, bpc, clusters;
948         u64 blkno;
949         int i, ret = 0;
950         size_t cplen, blocksize;
951         struct buffer_head *bh = NULL;
952         struct ocfs2_extent_list *el;
953
954         el = &xv->xr_list;
955         clusters = le32_to_cpu(xv->xr_clusters);
956         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
957         blocksize = inode->i_sb->s_blocksize;
958
959         cpos = 0;
960         while (cpos < clusters) {
961                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
962                                                &num_clusters, el);
963                 if (ret) {
964                         mlog_errno(ret);
965                         goto out;
966                 }
967
968                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
969                 /* Copy ocfs2_xattr_value */
970                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
971                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
972                                                &bh, NULL);
973                         if (ret) {
974                                 mlog_errno(ret);
975                                 goto out;
976                         }
977
978                         cplen = len >= blocksize ? blocksize : len;
979                         memcpy(buffer, bh->b_data, cplen);
980                         len -= cplen;
981                         buffer += cplen;
982
983                         brelse(bh);
984                         bh = NULL;
985                         if (len == 0)
986                                 break;
987                 }
988                 cpos += num_clusters;
989         }
990 out:
991         return ret;
992 }
993
994 static int ocfs2_xattr_ibody_get(struct inode *inode,
995                                  int name_index,
996                                  const char *name,
997                                  void *buffer,
998                                  size_t buffer_size,
999                                  struct ocfs2_xattr_search *xs)
1000 {
1001         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1002         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1003         struct ocfs2_xattr_value_root *xv;
1004         size_t size;
1005         int ret = 0;
1006
1007         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1008                 return -ENODATA;
1009
1010         xs->end = (void *)di + inode->i_sb->s_blocksize;
1011         xs->header = (struct ocfs2_xattr_header *)
1012                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1013         xs->base = (void *)xs->header;
1014         xs->here = xs->header->xh_entries;
1015
1016         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1017         if (ret)
1018                 return ret;
1019         size = le64_to_cpu(xs->here->xe_value_size);
1020         if (buffer) {
1021                 if (size > buffer_size)
1022                         return -ERANGE;
1023                 if (ocfs2_xattr_is_local(xs->here)) {
1024                         memcpy(buffer, (void *)xs->base +
1025                                le16_to_cpu(xs->here->xe_name_offset) +
1026                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1027                 } else {
1028                         xv = (struct ocfs2_xattr_value_root *)
1029                                 (xs->base + le16_to_cpu(
1030                                  xs->here->xe_name_offset) +
1031                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1032                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1033                                                             buffer, size);
1034                         if (ret < 0) {
1035                                 mlog_errno(ret);
1036                                 return ret;
1037                         }
1038                 }
1039         }
1040
1041         return size;
1042 }
1043
1044 static int ocfs2_xattr_block_get(struct inode *inode,
1045                                  int name_index,
1046                                  const char *name,
1047                                  void *buffer,
1048                                  size_t buffer_size,
1049                                  struct ocfs2_xattr_search *xs)
1050 {
1051         struct ocfs2_xattr_block *xb;
1052         struct ocfs2_xattr_value_root *xv;
1053         size_t size;
1054         int ret = -ENODATA, name_offset, name_len, i;
1055         int uninitialized_var(block_off);
1056
1057         xs->bucket = ocfs2_xattr_bucket_new(inode);
1058         if (!xs->bucket) {
1059                 ret = -ENOMEM;
1060                 mlog_errno(ret);
1061                 goto cleanup;
1062         }
1063
1064         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1065         if (ret) {
1066                 mlog_errno(ret);
1067                 goto cleanup;
1068         }
1069
1070         if (xs->not_found) {
1071                 ret = -ENODATA;
1072                 goto cleanup;
1073         }
1074
1075         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1076         size = le64_to_cpu(xs->here->xe_value_size);
1077         if (buffer) {
1078                 ret = -ERANGE;
1079                 if (size > buffer_size)
1080                         goto cleanup;
1081
1082                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1083                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1084                 i = xs->here - xs->header->xh_entries;
1085
1086                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1087                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1088                                                                 bucket_xh(xs->bucket),
1089                                                                 i,
1090                                                                 &block_off,
1091                                                                 &name_offset);
1092                         xs->base = bucket_block(xs->bucket, block_off);
1093                 }
1094                 if (ocfs2_xattr_is_local(xs->here)) {
1095                         memcpy(buffer, (void *)xs->base +
1096                                name_offset + name_len, size);
1097                 } else {
1098                         xv = (struct ocfs2_xattr_value_root *)
1099                                 (xs->base + name_offset + name_len);
1100                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1101                                                             buffer, size);
1102                         if (ret < 0) {
1103                                 mlog_errno(ret);
1104                                 goto cleanup;
1105                         }
1106                 }
1107         }
1108         ret = size;
1109 cleanup:
1110         ocfs2_xattr_bucket_free(xs->bucket);
1111
1112         brelse(xs->xattr_bh);
1113         xs->xattr_bh = NULL;
1114         return ret;
1115 }
1116
1117 int ocfs2_xattr_get_nolock(struct inode *inode,
1118                            struct buffer_head *di_bh,
1119                            int name_index,
1120                            const char *name,
1121                            void *buffer,
1122                            size_t buffer_size)
1123 {
1124         int ret;
1125         struct ocfs2_dinode *di = NULL;
1126         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1127         struct ocfs2_xattr_search xis = {
1128                 .not_found = -ENODATA,
1129         };
1130         struct ocfs2_xattr_search xbs = {
1131                 .not_found = -ENODATA,
1132         };
1133
1134         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1135                 return -EOPNOTSUPP;
1136
1137         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1138                 ret = -ENODATA;
1139
1140         xis.inode_bh = xbs.inode_bh = di_bh;
1141         di = (struct ocfs2_dinode *)di_bh->b_data;
1142
1143         down_read(&oi->ip_xattr_sem);
1144         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1145                                     buffer_size, &xis);
1146         if (ret == -ENODATA && di->i_xattr_loc)
1147                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1148                                             buffer_size, &xbs);
1149         up_read(&oi->ip_xattr_sem);
1150
1151         return ret;
1152 }
1153
1154 /* ocfs2_xattr_get()
1155  *
1156  * Copy an extended attribute into the buffer provided.
1157  * Buffer is NULL to compute the size of buffer required.
1158  */
1159 static int ocfs2_xattr_get(struct inode *inode,
1160                            int name_index,
1161                            const char *name,
1162                            void *buffer,
1163                            size_t buffer_size)
1164 {
1165         int ret;
1166         struct buffer_head *di_bh = NULL;
1167
1168         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1169         if (ret < 0) {
1170                 mlog_errno(ret);
1171                 return ret;
1172         }
1173         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1174                                      name, buffer, buffer_size);
1175
1176         ocfs2_inode_unlock(inode, 0);
1177
1178         brelse(di_bh);
1179
1180         return ret;
1181 }
1182
1183 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1184                                            handle_t *handle,
1185                                            struct ocfs2_xattr_value_root *xv,
1186                                            const void *value,
1187                                            int value_len)
1188 {
1189         int ret = 0, i, cp_len;
1190         u16 blocksize = inode->i_sb->s_blocksize;
1191         u32 p_cluster, num_clusters;
1192         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1193         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1194         u64 blkno;
1195         struct buffer_head *bh = NULL;
1196
1197         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1198
1199         while (cpos < clusters) {
1200                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1201                                                &num_clusters, &xv->xr_list);
1202                 if (ret) {
1203                         mlog_errno(ret);
1204                         goto out;
1205                 }
1206
1207                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1208
1209                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1210                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1211                                                &bh, NULL);
1212                         if (ret) {
1213                                 mlog_errno(ret);
1214                                 goto out;
1215                         }
1216
1217                         ret = ocfs2_journal_access(handle,
1218                                                    INODE_CACHE(inode),
1219                                                    bh,
1220                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1221                         if (ret < 0) {
1222                                 mlog_errno(ret);
1223                                 goto out;
1224                         }
1225
1226                         cp_len = value_len > blocksize ? blocksize : value_len;
1227                         memcpy(bh->b_data, value, cp_len);
1228                         value_len -= cp_len;
1229                         value += cp_len;
1230                         if (cp_len < blocksize)
1231                                 memset(bh->b_data + cp_len, 0,
1232                                        blocksize - cp_len);
1233
1234                         ret = ocfs2_journal_dirty(handle, bh);
1235                         if (ret < 0) {
1236                                 mlog_errno(ret);
1237                                 goto out;
1238                         }
1239                         brelse(bh);
1240                         bh = NULL;
1241
1242                         /*
1243                          * XXX: do we need to empty all the following
1244                          * blocks in this cluster?
1245                          */
1246                         if (!value_len)
1247                                 break;
1248                 }
1249                 cpos += num_clusters;
1250         }
1251 out:
1252         brelse(bh);
1253
1254         return ret;
1255 }
1256
1257 static int ocfs2_xattr_cleanup(struct inode *inode,
1258                                handle_t *handle,
1259                                struct ocfs2_xattr_info *xi,
1260                                struct ocfs2_xattr_search *xs,
1261                                struct ocfs2_xattr_value_buf *vb,
1262                                size_t offs)
1263 {
1264         int ret = 0;
1265         size_t name_len = strlen(xi->name);
1266         void *val = xs->base + offs;
1267         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1268
1269         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1270                             OCFS2_JOURNAL_ACCESS_WRITE);
1271         if (ret) {
1272                 mlog_errno(ret);
1273                 goto out;
1274         }
1275         /* Decrease xattr count */
1276         le16_add_cpu(&xs->header->xh_count, -1);
1277         /* Remove the xattr entry and tree root which has already be set*/
1278         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1279         memset(val, 0, size);
1280
1281         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1282         if (ret < 0)
1283                 mlog_errno(ret);
1284 out:
1285         return ret;
1286 }
1287
1288 static int ocfs2_xattr_update_entry(struct inode *inode,
1289                                     handle_t *handle,
1290                                     struct ocfs2_xattr_info *xi,
1291                                     struct ocfs2_xattr_search *xs,
1292                                     struct ocfs2_xattr_value_buf *vb,
1293                                     size_t offs)
1294 {
1295         int ret;
1296
1297         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1298                             OCFS2_JOURNAL_ACCESS_WRITE);
1299         if (ret) {
1300                 mlog_errno(ret);
1301                 goto out;
1302         }
1303
1304         xs->here->xe_name_offset = cpu_to_le16(offs);
1305         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1306         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1307                 ocfs2_xattr_set_local(xs->here, 1);
1308         else
1309                 ocfs2_xattr_set_local(xs->here, 0);
1310         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1311
1312         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1313         if (ret < 0)
1314                 mlog_errno(ret);
1315 out:
1316         return ret;
1317 }
1318
1319 /*
1320  * ocfs2_xattr_set_value_outside()
1321  *
1322  * Set large size value in B tree.
1323  */
1324 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1325                                          struct ocfs2_xattr_info *xi,
1326                                          struct ocfs2_xattr_search *xs,
1327                                          struct ocfs2_xattr_set_ctxt *ctxt,
1328                                          struct ocfs2_xattr_value_buf *vb,
1329                                          size_t offs)
1330 {
1331         size_t name_len = strlen(xi->name);
1332         void *val = xs->base + offs;
1333         struct ocfs2_xattr_value_root *xv = NULL;
1334         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1335         int ret = 0;
1336
1337         memset(val, 0, size);
1338         memcpy(val, xi->name, name_len);
1339         xv = (struct ocfs2_xattr_value_root *)
1340                 (val + OCFS2_XATTR_SIZE(name_len));
1341         xv->xr_clusters = 0;
1342         xv->xr_last_eb_blk = 0;
1343         xv->xr_list.l_tree_depth = 0;
1344         xv->xr_list.l_count = cpu_to_le16(1);
1345         xv->xr_list.l_next_free_rec = 0;
1346         vb->vb_xv = xv;
1347
1348         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1349         if (ret < 0) {
1350                 mlog_errno(ret);
1351                 return ret;
1352         }
1353         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1354         if (ret < 0) {
1355                 mlog_errno(ret);
1356                 return ret;
1357         }
1358         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1359                                               xi->value, xi->value_len);
1360         if (ret < 0)
1361                 mlog_errno(ret);
1362
1363         return ret;
1364 }
1365
1366 /*
1367  * ocfs2_xattr_set_entry_local()
1368  *
1369  * Set, replace or remove extended attribute in local.
1370  */
1371 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1372                                         struct ocfs2_xattr_info *xi,
1373                                         struct ocfs2_xattr_search *xs,
1374                                         struct ocfs2_xattr_entry *last,
1375                                         size_t min_offs)
1376 {
1377         size_t name_len = strlen(xi->name);
1378         int i;
1379
1380         if (xi->value && xs->not_found) {
1381                 /* Insert the new xattr entry. */
1382                 le16_add_cpu(&xs->header->xh_count, 1);
1383                 ocfs2_xattr_set_type(last, xi->name_index);
1384                 ocfs2_xattr_set_local(last, 1);
1385                 last->xe_name_len = name_len;
1386         } else {
1387                 void *first_val;
1388                 void *val;
1389                 size_t offs, size;
1390
1391                 first_val = xs->base + min_offs;
1392                 offs = le16_to_cpu(xs->here->xe_name_offset);
1393                 val = xs->base + offs;
1394
1395                 if (le64_to_cpu(xs->here->xe_value_size) >
1396                     OCFS2_XATTR_INLINE_SIZE)
1397                         size = OCFS2_XATTR_SIZE(name_len) +
1398                                 OCFS2_XATTR_ROOT_SIZE;
1399                 else
1400                         size = OCFS2_XATTR_SIZE(name_len) +
1401                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1402
1403                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1404                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1405                         /* The old and the new value have the
1406                            same size. Just replace the value. */
1407                         ocfs2_xattr_set_local(xs->here, 1);
1408                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1409                         /* Clear value bytes. */
1410                         memset(val + OCFS2_XATTR_SIZE(name_len),
1411                                0,
1412                                OCFS2_XATTR_SIZE(xi->value_len));
1413                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1414                                xi->value,
1415                                xi->value_len);
1416                         return;
1417                 }
1418                 /* Remove the old name+value. */
1419                 memmove(first_val + size, first_val, val - first_val);
1420                 memset(first_val, 0, size);
1421                 xs->here->xe_name_hash = 0;
1422                 xs->here->xe_name_offset = 0;
1423                 ocfs2_xattr_set_local(xs->here, 1);
1424                 xs->here->xe_value_size = 0;
1425
1426                 min_offs += size;
1427
1428                 /* Adjust all value offsets. */
1429                 last = xs->header->xh_entries;
1430                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1431                         size_t o = le16_to_cpu(last->xe_name_offset);
1432
1433                         if (o < offs)
1434                                 last->xe_name_offset = cpu_to_le16(o + size);
1435                         last += 1;
1436                 }
1437
1438                 if (!xi->value) {
1439                         /* Remove the old entry. */
1440                         last -= 1;
1441                         memmove(xs->here, xs->here + 1,
1442                                 (void *)last - (void *)xs->here);
1443                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1444                         le16_add_cpu(&xs->header->xh_count, -1);
1445                 }
1446         }
1447         if (xi->value) {
1448                 /* Insert the new name+value. */
1449                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1450                                 OCFS2_XATTR_SIZE(xi->value_len);
1451                 void *val = xs->base + min_offs - size;
1452
1453                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1454                 memset(val, 0, size);
1455                 memcpy(val, xi->name, name_len);
1456                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1457                        xi->value,
1458                        xi->value_len);
1459                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1460                 ocfs2_xattr_set_local(xs->here, 1);
1461                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1462         }
1463
1464         return;
1465 }
1466
1467 /*
1468  * ocfs2_xattr_set_entry()
1469  *
1470  * Set extended attribute entry into inode or block.
1471  *
1472  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1473  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1474  * then set value in B tree with set_value_outside().
1475  */
1476 static int ocfs2_xattr_set_entry(struct inode *inode,
1477                                  struct ocfs2_xattr_info *xi,
1478                                  struct ocfs2_xattr_search *xs,
1479                                  struct ocfs2_xattr_set_ctxt *ctxt,
1480                                  int flag)
1481 {
1482         struct ocfs2_xattr_entry *last;
1483         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1484         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1485         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1486         size_t size_l = 0;
1487         handle_t *handle = ctxt->handle;
1488         int free, i, ret;
1489         struct ocfs2_xattr_info xi_l = {
1490                 .name_index = xi->name_index,
1491                 .name = xi->name,
1492                 .value = xi->value,
1493                 .value_len = xi->value_len,
1494         };
1495         struct ocfs2_xattr_value_buf vb = {
1496                 .vb_bh = xs->xattr_bh,
1497                 .vb_access = ocfs2_journal_access_di,
1498         };
1499
1500         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1501                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1502                 vb.vb_access = ocfs2_journal_access_xb;
1503         } else
1504                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1505
1506         /* Compute min_offs, last and free space. */
1507         last = xs->header->xh_entries;
1508
1509         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1510                 size_t offs = le16_to_cpu(last->xe_name_offset);
1511                 if (offs < min_offs)
1512                         min_offs = offs;
1513                 last += 1;
1514         }
1515
1516         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1517         if (free < 0)
1518                 return -EIO;
1519
1520         if (!xs->not_found) {
1521                 size_t size = 0;
1522                 if (ocfs2_xattr_is_local(xs->here))
1523                         size = OCFS2_XATTR_SIZE(name_len) +
1524                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1525                 else
1526                         size = OCFS2_XATTR_SIZE(name_len) +
1527                                 OCFS2_XATTR_ROOT_SIZE;
1528                 free += (size + sizeof(struct ocfs2_xattr_entry));
1529         }
1530         /* Check free space in inode or block */
1531         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1532                 if (free < sizeof(struct ocfs2_xattr_entry) +
1533                            OCFS2_XATTR_SIZE(name_len) +
1534                            OCFS2_XATTR_ROOT_SIZE) {
1535                         ret = -ENOSPC;
1536                         goto out;
1537                 }
1538                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1539                 xi_l.value = (void *)&def_xv;
1540                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1541         } else if (xi->value) {
1542                 if (free < sizeof(struct ocfs2_xattr_entry) +
1543                            OCFS2_XATTR_SIZE(name_len) +
1544                            OCFS2_XATTR_SIZE(xi->value_len)) {
1545                         ret = -ENOSPC;
1546                         goto out;
1547                 }
1548         }
1549
1550         if (!xs->not_found) {
1551                 /* For existing extended attribute */
1552                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1553                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1554                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1555                 void *val = xs->base + offs;
1556
1557                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1558                         /* Replace existing local xattr with tree root */
1559                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1560                                                             ctxt, &vb, offs);
1561                         if (ret < 0)
1562                                 mlog_errno(ret);
1563                         goto out;
1564                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1565                         /* For existing xattr which has value outside */
1566                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1567                                 (val + OCFS2_XATTR_SIZE(name_len));
1568
1569                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1570                                 /*
1571                                  * If new value need set outside also,
1572                                  * first truncate old value to new value,
1573                                  * then set new value with set_value_outside().
1574                                  */
1575                                 ret = ocfs2_xattr_value_truncate(inode,
1576                                                                  &vb,
1577                                                                  xi->value_len,
1578                                                                  ctxt);
1579                                 if (ret < 0) {
1580                                         mlog_errno(ret);
1581                                         goto out;
1582                                 }
1583
1584                                 ret = ocfs2_xattr_update_entry(inode,
1585                                                                handle,
1586                                                                xi,
1587                                                                xs,
1588                                                                &vb,
1589                                                                offs);
1590                                 if (ret < 0) {
1591                                         mlog_errno(ret);
1592                                         goto out;
1593                                 }
1594
1595                                 ret = __ocfs2_xattr_set_value_outside(inode,
1596                                                                 handle,
1597                                                                 vb.vb_xv,
1598                                                                 xi->value,
1599                                                                 xi->value_len);
1600                                 if (ret < 0)
1601                                         mlog_errno(ret);
1602                                 goto out;
1603                         } else {
1604                                 /*
1605                                  * If new value need set in local,
1606                                  * just trucate old value to zero.
1607                                  */
1608                                  ret = ocfs2_xattr_value_truncate(inode,
1609                                                                   &vb,
1610                                                                   0,
1611                                                                   ctxt);
1612                                 if (ret < 0)
1613                                         mlog_errno(ret);
1614                         }
1615                 }
1616         }
1617
1618         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1619                                       OCFS2_JOURNAL_ACCESS_WRITE);
1620         if (ret) {
1621                 mlog_errno(ret);
1622                 goto out;
1623         }
1624
1625         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1626                 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1627                                    OCFS2_JOURNAL_ACCESS_WRITE);
1628                 if (ret) {
1629                         mlog_errno(ret);
1630                         goto out;
1631                 }
1632         }
1633
1634         /*
1635          * Set value in local, include set tree root in local.
1636          * This is the first step for value size >INLINE_SIZE.
1637          */
1638         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1639
1640         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1641                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1642                 if (ret < 0) {
1643                         mlog_errno(ret);
1644                         goto out;
1645                 }
1646         }
1647
1648         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1649             (flag & OCFS2_INLINE_XATTR_FL)) {
1650                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1651                 unsigned int xattrsize = osb->s_xattr_inline_size;
1652
1653                 /*
1654                  * Adjust extent record count or inline data size
1655                  * to reserve space for extended attribute.
1656                  */
1657                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1658                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1659                         le16_add_cpu(&idata->id_count, -xattrsize);
1660                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1661                         struct ocfs2_extent_list *el = &di->id2.i_list;
1662                         le16_add_cpu(&el->l_count, -(xattrsize /
1663                                         sizeof(struct ocfs2_extent_rec)));
1664                 }
1665                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1666         }
1667         /* Update xattr flag */
1668         spin_lock(&oi->ip_lock);
1669         oi->ip_dyn_features |= flag;
1670         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1671         spin_unlock(&oi->ip_lock);
1672
1673         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1674         if (ret < 0)
1675                 mlog_errno(ret);
1676
1677         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1678                 /*
1679                  * Set value outside in B tree.
1680                  * This is the second step for value size > INLINE_SIZE.
1681                  */
1682                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1683                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1684                                                     &vb, offs);
1685                 if (ret < 0) {
1686                         int ret2;
1687
1688                         mlog_errno(ret);
1689                         /*
1690                          * If set value outside failed, we have to clean
1691                          * the junk tree root we have already set in local.
1692                          */
1693                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1694                                                    xi, xs, &vb, offs);
1695                         if (ret2 < 0)
1696                                 mlog_errno(ret2);
1697                 }
1698         }
1699 out:
1700         return ret;
1701 }
1702
1703 static int ocfs2_remove_value_outside(struct inode*inode,
1704                                       struct ocfs2_xattr_value_buf *vb,
1705                                       struct ocfs2_xattr_header *header)
1706 {
1707         int ret = 0, i;
1708         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1709         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1710
1711         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1712
1713         ctxt.handle = ocfs2_start_trans(osb,
1714                                         ocfs2_remove_extent_credits(osb->sb));
1715         if (IS_ERR(ctxt.handle)) {
1716                 ret = PTR_ERR(ctxt.handle);
1717                 mlog_errno(ret);
1718                 goto out;
1719         }
1720
1721         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1722                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1723
1724                 if (!ocfs2_xattr_is_local(entry)) {
1725                         void *val;
1726
1727                         val = (void *)header +
1728                                 le16_to_cpu(entry->xe_name_offset);
1729                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1730                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1731                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1732                         if (ret < 0) {
1733                                 mlog_errno(ret);
1734                                 break;
1735                         }
1736                 }
1737         }
1738
1739         ocfs2_commit_trans(osb, ctxt.handle);
1740         ocfs2_schedule_truncate_log_flush(osb, 1);
1741         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1742 out:
1743         return ret;
1744 }
1745
1746 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1747                                     struct buffer_head *di_bh)
1748 {
1749
1750         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1751         struct ocfs2_xattr_header *header;
1752         int ret;
1753         struct ocfs2_xattr_value_buf vb = {
1754                 .vb_bh = di_bh,
1755                 .vb_access = ocfs2_journal_access_di,
1756         };
1757
1758         header = (struct ocfs2_xattr_header *)
1759                  ((void *)di + inode->i_sb->s_blocksize -
1760                  le16_to_cpu(di->i_xattr_inline_size));
1761
1762         ret = ocfs2_remove_value_outside(inode, &vb, header);
1763
1764         return ret;
1765 }
1766
1767 static int ocfs2_xattr_block_remove(struct inode *inode,
1768                                     struct buffer_head *blk_bh)
1769 {
1770         struct ocfs2_xattr_block *xb;
1771         int ret = 0;
1772         struct ocfs2_xattr_value_buf vb = {
1773                 .vb_bh = blk_bh,
1774                 .vb_access = ocfs2_journal_access_xb,
1775         };
1776
1777         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1778         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1779                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1780                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1781         } else
1782                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1783
1784         return ret;
1785 }
1786
1787 static int ocfs2_xattr_free_block(struct inode *inode,
1788                                   u64 block)
1789 {
1790         struct inode *xb_alloc_inode;
1791         struct buffer_head *xb_alloc_bh = NULL;
1792         struct buffer_head *blk_bh = NULL;
1793         struct ocfs2_xattr_block *xb;
1794         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1795         handle_t *handle;
1796         int ret = 0;
1797         u64 blk, bg_blkno;
1798         u16 bit;
1799
1800         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1801         if (ret < 0) {
1802                 mlog_errno(ret);
1803                 goto out;
1804         }
1805
1806         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1807         if (ret < 0) {
1808                 mlog_errno(ret);
1809                 goto out;
1810         }
1811
1812         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1813         blk = le64_to_cpu(xb->xb_blkno);
1814         bit = le16_to_cpu(xb->xb_suballoc_bit);
1815         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1816
1817         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1818                                 EXTENT_ALLOC_SYSTEM_INODE,
1819                                 le16_to_cpu(xb->xb_suballoc_slot));
1820         if (!xb_alloc_inode) {
1821                 ret = -ENOMEM;
1822                 mlog_errno(ret);
1823                 goto out;
1824         }
1825         mutex_lock(&xb_alloc_inode->i_mutex);
1826
1827         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1828         if (ret < 0) {
1829                 mlog_errno(ret);
1830                 goto out_mutex;
1831         }
1832
1833         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1834         if (IS_ERR(handle)) {
1835                 ret = PTR_ERR(handle);
1836                 mlog_errno(ret);
1837                 goto out_unlock;
1838         }
1839
1840         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1841                                        bit, bg_blkno, 1);
1842         if (ret < 0)
1843                 mlog_errno(ret);
1844
1845         ocfs2_commit_trans(osb, handle);
1846 out_unlock:
1847         ocfs2_inode_unlock(xb_alloc_inode, 1);
1848         brelse(xb_alloc_bh);
1849 out_mutex:
1850         mutex_unlock(&xb_alloc_inode->i_mutex);
1851         iput(xb_alloc_inode);
1852 out:
1853         brelse(blk_bh);
1854         return ret;
1855 }
1856
1857 /*
1858  * ocfs2_xattr_remove()
1859  *
1860  * Free extended attribute resources associated with this inode.
1861  */
1862 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1863 {
1864         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1865         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1866         handle_t *handle;
1867         int ret;
1868
1869         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1870                 return 0;
1871
1872         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1873                 return 0;
1874
1875         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1876                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1877                 if (ret < 0) {
1878                         mlog_errno(ret);
1879                         goto out;
1880                 }
1881         }
1882
1883         if (di->i_xattr_loc) {
1884                 ret = ocfs2_xattr_free_block(inode,
1885                                              le64_to_cpu(di->i_xattr_loc));
1886                 if (ret < 0) {
1887                         mlog_errno(ret);
1888                         goto out;
1889                 }
1890         }
1891
1892         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1893                                    OCFS2_INODE_UPDATE_CREDITS);
1894         if (IS_ERR(handle)) {
1895                 ret = PTR_ERR(handle);
1896                 mlog_errno(ret);
1897                 goto out;
1898         }
1899         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1900                                       OCFS2_JOURNAL_ACCESS_WRITE);
1901         if (ret) {
1902                 mlog_errno(ret);
1903                 goto out_commit;
1904         }
1905
1906         di->i_xattr_loc = 0;
1907
1908         spin_lock(&oi->ip_lock);
1909         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1910         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1911         spin_unlock(&oi->ip_lock);
1912
1913         ret = ocfs2_journal_dirty(handle, di_bh);
1914         if (ret < 0)
1915                 mlog_errno(ret);
1916 out_commit:
1917         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1918 out:
1919         return ret;
1920 }
1921
1922 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1923                                         struct ocfs2_dinode *di)
1924 {
1925         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1926         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1927         int free;
1928
1929         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1930                 return 0;
1931
1932         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1933                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1934                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1935         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1936                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1937                         le64_to_cpu(di->i_size);
1938         } else {
1939                 struct ocfs2_extent_list *el = &di->id2.i_list;
1940                 free = (le16_to_cpu(el->l_count) -
1941                         le16_to_cpu(el->l_next_free_rec)) *
1942                         sizeof(struct ocfs2_extent_rec);
1943         }
1944         if (free >= xattrsize)
1945                 return 1;
1946
1947         return 0;
1948 }
1949
1950 /*
1951  * ocfs2_xattr_ibody_find()
1952  *
1953  * Find extended attribute in inode block and
1954  * fill search info into struct ocfs2_xattr_search.
1955  */
1956 static int ocfs2_xattr_ibody_find(struct inode *inode,
1957                                   int name_index,
1958                                   const char *name,
1959                                   struct ocfs2_xattr_search *xs)
1960 {
1961         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1962         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1963         int ret;
1964         int has_space = 0;
1965
1966         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1967                 return 0;
1968
1969         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1970                 down_read(&oi->ip_alloc_sem);
1971                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1972                 up_read(&oi->ip_alloc_sem);
1973                 if (!has_space)
1974                         return 0;
1975         }
1976
1977         xs->xattr_bh = xs->inode_bh;
1978         xs->end = (void *)di + inode->i_sb->s_blocksize;
1979         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1980                 xs->header = (struct ocfs2_xattr_header *)
1981                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1982         else
1983                 xs->header = (struct ocfs2_xattr_header *)
1984                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1985         xs->base = (void *)xs->header;
1986         xs->here = xs->header->xh_entries;
1987
1988         /* Find the named attribute. */
1989         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1990                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1991                 if (ret && ret != -ENODATA)
1992                         return ret;
1993                 xs->not_found = ret;
1994         }
1995
1996         return 0;
1997 }
1998
1999 /*
2000  * ocfs2_xattr_ibody_set()
2001  *
2002  * Set, replace or remove an extended attribute into inode block.
2003  *
2004  */
2005 static int ocfs2_xattr_ibody_set(struct inode *inode,
2006                                  struct ocfs2_xattr_info *xi,
2007                                  struct ocfs2_xattr_search *xs,
2008                                  struct ocfs2_xattr_set_ctxt *ctxt)
2009 {
2010         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2011         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2012         int ret;
2013
2014         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2015                 return -ENOSPC;
2016
2017         down_write(&oi->ip_alloc_sem);
2018         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2019                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2020                         ret = -ENOSPC;
2021                         goto out;
2022                 }
2023         }
2024
2025         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2026                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2027 out:
2028         up_write(&oi->ip_alloc_sem);
2029
2030         return ret;
2031 }
2032
2033 /*
2034  * ocfs2_xattr_block_find()
2035  *
2036  * Find extended attribute in external block and
2037  * fill search info into struct ocfs2_xattr_search.
2038  */
2039 static int ocfs2_xattr_block_find(struct inode *inode,
2040                                   int name_index,
2041                                   const char *name,
2042                                   struct ocfs2_xattr_search *xs)
2043 {
2044         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2045         struct buffer_head *blk_bh = NULL;
2046         struct ocfs2_xattr_block *xb;
2047         int ret = 0;
2048
2049         if (!di->i_xattr_loc)
2050                 return ret;
2051
2052         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2053                                      &blk_bh);
2054         if (ret < 0) {
2055                 mlog_errno(ret);
2056                 return ret;
2057         }
2058
2059         xs->xattr_bh = blk_bh;
2060         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2061
2062         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2063                 xs->header = &xb->xb_attrs.xb_header;
2064                 xs->base = (void *)xs->header;
2065                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2066                 xs->here = xs->header->xh_entries;
2067
2068                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2069         } else
2070                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2071                                                    name_index,
2072                                                    name, xs);
2073
2074         if (ret && ret != -ENODATA) {
2075                 xs->xattr_bh = NULL;
2076                 goto cleanup;
2077         }
2078         xs->not_found = ret;
2079         return 0;
2080 cleanup:
2081         brelse(blk_bh);
2082
2083         return ret;
2084 }
2085
2086 /*
2087  * ocfs2_xattr_block_set()
2088  *
2089  * Set, replace or remove an extended attribute into external block.
2090  *
2091  */
2092 static int ocfs2_xattr_block_set(struct inode *inode,
2093                                  struct ocfs2_xattr_info *xi,
2094                                  struct ocfs2_xattr_search *xs,
2095                                  struct ocfs2_xattr_set_ctxt *ctxt)
2096 {
2097         struct buffer_head *new_bh = NULL;
2098         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2099         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2100         handle_t *handle = ctxt->handle;
2101         struct ocfs2_xattr_block *xblk = NULL;
2102         u16 suballoc_bit_start;
2103         u32 num_got;
2104         u64 first_blkno;
2105         int ret;
2106
2107         if (!xs->xattr_bh) {
2108                 ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode),
2109                                               xs->inode_bh,
2110                                               OCFS2_JOURNAL_ACCESS_CREATE);
2111                 if (ret < 0) {
2112                         mlog_errno(ret);
2113                         goto end;
2114                 }
2115
2116                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2117                                            &suballoc_bit_start, &num_got,
2118                                            &first_blkno);
2119                 if (ret < 0) {
2120                         mlog_errno(ret);
2121                         goto end;
2122                 }
2123
2124                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2125                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2126
2127                 ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2128                                               new_bh,
2129                                               OCFS2_JOURNAL_ACCESS_CREATE);
2130                 if (ret < 0) {
2131                         mlog_errno(ret);
2132                         goto end;
2133                 }
2134
2135                 /* Initialize ocfs2_xattr_block */
2136                 xs->xattr_bh = new_bh;
2137                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2138                 memset(xblk, 0, inode->i_sb->s_blocksize);
2139                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2140                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2141                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2142                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2143                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2144
2145                 xs->header = &xblk->xb_attrs.xb_header;
2146                 xs->base = (void *)xs->header;
2147                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2148                 xs->here = xs->header->xh_entries;
2149
2150                 ret = ocfs2_journal_dirty(handle, new_bh);
2151                 if (ret < 0) {
2152                         mlog_errno(ret);
2153                         goto end;
2154                 }
2155                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2156                 ocfs2_journal_dirty(handle, xs->inode_bh);
2157         } else
2158                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2159
2160         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2161                 /* Set extended attribute into external block */
2162                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2163                                             OCFS2_HAS_XATTR_FL);
2164                 if (!ret || ret != -ENOSPC)
2165                         goto end;
2166
2167                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2168                 if (ret)
2169                         goto end;
2170         }
2171
2172         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2173
2174 end:
2175
2176         return ret;
2177 }
2178
2179 /* Check whether the new xattr can be inserted into the inode. */
2180 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2181                                        struct ocfs2_xattr_info *xi,
2182                                        struct ocfs2_xattr_search *xs)
2183 {
2184         u64 value_size;
2185         struct ocfs2_xattr_entry *last;
2186         int free, i;
2187         size_t min_offs = xs->end - xs->base;
2188
2189         if (!xs->header)
2190                 return 0;
2191
2192         last = xs->header->xh_entries;
2193
2194         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2195                 size_t offs = le16_to_cpu(last->xe_name_offset);
2196                 if (offs < min_offs)
2197                         min_offs = offs;
2198                 last += 1;
2199         }
2200
2201         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2202         if (free < 0)
2203                 return 0;
2204
2205         BUG_ON(!xs->not_found);
2206
2207         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2208                 value_size = OCFS2_XATTR_ROOT_SIZE;
2209         else
2210                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2211
2212         if (free >= sizeof(struct ocfs2_xattr_entry) +
2213                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2214                 return 1;
2215
2216         return 0;
2217 }
2218
2219 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2220                                      struct ocfs2_dinode *di,
2221                                      struct ocfs2_xattr_info *xi,
2222                                      struct ocfs2_xattr_search *xis,
2223                                      struct ocfs2_xattr_search *xbs,
2224                                      int *clusters_need,
2225                                      int *meta_need,
2226                                      int *credits_need)
2227 {
2228         int ret = 0, old_in_xb = 0;
2229         int clusters_add = 0, meta_add = 0, credits = 0;
2230         struct buffer_head *bh = NULL;
2231         struct ocfs2_xattr_block *xb = NULL;
2232         struct ocfs2_xattr_entry *xe = NULL;
2233         struct ocfs2_xattr_value_root *xv = NULL;
2234         char *base = NULL;
2235         int name_offset, name_len = 0;
2236         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2237                                                     xi->value_len);
2238         u64 value_size;
2239
2240         /*
2241          * Calculate the clusters we need to write.
2242          * No matter whether we replace an old one or add a new one,
2243          * we need this for writing.
2244          */
2245         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2246                 credits += new_clusters *
2247                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2248
2249         if (xis->not_found && xbs->not_found) {
2250                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2251
2252                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2253                         clusters_add += new_clusters;
2254                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2255                                                         &def_xv.xv.xr_list,
2256                                                         new_clusters);
2257                 }
2258
2259                 goto meta_guess;
2260         }
2261
2262         if (!xis->not_found) {
2263                 xe = xis->here;
2264                 name_offset = le16_to_cpu(xe->xe_name_offset);
2265                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2266                 base = xis->base;
2267                 credits += OCFS2_INODE_UPDATE_CREDITS;
2268         } else {
2269                 int i, block_off = 0;
2270                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2271                 xe = xbs->here;
2272                 name_offset = le16_to_cpu(xe->xe_name_offset);
2273                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2274                 i = xbs->here - xbs->header->xh_entries;
2275                 old_in_xb = 1;
2276
2277                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2278                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2279                                                         bucket_xh(xbs->bucket),
2280                                                         i, &block_off,
2281                                                         &name_offset);
2282                         base = bucket_block(xbs->bucket, block_off);
2283                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2284                 } else {
2285                         base = xbs->base;
2286                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2287                 }
2288         }
2289
2290         /*
2291          * delete a xattr doesn't need metadata and cluster allocation.
2292          * so just calculate the credits and return.
2293          *
2294          * The credits for removing the value tree will be extended
2295          * by ocfs2_remove_extent itself.
2296          */
2297         if (!xi->value) {
2298                 if (!ocfs2_xattr_is_local(xe))
2299                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2300
2301                 goto out;
2302         }
2303
2304         /* do cluster allocation guess first. */
2305         value_size = le64_to_cpu(xe->xe_value_size);
2306
2307         if (old_in_xb) {
2308                 /*
2309                  * In xattr set, we always try to set the xe in inode first,
2310                  * so if it can be inserted into inode successfully, the old
2311                  * one will be removed from the xattr block, and this xattr
2312                  * will be inserted into inode as a new xattr in inode.
2313                  */
2314                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2315                         clusters_add += new_clusters;
2316                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2317                                     OCFS2_INODE_UPDATE_CREDITS;
2318                         if (!ocfs2_xattr_is_local(xe))
2319                                 credits += ocfs2_calc_extend_credits(
2320                                                         inode->i_sb,
2321                                                         &def_xv.xv.xr_list,
2322                                                         new_clusters);
2323                         goto out;
2324                 }
2325         }
2326
2327         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2328                 /* the new values will be stored outside. */
2329                 u32 old_clusters = 0;
2330
2331                 if (!ocfs2_xattr_is_local(xe)) {
2332                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2333                                                                  value_size);
2334                         xv = (struct ocfs2_xattr_value_root *)
2335                              (base + name_offset + name_len);
2336                         value_size = OCFS2_XATTR_ROOT_SIZE;
2337                 } else
2338                         xv = &def_xv.xv;
2339
2340                 if (old_clusters >= new_clusters) {
2341                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2342                         goto out;
2343                 } else {
2344                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2345                         clusters_add += new_clusters - old_clusters;
2346                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2347                                                              &xv->xr_list,
2348                                                              new_clusters -
2349                                                              old_clusters);
2350                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2351                                 goto out;
2352                 }
2353         } else {
2354                 /*
2355                  * Now the new value will be stored inside. So if the new
2356                  * value is smaller than the size of value root or the old
2357                  * value, we don't need any allocation, otherwise we have
2358                  * to guess metadata allocation.
2359                  */
2360                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2361                     (!ocfs2_xattr_is_local(xe) &&
2362                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2363                         goto out;
2364         }
2365
2366 meta_guess:
2367         /* calculate metadata allocation. */
2368         if (di->i_xattr_loc) {
2369                 if (!xbs->xattr_bh) {
2370                         ret = ocfs2_read_xattr_block(inode,
2371                                                      le64_to_cpu(di->i_xattr_loc),
2372                                                      &bh);
2373                         if (ret) {
2374                                 mlog_errno(ret);
2375                                 goto out;
2376                         }
2377
2378                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2379                 } else
2380                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2381
2382                 /*
2383                  * If there is already an xattr tree, good, we can calculate
2384                  * like other b-trees. Otherwise we may have the chance of
2385                  * create a tree, the credit calculation is borrowed from
2386                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2387                  * new tree will be cluster based, so no meta is needed.
2388                  */
2389                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2390                         struct ocfs2_extent_list *el =
2391                                  &xb->xb_attrs.xb_root.xt_list;
2392                         meta_add += ocfs2_extend_meta_needed(el);
2393                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2394                                                              el, 1);
2395                 } else
2396                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2397
2398                 /*
2399                  * This cluster will be used either for new bucket or for
2400                  * new xattr block.
2401                  * If the cluster size is the same as the bucket size, one
2402                  * more is needed since we may need to extend the bucket
2403                  * also.
2404                  */
2405                 clusters_add += 1;
2406                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2407                 if (OCFS2_XATTR_BUCKET_SIZE ==
2408                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2409                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2410                         clusters_add += 1;
2411                 }
2412         } else {
2413                 meta_add += 1;
2414                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2415         }
2416 out:
2417         if (clusters_need)
2418                 *clusters_need = clusters_add;
2419         if (meta_need)
2420                 *meta_need = meta_add;
2421         if (credits_need)
2422                 *credits_need = credits;
2423         brelse(bh);
2424         return ret;
2425 }
2426
2427 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2428                                      struct ocfs2_dinode *di,
2429                                      struct ocfs2_xattr_info *xi,
2430                                      struct ocfs2_xattr_search *xis,
2431                                      struct ocfs2_xattr_search *xbs,
2432                                      struct ocfs2_xattr_set_ctxt *ctxt,
2433                                      int *credits)
2434 {
2435         int clusters_add, meta_add, ret;
2436         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2437
2438         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2439
2440         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2441
2442         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2443                                         &clusters_add, &meta_add, credits);
2444         if (ret) {
2445                 mlog_errno(ret);
2446                 return ret;
2447         }
2448
2449         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2450              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2451
2452         if (meta_add) {
2453                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2454                                                         &ctxt->meta_ac);
2455                 if (ret) {
2456                         mlog_errno(ret);
2457                         goto out;
2458                 }
2459         }
2460
2461         if (clusters_add) {
2462                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2463                 if (ret)
2464                         mlog_errno(ret);
2465         }
2466 out:
2467         if (ret) {
2468                 if (ctxt->meta_ac) {
2469                         ocfs2_free_alloc_context(ctxt->meta_ac);
2470                         ctxt->meta_ac = NULL;
2471                 }
2472
2473                 /*
2474                  * We cannot have an error and a non null ctxt->data_ac.
2475                  */
2476         }
2477
2478         return ret;
2479 }
2480
2481 static int __ocfs2_xattr_set_handle(struct inode *inode,
2482                                     struct ocfs2_dinode *di,
2483                                     struct ocfs2_xattr_info *xi,
2484                                     struct ocfs2_xattr_search *xis,
2485                                     struct ocfs2_xattr_search *xbs,
2486                                     struct ocfs2_xattr_set_ctxt *ctxt)
2487 {
2488         int ret = 0, credits, old_found;
2489
2490         if (!xi->value) {
2491                 /* Remove existing extended attribute */
2492                 if (!xis->not_found)
2493                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2494                 else if (!xbs->not_found)
2495                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2496         } else {
2497                 /* We always try to set extended attribute into inode first*/
2498                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2499                 if (!ret && !xbs->not_found) {
2500                         /*
2501                          * If succeed and that extended attribute existing in
2502                          * external block, then we will remove it.
2503                          */
2504                         xi->value = NULL;
2505                         xi->value_len = 0;
2506
2507                         old_found = xis->not_found;
2508                         xis->not_found = -ENODATA;
2509                         ret = ocfs2_calc_xattr_set_need(inode,
2510                                                         di,
2511                                                         xi,
2512                                                         xis,
2513                                                         xbs,
2514                                                         NULL,
2515                                                         NULL,
2516                                                         &credits);
2517                         xis->not_found = old_found;
2518                         if (ret) {
2519                                 mlog_errno(ret);
2520                                 goto out;
2521                         }
2522
2523                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2524                                         ctxt->handle->h_buffer_credits);
2525                         if (ret) {
2526                                 mlog_errno(ret);
2527                                 goto out;
2528                         }
2529                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2530                 } else if (ret == -ENOSPC) {
2531                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2532                                 ret = ocfs2_xattr_block_find(inode,
2533                                                              xi->name_index,
2534                                                              xi->name, xbs);
2535                                 if (ret)
2536                                         goto out;
2537
2538                                 old_found = xis->not_found;
2539                                 xis->not_found = -ENODATA;
2540                                 ret = ocfs2_calc_xattr_set_need(inode,
2541                                                                 di,
2542                                                                 xi,
2543                                                                 xis,
2544                                                                 xbs,
2545                                                                 NULL,
2546                                                                 NULL,
2547                                                                 &credits);
2548                                 xis->not_found = old_found;
2549                                 if (ret) {
2550                                         mlog_errno(ret);
2551                                         goto out;
2552                                 }
2553
2554                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2555                                         ctxt->handle->h_buffer_credits);
2556                                 if (ret) {
2557                                         mlog_errno(ret);
2558                                         goto out;
2559                                 }
2560                         }
2561                         /*
2562                          * If no space in inode, we will set extended attribute
2563                          * into external block.
2564                          */
2565                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2566                         if (ret)
2567                                 goto out;
2568                         if (!xis->not_found) {
2569                                 /*
2570                                  * If succeed and that extended attribute
2571                                  * existing in inode, we will remove it.
2572                                  */
2573                                 xi->value = NULL;
2574                                 xi->value_len = 0;
2575                                 xbs->not_found = -ENODATA;
2576                                 ret = ocfs2_calc_xattr_set_need(inode,
2577                                                                 di,
2578                                                                 xi,
2579                                                                 xis,
2580                                                                 xbs,
2581                                                                 NULL,
2582                                                                 NULL,
2583                                                                 &credits);
2584                                 if (ret) {
2585                                         mlog_errno(ret);
2586                                         goto out;
2587                                 }
2588
2589                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2590                                                 ctxt->handle->h_buffer_credits);
2591                                 if (ret) {
2592                                         mlog_errno(ret);
2593                                         goto out;
2594                                 }
2595                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2596                                                             xis, ctxt);
2597                         }
2598                 }
2599         }
2600
2601         if (!ret) {
2602                 /* Update inode ctime. */
2603                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2604                                               xis->inode_bh,
2605                                               OCFS2_JOURNAL_ACCESS_WRITE);
2606                 if (ret) {
2607                         mlog_errno(ret);
2608                         goto out;
2609                 }
2610
2611                 inode->i_ctime = CURRENT_TIME;
2612                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2613                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2614                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2615         }
2616 out:
2617         return ret;
2618 }
2619
2620 /*
2621  * This function only called duing creating inode
2622  * for init security/acl xattrs of the new inode.
2623  * All transanction credits have been reserved in mknod.
2624  */
2625 int ocfs2_xattr_set_handle(handle_t *handle,
2626                            struct inode *inode,
2627                            struct buffer_head *di_bh,
2628                            int name_index,
2629                            const char *name,
2630                            const void *value,
2631                            size_t value_len,
2632                            int flags,
2633                            struct ocfs2_alloc_context *meta_ac,
2634                            struct ocfs2_alloc_context *data_ac)
2635 {
2636         struct ocfs2_dinode *di;
2637         int ret;
2638
2639         struct ocfs2_xattr_info xi = {
2640                 .name_index = name_index,
2641                 .name = name,
2642                 .value = value,
2643                 .value_len = value_len,
2644         };
2645
2646         struct ocfs2_xattr_search xis = {
2647                 .not_found = -ENODATA,
2648         };
2649
2650         struct ocfs2_xattr_search xbs = {
2651                 .not_found = -ENODATA,
2652         };
2653
2654         struct ocfs2_xattr_set_ctxt ctxt = {
2655                 .handle = handle,
2656                 .meta_ac = meta_ac,
2657                 .data_ac = data_ac,
2658         };
2659
2660         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2661                 return -EOPNOTSUPP;
2662
2663         /*
2664          * In extreme situation, may need xattr bucket when
2665          * block size is too small. And we have already reserved
2666          * the credits for bucket in mknod.
2667          */
2668         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2669                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2670                 if (!xbs.bucket) {
2671                         mlog_errno(-ENOMEM);
2672                         return -ENOMEM;
2673                 }
2674         }
2675
2676         xis.inode_bh = xbs.inode_bh = di_bh;
2677         di = (struct ocfs2_dinode *)di_bh->b_data;
2678
2679         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2680
2681         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2682         if (ret)
2683                 goto cleanup;
2684         if (xis.not_found) {
2685                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2686                 if (ret)
2687                         goto cleanup;
2688         }
2689
2690         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2691
2692 cleanup:
2693         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2694         brelse(xbs.xattr_bh);
2695         ocfs2_xattr_bucket_free(xbs.bucket);
2696
2697         return ret;
2698 }
2699
2700 /*
2701  * ocfs2_xattr_set()
2702  *
2703  * Set, replace or remove an extended attribute for this inode.
2704  * value is NULL to remove an existing extended attribute, else either
2705  * create or replace an extended attribute.
2706  */
2707 int ocfs2_xattr_set(struct inode *inode,
2708                     int name_index,
2709                     const char *name,
2710                     const void *value,
2711                     size_t value_len,
2712                     int flags)
2713 {
2714         struct buffer_head *di_bh = NULL;
2715         struct ocfs2_dinode *di;
2716         int ret, credits;
2717         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2718         struct inode *tl_inode = osb->osb_tl_inode;
2719         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2720
2721         struct ocfs2_xattr_info xi = {
2722                 .name_index = name_index,
2723                 .name = name,
2724                 .value = value,
2725                 .value_len = value_len,
2726         };
2727
2728         struct ocfs2_xattr_search xis = {
2729                 .not_found = -ENODATA,
2730         };
2731
2732         struct ocfs2_xattr_search xbs = {
2733                 .not_found = -ENODATA,
2734         };
2735
2736         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2737                 return -EOPNOTSUPP;
2738
2739         /*
2740          * Only xbs will be used on indexed trees.  xis doesn't need a
2741          * bucket.
2742          */
2743         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2744         if (!xbs.bucket) {
2745                 mlog_errno(-ENOMEM);
2746                 return -ENOMEM;
2747         }
2748
2749         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2750         if (ret < 0) {
2751                 mlog_errno(ret);
2752                 goto cleanup_nolock;
2753         }
2754         xis.inode_bh = xbs.inode_bh = di_bh;
2755         di = (struct ocfs2_dinode *)di_bh->b_data;
2756
2757         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2758         /*
2759          * Scan inode and external block to find the same name
2760          * extended attribute and collect search infomation.
2761          */
2762         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2763         if (ret)
2764                 goto cleanup;
2765         if (xis.not_found) {
2766                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2767                 if (ret)
2768                         goto cleanup;
2769         }
2770
2771         if (xis.not_found && xbs.not_found) {
2772                 ret = -ENODATA;
2773                 if (flags & XATTR_REPLACE)
2774                         goto cleanup;
2775                 ret = 0;
2776                 if (!value)
2777                         goto cleanup;
2778         } else {
2779                 ret = -EEXIST;
2780                 if (flags & XATTR_CREATE)
2781                         goto cleanup;
2782         }
2783
2784
2785         mutex_lock(&tl_inode->i_mutex);
2786
2787         if (ocfs2_truncate_log_needs_flush(osb)) {
2788                 ret = __ocfs2_flush_truncate_log(osb);
2789                 if (ret < 0) {
2790                         mutex_unlock(&tl_inode->i_mutex);
2791                         mlog_errno(ret);
2792                         goto cleanup;
2793                 }
2794         }
2795         mutex_unlock(&tl_inode->i_mutex);
2796
2797         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2798                                         &xbs, &ctxt, &credits);
2799         if (ret) {
2800                 mlog_errno(ret);
2801                 goto cleanup;
2802         }
2803
2804         /* we need to update inode's ctime field, so add credit for it. */
2805         credits += OCFS2_INODE_UPDATE_CREDITS;
2806         ctxt.handle = ocfs2_start_trans(osb, credits);
2807         if (IS_ERR(ctxt.handle)) {
2808                 ret = PTR_ERR(ctxt.handle);
2809                 mlog_errno(ret);
2810                 goto cleanup;
2811         }
2812
2813         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2814
2815         ocfs2_commit_trans(osb, ctxt.handle);
2816
2817         if (ctxt.data_ac)
2818                 ocfs2_free_alloc_context(ctxt.data_ac);
2819         if (ctxt.meta_ac)
2820                 ocfs2_free_alloc_context(ctxt.meta_ac);
2821         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2822                 ocfs2_schedule_truncate_log_flush(osb, 1);
2823         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2824 cleanup:
2825         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2826         ocfs2_inode_unlock(inode, 1);
2827 cleanup_nolock:
2828         brelse(di_bh);
2829         brelse(xbs.xattr_bh);
2830         ocfs2_xattr_bucket_free(xbs.bucket);
2831
2832         return ret;
2833 }
2834
2835 /*
2836  * Find the xattr extent rec which may contains name_hash.
2837  * e_cpos will be the first name hash of the xattr rec.
2838  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2839  */
2840 static int ocfs2_xattr_get_rec(struct inode *inode,
2841                                u32 name_hash,
2842                                u64 *p_blkno,
2843                                u32 *e_cpos,
2844                                u32 *num_clusters,
2845                                struct ocfs2_extent_list *el)
2846 {
2847         int ret = 0, i;
2848         struct buffer_head *eb_bh = NULL;
2849         struct ocfs2_extent_block *eb;
2850         struct ocfs2_extent_rec *rec = NULL;
2851         u64 e_blkno = 0;
2852
2853         if (el->l_tree_depth) {
2854                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
2855                                       &eb_bh);
2856                 if (ret) {
2857                         mlog_errno(ret);
2858                         goto out;
2859                 }
2860
2861                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2862                 el = &eb->h_list;
2863
2864                 if (el->l_tree_depth) {
2865                         ocfs2_error(inode->i_sb,
2866                                     "Inode %lu has non zero tree depth in "
2867                                     "xattr tree block %llu\n", inode->i_ino,
2868                                     (unsigned long long)eb_bh->b_blocknr);
2869                         ret = -EROFS;
2870                         goto out;
2871                 }
2872         }
2873
2874         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2875                 rec = &el->l_recs[i];
2876
2877                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2878                         e_blkno = le64_to_cpu(rec->e_blkno);
2879                         break;
2880                 }
2881         }
2882
2883         if (!e_blkno) {
2884                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2885                             "record (%u, %u, 0) in xattr", inode->i_ino,
2886                             le32_to_cpu(rec->e_cpos),
2887                             ocfs2_rec_clusters(el, rec));
2888                 ret = -EROFS;
2889                 goto out;
2890         }
2891
2892         *p_blkno = le64_to_cpu(rec->e_blkno);
2893         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2894         if (e_cpos)
2895                 *e_cpos = le32_to_cpu(rec->e_cpos);
2896 out:
2897         brelse(eb_bh);
2898         return ret;
2899 }
2900
2901 typedef int (xattr_bucket_func)(struct inode *inode,
2902                                 struct ocfs2_xattr_bucket *bucket,
2903                                 void *para);
2904
2905 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2906                                    struct ocfs2_xattr_bucket *bucket,
2907                                    int name_index,
2908                                    const char *name,
2909                                    u32 name_hash,
2910                                    u16 *xe_index,
2911                                    int *found)
2912 {
2913         int i, ret = 0, cmp = 1, block_off, new_offset;
2914         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2915         size_t name_len = strlen(name);
2916         struct ocfs2_xattr_entry *xe = NULL;
2917         char *xe_name;
2918
2919         /*
2920          * We don't use binary search in the bucket because there
2921          * may be multiple entries with the same name hash.
2922          */
2923         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2924                 xe = &xh->xh_entries[i];
2925
2926                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2927                         continue;
2928                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2929                         break;
2930
2931                 cmp = name_index - ocfs2_xattr_get_type(xe);
2932                 if (!cmp)
2933                         cmp = name_len - xe->xe_name_len;
2934                 if (cmp)
2935                         continue;
2936
2937                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2938                                                         xh,
2939                                                         i,
2940                                                         &block_off,
2941                                                         &new_offset);
2942                 if (ret) {
2943                         mlog_errno(ret);
2944                         break;
2945                 }
2946
2947
2948                 xe_name = bucket_block(bucket, block_off) + new_offset;
2949                 if (!memcmp(name, xe_name, name_len)) {
2950                         *xe_index = i;
2951                         *found = 1;
2952                         ret = 0;
2953                         break;
2954                 }
2955         }
2956
2957         return ret;
2958 }
2959
2960 /*
2961  * Find the specified xattr entry in a series of buckets.
2962  * This series start from p_blkno and last for num_clusters.
2963  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2964  * the num of the valid buckets.
2965  *
2966  * Return the buffer_head this xattr should reside in. And if the xattr's
2967  * hash is in the gap of 2 buckets, return the lower bucket.
2968  */
2969 static int ocfs2_xattr_bucket_find(struct inode *inode,
2970                                    int name_index,
2971                                    const char *name,
2972                                    u32 name_hash,
2973                                    u64 p_blkno,
2974                                    u32 first_hash,
2975                                    u32 num_clusters,
2976                                    struct ocfs2_xattr_search *xs)
2977 {
2978         int ret, found = 0;
2979         struct ocfs2_xattr_header *xh = NULL;
2980         struct ocfs2_xattr_entry *xe = NULL;
2981         u16 index = 0;
2982         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2983         int low_bucket = 0, bucket, high_bucket;
2984         struct ocfs2_xattr_bucket *search;
2985         u32 last_hash;
2986         u64 blkno, lower_blkno = 0;
2987
2988         search = ocfs2_xattr_bucket_new(inode);
2989         if (!search) {
2990                 ret = -ENOMEM;
2991                 mlog_errno(ret);
2992                 goto out;
2993         }
2994
2995         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2996         if (ret) {
2997                 mlog_errno(ret);
2998                 goto out;
2999         }
3000
3001         xh = bucket_xh(search);
3002         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3003         while (low_bucket <= high_bucket) {
3004                 ocfs2_xattr_bucket_relse(search);
3005
3006                 bucket = (low_bucket + high_bucket) / 2;
3007                 blkno = p_blkno + bucket * blk_per_bucket;
3008                 ret = ocfs2_read_xattr_bucket(search, blkno);
3009                 if (ret) {
3010                         mlog_errno(ret);
3011                         goto out;
3012                 }
3013
3014                 xh = bucket_xh(search);
3015                 xe = &xh->xh_entries[0];
3016                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3017                         high_bucket = bucket - 1;
3018                         continue;
3019                 }
3020
3021                 /*
3022                  * Check whether the hash of the last entry in our
3023                  * bucket is larger than the search one. for an empty
3024                  * bucket, the last one is also the first one.
3025                  */
3026                 if (xh->xh_count)
3027                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3028
3029                 last_hash = le32_to_cpu(xe->xe_name_hash);
3030
3031                 /* record lower_blkno which may be the insert place. */
3032                 lower_blkno = blkno;
3033
3034                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3035                         low_bucket = bucket + 1;
3036                         continue;
3037                 }
3038
3039                 /* the searched xattr should reside in this bucket if exists. */
3040                 ret = ocfs2_find_xe_in_bucket(inode, search,
3041                                               name_index, name, name_hash,
3042                                               &index, &found);
3043                 if (ret) {
3044                         mlog_errno(ret);
3045                         goto out;
3046                 }
3047                 break;
3048         }
3049
3050         /*
3051          * Record the bucket we have found.
3052          * When the xattr's hash value is in the gap of 2 buckets, we will
3053          * always set it to the previous bucket.
3054          */
3055         if (!lower_blkno)
3056                 lower_blkno = p_blkno;
3057
3058         /* This should be in cache - we just read it during the search */
3059         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3060         if (ret) {
3061                 mlog_errno(ret);
3062                 goto out;
3063         }
3064
3065         xs->header = bucket_xh(xs->bucket);
3066         xs->base = bucket_block(xs->bucket, 0);
3067         xs->end = xs->base + inode->i_sb->s_blocksize;
3068
3069         if (found) {
3070                 xs->here = &xs->header->xh_entries[index];
3071                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3072                      (unsigned long long)bucket_blkno(xs->bucket), index);
3073         } else
3074                 ret = -ENODATA;
3075
3076 out:
3077         ocfs2_xattr_bucket_free(search);
3078         return ret;
3079 }
3080
3081 static int ocfs2_xattr_index_block_find(struct inode *inode,
3082                                         struct buffer_head *root_bh,
3083                                         int name_index,
3084                                         const char *name,
3085                                         struct ocfs2_xattr_search *xs)
3086 {
3087         int ret;
3088         struct ocfs2_xattr_block *xb =
3089                         (struct ocfs2_xattr_block *)root_bh->b_data;
3090         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3091         struct ocfs2_extent_list *el = &xb_root->xt_list;
3092         u64 p_blkno = 0;
3093         u32 first_hash, num_clusters = 0;
3094         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3095
3096         if (le16_to_cpu(el->l_next_free_rec) == 0)
3097                 return -ENODATA;
3098
3099         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3100              name, name_hash, name_index);
3101
3102         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3103                                   &num_clusters, el);
3104         if (ret) {
3105                 mlog_errno(ret);
3106                 goto out;
3107         }
3108
3109         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3110
3111         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3112              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3113              first_hash);
3114
3115         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3116                                       p_blkno, first_hash, num_clusters, xs);
3117
3118 out:
3119         return ret;
3120 }
3121
3122 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3123                                        u64 blkno,
3124                                        u32 clusters,
3125                                        xattr_bucket_func *func,
3126                                        void *para)
3127 {
3128         int i, ret = 0;
3129         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3130         u32 num_buckets = clusters * bpc;
3131         struct ocfs2_xattr_bucket *bucket;
3132
3133         bucket = ocfs2_xattr_bucket_new(inode);
3134         if (!bucket) {
3135                 mlog_errno(-ENOMEM);
3136                 return -ENOMEM;
3137         }
3138
3139         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3140              clusters, (unsigned long long)blkno);
3141
3142         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3143                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3144                 if (ret) {
3145                         mlog_errno(ret);
3146                         break;
3147                 }
3148
3149                 /*
3150                  * The real bucket num in this series of blocks is stored
3151                  * in the 1st bucket.
3152                  */
3153                 if (i == 0)
3154                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3155
3156                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3157                      (unsigned long long)blkno,
3158                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3159                 if (func) {
3160                         ret = func(inode, bucket, para);
3161                         if (ret && ret != -ERANGE)
3162                                 mlog_errno(ret);
3163                         /* Fall through to bucket_relse() */
3164                 }
3165
3166                 ocfs2_xattr_bucket_relse(bucket);
3167                 if (ret)
3168                         break;
3169         }
3170
3171         ocfs2_xattr_bucket_free(bucket);
3172         return ret;
3173 }
3174
3175 struct ocfs2_xattr_tree_list {
3176         char *buffer;
3177         size_t buffer_size;
3178         size_t result;
3179 };
3180
3181 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3182                                              struct ocfs2_xattr_header *xh,
3183                                              int index,
3184                                              int *block_off,
3185                                              int *new_offset)
3186 {
3187         u16 name_offset;
3188
3189         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3190                 return -EINVAL;
3191
3192         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3193
3194         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3195         *new_offset = name_offset % inode->i_sb->s_blocksize;
3196
3197         return 0;
3198 }
3199
3200 static int ocfs2_list_xattr_bucket(struct inode *inode,
3201                                    struct ocfs2_xattr_bucket *bucket,
3202                                    void *para)
3203 {
3204         int ret = 0, type;
3205         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3206         int i, block_off, new_offset;
3207         const char *prefix, *name;
3208
3209         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3210                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3211                 type = ocfs2_xattr_get_type(entry);
3212                 prefix = ocfs2_xattr_prefix(type);
3213
3214                 if (prefix) {
3215                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3216                                                                 bucket_xh(bucket),
3217                                                                 i,
3218                                                                 &block_off,
3219                                                                 &new_offset);
3220                         if (ret)
3221                                 break;
3222
3223                         name = (const char *)bucket_block(bucket, block_off) +
3224                                 new_offset;
3225                         ret = ocfs2_xattr_list_entry(xl->buffer,
3226                                                      xl->buffer_size,
3227                                                      &xl->result,
3228                                                      prefix, name,
3229                                                      entry->xe_name_len);
3230                         if (ret)
3231                                 break;
3232                 }
3233         }
3234
3235         return ret;
3236 }
3237
3238 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3239                                              struct ocfs2_xattr_tree_root *xt,
3240                                              char *buffer,
3241                                              size_t buffer_size)
3242 {
3243         struct ocfs2_extent_list *el = &xt->xt_list;
3244         int ret = 0;
3245         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3246         u64 p_blkno = 0;
3247         struct ocfs2_xattr_tree_list xl = {
3248                 .buffer = buffer,
3249                 .buffer_size = buffer_size,
3250                 .result = 0,
3251         };
3252
3253         if (le16_to_cpu(el->l_next_free_rec) == 0)
3254                 return 0;
3255
3256         while (name_hash > 0) {
3257                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3258                                           &e_cpos, &num_clusters, el);
3259                 if (ret) {
3260                         mlog_errno(ret);
3261                         goto out;
3262                 }
3263
3264                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3265                                                   ocfs2_list_xattr_bucket,
3266                                                   &xl);
3267                 if (ret) {
3268                         if (ret != -ERANGE)
3269                                 mlog_errno(ret);
3270                         goto out;
3271                 }
3272
3273                 if (e_cpos == 0)
3274                         break;
3275
3276                 name_hash = e_cpos - 1;
3277         }
3278
3279         ret = xl.result;
3280 out:
3281         return ret;
3282 }
3283
3284 static int cmp_xe(const void *a, const void *b)
3285 {
3286         const struct ocfs2_xattr_entry *l = a, *r = b;
3287         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3288         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3289
3290         if (l_hash > r_hash)
3291                 return 1;
3292         if (l_hash < r_hash)
3293                 return -1;
3294         return 0;
3295 }
3296
3297 static void swap_xe(void *a, void *b, int size)
3298 {
3299         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3300
3301         tmp = *l;
3302         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3303         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3304 }
3305
3306 /*
3307  * When the ocfs2_xattr_block is filled up, new bucket will be created
3308  * and all the xattr entries will be moved to the new bucket.
3309  * The header goes at the start of the bucket, and the names+values are
3310  * filled from the end.  This is why *target starts as the last buffer.
3311  * Note: we need to sort the entries since they are not saved in order
3312  * in the ocfs2_xattr_block.
3313  */
3314 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3315                                            struct buffer_head *xb_bh,
3316                                            struct ocfs2_xattr_bucket *bucket)
3317 {
3318         int i, blocksize = inode->i_sb->s_blocksize;
3319         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3320         u16 offset, size, off_change;
3321         struct ocfs2_xattr_entry *xe;
3322         struct ocfs2_xattr_block *xb =
3323                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3324         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3325         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3326         u16 count = le16_to_cpu(xb_xh->xh_count);
3327         char *src = xb_bh->b_data;
3328         char *target = bucket_block(bucket, blks - 1);
3329
3330         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3331              (unsigned long long)xb_bh->b_blocknr,
3332              (unsigned long long)bucket_blkno(bucket));
3333
3334         for (i = 0; i < blks; i++)
3335                 memset(bucket_block(bucket, i), 0, blocksize);
3336
3337         /*
3338          * Since the xe_name_offset is based on ocfs2_xattr_header,
3339          * there is a offset change corresponding to the change of
3340          * ocfs2_xattr_header's position.
3341          */
3342         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3343         xe = &xb_xh->xh_entries[count - 1];
3344         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3345         size = blocksize - offset;
3346
3347         /* copy all the names and values. */
3348         memcpy(target + offset, src + offset, size);
3349
3350         /* Init new header now. */
3351         xh->xh_count = xb_xh->xh_count;
3352         xh->xh_num_buckets = cpu_to_le16(1);
3353         xh->xh_name_value_len = cpu_to_le16(size);
3354         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3355
3356         /* copy all the entries. */
3357         target = bucket_block(bucket, 0);
3358         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3359         size = count * sizeof(struct ocfs2_xattr_entry);
3360         memcpy(target + offset, (char *)xb_xh + offset, size);
3361
3362         /* Change the xe offset for all the xe because of the move. */
3363         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3364                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3365         for (i = 0; i < count; i++)
3366                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3367
3368         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3369              offset, size, off_change);
3370
3371         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3372              cmp_xe, swap_xe);
3373 }
3374
3375 /*
3376  * After we move xattr from block to index btree, we have to
3377  * update ocfs2_xattr_search to the new xe and base.
3378  *
3379  * When the entry is in xattr block, xattr_bh indicates the storage place.
3380  * While if the entry is in index b-tree, "bucket" indicates the
3381  * real place of the xattr.
3382  */
3383 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3384                                             struct ocfs2_xattr_search *xs,
3385                                             struct buffer_head *old_bh)
3386 {
3387         char *buf = old_bh->b_data;
3388         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3389         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3390         int i;
3391
3392         xs->header = bucket_xh(xs->bucket);
3393         xs->base = bucket_block(xs->bucket, 0);
3394         xs->end = xs->base + inode->i_sb->s_blocksize;
3395
3396         if (xs->not_found)
3397                 return;
3398
3399         i = xs->here - old_xh->xh_entries;
3400         xs->here = &xs->header->xh_entries[i];
3401 }
3402
3403 static int ocfs2_xattr_create_index_block(struct inode *inode,
3404                                           struct ocfs2_xattr_search *xs,
3405                                           struct ocfs2_xattr_set_ctxt *ctxt)
3406 {
3407         int ret;
3408         u32 bit_off, len;
3409         u64 blkno;
3410         handle_t *handle = ctxt->handle;
3411         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3412         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3413         struct buffer_head *xb_bh = xs->xattr_bh;
3414         struct ocfs2_xattr_block *xb =
3415                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3416         struct ocfs2_xattr_tree_root *xr;
3417         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3418
3419         mlog(0, "create xattr index block for %llu\n",
3420              (unsigned long long)xb_bh->b_blocknr);
3421
3422         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3423         BUG_ON(!xs->bucket);
3424
3425         /*
3426          * XXX:
3427          * We can use this lock for now, and maybe move to a dedicated mutex
3428          * if performance becomes a problem later.
3429          */
3430         down_write(&oi->ip_alloc_sem);
3431
3432         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3433                                       OCFS2_JOURNAL_ACCESS_WRITE);
3434         if (ret) {
3435                 mlog_errno(ret);
3436                 goto out;
3437         }
3438
3439         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3440                                      1, 1, &bit_off, &len);
3441         if (ret) {
3442                 mlog_errno(ret);
3443                 goto out;
3444         }
3445
3446         /*
3447          * The bucket may spread in many blocks, and
3448          * we will only touch the 1st block and the last block
3449          * in the whole bucket(one for entry and one for data).
3450          */
3451         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3452
3453         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3454              (unsigned long long)blkno);
3455
3456         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3457         if (ret) {
3458                 mlog_errno(ret);
3459                 goto out;
3460         }
3461
3462         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3463                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3464         if (ret) {
3465                 mlog_errno(ret);
3466                 goto out;
3467         }
3468
3469         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3470         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3471
3472         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3473
3474         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3475         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3476                offsetof(struct ocfs2_xattr_block, xb_attrs));
3477
3478         xr = &xb->xb_attrs.xb_root;
3479         xr->xt_clusters = cpu_to_le32(1);
3480         xr->xt_last_eb_blk = 0;
3481         xr->xt_list.l_tree_depth = 0;
3482         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3483         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3484
3485         xr->xt_list.l_recs[0].e_cpos = 0;
3486         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3487         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3488
3489         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3490
3491         ocfs2_journal_dirty(handle, xb_bh);
3492
3493 out:
3494         up_write(&oi->ip_alloc_sem);
3495
3496         return ret;
3497 }
3498
3499 static int cmp_xe_offset(const void *a, const void *b)
3500 {
3501         const struct ocfs2_xattr_entry *l = a, *r = b;
3502         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3503         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3504
3505         if (l_name_offset < r_name_offset)
3506                 return 1;
3507         if (l_name_offset > r_name_offset)
3508                 return -1;
3509         return 0;
3510 }
3511
3512 /*
3513  * defrag a xattr bucket if we find that the bucket has some
3514  * holes beteen name/value pairs.
3515  * We will move all the name/value pairs to the end of the bucket
3516  * so that we can spare some space for insertion.
3517  */
3518 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3519                                      handle_t *handle,
3520                                      struct ocfs2_xattr_bucket *bucket)
3521 {
3522         int ret, i;
3523         size_t end, offset, len, value_len;
3524         struct ocfs2_xattr_header *xh;
3525         char *entries, *buf, *bucket_buf = NULL;
3526         u64 blkno = bucket_blkno(bucket);
3527         u16 xh_free_start;
3528         size_t blocksize = inode->i_sb->s_blocksize;
3529         struct ocfs2_xattr_entry *xe;
3530
3531         /*
3532          * In order to make the operation more efficient and generic,
3533          * we copy all the blocks into a contiguous memory and do the
3534          * defragment there, so if anything is error, we will not touch
3535          * the real block.
3536          */
3537         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3538         if (!bucket_buf) {
3539                 ret = -EIO;
3540                 goto out;
3541         }
3542
3543         buf = bucket_buf;
3544         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3545                 memcpy(buf, bucket_block(bucket, i), blocksize);
3546
3547         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3548                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3549         if (ret < 0) {
3550                 mlog_errno(ret);
3551                 goto out;
3552         }
3553
3554         xh = (struct ocfs2_xattr_header *)bucket_buf;
3555         entries = (char *)xh->xh_entries;
3556         xh_free_start = le16_to_cpu(xh->xh_free_start);
3557
3558         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3559              "xh_free_start = %u, xh_name_value_len = %u.\n",
3560              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3561              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3562
3563         /*
3564          * sort all the entries by their offset.
3565          * the largest will be the first, so that we can
3566          * move them to the end one by one.
3567          */
3568         sort(entries, le16_to_cpu(xh->xh_count),
3569              sizeof(struct ocfs2_xattr_entry),
3570              cmp_xe_offset, swap_xe);
3571
3572         /* Move all name/values to the end of the bucket. */
3573         xe = xh->xh_entries;
3574         end = OCFS2_XATTR_BUCKET_SIZE;
3575         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3576                 offset = le16_to_cpu(xe->xe_name_offset);
3577                 if (ocfs2_xattr_is_local(xe))
3578                         value_len = OCFS2_XATTR_SIZE(
3579                                         le64_to_cpu(xe->xe_value_size));
3580                 else
3581                         value_len = OCFS2_XATTR_ROOT_SIZE;
3582                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3583
3584                 /*
3585                  * We must make sure that the name/value pair
3586                  * exist in the same block. So adjust end to
3587                  * the previous block end if needed.
3588                  */
3589                 if (((end - len) / blocksize !=
3590                         (end - 1) / blocksize))
3591                         end = end - end % blocksize;
3592
3593                 if (end > offset + len) {
3594                         memmove(bucket_buf + end - len,
3595                                 bucket_buf + offset, len);
3596                         xe->xe_name_offset = cpu_to_le16(end - len);
3597                 }
3598
3599                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3600                                 "bucket %llu\n", (unsigned long long)blkno);
3601
3602                 end -= len;
3603         }
3604
3605         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3606                         "bucket %llu\n", (unsigned long long)blkno);
3607
3608         if (xh_free_start == end)
3609                 goto out;
3610
3611         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3612         xh->xh_free_start = cpu_to_le16(end);
3613
3614         /* sort the entries by their name_hash. */
3615         sort(entries, le16_to_cpu(xh->xh_count),
3616              sizeof(struct ocfs2_xattr_entry),
3617              cmp_xe, swap_xe);
3618
3619         buf = bucket_buf;
3620         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3621                 memcpy(bucket_block(bucket, i), buf, blocksize);
3622         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3623
3624 out:
3625         kfree(bucket_buf);
3626         return ret;
3627 }
3628
3629 /*
3630  * prev_blkno points to the start of an existing extent.  new_blkno
3631  * points to a newly allocated extent.  Because we know each of our
3632  * clusters contains more than bucket, we can easily split one cluster
3633  * at a bucket boundary.  So we take the last cluster of the existing
3634  * extent and split it down the middle.  We move the last half of the
3635  * buckets in the last cluster of the existing extent over to the new
3636  * extent.
3637  *
3638  * first_bh is the buffer at prev_blkno so we can update the existing
3639  * extent's bucket count.  header_bh is the bucket were we were hoping
3640  * to insert our xattr.  If the bucket move places the target in the new
3641  * extent, we'll update first_bh and header_bh after modifying the old
3642  * extent.
3643  *
3644  * first_hash will be set as the 1st xe's name_hash in the new extent.
3645  */
3646 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3647                                                handle_t *handle,
3648                                                struct ocfs2_xattr_bucket *first,
3649                                                struct ocfs2_xattr_bucket *target,
3650                                                u64 new_blkno,
3651                                                u32 num_clusters,
3652                                                u32 *first_hash)
3653 {
3654         int ret;
3655         struct super_block *sb = inode->i_sb;
3656         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3657         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3658         int to_move = num_buckets / 2;
3659         u64 src_blkno;
3660         u64 last_cluster_blkno = bucket_blkno(first) +
3661                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3662
3663         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3664         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3665
3666         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3667              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3668
3669         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3670                                      last_cluster_blkno, new_blkno,
3671                                      to_move, first_hash);
3672         if (ret) {
3673                 mlog_errno(ret);
3674                 goto out;
3675         }
3676
3677         /* This is the first bucket that got moved */
3678         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3679
3680         /*
3681          * If the target bucket was part of the moved buckets, we need to
3682          * update first and target.
3683          */
3684         if (bucket_blkno(target) >= src_blkno) {
3685                 /* Find the block for the new target bucket */
3686                 src_blkno = new_blkno +
3687                         (bucket_blkno(target) - src_blkno);
3688
3689                 ocfs2_xattr_bucket_relse(first);
3690                 ocfs2_xattr_bucket_relse(target);
3691
3692                 /*
3693                  * These shouldn't fail - the buffers are in the
3694                  * journal from ocfs2_cp_xattr_bucket().
3695                  */
3696                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3697                 if (ret) {
3698                         mlog_errno(ret);
3699                         goto out;
3700                 }
3701                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3702                 if (ret)
3703                         mlog_errno(ret);
3704
3705         }
3706
3707 out:
3708         return ret;
3709 }
3710
3711 /*
3712  * Find the suitable pos when we divide a bucket into 2.
3713  * We have to make sure the xattrs with the same hash value exist
3714  * in the same bucket.
3715  *
3716  * If this ocfs2_xattr_header covers more than one hash value, find a
3717  * place where the hash value changes.  Try to find the most even split.
3718  * The most common case is that all entries have different hash values,
3719  * and the first check we make will find a place to split.
3720  */
3721 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3722 {
3723         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3724         int count = le16_to_cpu(xh->xh_count);
3725         int delta, middle = count / 2;
3726
3727         /*
3728          * We start at the middle.  Each step gets farther away in both
3729          * directions.  We therefore hit the change in hash value
3730          * nearest to the middle.  Note that this loop does not execute for
3731          * count < 2.
3732          */
3733         for (delta = 0; delta < middle; delta++) {
3734                 /* Let's check delta earlier than middle */
3735                 if (cmp_xe(&entries[middle - delta - 1],
3736                            &entries[middle - delta]))
3737                         return middle - delta;
3738
3739                 /* For even counts, don't walk off the end */
3740                 if ((middle + delta + 1) == count)
3741                         continue;
3742
3743                 /* Now try delta past middle */
3744                 if (cmp_xe(&entries[middle + delta],
3745                            &entries[middle + delta + 1]))
3746                         return middle + delta + 1;
3747         }
3748
3749         /* Every entry had the same hash */
3750         return count;
3751 }
3752
3753 /*
3754  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3755  * first_hash will record the 1st hash of the new bucket.
3756  *
3757  * Normally half of the xattrs will be moved.  But we have to make
3758  * sure that the xattrs with the same hash value are stored in the
3759  * same bucket. If all the xattrs in this bucket have the same hash
3760  * value, the new bucket will be initialized as an empty one and the
3761  * first_hash will be initialized as (hash_value+1).
3762  */
3763 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3764                                     handle_t *handle,
3765                                     u64 blk,
3766                                     u64 new_blk,
3767                                     u32 *first_hash,
3768                                     int new_bucket_head)
3769 {
3770         int ret, i;
3771         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3772         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3773         struct ocfs2_xattr_header *xh;
3774         struct ocfs2_xattr_entry *xe;
3775         int blocksize = inode->i_sb->s_blocksize;
3776
3777         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3778              (unsigned long long)blk, (unsigned long long)new_blk);
3779
3780         s_bucket = ocfs2_xattr_bucket_new(inode);
3781         t_bucket = ocfs2_xattr_bucket_new(inode);
3782         if (!s_bucket || !t_bucket) {
3783                 ret = -ENOMEM;
3784                 mlog_errno(ret);
3785                 goto out;
3786         }
3787
3788         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3789         if (ret) {
3790                 mlog_errno(ret);
3791                 goto out;
3792         }
3793
3794         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3795                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3796         if (ret) {
3797                 mlog_errno(ret);
3798                 goto out;
3799         }
3800
3801         /*
3802          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3803          * there's no need to read it.
3804          */
3805         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3806         if (ret) {
3807                 mlog_errno(ret);
3808                 goto out;
3809         }
3810
3811         /*
3812          * Hey, if we're overwriting t_bucket, what difference does
3813          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3814          * same part of ocfs2_cp_xattr_bucket().
3815          */
3816         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3817                                                 new_bucket_head ?
3818                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3819                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3820         if (ret) {
3821                 mlog_errno(ret);
3822                 goto out;
3823         }
3824
3825         xh = bucket_xh(s_bucket);
3826         count = le16_to_cpu(xh->xh_count);
3827         start = ocfs2_xattr_find_divide_pos(xh);
3828
3829         if (start == count) {
3830                 xe = &xh->xh_entries[start-1];
3831
3832                 /*
3833                  * initialized a new empty bucket here.
3834                  * The hash value is set as one larger than
3835                  * that of the last entry in the previous bucket.
3836                  */
3837                 for (i = 0; i < t_bucket->bu_blocks; i++)
3838                         memset(bucket_block(t_bucket, i), 0, blocksize);
3839
3840                 xh = bucket_xh(t_bucket);
3841                 xh->xh_free_start = cpu_to_le16(blocksize);
3842                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3843                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3844
3845                 goto set_num_buckets;
3846         }
3847
3848         /* copy the whole bucket to the new first. */
3849         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3850
3851         /* update the new bucket. */
3852         xh = bucket_xh(t_bucket);
3853
3854         /*
3855          * Calculate the total name/value len and xh_free_start for
3856          * the old bucket first.
3857          */
3858         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3859         name_value_len = 0;
3860         for (i = 0; i < start; i++) {
3861                 xe = &xh->xh_entries[i];
3862                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3863                 if (ocfs2_xattr_is_local(xe))
3864                         xe_len +=
3865                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3866                 else
3867                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3868                 name_value_len += xe_len;
3869                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3870                         name_offset = le16_to_cpu(xe->xe_name_offset);
3871         }
3872
3873         /*
3874          * Now begin the modification to the new bucket.
3875          *
3876          * In the new bucket, We just move the xattr entry to the beginning
3877          * and don't touch the name/value. So there will be some holes in the
3878          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3879          * called.
3880          */
3881         xe = &xh->xh_entries[start];
3882         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3883         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3884              (int)((char *)xe - (char *)xh),
3885              (int)((char *)xh->xh_entries - (char *)xh));
3886         memmove((char *)xh->xh_entries, (char *)xe, len);
3887         xe = &xh->xh_entries[count - start];
3888         len = sizeof(struct ocfs2_xattr_entry) * start;
3889         memset((char *)xe, 0, len);
3890
3891         le16_add_cpu(&xh->xh_count, -start);
3892         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3893
3894         /* Calculate xh_free_start for the new bucket. */
3895         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3896         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3897                 xe = &xh->xh_entries[i];
3898                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3899                 if (ocfs2_xattr_is_local(xe))
3900                         xe_len +=
3901                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3902                 else
3903                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3904                 if (le16_to_cpu(xe->xe_name_offset) <
3905                     le16_to_cpu(xh->xh_free_start))
3906                         xh->xh_free_start = xe->xe_name_offset;
3907         }
3908
3909 set_num_buckets:
3910         /* set xh->xh_num_buckets for the new xh. */
3911         if (new_bucket_head)
3912                 xh->xh_num_buckets = cpu_to_le16(1);
3913         else
3914                 xh->xh_num_buckets = 0;
3915
3916         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3917
3918         /* store the first_hash of the new bucket. */
3919         if (first_hash)
3920                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3921
3922         /*
3923          * Now only update the 1st block of the old bucket.  If we
3924          * just added a new empty bucket, there is no need to modify
3925          * it.
3926          */
3927         if (start == count)
3928                 goto out;
3929
3930         xh = bucket_xh(s_bucket);
3931         memset(&xh->xh_entries[start], 0,
3932                sizeof(struct ocfs2_xattr_entry) * (count - start));
3933         xh->xh_count = cpu_to_le16(start);
3934         xh->xh_free_start = cpu_to_le16(name_offset);
3935         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3936
3937         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3938
3939 out:
3940         ocfs2_xattr_bucket_free(s_bucket);
3941         ocfs2_xattr_bucket_free(t_bucket);
3942
3943         return ret;
3944 }
3945
3946 /*
3947  * Copy xattr from one bucket to another bucket.
3948  *
3949  * The caller must make sure that the journal transaction
3950  * has enough space for journaling.
3951  */
3952 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3953                                  handle_t *handle,
3954                                  u64 s_blkno,
3955                                  u64 t_blkno,
3956                                  int t_is_new)
3957 {
3958         int ret;
3959         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3960
3961         BUG_ON(s_blkno == t_blkno);
3962
3963         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3964              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3965              t_is_new);
3966
3967         s_bucket = ocfs2_xattr_bucket_new(inode);
3968         t_bucket = ocfs2_xattr_bucket_new(inode);
3969         if (!s_bucket || !t_bucket) {
3970                 ret = -ENOMEM;
3971                 mlog_errno(ret);
3972                 goto out;
3973         }
3974
3975         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3976         if (ret)
3977                 goto out;
3978
3979         /*
3980          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3981          * there's no need to read it.
3982          */
3983         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3984         if (ret)
3985                 goto out;
3986
3987         /*
3988          * Hey, if we're overwriting t_bucket, what difference does
3989          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3990          * cluster to fill, we came here from
3991          * ocfs2_mv_xattr_buckets(), and it is really new -
3992          * ACCESS_CREATE is required.  But we also might have moved data
3993          * out of t_bucket before extending back into it.
3994          * ocfs2_add_new_xattr_bucket() can do this - its call to
3995          * ocfs2_add_new_xattr_cluster() may have created a new extent
3996          * and copied out the end of the old extent.  Then it re-extends
3997          * the old extent back to create space for new xattrs.  That's
3998          * how we get here, and the bucket isn't really new.
3999          */
4000         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4001                                                 t_is_new ?
4002                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4003                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4004         if (ret)
4005                 goto out;
4006
4007         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4008         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4009
4010 out:
4011         ocfs2_xattr_bucket_free(t_bucket);
4012         ocfs2_xattr_bucket_free(s_bucket);
4013
4014         return ret;
4015 }
4016
4017 /*
4018  * src_blk points to the start of an existing extent.  last_blk points to
4019  * last cluster in that extent.  to_blk points to a newly allocated
4020  * extent.  We copy the buckets from the cluster at last_blk to the new
4021  * extent.  If start_bucket is non-zero, we skip that many buckets before
4022  * we start copying.  The new extent's xh_num_buckets gets set to the
4023  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4024  * by the same amount.
4025  */
4026 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4027                                   u64 src_blk, u64 last_blk, u64 to_blk,
4028                                   unsigned int start_bucket,
4029                                   u32 *first_hash)
4030 {
4031         int i, ret, credits;
4032         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4033         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4034         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4035         struct ocfs2_xattr_bucket *old_first, *new_first;
4036
4037         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4038              (unsigned long long)last_blk, (unsigned long long)to_blk);
4039
4040         BUG_ON(start_bucket >= num_buckets);
4041         if (start_bucket) {
4042                 num_buckets -= start_bucket;
4043                 last_blk += (start_bucket * blks_per_bucket);
4044         }
4045
4046         /* The first bucket of the original extent */
4047         old_first = ocfs2_xattr_bucket_new(inode);
4048         /* The first bucket of the new extent */
4049         new_first = ocfs2_xattr_bucket_new(inode);
4050         if (!old_first || !new_first) {
4051                 ret = -ENOMEM;
4052                 mlog_errno(ret);
4053                 goto out;
4054         }
4055
4056         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4057         if (ret) {
4058                 mlog_errno(ret);
4059                 goto out;
4060         }
4061
4062         /*
4063          * We need to update the first bucket of the old extent and all
4064          * the buckets going to the new extent.
4065          */
4066         credits = ((num_buckets + 1) * blks_per_bucket) +
4067                 handle->h_buffer_credits;
4068         ret = ocfs2_extend_trans(handle, credits);
4069         if (ret) {
4070                 mlog_errno(ret);
4071                 goto out;
4072         }
4073
4074         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4075                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4076         if (ret) {
4077                 mlog_errno(ret);
4078                 goto out;
4079         }
4080
4081         for (i = 0; i < num_buckets; i++) {
4082                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4083                                             last_blk + (i * blks_per_bucket),
4084                                             to_blk + (i * blks_per_bucket),
4085                                             1);
4086                 if (ret) {
4087                         mlog_errno(ret);
4088                         goto out;
4089                 }
4090         }
4091
4092         /*
4093          * Get the new bucket ready before we dirty anything
4094          * (This actually shouldn't fail, because we already dirtied
4095          * it once in ocfs2_cp_xattr_bucket()).
4096          */
4097         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4098         if (ret) {
4099                 mlog_errno(ret);
4100                 goto out;
4101         }
4102         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4103                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4104         if (ret) {
4105                 mlog_errno(ret);
4106                 goto out;
4107         }
4108
4109         /* Now update the headers */
4110         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4111         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4112
4113         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4114         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4115
4116         if (first_hash)
4117                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4118
4119 out:
4120         ocfs2_xattr_bucket_free(new_first);
4121         ocfs2_xattr_bucket_free(old_first);
4122         return ret;
4123 }
4124
4125 /*
4126  * Move some xattrs in this cluster to the new cluster.
4127  * This function should only be called when bucket size == cluster size.
4128  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4129  */
4130 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4131                                       handle_t *handle,
4132                                       u64 prev_blk,
4133                                       u64 new_blk,
4134                                       u32 *first_hash)
4135 {
4136         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4137         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4138
4139         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4140
4141         ret = ocfs2_extend_trans(handle, credits);
4142         if (ret) {
4143                 mlog_errno(ret);
4144                 return ret;
4145         }
4146
4147         /* Move half of the xattr in start_blk to the next bucket. */
4148         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4149                                           new_blk, first_hash, 1);
4150 }
4151
4152 /*
4153  * Move some xattrs from the old cluster to the new one since they are not
4154  * contiguous in ocfs2 xattr tree.
4155  *
4156  * new_blk starts a new separate cluster, and we will move some xattrs from
4157  * prev_blk to it. v_start will be set as the first name hash value in this
4158  * new cluster so that it can be used as e_cpos during tree insertion and
4159  * don't collide with our original b-tree operations. first_bh and header_bh
4160  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4161  * to extend the insert bucket.
4162  *
4163  * The problem is how much xattr should we move to the new one and when should
4164  * we update first_bh and header_bh?
4165  * 1. If cluster size > bucket size, that means the previous cluster has more
4166  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4167  *    update the first_bh and header_bh if the insert bucket has been moved
4168  *    to the new cluster.
4169  * 2. If cluster_size == bucket_size:
4170  *    a) If the previous extent rec has more than one cluster and the insert
4171  *       place isn't in the last cluster, copy the entire last cluster to the
4172  *       new one. This time, we don't need to upate the first_bh and header_bh
4173  *       since they will not be moved into the new cluster.
4174  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4175  *       the new one. And we set the extend flag to zero if the insert place is
4176  *       moved into the new allocated cluster since no extend is needed.
4177  */
4178 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4179                                             handle_t *handle,
4180                                             struct ocfs2_xattr_bucket *first,
4181                                             struct ocfs2_xattr_bucket *target,
4182                                             u64 new_blk,
4183                                             u32 prev_clusters,
4184                                             u32 *v_start,
4185                                             int *extend)
4186 {
4187         int ret;
4188
4189         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4190              (unsigned long long)bucket_blkno(first), prev_clusters,
4191              (unsigned long long)new_blk);
4192
4193         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4194                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4195                                                           handle,
4196                                                           first, target,
4197                                                           new_blk,
4198                                                           prev_clusters,
4199                                                           v_start);
4200                 if (ret)
4201                         mlog_errno(ret);
4202         } else {
4203                 /* The start of the last cluster in the first extent */
4204                 u64 last_blk = bucket_blkno(first) +
4205                         ((prev_clusters - 1) *
4206                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4207
4208                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4209                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4210                                                      bucket_blkno(first),
4211                                                      last_blk, new_blk, 0,
4212                                                      v_start);
4213                         if (ret)
4214                                 mlog_errno(ret);
4215                 } else {
4216                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4217                                                          last_blk, new_blk,
4218                                                          v_start);
4219                         if (ret)
4220                                 mlog_errno(ret);
4221
4222                         if ((bucket_blkno(target) == last_blk) && extend)
4223                                 *extend = 0;
4224                 }
4225         }
4226
4227         return ret;
4228 }
4229
4230 /*
4231  * Add a new cluster for xattr storage.
4232  *
4233  * If the new cluster is contiguous with the previous one, it will be
4234  * appended to the same extent record, and num_clusters will be updated.
4235  * If not, we will insert a new extent for it and move some xattrs in
4236  * the last cluster into the new allocated one.
4237  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4238  * lose the benefits of hashing because we'll have to search large leaves.
4239  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4240  * if it's bigger).
4241  *
4242  * first_bh is the first block of the previous extent rec and header_bh
4243  * indicates the bucket we will insert the new xattrs. They will be updated
4244  * when the header_bh is moved into the new cluster.
4245  */
4246 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4247                                        struct buffer_head *root_bh,
4248                                        struct ocfs2_xattr_bucket *first,
4249                                        struct ocfs2_xattr_bucket *target,
4250                                        u32 *num_clusters,
4251                                        u32 prev_cpos,
4252                                        int *extend,
4253                                        struct ocfs2_xattr_set_ctxt *ctxt)
4254 {
4255         int ret;
4256         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4257         u32 prev_clusters = *num_clusters;
4258         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4259         u64 block;
4260         handle_t *handle = ctxt->handle;
4261         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4262         struct ocfs2_extent_tree et;
4263
4264         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4265              "previous xattr blkno = %llu\n",
4266              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4267              prev_cpos, (unsigned long long)bucket_blkno(first));
4268
4269         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4270
4271         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4272                                       OCFS2_JOURNAL_ACCESS_WRITE);
4273         if (ret < 0) {
4274                 mlog_errno(ret);
4275                 goto leave;
4276         }
4277
4278         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4279                                      clusters_to_add, &bit_off, &num_bits);
4280         if (ret < 0) {
4281                 if (ret != -ENOSPC)
4282                         mlog_errno(ret);
4283                 goto leave;
4284         }
4285
4286         BUG_ON(num_bits > clusters_to_add);
4287
4288         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4289         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4290              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4291
4292         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4293             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4294              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4295                 /*
4296                  * If this cluster is contiguous with the old one and
4297                  * adding this new cluster, we don't surpass the limit of
4298                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4299                  * initialized and used like other buckets in the previous
4300                  * cluster.
4301                  * So add it as a contiguous one. The caller will handle
4302                  * its init process.
4303                  */
4304                 v_start = prev_cpos + prev_clusters;
4305                 *num_clusters = prev_clusters + num_bits;
4306                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4307                      num_bits);
4308         } else {
4309                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4310                                                        handle,
4311                                                        first,
4312                                                        target,
4313                                                        block,
4314                                                        prev_clusters,
4315                                                        &v_start,
4316                                                        extend);
4317                 if (ret) {
4318                         mlog_errno(ret);
4319                         goto leave;
4320                 }
4321         }
4322
4323         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4324              num_bits, (unsigned long long)block, v_start);
4325         ret = ocfs2_insert_extent(handle, &et, v_start, block,
4326                                   num_bits, 0, ctxt->meta_ac);
4327         if (ret < 0) {
4328                 mlog_errno(ret);
4329                 goto leave;
4330         }
4331
4332         ret = ocfs2_journal_dirty(handle, root_bh);
4333         if (ret < 0)
4334                 mlog_errno(ret);
4335
4336 leave:
4337         return ret;
4338 }
4339
4340 /*
4341  * We are given an extent.  'first' is the bucket at the very front of
4342  * the extent.  The extent has space for an additional bucket past
4343  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4344  * of the target bucket.  We wish to shift every bucket past the target
4345  * down one, filling in that additional space.  When we get back to the
4346  * target, we split the target between itself and the now-empty bucket
4347  * at target+1 (aka, target_blkno + blks_per_bucket).
4348  */
4349 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4350                                      handle_t *handle,
4351                                      struct ocfs2_xattr_bucket *first,
4352                                      u64 target_blk,
4353                                      u32 num_clusters)
4354 {
4355         int ret, credits;
4356         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4357         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4358         u64 end_blk;
4359         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4360
4361         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4362              "from %llu, len = %u\n", (unsigned long long)target_blk,
4363              (unsigned long long)bucket_blkno(first), num_clusters);
4364
4365         /* The extent must have room for an additional bucket */
4366         BUG_ON(new_bucket >=
4367                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4368
4369         /* end_blk points to the last existing bucket */
4370         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4371
4372         /*
4373          * end_blk is the start of the last existing bucket.
4374          * Thus, (end_blk - target_blk) covers the target bucket and
4375          * every bucket after it up to, but not including, the last
4376          * existing bucket.  Then we add the last existing bucket, the
4377          * new bucket, and the first bucket (3 * blk_per_bucket).
4378          */
4379         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4380                   handle->h_buffer_credits;
4381         ret = ocfs2_extend_trans(handle, credits);
4382         if (ret) {
4383                 mlog_errno(ret);
4384                 goto out;
4385         }
4386
4387         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4388                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4389         if (ret) {
4390                 mlog_errno(ret);
4391                 goto out;
4392         }
4393
4394         while (end_blk != target_blk) {
4395                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4396                                             end_blk + blk_per_bucket, 0);
4397                 if (ret)
4398                         goto out;
4399                 end_blk -= blk_per_bucket;
4400         }
4401
4402         /* Move half of the xattr in target_blkno to the next bucket. */
4403         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4404                                         target_blk + blk_per_bucket, NULL, 0);
4405
4406         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4407         ocfs2_xattr_bucket_journal_dirty(handle, first);
4408
4409 out:
4410         return ret;
4411 }
4412
4413 /*
4414  * Add new xattr bucket in an extent record and adjust the buckets
4415  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4416  * bucket we want to insert into.
4417  *
4418  * In the easy case, we will move all the buckets after target down by
4419  * one. Half of target's xattrs will be moved to the next bucket.
4420  *
4421  * If current cluster is full, we'll allocate a new one.  This may not
4422  * be contiguous.  The underlying calls will make sure that there is
4423  * space for the insert, shifting buckets around if necessary.
4424  * 'target' may be moved by those calls.
4425  */
4426 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4427                                       struct buffer_head *xb_bh,
4428                                       struct ocfs2_xattr_bucket *target,
4429                                       struct ocfs2_xattr_set_ctxt *ctxt)
4430 {
4431         struct ocfs2_xattr_block *xb =
4432                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4433         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4434         struct ocfs2_extent_list *el = &xb_root->xt_list;
4435         u32 name_hash =
4436                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4437         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4438         int ret, num_buckets, extend = 1;
4439         u64 p_blkno;
4440         u32 e_cpos, num_clusters;
4441         /* The bucket at the front of the extent */
4442         struct ocfs2_xattr_bucket *first;
4443
4444         mlog(0, "Add new xattr bucket starting from %llu\n",
4445              (unsigned long long)bucket_blkno(target));
4446
4447         /* The first bucket of the original extent */
4448         first = ocfs2_xattr_bucket_new(inode);
4449         if (!first) {
4450                 ret = -ENOMEM;
4451                 mlog_errno(ret);
4452                 goto out;
4453         }
4454
4455         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4456                                   &num_clusters, el);
4457         if (ret) {
4458                 mlog_errno(ret);
4459                 goto out;
4460         }
4461
4462         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4463         if (ret) {
4464                 mlog_errno(ret);
4465                 goto out;
4466         }
4467
4468         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4469         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4470                 /*
4471                  * This can move first+target if the target bucket moves
4472                  * to the new extent.
4473                  */
4474                 ret = ocfs2_add_new_xattr_cluster(inode,
4475                                                   xb_bh,
4476                                                   first,
4477                                                   target,
4478                                                   &num_clusters,
4479                                                   e_cpos,
4480                                                   &extend,
4481                                                   ctxt);
4482                 if (ret) {
4483                         mlog_errno(ret);
4484                         goto out;
4485                 }
4486         }
4487
4488         if (extend) {
4489                 ret = ocfs2_extend_xattr_bucket(inode,
4490                                                 ctxt->handle,
4491                                                 first,
4492                                                 bucket_blkno(target),
4493                                                 num_clusters);
4494                 if (ret)
4495                         mlog_errno(ret);
4496         }
4497
4498 out:
4499         ocfs2_xattr_bucket_free(first);
4500
4501         return ret;
4502 }
4503
4504 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4505                                         struct ocfs2_xattr_bucket *bucket,
4506                                         int offs)
4507 {
4508         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4509
4510         offs = offs % inode->i_sb->s_blocksize;
4511         return bucket_block(bucket, block_off) + offs;
4512 }
4513
4514 /*
4515  * Handle the normal xattr set, including replace, delete and new.
4516  *
4517  * Note: "local" indicates the real data's locality. So we can't
4518  * just its bucket locality by its length.
4519  */
4520 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4521                                          struct ocfs2_xattr_info *xi,
4522                                          struct ocfs2_xattr_search *xs,
4523                                          u32 name_hash,
4524                                          int local)
4525 {
4526         struct ocfs2_xattr_entry *last, *xe;
4527         int name_len = strlen(xi->name);
4528         struct ocfs2_xattr_header *xh = xs->header;
4529         u16 count = le16_to_cpu(xh->xh_count), start;
4530         size_t blocksize = inode->i_sb->s_blocksize;
4531         char *val;
4532         size_t offs, size, new_size;
4533
4534         last = &xh->xh_entries[count];
4535         if (!xs->not_found) {
4536                 xe = xs->here;
4537                 offs = le16_to_cpu(xe->xe_name_offset);
4538                 if (ocfs2_xattr_is_local(xe))
4539                         size = OCFS2_XATTR_SIZE(name_len) +
4540                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4541                 else
4542                         size = OCFS2_XATTR_SIZE(name_len) +
4543                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4544
4545                 /*
4546                  * If the new value will be stored outside, xi->value has been
4547                  * initalized as an empty ocfs2_xattr_value_root, and the same
4548                  * goes with xi->value_len, so we can set new_size safely here.
4549                  * See ocfs2_xattr_set_in_bucket.
4550                  */
4551                 new_size = OCFS2_XATTR_SIZE(name_len) +
4552                            OCFS2_XATTR_SIZE(xi->value_len);
4553
4554                 le16_add_cpu(&xh->xh_name_value_len, -size);
4555                 if (xi->value) {
4556                         if (new_size > size)
4557                                 goto set_new_name_value;
4558
4559                         /* Now replace the old value with new one. */
4560                         if (local)
4561                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4562                         else
4563                                 xe->xe_value_size = 0;
4564
4565                         val = ocfs2_xattr_bucket_get_val(inode,
4566                                                          xs->bucket, offs);
4567                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4568                                size - OCFS2_XATTR_SIZE(name_len));
4569                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4570                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4571                                        xi->value, xi->value_len);
4572
4573                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4574                         ocfs2_xattr_set_local(xe, local);
4575                         return;
4576                 } else {
4577                         /*
4578                          * Remove the old entry if there is more than one.
4579                          * We don't remove the last entry so that we can
4580                          * use it to indicate the hash value of the empty
4581                          * bucket.
4582                          */
4583                         last -= 1;
4584                         le16_add_cpu(&xh->xh_count, -1);
4585                         if (xh->xh_count) {
4586                                 memmove(xe, xe + 1,
4587                                         (void *)last - (void *)xe);
4588                                 memset(last, 0,
4589                                        sizeof(struct ocfs2_xattr_entry));
4590                         } else
4591                                 xh->xh_free_start =
4592                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4593
4594                         return;
4595                 }
4596         } else {
4597                 /* find a new entry for insert. */
4598                 int low = 0, high = count - 1, tmp;
4599                 struct ocfs2_xattr_entry *tmp_xe;
4600
4601                 while (low <= high && count) {
4602                         tmp = (low + high) / 2;
4603                         tmp_xe = &xh->xh_entries[tmp];
4604
4605                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4606                                 low = tmp + 1;
4607                         else if (name_hash <
4608                                  le32_to_cpu(tmp_xe->xe_name_hash))
4609                                 high = tmp - 1;
4610                         else {
4611                                 low = tmp;
4612                                 break;
4613                         }
4614                 }
4615
4616                 xe = &xh->xh_entries[low];
4617                 if (low != count)
4618                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4619
4620                 le16_add_cpu(&xh->xh_count, 1);
4621                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4622                 xe->xe_name_hash = cpu_to_le32(name_hash);
4623                 xe->xe_name_len = name_len;
4624                 ocfs2_xattr_set_type(xe, xi->name_index);
4625         }
4626
4627 set_new_name_value:
4628         /* Insert the new name+value. */
4629         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4630
4631         /*
4632          * We must make sure that the name/value pair
4633          * exists in the same block.
4634          */
4635         offs = le16_to_cpu(xh->xh_free_start);
4636         start = offs - size;
4637
4638         if (start >> inode->i_sb->s_blocksize_bits !=
4639             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4640                 offs = offs - offs % blocksize;
4641                 xh->xh_free_start = cpu_to_le16(offs);
4642         }
4643
4644         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4645         xe->xe_name_offset = cpu_to_le16(offs - size);
4646
4647         memset(val, 0, size);
4648         memcpy(val, xi->name, name_len);
4649         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4650
4651         xe->xe_value_size = cpu_to_le64(xi->value_len);
4652         ocfs2_xattr_set_local(xe, local);
4653         xs->here = xe;
4654         le16_add_cpu(&xh->xh_free_start, -size);
4655         le16_add_cpu(&xh->xh_name_value_len, size);
4656
4657         return;
4658 }
4659
4660 /*
4661  * Set the xattr entry in the specified bucket.
4662  * The bucket is indicated by xs->bucket and it should have the enough
4663  * space for the xattr insertion.
4664  */
4665 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4666                                            handle_t *handle,
4667                                            struct ocfs2_xattr_info *xi,
4668                                            struct ocfs2_xattr_search *xs,
4669                                            u32 name_hash,
4670                                            int local)
4671 {
4672         int ret;
4673         u64 blkno;
4674
4675         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4676              (unsigned long)xi->value_len, xi->name_index,
4677              (unsigned long long)bucket_blkno(xs->bucket));
4678
4679         if (!xs->bucket->bu_bhs[1]) {
4680                 blkno = bucket_blkno(xs->bucket);
4681                 ocfs2_xattr_bucket_relse(xs->bucket);
4682                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4683                 if (ret) {
4684                         mlog_errno(ret);
4685                         goto out;
4686                 }
4687         }
4688
4689         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4690                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4691         if (ret < 0) {
4692                 mlog_errno(ret);
4693                 goto out;
4694         }
4695
4696         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4697         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4698
4699 out:
4700         return ret;
4701 }
4702
4703 /*
4704  * Truncate the specified xe_off entry in xattr bucket.
4705  * bucket is indicated by header_bh and len is the new length.
4706  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4707  *
4708  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4709  */
4710 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4711                                              struct ocfs2_xattr_bucket *bucket,
4712                                              int xe_off,
4713                                              int len,
4714                                              struct ocfs2_xattr_set_ctxt *ctxt)
4715 {
4716         int ret, offset;
4717         u64 value_blk;
4718         struct ocfs2_xattr_entry *xe;
4719         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4720         size_t blocksize = inode->i_sb->s_blocksize;
4721         struct ocfs2_xattr_value_buf vb = {
4722                 .vb_access = ocfs2_journal_access,
4723         };
4724
4725         xe = &xh->xh_entries[xe_off];
4726
4727         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4728
4729         offset = le16_to_cpu(xe->xe_name_offset) +
4730                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4731
4732         value_blk = offset / blocksize;
4733
4734         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4735         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4736
4737         vb.vb_bh = bucket->bu_bhs[value_blk];
4738         BUG_ON(!vb.vb_bh);
4739
4740         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4741                 (vb.vb_bh->b_data + offset % blocksize);
4742
4743         /*
4744          * From here on out we have to dirty the bucket.  The generic
4745          * value calls only modify one of the bucket's bhs, but we need
4746          * to send the bucket at once.  So if they error, they *could* have
4747          * modified something.  We have to assume they did, and dirty
4748          * the whole bucket.  This leaves us in a consistent state.
4749          */
4750         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4751              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4752         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4753         if (ret) {
4754                 mlog_errno(ret);
4755                 goto out;
4756         }
4757
4758         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4759                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4760         if (ret) {
4761                 mlog_errno(ret);
4762                 goto out;
4763         }
4764
4765         xe->xe_value_size = cpu_to_le64(len);
4766
4767         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4768
4769 out:
4770         return ret;
4771 }
4772
4773 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4774                                         struct ocfs2_xattr_search *xs,
4775                                         int len,
4776                                         struct ocfs2_xattr_set_ctxt *ctxt)
4777 {
4778         int ret, offset;
4779         struct ocfs2_xattr_entry *xe = xs->here;
4780         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4781
4782         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4783
4784         offset = xe - xh->xh_entries;
4785         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4786                                                 offset, len, ctxt);
4787         if (ret)
4788                 mlog_errno(ret);
4789
4790         return ret;
4791 }
4792
4793 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4794                                                 handle_t *handle,
4795                                                 struct ocfs2_xattr_search *xs,
4796                                                 char *val,
4797                                                 int value_len)
4798 {
4799         int ret, offset, block_off;
4800         struct ocfs2_xattr_value_root *xv;
4801         struct ocfs2_xattr_entry *xe = xs->here;
4802         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4803         void *base;
4804
4805         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4806
4807         ret = ocfs2_xattr_bucket_get_name_value(inode, xh,
4808                                                 xe - xh->xh_entries,
4809                                                 &block_off,
4810                                                 &offset);
4811         if (ret) {
4812                 mlog_errno(ret);
4813                 goto out;
4814         }
4815
4816         base = bucket_block(xs->bucket, block_off);
4817         xv = (struct ocfs2_xattr_value_root *)(base + offset +
4818                  OCFS2_XATTR_SIZE(xe->xe_name_len));
4819
4820         ret = __ocfs2_xattr_set_value_outside(inode, handle,
4821                                               xv, val, value_len);
4822         if (ret)
4823                 mlog_errno(ret);
4824 out:
4825         return ret;
4826 }
4827
4828 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4829                                   struct buffer_head *root_bh,
4830                                   u64 blkno,
4831                                   u32 cpos,
4832                                   u32 len)
4833 {
4834         int ret;
4835         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4836         struct inode *tl_inode = osb->osb_tl_inode;
4837         handle_t *handle;
4838         struct ocfs2_xattr_block *xb =
4839                         (struct ocfs2_xattr_block *)root_bh->b_data;
4840         struct ocfs2_alloc_context *meta_ac = NULL;
4841         struct ocfs2_cached_dealloc_ctxt dealloc;
4842         struct ocfs2_extent_tree et;
4843
4844         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4845
4846         ocfs2_init_dealloc_ctxt(&dealloc);
4847
4848         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4849              cpos, len, (unsigned long long)blkno);
4850
4851         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
4852                                                len);
4853
4854         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4855         if (ret) {
4856                 mlog_errno(ret);
4857                 return ret;
4858         }
4859
4860         mutex_lock(&tl_inode->i_mutex);
4861
4862         if (ocfs2_truncate_log_needs_flush(osb)) {
4863                 ret = __ocfs2_flush_truncate_log(osb);
4864                 if (ret < 0) {
4865                         mlog_errno(ret);
4866                         goto out;
4867                 }
4868         }
4869
4870         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4871         if (IS_ERR(handle)) {
4872                 ret = -ENOMEM;
4873                 mlog_errno(ret);
4874                 goto out;
4875         }
4876
4877         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4878                                       OCFS2_JOURNAL_ACCESS_WRITE);
4879         if (ret) {
4880                 mlog_errno(ret);
4881                 goto out_commit;
4882         }
4883
4884         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
4885                                   &dealloc);
4886         if (ret) {
4887                 mlog_errno(ret);
4888                 goto out_commit;
4889         }
4890
4891         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4892
4893         ret = ocfs2_journal_dirty(handle, root_bh);
4894         if (ret) {
4895                 mlog_errno(ret);
4896                 goto out_commit;
4897         }
4898
4899         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4900         if (ret)
4901                 mlog_errno(ret);
4902
4903 out_commit:
4904         ocfs2_commit_trans(osb, handle);
4905 out:
4906         ocfs2_schedule_truncate_log_flush(osb, 1);
4907
4908         mutex_unlock(&tl_inode->i_mutex);
4909
4910         if (meta_ac)
4911                 ocfs2_free_alloc_context(meta_ac);
4912
4913         ocfs2_run_deallocs(osb, &dealloc);
4914
4915         return ret;
4916 }
4917
4918 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4919                                          handle_t *handle,
4920                                          struct ocfs2_xattr_search *xs)
4921 {
4922         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4923         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4924                                                 le16_to_cpu(xh->xh_count) - 1];
4925         int ret = 0;
4926
4927         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4928                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4929         if (ret) {
4930                 mlog_errno(ret);
4931                 return;
4932         }
4933
4934         /* Remove the old entry. */
4935         memmove(xs->here, xs->here + 1,
4936                 (void *)last - (void *)xs->here);
4937         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4938         le16_add_cpu(&xh->xh_count, -1);
4939
4940         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4941 }
4942
4943 /*
4944  * Set the xattr name/value in the bucket specified in xs.
4945  *
4946  * As the new value in xi may be stored in the bucket or in an outside cluster,
4947  * we divide the whole process into 3 steps:
4948  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4949  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4950  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4951  * 4. If the clusters for the new outside value can't be allocated, we need
4952  *    to free the xattr we allocated in set.
4953  */
4954 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4955                                      struct ocfs2_xattr_info *xi,
4956                                      struct ocfs2_xattr_search *xs,
4957                                      struct ocfs2_xattr_set_ctxt *ctxt)
4958 {
4959         int ret, local = 1;
4960         size_t value_len;
4961         char *val = (char *)xi->value;
4962         struct ocfs2_xattr_entry *xe = xs->here;
4963         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4964                                               strlen(xi->name));
4965
4966         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4967                 /*
4968                  * We need to truncate the xattr storage first.
4969                  *
4970                  * If both the old and new value are stored to
4971                  * outside block, we only need to truncate
4972                  * the storage and then set the value outside.
4973                  *
4974                  * If the new value should be stored within block,
4975                  * we should free all the outside block first and
4976                  * the modification to the xattr block will be done
4977                  * by following steps.
4978                  */
4979                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4980                         value_len = xi->value_len;
4981                 else
4982                         value_len = 0;
4983
4984                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4985                                                            value_len,
4986                                                            ctxt);
4987                 if (ret)
4988                         goto out;
4989
4990                 if (value_len)
4991                         goto set_value_outside;
4992         }
4993
4994         value_len = xi->value_len;
4995         /* So we have to handle the inside block change now. */
4996         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4997                 /*
4998                  * If the new value will be stored outside of block,
4999                  * initalize a new empty value root and insert it first.
5000                  */
5001                 local = 0;
5002                 xi->value = &def_xv;
5003                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5004         }
5005
5006         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5007                                               name_hash, local);
5008         if (ret) {
5009                 mlog_errno(ret);
5010                 goto out;
5011         }
5012
5013         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5014                 goto out;
5015
5016         /* allocate the space now for the outside block storage. */
5017         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5018                                                    value_len, ctxt);
5019         if (ret) {
5020                 mlog_errno(ret);
5021
5022                 if (xs->not_found) {
5023                         /*
5024                          * We can't allocate enough clusters for outside
5025                          * storage and we have allocated xattr already,
5026                          * so need to remove it.
5027                          */
5028                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5029                 }
5030                 goto out;
5031         }
5032
5033 set_value_outside:
5034         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5035                                                    xs, val, value_len);
5036 out:
5037         return ret;
5038 }
5039
5040 /*
5041  * check whether the xattr bucket is filled up with the same hash value.
5042  * If we want to insert the xattr with the same hash, return -ENOSPC.
5043  * If we want to insert a xattr with different hash value, go ahead
5044  * and ocfs2_divide_xattr_bucket will handle this.
5045  */
5046 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5047                                               struct ocfs2_xattr_bucket *bucket,
5048                                               const char *name)
5049 {
5050         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5051         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5052
5053         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5054                 return 0;
5055
5056         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5057             xh->xh_entries[0].xe_name_hash) {
5058                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5059                      "hash = %u\n",
5060                      (unsigned long long)bucket_blkno(bucket),
5061                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5062                 return -ENOSPC;
5063         }
5064
5065         return 0;
5066 }
5067
5068 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5069                                              struct ocfs2_xattr_info *xi,
5070                                              struct ocfs2_xattr_search *xs,
5071                                              struct ocfs2_xattr_set_ctxt *ctxt)
5072 {
5073         struct ocfs2_xattr_header *xh;
5074         struct ocfs2_xattr_entry *xe;
5075         u16 count, header_size, xh_free_start;
5076         int free, max_free, need, old;
5077         size_t value_size = 0, name_len = strlen(xi->name);
5078         size_t blocksize = inode->i_sb->s_blocksize;
5079         int ret, allocation = 0;
5080
5081         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5082
5083 try_again:
5084         xh = xs->header;
5085         count = le16_to_cpu(xh->xh_count);
5086         xh_free_start = le16_to_cpu(xh->xh_free_start);
5087         header_size = sizeof(struct ocfs2_xattr_header) +
5088                         count * sizeof(struct ocfs2_xattr_entry);
5089         max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5090                 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5091
5092         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5093                         "of %u which exceed block size\n",
5094                         (unsigned long long)bucket_blkno(xs->bucket),
5095                         header_size);
5096
5097         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5098                 value_size = OCFS2_XATTR_ROOT_SIZE;
5099         else if (xi->value)
5100                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5101
5102         if (xs->not_found)
5103                 need = sizeof(struct ocfs2_xattr_entry) +
5104                         OCFS2_XATTR_SIZE(name_len) + value_size;
5105         else {
5106                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5107
5108                 /*
5109                  * We only replace the old value if the new length is smaller
5110                  * than the old one. Otherwise we will allocate new space in the
5111                  * bucket to store it.
5112                  */
5113                 xe = xs->here;
5114                 if (ocfs2_xattr_is_local(xe))
5115                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5116                 else
5117                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5118
5119                 if (old >= value_size)
5120                         need = 0;
5121         }
5122
5123         free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5124         /*
5125          * We need to make sure the new name/value pair
5126          * can exist in the same block.
5127          */
5128         if (xh_free_start % blocksize < need)
5129                 free -= xh_free_start % blocksize;
5130
5131         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5132              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5133              " %u\n", xs->not_found,
5134              (unsigned long long)bucket_blkno(xs->bucket),
5135              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5136              le16_to_cpu(xh->xh_name_value_len));
5137
5138         if (free < need ||
5139             (xs->not_found &&
5140              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5141                 if (need <= max_free &&
5142                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5143                         /*
5144                          * We can create the space by defragment. Since only the
5145                          * name/value will be moved, the xe shouldn't be changed
5146                          * in xs.
5147                          */
5148                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5149                                                         xs->bucket);
5150                         if (ret) {
5151                                 mlog_errno(ret);
5152                                 goto out;
5153                         }
5154
5155                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5156                         free = xh_free_start - header_size
5157                                 - OCFS2_XATTR_HEADER_GAP;
5158                         if (xh_free_start % blocksize < need)
5159                                 free -= xh_free_start % blocksize;
5160
5161                         if (free >= need)
5162                                 goto xattr_set;
5163
5164                         mlog(0, "Can't get enough space for xattr insert by "
5165                              "defragment. Need %u bytes, but we have %d, so "
5166                              "allocate new bucket for it.\n", need, free);
5167                 }
5168
5169                 /*
5170                  * We have to add new buckets or clusters and one
5171                  * allocation should leave us enough space for insert.
5172                  */
5173                 BUG_ON(allocation);
5174
5175                 /*
5176                  * We do not allow for overlapping ranges between buckets. And
5177                  * the maximum number of collisions we will allow for then is
5178                  * one bucket's worth, so check it here whether we need to
5179                  * add a new bucket for the insert.
5180                  */
5181                 ret = ocfs2_check_xattr_bucket_collision(inode,
5182                                                          xs->bucket,
5183                                                          xi->name);
5184                 if (ret) {
5185                         mlog_errno(ret);
5186                         goto out;
5187                 }
5188
5189                 ret = ocfs2_add_new_xattr_bucket(inode,
5190                                                  xs->xattr_bh,
5191                                                  xs->bucket,
5192                                                  ctxt);
5193                 if (ret) {
5194                         mlog_errno(ret);
5195                         goto out;
5196                 }
5197
5198                 /*
5199                  * ocfs2_add_new_xattr_bucket() will have updated
5200                  * xs->bucket if it moved, but it will not have updated
5201                  * any of the other search fields.  Thus, we drop it and
5202                  * re-search.  Everything should be cached, so it'll be
5203                  * quick.
5204                  */
5205                 ocfs2_xattr_bucket_relse(xs->bucket);
5206                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5207                                                    xi->name_index,
5208                                                    xi->name, xs);
5209                 if (ret && ret != -ENODATA)
5210                         goto out;
5211                 xs->not_found = ret;
5212                 allocation = 1;
5213                 goto try_again;
5214         }
5215
5216 xattr_set:
5217         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5218 out:
5219         mlog_exit(ret);
5220         return ret;
5221 }
5222
5223 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5224                                         struct ocfs2_xattr_bucket *bucket,
5225                                         void *para)
5226 {
5227         int ret = 0;
5228         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5229         u16 i;
5230         struct ocfs2_xattr_entry *xe;
5231         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5232         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5233         int credits = ocfs2_remove_extent_credits(osb->sb) +
5234                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5235
5236
5237         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5238
5239         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5240                 xe = &xh->xh_entries[i];
5241                 if (ocfs2_xattr_is_local(xe))
5242                         continue;
5243
5244                 ctxt.handle = ocfs2_start_trans(osb, credits);
5245                 if (IS_ERR(ctxt.handle)) {
5246                         ret = PTR_ERR(ctxt.handle);
5247                         mlog_errno(ret);
5248                         break;
5249                 }
5250
5251                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5252                                                         i, 0, &ctxt);
5253
5254                 ocfs2_commit_trans(osb, ctxt.handle);
5255                 if (ret) {
5256                         mlog_errno(ret);
5257                         break;
5258                 }
5259         }
5260
5261         ocfs2_schedule_truncate_log_flush(osb, 1);
5262         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5263         return ret;
5264 }
5265
5266 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5267                                           struct buffer_head *xb_bh)
5268 {
5269         struct ocfs2_xattr_block *xb =
5270                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5271         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5272         int ret = 0;
5273         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5274         u64 p_blkno;
5275
5276         if (le16_to_cpu(el->l_next_free_rec) == 0)
5277                 return 0;
5278
5279         while (name_hash > 0) {
5280                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5281                                           &e_cpos, &num_clusters, el);
5282                 if (ret) {
5283                         mlog_errno(ret);
5284                         goto out;
5285                 }
5286
5287                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5288                                                   ocfs2_delete_xattr_in_bucket,
5289                                                   NULL);
5290                 if (ret) {
5291                         mlog_errno(ret);
5292                         goto out;
5293                 }
5294
5295                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5296                                              p_blkno, e_cpos, num_clusters);
5297                 if (ret) {
5298                         mlog_errno(ret);
5299                         break;
5300                 }
5301
5302                 if (e_cpos == 0)
5303                         break;
5304
5305                 name_hash = e_cpos - 1;
5306         }
5307
5308 out:
5309         return ret;
5310 }
5311
5312 /*
5313  * 'security' attributes support
5314  */
5315 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5316                                         size_t list_size, const char *name,
5317                                         size_t name_len)
5318 {
5319         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5320         const size_t total_len = prefix_len + name_len + 1;
5321
5322         if (list && total_len <= list_size) {
5323                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5324                 memcpy(list + prefix_len, name, name_len);
5325                 list[prefix_len + name_len] = '\0';
5326         }
5327         return total_len;
5328 }
5329
5330 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5331                                     void *buffer, size_t size)
5332 {
5333         if (strcmp(name, "") == 0)
5334                 return -EINVAL;
5335         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5336                                buffer, size);
5337 }
5338
5339 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5340                                     const void *value, size_t size, int flags)
5341 {
5342         if (strcmp(name, "") == 0)
5343                 return -EINVAL;
5344
5345         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5346                                size, flags);
5347 }
5348
5349 int ocfs2_init_security_get(struct inode *inode,
5350                             struct inode *dir,
5351                             struct ocfs2_security_xattr_info *si)
5352 {
5353         /* check whether ocfs2 support feature xattr */
5354         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5355                 return -EOPNOTSUPP;
5356         return security_inode_init_security(inode, dir, &si->name, &si->value,
5357                                             &si->value_len);
5358 }
5359
5360 int ocfs2_init_security_set(handle_t *handle,
5361                             struct inode *inode,
5362                             struct buffer_head *di_bh,
5363                             struct ocfs2_security_xattr_info *si,
5364                             struct ocfs2_alloc_context *xattr_ac,
5365                             struct ocfs2_alloc_context *data_ac)
5366 {
5367         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5368                                      OCFS2_XATTR_INDEX_SECURITY,
5369                                      si->name, si->value, si->value_len, 0,
5370                                      xattr_ac, data_ac);
5371 }
5372
5373 struct xattr_handler ocfs2_xattr_security_handler = {
5374         .prefix = XATTR_SECURITY_PREFIX,
5375         .list   = ocfs2_xattr_security_list,
5376         .get    = ocfs2_xattr_security_get,
5377         .set    = ocfs2_xattr_security_set,
5378 };
5379
5380 /*
5381  * 'trusted' attributes support
5382  */
5383 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5384                                        size_t list_size, const char *name,
5385                                        size_t name_len)
5386 {
5387         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5388         const size_t total_len = prefix_len + name_len + 1;
5389
5390         if (list && total_len <= list_size) {
5391                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5392                 memcpy(list + prefix_len, name, name_len);
5393                 list[prefix_len + name_len] = '\0';
5394         }
5395         return total_len;
5396 }
5397
5398 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5399                                    void *buffer, size_t size)
5400 {
5401         if (strcmp(name, "") == 0)
5402                 return -EINVAL;
5403         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5404                                buffer, size);
5405 }
5406
5407 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5408                                    const void *value, size_t size, int flags)
5409 {
5410         if (strcmp(name, "") == 0)
5411                 return -EINVAL;
5412
5413         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5414                                size, flags);
5415 }
5416
5417 struct xattr_handler ocfs2_xattr_trusted_handler = {
5418         .prefix = XATTR_TRUSTED_PREFIX,
5419         .list   = ocfs2_xattr_trusted_list,
5420         .get    = ocfs2_xattr_trusted_get,
5421         .set    = ocfs2_xattr_trusted_set,
5422 };
5423
5424 /*
5425  * 'user' attributes support
5426  */
5427 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5428                                     size_t list_size, const char *name,
5429                                     size_t name_len)
5430 {
5431         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5432         const size_t total_len = prefix_len + name_len + 1;
5433         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5434
5435         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5436                 return 0;
5437
5438         if (list && total_len <= list_size) {
5439                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5440                 memcpy(list + prefix_len, name, name_len);
5441                 list[prefix_len + name_len] = '\0';
5442         }
5443         return total_len;
5444 }
5445
5446 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5447                                 void *buffer, size_t size)
5448 {
5449         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5450
5451         if (strcmp(name, "") == 0)
5452                 return -EINVAL;
5453         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5454                 return -EOPNOTSUPP;
5455         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5456                                buffer, size);
5457 }
5458
5459 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5460                                 const void *value, size_t size, int flags)
5461 {
5462         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5463
5464         if (strcmp(name, "") == 0)
5465                 return -EINVAL;
5466         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5467                 return -EOPNOTSUPP;
5468
5469         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5470                                size, flags);
5471 }
5472
5473 struct xattr_handler ocfs2_xattr_user_handler = {
5474         .prefix = XATTR_USER_PREFIX,
5475         .list   = ocfs2_xattr_user_list,
5476         .get    = ocfs2_xattr_user_get,
5477         .set    = ocfs2_xattr_user_set,
5478 };