945ca697eb25da350782526dff7cef3519b3e362
[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 #include "refcounttree.h"
59 #include "acl.h"
60
61 struct ocfs2_xattr_def_value_root {
62         struct ocfs2_xattr_value_root   xv;
63         struct ocfs2_extent_rec         er;
64 };
65
66 struct ocfs2_xattr_bucket {
67         /* The inode these xattrs are associated with */
68         struct inode *bu_inode;
69
70         /* The actual buffers that make up the bucket */
71         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
72
73         /* How many blocks make up one bucket for this filesystem */
74         int bu_blocks;
75 };
76
77 struct ocfs2_xattr_set_ctxt {
78         handle_t *handle;
79         struct ocfs2_alloc_context *meta_ac;
80         struct ocfs2_alloc_context *data_ac;
81         struct ocfs2_cached_dealloc_ctxt dealloc;
82 };
83
84 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
85 #define OCFS2_XATTR_INLINE_SIZE 80
86 #define OCFS2_XATTR_HEADER_GAP  4
87 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
88                                          - sizeof(struct ocfs2_xattr_header) \
89                                          - OCFS2_XATTR_HEADER_GAP)
90 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
91                                          - sizeof(struct ocfs2_xattr_block) \
92                                          - sizeof(struct ocfs2_xattr_header) \
93                                          - OCFS2_XATTR_HEADER_GAP)
94
95 static struct ocfs2_xattr_def_value_root def_xv = {
96         .xv.xr_list.l_count = cpu_to_le16(1),
97 };
98
99 struct xattr_handler *ocfs2_xattr_handlers[] = {
100         &ocfs2_xattr_user_handler,
101         &ocfs2_xattr_acl_access_handler,
102         &ocfs2_xattr_acl_default_handler,
103         &ocfs2_xattr_trusted_handler,
104         &ocfs2_xattr_security_handler,
105         NULL
106 };
107
108 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
109         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
110         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
111                                         = &ocfs2_xattr_acl_access_handler,
112         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
113                                         = &ocfs2_xattr_acl_default_handler,
114         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
115         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
116 };
117
118 struct ocfs2_xattr_info {
119         int name_index;
120         const char *name;
121         const void *value;
122         size_t value_len;
123 };
124
125 struct ocfs2_xattr_search {
126         struct buffer_head *inode_bh;
127         /*
128          * xattr_bh point to the block buffer head which has extended attribute
129          * when extended attribute in inode, xattr_bh is equal to inode_bh.
130          */
131         struct buffer_head *xattr_bh;
132         struct ocfs2_xattr_header *header;
133         struct ocfs2_xattr_bucket *bucket;
134         void *base;
135         void *end;
136         struct ocfs2_xattr_entry *here;
137         int not_found;
138 };
139
140 /* Operations on struct ocfs2_xa_entry */
141 struct ocfs2_xa_loc;
142 struct ocfs2_xa_loc_operations {
143         /*
144          * Return a pointer to the appropriate buffer in loc->xl_storage
145          * at the given offset from loc->xl_header.
146          */
147         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
148
149         /*
150          * Remove the name+value at this location.  Do whatever is
151          * appropriate with the remaining name+value pairs.
152          */
153         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
154 };
155
156 /*
157  * Describes an xattr entry location.  This is a memory structure
158  * tracking the on-disk structure.
159  */
160 struct ocfs2_xa_loc {
161         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
162         struct ocfs2_xattr_header *xl_header;
163
164         /* Bytes from xl_header to the end of the storage */
165         int xl_size;
166
167         /*
168          * The ocfs2_xattr_entry this location describes.  If this is
169          * NULL, this location describes the on-disk structure where it
170          * would have been.
171          */
172         struct ocfs2_xattr_entry *xl_entry;
173
174         /*
175          * Internal housekeeping
176          */
177
178         /* Buffer(s) containing this entry */
179         void *xl_storage;
180
181         /* Operations on the storage backing this location */
182         const struct ocfs2_xa_loc_operations *xl_ops;
183 };
184
185 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
186                                              struct ocfs2_xattr_header *xh,
187                                              int index,
188                                              int *block_off,
189                                              int *new_offset);
190
191 static int ocfs2_xattr_block_find(struct inode *inode,
192                                   int name_index,
193                                   const char *name,
194                                   struct ocfs2_xattr_search *xs);
195 static int ocfs2_xattr_index_block_find(struct inode *inode,
196                                         struct buffer_head *root_bh,
197                                         int name_index,
198                                         const char *name,
199                                         struct ocfs2_xattr_search *xs);
200
201 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
202                                         struct buffer_head *blk_bh,
203                                         char *buffer,
204                                         size_t buffer_size);
205
206 static int ocfs2_xattr_create_index_block(struct inode *inode,
207                                           struct ocfs2_xattr_search *xs,
208                                           struct ocfs2_xattr_set_ctxt *ctxt);
209
210 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
211                                              struct ocfs2_xattr_info *xi,
212                                              struct ocfs2_xattr_search *xs,
213                                              struct ocfs2_xattr_set_ctxt *ctxt);
214
215 typedef int (xattr_tree_rec_func)(struct inode *inode,
216                                   struct buffer_head *root_bh,
217                                   u64 blkno, u32 cpos, u32 len, void *para);
218 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
219                                            struct buffer_head *root_bh,
220                                            xattr_tree_rec_func *rec_func,
221                                            void *para);
222 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
223                                         struct ocfs2_xattr_bucket *bucket,
224                                         void *para);
225 static int ocfs2_rm_xattr_cluster(struct inode *inode,
226                                   struct buffer_head *root_bh,
227                                   u64 blkno,
228                                   u32 cpos,
229                                   u32 len,
230                                   void *para);
231
232 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
233                                   u64 src_blk, u64 last_blk, u64 to_blk,
234                                   unsigned int start_bucket,
235                                   u32 *first_hash);
236 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
237                                         struct ocfs2_dinode *di,
238                                         struct ocfs2_xattr_info *xi,
239                                         struct ocfs2_xattr_search *xis,
240                                         struct ocfs2_xattr_search *xbs,
241                                         struct ocfs2_refcount_tree **ref_tree,
242                                         int *meta_need,
243                                         int *credits);
244 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
245                                            struct ocfs2_xattr_bucket *bucket,
246                                            int offset,
247                                            struct ocfs2_xattr_value_root **xv,
248                                            struct buffer_head **bh);
249
250 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
251 {
252         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
253 }
254
255 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
256 {
257         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
258 }
259
260 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
261 {
262         u16 len = sb->s_blocksize -
263                  offsetof(struct ocfs2_xattr_header, xh_entries);
264
265         return len / sizeof(struct ocfs2_xattr_entry);
266 }
267
268 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
269 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
270 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
271
272 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
273 {
274         struct ocfs2_xattr_bucket *bucket;
275         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
276
277         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
278
279         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
280         if (bucket) {
281                 bucket->bu_inode = inode;
282                 bucket->bu_blocks = blks;
283         }
284
285         return bucket;
286 }
287
288 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
289 {
290         int i;
291
292         for (i = 0; i < bucket->bu_blocks; i++) {
293                 brelse(bucket->bu_bhs[i]);
294                 bucket->bu_bhs[i] = NULL;
295         }
296 }
297
298 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
299 {
300         if (bucket) {
301                 ocfs2_xattr_bucket_relse(bucket);
302                 bucket->bu_inode = NULL;
303                 kfree(bucket);
304         }
305 }
306
307 /*
308  * A bucket that has never been written to disk doesn't need to be
309  * read.  We just need the buffer_heads.  Don't call this for
310  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
311  * them fully.
312  */
313 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
314                                    u64 xb_blkno)
315 {
316         int i, rc = 0;
317
318         for (i = 0; i < bucket->bu_blocks; i++) {
319                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
320                                               xb_blkno + i);
321                 if (!bucket->bu_bhs[i]) {
322                         rc = -EIO;
323                         mlog_errno(rc);
324                         break;
325                 }
326
327                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
328                                            bucket->bu_bhs[i]))
329                         ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
330                                                       bucket->bu_bhs[i]);
331         }
332
333         if (rc)
334                 ocfs2_xattr_bucket_relse(bucket);
335         return rc;
336 }
337
338 /* Read the xattr bucket at xb_blkno */
339 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
340                                    u64 xb_blkno)
341 {
342         int rc;
343
344         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
345                                bucket->bu_blocks, bucket->bu_bhs, 0,
346                                NULL);
347         if (!rc) {
348                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
349                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
350                                                  bucket->bu_bhs,
351                                                  bucket->bu_blocks,
352                                                  &bucket_xh(bucket)->xh_check);
353                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
354                 if (rc)
355                         mlog_errno(rc);
356         }
357
358         if (rc)
359                 ocfs2_xattr_bucket_relse(bucket);
360         return rc;
361 }
362
363 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
364                                              struct ocfs2_xattr_bucket *bucket,
365                                              int type)
366 {
367         int i, rc = 0;
368
369         for (i = 0; i < bucket->bu_blocks; i++) {
370                 rc = ocfs2_journal_access(handle,
371                                           INODE_CACHE(bucket->bu_inode),
372                                           bucket->bu_bhs[i], type);
373                 if (rc) {
374                         mlog_errno(rc);
375                         break;
376                 }
377         }
378
379         return rc;
380 }
381
382 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
383                                              struct ocfs2_xattr_bucket *bucket)
384 {
385         int i;
386
387         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
388         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
389                                    bucket->bu_bhs, bucket->bu_blocks,
390                                    &bucket_xh(bucket)->xh_check);
391         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
392
393         for (i = 0; i < bucket->bu_blocks; i++)
394                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
395 }
396
397 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
398                                          struct ocfs2_xattr_bucket *src)
399 {
400         int i;
401         int blocksize = src->bu_inode->i_sb->s_blocksize;
402
403         BUG_ON(dest->bu_blocks != src->bu_blocks);
404         BUG_ON(dest->bu_inode != src->bu_inode);
405
406         for (i = 0; i < src->bu_blocks; i++) {
407                 memcpy(bucket_block(dest, i), bucket_block(src, i),
408                        blocksize);
409         }
410 }
411
412 static int ocfs2_validate_xattr_block(struct super_block *sb,
413                                       struct buffer_head *bh)
414 {
415         int rc;
416         struct ocfs2_xattr_block *xb =
417                 (struct ocfs2_xattr_block *)bh->b_data;
418
419         mlog(0, "Validating xattr block %llu\n",
420              (unsigned long long)bh->b_blocknr);
421
422         BUG_ON(!buffer_uptodate(bh));
423
424         /*
425          * If the ecc fails, we return the error but otherwise
426          * leave the filesystem running.  We know any error is
427          * local to this block.
428          */
429         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
430         if (rc)
431                 return rc;
432
433         /*
434          * Errors after here are fatal
435          */
436
437         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
438                 ocfs2_error(sb,
439                             "Extended attribute block #%llu has bad "
440                             "signature %.*s",
441                             (unsigned long long)bh->b_blocknr, 7,
442                             xb->xb_signature);
443                 return -EINVAL;
444         }
445
446         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
447                 ocfs2_error(sb,
448                             "Extended attribute block #%llu has an "
449                             "invalid xb_blkno of %llu",
450                             (unsigned long long)bh->b_blocknr,
451                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
452                 return -EINVAL;
453         }
454
455         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
456                 ocfs2_error(sb,
457                             "Extended attribute block #%llu has an invalid "
458                             "xb_fs_generation of #%u",
459                             (unsigned long long)bh->b_blocknr,
460                             le32_to_cpu(xb->xb_fs_generation));
461                 return -EINVAL;
462         }
463
464         return 0;
465 }
466
467 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
468                                   struct buffer_head **bh)
469 {
470         int rc;
471         struct buffer_head *tmp = *bh;
472
473         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
474                               ocfs2_validate_xattr_block);
475
476         /* If ocfs2_read_block() got us a new bh, pass it up. */
477         if (!rc && !*bh)
478                 *bh = tmp;
479
480         return rc;
481 }
482
483 static inline const char *ocfs2_xattr_prefix(int name_index)
484 {
485         struct xattr_handler *handler = NULL;
486
487         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
488                 handler = ocfs2_xattr_handler_map[name_index];
489
490         return handler ? handler->prefix : NULL;
491 }
492
493 static u32 ocfs2_xattr_name_hash(struct inode *inode,
494                                  const char *name,
495                                  int name_len)
496 {
497         /* Get hash value of uuid from super block */
498         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
499         int i;
500
501         /* hash extended attribute name */
502         for (i = 0; i < name_len; i++) {
503                 hash = (hash << OCFS2_HASH_SHIFT) ^
504                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
505                        *name++;
506         }
507
508         return hash;
509 }
510
511 /*
512  * ocfs2_xattr_hash_entry()
513  *
514  * Compute the hash of an extended attribute.
515  */
516 static void ocfs2_xattr_hash_entry(struct inode *inode,
517                                    struct ocfs2_xattr_header *header,
518                                    struct ocfs2_xattr_entry *entry)
519 {
520         u32 hash = 0;
521         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
522
523         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
524         entry->xe_name_hash = cpu_to_le32(hash);
525
526         return;
527 }
528
529 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
530 {
531         int size = 0;
532
533         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
534                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
535         else
536                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
537         size += sizeof(struct ocfs2_xattr_entry);
538
539         return size;
540 }
541
542 int ocfs2_calc_security_init(struct inode *dir,
543                              struct ocfs2_security_xattr_info *si,
544                              int *want_clusters,
545                              int *xattr_credits,
546                              struct ocfs2_alloc_context **xattr_ac)
547 {
548         int ret = 0;
549         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
550         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
551                                                  si->value_len);
552
553         /*
554          * The max space of security xattr taken inline is
555          * 256(name) + 80(value) + 16(entry) = 352 bytes,
556          * So reserve one metadata block for it is ok.
557          */
558         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
559             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
560                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
561                 if (ret) {
562                         mlog_errno(ret);
563                         return ret;
564                 }
565                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
566         }
567
568         /* reserve clusters for xattr value which will be set in B tree*/
569         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
570                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
571                                                             si->value_len);
572
573                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
574                                                            new_clusters);
575                 *want_clusters += new_clusters;
576         }
577         return ret;
578 }
579
580 int ocfs2_calc_xattr_init(struct inode *dir,
581                           struct buffer_head *dir_bh,
582                           int mode,
583                           struct ocfs2_security_xattr_info *si,
584                           int *want_clusters,
585                           int *xattr_credits,
586                           int *want_meta)
587 {
588         int ret = 0;
589         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
590         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
591
592         if (si->enable)
593                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
594                                                      si->value_len);
595
596         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
597                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
598                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
599                                         "", NULL, 0);
600                 if (acl_len > 0) {
601                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
602                         if (S_ISDIR(mode))
603                                 a_size <<= 1;
604                 } else if (acl_len != 0 && acl_len != -ENODATA) {
605                         mlog_errno(ret);
606                         return ret;
607                 }
608         }
609
610         if (!(s_size + a_size))
611                 return ret;
612
613         /*
614          * The max space of security xattr taken inline is
615          * 256(name) + 80(value) + 16(entry) = 352 bytes,
616          * The max space of acl xattr taken inline is
617          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
618          * when blocksize = 512, may reserve one more cluser for
619          * xattr bucket, otherwise reserve one metadata block
620          * for them is ok.
621          * If this is a new directory with inline data,
622          * we choose to reserve the entire inline area for
623          * directory contents and force an external xattr block.
624          */
625         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
626             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
627             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
628                 *want_meta = *want_meta + 1;
629                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
630         }
631
632         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
633             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
634                 *want_clusters += 1;
635                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
636         }
637
638         /*
639          * reserve credits and clusters for xattrs which has large value
640          * and have to be set outside
641          */
642         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
643                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
644                                                         si->value_len);
645                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
646                                                            new_clusters);
647                 *want_clusters += new_clusters;
648         }
649         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
650             acl_len > OCFS2_XATTR_INLINE_SIZE) {
651                 /* for directory, it has DEFAULT and ACCESS two types of acls */
652                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
653                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
654                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
655                                                            new_clusters);
656                 *want_clusters += new_clusters;
657         }
658
659         return ret;
660 }
661
662 static int ocfs2_xattr_extend_allocation(struct inode *inode,
663                                          u32 clusters_to_add,
664                                          struct ocfs2_xattr_value_buf *vb,
665                                          struct ocfs2_xattr_set_ctxt *ctxt)
666 {
667         int status = 0;
668         handle_t *handle = ctxt->handle;
669         enum ocfs2_alloc_restarted why;
670         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
671         struct ocfs2_extent_tree et;
672
673         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
674
675         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
676
677         status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
678                               OCFS2_JOURNAL_ACCESS_WRITE);
679         if (status < 0) {
680                 mlog_errno(status);
681                 goto leave;
682         }
683
684         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
685         status = ocfs2_add_clusters_in_btree(handle,
686                                              &et,
687                                              &logical_start,
688                                              clusters_to_add,
689                                              0,
690                                              ctxt->data_ac,
691                                              ctxt->meta_ac,
692                                              &why);
693         if (status < 0) {
694                 mlog_errno(status);
695                 goto leave;
696         }
697
698         status = ocfs2_journal_dirty(handle, vb->vb_bh);
699         if (status < 0) {
700                 mlog_errno(status);
701                 goto leave;
702         }
703
704         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
705
706         /*
707          * We should have already allocated enough space before the transaction,
708          * so no need to restart.
709          */
710         BUG_ON(why != RESTART_NONE || clusters_to_add);
711
712 leave:
713
714         return status;
715 }
716
717 static int __ocfs2_remove_xattr_range(struct inode *inode,
718                                       struct ocfs2_xattr_value_buf *vb,
719                                       u32 cpos, u32 phys_cpos, u32 len,
720                                       unsigned int ext_flags,
721                                       struct ocfs2_xattr_set_ctxt *ctxt)
722 {
723         int ret;
724         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
725         handle_t *handle = ctxt->handle;
726         struct ocfs2_extent_tree et;
727
728         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
729
730         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
731                             OCFS2_JOURNAL_ACCESS_WRITE);
732         if (ret) {
733                 mlog_errno(ret);
734                 goto out;
735         }
736
737         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
738                                   &ctxt->dealloc);
739         if (ret) {
740                 mlog_errno(ret);
741                 goto out;
742         }
743
744         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
745
746         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
747         if (ret) {
748                 mlog_errno(ret);
749                 goto out;
750         }
751
752         if (ext_flags & OCFS2_EXT_REFCOUNTED)
753                 ret = ocfs2_decrease_refcount(inode, handle,
754                                         ocfs2_blocks_to_clusters(inode->i_sb,
755                                                                  phys_blkno),
756                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
757         else
758                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
759                                                   phys_blkno, len);
760         if (ret)
761                 mlog_errno(ret);
762
763 out:
764         return ret;
765 }
766
767 static int ocfs2_xattr_shrink_size(struct inode *inode,
768                                    u32 old_clusters,
769                                    u32 new_clusters,
770                                    struct ocfs2_xattr_value_buf *vb,
771                                    struct ocfs2_xattr_set_ctxt *ctxt)
772 {
773         int ret = 0;
774         unsigned int ext_flags;
775         u32 trunc_len, cpos, phys_cpos, alloc_size;
776         u64 block;
777
778         if (old_clusters <= new_clusters)
779                 return 0;
780
781         cpos = new_clusters;
782         trunc_len = old_clusters - new_clusters;
783         while (trunc_len) {
784                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
785                                                &alloc_size,
786                                                &vb->vb_xv->xr_list, &ext_flags);
787                 if (ret) {
788                         mlog_errno(ret);
789                         goto out;
790                 }
791
792                 if (alloc_size > trunc_len)
793                         alloc_size = trunc_len;
794
795                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
796                                                  phys_cpos, alloc_size,
797                                                  ext_flags, ctxt);
798                 if (ret) {
799                         mlog_errno(ret);
800                         goto out;
801                 }
802
803                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
804                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
805                                                        block, alloc_size);
806                 cpos += alloc_size;
807                 trunc_len -= alloc_size;
808         }
809
810 out:
811         return ret;
812 }
813
814 static int ocfs2_xattr_value_truncate(struct inode *inode,
815                                       struct ocfs2_xattr_value_buf *vb,
816                                       int len,
817                                       struct ocfs2_xattr_set_ctxt *ctxt)
818 {
819         int ret;
820         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
821         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
822
823         if (new_clusters == old_clusters)
824                 return 0;
825
826         if (new_clusters > old_clusters)
827                 ret = ocfs2_xattr_extend_allocation(inode,
828                                                     new_clusters - old_clusters,
829                                                     vb, ctxt);
830         else
831                 ret = ocfs2_xattr_shrink_size(inode,
832                                               old_clusters, new_clusters,
833                                               vb, ctxt);
834
835         return ret;
836 }
837
838 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
839                                   size_t *result, const char *prefix,
840                                   const char *name, int name_len)
841 {
842         char *p = buffer + *result;
843         int prefix_len = strlen(prefix);
844         int total_len = prefix_len + name_len + 1;
845
846         *result += total_len;
847
848         /* we are just looking for how big our buffer needs to be */
849         if (!size)
850                 return 0;
851
852         if (*result > size)
853                 return -ERANGE;
854
855         memcpy(p, prefix, prefix_len);
856         memcpy(p + prefix_len, name, name_len);
857         p[prefix_len + name_len] = '\0';
858
859         return 0;
860 }
861
862 static int ocfs2_xattr_list_entries(struct inode *inode,
863                                     struct ocfs2_xattr_header *header,
864                                     char *buffer, size_t buffer_size)
865 {
866         size_t result = 0;
867         int i, type, ret;
868         const char *prefix, *name;
869
870         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
871                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
872                 type = ocfs2_xattr_get_type(entry);
873                 prefix = ocfs2_xattr_prefix(type);
874
875                 if (prefix) {
876                         name = (const char *)header +
877                                 le16_to_cpu(entry->xe_name_offset);
878
879                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
880                                                      &result, prefix, name,
881                                                      entry->xe_name_len);
882                         if (ret)
883                                 return ret;
884                 }
885         }
886
887         return result;
888 }
889
890 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
891                                          struct ocfs2_dinode *di)
892 {
893         struct ocfs2_xattr_header *xh;
894         int i;
895
896         xh = (struct ocfs2_xattr_header *)
897                  ((void *)di + inode->i_sb->s_blocksize -
898                  le16_to_cpu(di->i_xattr_inline_size));
899
900         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
901                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
902                         return 1;
903
904         return 0;
905 }
906
907 static int ocfs2_xattr_ibody_list(struct inode *inode,
908                                   struct ocfs2_dinode *di,
909                                   char *buffer,
910                                   size_t buffer_size)
911 {
912         struct ocfs2_xattr_header *header = NULL;
913         struct ocfs2_inode_info *oi = OCFS2_I(inode);
914         int ret = 0;
915
916         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
917                 return ret;
918
919         header = (struct ocfs2_xattr_header *)
920                  ((void *)di + inode->i_sb->s_blocksize -
921                  le16_to_cpu(di->i_xattr_inline_size));
922
923         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
924
925         return ret;
926 }
927
928 static int ocfs2_xattr_block_list(struct inode *inode,
929                                   struct ocfs2_dinode *di,
930                                   char *buffer,
931                                   size_t buffer_size)
932 {
933         struct buffer_head *blk_bh = NULL;
934         struct ocfs2_xattr_block *xb;
935         int ret = 0;
936
937         if (!di->i_xattr_loc)
938                 return ret;
939
940         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
941                                      &blk_bh);
942         if (ret < 0) {
943                 mlog_errno(ret);
944                 return ret;
945         }
946
947         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
948         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
949                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
950                 ret = ocfs2_xattr_list_entries(inode, header,
951                                                buffer, buffer_size);
952         } else
953                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
954                                                    buffer, buffer_size);
955
956         brelse(blk_bh);
957
958         return ret;
959 }
960
961 ssize_t ocfs2_listxattr(struct dentry *dentry,
962                         char *buffer,
963                         size_t size)
964 {
965         int ret = 0, i_ret = 0, b_ret = 0;
966         struct buffer_head *di_bh = NULL;
967         struct ocfs2_dinode *di = NULL;
968         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
969
970         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
971                 return -EOPNOTSUPP;
972
973         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
974                 return ret;
975
976         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
977         if (ret < 0) {
978                 mlog_errno(ret);
979                 return ret;
980         }
981
982         di = (struct ocfs2_dinode *)di_bh->b_data;
983
984         down_read(&oi->ip_xattr_sem);
985         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
986         if (i_ret < 0)
987                 b_ret = 0;
988         else {
989                 if (buffer) {
990                         buffer += i_ret;
991                         size -= i_ret;
992                 }
993                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
994                                                buffer, size);
995                 if (b_ret < 0)
996                         i_ret = 0;
997         }
998         up_read(&oi->ip_xattr_sem);
999         ocfs2_inode_unlock(dentry->d_inode, 0);
1000
1001         brelse(di_bh);
1002
1003         return i_ret + b_ret;
1004 }
1005
1006 static int ocfs2_xattr_find_entry(int name_index,
1007                                   const char *name,
1008                                   struct ocfs2_xattr_search *xs)
1009 {
1010         struct ocfs2_xattr_entry *entry;
1011         size_t name_len;
1012         int i, cmp = 1;
1013
1014         if (name == NULL)
1015                 return -EINVAL;
1016
1017         name_len = strlen(name);
1018         entry = xs->here;
1019         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1020                 cmp = name_index - ocfs2_xattr_get_type(entry);
1021                 if (!cmp)
1022                         cmp = name_len - entry->xe_name_len;
1023                 if (!cmp)
1024                         cmp = memcmp(name, (xs->base +
1025                                      le16_to_cpu(entry->xe_name_offset)),
1026                                      name_len);
1027                 if (cmp == 0)
1028                         break;
1029                 entry += 1;
1030         }
1031         xs->here = entry;
1032
1033         return cmp ? -ENODATA : 0;
1034 }
1035
1036 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1037                                          struct ocfs2_xattr_value_root *xv,
1038                                          void *buffer,
1039                                          size_t len)
1040 {
1041         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1042         u64 blkno;
1043         int i, ret = 0;
1044         size_t cplen, blocksize;
1045         struct buffer_head *bh = NULL;
1046         struct ocfs2_extent_list *el;
1047
1048         el = &xv->xr_list;
1049         clusters = le32_to_cpu(xv->xr_clusters);
1050         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1051         blocksize = inode->i_sb->s_blocksize;
1052
1053         cpos = 0;
1054         while (cpos < clusters) {
1055                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1056                                                &num_clusters, el, NULL);
1057                 if (ret) {
1058                         mlog_errno(ret);
1059                         goto out;
1060                 }
1061
1062                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1063                 /* Copy ocfs2_xattr_value */
1064                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1065                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1066                                                &bh, NULL);
1067                         if (ret) {
1068                                 mlog_errno(ret);
1069                                 goto out;
1070                         }
1071
1072                         cplen = len >= blocksize ? blocksize : len;
1073                         memcpy(buffer, bh->b_data, cplen);
1074                         len -= cplen;
1075                         buffer += cplen;
1076
1077                         brelse(bh);
1078                         bh = NULL;
1079                         if (len == 0)
1080                                 break;
1081                 }
1082                 cpos += num_clusters;
1083         }
1084 out:
1085         return ret;
1086 }
1087
1088 static int ocfs2_xattr_ibody_get(struct inode *inode,
1089                                  int name_index,
1090                                  const char *name,
1091                                  void *buffer,
1092                                  size_t buffer_size,
1093                                  struct ocfs2_xattr_search *xs)
1094 {
1095         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1096         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1097         struct ocfs2_xattr_value_root *xv;
1098         size_t size;
1099         int ret = 0;
1100
1101         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1102                 return -ENODATA;
1103
1104         xs->end = (void *)di + inode->i_sb->s_blocksize;
1105         xs->header = (struct ocfs2_xattr_header *)
1106                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1107         xs->base = (void *)xs->header;
1108         xs->here = xs->header->xh_entries;
1109
1110         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1111         if (ret)
1112                 return ret;
1113         size = le64_to_cpu(xs->here->xe_value_size);
1114         if (buffer) {
1115                 if (size > buffer_size)
1116                         return -ERANGE;
1117                 if (ocfs2_xattr_is_local(xs->here)) {
1118                         memcpy(buffer, (void *)xs->base +
1119                                le16_to_cpu(xs->here->xe_name_offset) +
1120                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1121                 } else {
1122                         xv = (struct ocfs2_xattr_value_root *)
1123                                 (xs->base + le16_to_cpu(
1124                                  xs->here->xe_name_offset) +
1125                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1126                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1127                                                             buffer, size);
1128                         if (ret < 0) {
1129                                 mlog_errno(ret);
1130                                 return ret;
1131                         }
1132                 }
1133         }
1134
1135         return size;
1136 }
1137
1138 static int ocfs2_xattr_block_get(struct inode *inode,
1139                                  int name_index,
1140                                  const char *name,
1141                                  void *buffer,
1142                                  size_t buffer_size,
1143                                  struct ocfs2_xattr_search *xs)
1144 {
1145         struct ocfs2_xattr_block *xb;
1146         struct ocfs2_xattr_value_root *xv;
1147         size_t size;
1148         int ret = -ENODATA, name_offset, name_len, i;
1149         int uninitialized_var(block_off);
1150
1151         xs->bucket = ocfs2_xattr_bucket_new(inode);
1152         if (!xs->bucket) {
1153                 ret = -ENOMEM;
1154                 mlog_errno(ret);
1155                 goto cleanup;
1156         }
1157
1158         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1159         if (ret) {
1160                 mlog_errno(ret);
1161                 goto cleanup;
1162         }
1163
1164         if (xs->not_found) {
1165                 ret = -ENODATA;
1166                 goto cleanup;
1167         }
1168
1169         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1170         size = le64_to_cpu(xs->here->xe_value_size);
1171         if (buffer) {
1172                 ret = -ERANGE;
1173                 if (size > buffer_size)
1174                         goto cleanup;
1175
1176                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1177                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1178                 i = xs->here - xs->header->xh_entries;
1179
1180                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1181                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1182                                                                 bucket_xh(xs->bucket),
1183                                                                 i,
1184                                                                 &block_off,
1185                                                                 &name_offset);
1186                         xs->base = bucket_block(xs->bucket, block_off);
1187                 }
1188                 if (ocfs2_xattr_is_local(xs->here)) {
1189                         memcpy(buffer, (void *)xs->base +
1190                                name_offset + name_len, size);
1191                 } else {
1192                         xv = (struct ocfs2_xattr_value_root *)
1193                                 (xs->base + name_offset + name_len);
1194                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1195                                                             buffer, size);
1196                         if (ret < 0) {
1197                                 mlog_errno(ret);
1198                                 goto cleanup;
1199                         }
1200                 }
1201         }
1202         ret = size;
1203 cleanup:
1204         ocfs2_xattr_bucket_free(xs->bucket);
1205
1206         brelse(xs->xattr_bh);
1207         xs->xattr_bh = NULL;
1208         return ret;
1209 }
1210
1211 int ocfs2_xattr_get_nolock(struct inode *inode,
1212                            struct buffer_head *di_bh,
1213                            int name_index,
1214                            const char *name,
1215                            void *buffer,
1216                            size_t buffer_size)
1217 {
1218         int ret;
1219         struct ocfs2_dinode *di = NULL;
1220         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1221         struct ocfs2_xattr_search xis = {
1222                 .not_found = -ENODATA,
1223         };
1224         struct ocfs2_xattr_search xbs = {
1225                 .not_found = -ENODATA,
1226         };
1227
1228         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1229                 return -EOPNOTSUPP;
1230
1231         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1232                 ret = -ENODATA;
1233
1234         xis.inode_bh = xbs.inode_bh = di_bh;
1235         di = (struct ocfs2_dinode *)di_bh->b_data;
1236
1237         down_read(&oi->ip_xattr_sem);
1238         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1239                                     buffer_size, &xis);
1240         if (ret == -ENODATA && di->i_xattr_loc)
1241                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1242                                             buffer_size, &xbs);
1243         up_read(&oi->ip_xattr_sem);
1244
1245         return ret;
1246 }
1247
1248 /* ocfs2_xattr_get()
1249  *
1250  * Copy an extended attribute into the buffer provided.
1251  * Buffer is NULL to compute the size of buffer required.
1252  */
1253 static int ocfs2_xattr_get(struct inode *inode,
1254                            int name_index,
1255                            const char *name,
1256                            void *buffer,
1257                            size_t buffer_size)
1258 {
1259         int ret;
1260         struct buffer_head *di_bh = NULL;
1261
1262         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1263         if (ret < 0) {
1264                 mlog_errno(ret);
1265                 return ret;
1266         }
1267         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1268                                      name, buffer, buffer_size);
1269
1270         ocfs2_inode_unlock(inode, 0);
1271
1272         brelse(di_bh);
1273
1274         return ret;
1275 }
1276
1277 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1278                                            handle_t *handle,
1279                                            struct ocfs2_xattr_value_buf *vb,
1280                                            const void *value,
1281                                            int value_len)
1282 {
1283         int ret = 0, i, cp_len;
1284         u16 blocksize = inode->i_sb->s_blocksize;
1285         u32 p_cluster, num_clusters;
1286         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1287         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1288         u64 blkno;
1289         struct buffer_head *bh = NULL;
1290         unsigned int ext_flags;
1291         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1292
1293         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1294
1295         while (cpos < clusters) {
1296                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1297                                                &num_clusters, &xv->xr_list,
1298                                                &ext_flags);
1299                 if (ret) {
1300                         mlog_errno(ret);
1301                         goto out;
1302                 }
1303
1304                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1305
1306                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1307
1308                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1309                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1310                                                &bh, NULL);
1311                         if (ret) {
1312                                 mlog_errno(ret);
1313                                 goto out;
1314                         }
1315
1316                         ret = ocfs2_journal_access(handle,
1317                                                    INODE_CACHE(inode),
1318                                                    bh,
1319                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1320                         if (ret < 0) {
1321                                 mlog_errno(ret);
1322                                 goto out;
1323                         }
1324
1325                         cp_len = value_len > blocksize ? blocksize : value_len;
1326                         memcpy(bh->b_data, value, cp_len);
1327                         value_len -= cp_len;
1328                         value += cp_len;
1329                         if (cp_len < blocksize)
1330                                 memset(bh->b_data + cp_len, 0,
1331                                        blocksize - cp_len);
1332
1333                         ret = ocfs2_journal_dirty(handle, bh);
1334                         if (ret < 0) {
1335                                 mlog_errno(ret);
1336                                 goto out;
1337                         }
1338                         brelse(bh);
1339                         bh = NULL;
1340
1341                         /*
1342                          * XXX: do we need to empty all the following
1343                          * blocks in this cluster?
1344                          */
1345                         if (!value_len)
1346                                 break;
1347                 }
1348                 cpos += num_clusters;
1349         }
1350 out:
1351         brelse(bh);
1352
1353         return ret;
1354 }
1355
1356 static int ocfs2_xattr_cleanup(struct inode *inode,
1357                                handle_t *handle,
1358                                struct ocfs2_xattr_info *xi,
1359                                struct ocfs2_xattr_search *xs,
1360                                struct ocfs2_xattr_value_buf *vb,
1361                                size_t offs)
1362 {
1363         int ret = 0;
1364         size_t name_len = strlen(xi->name);
1365         void *val = xs->base + offs;
1366         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1367
1368         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1369                             OCFS2_JOURNAL_ACCESS_WRITE);
1370         if (ret) {
1371                 mlog_errno(ret);
1372                 goto out;
1373         }
1374         /* Decrease xattr count */
1375         le16_add_cpu(&xs->header->xh_count, -1);
1376         /* Remove the xattr entry and tree root which has already be set*/
1377         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1378         memset(val, 0, size);
1379
1380         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1381         if (ret < 0)
1382                 mlog_errno(ret);
1383 out:
1384         return ret;
1385 }
1386
1387 static int ocfs2_xattr_update_entry(struct inode *inode,
1388                                     handle_t *handle,
1389                                     struct ocfs2_xattr_info *xi,
1390                                     struct ocfs2_xattr_search *xs,
1391                                     struct ocfs2_xattr_value_buf *vb,
1392                                     size_t offs)
1393 {
1394         int ret;
1395
1396         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1397                             OCFS2_JOURNAL_ACCESS_WRITE);
1398         if (ret) {
1399                 mlog_errno(ret);
1400                 goto out;
1401         }
1402
1403         xs->here->xe_name_offset = cpu_to_le16(offs);
1404         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1405         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1406                 ocfs2_xattr_set_local(xs->here, 1);
1407         else
1408                 ocfs2_xattr_set_local(xs->here, 0);
1409         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1410
1411         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1412         if (ret < 0)
1413                 mlog_errno(ret);
1414 out:
1415         return ret;
1416 }
1417
1418 /*
1419  * ocfs2_xattr_set_value_outside()
1420  *
1421  * Set large size value in B tree.
1422  */
1423 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1424                                          struct ocfs2_xattr_info *xi,
1425                                          struct ocfs2_xattr_search *xs,
1426                                          struct ocfs2_xattr_set_ctxt *ctxt,
1427                                          struct ocfs2_xattr_value_buf *vb,
1428                                          size_t offs)
1429 {
1430         size_t name_len = strlen(xi->name);
1431         void *val = xs->base + offs;
1432         struct ocfs2_xattr_value_root *xv = NULL;
1433         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1434         int ret = 0;
1435
1436         memset(val, 0, size);
1437         memcpy(val, xi->name, name_len);
1438         xv = (struct ocfs2_xattr_value_root *)
1439                 (val + OCFS2_XATTR_SIZE(name_len));
1440         xv->xr_clusters = 0;
1441         xv->xr_last_eb_blk = 0;
1442         xv->xr_list.l_tree_depth = 0;
1443         xv->xr_list.l_count = cpu_to_le16(1);
1444         xv->xr_list.l_next_free_rec = 0;
1445         vb->vb_xv = xv;
1446
1447         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1448         if (ret < 0) {
1449                 mlog_errno(ret);
1450                 return ret;
1451         }
1452         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1453         if (ret < 0) {
1454                 mlog_errno(ret);
1455                 return ret;
1456         }
1457         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb,
1458                                               xi->value, xi->value_len);
1459         if (ret < 0)
1460                 mlog_errno(ret);
1461
1462         return ret;
1463 }
1464
1465 /*
1466  * Wipe the name+value pair and allow the storage to reclaim it.  This
1467  * must be followed by either removal of the entry or a call to
1468  * ocfs2_xa_add_namevalue().
1469  */
1470 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1471 {
1472         loc->xl_ops->xlo_wipe_namevalue(loc);
1473 }
1474
1475 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1476                                            int offset)
1477 {
1478         BUG_ON(offset >= loc->xl_size);
1479         return (char *)loc->xl_header + offset;
1480 }
1481
1482 /*
1483  * Block storage for xattrs keeps the name+value pairs compacted.  When
1484  * we remove one, we have to shift any that preceded it towards the end.
1485  */
1486 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1487 {
1488         int i, offset;
1489         int namevalue_offset, first_namevalue_offset, namevalue_size;
1490         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1491         struct ocfs2_xattr_header *xh = loc->xl_header;
1492         u64 value_size = le64_to_cpu(entry->xe_value_size);
1493         int count = le16_to_cpu(xh->xh_count);
1494
1495         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1496         namevalue_size = OCFS2_XATTR_SIZE(entry->xe_name_len);
1497         if (value_size > OCFS2_XATTR_INLINE_SIZE)
1498                 namevalue_size += OCFS2_XATTR_ROOT_SIZE;
1499         else
1500                 namevalue_size += OCFS2_XATTR_SIZE(value_size);
1501
1502         for (i = 0, first_namevalue_offset = loc->xl_size;
1503              i < count; i++) {
1504                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1505                 if (offset < first_namevalue_offset)
1506                         first_namevalue_offset = offset;
1507         }
1508
1509         /* Shift the name+value pairs */
1510         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1511                 (char *)xh + first_namevalue_offset,
1512                 namevalue_offset - first_namevalue_offset);
1513         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1514
1515         /* Now tell xh->xh_entries about it */
1516         for (i = 0; i < count; i++) {
1517                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1518                 if (offset < namevalue_offset)
1519                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1520                                      namevalue_size);
1521         }
1522
1523         /*
1524          * Note that we don't update xh_free_start or xh_name_value_len
1525          * because they're not used in block-stored xattrs.
1526          */
1527 }
1528
1529 /*
1530  * Operations for xattrs stored in blocks.  This includes inline inode
1531  * storage and unindexed ocfs2_xattr_blocks.
1532  */
1533 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1534         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1535         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1536 };
1537
1538 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1539                                             int offset)
1540 {
1541         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1542         int block, block_offset;
1543
1544         BUG_ON(offset >= OCFS2_XATTR_BUCKET_SIZE);
1545
1546         /* The header is at the front of the bucket */
1547         block = offset >> bucket->bu_inode->i_sb->s_blocksize_bits;
1548         block_offset = offset % bucket->bu_inode->i_sb->s_blocksize;
1549
1550         return bucket_block(bucket, block) + block_offset;
1551 }
1552
1553 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1554 {
1555         int namevalue_size;
1556         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1557         u64 value_size = le64_to_cpu(entry->xe_value_size);
1558
1559         namevalue_size = OCFS2_XATTR_SIZE(entry->xe_name_len);
1560         if (value_size > OCFS2_XATTR_INLINE_SIZE)
1561                 namevalue_size += OCFS2_XATTR_ROOT_SIZE;
1562         else
1563                 namevalue_size += OCFS2_XATTR_SIZE(value_size);
1564
1565         le16_add_cpu(&loc->xl_header->xh_name_value_len, -namevalue_size);
1566 }
1567
1568 /* Operations for xattrs stored in buckets. */
1569 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1570         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1571         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1572 };
1573
1574 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1575 {
1576         ocfs2_xa_wipe_namevalue(loc);
1577 }
1578
1579 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
1580                                      struct inode *inode,
1581                                      struct buffer_head *bh,
1582                                      struct ocfs2_xattr_entry *entry)
1583 {
1584         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
1585
1586         loc->xl_ops = &ocfs2_xa_block_loc_ops;
1587         loc->xl_storage = bh;
1588         loc->xl_entry = entry;
1589
1590         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1591                 loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
1592         else {
1593                 BUG_ON(entry);
1594                 loc->xl_size = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1595         }
1596         loc->xl_header =
1597                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
1598                                               loc->xl_size);
1599 }
1600
1601 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
1602                                           struct buffer_head *bh,
1603                                           struct ocfs2_xattr_entry *entry)
1604 {
1605         struct ocfs2_xattr_block *xb =
1606                 (struct ocfs2_xattr_block *)bh->b_data;
1607
1608         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
1609
1610         loc->xl_ops = &ocfs2_xa_block_loc_ops;
1611         loc->xl_storage = bh;
1612         loc->xl_header = &(xb->xb_attrs.xb_header);
1613         loc->xl_entry = entry;
1614         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
1615                                              xb_attrs.xb_header);
1616 }
1617
1618 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
1619                                            struct ocfs2_xattr_bucket *bucket,
1620                                            struct ocfs2_xattr_entry *entry)
1621 {
1622         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
1623         loc->xl_storage = bucket;
1624         loc->xl_header = bucket_xh(bucket);
1625         loc->xl_entry = entry;
1626         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
1627 }
1628
1629 /*
1630  * ocfs2_xattr_set_entry_local()
1631  *
1632  * Set, replace or remove extended attribute in local.
1633  */
1634 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1635                                         struct ocfs2_xattr_info *xi,
1636                                         struct ocfs2_xattr_search *xs,
1637                                         struct ocfs2_xattr_entry *last,
1638                                         size_t min_offs)
1639 {
1640         size_t name_len = strlen(xi->name);
1641         int i;
1642         struct ocfs2_xa_loc loc;
1643
1644         if (xs->xattr_bh == xs->inode_bh)
1645                 ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
1646                                          xs->not_found ? NULL : xs->here);
1647         else
1648                 ocfs2_init_xattr_block_xa_loc(&loc, xs->xattr_bh,
1649                                               xs->not_found ? NULL : xs->here);
1650         if (xi->value && xs->not_found) {
1651                 /* Insert the new xattr entry. */
1652                 le16_add_cpu(&xs->header->xh_count, 1);
1653                 ocfs2_xattr_set_type(last, xi->name_index);
1654                 ocfs2_xattr_set_local(last, 1);
1655                 last->xe_name_len = name_len;
1656         } else {
1657                 void *first_val;
1658                 void *val;
1659                 size_t offs, size;
1660
1661                 first_val = xs->base + min_offs;
1662                 offs = le16_to_cpu(xs->here->xe_name_offset);
1663                 val = xs->base + offs;
1664
1665                 if (le64_to_cpu(xs->here->xe_value_size) >
1666                     OCFS2_XATTR_INLINE_SIZE)
1667                         size = OCFS2_XATTR_SIZE(name_len) +
1668                                 OCFS2_XATTR_ROOT_SIZE;
1669                 else
1670                         size = OCFS2_XATTR_SIZE(name_len) +
1671                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1672
1673                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1674                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1675                         /* The old and the new value have the
1676                            same size. Just replace the value. */
1677                         ocfs2_xattr_set_local(xs->here, 1);
1678                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1679                         /* Clear value bytes. */
1680                         memset(val + OCFS2_XATTR_SIZE(name_len),
1681                                0,
1682                                OCFS2_XATTR_SIZE(xi->value_len));
1683                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1684                                xi->value,
1685                                xi->value_len);
1686                         return;
1687                 }
1688
1689                 /* Remove the old name+value. */
1690                 ocfs2_xa_wipe_namevalue(&loc);
1691                 xs->here->xe_name_hash = 0;
1692                 xs->here->xe_name_offset = 0;
1693                 ocfs2_xattr_set_local(xs->here, 1);
1694                 xs->here->xe_value_size = 0;
1695
1696                 min_offs += size;
1697
1698                 if (!xi->value) {
1699                         /* Remove the old entry. */
1700                         i = le16_to_cpu(xs->header->xh_count) - 1;
1701                         last = &xs->header->xh_entries[i];
1702                         xs->header->xh_count = cpu_to_le16(i);
1703
1704                         memmove(xs->here, xs->here + 1,
1705                                 (void *)last - (void *)xs->here);
1706                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1707                 }
1708         }
1709         if (xi->value) {
1710                 /* Insert the new name+value. */
1711                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1712                                 OCFS2_XATTR_SIZE(xi->value_len);
1713                 void *val = xs->base + min_offs - size;
1714
1715                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1716                 memset(val, 0, size);
1717                 memcpy(val, xi->name, name_len);
1718                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1719                        xi->value,
1720                        xi->value_len);
1721                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1722                 ocfs2_xattr_set_local(xs->here, 1);
1723                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1724         }
1725
1726         return;
1727 }
1728
1729 /*
1730  * ocfs2_xattr_set_entry()
1731  *
1732  * Set extended attribute entry into inode or block.
1733  *
1734  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1735  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1736  * then set value in B tree with set_value_outside().
1737  */
1738 static int ocfs2_xattr_set_entry(struct inode *inode,
1739                                  struct ocfs2_xattr_info *xi,
1740                                  struct ocfs2_xattr_search *xs,
1741                                  struct ocfs2_xattr_set_ctxt *ctxt,
1742                                  int flag)
1743 {
1744         struct ocfs2_xattr_entry *last;
1745         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1746         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1747         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1748         size_t size_l = 0;
1749         handle_t *handle = ctxt->handle;
1750         int free, i, ret;
1751         struct ocfs2_xattr_info xi_l = {
1752                 .name_index = xi->name_index,
1753                 .name = xi->name,
1754                 .value = xi->value,
1755                 .value_len = xi->value_len,
1756         };
1757         struct ocfs2_xattr_value_buf vb = {
1758                 .vb_bh = xs->xattr_bh,
1759                 .vb_access = ocfs2_journal_access_di,
1760         };
1761
1762         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1763                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1764                 vb.vb_access = ocfs2_journal_access_xb;
1765         } else
1766                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1767
1768         /* Compute min_offs, last and free space. */
1769         last = xs->header->xh_entries;
1770
1771         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1772                 size_t offs = le16_to_cpu(last->xe_name_offset);
1773                 if (offs < min_offs)
1774                         min_offs = offs;
1775                 last += 1;
1776         }
1777
1778         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1779         if (free < 0)
1780                 return -EIO;
1781
1782         if (!xs->not_found) {
1783                 size_t size = 0;
1784                 if (ocfs2_xattr_is_local(xs->here))
1785                         size = OCFS2_XATTR_SIZE(name_len) +
1786                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1787                 else
1788                         size = OCFS2_XATTR_SIZE(name_len) +
1789                                 OCFS2_XATTR_ROOT_SIZE;
1790                 free += (size + sizeof(struct ocfs2_xattr_entry));
1791         }
1792         /* Check free space in inode or block */
1793         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1794                 if (free < sizeof(struct ocfs2_xattr_entry) +
1795                            OCFS2_XATTR_SIZE(name_len) +
1796                            OCFS2_XATTR_ROOT_SIZE) {
1797                         ret = -ENOSPC;
1798                         goto out;
1799                 }
1800                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1801                 xi_l.value = (void *)&def_xv;
1802                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1803         } else if (xi->value) {
1804                 if (free < sizeof(struct ocfs2_xattr_entry) +
1805                            OCFS2_XATTR_SIZE(name_len) +
1806                            OCFS2_XATTR_SIZE(xi->value_len)) {
1807                         ret = -ENOSPC;
1808                         goto out;
1809                 }
1810         }
1811
1812         if (!xs->not_found) {
1813                 /* For existing extended attribute */
1814                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1815                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1816                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1817                 void *val = xs->base + offs;
1818
1819                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1820                         /* Replace existing local xattr with tree root */
1821                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1822                                                             ctxt, &vb, offs);
1823                         if (ret < 0)
1824                                 mlog_errno(ret);
1825                         goto out;
1826                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1827                         /* For existing xattr which has value outside */
1828                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1829                                 (val + OCFS2_XATTR_SIZE(name_len));
1830
1831                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1832                                 /*
1833                                  * If new value need set outside also,
1834                                  * first truncate old value to new value,
1835                                  * then set new value with set_value_outside().
1836                                  */
1837                                 ret = ocfs2_xattr_value_truncate(inode,
1838                                                                  &vb,
1839                                                                  xi->value_len,
1840                                                                  ctxt);
1841                                 if (ret < 0) {
1842                                         mlog_errno(ret);
1843                                         goto out;
1844                                 }
1845
1846                                 ret = ocfs2_xattr_update_entry(inode,
1847                                                                handle,
1848                                                                xi,
1849                                                                xs,
1850                                                                &vb,
1851                                                                offs);
1852                                 if (ret < 0) {
1853                                         mlog_errno(ret);
1854                                         goto out;
1855                                 }
1856
1857                                 ret = __ocfs2_xattr_set_value_outside(inode,
1858                                                                 handle,
1859                                                                 &vb,
1860                                                                 xi->value,
1861                                                                 xi->value_len);
1862                                 if (ret < 0)
1863                                         mlog_errno(ret);
1864                                 goto out;
1865                         } else {
1866                                 /*
1867                                  * If new value need set in local,
1868                                  * just trucate old value to zero.
1869                                  */
1870                                  ret = ocfs2_xattr_value_truncate(inode,
1871                                                                   &vb,
1872                                                                   0,
1873                                                                   ctxt);
1874                                 if (ret < 0)
1875                                         mlog_errno(ret);
1876                         }
1877                 }
1878         }
1879
1880         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1881                                       OCFS2_JOURNAL_ACCESS_WRITE);
1882         if (ret) {
1883                 mlog_errno(ret);
1884                 goto out;
1885         }
1886
1887         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1888                 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1889                                    OCFS2_JOURNAL_ACCESS_WRITE);
1890                 if (ret) {
1891                         mlog_errno(ret);
1892                         goto out;
1893                 }
1894         }
1895
1896         /*
1897          * Set value in local, include set tree root in local.
1898          * This is the first step for value size >INLINE_SIZE.
1899          */
1900         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1901
1902         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1903                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1904                 if (ret < 0) {
1905                         mlog_errno(ret);
1906                         goto out;
1907                 }
1908         }
1909
1910         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1911             (flag & OCFS2_INLINE_XATTR_FL)) {
1912                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1913                 unsigned int xattrsize = osb->s_xattr_inline_size;
1914
1915                 /*
1916                  * Adjust extent record count or inline data size
1917                  * to reserve space for extended attribute.
1918                  */
1919                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1920                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1921                         le16_add_cpu(&idata->id_count, -xattrsize);
1922                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1923                         struct ocfs2_extent_list *el = &di->id2.i_list;
1924                         le16_add_cpu(&el->l_count, -(xattrsize /
1925                                         sizeof(struct ocfs2_extent_rec)));
1926                 }
1927                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1928         }
1929         /* Update xattr flag */
1930         spin_lock(&oi->ip_lock);
1931         oi->ip_dyn_features |= flag;
1932         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1933         spin_unlock(&oi->ip_lock);
1934
1935         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1936         if (ret < 0)
1937                 mlog_errno(ret);
1938
1939         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1940                 /*
1941                  * Set value outside in B tree.
1942                  * This is the second step for value size > INLINE_SIZE.
1943                  */
1944                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1945                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1946                                                     &vb, offs);
1947                 if (ret < 0) {
1948                         int ret2;
1949
1950                         mlog_errno(ret);
1951                         /*
1952                          * If set value outside failed, we have to clean
1953                          * the junk tree root we have already set in local.
1954                          */
1955                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1956                                                    xi, xs, &vb, offs);
1957                         if (ret2 < 0)
1958                                 mlog_errno(ret2);
1959                 }
1960         }
1961 out:
1962         return ret;
1963 }
1964
1965 /*
1966  * In xattr remove, if it is stored outside and refcounted, we may have
1967  * the chance to split the refcount tree. So need the allocators.
1968  */
1969 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
1970                                         struct ocfs2_xattr_value_root *xv,
1971                                         struct ocfs2_caching_info *ref_ci,
1972                                         struct buffer_head *ref_root_bh,
1973                                         struct ocfs2_alloc_context **meta_ac,
1974                                         int *ref_credits)
1975 {
1976         int ret, meta_add = 0;
1977         u32 p_cluster, num_clusters;
1978         unsigned int ext_flags;
1979
1980         *ref_credits = 0;
1981         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
1982                                        &num_clusters,
1983                                        &xv->xr_list,
1984                                        &ext_flags);
1985         if (ret) {
1986                 mlog_errno(ret);
1987                 goto out;
1988         }
1989
1990         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
1991                 goto out;
1992
1993         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
1994                                                  ref_root_bh, xv,
1995                                                  &meta_add, ref_credits);
1996         if (ret) {
1997                 mlog_errno(ret);
1998                 goto out;
1999         }
2000
2001         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2002                                                 meta_add, meta_ac);
2003         if (ret)
2004                 mlog_errno(ret);
2005
2006 out:
2007         return ret;
2008 }
2009
2010 static int ocfs2_remove_value_outside(struct inode*inode,
2011                                       struct ocfs2_xattr_value_buf *vb,
2012                                       struct ocfs2_xattr_header *header,
2013                                       struct ocfs2_caching_info *ref_ci,
2014                                       struct buffer_head *ref_root_bh)
2015 {
2016         int ret = 0, i, ref_credits;
2017         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2018         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2019         void *val;
2020
2021         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2022
2023         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2024                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2025
2026                 if (ocfs2_xattr_is_local(entry))
2027                         continue;
2028
2029                 val = (void *)header +
2030                         le16_to_cpu(entry->xe_name_offset);
2031                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2032                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2033
2034                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2035                                                          ref_ci, ref_root_bh,
2036                                                          &ctxt.meta_ac,
2037                                                          &ref_credits);
2038
2039                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2040                                         ocfs2_remove_extent_credits(osb->sb));
2041                 if (IS_ERR(ctxt.handle)) {
2042                         ret = PTR_ERR(ctxt.handle);
2043                         mlog_errno(ret);
2044                         break;
2045                 }
2046
2047                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2048                 if (ret < 0) {
2049                         mlog_errno(ret);
2050                         break;
2051                 }
2052
2053                 ocfs2_commit_trans(osb, ctxt.handle);
2054                 if (ctxt.meta_ac) {
2055                         ocfs2_free_alloc_context(ctxt.meta_ac);
2056                         ctxt.meta_ac = NULL;
2057                 }
2058         }
2059
2060         if (ctxt.meta_ac)
2061                 ocfs2_free_alloc_context(ctxt.meta_ac);
2062         ocfs2_schedule_truncate_log_flush(osb, 1);
2063         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2064         return ret;
2065 }
2066
2067 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2068                                     struct buffer_head *di_bh,
2069                                     struct ocfs2_caching_info *ref_ci,
2070                                     struct buffer_head *ref_root_bh)
2071 {
2072
2073         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2074         struct ocfs2_xattr_header *header;
2075         int ret;
2076         struct ocfs2_xattr_value_buf vb = {
2077                 .vb_bh = di_bh,
2078                 .vb_access = ocfs2_journal_access_di,
2079         };
2080
2081         header = (struct ocfs2_xattr_header *)
2082                  ((void *)di + inode->i_sb->s_blocksize -
2083                  le16_to_cpu(di->i_xattr_inline_size));
2084
2085         ret = ocfs2_remove_value_outside(inode, &vb, header,
2086                                          ref_ci, ref_root_bh);
2087
2088         return ret;
2089 }
2090
2091 struct ocfs2_rm_xattr_bucket_para {
2092         struct ocfs2_caching_info *ref_ci;
2093         struct buffer_head *ref_root_bh;
2094 };
2095
2096 static int ocfs2_xattr_block_remove(struct inode *inode,
2097                                     struct buffer_head *blk_bh,
2098                                     struct ocfs2_caching_info *ref_ci,
2099                                     struct buffer_head *ref_root_bh)
2100 {
2101         struct ocfs2_xattr_block *xb;
2102         int ret = 0;
2103         struct ocfs2_xattr_value_buf vb = {
2104                 .vb_bh = blk_bh,
2105                 .vb_access = ocfs2_journal_access_xb,
2106         };
2107         struct ocfs2_rm_xattr_bucket_para args = {
2108                 .ref_ci = ref_ci,
2109                 .ref_root_bh = ref_root_bh,
2110         };
2111
2112         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2113         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2114                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2115                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2116                                                  ref_ci, ref_root_bh);
2117         } else
2118                 ret = ocfs2_iterate_xattr_index_block(inode,
2119                                                 blk_bh,
2120                                                 ocfs2_rm_xattr_cluster,
2121                                                 &args);
2122
2123         return ret;
2124 }
2125
2126 static int ocfs2_xattr_free_block(struct inode *inode,
2127                                   u64 block,
2128                                   struct ocfs2_caching_info *ref_ci,
2129                                   struct buffer_head *ref_root_bh)
2130 {
2131         struct inode *xb_alloc_inode;
2132         struct buffer_head *xb_alloc_bh = NULL;
2133         struct buffer_head *blk_bh = NULL;
2134         struct ocfs2_xattr_block *xb;
2135         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2136         handle_t *handle;
2137         int ret = 0;
2138         u64 blk, bg_blkno;
2139         u16 bit;
2140
2141         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2142         if (ret < 0) {
2143                 mlog_errno(ret);
2144                 goto out;
2145         }
2146
2147         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2148         if (ret < 0) {
2149                 mlog_errno(ret);
2150                 goto out;
2151         }
2152
2153         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2154         blk = le64_to_cpu(xb->xb_blkno);
2155         bit = le16_to_cpu(xb->xb_suballoc_bit);
2156         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2157
2158         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2159                                 EXTENT_ALLOC_SYSTEM_INODE,
2160                                 le16_to_cpu(xb->xb_suballoc_slot));
2161         if (!xb_alloc_inode) {
2162                 ret = -ENOMEM;
2163                 mlog_errno(ret);
2164                 goto out;
2165         }
2166         mutex_lock(&xb_alloc_inode->i_mutex);
2167
2168         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2169         if (ret < 0) {
2170                 mlog_errno(ret);
2171                 goto out_mutex;
2172         }
2173
2174         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2175         if (IS_ERR(handle)) {
2176                 ret = PTR_ERR(handle);
2177                 mlog_errno(ret);
2178                 goto out_unlock;
2179         }
2180
2181         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2182                                        bit, bg_blkno, 1);
2183         if (ret < 0)
2184                 mlog_errno(ret);
2185
2186         ocfs2_commit_trans(osb, handle);
2187 out_unlock:
2188         ocfs2_inode_unlock(xb_alloc_inode, 1);
2189         brelse(xb_alloc_bh);
2190 out_mutex:
2191         mutex_unlock(&xb_alloc_inode->i_mutex);
2192         iput(xb_alloc_inode);
2193 out:
2194         brelse(blk_bh);
2195         return ret;
2196 }
2197
2198 /*
2199  * ocfs2_xattr_remove()
2200  *
2201  * Free extended attribute resources associated with this inode.
2202  */
2203 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2204 {
2205         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2206         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2207         struct ocfs2_refcount_tree *ref_tree = NULL;
2208         struct buffer_head *ref_root_bh = NULL;
2209         struct ocfs2_caching_info *ref_ci = NULL;
2210         handle_t *handle;
2211         int ret;
2212
2213         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2214                 return 0;
2215
2216         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2217                 return 0;
2218
2219         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2220                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2221                                                le64_to_cpu(di->i_refcount_loc),
2222                                                1, &ref_tree, &ref_root_bh);
2223                 if (ret) {
2224                         mlog_errno(ret);
2225                         goto out;
2226                 }
2227                 ref_ci = &ref_tree->rf_ci;
2228
2229         }
2230
2231         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2232                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2233                                                ref_ci, ref_root_bh);
2234                 if (ret < 0) {
2235                         mlog_errno(ret);
2236                         goto out;
2237                 }
2238         }
2239
2240         if (di->i_xattr_loc) {
2241                 ret = ocfs2_xattr_free_block(inode,
2242                                              le64_to_cpu(di->i_xattr_loc),
2243                                              ref_ci, ref_root_bh);
2244                 if (ret < 0) {
2245                         mlog_errno(ret);
2246                         goto out;
2247                 }
2248         }
2249
2250         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2251                                    OCFS2_INODE_UPDATE_CREDITS);
2252         if (IS_ERR(handle)) {
2253                 ret = PTR_ERR(handle);
2254                 mlog_errno(ret);
2255                 goto out;
2256         }
2257         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2258                                       OCFS2_JOURNAL_ACCESS_WRITE);
2259         if (ret) {
2260                 mlog_errno(ret);
2261                 goto out_commit;
2262         }
2263
2264         di->i_xattr_loc = 0;
2265
2266         spin_lock(&oi->ip_lock);
2267         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2268         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2269         spin_unlock(&oi->ip_lock);
2270
2271         ret = ocfs2_journal_dirty(handle, di_bh);
2272         if (ret < 0)
2273                 mlog_errno(ret);
2274 out_commit:
2275         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2276 out:
2277         if (ref_tree)
2278                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2279         brelse(ref_root_bh);
2280         return ret;
2281 }
2282
2283 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2284                                         struct ocfs2_dinode *di)
2285 {
2286         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2287         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2288         int free;
2289
2290         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2291                 return 0;
2292
2293         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2294                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2295                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2296         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2297                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2298                         le64_to_cpu(di->i_size);
2299         } else {
2300                 struct ocfs2_extent_list *el = &di->id2.i_list;
2301                 free = (le16_to_cpu(el->l_count) -
2302                         le16_to_cpu(el->l_next_free_rec)) *
2303                         sizeof(struct ocfs2_extent_rec);
2304         }
2305         if (free >= xattrsize)
2306                 return 1;
2307
2308         return 0;
2309 }
2310
2311 /*
2312  * ocfs2_xattr_ibody_find()
2313  *
2314  * Find extended attribute in inode block and
2315  * fill search info into struct ocfs2_xattr_search.
2316  */
2317 static int ocfs2_xattr_ibody_find(struct inode *inode,
2318                                   int name_index,
2319                                   const char *name,
2320                                   struct ocfs2_xattr_search *xs)
2321 {
2322         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2323         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2324         int ret;
2325         int has_space = 0;
2326
2327         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2328                 return 0;
2329
2330         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2331                 down_read(&oi->ip_alloc_sem);
2332                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2333                 up_read(&oi->ip_alloc_sem);
2334                 if (!has_space)
2335                         return 0;
2336         }
2337
2338         xs->xattr_bh = xs->inode_bh;
2339         xs->end = (void *)di + inode->i_sb->s_blocksize;
2340         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2341                 xs->header = (struct ocfs2_xattr_header *)
2342                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2343         else
2344                 xs->header = (struct ocfs2_xattr_header *)
2345                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2346         xs->base = (void *)xs->header;
2347         xs->here = xs->header->xh_entries;
2348
2349         /* Find the named attribute. */
2350         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2351                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2352                 if (ret && ret != -ENODATA)
2353                         return ret;
2354                 xs->not_found = ret;
2355         }
2356
2357         return 0;
2358 }
2359
2360 /*
2361  * ocfs2_xattr_ibody_set()
2362  *
2363  * Set, replace or remove an extended attribute into inode block.
2364  *
2365  */
2366 static int ocfs2_xattr_ibody_set(struct inode *inode,
2367                                  struct ocfs2_xattr_info *xi,
2368                                  struct ocfs2_xattr_search *xs,
2369                                  struct ocfs2_xattr_set_ctxt *ctxt)
2370 {
2371         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2372         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2373         int ret;
2374
2375         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2376                 return -ENOSPC;
2377
2378         down_write(&oi->ip_alloc_sem);
2379         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2380                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2381                         ret = -ENOSPC;
2382                         goto out;
2383                 }
2384         }
2385
2386         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2387                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2388 out:
2389         up_write(&oi->ip_alloc_sem);
2390
2391         return ret;
2392 }
2393
2394 /*
2395  * ocfs2_xattr_block_find()
2396  *
2397  * Find extended attribute in external block and
2398  * fill search info into struct ocfs2_xattr_search.
2399  */
2400 static int ocfs2_xattr_block_find(struct inode *inode,
2401                                   int name_index,
2402                                   const char *name,
2403                                   struct ocfs2_xattr_search *xs)
2404 {
2405         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2406         struct buffer_head *blk_bh = NULL;
2407         struct ocfs2_xattr_block *xb;
2408         int ret = 0;
2409
2410         if (!di->i_xattr_loc)
2411                 return ret;
2412
2413         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2414                                      &blk_bh);
2415         if (ret < 0) {
2416                 mlog_errno(ret);
2417                 return ret;
2418         }
2419
2420         xs->xattr_bh = blk_bh;
2421         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2422
2423         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2424                 xs->header = &xb->xb_attrs.xb_header;
2425                 xs->base = (void *)xs->header;
2426                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2427                 xs->here = xs->header->xh_entries;
2428
2429                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2430         } else
2431                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2432                                                    name_index,
2433                                                    name, xs);
2434
2435         if (ret && ret != -ENODATA) {
2436                 xs->xattr_bh = NULL;
2437                 goto cleanup;
2438         }
2439         xs->not_found = ret;
2440         return 0;
2441 cleanup:
2442         brelse(blk_bh);
2443
2444         return ret;
2445 }
2446
2447 static int ocfs2_create_xattr_block(handle_t *handle,
2448                                     struct inode *inode,
2449                                     struct buffer_head *inode_bh,
2450                                     struct ocfs2_alloc_context *meta_ac,
2451                                     struct buffer_head **ret_bh,
2452                                     int indexed)
2453 {
2454         int ret;
2455         u16 suballoc_bit_start;
2456         u32 num_got;
2457         u64 first_blkno;
2458         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2459         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2460         struct buffer_head *new_bh = NULL;
2461         struct ocfs2_xattr_block *xblk;
2462
2463         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), inode_bh,
2464                                       OCFS2_JOURNAL_ACCESS_CREATE);
2465         if (ret < 0) {
2466                 mlog_errno(ret);
2467                 goto end;
2468         }
2469
2470         ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
2471                                    &suballoc_bit_start, &num_got,
2472                                    &first_blkno);
2473         if (ret < 0) {
2474                 mlog_errno(ret);
2475                 goto end;
2476         }
2477
2478         new_bh = sb_getblk(inode->i_sb, first_blkno);
2479         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2480
2481         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2482                                       new_bh,
2483                                       OCFS2_JOURNAL_ACCESS_CREATE);
2484         if (ret < 0) {
2485                 mlog_errno(ret);
2486                 goto end;
2487         }
2488
2489         /* Initialize ocfs2_xattr_block */
2490         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2491         memset(xblk, 0, inode->i_sb->s_blocksize);
2492         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2493         xblk->xb_suballoc_slot = cpu_to_le16(meta_ac->ac_alloc_slot);
2494         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2495         xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2496         xblk->xb_blkno = cpu_to_le64(first_blkno);
2497
2498         if (indexed) {
2499                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2500                 xr->xt_clusters = cpu_to_le32(1);
2501                 xr->xt_last_eb_blk = 0;
2502                 xr->xt_list.l_tree_depth = 0;
2503                 xr->xt_list.l_count = cpu_to_le16(
2504                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2505                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2506                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2507         }
2508
2509         ret = ocfs2_journal_dirty(handle, new_bh);
2510         if (ret < 0) {
2511                 mlog_errno(ret);
2512                 goto end;
2513         }
2514         di->i_xattr_loc = cpu_to_le64(first_blkno);
2515         ocfs2_journal_dirty(handle, inode_bh);
2516
2517         *ret_bh = new_bh;
2518         new_bh = NULL;
2519
2520 end:
2521         brelse(new_bh);
2522         return ret;
2523 }
2524
2525 /*
2526  * ocfs2_xattr_block_set()
2527  *
2528  * Set, replace or remove an extended attribute into external block.
2529  *
2530  */
2531 static int ocfs2_xattr_block_set(struct inode *inode,
2532                                  struct ocfs2_xattr_info *xi,
2533                                  struct ocfs2_xattr_search *xs,
2534                                  struct ocfs2_xattr_set_ctxt *ctxt)
2535 {
2536         struct buffer_head *new_bh = NULL;
2537         handle_t *handle = ctxt->handle;
2538         struct ocfs2_xattr_block *xblk = NULL;
2539         int ret;
2540
2541         if (!xs->xattr_bh) {
2542                 ret = ocfs2_create_xattr_block(handle, inode, xs->inode_bh,
2543                                                ctxt->meta_ac, &new_bh, 0);
2544                 if (ret) {
2545                         mlog_errno(ret);
2546                         goto end;
2547                 }
2548
2549                 xs->xattr_bh = new_bh;
2550                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2551                 xs->header = &xblk->xb_attrs.xb_header;
2552                 xs->base = (void *)xs->header;
2553                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2554                 xs->here = xs->header->xh_entries;
2555         } else
2556                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2557
2558         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2559                 /* Set extended attribute into external block */
2560                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2561                                             OCFS2_HAS_XATTR_FL);
2562                 if (!ret || ret != -ENOSPC)
2563                         goto end;
2564
2565                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2566                 if (ret)
2567                         goto end;
2568         }
2569
2570         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2571
2572 end:
2573
2574         return ret;
2575 }
2576
2577 /* Check whether the new xattr can be inserted into the inode. */
2578 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2579                                        struct ocfs2_xattr_info *xi,
2580                                        struct ocfs2_xattr_search *xs)
2581 {
2582         u64 value_size;
2583         struct ocfs2_xattr_entry *last;
2584         int free, i;
2585         size_t min_offs = xs->end - xs->base;
2586
2587         if (!xs->header)
2588                 return 0;
2589
2590         last = xs->header->xh_entries;
2591
2592         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2593                 size_t offs = le16_to_cpu(last->xe_name_offset);
2594                 if (offs < min_offs)
2595                         min_offs = offs;
2596                 last += 1;
2597         }
2598
2599         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2600         if (free < 0)
2601                 return 0;
2602
2603         BUG_ON(!xs->not_found);
2604
2605         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2606                 value_size = OCFS2_XATTR_ROOT_SIZE;
2607         else
2608                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2609
2610         if (free >= sizeof(struct ocfs2_xattr_entry) +
2611                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2612                 return 1;
2613
2614         return 0;
2615 }
2616
2617 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2618                                      struct ocfs2_dinode *di,
2619                                      struct ocfs2_xattr_info *xi,
2620                                      struct ocfs2_xattr_search *xis,
2621                                      struct ocfs2_xattr_search *xbs,
2622                                      int *clusters_need,
2623                                      int *meta_need,
2624                                      int *credits_need)
2625 {
2626         int ret = 0, old_in_xb = 0;
2627         int clusters_add = 0, meta_add = 0, credits = 0;
2628         struct buffer_head *bh = NULL;
2629         struct ocfs2_xattr_block *xb = NULL;
2630         struct ocfs2_xattr_entry *xe = NULL;
2631         struct ocfs2_xattr_value_root *xv = NULL;
2632         char *base = NULL;
2633         int name_offset, name_len = 0;
2634         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2635                                                     xi->value_len);
2636         u64 value_size;
2637
2638         /*
2639          * Calculate the clusters we need to write.
2640          * No matter whether we replace an old one or add a new one,
2641          * we need this for writing.
2642          */
2643         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2644                 credits += new_clusters *
2645                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2646
2647         if (xis->not_found && xbs->not_found) {
2648                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2649
2650                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2651                         clusters_add += new_clusters;
2652                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2653                                                         &def_xv.xv.xr_list,
2654                                                         new_clusters);
2655                 }
2656
2657                 goto meta_guess;
2658         }
2659
2660         if (!xis->not_found) {
2661                 xe = xis->here;
2662                 name_offset = le16_to_cpu(xe->xe_name_offset);
2663                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2664                 base = xis->base;
2665                 credits += OCFS2_INODE_UPDATE_CREDITS;
2666         } else {
2667                 int i, block_off = 0;
2668                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2669                 xe = xbs->here;
2670                 name_offset = le16_to_cpu(xe->xe_name_offset);
2671                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2672                 i = xbs->here - xbs->header->xh_entries;
2673                 old_in_xb = 1;
2674
2675                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2676                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
2677                                                         bucket_xh(xbs->bucket),
2678                                                         i, &block_off,
2679                                                         &name_offset);
2680                         base = bucket_block(xbs->bucket, block_off);
2681                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2682                 } else {
2683                         base = xbs->base;
2684                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2685                 }
2686         }
2687
2688         /*
2689          * delete a xattr doesn't need metadata and cluster allocation.
2690          * so just calculate the credits and return.
2691          *
2692          * The credits for removing the value tree will be extended
2693          * by ocfs2_remove_extent itself.
2694          */
2695         if (!xi->value) {
2696                 if (!ocfs2_xattr_is_local(xe))
2697                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2698
2699                 goto out;
2700         }
2701
2702         /* do cluster allocation guess first. */
2703         value_size = le64_to_cpu(xe->xe_value_size);
2704
2705         if (old_in_xb) {
2706                 /*
2707                  * In xattr set, we always try to set the xe in inode first,
2708                  * so if it can be inserted into inode successfully, the old
2709                  * one will be removed from the xattr block, and this xattr
2710                  * will be inserted into inode as a new xattr in inode.
2711                  */
2712                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2713                         clusters_add += new_clusters;
2714                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2715                                     OCFS2_INODE_UPDATE_CREDITS;
2716                         if (!ocfs2_xattr_is_local(xe))
2717                                 credits += ocfs2_calc_extend_credits(
2718                                                         inode->i_sb,
2719                                                         &def_xv.xv.xr_list,
2720                                                         new_clusters);
2721                         goto out;
2722                 }
2723         }
2724
2725         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2726                 /* the new values will be stored outside. */
2727                 u32 old_clusters = 0;
2728
2729                 if (!ocfs2_xattr_is_local(xe)) {
2730                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2731                                                                  value_size);
2732                         xv = (struct ocfs2_xattr_value_root *)
2733                              (base + name_offset + name_len);
2734                         value_size = OCFS2_XATTR_ROOT_SIZE;
2735                 } else
2736                         xv = &def_xv.xv;
2737
2738                 if (old_clusters >= new_clusters) {
2739                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2740                         goto out;
2741                 } else {
2742                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2743                         clusters_add += new_clusters - old_clusters;
2744                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2745                                                              &xv->xr_list,
2746                                                              new_clusters -
2747                                                              old_clusters);
2748                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2749                                 goto out;
2750                 }
2751         } else {
2752                 /*
2753                  * Now the new value will be stored inside. So if the new
2754                  * value is smaller than the size of value root or the old
2755                  * value, we don't need any allocation, otherwise we have
2756                  * to guess metadata allocation.
2757                  */
2758                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2759                     (!ocfs2_xattr_is_local(xe) &&
2760                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2761                         goto out;
2762         }
2763
2764 meta_guess:
2765         /* calculate metadata allocation. */
2766         if (di->i_xattr_loc) {
2767                 if (!xbs->xattr_bh) {
2768                         ret = ocfs2_read_xattr_block(inode,
2769                                                      le64_to_cpu(di->i_xattr_loc),
2770                                                      &bh);
2771                         if (ret) {
2772                                 mlog_errno(ret);
2773                                 goto out;
2774                         }
2775
2776                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2777                 } else
2778                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2779
2780                 /*
2781                  * If there is already an xattr tree, good, we can calculate
2782                  * like other b-trees. Otherwise we may have the chance of
2783                  * create a tree, the credit calculation is borrowed from
2784                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2785                  * new tree will be cluster based, so no meta is needed.
2786                  */
2787                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2788                         struct ocfs2_extent_list *el =
2789                                  &xb->xb_attrs.xb_root.xt_list;
2790                         meta_add += ocfs2_extend_meta_needed(el);
2791                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2792                                                              el, 1);
2793                 } else
2794                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2795
2796                 /*
2797                  * This cluster will be used either for new bucket or for
2798                  * new xattr block.
2799                  * If the cluster size is the same as the bucket size, one
2800                  * more is needed since we may need to extend the bucket
2801                  * also.
2802                  */
2803                 clusters_add += 1;
2804                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2805                 if (OCFS2_XATTR_BUCKET_SIZE ==
2806                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2807                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2808                         clusters_add += 1;
2809                 }
2810         } else {
2811                 meta_add += 1;
2812                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2813         }
2814 out:
2815         if (clusters_need)
2816                 *clusters_need = clusters_add;
2817         if (meta_need)
2818                 *meta_need = meta_add;
2819         if (credits_need)
2820                 *credits_need = credits;
2821         brelse(bh);
2822         return ret;
2823 }
2824
2825 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2826                                      struct ocfs2_dinode *di,
2827                                      struct ocfs2_xattr_info *xi,
2828                                      struct ocfs2_xattr_search *xis,
2829                                      struct ocfs2_xattr_search *xbs,
2830                                      struct ocfs2_xattr_set_ctxt *ctxt,
2831                                      int extra_meta,
2832                                      int *credits)
2833 {
2834         int clusters_add, meta_add, ret;
2835         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2836
2837         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2838
2839         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2840
2841         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2842                                         &clusters_add, &meta_add, credits);
2843         if (ret) {
2844                 mlog_errno(ret);
2845                 return ret;
2846         }
2847
2848         meta_add += extra_meta;
2849         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2850              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2851
2852         if (meta_add) {
2853                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2854                                                         &ctxt->meta_ac);
2855                 if (ret) {
2856                         mlog_errno(ret);
2857                         goto out;
2858                 }
2859         }
2860
2861         if (clusters_add) {
2862                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2863                 if (ret)
2864                         mlog_errno(ret);
2865         }
2866 out:
2867         if (ret) {
2868                 if (ctxt->meta_ac) {
2869                         ocfs2_free_alloc_context(ctxt->meta_ac);
2870                         ctxt->meta_ac = NULL;
2871                 }
2872
2873                 /*
2874                  * We cannot have an error and a non null ctxt->data_ac.
2875                  */
2876         }
2877
2878         return ret;
2879 }
2880
2881 static int __ocfs2_xattr_set_handle(struct inode *inode,
2882                                     struct ocfs2_dinode *di,
2883                                     struct ocfs2_xattr_info *xi,
2884                                     struct ocfs2_xattr_search *xis,
2885                                     struct ocfs2_xattr_search *xbs,
2886                                     struct ocfs2_xattr_set_ctxt *ctxt)
2887 {
2888         int ret = 0, credits, old_found;
2889
2890         if (!xi->value) {
2891                 /* Remove existing extended attribute */
2892                 if (!xis->not_found)
2893                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2894                 else if (!xbs->not_found)
2895                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2896         } else {
2897                 /* We always try to set extended attribute into inode first*/
2898                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2899                 if (!ret && !xbs->not_found) {
2900                         /*
2901                          * If succeed and that extended attribute existing in
2902                          * external block, then we will remove it.
2903                          */
2904                         xi->value = NULL;
2905                         xi->value_len = 0;
2906
2907                         old_found = xis->not_found;
2908                         xis->not_found = -ENODATA;
2909                         ret = ocfs2_calc_xattr_set_need(inode,
2910                                                         di,
2911                                                         xi,
2912                                                         xis,
2913                                                         xbs,
2914                                                         NULL,
2915                                                         NULL,
2916                                                         &credits);
2917                         xis->not_found = old_found;
2918                         if (ret) {
2919                                 mlog_errno(ret);
2920                                 goto out;
2921                         }
2922
2923                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2924                                         ctxt->handle->h_buffer_credits);
2925                         if (ret) {
2926                                 mlog_errno(ret);
2927                                 goto out;
2928                         }
2929                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2930                 } else if (ret == -ENOSPC) {
2931                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2932                                 ret = ocfs2_xattr_block_find(inode,
2933                                                              xi->name_index,
2934                                                              xi->name, xbs);
2935                                 if (ret)
2936                                         goto out;
2937
2938                                 old_found = xis->not_found;
2939                                 xis->not_found = -ENODATA;
2940                                 ret = ocfs2_calc_xattr_set_need(inode,
2941                                                                 di,
2942                                                                 xi,
2943                                                                 xis,
2944                                                                 xbs,
2945                                                                 NULL,
2946                                                                 NULL,
2947                                                                 &credits);
2948                                 xis->not_found = old_found;
2949                                 if (ret) {
2950                                         mlog_errno(ret);
2951                                         goto out;
2952                                 }
2953
2954                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2955                                         ctxt->handle->h_buffer_credits);
2956                                 if (ret) {
2957                                         mlog_errno(ret);
2958                                         goto out;
2959                                 }
2960                         }
2961                         /*
2962                          * If no space in inode, we will set extended attribute
2963                          * into external block.
2964                          */
2965                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2966                         if (ret)
2967                                 goto out;
2968                         if (!xis->not_found) {
2969                                 /*
2970                                  * If succeed and that extended attribute
2971                                  * existing in inode, we will remove it.
2972                                  */
2973                                 xi->value = NULL;
2974                                 xi->value_len = 0;
2975                                 xbs->not_found = -ENODATA;
2976                                 ret = ocfs2_calc_xattr_set_need(inode,
2977                                                                 di,
2978                                                                 xi,
2979                                                                 xis,
2980                                                                 xbs,
2981                                                                 NULL,
2982                                                                 NULL,
2983                                                                 &credits);
2984                                 if (ret) {
2985                                         mlog_errno(ret);
2986                                         goto out;
2987                                 }
2988
2989                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2990                                                 ctxt->handle->h_buffer_credits);
2991                                 if (ret) {
2992                                         mlog_errno(ret);
2993                                         goto out;
2994                                 }
2995                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2996                                                             xis, ctxt);
2997                         }
2998                 }
2999         }
3000
3001         if (!ret) {
3002                 /* Update inode ctime. */
3003                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3004                                               xis->inode_bh,
3005                                               OCFS2_JOURNAL_ACCESS_WRITE);
3006                 if (ret) {
3007                         mlog_errno(ret);
3008                         goto out;
3009                 }
3010
3011                 inode->i_ctime = CURRENT_TIME;
3012                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3013                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3014                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3015         }
3016 out:
3017         return ret;
3018 }
3019
3020 /*
3021  * This function only called duing creating inode
3022  * for init security/acl xattrs of the new inode.
3023  * All transanction credits have been reserved in mknod.
3024  */
3025 int ocfs2_xattr_set_handle(handle_t *handle,
3026                            struct inode *inode,
3027                            struct buffer_head *di_bh,
3028                            int name_index,
3029                            const char *name,
3030                            const void *value,
3031                            size_t value_len,
3032                            int flags,
3033                            struct ocfs2_alloc_context *meta_ac,
3034                            struct ocfs2_alloc_context *data_ac)
3035 {
3036         struct ocfs2_dinode *di;
3037         int ret;
3038
3039         struct ocfs2_xattr_info xi = {
3040                 .name_index = name_index,
3041                 .name = name,
3042                 .value = value,
3043                 .value_len = value_len,
3044         };
3045
3046         struct ocfs2_xattr_search xis = {
3047                 .not_found = -ENODATA,
3048         };
3049
3050         struct ocfs2_xattr_search xbs = {
3051                 .not_found = -ENODATA,
3052         };
3053
3054         struct ocfs2_xattr_set_ctxt ctxt = {
3055                 .handle = handle,
3056                 .meta_ac = meta_ac,
3057                 .data_ac = data_ac,
3058         };
3059
3060         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3061                 return -EOPNOTSUPP;
3062
3063         /*
3064          * In extreme situation, may need xattr bucket when
3065          * block size is too small. And we have already reserved
3066          * the credits for bucket in mknod.
3067          */
3068         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3069                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
3070                 if (!xbs.bucket) {
3071                         mlog_errno(-ENOMEM);
3072                         return -ENOMEM;
3073                 }
3074         }
3075
3076         xis.inode_bh = xbs.inode_bh = di_bh;
3077         di = (struct ocfs2_dinode *)di_bh->b_data;
3078
3079         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3080
3081         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3082         if (ret)
3083                 goto cleanup;
3084         if (xis.not_found) {
3085                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3086                 if (ret)
3087                         goto cleanup;
3088         }
3089
3090         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3091
3092 cleanup:
3093         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3094         brelse(xbs.xattr_bh);
3095         ocfs2_xattr_bucket_free(xbs.bucket);
3096
3097         return ret;
3098 }
3099
3100 /*
3101  * ocfs2_xattr_set()
3102  *
3103  * Set, replace or remove an extended attribute for this inode.
3104  * value is NULL to remove an existing extended attribute, else either
3105  * create or replace an extended attribute.
3106  */
3107 int ocfs2_xattr_set(struct inode *inode,
3108                     int name_index,
3109                     const char *name,
3110                     const void *value,
3111                     size_t value_len,
3112                     int flags)
3113 {
3114         struct buffer_head *di_bh = NULL;
3115         struct ocfs2_dinode *di;
3116         int ret, credits, ref_meta = 0, ref_credits = 0;
3117         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3118         struct inode *tl_inode = osb->osb_tl_inode;
3119         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3120         struct ocfs2_refcount_tree *ref_tree = NULL;
3121
3122         struct ocfs2_xattr_info xi = {
3123                 .name_index = name_index,
3124                 .name = name,
3125                 .value = value,
3126                 .value_len = value_len,
3127         };
3128
3129         struct ocfs2_xattr_search xis = {
3130                 .not_found = -ENODATA,
3131         };
3132
3133         struct ocfs2_xattr_search xbs = {
3134                 .not_found = -ENODATA,
3135         };
3136
3137         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3138                 return -EOPNOTSUPP;
3139
3140         /*
3141          * Only xbs will be used on indexed trees.  xis doesn't need a
3142          * bucket.
3143          */
3144         xbs.bucket = ocfs2_xattr_bucket_new(inode);
3145         if (!xbs.bucket) {
3146                 mlog_errno(-ENOMEM);
3147                 return -ENOMEM;
3148         }
3149
3150         ret = ocfs2_inode_lock(inode, &di_bh, 1);
3151         if (ret < 0) {
3152                 mlog_errno(ret);
3153                 goto cleanup_nolock;
3154         }
3155         xis.inode_bh = xbs.inode_bh = di_bh;
3156         di = (struct ocfs2_dinode *)di_bh->b_data;
3157
3158         down_write(&OCFS2_I(inode)->ip_xattr_sem);
3159         /*
3160          * Scan inode and external block to find the same name
3161          * extended attribute and collect search infomation.
3162          */
3163         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3164         if (ret)
3165                 goto cleanup;
3166         if (xis.not_found) {
3167                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3168                 if (ret)
3169                         goto cleanup;
3170         }
3171
3172         if (xis.not_found && xbs.not_found) {
3173                 ret = -ENODATA;
3174                 if (flags & XATTR_REPLACE)
3175                         goto cleanup;
3176                 ret = 0;
3177                 if (!value)
3178                         goto cleanup;
3179         } else {
3180                 ret = -EEXIST;
3181                 if (flags & XATTR_CREATE)
3182                         goto cleanup;
3183         }
3184
3185         /* Check whether the value is refcounted and do some prepartion. */
3186         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3187             (!xis.not_found || !xbs.not_found)) {
3188                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3189                                                    &xis, &xbs, &ref_tree,
3190                                                    &ref_meta, &ref_credits);
3191                 if (ret) {
3192                         mlog_errno(ret);
3193                         goto cleanup;
3194                 }
3195         }
3196
3197         mutex_lock(&tl_inode->i_mutex);
3198
3199         if (ocfs2_truncate_log_needs_flush(osb)) {
3200                 ret = __ocfs2_flush_truncate_log(osb);
3201                 if (ret < 0) {
3202                         mutex_unlock(&tl_inode->i_mutex);
3203                         mlog_errno(ret);
3204                         goto cleanup;
3205                 }
3206         }
3207         mutex_unlock(&tl_inode->i_mutex);
3208
3209         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3210                                         &xbs, &ctxt, ref_meta, &credits);
3211         if (ret) {
3212                 mlog_errno(ret);
3213                 goto cleanup;
3214         }
3215
3216         /* we need to update inode's ctime field, so add credit for it. */
3217         credits += OCFS2_INODE_UPDATE_CREDITS;
3218         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3219         if (IS_ERR(ctxt.handle)) {
3220                 ret = PTR_ERR(ctxt.handle);
3221                 mlog_errno(ret);
3222                 goto cleanup;
3223         }
3224
3225         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3226
3227         ocfs2_commit_trans(osb, ctxt.handle);
3228
3229         if (ctxt.data_ac)
3230                 ocfs2_free_alloc_context(ctxt.data_ac);
3231         if (ctxt.meta_ac)
3232                 ocfs2_free_alloc_context(ctxt.meta_ac);
3233         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3234                 ocfs2_schedule_truncate_log_flush(osb, 1);
3235         ocfs2_run_deallocs(osb, &ctxt.dealloc);
3236
3237 cleanup:
3238         if (ref_tree)
3239                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3240         up_write(&OCFS2_I(inode)->ip_xattr_sem);
3241         if (!value && !ret) {
3242                 ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3243                 if (ret)
3244                         mlog_errno(ret);
3245         }
3246         ocfs2_inode_unlock(inode, 1);
3247 cleanup_nolock:
3248         brelse(di_bh);
3249         brelse(xbs.xattr_bh);
3250         ocfs2_xattr_bucket_free(xbs.bucket);
3251
3252         return ret;
3253 }
3254
3255 /*
3256  * Find the xattr extent rec which may contains name_hash.
3257  * e_cpos will be the first name hash of the xattr rec.
3258  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3259  */
3260 static int ocfs2_xattr_get_rec(struct inode *inode,
3261                                u32 name_hash,
3262                                u64 *p_blkno,
3263                                u32 *e_cpos,
3264                                u32 *num_clusters,
3265                                struct ocfs2_extent_list *el)
3266 {
3267         int ret = 0, i;
3268         struct buffer_head *eb_bh = NULL;
3269         struct ocfs2_extent_block *eb;
3270         struct ocfs2_extent_rec *rec = NULL;
3271         u64 e_blkno = 0;
3272
3273         if (el->l_tree_depth) {
3274                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3275                                       &eb_bh);
3276                 if (ret) {
3277                         mlog_errno(ret);
3278                         goto out;
3279                 }
3280
3281                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3282                 el = &eb->h_list;
3283
3284                 if (el->l_tree_depth) {
3285                         ocfs2_error(inode->i_sb,
3286                                     "Inode %lu has non zero tree depth in "
3287                                     "xattr tree block %llu\n", inode->i_ino,
3288                                     (unsigned long long)eb_bh->b_blocknr);
3289                         ret = -EROFS;
3290                         goto out;
3291                 }
3292         }
3293
3294         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3295                 rec = &el->l_recs[i];
3296
3297                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3298                         e_blkno = le64_to_cpu(rec->e_blkno);
3299                         break;
3300                 }
3301         }
3302
3303         if (!e_blkno) {
3304                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3305                             "record (%u, %u, 0) in xattr", inode->i_ino,
3306                             le32_to_cpu(rec->e_cpos),
3307                             ocfs2_rec_clusters(el, rec));
3308                 ret = -EROFS;
3309                 goto out;
3310         }
3311
3312         *p_blkno = le64_to_cpu(rec->e_blkno);
3313         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3314         if (e_cpos)
3315                 *e_cpos = le32_to_cpu(rec->e_cpos);
3316 out:
3317         brelse(eb_bh);
3318         return ret;
3319 }
3320
3321 typedef int (xattr_bucket_func)(struct inode *inode,
3322                                 struct ocfs2_xattr_bucket *bucket,
3323                                 void *para);
3324
3325 static int ocfs2_find_xe_in_bucket(struct inode *inode,
3326                                    struct ocfs2_xattr_bucket *bucket,
3327                                    int name_index,
3328                                    const char *name,
3329                                    u32 name_hash,
3330                                    u16 *xe_index,
3331                                    int *found)
3332 {
3333         int i, ret = 0, cmp = 1, block_off, new_offset;
3334         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3335         size_t name_len = strlen(name);
3336         struct ocfs2_xattr_entry *xe = NULL;
3337         char *xe_name;
3338
3339         /*
3340          * We don't use binary search in the bucket because there
3341          * may be multiple entries with the same name hash.
3342          */
3343         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3344                 xe = &xh->xh_entries[i];
3345
3346                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
3347                         continue;
3348                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3349                         break;
3350
3351                 cmp = name_index - ocfs2_xattr_get_type(xe);
3352                 if (!cmp)
3353                         cmp = name_len - xe->xe_name_len;
3354                 if (cmp)
3355                         continue;
3356
3357                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3358                                                         xh,
3359                                                         i,
3360                                                         &block_off,
3361                                                         &new_offset);
3362                 if (ret) {
3363                         mlog_errno(ret);
3364                         break;
3365                 }
3366
3367
3368                 xe_name = bucket_block(bucket, block_off) + new_offset;
3369                 if (!memcmp(name, xe_name, name_len)) {
3370                         *xe_index = i;
3371                         *found = 1;
3372                         ret = 0;
3373                         break;
3374                 }
3375         }
3376
3377         return ret;
3378 }
3379
3380 /*
3381  * Find the specified xattr entry in a series of buckets.
3382  * This series start from p_blkno and last for num_clusters.
3383  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3384  * the num of the valid buckets.
3385  *
3386  * Return the buffer_head this xattr should reside in. And if the xattr's
3387  * hash is in the gap of 2 buckets, return the lower bucket.
3388  */
3389 static int ocfs2_xattr_bucket_find(struct inode *inode,
3390                                    int name_index,
3391                                    const char *name,
3392                                    u32 name_hash,
3393                                    u64 p_blkno,
3394                                    u32 first_hash,
3395                                    u32 num_clusters,
3396                                    struct ocfs2_xattr_search *xs)
3397 {
3398         int ret, found = 0;
3399         struct ocfs2_xattr_header *xh = NULL;
3400         struct ocfs2_xattr_entry *xe = NULL;
3401         u16 index = 0;
3402         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3403         int low_bucket = 0, bucket, high_bucket;
3404         struct ocfs2_xattr_bucket *search;
3405         u32 last_hash;
3406         u64 blkno, lower_blkno = 0;
3407
3408         search = ocfs2_xattr_bucket_new(inode);
3409         if (!search) {
3410                 ret = -ENOMEM;
3411                 mlog_errno(ret);
3412                 goto out;
3413         }
3414
3415         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3416         if (ret) {
3417                 mlog_errno(ret);
3418                 goto out;
3419         }
3420
3421         xh = bucket_xh(search);
3422         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3423         while (low_bucket <= high_bucket) {
3424                 ocfs2_xattr_bucket_relse(search);
3425
3426                 bucket = (low_bucket + high_bucket) / 2;
3427                 blkno = p_blkno + bucket * blk_per_bucket;
3428                 ret = ocfs2_read_xattr_bucket(search, blkno);
3429                 if (ret) {
3430                         mlog_errno(ret);
3431                         goto out;
3432                 }
3433
3434                 xh = bucket_xh(search);
3435                 xe = &xh->xh_entries[0];
3436                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3437                         high_bucket = bucket - 1;
3438                         continue;
3439                 }
3440
3441                 /*
3442                  * Check whether the hash of the last entry in our
3443                  * bucket is larger than the search one. for an empty
3444                  * bucket, the last one is also the first one.
3445                  */
3446                 if (xh->xh_count)
3447                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3448
3449                 last_hash = le32_to_cpu(xe->xe_name_hash);
3450
3451                 /* record lower_blkno which may be the insert place. */
3452                 lower_blkno = blkno;
3453
3454                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3455                         low_bucket = bucket + 1;
3456                         continue;
3457                 }
3458
3459                 /* the searched xattr should reside in this bucket if exists. */
3460                 ret = ocfs2_find_xe_in_bucket(inode, search,
3461                                               name_index, name, name_hash,
3462                                               &index, &found);
3463                 if (ret) {
3464                         mlog_errno(ret);
3465                         goto out;
3466                 }
3467                 break;
3468         }
3469
3470         /*
3471          * Record the bucket we have found.
3472          * When the xattr's hash value is in the gap of 2 buckets, we will
3473          * always set it to the previous bucket.
3474          */
3475         if (!lower_blkno)
3476                 lower_blkno = p_blkno;
3477
3478         /* This should be in cache - we just read it during the search */
3479         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3480         if (ret) {
3481                 mlog_errno(ret);
3482                 goto out;
3483         }
3484
3485         xs->header = bucket_xh(xs->bucket);
3486         xs->base = bucket_block(xs->bucket, 0);
3487         xs->end = xs->base + inode->i_sb->s_blocksize;
3488
3489         if (found) {
3490                 xs->here = &xs->header->xh_entries[index];
3491                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3492                      (unsigned long long)bucket_blkno(xs->bucket), index);
3493         } else
3494                 ret = -ENODATA;
3495
3496 out:
3497         ocfs2_xattr_bucket_free(search);
3498         return ret;
3499 }
3500
3501 static int ocfs2_xattr_index_block_find(struct inode *inode,
3502                                         struct buffer_head *root_bh,
3503                                         int name_index,
3504                                         const char *name,
3505                                         struct ocfs2_xattr_search *xs)
3506 {
3507         int ret;
3508         struct ocfs2_xattr_block *xb =
3509                         (struct ocfs2_xattr_block *)root_bh->b_data;
3510         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3511         struct ocfs2_extent_list *el = &xb_root->xt_list;
3512         u64 p_blkno = 0;
3513         u32 first_hash, num_clusters = 0;
3514         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3515
3516         if (le16_to_cpu(el->l_next_free_rec) == 0)
3517                 return -ENODATA;
3518
3519         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3520              name, name_hash, name_index);
3521
3522         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3523                                   &num_clusters, el);
3524         if (ret) {
3525                 mlog_errno(ret);
3526                 goto out;
3527         }
3528
3529         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3530
3531         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3532              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3533              first_hash);
3534
3535         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3536                                       p_blkno, first_hash, num_clusters, xs);
3537
3538 out:
3539         return ret;
3540 }
3541
3542 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3543                                        u64 blkno,
3544                                        u32 clusters,
3545                                        xattr_bucket_func *func,
3546                                        void *para)
3547 {
3548         int i, ret = 0;
3549         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3550         u32 num_buckets = clusters * bpc;
3551         struct ocfs2_xattr_bucket *bucket;
3552
3553         bucket = ocfs2_xattr_bucket_new(inode);
3554         if (!bucket) {
3555                 mlog_errno(-ENOMEM);
3556                 return -ENOMEM;
3557         }
3558
3559         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3560              clusters, (unsigned long long)blkno);
3561
3562         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3563                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3564                 if (ret) {
3565                         mlog_errno(ret);
3566                         break;
3567                 }
3568
3569                 /*
3570                  * The real bucket num in this series of blocks is stored
3571                  * in the 1st bucket.
3572                  */
3573                 if (i == 0)
3574                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3575
3576                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3577                      (unsigned long long)blkno,
3578                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3579                 if (func) {
3580                         ret = func(inode, bucket, para);
3581                         if (ret && ret != -ERANGE)
3582                                 mlog_errno(ret);
3583                         /* Fall through to bucket_relse() */
3584                 }
3585
3586                 ocfs2_xattr_bucket_relse(bucket);
3587                 if (ret)
3588                         break;
3589         }
3590
3591         ocfs2_xattr_bucket_free(bucket);
3592         return ret;
3593 }
3594
3595 struct ocfs2_xattr_tree_list {
3596         char *buffer;
3597         size_t buffer_size;
3598         size_t result;
3599 };
3600
3601 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3602                                              struct ocfs2_xattr_header *xh,
3603                                              int index,
3604                                              int *block_off,
3605                                              int *new_offset)
3606 {
3607         u16 name_offset;
3608
3609         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3610                 return -EINVAL;
3611
3612         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3613
3614         *block_off = name_offset >> sb->s_blocksize_bits;
3615         *new_offset = name_offset % sb->s_blocksize;
3616
3617         return 0;
3618 }
3619
3620 static int ocfs2_list_xattr_bucket(struct inode *inode,
3621                                    struct ocfs2_xattr_bucket *bucket,
3622                                    void *para)
3623 {
3624         int ret = 0, type;
3625         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3626         int i, block_off, new_offset;
3627         const char *prefix, *name;
3628
3629         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3630                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3631                 type = ocfs2_xattr_get_type(entry);
3632                 prefix = ocfs2_xattr_prefix(type);
3633
3634                 if (prefix) {
3635                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3636                                                                 bucket_xh(bucket),
3637                                                                 i,
3638                                                                 &block_off,
3639                                                                 &new_offset);
3640                         if (ret)
3641                                 break;
3642
3643                         name = (const char *)bucket_block(bucket, block_off) +
3644                                 new_offset;
3645                         ret = ocfs2_xattr_list_entry(xl->buffer,
3646                                                      xl->buffer_size,
3647                                                      &xl->result,
3648                                                      prefix, name,
3649                                                      entry->xe_name_len);
3650                         if (ret)
3651                                 break;
3652                 }
3653         }
3654
3655         return ret;
3656 }
3657
3658 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
3659                                            struct buffer_head *blk_bh,
3660                                            xattr_tree_rec_func *rec_func,
3661                                            void *para)
3662 {
3663         struct ocfs2_xattr_block *xb =
3664                         (struct ocfs2_xattr_block *)blk_bh->b_data;
3665         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
3666         int ret = 0;
3667         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3668         u64 p_blkno = 0;
3669
3670         if (!el->l_next_free_rec || !rec_func)
3671                 return 0;
3672
3673         while (name_hash > 0) {
3674                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3675                                           &e_cpos, &num_clusters, el);
3676                 if (ret) {
3677                         mlog_errno(ret);
3678                         break;
3679                 }
3680
3681                 ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
3682                                num_clusters, para);
3683                 if (ret) {
3684                         if (ret != -ERANGE)
3685                                 mlog_errno(ret);
3686                         break;
3687                 }
3688
3689                 if (e_cpos == 0)
3690                         break;
3691
3692                 name_hash = e_cpos - 1;
3693         }
3694
3695         return ret;
3696
3697 }
3698
3699 static int ocfs2_list_xattr_tree_rec(struct inode *inode,
3700                                      struct buffer_head *root_bh,
3701                                      u64 blkno, u32 cpos, u32 len, void *para)
3702 {
3703         return ocfs2_iterate_xattr_buckets(inode, blkno, len,
3704                                            ocfs2_list_xattr_bucket, para);
3705 }
3706
3707 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3708                                              struct buffer_head *blk_bh,
3709                                              char *buffer,
3710                                              size_t buffer_size)
3711 {
3712         int ret;
3713         struct ocfs2_xattr_tree_list xl = {
3714                 .buffer = buffer,
3715                 .buffer_size = buffer_size,
3716                 .result = 0,
3717         };
3718
3719         ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
3720                                               ocfs2_list_xattr_tree_rec, &xl);
3721         if (ret) {
3722                 mlog_errno(ret);
3723                 goto out;
3724         }
3725
3726         ret = xl.result;
3727 out:
3728         return ret;
3729 }
3730
3731 static int cmp_xe(const void *a, const void *b)
3732 {
3733         const struct ocfs2_xattr_entry *l = a, *r = b;
3734         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3735         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3736
3737         if (l_hash > r_hash)
3738                 return 1;
3739         if (l_hash < r_hash)
3740                 return -1;
3741         return 0;
3742 }
3743
3744 static void swap_xe(void *a, void *b, int size)
3745 {
3746         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3747
3748         tmp = *l;
3749         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3750         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3751 }
3752
3753 /*
3754  * When the ocfs2_xattr_block is filled up, new bucket will be created
3755  * and all the xattr entries will be moved to the new bucket.
3756  * The header goes at the start of the bucket, and the names+values are
3757  * filled from the end.  This is why *target starts as the last buffer.
3758  * Note: we need to sort the entries since they are not saved in order
3759  * in the ocfs2_xattr_block.
3760  */
3761 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3762                                            struct buffer_head *xb_bh,
3763                                            struct ocfs2_xattr_bucket *bucket)
3764 {
3765         int i, blocksize = inode->i_sb->s_blocksize;
3766         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3767         u16 offset, size, off_change;
3768         struct ocfs2_xattr_entry *xe;
3769         struct ocfs2_xattr_block *xb =
3770                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3771         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3772         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3773         u16 count = le16_to_cpu(xb_xh->xh_count);
3774         char *src = xb_bh->b_data;
3775         char *target = bucket_block(bucket, blks - 1);
3776
3777         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3778              (unsigned long long)xb_bh->b_blocknr,
3779              (unsigned long long)bucket_blkno(bucket));
3780
3781         for (i = 0; i < blks; i++)
3782                 memset(bucket_block(bucket, i), 0, blocksize);
3783
3784         /*
3785          * Since the xe_name_offset is based on ocfs2_xattr_header,
3786          * there is a offset change corresponding to the change of
3787          * ocfs2_xattr_header's position.
3788          */
3789         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3790         xe = &xb_xh->xh_entries[count - 1];
3791         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3792         size = blocksize - offset;
3793
3794         /* copy all the names and values. */
3795         memcpy(target + offset, src + offset, size);
3796
3797         /* Init new header now. */
3798         xh->xh_count = xb_xh->xh_count;
3799         xh->xh_num_buckets = cpu_to_le16(1);
3800         xh->xh_name_value_len = cpu_to_le16(size);
3801         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3802
3803         /* copy all the entries. */
3804         target = bucket_block(bucket, 0);
3805         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3806         size = count * sizeof(struct ocfs2_xattr_entry);
3807         memcpy(target + offset, (char *)xb_xh + offset, size);
3808
3809         /* Change the xe offset for all the xe because of the move. */
3810         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3811                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3812         for (i = 0; i < count; i++)
3813                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3814
3815         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3816              offset, size, off_change);
3817
3818         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3819              cmp_xe, swap_xe);
3820 }
3821
3822 /*
3823  * After we move xattr from block to index btree, we have to
3824  * update ocfs2_xattr_search to the new xe and base.
3825  *
3826  * When the entry is in xattr block, xattr_bh indicates the storage place.
3827  * While if the entry is in index b-tree, "bucket" indicates the
3828  * real place of the xattr.
3829  */
3830 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3831                                             struct ocfs2_xattr_search *xs,
3832                                             struct buffer_head *old_bh)
3833 {
3834         char *buf = old_bh->b_data;
3835         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3836         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3837         int i;
3838
3839         xs->header = bucket_xh(xs->bucket);
3840         xs->base = bucket_block(xs->bucket, 0);
3841         xs->end = xs->base + inode->i_sb->s_blocksize;
3842
3843         if (xs->not_found)
3844                 return;
3845
3846         i = xs->here - old_xh->xh_entries;
3847         xs->here = &xs->header->xh_entries[i];
3848 }
3849
3850 static int ocfs2_xattr_create_index_block(struct inode *inode,
3851                                           struct ocfs2_xattr_search *xs,
3852                                           struct ocfs2_xattr_set_ctxt *ctxt)
3853 {
3854         int ret;
3855         u32 bit_off, len;
3856         u64 blkno;
3857         handle_t *handle = ctxt->handle;
3858         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3859         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3860         struct buffer_head *xb_bh = xs->xattr_bh;
3861         struct ocfs2_xattr_block *xb =
3862                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3863         struct ocfs2_xattr_tree_root *xr;
3864         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3865
3866         mlog(0, "create xattr index block for %llu\n",
3867              (unsigned long long)xb_bh->b_blocknr);
3868
3869         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3870         BUG_ON(!xs->bucket);
3871
3872         /*
3873          * XXX:
3874          * We can use this lock for now, and maybe move to a dedicated mutex
3875          * if performance becomes a problem later.
3876          */
3877         down_write(&oi->ip_alloc_sem);
3878
3879         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3880                                       OCFS2_JOURNAL_ACCESS_WRITE);
3881         if (ret) {
3882                 mlog_errno(ret);
3883                 goto out;
3884         }
3885
3886         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3887                                      1, 1, &bit_off, &len);
3888         if (ret) {
3889                 mlog_errno(ret);
3890                 goto out;
3891         }
3892
3893         /*
3894          * The bucket may spread in many blocks, and
3895          * we will only touch the 1st block and the last block
3896          * in the whole bucket(one for entry and one for data).
3897          */
3898         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3899
3900         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3901              (unsigned long long)blkno);
3902
3903         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3904         if (ret) {
3905                 mlog_errno(ret);
3906                 goto out;
3907         }
3908
3909         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3910                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3911         if (ret) {
3912                 mlog_errno(ret);
3913                 goto out;
3914         }
3915
3916         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3917         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3918
3919         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3920
3921         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3922         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3923                offsetof(struct ocfs2_xattr_block, xb_attrs));
3924
3925         xr = &xb->xb_attrs.xb_root;
3926         xr->xt_clusters = cpu_to_le32(1);
3927         xr->xt_last_eb_blk = 0;
3928         xr->xt_list.l_tree_depth = 0;
3929         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3930         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3931
3932         xr->xt_list.l_recs[0].e_cpos = 0;
3933         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3934         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3935
3936         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3937
3938         ocfs2_journal_dirty(handle, xb_bh);
3939
3940 out:
3941         up_write(&oi->ip_alloc_sem);
3942
3943         return ret;
3944 }
3945
3946 static int cmp_xe_offset(const void *a, const void *b)
3947 {
3948         const struct ocfs2_xattr_entry *l = a, *r = b;
3949         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3950         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3951
3952         if (l_name_offset < r_name_offset)
3953                 return 1;
3954         if (l_name_offset > r_name_offset)
3955                 return -1;
3956         return 0;
3957 }
3958
3959 /*
3960  * defrag a xattr bucket if we find that the bucket has some
3961  * holes beteen name/value pairs.
3962  * We will move all the name/value pairs to the end of the bucket
3963  * so that we can spare some space for insertion.
3964  */
3965 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3966                                      handle_t *handle,
3967                                      struct ocfs2_xattr_bucket *bucket)
3968 {
3969         int ret, i;
3970         size_t end, offset, len, value_len;
3971         struct ocfs2_xattr_header *xh;
3972         char *entries, *buf, *bucket_buf = NULL;
3973         u64 blkno = bucket_blkno(bucket);
3974         u16 xh_free_start;
3975         size_t blocksize = inode->i_sb->s_blocksize;
3976         struct ocfs2_xattr_entry *xe;
3977
3978         /*
3979          * In order to make the operation more efficient and generic,
3980          * we copy all the blocks into a contiguous memory and do the
3981          * defragment there, so if anything is error, we will not touch
3982          * the real block.
3983          */
3984         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3985         if (!bucket_buf) {
3986                 ret = -EIO;
3987                 goto out;
3988         }
3989
3990         buf = bucket_buf;
3991         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3992                 memcpy(buf, bucket_block(bucket, i), blocksize);
3993
3994         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3995                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3996         if (ret < 0) {
3997                 mlog_errno(ret);
3998                 goto out;
3999         }
4000
4001         xh = (struct ocfs2_xattr_header *)bucket_buf;
4002         entries = (char *)xh->xh_entries;
4003         xh_free_start = le16_to_cpu(xh->xh_free_start);
4004
4005         mlog(0, "adjust xattr bucket in %llu, count = %u, "
4006              "xh_free_start = %u, xh_name_value_len = %u.\n",
4007              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4008              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4009
4010         /*
4011          * sort all the entries by their offset.
4012          * the largest will be the first, so that we can
4013          * move them to the end one by one.
4014          */
4015         sort(entries, le16_to_cpu(xh->xh_count),
4016              sizeof(struct ocfs2_xattr_entry),
4017              cmp_xe_offset, swap_xe);
4018
4019         /* Move all name/values to the end of the bucket. */
4020         xe = xh->xh_entries;
4021         end = OCFS2_XATTR_BUCKET_SIZE;
4022         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4023                 offset = le16_to_cpu(xe->xe_name_offset);
4024                 if (ocfs2_xattr_is_local(xe))
4025                         value_len = OCFS2_XATTR_SIZE(
4026                                         le64_to_cpu(xe->xe_value_size));
4027                 else
4028                         value_len = OCFS2_XATTR_ROOT_SIZE;
4029                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
4030
4031                 /*
4032                  * We must make sure that the name/value pair
4033                  * exist in the same block. So adjust end to
4034                  * the previous block end if needed.
4035                  */
4036                 if (((end - len) / blocksize !=
4037                         (end - 1) / blocksize))
4038                         end = end - end % blocksize;
4039
4040                 if (end > offset + len) {
4041                         memmove(bucket_buf + end - len,
4042                                 bucket_buf + offset, len);
4043                         xe->xe_name_offset = cpu_to_le16(end - len);
4044                 }
4045
4046                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4047                                 "bucket %llu\n", (unsigned long long)blkno);
4048
4049                 end -= len;
4050         }
4051
4052         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4053                         "bucket %llu\n", (unsigned long long)blkno);
4054
4055         if (xh_free_start == end)
4056                 goto out;
4057
4058         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4059         xh->xh_free_start = cpu_to_le16(end);
4060
4061         /* sort the entries by their name_hash. */
4062         sort(entries, le16_to_cpu(xh->xh_count),
4063              sizeof(struct ocfs2_xattr_entry),
4064              cmp_xe, swap_xe);
4065
4066         buf = bucket_buf;
4067         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4068                 memcpy(bucket_block(bucket, i), buf, blocksize);
4069         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4070
4071 out:
4072         kfree(bucket_buf);
4073         return ret;
4074 }
4075
4076 /*
4077  * prev_blkno points to the start of an existing extent.  new_blkno
4078  * points to a newly allocated extent.  Because we know each of our
4079  * clusters contains more than bucket, we can easily split one cluster
4080  * at a bucket boundary.  So we take the last cluster of the existing
4081  * extent and split it down the middle.  We move the last half of the
4082  * buckets in the last cluster of the existing extent over to the new
4083  * extent.
4084  *
4085  * first_bh is the buffer at prev_blkno so we can update the existing
4086  * extent's bucket count.  header_bh is the bucket were we were hoping
4087  * to insert our xattr.  If the bucket move places the target in the new
4088  * extent, we'll update first_bh and header_bh after modifying the old
4089  * extent.
4090  *
4091  * first_hash will be set as the 1st xe's name_hash in the new extent.
4092  */
4093 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4094                                                handle_t *handle,
4095                                                struct ocfs2_xattr_bucket *first,
4096                                                struct ocfs2_xattr_bucket *target,
4097                                                u64 new_blkno,
4098                                                u32 num_clusters,
4099                                                u32 *first_hash)
4100 {
4101         int ret;
4102         struct super_block *sb = inode->i_sb;
4103         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4104         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4105         int to_move = num_buckets / 2;
4106         u64 src_blkno;
4107         u64 last_cluster_blkno = bucket_blkno(first) +
4108                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4109
4110         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4111         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4112
4113         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
4114              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
4115
4116         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4117                                      last_cluster_blkno, new_blkno,
4118                                      to_move, first_hash);
4119         if (ret) {
4120                 mlog_errno(ret);
4121                 goto out;
4122         }
4123
4124         /* This is the first bucket that got moved */
4125         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4126
4127         /*
4128          * If the target bucket was part of the moved buckets, we need to
4129          * update first and target.
4130          */
4131         if (bucket_blkno(target) >= src_blkno) {
4132                 /* Find the block for the new target bucket */
4133                 src_blkno = new_blkno +
4134                         (bucket_blkno(target) - src_blkno);
4135
4136                 ocfs2_xattr_bucket_relse(first);
4137                 ocfs2_xattr_bucket_relse(target);
4138
4139                 /*
4140                  * These shouldn't fail - the buffers are in the
4141                  * journal from ocfs2_cp_xattr_bucket().
4142                  */
4143                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
4144                 if (ret) {
4145                         mlog_errno(ret);
4146                         goto out;
4147                 }
4148                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
4149                 if (ret)
4150                         mlog_errno(ret);
4151
4152         }
4153
4154 out:
4155         return ret;
4156 }
4157
4158 /*
4159  * Find the suitable pos when we divide a bucket into 2.
4160  * We have to make sure the xattrs with the same hash value exist
4161  * in the same bucket.
4162  *
4163  * If this ocfs2_xattr_header covers more than one hash value, find a
4164  * place where the hash value changes.  Try to find the most even split.
4165  * The most common case is that all entries have different hash values,
4166  * and the first check we make will find a place to split.
4167  */
4168 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4169 {
4170         struct ocfs2_xattr_entry *entries = xh->xh_entries;
4171         int count = le16_to_cpu(xh->xh_count);
4172         int delta, middle = count / 2;
4173
4174         /*
4175          * We start at the middle.  Each step gets farther away in both
4176          * directions.  We therefore hit the change in hash value
4177          * nearest to the middle.  Note that this loop does not execute for
4178          * count < 2.
4179          */
4180         for (delta = 0; delta < middle; delta++) {
4181                 /* Let's check delta earlier than middle */
4182                 if (cmp_xe(&entries[middle - delta - 1],
4183                            &entries[middle - delta]))
4184                         return middle - delta;
4185
4186                 /* For even counts, don't walk off the end */
4187                 if ((middle + delta + 1) == count)
4188                         continue;
4189
4190                 /* Now try delta past middle */
4191                 if (cmp_xe(&entries[middle + delta],
4192                            &entries[middle + delta + 1]))
4193                         return middle + delta + 1;
4194         }
4195
4196         /* Every entry had the same hash */
4197         return count;
4198 }
4199
4200 /*
4201  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4202  * first_hash will record the 1st hash of the new bucket.
4203  *
4204  * Normally half of the xattrs will be moved.  But we have to make
4205  * sure that the xattrs with the same hash value are stored in the
4206  * same bucket. If all the xattrs in this bucket have the same hash
4207  * value, the new bucket will be initialized as an empty one and the
4208  * first_hash will be initialized as (hash_value+1).
4209  */
4210 static int ocfs2_divide_xattr_bucket(struct inode *inode,
4211                                     handle_t *handle,
4212                                     u64 blk,
4213                                     u64 new_blk,
4214                                     u32 *first_hash,
4215                                     int new_bucket_head)
4216 {
4217         int ret, i;
4218         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
4219         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4220         struct ocfs2_xattr_header *xh;
4221         struct ocfs2_xattr_entry *xe;
4222         int blocksize = inode->i_sb->s_blocksize;
4223
4224         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
4225              (unsigned long long)blk, (unsigned long long)new_blk);
4226
4227         s_bucket = ocfs2_xattr_bucket_new(inode);
4228         t_bucket = ocfs2_xattr_bucket_new(inode);
4229         if (!s_bucket || !t_bucket) {
4230                 ret = -ENOMEM;
4231                 mlog_errno(ret);
4232                 goto out;
4233         }
4234
4235         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4236         if (ret) {
4237                 mlog_errno(ret);
4238                 goto out;
4239         }
4240
4241         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4242                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4243         if (ret) {
4244                 mlog_errno(ret);
4245                 goto out;
4246         }
4247
4248         /*
4249          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
4250          * there's no need to read it.
4251          */
4252         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4253         if (ret) {
4254                 mlog_errno(ret);
4255                 goto out;
4256         }
4257
4258         /*
4259          * Hey, if we're overwriting t_bucket, what difference does
4260          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
4261          * same part of ocfs2_cp_xattr_bucket().
4262          */
4263         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4264                                                 new_bucket_head ?
4265                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4266                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4267         if (ret) {
4268                 mlog_errno(ret);
4269                 goto out;
4270         }
4271
4272         xh = bucket_xh(s_bucket);
4273         count = le16_to_cpu(xh->xh_count);
4274         start = ocfs2_xattr_find_divide_pos(xh);
4275
4276         if (start == count) {
4277                 xe = &xh->xh_entries[start-1];
4278
4279                 /*
4280                  * initialized a new empty bucket here.
4281                  * The hash value is set as one larger than
4282                  * that of the last entry in the previous bucket.
4283                  */
4284                 for (i = 0; i < t_bucket->bu_blocks; i++)
4285                         memset(bucket_block(t_bucket, i), 0, blocksize);
4286
4287                 xh = bucket_xh(t_bucket);
4288                 xh->xh_free_start = cpu_to_le16(blocksize);
4289                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4290                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4291
4292                 goto set_num_buckets;
4293         }
4294
4295         /* copy the whole bucket to the new first. */
4296         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4297
4298         /* update the new bucket. */
4299         xh = bucket_xh(t_bucket);
4300
4301         /*
4302          * Calculate the total name/value len and xh_free_start for
4303          * the old bucket first.
4304          */
4305         name_offset = OCFS2_XATTR_BUCKET_SIZE;
4306         name_value_len = 0;
4307         for (i = 0; i < start; i++) {
4308                 xe = &xh->xh_entries[i];
4309                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
4310                 if (ocfs2_xattr_is_local(xe))
4311                         xe_len +=
4312                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4313                 else
4314                         xe_len += OCFS2_XATTR_ROOT_SIZE;
4315                 name_value_len += xe_len;
4316                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4317                         name_offset = le16_to_cpu(xe->xe_name_offset);
4318         }
4319
4320         /*
4321          * Now begin the modification to the new bucket.
4322          *
4323          * In the new bucket, We just move the xattr entry to the beginning
4324          * and don't touch the name/value. So there will be some holes in the
4325          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4326          * called.
4327          */
4328         xe = &xh->xh_entries[start];
4329         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4330         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
4331              (int)((char *)xe - (char *)xh),
4332              (int)((char *)xh->xh_entries - (char *)xh));
4333         memmove((char *)xh->xh_entries, (char *)xe, len);
4334         xe = &xh->xh_entries[count - start];
4335         len = sizeof(struct ocfs2_xattr_entry) * start;
4336         memset((char *)xe, 0, len);
4337
4338         le16_add_cpu(&xh->xh_count, -start);
4339         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4340
4341         /* Calculate xh_free_start for the new bucket. */
4342         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4343         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4344                 xe = &xh->xh_entries[i];
4345                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
4346                 if (ocfs2_xattr_is_local(xe))
4347                         xe_len +=
4348                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4349                 else
4350                         xe_len += OCFS2_XATTR_ROOT_SIZE;
4351                 if (le16_to_cpu(xe->xe_name_offset) <
4352                     le16_to_cpu(xh->xh_free_start))
4353                         xh->xh_free_start = xe->xe_name_offset;
4354         }
4355
4356 set_num_buckets:
4357         /* set xh->xh_num_buckets for the new xh. */
4358         if (new_bucket_head)
4359                 xh->xh_num_buckets = cpu_to_le16(1);
4360         else
4361                 xh->xh_num_buckets = 0;
4362
4363         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4364
4365         /* store the first_hash of the new bucket. */
4366         if (first_hash)
4367                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4368
4369         /*
4370          * Now only update the 1st block of the old bucket.  If we
4371          * just added a new empty bucket, there is no need to modify
4372          * it.
4373          */
4374         if (start == count)
4375                 goto out;
4376
4377         xh = bucket_xh(s_bucket);
4378         memset(&xh->xh_entries[start], 0,
4379                sizeof(struct ocfs2_xattr_entry) * (count - start));
4380         xh->xh_count = cpu_to_le16(start);
4381         xh->xh_free_start = cpu_to_le16(name_offset);
4382         xh->xh_name_value_len = cpu_to_le16(name_value_len);
4383
4384         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4385
4386 out:
4387         ocfs2_xattr_bucket_free(s_bucket);
4388         ocfs2_xattr_bucket_free(t_bucket);
4389
4390         return ret;
4391 }
4392
4393 /*
4394  * Copy xattr from one bucket to another bucket.
4395  *
4396  * The caller must make sure that the journal transaction
4397  * has enough space for journaling.
4398  */
4399 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4400                                  handle_t *handle,
4401                                  u64 s_blkno,
4402                                  u64 t_blkno,
4403                                  int t_is_new)
4404 {
4405         int ret;
4406         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4407
4408         BUG_ON(s_blkno == t_blkno);
4409
4410         mlog(0, "cp bucket %llu to %llu, target is %d\n",
4411              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4412              t_is_new);
4413
4414         s_bucket = ocfs2_xattr_bucket_new(inode);
4415         t_bucket = ocfs2_xattr_bucket_new(inode);
4416         if (!s_bucket || !t_bucket) {
4417                 ret = -ENOMEM;
4418                 mlog_errno(ret);
4419                 goto out;
4420         }
4421
4422         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4423         if (ret)
4424                 goto out;
4425
4426         /*
4427          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4428          * there's no need to read it.
4429          */
4430         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4431         if (ret)
4432                 goto out;
4433
4434         /*
4435          * Hey, if we're overwriting t_bucket, what difference does
4436          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4437          * cluster to fill, we came here from
4438          * ocfs2_mv_xattr_buckets(), and it is really new -
4439          * ACCESS_CREATE is required.  But we also might have moved data
4440          * out of t_bucket before extending back into it.
4441          * ocfs2_add_new_xattr_bucket() can do this - its call to
4442          * ocfs2_add_new_xattr_cluster() may have created a new extent
4443          * and copied out the end of the old extent.  Then it re-extends
4444          * the old extent back to create space for new xattrs.  That's
4445          * how we get here, and the bucket isn't really new.
4446          */
4447         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4448                                                 t_is_new ?
4449                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4450                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4451         if (ret)
4452                 goto out;
4453
4454         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4455         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4456
4457 out:
4458         ocfs2_xattr_bucket_free(t_bucket);
4459         ocfs2_xattr_bucket_free(s_bucket);
4460
4461         return ret;
4462 }
4463
4464 /*
4465  * src_blk points to the start of an existing extent.  last_blk points to
4466  * last cluster in that extent.  to_blk points to a newly allocated
4467  * extent.  We copy the buckets from the cluster at last_blk to the new
4468  * extent.  If start_bucket is non-zero, we skip that many buckets before
4469  * we start copying.  The new extent's xh_num_buckets gets set to the
4470  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4471  * by the same amount.
4472  */
4473 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4474                                   u64 src_blk, u64 last_blk, u64 to_blk,
4475                                   unsigned int start_bucket,
4476                                   u32 *first_hash)
4477 {
4478         int i, ret, credits;
4479         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4480         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4481         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4482         struct ocfs2_xattr_bucket *old_first, *new_first;
4483
4484         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4485              (unsigned long long)last_blk, (unsigned long long)to_blk);
4486
4487         BUG_ON(start_bucket >= num_buckets);
4488         if (start_bucket) {
4489                 num_buckets -= start_bucket;
4490                 last_blk += (start_bucket * blks_per_bucket);
4491         }
4492
4493         /* The first bucket of the original extent */
4494         old_first = ocfs2_xattr_bucket_new(inode);
4495         /* The first bucket of the new extent */
4496         new_first = ocfs2_xattr_bucket_new(inode);
4497         if (!old_first || !new_first) {
4498                 ret = -ENOMEM;
4499                 mlog_errno(ret);
4500                 goto out;
4501         }
4502
4503         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4504         if (ret) {
4505                 mlog_errno(ret);
4506                 goto out;
4507         }
4508
4509         /*
4510          * We need to update the first bucket of the old extent and all
4511          * the buckets going to the new extent.
4512          */
4513         credits = ((num_buckets + 1) * blks_per_bucket) +
4514                 handle->h_buffer_credits;
4515         ret = ocfs2_extend_trans(handle, credits);
4516         if (ret) {
4517                 mlog_errno(ret);
4518                 goto out;
4519         }
4520
4521         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4522                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4523         if (ret) {
4524                 mlog_errno(ret);
4525                 goto out;
4526         }
4527
4528         for (i = 0; i < num_buckets; i++) {
4529                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4530                                             last_blk + (i * blks_per_bucket),
4531                                             to_blk + (i * blks_per_bucket),
4532                                             1);
4533                 if (ret) {
4534                         mlog_errno(ret);
4535                         goto out;
4536                 }
4537         }
4538
4539         /*
4540          * Get the new bucket ready before we dirty anything
4541          * (This actually shouldn't fail, because we already dirtied
4542          * it once in ocfs2_cp_xattr_bucket()).
4543          */
4544         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4545         if (ret) {
4546                 mlog_errno(ret);
4547                 goto out;
4548         }
4549         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4550                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4551         if (ret) {
4552                 mlog_errno(ret);
4553                 goto out;
4554         }
4555
4556         /* Now update the headers */
4557         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4558         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4559
4560         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4561         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4562
4563         if (first_hash)
4564                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4565
4566 out:
4567         ocfs2_xattr_bucket_free(new_first);
4568         ocfs2_xattr_bucket_free(old_first);
4569         return ret;
4570 }
4571
4572 /*
4573  * Move some xattrs in this cluster to the new cluster.
4574  * This function should only be called when bucket size == cluster size.
4575  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4576  */
4577 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4578                                       handle_t *handle,
4579                                       u64 prev_blk,
4580                                       u64 new_blk,
4581                                       u32 *first_hash)
4582 {
4583         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4584         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4585
4586         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4587
4588         ret = ocfs2_extend_trans(handle, credits);
4589         if (ret) {
4590                 mlog_errno(ret);
4591                 return ret;
4592         }
4593
4594         /* Move half of the xattr in start_blk to the next bucket. */
4595         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4596                                           new_blk, first_hash, 1);
4597 }
4598
4599 /*
4600  * Move some xattrs from the old cluster to the new one since they are not
4601  * contiguous in ocfs2 xattr tree.
4602  *
4603  * new_blk starts a new separate cluster, and we will move some xattrs from
4604  * prev_blk to it. v_start will be set as the first name hash value in this
4605  * new cluster so that it can be used as e_cpos during tree insertion and
4606  * don't collide with our original b-tree operations. first_bh and header_bh
4607  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4608  * to extend the insert bucket.
4609  *
4610  * The problem is how much xattr should we move to the new one and when should
4611  * we update first_bh and header_bh?
4612  * 1. If cluster size > bucket size, that means the previous cluster has more
4613  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4614  *    update the first_bh and header_bh if the insert bucket has been moved
4615  *    to the new cluster.
4616  * 2. If cluster_size == bucket_size:
4617  *    a) If the previous extent rec has more than one cluster and the insert
4618  *       place isn't in the last cluster, copy the entire last cluster to the
4619  *       new one. This time, we don't need to upate the first_bh and header_bh
4620  *       since they will not be moved into the new cluster.
4621  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4622  *       the new one. And we set the extend flag to zero if the insert place is
4623  *       moved into the new allocated cluster since no extend is needed.
4624  */
4625 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4626                                             handle_t *handle,
4627                                             struct ocfs2_xattr_bucket *first,
4628                                             struct ocfs2_xattr_bucket *target,
4629                                             u64 new_blk,
4630                                             u32 prev_clusters,
4631                                             u32 *v_start,
4632                                             int *extend)
4633 {
4634         int ret;
4635
4636         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4637              (unsigned long long)bucket_blkno(first), prev_clusters,
4638              (unsigned long long)new_blk);
4639
4640         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4641                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4642                                                           handle,
4643                                                           first, target,
4644                                                           new_blk,
4645                                                           prev_clusters,
4646                                                           v_start);
4647                 if (ret)
4648                         mlog_errno(ret);
4649         } else {
4650                 /* The start of the last cluster in the first extent */
4651                 u64 last_blk = bucket_blkno(first) +
4652                         ((prev_clusters - 1) *
4653                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4654
4655                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4656                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4657                                                      bucket_blkno(first),
4658                                                      last_blk, new_blk, 0,
4659                                                      v_start);
4660                         if (ret)
4661                                 mlog_errno(ret);
4662                 } else {
4663                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4664                                                          last_blk, new_blk,
4665                                                          v_start);
4666                         if (ret)
4667                                 mlog_errno(ret);
4668
4669                         if ((bucket_blkno(target) == last_blk) && extend)
4670                                 *extend = 0;
4671                 }
4672         }
4673
4674         return ret;
4675 }
4676
4677 /*
4678  * Add a new cluster for xattr storage.
4679  *
4680  * If the new cluster is contiguous with the previous one, it will be
4681  * appended to the same extent record, and num_clusters will be updated.
4682  * If not, we will insert a new extent for it and move some xattrs in
4683  * the last cluster into the new allocated one.
4684  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4685  * lose the benefits of hashing because we'll have to search large leaves.
4686  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4687  * if it's bigger).
4688  *
4689  * first_bh is the first block of the previous extent rec and header_bh
4690  * indicates the bucket we will insert the new xattrs. They will be updated
4691  * when the header_bh is moved into the new cluster.
4692  */
4693 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4694                                        struct buffer_head *root_bh,
4695                                        struct ocfs2_xattr_bucket *first,
4696                                        struct ocfs2_xattr_bucket *target,
4697                                        u32 *num_clusters,
4698                                        u32 prev_cpos,
4699                                        int *extend,
4700                                        struct ocfs2_xattr_set_ctxt *ctxt)
4701 {
4702         int ret;
4703         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4704         u32 prev_clusters = *num_clusters;
4705         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4706         u64 block;
4707         handle_t *handle = ctxt->handle;
4708         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4709         struct ocfs2_extent_tree et;
4710
4711         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4712              "previous xattr blkno = %llu\n",
4713              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4714              prev_cpos, (unsigned long long)bucket_blkno(first));
4715
4716         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4717
4718         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4719                                       OCFS2_JOURNAL_ACCESS_WRITE);
4720         if (ret < 0) {
4721                 mlog_errno(ret);
4722                 goto leave;
4723         }
4724
4725         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4726                                      clusters_to_add, &bit_off, &num_bits);
4727         if (ret < 0) {
4728                 if (ret != -ENOSPC)
4729                         mlog_errno(ret);
4730                 goto leave;
4731         }
4732
4733         BUG_ON(num_bits > clusters_to_add);
4734
4735         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4736         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4737              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4738
4739         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4740             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4741              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4742                 /*
4743                  * If this cluster is contiguous with the old one and
4744                  * adding this new cluster, we don't surpass the limit of
4745                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4746                  * initialized and used like other buckets in the previous
4747                  * cluster.
4748                  * So add it as a contiguous one. The caller will handle
4749                  * its init process.
4750                  */
4751                 v_start = prev_cpos + prev_clusters;
4752                 *num_clusters = prev_clusters + num_bits;
4753                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4754                      num_bits);
4755         } else {
4756                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4757                                                        handle,
4758                                                        first,
4759                                                        target,
4760                                                        block,
4761                                                        prev_clusters,
4762                                                        &v_start,
4763                                                        extend);
4764                 if (ret) {
4765                         mlog_errno(ret);
4766                         goto leave;
4767                 }
4768         }
4769
4770         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4771              num_bits, (unsigned long long)block, v_start);
4772         ret = ocfs2_insert_extent(handle, &et, v_start, block,
4773                                   num_bits, 0, ctxt->meta_ac);
4774         if (ret < 0) {
4775                 mlog_errno(ret);
4776                 goto leave;
4777         }
4778
4779         ret = ocfs2_journal_dirty(handle, root_bh);
4780         if (ret < 0)
4781                 mlog_errno(ret);
4782
4783 leave:
4784         return ret;
4785 }
4786
4787 /*
4788  * We are given an extent.  'first' is the bucket at the very front of
4789  * the extent.  The extent has space for an additional bucket past
4790  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4791  * of the target bucket.  We wish to shift every bucket past the target
4792  * down one, filling in that additional space.  When we get back to the
4793  * target, we split the target between itself and the now-empty bucket
4794  * at target+1 (aka, target_blkno + blks_per_bucket).
4795  */
4796 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4797                                      handle_t *handle,
4798                                      struct ocfs2_xattr_bucket *first,
4799                                      u64 target_blk,
4800                                      u32 num_clusters)
4801 {
4802         int ret, credits;
4803         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4804         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4805         u64 end_blk;
4806         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4807
4808         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4809              "from %llu, len = %u\n", (unsigned long long)target_blk,
4810              (unsigned long long)bucket_blkno(first), num_clusters);
4811
4812         /* The extent must have room for an additional bucket */
4813         BUG_ON(new_bucket >=
4814                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4815
4816         /* end_blk points to the last existing bucket */
4817         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4818
4819         /*
4820          * end_blk is the start of the last existing bucket.
4821          * Thus, (end_blk - target_blk) covers the target bucket and
4822          * every bucket after it up to, but not including, the last
4823          * existing bucket.  Then we add the last existing bucket, the
4824          * new bucket, and the first bucket (3 * blk_per_bucket).
4825          */
4826         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4827                   handle->h_buffer_credits;
4828         ret = ocfs2_extend_trans(handle, credits);
4829         if (ret) {
4830                 mlog_errno(ret);
4831                 goto out;
4832         }
4833
4834         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4835                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4836         if (ret) {
4837                 mlog_errno(ret);
4838                 goto out;
4839         }
4840
4841         while (end_blk != target_blk) {
4842                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4843                                             end_blk + blk_per_bucket, 0);
4844                 if (ret)
4845                         goto out;
4846                 end_blk -= blk_per_bucket;
4847         }
4848
4849         /* Move half of the xattr in target_blkno to the next bucket. */
4850         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4851                                         target_blk + blk_per_bucket, NULL, 0);
4852
4853         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4854         ocfs2_xattr_bucket_journal_dirty(handle, first);
4855
4856 out:
4857         return ret;
4858 }
4859
4860 /*
4861  * Add new xattr bucket in an extent record and adjust the buckets
4862  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4863  * bucket we want to insert into.
4864  *
4865  * In the easy case, we will move all the buckets after target down by
4866  * one. Half of target's xattrs will be moved to the next bucket.
4867  *
4868  * If current cluster is full, we'll allocate a new one.  This may not
4869  * be contiguous.  The underlying calls will make sure that there is
4870  * space for the insert, shifting buckets around if necessary.
4871  * 'target' may be moved by those calls.
4872  */
4873 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4874                                       struct buffer_head *xb_bh,
4875                                       struct ocfs2_xattr_bucket *target,
4876                                       struct ocfs2_xattr_set_ctxt *ctxt)
4877 {
4878         struct ocfs2_xattr_block *xb =
4879                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4880         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4881         struct ocfs2_extent_list *el = &xb_root->xt_list;
4882         u32 name_hash =
4883                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4884         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4885         int ret, num_buckets, extend = 1;
4886         u64 p_blkno;
4887         u32 e_cpos, num_clusters;
4888         /* The bucket at the front of the extent */
4889         struct ocfs2_xattr_bucket *first;
4890
4891         mlog(0, "Add new xattr bucket starting from %llu\n",
4892              (unsigned long long)bucket_blkno(target));
4893
4894         /* The first bucket of the original extent */
4895         first = ocfs2_xattr_bucket_new(inode);
4896         if (!first) {
4897                 ret = -ENOMEM;
4898                 mlog_errno(ret);
4899                 goto out;
4900         }
4901
4902         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4903                                   &num_clusters, el);
4904         if (ret) {
4905                 mlog_errno(ret);
4906                 goto out;
4907         }
4908
4909         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4910         if (ret) {
4911                 mlog_errno(ret);
4912                 goto out;
4913         }
4914
4915         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4916         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4917                 /*
4918                  * This can move first+target if the target bucket moves
4919                  * to the new extent.
4920                  */
4921                 ret = ocfs2_add_new_xattr_cluster(inode,
4922                                                   xb_bh,
4923                                                   first,
4924                                                   target,
4925                                                   &num_clusters,
4926                                                   e_cpos,
4927                                                   &extend,
4928                                                   ctxt);
4929                 if (ret) {
4930                         mlog_errno(ret);
4931                         goto out;
4932                 }
4933         }
4934
4935         if (extend) {
4936                 ret = ocfs2_extend_xattr_bucket(inode,
4937                                                 ctxt->handle,
4938                                                 first,
4939                                                 bucket_blkno(target),
4940                                                 num_clusters);
4941                 if (ret)
4942                         mlog_errno(ret);
4943         }
4944
4945 out:
4946         ocfs2_xattr_bucket_free(first);
4947
4948         return ret;
4949 }
4950
4951 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4952                                         struct ocfs2_xattr_bucket *bucket,
4953                                         int offs)
4954 {
4955         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4956
4957         offs = offs % inode->i_sb->s_blocksize;
4958         return bucket_block(bucket, block_off) + offs;
4959 }
4960
4961 /*
4962  * Handle the normal xattr set, including replace, delete and new.
4963  *
4964  * Note: "local" indicates the real data's locality. So we can't
4965  * just its bucket locality by its length.
4966  */
4967 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4968                                          struct ocfs2_xattr_info *xi,
4969                                          struct ocfs2_xattr_search *xs,
4970                                          u32 name_hash,
4971                                          int local)
4972 {
4973         struct ocfs2_xattr_entry *last, *xe;
4974         int name_len = strlen(xi->name);
4975         struct ocfs2_xattr_header *xh = xs->header;
4976         u16 count = le16_to_cpu(xh->xh_count), start;
4977         size_t blocksize = inode->i_sb->s_blocksize;
4978         char *val;
4979         size_t offs, size, new_size;
4980         struct ocfs2_xa_loc loc;
4981
4982         ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
4983                                        xs->not_found ? NULL : xs->here);
4984         last = &xh->xh_entries[count];
4985         if (!xs->not_found) {
4986                 xe = xs->here;
4987                 offs = le16_to_cpu(xe->xe_name_offset);
4988                 if (ocfs2_xattr_is_local(xe))
4989                         size = OCFS2_XATTR_SIZE(name_len) +
4990                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4991                 else
4992                         size = OCFS2_XATTR_SIZE(name_len) +
4993                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4994
4995                 /*
4996                  * If the new value will be stored outside, xi->value has been
4997                  * initalized as an empty ocfs2_xattr_value_root, and the same
4998                  * goes with xi->value_len, so we can set new_size safely here.
4999                  * See ocfs2_xattr_set_in_bucket.
5000                  */
5001                 new_size = OCFS2_XATTR_SIZE(name_len) +
5002                            OCFS2_XATTR_SIZE(xi->value_len);
5003
5004                 ocfs2_xa_wipe_namevalue(&loc);
5005                 if (xi->value) {
5006                         if (new_size > size)
5007                                 goto set_new_name_value;
5008
5009                         /* Now replace the old value with new one. */
5010                         if (local)
5011                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
5012                         else
5013                                 xe->xe_value_size = 0;
5014
5015                         val = ocfs2_xattr_bucket_get_val(inode,
5016                                                          xs->bucket, offs);
5017                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
5018                                size - OCFS2_XATTR_SIZE(name_len));
5019                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
5020                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
5021                                        xi->value, xi->value_len);
5022
5023                         le16_add_cpu(&xh->xh_name_value_len, new_size);
5024                         ocfs2_xattr_set_local(xe, local);
5025                         return;
5026                 } else {
5027                         /*
5028                          * Remove the old entry if there is more than one.
5029                          * We don't remove the last entry so that we can
5030                          * use it to indicate the hash value of the empty
5031                          * bucket.
5032                          */
5033                         last -= 1;
5034                         le16_add_cpu(&xh->xh_count, -1);
5035                         if (xh->xh_count) {
5036                                 memmove(xe, xe + 1,
5037                                         (void *)last - (void *)xe);
5038                                 memset(last, 0,
5039                                        sizeof(struct ocfs2_xattr_entry));
5040                         } else
5041                                 xh->xh_free_start =
5042                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
5043
5044                         return;
5045                 }
5046         } else {
5047                 /* find a new entry for insert. */
5048                 int low = 0, high = count - 1, tmp;
5049                 struct ocfs2_xattr_entry *tmp_xe;
5050
5051                 while (low <= high && count) {
5052                         tmp = (low + high) / 2;
5053                         tmp_xe = &xh->xh_entries[tmp];
5054
5055                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
5056                                 low = tmp + 1;
5057                         else if (name_hash <
5058                                  le32_to_cpu(tmp_xe->xe_name_hash))
5059                                 high = tmp - 1;
5060                         else {
5061                                 low = tmp;
5062                                 break;
5063                         }
5064                 }
5065
5066                 xe = &xh->xh_entries[low];
5067                 if (low != count)
5068                         memmove(xe + 1, xe, (void *)last - (void *)xe);
5069
5070                 le16_add_cpu(&xh->xh_count, 1);
5071                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
5072                 xe->xe_name_hash = cpu_to_le32(name_hash);
5073                 xe->xe_name_len = name_len;
5074                 ocfs2_xattr_set_type(xe, xi->name_index);
5075         }
5076
5077 set_new_name_value:
5078         /* Insert the new name+value. */
5079         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
5080
5081         /*
5082          * We must make sure that the name/value pair
5083          * exists in the same block.
5084          */
5085         offs = le16_to_cpu(xh->xh_free_start);
5086         start = offs - size;
5087
5088         if (start >> inode->i_sb->s_blocksize_bits !=
5089             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
5090                 offs = offs - offs % blocksize;
5091                 xh->xh_free_start = cpu_to_le16(offs);
5092         }
5093
5094         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
5095         xe->xe_name_offset = cpu_to_le16(offs - size);
5096
5097         memset(val, 0, size);
5098         memcpy(val, xi->name, name_len);
5099         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
5100
5101         xe->xe_value_size = cpu_to_le64(xi->value_len);
5102         ocfs2_xattr_set_local(xe, local);
5103         xs->here = xe;
5104         le16_add_cpu(&xh->xh_free_start, -size);
5105         le16_add_cpu(&xh->xh_name_value_len, size);
5106
5107         return;
5108 }
5109
5110 /*
5111  * Set the xattr entry in the specified bucket.
5112  * The bucket is indicated by xs->bucket and it should have the enough
5113  * space for the xattr insertion.
5114  */
5115 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
5116                                            handle_t *handle,
5117                                            struct ocfs2_xattr_info *xi,
5118                                            struct ocfs2_xattr_search *xs,
5119                                            u32 name_hash,
5120                                            int local)
5121 {
5122         int ret;
5123         u64 blkno;
5124
5125         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
5126              (unsigned long)xi->value_len, xi->name_index,
5127              (unsigned long long)bucket_blkno(xs->bucket));
5128
5129         if (!xs->bucket->bu_bhs[1]) {
5130                 blkno = bucket_blkno(xs->bucket);
5131                 ocfs2_xattr_bucket_relse(xs->bucket);
5132                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
5133                 if (ret) {
5134                         mlog_errno(ret);
5135                         goto out;
5136                 }
5137         }
5138
5139         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
5140                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5141         if (ret < 0) {
5142                 mlog_errno(ret);
5143                 goto out;
5144         }
5145
5146         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
5147         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
5148
5149 out:
5150         return ret;
5151 }
5152
5153 /*
5154  * Truncate the specified xe_off entry in xattr bucket.
5155  * bucket is indicated by header_bh and len is the new length.
5156  * Both the ocfs2_xattr_value_root and the entry will be updated here.
5157  *
5158  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5159  */
5160 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5161                                              struct ocfs2_xattr_bucket *bucket,
5162                                              int xe_off,
5163                                              int len,
5164                                              struct ocfs2_xattr_set_ctxt *ctxt)
5165 {
5166         int ret, offset;
5167         u64 value_blk;
5168         struct ocfs2_xattr_entry *xe;
5169         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5170         size_t blocksize = inode->i_sb->s_blocksize;
5171         struct ocfs2_xattr_value_buf vb = {
5172                 .vb_access = ocfs2_journal_access,
5173         };
5174
5175         xe = &xh->xh_entries[xe_off];
5176
5177         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5178
5179         offset = le16_to_cpu(xe->xe_name_offset) +
5180                  OCFS2_XATTR_SIZE(xe->xe_name_len);
5181
5182         value_blk = offset / blocksize;
5183
5184         /* We don't allow ocfs2_xattr_value to be stored in different block. */
5185         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5186
5187         vb.vb_bh = bucket->bu_bhs[value_blk];
5188         BUG_ON(!vb.vb_bh);
5189
5190         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5191                 (vb.vb_bh->b_data + offset % blocksize);
5192
5193         /*
5194          * From here on out we have to dirty the bucket.  The generic
5195          * value calls only modify one of the bucket's bhs, but we need
5196          * to send the bucket at once.  So if they error, they *could* have
5197          * modified something.  We have to assume they did, and dirty
5198          * the whole bucket.  This leaves us in a consistent state.
5199          */
5200         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
5201              xe_off, (unsigned long long)bucket_blkno(bucket), len);
5202         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5203         if (ret) {
5204                 mlog_errno(ret);
5205                 goto out;
5206         }
5207
5208         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5209                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5210         if (ret) {
5211                 mlog_errno(ret);
5212                 goto out;
5213         }
5214
5215         xe->xe_value_size = cpu_to_le64(len);
5216
5217         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5218
5219 out:
5220         return ret;
5221 }
5222
5223 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
5224                                         struct ocfs2_xattr_search *xs,
5225                                         int len,
5226                                         struct ocfs2_xattr_set_ctxt *ctxt)
5227 {
5228         int ret, offset;
5229         struct ocfs2_xattr_entry *xe = xs->here;
5230         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
5231
5232         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
5233
5234         offset = xe - xh->xh_entries;
5235         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
5236                                                 offset, len, ctxt);
5237         if (ret)
5238                 mlog_errno(ret);
5239
5240         return ret;
5241 }
5242
5243 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
5244                                                 handle_t *handle,
5245                                                 struct ocfs2_xattr_search *xs,
5246                                                 char *val,
5247                                                 int value_len)
5248 {
5249         int ret, offset, block_off;
5250         struct ocfs2_xattr_value_root *xv;
5251         struct ocfs2_xattr_entry *xe = xs->here;
5252         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
5253         void *base;
5254         struct ocfs2_xattr_value_buf vb = {
5255                 .vb_access = ocfs2_journal_access,
5256         };
5257
5258         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
5259
5260         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb, xh,
5261                                                 xe - xh->xh_entries,
5262                                                 &block_off,
5263                                                 &offset);
5264         if (ret) {
5265                 mlog_errno(ret);
5266                 goto out;
5267         }
5268
5269         base = bucket_block(xs->bucket, block_off);
5270         xv = (struct ocfs2_xattr_value_root *)(base + offset +
5271                  OCFS2_XATTR_SIZE(xe->xe_name_len));
5272
5273         vb.vb_xv = xv;
5274         vb.vb_bh = xs->bucket->bu_bhs[block_off];
5275         ret = __ocfs2_xattr_set_value_outside(inode, handle,
5276                                               &vb, val, value_len);
5277         if (ret)
5278                 mlog_errno(ret);
5279 out:
5280         return ret;
5281 }
5282
5283 static int ocfs2_rm_xattr_cluster(struct inode *inode,
5284                                   struct buffer_head *root_bh,
5285                                   u64 blkno,
5286                                   u32 cpos,
5287                                   u32 len,
5288                                   void *para)
5289 {
5290         int ret;
5291         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5292         struct inode *tl_inode = osb->osb_tl_inode;
5293         handle_t *handle;
5294         struct ocfs2_xattr_block *xb =
5295                         (struct ocfs2_xattr_block *)root_bh->b_data;
5296         struct ocfs2_alloc_context *meta_ac = NULL;
5297         struct ocfs2_cached_dealloc_ctxt dealloc;
5298         struct ocfs2_extent_tree et;
5299
5300         ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5301                                           ocfs2_delete_xattr_in_bucket, para);
5302         if (ret) {
5303                 mlog_errno(ret);
5304                 return ret;
5305         }
5306
5307         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5308
5309         ocfs2_init_dealloc_ctxt(&dealloc);
5310
5311         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
5312              cpos, len, (unsigned long long)blkno);
5313
5314         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5315                                                len);
5316
5317         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5318         if (ret) {
5319                 mlog_errno(ret);
5320                 return ret;
5321         }
5322
5323         mutex_lock(&tl_inode->i_mutex);
5324
5325         if (ocfs2_truncate_log_needs_flush(osb)) {
5326                 ret = __ocfs2_flush_truncate_log(osb);
5327                 if (ret < 0) {
5328                         mlog_errno(ret);
5329                         goto out;
5330                 }
5331         }
5332
5333         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5334         if (IS_ERR(handle)) {
5335                 ret = -ENOMEM;
5336                 mlog_errno(ret);
5337                 goto out;
5338         }
5339
5340         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5341                                       OCFS2_JOURNAL_ACCESS_WRITE);
5342         if (ret) {
5343                 mlog_errno(ret);
5344                 goto out_commit;
5345         }
5346
5347         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5348                                   &dealloc);
5349         if (ret) {
5350                 mlog_errno(ret);
5351                 goto out_commit;
5352         }
5353
5354         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5355
5356         ret = ocfs2_journal_dirty(handle, root_bh);
5357         if (ret) {
5358                 mlog_errno(ret);
5359                 goto out_commit;
5360         }
5361
5362         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5363         if (ret)
5364                 mlog_errno(ret);
5365
5366 out_commit:
5367         ocfs2_commit_trans(osb, handle);
5368 out:
5369         ocfs2_schedule_truncate_log_flush(osb, 1);
5370
5371         mutex_unlock(&tl_inode->i_mutex);
5372
5373         if (meta_ac)
5374                 ocfs2_free_alloc_context(meta_ac);
5375
5376         ocfs2_run_deallocs(osb, &dealloc);
5377
5378         return ret;
5379 }
5380
5381 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
5382                                          handle_t *handle,
5383                                          struct ocfs2_xattr_search *xs)
5384 {
5385         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
5386         struct ocfs2_xattr_entry *last = &xh->xh_entries[
5387                                                 le16_to_cpu(xh->xh_count) - 1];
5388         int ret = 0;
5389
5390         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
5391                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5392         if (ret) {
5393                 mlog_errno(ret);
5394                 return;
5395         }
5396
5397         /* Remove the old entry. */
5398         memmove(xs->here, xs->here + 1,
5399                 (void *)last - (void *)xs->here);
5400         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
5401         le16_add_cpu(&xh->xh_count, -1);
5402
5403         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
5404 }
5405
5406 /*
5407  * Set the xattr name/value in the bucket specified in xs.
5408  *
5409  * As the new value in xi may be stored in the bucket or in an outside cluster,
5410  * we divide the whole process into 3 steps:
5411  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
5412  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
5413  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
5414  * 4. If the clusters for the new outside value can't be allocated, we need
5415  *    to free the xattr we allocated in set.
5416  */
5417 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
5418                                      struct ocfs2_xattr_info *xi,
5419                                      struct ocfs2_xattr_search *xs,
5420                                      struct ocfs2_xattr_set_ctxt *ctxt)
5421 {
5422         int ret, local = 1;
5423         size_t value_len;
5424         char *val = (char *)xi->value;
5425         struct ocfs2_xattr_entry *xe = xs->here;
5426         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
5427                                               strlen(xi->name));
5428
5429         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
5430                 /*
5431                  * We need to truncate the xattr storage first.
5432                  *
5433                  * If both the old and new value are stored to
5434                  * outside block, we only need to truncate
5435                  * the storage and then set the value outside.
5436                  *
5437                  * If the new value should be stored within block,
5438                  * we should free all the outside block first and
5439                  * the modification to the xattr block will be done
5440                  * by following steps.
5441                  */
5442                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5443                         value_len = xi->value_len;
5444                 else
5445                         value_len = 0;
5446
5447                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5448                                                            value_len,
5449                                                            ctxt);
5450                 if (ret)
5451                         goto out;
5452
5453                 if (value_len)
5454                         goto set_value_outside;
5455         }
5456
5457         value_len = xi->value_len;
5458         /* So we have to handle the inside block change now. */
5459         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
5460                 /*
5461                  * If the new value will be stored outside of block,
5462                  * initalize a new empty value root and insert it first.
5463                  */
5464                 local = 0;
5465                 xi->value = &def_xv;
5466                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5467         }
5468
5469         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5470                                               name_hash, local);
5471         if (ret) {
5472                 mlog_errno(ret);
5473                 goto out;
5474         }
5475
5476         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5477                 goto out;
5478
5479         /* allocate the space now for the outside block storage. */
5480         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5481                                                    value_len, ctxt);
5482         if (ret) {
5483                 mlog_errno(ret);
5484
5485                 if (xs->not_found) {
5486                         /*
5487                          * We can't allocate enough clusters for outside
5488                          * storage and we have allocated xattr already,
<