Revert "fs: ocfs2: fix possible null-pointer dereferences in ocfs2_xa_prepare_entry()"
[muen/linux.git] / fs / ocfs2 / xattr.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* -*- mode: c; c-basic-offset: 8; -*-
3  * vim: noexpandtab sw=8 ts=8 sts=0:
4  *
5  * xattr.c
6  *
7  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
8  *
9  * CREDITS:
10  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
11  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
12  */
13
14 #include <linux/capability.h>
15 #include <linux/fs.h>
16 #include <linux/types.h>
17 #include <linux/slab.h>
18 #include <linux/highmem.h>
19 #include <linux/pagemap.h>
20 #include <linux/uio.h>
21 #include <linux/sched.h>
22 #include <linux/splice.h>
23 #include <linux/mount.h>
24 #include <linux/writeback.h>
25 #include <linux/falloc.h>
26 #include <linux/sort.h>
27 #include <linux/init.h>
28 #include <linux/module.h>
29 #include <linux/string.h>
30 #include <linux/security.h>
31
32 #include <cluster/masklog.h>
33
34 #include "ocfs2.h"
35 #include "alloc.h"
36 #include "blockcheck.h"
37 #include "dlmglue.h"
38 #include "file.h"
39 #include "symlink.h"
40 #include "sysfile.h"
41 #include "inode.h"
42 #include "journal.h"
43 #include "ocfs2_fs.h"
44 #include "suballoc.h"
45 #include "uptodate.h"
46 #include "buffer_head_io.h"
47 #include "super.h"
48 #include "xattr.h"
49 #include "refcounttree.h"
50 #include "acl.h"
51 #include "ocfs2_trace.h"
52
53 struct ocfs2_xattr_def_value_root {
54         struct ocfs2_xattr_value_root   xv;
55         struct ocfs2_extent_rec         er;
56 };
57
58 struct ocfs2_xattr_bucket {
59         /* The inode these xattrs are associated with */
60         struct inode *bu_inode;
61
62         /* The actual buffers that make up the bucket */
63         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
64
65         /* How many blocks make up one bucket for this filesystem */
66         int bu_blocks;
67 };
68
69 struct ocfs2_xattr_set_ctxt {
70         handle_t *handle;
71         struct ocfs2_alloc_context *meta_ac;
72         struct ocfs2_alloc_context *data_ac;
73         struct ocfs2_cached_dealloc_ctxt dealloc;
74         int set_abort;
75 };
76
77 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
78 #define OCFS2_XATTR_INLINE_SIZE 80
79 #define OCFS2_XATTR_HEADER_GAP  4
80 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
81                                          - sizeof(struct ocfs2_xattr_header) \
82                                          - OCFS2_XATTR_HEADER_GAP)
83 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
84                                          - sizeof(struct ocfs2_xattr_block) \
85                                          - sizeof(struct ocfs2_xattr_header) \
86                                          - OCFS2_XATTR_HEADER_GAP)
87
88 static struct ocfs2_xattr_def_value_root def_xv = {
89         .xv.xr_list.l_count = cpu_to_le16(1),
90 };
91
92 const struct xattr_handler *ocfs2_xattr_handlers[] = {
93         &ocfs2_xattr_user_handler,
94         &posix_acl_access_xattr_handler,
95         &posix_acl_default_xattr_handler,
96         &ocfs2_xattr_trusted_handler,
97         &ocfs2_xattr_security_handler,
98         NULL
99 };
100
101 static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
102         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
103         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
104                                         = &posix_acl_access_xattr_handler,
105         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
106                                         = &posix_acl_default_xattr_handler,
107         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
108         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
109 };
110
111 struct ocfs2_xattr_info {
112         int             xi_name_index;
113         const char      *xi_name;
114         int             xi_name_len;
115         const void      *xi_value;
116         size_t          xi_value_len;
117 };
118
119 struct ocfs2_xattr_search {
120         struct buffer_head *inode_bh;
121         /*
122          * xattr_bh point to the block buffer head which has extended attribute
123          * when extended attribute in inode, xattr_bh is equal to inode_bh.
124          */
125         struct buffer_head *xattr_bh;
126         struct ocfs2_xattr_header *header;
127         struct ocfs2_xattr_bucket *bucket;
128         void *base;
129         void *end;
130         struct ocfs2_xattr_entry *here;
131         int not_found;
132 };
133
134 /* Operations on struct ocfs2_xa_entry */
135 struct ocfs2_xa_loc;
136 struct ocfs2_xa_loc_operations {
137         /*
138          * Journal functions
139          */
140         int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
141                                   int type);
142         void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
143
144         /*
145          * Return a pointer to the appropriate buffer in loc->xl_storage
146          * at the given offset from loc->xl_header.
147          */
148         void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
149
150         /* Can we reuse the existing entry for the new value? */
151         int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
152                              struct ocfs2_xattr_info *xi);
153
154         /* How much space is needed for the new value? */
155         int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
156                                struct ocfs2_xattr_info *xi);
157
158         /*
159          * Return the offset of the first name+value pair.  This is
160          * the start of our downward-filling free space.
161          */
162         int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
163
164         /*
165          * Remove the name+value at this location.  Do whatever is
166          * appropriate with the remaining name+value pairs.
167          */
168         void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
169
170         /* Fill xl_entry with a new entry */
171         void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
172
173         /* Add name+value storage to an entry */
174         void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
175
176         /*
177          * Initialize the value buf's access and bh fields for this entry.
178          * ocfs2_xa_fill_value_buf() will handle the xv pointer.
179          */
180         void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
181                                    struct ocfs2_xattr_value_buf *vb);
182 };
183
184 /*
185  * Describes an xattr entry location.  This is a memory structure
186  * tracking the on-disk structure.
187  */
188 struct ocfs2_xa_loc {
189         /* This xattr belongs to this inode */
190         struct inode *xl_inode;
191
192         /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
193         struct ocfs2_xattr_header *xl_header;
194
195         /* Bytes from xl_header to the end of the storage */
196         int xl_size;
197
198         /*
199          * The ocfs2_xattr_entry this location describes.  If this is
200          * NULL, this location describes the on-disk structure where it
201          * would have been.
202          */
203         struct ocfs2_xattr_entry *xl_entry;
204
205         /*
206          * Internal housekeeping
207          */
208
209         /* Buffer(s) containing this entry */
210         void *xl_storage;
211
212         /* Operations on the storage backing this location */
213         const struct ocfs2_xa_loc_operations *xl_ops;
214 };
215
216 /*
217  * Convenience functions to calculate how much space is needed for a
218  * given name+value pair
219  */
220 static int namevalue_size(int name_len, uint64_t value_len)
221 {
222         if (value_len > OCFS2_XATTR_INLINE_SIZE)
223                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
224         else
225                 return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
226 }
227
228 static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
229 {
230         return namevalue_size(xi->xi_name_len, xi->xi_value_len);
231 }
232
233 static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
234 {
235         u64 value_len = le64_to_cpu(xe->xe_value_size);
236
237         BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
238                ocfs2_xattr_is_local(xe));
239         return namevalue_size(xe->xe_name_len, value_len);
240 }
241
242
243 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
244                                              struct ocfs2_xattr_header *xh,
245                                              int index,
246                                              int *block_off,
247                                              int *new_offset);
248
249 static int ocfs2_xattr_block_find(struct inode *inode,
250                                   int name_index,
251                                   const char *name,
252                                   struct ocfs2_xattr_search *xs);
253 static int ocfs2_xattr_index_block_find(struct inode *inode,
254                                         struct buffer_head *root_bh,
255                                         int name_index,
256                                         const char *name,
257                                         struct ocfs2_xattr_search *xs);
258
259 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
260                                         struct buffer_head *blk_bh,
261                                         char *buffer,
262                                         size_t buffer_size);
263
264 static int ocfs2_xattr_create_index_block(struct inode *inode,
265                                           struct ocfs2_xattr_search *xs,
266                                           struct ocfs2_xattr_set_ctxt *ctxt);
267
268 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
269                                              struct ocfs2_xattr_info *xi,
270                                              struct ocfs2_xattr_search *xs,
271                                              struct ocfs2_xattr_set_ctxt *ctxt);
272
273 typedef int (xattr_tree_rec_func)(struct inode *inode,
274                                   struct buffer_head *root_bh,
275                                   u64 blkno, u32 cpos, u32 len, void *para);
276 static int ocfs2_iterate_xattr_index_block(struct inode *inode,
277                                            struct buffer_head *root_bh,
278                                            xattr_tree_rec_func *rec_func,
279                                            void *para);
280 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
281                                         struct ocfs2_xattr_bucket *bucket,
282                                         void *para);
283 static int ocfs2_rm_xattr_cluster(struct inode *inode,
284                                   struct buffer_head *root_bh,
285                                   u64 blkno,
286                                   u32 cpos,
287                                   u32 len,
288                                   void *para);
289
290 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
291                                   u64 src_blk, u64 last_blk, u64 to_blk,
292                                   unsigned int start_bucket,
293                                   u32 *first_hash);
294 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
295                                         struct ocfs2_dinode *di,
296                                         struct ocfs2_xattr_info *xi,
297                                         struct ocfs2_xattr_search *xis,
298                                         struct ocfs2_xattr_search *xbs,
299                                         struct ocfs2_refcount_tree **ref_tree,
300                                         int *meta_need,
301                                         int *credits);
302 static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
303                                            struct ocfs2_xattr_bucket *bucket,
304                                            int offset,
305                                            struct ocfs2_xattr_value_root **xv,
306                                            struct buffer_head **bh);
307
308 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
309 {
310         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
311 }
312
313 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
314 {
315         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
316 }
317
318 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
319 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
320 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
321
322 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
323 {
324         struct ocfs2_xattr_bucket *bucket;
325         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
326
327         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
328
329         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
330         if (bucket) {
331                 bucket->bu_inode = inode;
332                 bucket->bu_blocks = blks;
333         }
334
335         return bucket;
336 }
337
338 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
339 {
340         int i;
341
342         for (i = 0; i < bucket->bu_blocks; i++) {
343                 brelse(bucket->bu_bhs[i]);
344                 bucket->bu_bhs[i] = NULL;
345         }
346 }
347
348 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
349 {
350         if (bucket) {
351                 ocfs2_xattr_bucket_relse(bucket);
352                 bucket->bu_inode = NULL;
353                 kfree(bucket);
354         }
355 }
356
357 /*
358  * A bucket that has never been written to disk doesn't need to be
359  * read.  We just need the buffer_heads.  Don't call this for
360  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
361  * them fully.
362  */
363 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
364                                    u64 xb_blkno, int new)
365 {
366         int i, rc = 0;
367
368         for (i = 0; i < bucket->bu_blocks; i++) {
369                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
370                                               xb_blkno + i);
371                 if (!bucket->bu_bhs[i]) {
372                         rc = -ENOMEM;
373                         mlog_errno(rc);
374                         break;
375                 }
376
377                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
378                                            bucket->bu_bhs[i])) {
379                         if (new)
380                                 ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
381                                                               bucket->bu_bhs[i]);
382                         else {
383                                 set_buffer_uptodate(bucket->bu_bhs[i]);
384                                 ocfs2_set_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
385                                                           bucket->bu_bhs[i]);
386                         }
387                 }
388         }
389
390         if (rc)
391                 ocfs2_xattr_bucket_relse(bucket);
392         return rc;
393 }
394
395 /* Read the xattr bucket at xb_blkno */
396 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
397                                    u64 xb_blkno)
398 {
399         int rc;
400
401         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
402                                bucket->bu_blocks, bucket->bu_bhs, 0,
403                                NULL);
404         if (!rc) {
405                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
406                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
407                                                  bucket->bu_bhs,
408                                                  bucket->bu_blocks,
409                                                  &bucket_xh(bucket)->xh_check);
410                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
411                 if (rc)
412                         mlog_errno(rc);
413         }
414
415         if (rc)
416                 ocfs2_xattr_bucket_relse(bucket);
417         return rc;
418 }
419
420 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
421                                              struct ocfs2_xattr_bucket *bucket,
422                                              int type)
423 {
424         int i, rc = 0;
425
426         for (i = 0; i < bucket->bu_blocks; i++) {
427                 rc = ocfs2_journal_access(handle,
428                                           INODE_CACHE(bucket->bu_inode),
429                                           bucket->bu_bhs[i], type);
430                 if (rc) {
431                         mlog_errno(rc);
432                         break;
433                 }
434         }
435
436         return rc;
437 }
438
439 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
440                                              struct ocfs2_xattr_bucket *bucket)
441 {
442         int i;
443
444         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
445         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
446                                    bucket->bu_bhs, bucket->bu_blocks,
447                                    &bucket_xh(bucket)->xh_check);
448         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
449
450         for (i = 0; i < bucket->bu_blocks; i++)
451                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
452 }
453
454 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
455                                          struct ocfs2_xattr_bucket *src)
456 {
457         int i;
458         int blocksize = src->bu_inode->i_sb->s_blocksize;
459
460         BUG_ON(dest->bu_blocks != src->bu_blocks);
461         BUG_ON(dest->bu_inode != src->bu_inode);
462
463         for (i = 0; i < src->bu_blocks; i++) {
464                 memcpy(bucket_block(dest, i), bucket_block(src, i),
465                        blocksize);
466         }
467 }
468
469 static int ocfs2_validate_xattr_block(struct super_block *sb,
470                                       struct buffer_head *bh)
471 {
472         int rc;
473         struct ocfs2_xattr_block *xb =
474                 (struct ocfs2_xattr_block *)bh->b_data;
475
476         trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
477
478         BUG_ON(!buffer_uptodate(bh));
479
480         /*
481          * If the ecc fails, we return the error but otherwise
482          * leave the filesystem running.  We know any error is
483          * local to this block.
484          */
485         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
486         if (rc)
487                 return rc;
488
489         /*
490          * Errors after here are fatal
491          */
492
493         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
494                 return ocfs2_error(sb,
495                                    "Extended attribute block #%llu has bad signature %.*s\n",
496                                    (unsigned long long)bh->b_blocknr, 7,
497                                    xb->xb_signature);
498         }
499
500         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
501                 return ocfs2_error(sb,
502                                    "Extended attribute block #%llu has an invalid xb_blkno of %llu\n",
503                                    (unsigned long long)bh->b_blocknr,
504                                    (unsigned long long)le64_to_cpu(xb->xb_blkno));
505         }
506
507         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
508                 return ocfs2_error(sb,
509                                    "Extended attribute block #%llu has an invalid xb_fs_generation of #%u\n",
510                                    (unsigned long long)bh->b_blocknr,
511                                    le32_to_cpu(xb->xb_fs_generation));
512         }
513
514         return 0;
515 }
516
517 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
518                                   struct buffer_head **bh)
519 {
520         int rc;
521         struct buffer_head *tmp = *bh;
522
523         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
524                               ocfs2_validate_xattr_block);
525
526         /* If ocfs2_read_block() got us a new bh, pass it up. */
527         if (!rc && !*bh)
528                 *bh = tmp;
529
530         return rc;
531 }
532
533 static inline const char *ocfs2_xattr_prefix(int name_index)
534 {
535         const struct xattr_handler *handler = NULL;
536
537         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
538                 handler = ocfs2_xattr_handler_map[name_index];
539         return handler ? xattr_prefix(handler) : NULL;
540 }
541
542 static u32 ocfs2_xattr_name_hash(struct inode *inode,
543                                  const char *name,
544                                  int name_len)
545 {
546         /* Get hash value of uuid from super block */
547         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
548         int i;
549
550         /* hash extended attribute name */
551         for (i = 0; i < name_len; i++) {
552                 hash = (hash << OCFS2_HASH_SHIFT) ^
553                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
554                        *name++;
555         }
556
557         return hash;
558 }
559
560 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
561 {
562         return namevalue_size(name_len, value_len) +
563                 sizeof(struct ocfs2_xattr_entry);
564 }
565
566 static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
567 {
568         return namevalue_size_xi(xi) +
569                 sizeof(struct ocfs2_xattr_entry);
570 }
571
572 static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
573 {
574         return namevalue_size_xe(xe) +
575                 sizeof(struct ocfs2_xattr_entry);
576 }
577
578 int ocfs2_calc_security_init(struct inode *dir,
579                              struct ocfs2_security_xattr_info *si,
580                              int *want_clusters,
581                              int *xattr_credits,
582                              struct ocfs2_alloc_context **xattr_ac)
583 {
584         int ret = 0;
585         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
586         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
587                                                  si->value_len);
588
589         /*
590          * The max space of security xattr taken inline is
591          * 256(name) + 80(value) + 16(entry) = 352 bytes,
592          * So reserve one metadata block for it is ok.
593          */
594         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
595             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
596                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
597                 if (ret) {
598                         mlog_errno(ret);
599                         return ret;
600                 }
601                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
602         }
603
604         /* reserve clusters for xattr value which will be set in B tree*/
605         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
606                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
607                                                             si->value_len);
608
609                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
610                                                            new_clusters);
611                 *want_clusters += new_clusters;
612         }
613         return ret;
614 }
615
616 int ocfs2_calc_xattr_init(struct inode *dir,
617                           struct buffer_head *dir_bh,
618                           umode_t mode,
619                           struct ocfs2_security_xattr_info *si,
620                           int *want_clusters,
621                           int *xattr_credits,
622                           int *want_meta)
623 {
624         int ret = 0;
625         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
626         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
627
628         if (si->enable)
629                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
630                                                      si->value_len);
631
632         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
633                 down_read(&OCFS2_I(dir)->ip_xattr_sem);
634                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
635                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
636                                         "", NULL, 0);
637                 up_read(&OCFS2_I(dir)->ip_xattr_sem);
638                 if (acl_len > 0) {
639                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
640                         if (S_ISDIR(mode))
641                                 a_size <<= 1;
642                 } else if (acl_len != 0 && acl_len != -ENODATA) {
643                         ret = acl_len;
644                         mlog_errno(ret);
645                         return ret;
646                 }
647         }
648
649         if (!(s_size + a_size))
650                 return ret;
651
652         /*
653          * The max space of security xattr taken inline is
654          * 256(name) + 80(value) + 16(entry) = 352 bytes,
655          * The max space of acl xattr taken inline is
656          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
657          * when blocksize = 512, may reserve one more cluser for
658          * xattr bucket, otherwise reserve one metadata block
659          * for them is ok.
660          * If this is a new directory with inline data,
661          * we choose to reserve the entire inline area for
662          * directory contents and force an external xattr block.
663          */
664         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
665             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
666             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
667                 *want_meta = *want_meta + 1;
668                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
669         }
670
671         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
672             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
673                 *want_clusters += 1;
674                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
675         }
676
677         /*
678          * reserve credits and clusters for xattrs which has large value
679          * and have to be set outside
680          */
681         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
682                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
683                                                         si->value_len);
684                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
685                                                            new_clusters);
686                 *want_clusters += new_clusters;
687         }
688         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
689             acl_len > OCFS2_XATTR_INLINE_SIZE) {
690                 /* for directory, it has DEFAULT and ACCESS two types of acls */
691                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
692                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
693                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
694                                                            new_clusters);
695                 *want_clusters += new_clusters;
696         }
697
698         return ret;
699 }
700
701 static int ocfs2_xattr_extend_allocation(struct inode *inode,
702                                          u32 clusters_to_add,
703                                          struct ocfs2_xattr_value_buf *vb,
704                                          struct ocfs2_xattr_set_ctxt *ctxt)
705 {
706         int status = 0, credits;
707         handle_t *handle = ctxt->handle;
708         enum ocfs2_alloc_restarted why;
709         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
710         struct ocfs2_extent_tree et;
711
712         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
713
714         while (clusters_to_add) {
715                 trace_ocfs2_xattr_extend_allocation(clusters_to_add);
716
717                 status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
718                                        OCFS2_JOURNAL_ACCESS_WRITE);
719                 if (status < 0) {
720                         mlog_errno(status);
721                         break;
722                 }
723
724                 prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
725                 status = ocfs2_add_clusters_in_btree(handle,
726                                                      &et,
727                                                      &logical_start,
728                                                      clusters_to_add,
729                                                      0,
730                                                      ctxt->data_ac,
731                                                      ctxt->meta_ac,
732                                                      &why);
733                 if ((status < 0) && (status != -EAGAIN)) {
734                         if (status != -ENOSPC)
735                                 mlog_errno(status);
736                         break;
737                 }
738
739                 ocfs2_journal_dirty(handle, vb->vb_bh);
740
741                 clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
742                                          prev_clusters;
743
744                 if (why != RESTART_NONE && clusters_to_add) {
745                         /*
746                          * We can only fail in case the alloc file doesn't give
747                          * up enough clusters.
748                          */
749                         BUG_ON(why == RESTART_META);
750
751                         credits = ocfs2_calc_extend_credits(inode->i_sb,
752                                                             &vb->vb_xv->xr_list);
753                         status = ocfs2_extend_trans(handle, credits);
754                         if (status < 0) {
755                                 status = -ENOMEM;
756                                 mlog_errno(status);
757                                 break;
758                         }
759                 }
760         }
761
762         return status;
763 }
764
765 static int __ocfs2_remove_xattr_range(struct inode *inode,
766                                       struct ocfs2_xattr_value_buf *vb,
767                                       u32 cpos, u32 phys_cpos, u32 len,
768                                       unsigned int ext_flags,
769                                       struct ocfs2_xattr_set_ctxt *ctxt)
770 {
771         int ret;
772         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
773         handle_t *handle = ctxt->handle;
774         struct ocfs2_extent_tree et;
775
776         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
777
778         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
779                             OCFS2_JOURNAL_ACCESS_WRITE);
780         if (ret) {
781                 mlog_errno(ret);
782                 goto out;
783         }
784
785         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
786                                   &ctxt->dealloc);
787         if (ret) {
788                 mlog_errno(ret);
789                 goto out;
790         }
791
792         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
793         ocfs2_journal_dirty(handle, vb->vb_bh);
794
795         if (ext_flags & OCFS2_EXT_REFCOUNTED)
796                 ret = ocfs2_decrease_refcount(inode, handle,
797                                         ocfs2_blocks_to_clusters(inode->i_sb,
798                                                                  phys_blkno),
799                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
800         else
801                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
802                                                   phys_blkno, len);
803         if (ret)
804                 mlog_errno(ret);
805
806 out:
807         return ret;
808 }
809
810 static int ocfs2_xattr_shrink_size(struct inode *inode,
811                                    u32 old_clusters,
812                                    u32 new_clusters,
813                                    struct ocfs2_xattr_value_buf *vb,
814                                    struct ocfs2_xattr_set_ctxt *ctxt)
815 {
816         int ret = 0;
817         unsigned int ext_flags;
818         u32 trunc_len, cpos, phys_cpos, alloc_size;
819         u64 block;
820
821         if (old_clusters <= new_clusters)
822                 return 0;
823
824         cpos = new_clusters;
825         trunc_len = old_clusters - new_clusters;
826         while (trunc_len) {
827                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
828                                                &alloc_size,
829                                                &vb->vb_xv->xr_list, &ext_flags);
830                 if (ret) {
831                         mlog_errno(ret);
832                         goto out;
833                 }
834
835                 if (alloc_size > trunc_len)
836                         alloc_size = trunc_len;
837
838                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
839                                                  phys_cpos, alloc_size,
840                                                  ext_flags, ctxt);
841                 if (ret) {
842                         mlog_errno(ret);
843                         goto out;
844                 }
845
846                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
847                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
848                                                        block, alloc_size);
849                 cpos += alloc_size;
850                 trunc_len -= alloc_size;
851         }
852
853 out:
854         return ret;
855 }
856
857 static int ocfs2_xattr_value_truncate(struct inode *inode,
858                                       struct ocfs2_xattr_value_buf *vb,
859                                       int len,
860                                       struct ocfs2_xattr_set_ctxt *ctxt)
861 {
862         int ret;
863         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
864         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
865
866         if (new_clusters == old_clusters)
867                 return 0;
868
869         if (new_clusters > old_clusters)
870                 ret = ocfs2_xattr_extend_allocation(inode,
871                                                     new_clusters - old_clusters,
872                                                     vb, ctxt);
873         else
874                 ret = ocfs2_xattr_shrink_size(inode,
875                                               old_clusters, new_clusters,
876                                               vb, ctxt);
877
878         return ret;
879 }
880
881 static int ocfs2_xattr_list_entry(struct super_block *sb,
882                                   char *buffer, size_t size,
883                                   size_t *result, int type,
884                                   const char *name, int name_len)
885 {
886         char *p = buffer + *result;
887         const char *prefix;
888         int prefix_len;
889         int total_len;
890
891         switch(type) {
892         case OCFS2_XATTR_INDEX_USER:
893                 if (OCFS2_SB(sb)->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
894                         return 0;
895                 break;
896
897         case OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS:
898         case OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT:
899                 if (!(sb->s_flags & SB_POSIXACL))
900                         return 0;
901                 break;
902
903         case OCFS2_XATTR_INDEX_TRUSTED:
904                 if (!capable(CAP_SYS_ADMIN))
905                         return 0;
906                 break;
907         }
908
909         prefix = ocfs2_xattr_prefix(type);
910         if (!prefix)
911                 return 0;
912         prefix_len = strlen(prefix);
913         total_len = prefix_len + name_len + 1;
914         *result += total_len;
915
916         /* we are just looking for how big our buffer needs to be */
917         if (!size)
918                 return 0;
919
920         if (*result > size)
921                 return -ERANGE;
922
923         memcpy(p, prefix, prefix_len);
924         memcpy(p + prefix_len, name, name_len);
925         p[prefix_len + name_len] = '\0';
926
927         return 0;
928 }
929
930 static int ocfs2_xattr_list_entries(struct inode *inode,
931                                     struct ocfs2_xattr_header *header,
932                                     char *buffer, size_t buffer_size)
933 {
934         size_t result = 0;
935         int i, type, ret;
936         const char *name;
937
938         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
939                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
940                 type = ocfs2_xattr_get_type(entry);
941                 name = (const char *)header +
942                         le16_to_cpu(entry->xe_name_offset);
943
944                 ret = ocfs2_xattr_list_entry(inode->i_sb,
945                                              buffer, buffer_size,
946                                              &result, type, name,
947                                              entry->xe_name_len);
948                 if (ret)
949                         return ret;
950         }
951
952         return result;
953 }
954
955 int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
956                                          struct ocfs2_dinode *di)
957 {
958         struct ocfs2_xattr_header *xh;
959         int i;
960
961         xh = (struct ocfs2_xattr_header *)
962                  ((void *)di + inode->i_sb->s_blocksize -
963                  le16_to_cpu(di->i_xattr_inline_size));
964
965         for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
966                 if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
967                         return 1;
968
969         return 0;
970 }
971
972 static int ocfs2_xattr_ibody_list(struct inode *inode,
973                                   struct ocfs2_dinode *di,
974                                   char *buffer,
975                                   size_t buffer_size)
976 {
977         struct ocfs2_xattr_header *header = NULL;
978         struct ocfs2_inode_info *oi = OCFS2_I(inode);
979         int ret = 0;
980
981         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
982                 return ret;
983
984         header = (struct ocfs2_xattr_header *)
985                  ((void *)di + inode->i_sb->s_blocksize -
986                  le16_to_cpu(di->i_xattr_inline_size));
987
988         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
989
990         return ret;
991 }
992
993 static int ocfs2_xattr_block_list(struct inode *inode,
994                                   struct ocfs2_dinode *di,
995                                   char *buffer,
996                                   size_t buffer_size)
997 {
998         struct buffer_head *blk_bh = NULL;
999         struct ocfs2_xattr_block *xb;
1000         int ret = 0;
1001
1002         if (!di->i_xattr_loc)
1003                 return ret;
1004
1005         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
1006                                      &blk_bh);
1007         if (ret < 0) {
1008                 mlog_errno(ret);
1009                 return ret;
1010         }
1011
1012         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1013         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1014                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
1015                 ret = ocfs2_xattr_list_entries(inode, header,
1016                                                buffer, buffer_size);
1017         } else
1018                 ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1019                                                    buffer, buffer_size);
1020
1021         brelse(blk_bh);
1022
1023         return ret;
1024 }
1025
1026 ssize_t ocfs2_listxattr(struct dentry *dentry,
1027                         char *buffer,
1028                         size_t size)
1029 {
1030         int ret = 0, i_ret = 0, b_ret = 0;
1031         struct buffer_head *di_bh = NULL;
1032         struct ocfs2_dinode *di = NULL;
1033         struct ocfs2_inode_info *oi = OCFS2_I(d_inode(dentry));
1034
1035         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1036                 return -EOPNOTSUPP;
1037
1038         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1039                 return ret;
1040
1041         ret = ocfs2_inode_lock(d_inode(dentry), &di_bh, 0);
1042         if (ret < 0) {
1043                 mlog_errno(ret);
1044                 return ret;
1045         }
1046
1047         di = (struct ocfs2_dinode *)di_bh->b_data;
1048
1049         down_read(&oi->ip_xattr_sem);
1050         i_ret = ocfs2_xattr_ibody_list(d_inode(dentry), di, buffer, size);
1051         if (i_ret < 0)
1052                 b_ret = 0;
1053         else {
1054                 if (buffer) {
1055                         buffer += i_ret;
1056                         size -= i_ret;
1057                 }
1058                 b_ret = ocfs2_xattr_block_list(d_inode(dentry), di,
1059                                                buffer, size);
1060                 if (b_ret < 0)
1061                         i_ret = 0;
1062         }
1063         up_read(&oi->ip_xattr_sem);
1064         ocfs2_inode_unlock(d_inode(dentry), 0);
1065
1066         brelse(di_bh);
1067
1068         return i_ret + b_ret;
1069 }
1070
1071 static int ocfs2_xattr_find_entry(int name_index,
1072                                   const char *name,
1073                                   struct ocfs2_xattr_search *xs)
1074 {
1075         struct ocfs2_xattr_entry *entry;
1076         size_t name_len;
1077         int i, cmp = 1;
1078
1079         if (name == NULL)
1080                 return -EINVAL;
1081
1082         name_len = strlen(name);
1083         entry = xs->here;
1084         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1085                 cmp = name_index - ocfs2_xattr_get_type(entry);
1086                 if (!cmp)
1087                         cmp = name_len - entry->xe_name_len;
1088                 if (!cmp)
1089                         cmp = memcmp(name, (xs->base +
1090                                      le16_to_cpu(entry->xe_name_offset)),
1091                                      name_len);
1092                 if (cmp == 0)
1093                         break;
1094                 entry += 1;
1095         }
1096         xs->here = entry;
1097
1098         return cmp ? -ENODATA : 0;
1099 }
1100
1101 static int ocfs2_xattr_get_value_outside(struct inode *inode,
1102                                          struct ocfs2_xattr_value_root *xv,
1103                                          void *buffer,
1104                                          size_t len)
1105 {
1106         u32 cpos, p_cluster, num_clusters, bpc, clusters;
1107         u64 blkno;
1108         int i, ret = 0;
1109         size_t cplen, blocksize;
1110         struct buffer_head *bh = NULL;
1111         struct ocfs2_extent_list *el;
1112
1113         el = &xv->xr_list;
1114         clusters = le32_to_cpu(xv->xr_clusters);
1115         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1116         blocksize = inode->i_sb->s_blocksize;
1117
1118         cpos = 0;
1119         while (cpos < clusters) {
1120                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1121                                                &num_clusters, el, NULL);
1122                 if (ret) {
1123                         mlog_errno(ret);
1124                         goto out;
1125                 }
1126
1127                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1128                 /* Copy ocfs2_xattr_value */
1129                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1130                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1131                                                &bh, NULL);
1132                         if (ret) {
1133                                 mlog_errno(ret);
1134                                 goto out;
1135                         }
1136
1137                         cplen = len >= blocksize ? blocksize : len;
1138                         memcpy(buffer, bh->b_data, cplen);
1139                         len -= cplen;
1140                         buffer += cplen;
1141
1142                         brelse(bh);
1143                         bh = NULL;
1144                         if (len == 0)
1145                                 break;
1146                 }
1147                 cpos += num_clusters;
1148         }
1149 out:
1150         return ret;
1151 }
1152
1153 static int ocfs2_xattr_ibody_get(struct inode *inode,
1154                                  int name_index,
1155                                  const char *name,
1156                                  void *buffer,
1157                                  size_t buffer_size,
1158                                  struct ocfs2_xattr_search *xs)
1159 {
1160         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1161         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1162         struct ocfs2_xattr_value_root *xv;
1163         size_t size;
1164         int ret = 0;
1165
1166         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1167                 return -ENODATA;
1168
1169         xs->end = (void *)di + inode->i_sb->s_blocksize;
1170         xs->header = (struct ocfs2_xattr_header *)
1171                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1172         xs->base = (void *)xs->header;
1173         xs->here = xs->header->xh_entries;
1174
1175         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1176         if (ret)
1177                 return ret;
1178         size = le64_to_cpu(xs->here->xe_value_size);
1179         if (buffer) {
1180                 if (size > buffer_size)
1181                         return -ERANGE;
1182                 if (ocfs2_xattr_is_local(xs->here)) {
1183                         memcpy(buffer, (void *)xs->base +
1184                                le16_to_cpu(xs->here->xe_name_offset) +
1185                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1186                 } else {
1187                         xv = (struct ocfs2_xattr_value_root *)
1188                                 (xs->base + le16_to_cpu(
1189                                  xs->here->xe_name_offset) +
1190                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1191                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1192                                                             buffer, size);
1193                         if (ret < 0) {
1194                                 mlog_errno(ret);
1195                                 return ret;
1196                         }
1197                 }
1198         }
1199
1200         return size;
1201 }
1202
1203 static int ocfs2_xattr_block_get(struct inode *inode,
1204                                  int name_index,
1205                                  const char *name,
1206                                  void *buffer,
1207                                  size_t buffer_size,
1208                                  struct ocfs2_xattr_search *xs)
1209 {
1210         struct ocfs2_xattr_block *xb;
1211         struct ocfs2_xattr_value_root *xv;
1212         size_t size;
1213         int ret = -ENODATA, name_offset, name_len, i;
1214         int uninitialized_var(block_off);
1215
1216         xs->bucket = ocfs2_xattr_bucket_new(inode);
1217         if (!xs->bucket) {
1218                 ret = -ENOMEM;
1219                 mlog_errno(ret);
1220                 goto cleanup;
1221         }
1222
1223         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1224         if (ret) {
1225                 mlog_errno(ret);
1226                 goto cleanup;
1227         }
1228
1229         if (xs->not_found) {
1230                 ret = -ENODATA;
1231                 goto cleanup;
1232         }
1233
1234         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1235         size = le64_to_cpu(xs->here->xe_value_size);
1236         if (buffer) {
1237                 ret = -ERANGE;
1238                 if (size > buffer_size)
1239                         goto cleanup;
1240
1241                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1242                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1243                 i = xs->here - xs->header->xh_entries;
1244
1245                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1246                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1247                                                                 bucket_xh(xs->bucket),
1248                                                                 i,
1249                                                                 &block_off,
1250                                                                 &name_offset);
1251                         if (ret) {
1252                                 mlog_errno(ret);
1253                                 goto cleanup;
1254                         }
1255                         xs->base = bucket_block(xs->bucket, block_off);
1256                 }
1257                 if (ocfs2_xattr_is_local(xs->here)) {
1258                         memcpy(buffer, (void *)xs->base +
1259                                name_offset + name_len, size);
1260                 } else {
1261                         xv = (struct ocfs2_xattr_value_root *)
1262                                 (xs->base + name_offset + name_len);
1263                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1264                                                             buffer, size);
1265                         if (ret < 0) {
1266                                 mlog_errno(ret);
1267                                 goto cleanup;
1268                         }
1269                 }
1270         }
1271         ret = size;
1272 cleanup:
1273         ocfs2_xattr_bucket_free(xs->bucket);
1274
1275         brelse(xs->xattr_bh);
1276         xs->xattr_bh = NULL;
1277         return ret;
1278 }
1279
1280 int ocfs2_xattr_get_nolock(struct inode *inode,
1281                            struct buffer_head *di_bh,
1282                            int name_index,
1283                            const char *name,
1284                            void *buffer,
1285                            size_t buffer_size)
1286 {
1287         int ret;
1288         struct ocfs2_dinode *di = NULL;
1289         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1290         struct ocfs2_xattr_search xis = {
1291                 .not_found = -ENODATA,
1292         };
1293         struct ocfs2_xattr_search xbs = {
1294                 .not_found = -ENODATA,
1295         };
1296
1297         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1298                 return -EOPNOTSUPP;
1299
1300         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1301                 return -ENODATA;
1302
1303         xis.inode_bh = xbs.inode_bh = di_bh;
1304         di = (struct ocfs2_dinode *)di_bh->b_data;
1305
1306         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1307                                     buffer_size, &xis);
1308         if (ret == -ENODATA && di->i_xattr_loc)
1309                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1310                                             buffer_size, &xbs);
1311
1312         return ret;
1313 }
1314
1315 /* ocfs2_xattr_get()
1316  *
1317  * Copy an extended attribute into the buffer provided.
1318  * Buffer is NULL to compute the size of buffer required.
1319  */
1320 static int ocfs2_xattr_get(struct inode *inode,
1321                            int name_index,
1322                            const char *name,
1323                            void *buffer,
1324                            size_t buffer_size)
1325 {
1326         int ret, had_lock;
1327         struct buffer_head *di_bh = NULL;
1328         struct ocfs2_lock_holder oh;
1329
1330         had_lock = ocfs2_inode_lock_tracker(inode, &di_bh, 0, &oh);
1331         if (had_lock < 0) {
1332                 mlog_errno(had_lock);
1333                 return had_lock;
1334         }
1335         down_read(&OCFS2_I(inode)->ip_xattr_sem);
1336         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1337                                      name, buffer, buffer_size);
1338         up_read(&OCFS2_I(inode)->ip_xattr_sem);
1339
1340         ocfs2_inode_unlock_tracker(inode, 0, &oh, had_lock);
1341
1342         brelse(di_bh);
1343
1344         return ret;
1345 }
1346
1347 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1348                                            handle_t *handle,
1349                                            struct ocfs2_xattr_value_buf *vb,
1350                                            const void *value,
1351                                            int value_len)
1352 {
1353         int ret = 0, i, cp_len;
1354         u16 blocksize = inode->i_sb->s_blocksize;
1355         u32 p_cluster, num_clusters;
1356         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1357         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1358         u64 blkno;
1359         struct buffer_head *bh = NULL;
1360         unsigned int ext_flags;
1361         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1362
1363         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1364
1365         while (cpos < clusters) {
1366                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1367                                                &num_clusters, &xv->xr_list,
1368                                                &ext_flags);
1369                 if (ret) {
1370                         mlog_errno(ret);
1371                         goto out;
1372                 }
1373
1374                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1375
1376                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1377
1378                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1379                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1380                                                &bh, NULL);
1381                         if (ret) {
1382                                 mlog_errno(ret);
1383                                 goto out;
1384                         }
1385
1386                         ret = ocfs2_journal_access(handle,
1387                                                    INODE_CACHE(inode),
1388                                                    bh,
1389                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1390                         if (ret < 0) {
1391                                 mlog_errno(ret);
1392                                 goto out;
1393                         }
1394
1395                         cp_len = value_len > blocksize ? blocksize : value_len;
1396                         memcpy(bh->b_data, value, cp_len);
1397                         value_len -= cp_len;
1398                         value += cp_len;
1399                         if (cp_len < blocksize)
1400                                 memset(bh->b_data + cp_len, 0,
1401                                        blocksize - cp_len);
1402
1403                         ocfs2_journal_dirty(handle, bh);
1404                         brelse(bh);
1405                         bh = NULL;
1406
1407                         /*
1408                          * XXX: do we need to empty all the following
1409                          * blocks in this cluster?
1410                          */
1411                         if (!value_len)
1412                                 break;
1413                 }
1414                 cpos += num_clusters;
1415         }
1416 out:
1417         brelse(bh);
1418
1419         return ret;
1420 }
1421
1422 static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1423                                        int num_entries)
1424 {
1425         int free_space;
1426
1427         if (!needed_space)
1428                 return 0;
1429
1430         free_space = free_start -
1431                 sizeof(struct ocfs2_xattr_header) -
1432                 (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1433                 OCFS2_XATTR_HEADER_GAP;
1434         if (free_space < 0)
1435                 return -EIO;
1436         if (free_space < needed_space)
1437                 return -ENOSPC;
1438
1439         return 0;
1440 }
1441
1442 static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1443                                    int type)
1444 {
1445         return loc->xl_ops->xlo_journal_access(handle, loc, type);
1446 }
1447
1448 static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1449 {
1450         loc->xl_ops->xlo_journal_dirty(handle, loc);
1451 }
1452
1453 /* Give a pointer into the storage for the given offset */
1454 static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1455 {
1456         BUG_ON(offset >= loc->xl_size);
1457         return loc->xl_ops->xlo_offset_pointer(loc, offset);
1458 }
1459
1460 /*
1461  * Wipe the name+value pair and allow the storage to reclaim it.  This
1462  * must be followed by either removal of the entry or a call to
1463  * ocfs2_xa_add_namevalue().
1464  */
1465 static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1466 {
1467         loc->xl_ops->xlo_wipe_namevalue(loc);
1468 }
1469
1470 /*
1471  * Find lowest offset to a name+value pair.  This is the start of our
1472  * downward-growing free space.
1473  */
1474 static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1475 {
1476         return loc->xl_ops->xlo_get_free_start(loc);
1477 }
1478
1479 /* Can we reuse loc->xl_entry for xi? */
1480 static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1481                                     struct ocfs2_xattr_info *xi)
1482 {
1483         return loc->xl_ops->xlo_can_reuse(loc, xi);
1484 }
1485
1486 /* How much free space is needed to set the new value */
1487 static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1488                                 struct ocfs2_xattr_info *xi)
1489 {
1490         return loc->xl_ops->xlo_check_space(loc, xi);
1491 }
1492
1493 static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1494 {
1495         loc->xl_ops->xlo_add_entry(loc, name_hash);
1496         loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1497         /*
1498          * We can't leave the new entry's xe_name_offset at zero or
1499          * add_namevalue() will go nuts.  We set it to the size of our
1500          * storage so that it can never be less than any other entry.
1501          */
1502         loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1503 }
1504
1505 static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1506                                    struct ocfs2_xattr_info *xi)
1507 {
1508         int size = namevalue_size_xi(xi);
1509         int nameval_offset;
1510         char *nameval_buf;
1511
1512         loc->xl_ops->xlo_add_namevalue(loc, size);
1513         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1514         loc->xl_entry->xe_name_len = xi->xi_name_len;
1515         ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1516         ocfs2_xattr_set_local(loc->xl_entry,
1517                               xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1518
1519         nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1520         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1521         memset(nameval_buf, 0, size);
1522         memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1523 }
1524
1525 static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1526                                     struct ocfs2_xattr_value_buf *vb)
1527 {
1528         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1529         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1530
1531         /* Value bufs are for value trees */
1532         BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1533         BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1534                (name_size + OCFS2_XATTR_ROOT_SIZE));
1535
1536         loc->xl_ops->xlo_fill_value_buf(loc, vb);
1537         vb->vb_xv =
1538                 (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1539                                                         nameval_offset +
1540                                                         name_size);
1541 }
1542
1543 static int ocfs2_xa_block_journal_access(handle_t *handle,
1544                                          struct ocfs2_xa_loc *loc, int type)
1545 {
1546         struct buffer_head *bh = loc->xl_storage;
1547         ocfs2_journal_access_func access;
1548
1549         if (loc->xl_size == (bh->b_size -
1550                              offsetof(struct ocfs2_xattr_block,
1551                                       xb_attrs.xb_header)))
1552                 access = ocfs2_journal_access_xb;
1553         else
1554                 access = ocfs2_journal_access_di;
1555         return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1556 }
1557
1558 static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1559                                          struct ocfs2_xa_loc *loc)
1560 {
1561         struct buffer_head *bh = loc->xl_storage;
1562
1563         ocfs2_journal_dirty(handle, bh);
1564 }
1565
1566 static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1567                                            int offset)
1568 {
1569         return (char *)loc->xl_header + offset;
1570 }
1571
1572 static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1573                                     struct ocfs2_xattr_info *xi)
1574 {
1575         /*
1576          * Block storage is strict.  If the sizes aren't exact, we will
1577          * remove the old one and reinsert the new.
1578          */
1579         return namevalue_size_xe(loc->xl_entry) ==
1580                 namevalue_size_xi(xi);
1581 }
1582
1583 static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1584 {
1585         struct ocfs2_xattr_header *xh = loc->xl_header;
1586         int i, count = le16_to_cpu(xh->xh_count);
1587         int offset, free_start = loc->xl_size;
1588
1589         for (i = 0; i < count; i++) {
1590                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1591                 if (offset < free_start)
1592                         free_start = offset;
1593         }
1594
1595         return free_start;
1596 }
1597
1598 static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1599                                       struct ocfs2_xattr_info *xi)
1600 {
1601         int count = le16_to_cpu(loc->xl_header->xh_count);
1602         int free_start = ocfs2_xa_get_free_start(loc);
1603         int needed_space = ocfs2_xi_entry_usage(xi);
1604
1605         /*
1606          * Block storage will reclaim the original entry before inserting
1607          * the new value, so we only need the difference.  If the new
1608          * entry is smaller than the old one, we don't need anything.
1609          */
1610         if (loc->xl_entry) {
1611                 /* Don't need space if we're reusing! */
1612                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1613                         needed_space = 0;
1614                 else
1615                         needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1616         }
1617         if (needed_space < 0)
1618                 needed_space = 0;
1619         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1620 }
1621
1622 /*
1623  * Block storage for xattrs keeps the name+value pairs compacted.  When
1624  * we remove one, we have to shift any that preceded it towards the end.
1625  */
1626 static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1627 {
1628         int i, offset;
1629         int namevalue_offset, first_namevalue_offset, namevalue_size;
1630         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1631         struct ocfs2_xattr_header *xh = loc->xl_header;
1632         int count = le16_to_cpu(xh->xh_count);
1633
1634         namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1635         namevalue_size = namevalue_size_xe(entry);
1636         first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1637
1638         /* Shift the name+value pairs */
1639         memmove((char *)xh + first_namevalue_offset + namevalue_size,
1640                 (char *)xh + first_namevalue_offset,
1641                 namevalue_offset - first_namevalue_offset);
1642         memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1643
1644         /* Now tell xh->xh_entries about it */
1645         for (i = 0; i < count; i++) {
1646                 offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1647                 if (offset <= namevalue_offset)
1648                         le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1649                                      namevalue_size);
1650         }
1651
1652         /*
1653          * Note that we don't update xh_free_start or xh_name_value_len
1654          * because they're not used in block-stored xattrs.
1655          */
1656 }
1657
1658 static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1659 {
1660         int count = le16_to_cpu(loc->xl_header->xh_count);
1661         loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1662         le16_add_cpu(&loc->xl_header->xh_count, 1);
1663         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1664 }
1665
1666 static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1667 {
1668         int free_start = ocfs2_xa_get_free_start(loc);
1669
1670         loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1671 }
1672
1673 static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1674                                           struct ocfs2_xattr_value_buf *vb)
1675 {
1676         struct buffer_head *bh = loc->xl_storage;
1677
1678         if (loc->xl_size == (bh->b_size -
1679                              offsetof(struct ocfs2_xattr_block,
1680                                       xb_attrs.xb_header)))
1681                 vb->vb_access = ocfs2_journal_access_xb;
1682         else
1683                 vb->vb_access = ocfs2_journal_access_di;
1684         vb->vb_bh = bh;
1685 }
1686
1687 /*
1688  * Operations for xattrs stored in blocks.  This includes inline inode
1689  * storage and unindexed ocfs2_xattr_blocks.
1690  */
1691 static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1692         .xlo_journal_access     = ocfs2_xa_block_journal_access,
1693         .xlo_journal_dirty      = ocfs2_xa_block_journal_dirty,
1694         .xlo_offset_pointer     = ocfs2_xa_block_offset_pointer,
1695         .xlo_check_space        = ocfs2_xa_block_check_space,
1696         .xlo_can_reuse          = ocfs2_xa_block_can_reuse,
1697         .xlo_get_free_start     = ocfs2_xa_block_get_free_start,
1698         .xlo_wipe_namevalue     = ocfs2_xa_block_wipe_namevalue,
1699         .xlo_add_entry          = ocfs2_xa_block_add_entry,
1700         .xlo_add_namevalue      = ocfs2_xa_block_add_namevalue,
1701         .xlo_fill_value_buf     = ocfs2_xa_block_fill_value_buf,
1702 };
1703
1704 static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1705                                           struct ocfs2_xa_loc *loc, int type)
1706 {
1707         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1708
1709         return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1710 }
1711
1712 static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1713                                           struct ocfs2_xa_loc *loc)
1714 {
1715         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1716
1717         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1718 }
1719
1720 static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1721                                             int offset)
1722 {
1723         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1724         int block, block_offset;
1725
1726         /* The header is at the front of the bucket */
1727         block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1728         block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1729
1730         return bucket_block(bucket, block) + block_offset;
1731 }
1732
1733 static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1734                                      struct ocfs2_xattr_info *xi)
1735 {
1736         return namevalue_size_xe(loc->xl_entry) >=
1737                 namevalue_size_xi(xi);
1738 }
1739
1740 static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1741 {
1742         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1743         return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1744 }
1745
1746 static int ocfs2_bucket_align_free_start(struct super_block *sb,
1747                                          int free_start, int size)
1748 {
1749         /*
1750          * We need to make sure that the name+value pair fits within
1751          * one block.
1752          */
1753         if (((free_start - size) >> sb->s_blocksize_bits) !=
1754             ((free_start - 1) >> sb->s_blocksize_bits))
1755                 free_start -= free_start % sb->s_blocksize;
1756
1757         return free_start;
1758 }
1759
1760 static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1761                                        struct ocfs2_xattr_info *xi)
1762 {
1763         int rc;
1764         int count = le16_to_cpu(loc->xl_header->xh_count);
1765         int free_start = ocfs2_xa_get_free_start(loc);
1766         int needed_space = ocfs2_xi_entry_usage(xi);
1767         int size = namevalue_size_xi(xi);
1768         struct super_block *sb = loc->xl_inode->i_sb;
1769
1770         /*
1771          * Bucket storage does not reclaim name+value pairs it cannot
1772          * reuse.  They live as holes until the bucket fills, and then
1773          * the bucket is defragmented.  However, the bucket can reclaim
1774          * the ocfs2_xattr_entry.
1775          */
1776         if (loc->xl_entry) {
1777                 /* Don't need space if we're reusing! */
1778                 if (ocfs2_xa_can_reuse_entry(loc, xi))
1779                         needed_space = 0;
1780                 else
1781                         needed_space -= sizeof(struct ocfs2_xattr_entry);
1782         }
1783         BUG_ON(needed_space < 0);
1784
1785         if (free_start < size) {
1786                 if (needed_space)
1787                         return -ENOSPC;
1788         } else {
1789                 /*
1790                  * First we check if it would fit in the first place.
1791                  * Below, we align the free start to a block.  This may
1792                  * slide us below the minimum gap.  By checking unaligned
1793                  * first, we avoid that error.
1794                  */
1795                 rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1796                                                  count);
1797                 if (rc)
1798                         return rc;
1799                 free_start = ocfs2_bucket_align_free_start(sb, free_start,
1800                                                            size);
1801         }
1802         return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1803 }
1804
1805 static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1806 {
1807         le16_add_cpu(&loc->xl_header->xh_name_value_len,
1808                      -namevalue_size_xe(loc->xl_entry));
1809 }
1810
1811 static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1812 {
1813         struct ocfs2_xattr_header *xh = loc->xl_header;
1814         int count = le16_to_cpu(xh->xh_count);
1815         int low = 0, high = count - 1, tmp;
1816         struct ocfs2_xattr_entry *tmp_xe;
1817
1818         /*
1819          * We keep buckets sorted by name_hash, so we need to find
1820          * our insert place.
1821          */
1822         while (low <= high && count) {
1823                 tmp = (low + high) / 2;
1824                 tmp_xe = &xh->xh_entries[tmp];
1825
1826                 if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1827                         low = tmp + 1;
1828                 else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1829                         high = tmp - 1;
1830                 else {
1831                         low = tmp;
1832                         break;
1833                 }
1834         }
1835
1836         if (low != count)
1837                 memmove(&xh->xh_entries[low + 1],
1838                         &xh->xh_entries[low],
1839                         ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1840
1841         le16_add_cpu(&xh->xh_count, 1);
1842         loc->xl_entry = &xh->xh_entries[low];
1843         memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1844 }
1845
1846 static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1847 {
1848         int free_start = ocfs2_xa_get_free_start(loc);
1849         struct ocfs2_xattr_header *xh = loc->xl_header;
1850         struct super_block *sb = loc->xl_inode->i_sb;
1851         int nameval_offset;
1852
1853         free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1854         nameval_offset = free_start - size;
1855         loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1856         xh->xh_free_start = cpu_to_le16(nameval_offset);
1857         le16_add_cpu(&xh->xh_name_value_len, size);
1858
1859 }
1860
1861 static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1862                                            struct ocfs2_xattr_value_buf *vb)
1863 {
1864         struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1865         struct super_block *sb = loc->xl_inode->i_sb;
1866         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1867         int size = namevalue_size_xe(loc->xl_entry);
1868         int block_offset = nameval_offset >> sb->s_blocksize_bits;
1869
1870         /* Values are not allowed to straddle block boundaries */
1871         BUG_ON(block_offset !=
1872                ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1873         /* We expect the bucket to be filled in */
1874         BUG_ON(!bucket->bu_bhs[block_offset]);
1875
1876         vb->vb_access = ocfs2_journal_access;
1877         vb->vb_bh = bucket->bu_bhs[block_offset];
1878 }
1879
1880 /* Operations for xattrs stored in buckets. */
1881 static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1882         .xlo_journal_access     = ocfs2_xa_bucket_journal_access,
1883         .xlo_journal_dirty      = ocfs2_xa_bucket_journal_dirty,
1884         .xlo_offset_pointer     = ocfs2_xa_bucket_offset_pointer,
1885         .xlo_check_space        = ocfs2_xa_bucket_check_space,
1886         .xlo_can_reuse          = ocfs2_xa_bucket_can_reuse,
1887         .xlo_get_free_start     = ocfs2_xa_bucket_get_free_start,
1888         .xlo_wipe_namevalue     = ocfs2_xa_bucket_wipe_namevalue,
1889         .xlo_add_entry          = ocfs2_xa_bucket_add_entry,
1890         .xlo_add_namevalue      = ocfs2_xa_bucket_add_namevalue,
1891         .xlo_fill_value_buf     = ocfs2_xa_bucket_fill_value_buf,
1892 };
1893
1894 static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1895 {
1896         struct ocfs2_xattr_value_buf vb;
1897
1898         if (ocfs2_xattr_is_local(loc->xl_entry))
1899                 return 0;
1900
1901         ocfs2_xa_fill_value_buf(loc, &vb);
1902         return le32_to_cpu(vb.vb_xv->xr_clusters);
1903 }
1904
1905 static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1906                                    struct ocfs2_xattr_set_ctxt *ctxt)
1907 {
1908         int trunc_rc, access_rc;
1909         struct ocfs2_xattr_value_buf vb;
1910
1911         ocfs2_xa_fill_value_buf(loc, &vb);
1912         trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1913                                               ctxt);
1914
1915         /*
1916          * The caller of ocfs2_xa_value_truncate() has already called
1917          * ocfs2_xa_journal_access on the loc.  However, The truncate code
1918          * calls ocfs2_extend_trans().  This may commit the previous
1919          * transaction and open a new one.  If this is a bucket, truncate
1920          * could leave only vb->vb_bh set up for journaling.  Meanwhile,
1921          * the caller is expecting to dirty the entire bucket.  So we must
1922          * reset the journal work.  We do this even if truncate has failed,
1923          * as it could have failed after committing the extend.
1924          */
1925         access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1926                                             OCFS2_JOURNAL_ACCESS_WRITE);
1927
1928         /* Errors in truncate take precedence */
1929         return trunc_rc ? trunc_rc : access_rc;
1930 }
1931
1932 static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1933 {
1934         int index, count;
1935         struct ocfs2_xattr_header *xh = loc->xl_header;
1936         struct ocfs2_xattr_entry *entry = loc->xl_entry;
1937
1938         ocfs2_xa_wipe_namevalue(loc);
1939         loc->xl_entry = NULL;
1940
1941         le16_add_cpu(&xh->xh_count, -1);
1942         count = le16_to_cpu(xh->xh_count);
1943
1944         /*
1945          * Only zero out the entry if there are more remaining.  This is
1946          * important for an empty bucket, as it keeps track of the
1947          * bucket's hash value.  It doesn't hurt empty block storage.
1948          */
1949         if (count) {
1950                 index = ((char *)entry - (char *)&xh->xh_entries) /
1951                         sizeof(struct ocfs2_xattr_entry);
1952                 memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1953                         (count - index) * sizeof(struct ocfs2_xattr_entry));
1954                 memset(&xh->xh_entries[count], 0,
1955                        sizeof(struct ocfs2_xattr_entry));
1956         }
1957 }
1958
1959 /*
1960  * If we have a problem adjusting the size of an external value during
1961  * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1962  * in an intermediate state.  For example, the value may be partially
1963  * truncated.
1964  *
1965  * If the value tree hasn't changed, the extend/truncate went nowhere.
1966  * We have nothing to do.  The caller can treat it as a straight error.
1967  *
1968  * If the value tree got partially truncated, we now have a corrupted
1969  * extended attribute.  We're going to wipe its entry and leak the
1970  * clusters.  Better to leak some storage than leave a corrupt entry.
1971  *
1972  * If the value tree grew, it obviously didn't grow enough for the
1973  * new entry.  We're not going to try and reclaim those clusters either.
1974  * If there was already an external value there (orig_clusters != 0),
1975  * the new clusters are attached safely and we can just leave the old
1976  * value in place.  If there was no external value there, we remove
1977  * the entry.
1978  *
1979  * This way, the xattr block we store in the journal will be consistent.
1980  * If the size change broke because of the journal, no changes will hit
1981  * disk anyway.
1982  */
1983 static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1984                                             const char *what,
1985                                             unsigned int orig_clusters)
1986 {
1987         unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1988         char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1989                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
1990
1991         if (new_clusters < orig_clusters) {
1992                 mlog(ML_ERROR,
1993                      "Partial truncate while %s xattr %.*s.  Leaking "
1994                      "%u clusters and removing the entry\n",
1995                      what, loc->xl_entry->xe_name_len, nameval_buf,
1996                      orig_clusters - new_clusters);
1997                 ocfs2_xa_remove_entry(loc);
1998         } else if (!orig_clusters) {
1999                 mlog(ML_ERROR,
2000                      "Unable to allocate an external value for xattr "
2001                      "%.*s safely.  Leaking %u clusters and removing the "
2002                      "entry\n",
2003                      loc->xl_entry->xe_name_len, nameval_buf,
2004                      new_clusters - orig_clusters);
2005                 ocfs2_xa_remove_entry(loc);
2006         } else if (new_clusters > orig_clusters)
2007                 mlog(ML_ERROR,
2008                      "Unable to grow xattr %.*s safely.  %u new clusters "
2009                      "have been added, but the value will not be "
2010                      "modified\n",
2011                      loc->xl_entry->xe_name_len, nameval_buf,
2012                      new_clusters - orig_clusters);
2013 }
2014
2015 static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
2016                            struct ocfs2_xattr_set_ctxt *ctxt)
2017 {
2018         int rc = 0;
2019         unsigned int orig_clusters;
2020
2021         if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2022                 orig_clusters = ocfs2_xa_value_clusters(loc);
2023                 rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2024                 if (rc) {
2025                         mlog_errno(rc);
2026                         /*
2027                          * Since this is remove, we can return 0 if
2028                          * ocfs2_xa_cleanup_value_truncate() is going to
2029                          * wipe the entry anyway.  So we check the
2030                          * cluster count as well.
2031                          */
2032                         if (orig_clusters != ocfs2_xa_value_clusters(loc))
2033                                 rc = 0;
2034                         ocfs2_xa_cleanup_value_truncate(loc, "removing",
2035                                                         orig_clusters);
2036                         if (rc)
2037                                 goto out;
2038                 }
2039         }
2040
2041         ocfs2_xa_remove_entry(loc);
2042
2043 out:
2044         return rc;
2045 }
2046
2047 static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2048 {
2049         int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2050         char *nameval_buf;
2051
2052         nameval_buf = ocfs2_xa_offset_pointer(loc,
2053                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2054         memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2055 }
2056
2057 /*
2058  * Take an existing entry and make it ready for the new value.  This
2059  * won't allocate space, but it may free space.  It should be ready for
2060  * ocfs2_xa_prepare_entry() to finish the work.
2061  */
2062 static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2063                                 struct ocfs2_xattr_info *xi,
2064                                 struct ocfs2_xattr_set_ctxt *ctxt)
2065 {
2066         int rc = 0;
2067         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2068         unsigned int orig_clusters;
2069         char *nameval_buf;
2070         int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2071         int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2072
2073         BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2074                name_size);
2075
2076         nameval_buf = ocfs2_xa_offset_pointer(loc,
2077                                 le16_to_cpu(loc->xl_entry->xe_name_offset));
2078         if (xe_local) {
2079                 memset(nameval_buf + name_size, 0,
2080                        namevalue_size_xe(loc->xl_entry) - name_size);
2081                 if (!xi_local)
2082                         ocfs2_xa_install_value_root(loc);
2083         } else {
2084                 orig_clusters = ocfs2_xa_value_clusters(loc);
2085                 if (xi_local) {
2086                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2087                         if (rc < 0)
2088                                 mlog_errno(rc);
2089                         else
2090                                 memset(nameval_buf + name_size, 0,
2091                                        namevalue_size_xe(loc->xl_entry) -
2092                                        name_size);
2093                 } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2094                            xi->xi_value_len) {
2095                         rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2096                                                      ctxt);
2097                         if (rc < 0)
2098                                 mlog_errno(rc);
2099                 }
2100
2101                 if (rc) {
2102                         ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2103                                                         orig_clusters);
2104                         goto out;
2105                 }
2106         }
2107
2108         loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2109         ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2110
2111 out:
2112         return rc;
2113 }
2114
2115 /*
2116  * Prepares loc->xl_entry to receive the new xattr.  This includes
2117  * properly setting up the name+value pair region.  If loc->xl_entry
2118  * already exists, it will take care of modifying it appropriately.
2119  *
2120  * Note that this modifies the data.  You did journal_access already,
2121  * right?
2122  */
2123 static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2124                                   struct ocfs2_xattr_info *xi,
2125                                   u32 name_hash,
2126                                   struct ocfs2_xattr_set_ctxt *ctxt)
2127 {
2128         int rc = 0;
2129         unsigned int orig_clusters;
2130         __le64 orig_value_size = 0;
2131
2132         rc = ocfs2_xa_check_space(loc, xi);
2133         if (rc)
2134                 goto out;
2135
2136         if (loc->xl_entry) {
2137                 if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2138                         orig_value_size = loc->xl_entry->xe_value_size;
2139                         rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2140                         if (rc)
2141                                 goto out;
2142                         goto alloc_value;
2143                 }
2144
2145                 if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2146                         orig_clusters = ocfs2_xa_value_clusters(loc);
2147                         rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2148                         if (rc) {
2149                                 mlog_errno(rc);
2150                                 ocfs2_xa_cleanup_value_truncate(loc,
2151                                                                 "overwriting",
2152                                                                 orig_clusters);
2153                                 goto out;
2154                         }
2155                 }
2156                 ocfs2_xa_wipe_namevalue(loc);
2157         } else
2158                 ocfs2_xa_add_entry(loc, name_hash);
2159
2160         /*
2161          * If we get here, we have a blank entry.  Fill it.  We grow our
2162          * name+value pair back from the end.
2163          */
2164         ocfs2_xa_add_namevalue(loc, xi);
2165         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2166                 ocfs2_xa_install_value_root(loc);
2167
2168 alloc_value:
2169         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2170                 orig_clusters = ocfs2_xa_value_clusters(loc);
2171                 rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2172                 if (rc < 0) {
2173                         ctxt->set_abort = 1;
2174                         ocfs2_xa_cleanup_value_truncate(loc, "growing",
2175                                                         orig_clusters);
2176                         /*
2177                          * If we were growing an existing value,
2178                          * ocfs2_xa_cleanup_value_truncate() won't remove
2179                          * the entry. We need to restore the original value
2180                          * size.
2181                          */
2182                         if (loc->xl_entry) {
2183                                 BUG_ON(!orig_value_size);
2184                                 loc->xl_entry->xe_value_size = orig_value_size;
2185                         }
2186                         mlog_errno(rc);
2187                 }
2188         }
2189
2190 out:
2191         return rc;
2192 }
2193
2194 /*
2195  * Store the value portion of the name+value pair.  This will skip
2196  * values that are stored externally.  Their tree roots were set up
2197  * by ocfs2_xa_prepare_entry().
2198  */
2199 static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2200                                 struct ocfs2_xattr_info *xi,
2201                                 struct ocfs2_xattr_set_ctxt *ctxt)
2202 {
2203         int rc = 0;
2204         int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2205         int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2206         char *nameval_buf;
2207         struct ocfs2_xattr_value_buf vb;
2208
2209         nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2210         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2211                 ocfs2_xa_fill_value_buf(loc, &vb);
2212                 rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2213                                                      ctxt->handle, &vb,
2214                                                      xi->xi_value,
2215                                                      xi->xi_value_len);
2216         } else
2217                 memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2218
2219         return rc;
2220 }
2221
2222 static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2223                         struct ocfs2_xattr_info *xi,
2224                         struct ocfs2_xattr_set_ctxt *ctxt)
2225 {
2226         int ret;
2227         u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2228                                               xi->xi_name_len);
2229
2230         ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2231                                       OCFS2_JOURNAL_ACCESS_WRITE);
2232         if (ret) {
2233                 mlog_errno(ret);
2234                 goto out;
2235         }
2236
2237         /*
2238          * From here on out, everything is going to modify the buffer a
2239          * little.  Errors are going to leave the xattr header in a
2240          * sane state.  Thus, even with errors we dirty the sucker.
2241          */
2242
2243         /* Don't worry, we are never called with !xi_value and !xl_entry */
2244         if (!xi->xi_value) {
2245                 ret = ocfs2_xa_remove(loc, ctxt);
2246                 goto out_dirty;
2247         }
2248
2249         ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2250         if (ret) {
2251                 if (ret != -ENOSPC)
2252                         mlog_errno(ret);
2253                 goto out_dirty;
2254         }
2255
2256         ret = ocfs2_xa_store_value(loc, xi, ctxt);
2257         if (ret)
2258                 mlog_errno(ret);
2259
2260 out_dirty:
2261         ocfs2_xa_journal_dirty(ctxt->handle, loc);
2262
2263 out:
2264         return ret;
2265 }
2266
2267 static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2268                                      struct inode *inode,
2269                                      struct buffer_head *bh,
2270                                      struct ocfs2_xattr_entry *entry)
2271 {
2272         struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2273
2274         BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2275
2276         loc->xl_inode = inode;
2277         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2278         loc->xl_storage = bh;
2279         loc->xl_entry = entry;
2280         loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2281         loc->xl_header =
2282                 (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2283                                               loc->xl_size);
2284 }
2285
2286 static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2287                                           struct inode *inode,
2288                                           struct buffer_head *bh,
2289                                           struct ocfs2_xattr_entry *entry)
2290 {
2291         struct ocfs2_xattr_block *xb =
2292                 (struct ocfs2_xattr_block *)bh->b_data;
2293
2294         BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2295
2296         loc->xl_inode = inode;
2297         loc->xl_ops = &ocfs2_xa_block_loc_ops;
2298         loc->xl_storage = bh;
2299         loc->xl_header = &(xb->xb_attrs.xb_header);
2300         loc->xl_entry = entry;
2301         loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2302                                              xb_attrs.xb_header);
2303 }
2304
2305 static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2306                                            struct ocfs2_xattr_bucket *bucket,
2307                                            struct ocfs2_xattr_entry *entry)
2308 {
2309         loc->xl_inode = bucket->bu_inode;
2310         loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2311         loc->xl_storage = bucket;
2312         loc->xl_header = bucket_xh(bucket);
2313         loc->xl_entry = entry;
2314         loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2315 }
2316
2317 /*
2318  * In xattr remove, if it is stored outside and refcounted, we may have
2319  * the chance to split the refcount tree. So need the allocators.
2320  */
2321 static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2322                                         struct ocfs2_xattr_value_root *xv,
2323                                         struct ocfs2_caching_info *ref_ci,
2324                                         struct buffer_head *ref_root_bh,
2325                                         struct ocfs2_alloc_context **meta_ac,
2326                                         int *ref_credits)
2327 {
2328         int ret, meta_add = 0;
2329         u32 p_cluster, num_clusters;
2330         unsigned int ext_flags;
2331
2332         *ref_credits = 0;
2333         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2334                                        &num_clusters,
2335                                        &xv->xr_list,
2336                                        &ext_flags);
2337         if (ret) {
2338                 mlog_errno(ret);
2339                 goto out;
2340         }
2341
2342         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2343                 goto out;
2344
2345         ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2346                                                  ref_root_bh, xv,
2347                                                  &meta_add, ref_credits);
2348         if (ret) {
2349                 mlog_errno(ret);
2350                 goto out;
2351         }
2352
2353         ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2354                                                 meta_add, meta_ac);
2355         if (ret)
2356                 mlog_errno(ret);
2357
2358 out:
2359         return ret;
2360 }
2361
2362 static int ocfs2_remove_value_outside(struct inode*inode,
2363                                       struct ocfs2_xattr_value_buf *vb,
2364                                       struct ocfs2_xattr_header *header,
2365                                       struct ocfs2_caching_info *ref_ci,
2366                                       struct buffer_head *ref_root_bh)
2367 {
2368         int ret = 0, i, ref_credits;
2369         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2370         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2371         void *val;
2372
2373         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2374
2375         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2376                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2377
2378                 if (ocfs2_xattr_is_local(entry))
2379                         continue;
2380
2381                 val = (void *)header +
2382                         le16_to_cpu(entry->xe_name_offset);
2383                 vb->vb_xv = (struct ocfs2_xattr_value_root *)
2384                         (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2385
2386                 ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2387                                                          ref_ci, ref_root_bh,
2388                                                          &ctxt.meta_ac,
2389                                                          &ref_credits);
2390
2391                 ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2392                                         ocfs2_remove_extent_credits(osb->sb));
2393                 if (IS_ERR(ctxt.handle)) {
2394                         ret = PTR_ERR(ctxt.handle);
2395                         mlog_errno(ret);
2396                         break;
2397                 }
2398
2399                 ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2400
2401                 ocfs2_commit_trans(osb, ctxt.handle);
2402                 if (ctxt.meta_ac) {
2403                         ocfs2_free_alloc_context(ctxt.meta_ac);
2404                         ctxt.meta_ac = NULL;
2405                 }
2406
2407                 if (ret < 0) {
2408                         mlog_errno(ret);
2409                         break;
2410                 }
2411
2412         }
2413
2414         if (ctxt.meta_ac)
2415                 ocfs2_free_alloc_context(ctxt.meta_ac);
2416         ocfs2_schedule_truncate_log_flush(osb, 1);
2417         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2418         return ret;
2419 }
2420
2421 static int ocfs2_xattr_ibody_remove(struct inode *inode,
2422                                     struct buffer_head *di_bh,
2423                                     struct ocfs2_caching_info *ref_ci,
2424                                     struct buffer_head *ref_root_bh)
2425 {
2426
2427         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2428         struct ocfs2_xattr_header *header;
2429         int ret;
2430         struct ocfs2_xattr_value_buf vb = {
2431                 .vb_bh = di_bh,
2432                 .vb_access = ocfs2_journal_access_di,
2433         };
2434
2435         header = (struct ocfs2_xattr_header *)
2436                  ((void *)di + inode->i_sb->s_blocksize -
2437                  le16_to_cpu(di->i_xattr_inline_size));
2438
2439         ret = ocfs2_remove_value_outside(inode, &vb, header,
2440                                          ref_ci, ref_root_bh);
2441
2442         return ret;
2443 }
2444
2445 struct ocfs2_rm_xattr_bucket_para {
2446         struct ocfs2_caching_info *ref_ci;
2447         struct buffer_head *ref_root_bh;
2448 };
2449
2450 static int ocfs2_xattr_block_remove(struct inode *inode,
2451                                     struct buffer_head *blk_bh,
2452                                     struct ocfs2_caching_info *ref_ci,
2453                                     struct buffer_head *ref_root_bh)
2454 {
2455         struct ocfs2_xattr_block *xb;
2456         int ret = 0;
2457         struct ocfs2_xattr_value_buf vb = {
2458                 .vb_bh = blk_bh,
2459                 .vb_access = ocfs2_journal_access_xb,
2460         };
2461         struct ocfs2_rm_xattr_bucket_para args = {
2462                 .ref_ci = ref_ci,
2463                 .ref_root_bh = ref_root_bh,
2464         };
2465
2466         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2467         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2468                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2469                 ret = ocfs2_remove_value_outside(inode, &vb, header,
2470                                                  ref_ci, ref_root_bh);
2471         } else
2472                 ret = ocfs2_iterate_xattr_index_block(inode,
2473                                                 blk_bh,
2474                                                 ocfs2_rm_xattr_cluster,
2475                                                 &args);
2476
2477         return ret;
2478 }
2479
2480 static int ocfs2_xattr_free_block(struct inode *inode,
2481                                   u64 block,
2482                                   struct ocfs2_caching_info *ref_ci,
2483                                   struct buffer_head *ref_root_bh)
2484 {
2485         struct inode *xb_alloc_inode;
2486         struct buffer_head *xb_alloc_bh = NULL;
2487         struct buffer_head *blk_bh = NULL;
2488         struct ocfs2_xattr_block *xb;
2489         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2490         handle_t *handle;
2491         int ret = 0;
2492         u64 blk, bg_blkno;
2493         u16 bit;
2494
2495         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2496         if (ret < 0) {
2497                 mlog_errno(ret);
2498                 goto out;
2499         }
2500
2501         ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2502         if (ret < 0) {
2503                 mlog_errno(ret);
2504                 goto out;
2505         }
2506
2507         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2508         blk = le64_to_cpu(xb->xb_blkno);
2509         bit = le16_to_cpu(xb->xb_suballoc_bit);
2510         if (xb->xb_suballoc_loc)
2511                 bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2512         else
2513                 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2514
2515         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2516                                 EXTENT_ALLOC_SYSTEM_INODE,
2517                                 le16_to_cpu(xb->xb_suballoc_slot));
2518         if (!xb_alloc_inode) {
2519                 ret = -ENOMEM;
2520                 mlog_errno(ret);
2521                 goto out;
2522         }
2523         inode_lock(xb_alloc_inode);
2524
2525         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2526         if (ret < 0) {
2527                 mlog_errno(ret);
2528                 goto out_mutex;
2529         }
2530
2531         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2532         if (IS_ERR(handle)) {
2533                 ret = PTR_ERR(handle);
2534                 mlog_errno(ret);
2535                 goto out_unlock;
2536         }
2537
2538         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2539                                        bit, bg_blkno, 1);
2540         if (ret < 0)
2541                 mlog_errno(ret);
2542
2543         ocfs2_commit_trans(osb, handle);
2544 out_unlock:
2545         ocfs2_inode_unlock(xb_alloc_inode, 1);
2546         brelse(xb_alloc_bh);
2547 out_mutex:
2548         inode_unlock(xb_alloc_inode);
2549         iput(xb_alloc_inode);
2550 out:
2551         brelse(blk_bh);
2552         return ret;
2553 }
2554
2555 /*
2556  * ocfs2_xattr_remove()
2557  *
2558  * Free extended attribute resources associated with this inode.
2559  */
2560 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2561 {
2562         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2563         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2564         struct ocfs2_refcount_tree *ref_tree = NULL;
2565         struct buffer_head *ref_root_bh = NULL;
2566         struct ocfs2_caching_info *ref_ci = NULL;
2567         handle_t *handle;
2568         int ret;
2569
2570         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2571                 return 0;
2572
2573         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2574                 return 0;
2575
2576         if (ocfs2_is_refcount_inode(inode)) {
2577                 ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2578                                                le64_to_cpu(di->i_refcount_loc),
2579                                                1, &ref_tree, &ref_root_bh);
2580                 if (ret) {
2581                         mlog_errno(ret);
2582                         goto out;
2583                 }
2584                 ref_ci = &ref_tree->rf_ci;
2585
2586         }
2587
2588         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2589                 ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2590                                                ref_ci, ref_root_bh);
2591                 if (ret < 0) {
2592                         mlog_errno(ret);
2593                         goto out;
2594                 }
2595         }
2596
2597         if (di->i_xattr_loc) {
2598                 ret = ocfs2_xattr_free_block(inode,
2599                                              le64_to_cpu(di->i_xattr_loc),
2600                                              ref_ci, ref_root_bh);
2601                 if (ret < 0) {
2602                         mlog_errno(ret);
2603                         goto out;
2604                 }
2605         }
2606
2607         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2608                                    OCFS2_INODE_UPDATE_CREDITS);
2609         if (IS_ERR(handle)) {
2610                 ret = PTR_ERR(handle);
2611                 mlog_errno(ret);
2612                 goto out;
2613         }
2614         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2615                                       OCFS2_JOURNAL_ACCESS_WRITE);
2616         if (ret) {
2617                 mlog_errno(ret);
2618                 goto out_commit;
2619         }
2620
2621         di->i_xattr_loc = 0;
2622
2623         spin_lock(&oi->ip_lock);
2624         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2625         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2626         spin_unlock(&oi->ip_lock);
2627         ocfs2_update_inode_fsync_trans(handle, inode, 0);
2628
2629         ocfs2_journal_dirty(handle, di_bh);
2630 out_commit:
2631         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2632 out:
2633         if (ref_tree)
2634                 ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2635         brelse(ref_root_bh);
2636         return ret;
2637 }
2638
2639 static int ocfs2_xattr_has_space_inline(struct inode *inode,
2640                                         struct ocfs2_dinode *di)
2641 {
2642         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2643         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2644         int free;
2645
2646         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2647                 return 0;
2648
2649         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2650                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2651                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2652         } else if (ocfs2_inode_is_fast_symlink(inode)) {
2653                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
2654                         le64_to_cpu(di->i_size);
2655         } else {
2656                 struct ocfs2_extent_list *el = &di->id2.i_list;
2657                 free = (le16_to_cpu(el->l_count) -
2658                         le16_to_cpu(el->l_next_free_rec)) *
2659                         sizeof(struct ocfs2_extent_rec);
2660         }
2661         if (free >= xattrsize)
2662                 return 1;
2663
2664         return 0;
2665 }
2666
2667 /*
2668  * ocfs2_xattr_ibody_find()
2669  *
2670  * Find extended attribute in inode block and
2671  * fill search info into struct ocfs2_xattr_search.
2672  */
2673 static int ocfs2_xattr_ibody_find(struct inode *inode,
2674                                   int name_index,
2675                                   const char *name,
2676                                   struct ocfs2_xattr_search *xs)
2677 {
2678         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2679         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2680         int ret;
2681         int has_space = 0;
2682
2683         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2684                 return 0;
2685
2686         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2687                 down_read(&oi->ip_alloc_sem);
2688                 has_space = ocfs2_xattr_has_space_inline(inode, di);
2689                 up_read(&oi->ip_alloc_sem);
2690                 if (!has_space)
2691                         return 0;
2692         }
2693
2694         xs->xattr_bh = xs->inode_bh;
2695         xs->end = (void *)di + inode->i_sb->s_blocksize;
2696         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2697                 xs->header = (struct ocfs2_xattr_header *)
2698                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2699         else
2700                 xs->header = (struct ocfs2_xattr_header *)
2701                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2702         xs->base = (void *)xs->header;
2703         xs->here = xs->header->xh_entries;
2704
2705         /* Find the named attribute. */
2706         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2707                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2708                 if (ret && ret != -ENODATA)
2709                         return ret;
2710                 xs->not_found = ret;
2711         }
2712
2713         return 0;
2714 }
2715
2716 static int ocfs2_xattr_ibody_init(struct inode *inode,
2717                                   struct buffer_head *di_bh,
2718                                   struct ocfs2_xattr_set_ctxt *ctxt)
2719 {
2720         int ret;
2721         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2722         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2723         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2724         unsigned int xattrsize = osb->s_xattr_inline_size;
2725
2726         if (!ocfs2_xattr_has_space_inline(inode, di)) {
2727                 ret = -ENOSPC;
2728                 goto out;
2729         }
2730
2731         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2732                                       OCFS2_JOURNAL_ACCESS_WRITE);
2733         if (ret) {
2734                 mlog_errno(ret);
2735                 goto out;
2736         }
2737
2738         /*
2739          * Adjust extent record count or inline data size
2740          * to reserve space for extended attribute.
2741          */
2742         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2743                 struct ocfs2_inline_data *idata = &di->id2.i_data;
2744                 le16_add_cpu(&idata->id_count, -xattrsize);
2745         } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2746                 struct ocfs2_extent_list *el = &di->id2.i_list;
2747                 le16_add_cpu(&el->l_count, -(xattrsize /
2748                                              sizeof(struct ocfs2_extent_rec)));
2749         }
2750         di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2751
2752         spin_lock(&oi->ip_lock);
2753         oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2754         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2755         spin_unlock(&oi->ip_lock);
2756
2757         ocfs2_journal_dirty(ctxt->handle, di_bh);
2758
2759 out:
2760         return ret;
2761 }
2762
2763 /*
2764  * ocfs2_xattr_ibody_set()
2765  *
2766  * Set, replace or remove an extended attribute into inode block.
2767  *
2768  */
2769 static int ocfs2_xattr_ibody_set(struct inode *inode,
2770                                  struct ocfs2_xattr_info *xi,
2771                                  struct ocfs2_xattr_search *xs,
2772                                  struct ocfs2_xattr_set_ctxt *ctxt)
2773 {
2774         int ret;
2775         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2776         struct ocfs2_xa_loc loc;
2777
2778         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2779                 return -ENOSPC;
2780
2781         down_write(&oi->ip_alloc_sem);
2782         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2783                 ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2784                 if (ret) {
2785                         if (ret != -ENOSPC)
2786                                 mlog_errno(ret);
2787                         goto out;
2788                 }
2789         }
2790
2791         ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2792                                  xs->not_found ? NULL : xs->here);
2793         ret = ocfs2_xa_set(&loc, xi, ctxt);
2794         if (ret) {
2795                 if (ret != -ENOSPC)
2796                         mlog_errno(ret);
2797                 goto out;
2798         }
2799         xs->here = loc.xl_entry;
2800
2801 out:
2802         up_write(&oi->ip_alloc_sem);
2803
2804         return ret;
2805 }
2806
2807 /*
2808  * ocfs2_xattr_block_find()
2809  *
2810  * Find extended attribute in external block and
2811  * fill search info into struct ocfs2_xattr_search.
2812  */
2813 static int ocfs2_xattr_block_find(struct inode *inode,
2814                                   int name_index,
2815                                   const char *name,
2816                                   struct ocfs2_xattr_search *xs)
2817 {
2818         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2819         struct buffer_head *blk_bh = NULL;
2820         struct ocfs2_xattr_block *xb;
2821         int ret = 0;
2822
2823         if (!di->i_xattr_loc)
2824                 return ret;
2825
2826         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2827                                      &blk_bh);
2828         if (ret < 0) {
2829                 mlog_errno(ret);
2830                 return ret;
2831         }
2832
2833         xs->xattr_bh = blk_bh;
2834         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2835
2836         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2837                 xs->header = &xb->xb_attrs.xb_header;
2838                 xs->base = (void *)xs->header;
2839                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2840                 xs->here = xs->header->xh_entries;
2841
2842                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2843         } else
2844                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2845                                                    name_index,
2846                                                    name, xs);
2847
2848         if (ret && ret != -ENODATA) {
2849                 xs->xattr_bh = NULL;
2850                 goto cleanup;
2851         }
2852         xs->not_found = ret;
2853         return 0;
2854 cleanup:
2855         brelse(blk_bh);
2856
2857         return ret;
2858 }
2859
2860 static int ocfs2_create_xattr_block(struct inode *inode,
2861                                     struct buffer_head *inode_bh,
2862                                     struct ocfs2_xattr_set_ctxt *ctxt,
2863                                     int indexed,
2864                                     struct buffer_head **ret_bh)
2865 {
2866         int ret;
2867         u16 suballoc_bit_start;
2868         u32 num_got;
2869         u64 suballoc_loc, first_blkno;
2870         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2871         struct buffer_head *new_bh = NULL;
2872         struct ocfs2_xattr_block *xblk;
2873
2874         ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2875                                       inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2876         if (ret < 0) {
2877                 mlog_errno(ret);
2878                 goto end;
2879         }
2880
2881         ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2882                                    &suballoc_loc, &suballoc_bit_start,
2883                                    &num_got, &first_blkno);
2884         if (ret < 0) {
2885                 mlog_errno(ret);
2886                 goto end;
2887         }
2888
2889         new_bh = sb_getblk(inode->i_sb, first_blkno);
2890         if (!new_bh) {
2891                 ret = -ENOMEM;
2892                 mlog_errno(ret);
2893                 goto end;
2894         }
2895
2896         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2897
2898         ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2899                                       new_bh,
2900                                       OCFS2_JOURNAL_ACCESS_CREATE);
2901         if (ret < 0) {
2902                 mlog_errno(ret);
2903                 goto end;
2904         }
2905
2906         /* Initialize ocfs2_xattr_block */
2907         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2908         memset(xblk, 0, inode->i_sb->s_blocksize);
2909         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2910         xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2911         xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2912         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2913         xblk->xb_fs_generation =
2914                 cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2915         xblk->xb_blkno = cpu_to_le64(first_blkno);
2916         if (indexed) {
2917                 struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2918                 xr->xt_clusters = cpu_to_le32(1);
2919                 xr->xt_last_eb_blk = 0;
2920                 xr->xt_list.l_tree_depth = 0;
2921                 xr->xt_list.l_count = cpu_to_le16(
2922                                         ocfs2_xattr_recs_per_xb(inode->i_sb));
2923                 xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2924                 xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2925         }
2926         ocfs2_journal_dirty(ctxt->handle, new_bh);
2927
2928         /* Add it to the inode */
2929         di->i_xattr_loc = cpu_to_le64(first_blkno);
2930
2931         spin_lock(&OCFS2_I(inode)->ip_lock);
2932         OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2933         di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2934         spin_unlock(&OCFS2_I(inode)->ip_lock);
2935
2936         ocfs2_journal_dirty(ctxt->handle, inode_bh);
2937
2938         *ret_bh = new_bh;
2939         new_bh = NULL;
2940
2941 end:
2942         brelse(new_bh);
2943         return ret;
2944 }
2945
2946 /*
2947  * ocfs2_xattr_block_set()
2948  *
2949  * Set, replace or remove an extended attribute into external block.
2950  *
2951  */
2952 static int ocfs2_xattr_block_set(struct inode *inode,
2953                                  struct ocfs2_xattr_info *xi,
2954                                  struct ocfs2_xattr_search *xs,
2955                                  struct ocfs2_xattr_set_ctxt *ctxt)
2956 {
2957         struct buffer_head *new_bh = NULL;
2958         struct ocfs2_xattr_block *xblk = NULL;
2959         int ret;
2960         struct ocfs2_xa_loc loc;
2961
2962         if (!xs->xattr_bh) {
2963                 ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2964                                                0, &new_bh);
2965                 if (ret) {
2966                         mlog_errno(ret);
2967                         goto end;
2968                 }
2969
2970                 xs->xattr_bh = new_bh;
2971                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2972                 xs->header = &xblk->xb_attrs.xb_header;
2973                 xs->base = (void *)xs->header;
2974                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2975                 xs->here = xs->header->xh_entries;
2976         } else
2977                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2978
2979         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2980                 ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2981                                               xs->not_found ? NULL : xs->here);
2982
2983                 ret = ocfs2_xa_set(&loc, xi, ctxt);
2984                 if (!ret)
2985                         xs->here = loc.xl_entry;
2986                 else if ((ret != -ENOSPC) || ctxt->set_abort)
2987                         goto end;
2988                 else {
2989                         ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2990                         if (ret)
2991                                 goto end;
2992                 }
2993         }
2994
2995         if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2996                 ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2997
2998 end:
2999         return ret;
3000 }
3001
3002 /* Check whether the new xattr can be inserted into the inode. */
3003 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
3004                                        struct ocfs2_xattr_info *xi,
3005                                        struct ocfs2_xattr_search *xs)
3006 {
3007         struct ocfs2_xattr_entry *last;
3008         int free, i;
3009         size_t min_offs = xs->end - xs->base;
3010
3011         if (!xs->header)
3012                 return 0;
3013
3014         last = xs->header->xh_entries;
3015
3016         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
3017                 size_t offs = le16_to_cpu(last->xe_name_offset);
3018                 if (offs < min_offs)
3019                         min_offs = offs;
3020                 last += 1;
3021         }
3022
3023         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3024         if (free < 0)
3025                 return 0;
3026
3027         BUG_ON(!xs->not_found);
3028
3029         if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3030                 return 1;
3031
3032         return 0;
3033 }
3034
3035 static int ocfs2_calc_xattr_set_need(struct inode *inode,
3036                                      struct ocfs2_dinode *di,
3037                                      struct ocfs2_xattr_info *xi,
3038                                      struct ocfs2_xattr_search *xis,
3039                                      struct ocfs2_xattr_search *xbs,
3040                                      int *clusters_need,
3041                                      int *meta_need,
3042                                      int *credits_need)
3043 {
3044         int ret = 0, old_in_xb = 0;
3045         int clusters_add = 0, meta_add = 0, credits = 0;
3046         struct buffer_head *bh = NULL;
3047         struct ocfs2_xattr_block *xb = NULL;
3048         struct ocfs2_xattr_entry *xe = NULL;
3049         struct ocfs2_xattr_value_root *xv = NULL;
3050         char *base = NULL;
3051         int name_offset, name_len = 0;
3052         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3053                                                     xi->xi_value_len);
3054         u64 value_size;
3055
3056         /*
3057          * Calculate the clusters we need to write.
3058          * No matter whether we replace an old one or add a new one,
3059          * we need this for writing.
3060          */
3061         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3062                 credits += new_clusters *
3063                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
3064
3065         if (xis->not_found && xbs->not_found) {
3066                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3067
3068                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3069                         clusters_add += new_clusters;
3070                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3071                                                         &def_xv.xv.xr_list);
3072                 }
3073
3074                 goto meta_guess;
3075         }
3076
3077         if (!xis->not_found) {
3078                 xe = xis->here;
3079                 name_offset = le16_to_cpu(xe->xe_name_offset);
3080                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3081                 base = xis->base;
3082                 credits += OCFS2_INODE_UPDATE_CREDITS;
3083         } else {
3084                 int i, block_off = 0;
3085                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3086                 xe = xbs->here;
3087                 name_offset = le16_to_cpu(xe->xe_name_offset);
3088                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3089                 i = xbs->here - xbs->header->xh_entries;
3090                 old_in_xb = 1;
3091
3092                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3093                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3094                                                         bucket_xh(xbs->bucket),
3095                                                         i, &block_off,
3096                                                         &name_offset);
3097                         base = bucket_block(xbs->bucket, block_off);
3098                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3099                 } else {
3100                         base = xbs->base;
3101                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3102                 }
3103         }
3104
3105         /*
3106          * delete a xattr doesn't need metadata and cluster allocation.
3107          * so just calculate the credits and return.
3108          *
3109          * The credits for removing the value tree will be extended
3110          * by ocfs2_remove_extent itself.
3111          */
3112         if (!xi->xi_value) {
3113                 if (!ocfs2_xattr_is_local(xe))
3114                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3115
3116                 goto out;
3117         }
3118
3119         /* do cluster allocation guess first. */
3120         value_size = le64_to_cpu(xe->xe_value_size);
3121
3122         if (old_in_xb) {
3123                 /*
3124                  * In xattr set, we always try to set the xe in inode first,
3125                  * so if it can be inserted into inode successfully, the old
3126                  * one will be removed from the xattr block, and this xattr
3127                  * will be inserted into inode as a new xattr in inode.
3128                  */
3129                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3130                         clusters_add += new_clusters;
3131                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
3132                                     OCFS2_INODE_UPDATE_CREDITS;
3133                         if (!ocfs2_xattr_is_local(xe))
3134                                 credits += ocfs2_calc_extend_credits(
3135                                                         inode->i_sb,
3136                                                         &def_xv.xv.xr_list);
3137                         goto out;
3138                 }
3139         }
3140
3141         if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3142                 /* the new values will be stored outside. */
3143                 u32 old_clusters = 0;
3144
3145                 if (!ocfs2_xattr_is_local(xe)) {
3146                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
3147                                                                  value_size);
3148                         xv = (struct ocfs2_xattr_value_root *)
3149                              (base + name_offset + name_len);
3150                         value_size = OCFS2_XATTR_ROOT_SIZE;
3151                 } else
3152                         xv = &def_xv.xv;
3153
3154                 if (old_clusters >= new_clusters) {
3155                         credits += ocfs2_remove_extent_credits(inode->i_sb);
3156                         goto out;
3157                 } else {
3158                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3159                         clusters_add += new_clusters - old_clusters;
3160                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3161                                                              &xv->xr_list);
3162                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3163                                 goto out;
3164                 }
3165         } else {
3166                 /*
3167                  * Now the new value will be stored inside. So if the new
3168                  * value is smaller than the size of value root or the old
3169                  * value, we don't need any allocation, otherwise we have
3170                  * to guess metadata allocation.
3171                  */
3172                 if ((ocfs2_xattr_is_local(xe) &&
3173                      (value_size >= xi->xi_value_len)) ||
3174                     (!ocfs2_xattr_is_local(xe) &&
3175                      OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3176                         goto out;
3177         }
3178
3179 meta_guess:
3180         /* calculate metadata allocation. */
3181         if (di->i_xattr_loc) {
3182                 if (!xbs->xattr_bh) {
3183                         ret = ocfs2_read_xattr_block(inode,
3184                                                      le64_to_cpu(di->i_xattr_loc),
3185                                                      &bh);
3186                         if (ret) {
3187                                 mlog_errno(ret);
3188                                 goto out;
3189                         }
3190
3191                         xb = (struct ocfs2_xattr_block *)bh->b_data;
3192                 } else
3193                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3194
3195                 /*
3196                  * If there is already an xattr tree, good, we can calculate
3197                  * like other b-trees. Otherwise we may have the chance of
3198                  * create a tree, the credit calculation is borrowed from
3199                  * ocfs2_calc_extend_credits with root_el = NULL. And the
3200                  * new tree will be cluster based, so no meta is needed.
3201                  */
3202                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3203                         struct ocfs2_extent_list *el =
3204                                  &xb->xb_attrs.xb_root.xt_list;
3205                         meta_add += ocfs2_extend_meta_needed(el);
3206                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3207                                                              el);
3208                 } else
3209                         credits += OCFS2_SUBALLOC_ALLOC + 1;
3210
3211                 /*
3212                  * This cluster will be used either for new bucket or for
3213                  * new xattr block.
3214                  * If the cluster size is the same as the bucket size, one
3215                  * more is needed since we may need to extend the bucket
3216                  * also.
3217                  */
3218                 clusters_add += 1;
3219                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3220                 if (OCFS2_XATTR_BUCKET_SIZE ==
3221                         OCFS2_SB(inode->i_sb)->s_clustersize) {
3222                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3223                         clusters_add += 1;
3224                 }
3225         } else {
3226                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3227                 if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3228                         struct ocfs2_extent_list *el = &def_xv.xv.xr_list;
3229                         meta_add += ocfs2_extend_meta_needed(el);
3230                         credits += ocfs2_calc_extend_credits(inode->i_sb,
3231                                                              el);
3232                 } else {
3233                         meta_add += 1;
3234                 }
3235         }
3236 out:
3237         if (clusters_need)
3238                 *clusters_need = clusters_add;
3239         if (meta_need)
3240                 *meta_need = meta_add;
3241         if (credits_need)
3242                 *credits_need = credits;
3243         brelse(bh);
3244         return ret;
3245 }
3246
3247 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3248                                      struct ocfs2_dinode *di,
3249                                      struct ocfs2_xattr_info *xi,
3250                                      struct ocfs2_xattr_search *xis,
3251                                      struct ocfs2_xattr_search *xbs,
3252                                      struct ocfs2_xattr_set_ctxt *ctxt,
3253                                      int extra_meta,
3254                                      int *credits)
3255 {
3256         int clusters_add, meta_add, ret;
3257         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3258
3259         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3260
3261         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3262
3263         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3264                                         &clusters_add, &meta_add, credits);
3265         if (ret) {
3266                 mlog_errno(ret);
3267                 return ret;
3268         }
3269
3270         meta_add += extra_meta;
3271         trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3272                                         clusters_add, *credits);
3273
3274         if (meta_add) {
3275                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3276                                                         &ctxt->meta_ac);
3277                 if (ret) {
3278                         mlog_errno(ret);
3279                         goto out;
3280                 }
3281         }
3282
3283         if (clusters_add) {
3284                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3285                 if (ret)
3286                         mlog_errno(ret);
3287         }
3288 out:
3289         if (ret) {
3290                 if (ctxt->meta_ac) {
3291                         ocfs2_free_alloc_context(ctxt->meta_ac);
3292                         ctxt->meta_ac = NULL;
3293                 }
3294
3295                 /*
3296                  * We cannot have an error and a non null ctxt->data_ac.
3297                  */
3298         }
3299
3300         return ret;
3301 }
3302
3303 static int __ocfs2_xattr_set_handle(struct inode *inode,
3304                                     struct ocfs2_dinode *di,
3305                                     struct ocfs2_xattr_info *xi,
3306                                     struct ocfs2_xattr_search *xis,
3307           &