Root/fs/ocfs2/xattr.c

1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * xattr.c
5 *
6 * Copyright (C) 2004, 2008 Oracle. All rights reserved.
7 *
8 * CREDITS:
9 * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
10 * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public
14 * License version 2 as published by the Free Software Foundation.
15 *
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
20 */
21
22#include <linux/capability.h>
23#include <linux/fs.h>
24#include <linux/types.h>
25#include <linux/slab.h>
26#include <linux/highmem.h>
27#include <linux/pagemap.h>
28#include <linux/uio.h>
29#include <linux/sched.h>
30#include <linux/splice.h>
31#include <linux/mount.h>
32#include <linux/writeback.h>
33#include <linux/falloc.h>
34#include <linux/sort.h>
35#include <linux/init.h>
36#include <linux/module.h>
37#include <linux/string.h>
38#include <linux/security.h>
39
40#include <cluster/masklog.h>
41
42#include "ocfs2.h"
43#include "alloc.h"
44#include "blockcheck.h"
45#include "dlmglue.h"
46#include "file.h"
47#include "symlink.h"
48#include "sysfile.h"
49#include "inode.h"
50#include "journal.h"
51#include "ocfs2_fs.h"
52#include "suballoc.h"
53#include "uptodate.h"
54#include "buffer_head_io.h"
55#include "super.h"
56#include "xattr.h"
57#include "refcounttree.h"
58#include "acl.h"
59#include "ocfs2_trace.h"
60
61struct ocfs2_xattr_def_value_root {
62    struct ocfs2_xattr_value_root xv;
63    struct ocfs2_extent_rec er;
64};
65
66struct ocfs2_xattr_bucket {
67    /* The inode these xattrs are associated with */
68    struct inode *bu_inode;
69
70    /* The actual buffers that make up the bucket */
71    struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
72
73    /* How many blocks make up one bucket for this filesystem */
74    int bu_blocks;
75};
76
77struct ocfs2_xattr_set_ctxt {
78    handle_t *handle;
79    struct ocfs2_alloc_context *meta_ac;
80    struct ocfs2_alloc_context *data_ac;
81    struct ocfs2_cached_dealloc_ctxt dealloc;
82    int set_abort;
83};
84
85#define OCFS2_XATTR_ROOT_SIZE (sizeof(struct ocfs2_xattr_def_value_root))
86#define OCFS2_XATTR_INLINE_SIZE 80
87#define OCFS2_XATTR_HEADER_GAP 4
88#define OCFS2_XATTR_FREE_IN_IBODY (OCFS2_MIN_XATTR_INLINE_SIZE \
89                     - sizeof(struct ocfs2_xattr_header) \
90                     - OCFS2_XATTR_HEADER_GAP)
91#define OCFS2_XATTR_FREE_IN_BLOCK(ptr) ((ptr)->i_sb->s_blocksize \
92                     - sizeof(struct ocfs2_xattr_block) \
93                     - sizeof(struct ocfs2_xattr_header) \
94                     - OCFS2_XATTR_HEADER_GAP)
95
96static struct ocfs2_xattr_def_value_root def_xv = {
97    .xv.xr_list.l_count = cpu_to_le16(1),
98};
99
100const struct xattr_handler *ocfs2_xattr_handlers[] = {
101    &ocfs2_xattr_user_handler,
102    &ocfs2_xattr_acl_access_handler,
103    &ocfs2_xattr_acl_default_handler,
104    &ocfs2_xattr_trusted_handler,
105    &ocfs2_xattr_security_handler,
106    NULL
107};
108
109static const struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
110    [OCFS2_XATTR_INDEX_USER] = &ocfs2_xattr_user_handler,
111    [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
112                    = &ocfs2_xattr_acl_access_handler,
113    [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
114                    = &ocfs2_xattr_acl_default_handler,
115    [OCFS2_XATTR_INDEX_TRUSTED] = &ocfs2_xattr_trusted_handler,
116    [OCFS2_XATTR_INDEX_SECURITY] = &ocfs2_xattr_security_handler,
117};
118
119struct ocfs2_xattr_info {
120    int xi_name_index;
121    const char *xi_name;
122    int xi_name_len;
123    const void *xi_value;
124    size_t xi_value_len;
125};
126
127struct ocfs2_xattr_search {
128    struct buffer_head *inode_bh;
129    /*
130     * xattr_bh point to the block buffer head which has extended attribute
131     * when extended attribute in inode, xattr_bh is equal to inode_bh.
132     */
133    struct buffer_head *xattr_bh;
134    struct ocfs2_xattr_header *header;
135    struct ocfs2_xattr_bucket *bucket;
136    void *base;
137    void *end;
138    struct ocfs2_xattr_entry *here;
139    int not_found;
140};
141
142/* Operations on struct ocfs2_xa_entry */
143struct ocfs2_xa_loc;
144struct ocfs2_xa_loc_operations {
145    /*
146     * Journal functions
147     */
148    int (*xlo_journal_access)(handle_t *handle, struct ocfs2_xa_loc *loc,
149                  int type);
150    void (*xlo_journal_dirty)(handle_t *handle, struct ocfs2_xa_loc *loc);
151
152    /*
153     * Return a pointer to the appropriate buffer in loc->xl_storage
154     * at the given offset from loc->xl_header.
155     */
156    void *(*xlo_offset_pointer)(struct ocfs2_xa_loc *loc, int offset);
157
158    /* Can we reuse the existing entry for the new value? */
159    int (*xlo_can_reuse)(struct ocfs2_xa_loc *loc,
160                 struct ocfs2_xattr_info *xi);
161
162    /* How much space is needed for the new value? */
163    int (*xlo_check_space)(struct ocfs2_xa_loc *loc,
164                   struct ocfs2_xattr_info *xi);
165
166    /*
167     * Return the offset of the first name+value pair. This is
168     * the start of our downward-filling free space.
169     */
170    int (*xlo_get_free_start)(struct ocfs2_xa_loc *loc);
171
172    /*
173     * Remove the name+value at this location. Do whatever is
174     * appropriate with the remaining name+value pairs.
175     */
176    void (*xlo_wipe_namevalue)(struct ocfs2_xa_loc *loc);
177
178    /* Fill xl_entry with a new entry */
179    void (*xlo_add_entry)(struct ocfs2_xa_loc *loc, u32 name_hash);
180
181    /* Add name+value storage to an entry */
182    void (*xlo_add_namevalue)(struct ocfs2_xa_loc *loc, int size);
183
184    /*
185     * Initialize the value buf's access and bh fields for this entry.
186     * ocfs2_xa_fill_value_buf() will handle the xv pointer.
187     */
188    void (*xlo_fill_value_buf)(struct ocfs2_xa_loc *loc,
189                   struct ocfs2_xattr_value_buf *vb);
190};
191
192/*
193 * Describes an xattr entry location. This is a memory structure
194 * tracking the on-disk structure.
195 */
196struct ocfs2_xa_loc {
197    /* This xattr belongs to this inode */
198    struct inode *xl_inode;
199
200    /* The ocfs2_xattr_header inside the on-disk storage. Not NULL. */
201    struct ocfs2_xattr_header *xl_header;
202
203    /* Bytes from xl_header to the end of the storage */
204    int xl_size;
205
206    /*
207     * The ocfs2_xattr_entry this location describes. If this is
208     * NULL, this location describes the on-disk structure where it
209     * would have been.
210     */
211    struct ocfs2_xattr_entry *xl_entry;
212
213    /*
214     * Internal housekeeping
215     */
216
217    /* Buffer(s) containing this entry */
218    void *xl_storage;
219
220    /* Operations on the storage backing this location */
221    const struct ocfs2_xa_loc_operations *xl_ops;
222};
223
224/*
225 * Convenience functions to calculate how much space is needed for a
226 * given name+value pair
227 */
228static int namevalue_size(int name_len, uint64_t value_len)
229{
230    if (value_len > OCFS2_XATTR_INLINE_SIZE)
231        return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
232    else
233        return OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
234}
235
236static int namevalue_size_xi(struct ocfs2_xattr_info *xi)
237{
238    return namevalue_size(xi->xi_name_len, xi->xi_value_len);
239}
240
241static int namevalue_size_xe(struct ocfs2_xattr_entry *xe)
242{
243    u64 value_len = le64_to_cpu(xe->xe_value_size);
244
245    BUG_ON((value_len > OCFS2_XATTR_INLINE_SIZE) &&
246           ocfs2_xattr_is_local(xe));
247    return namevalue_size(xe->xe_name_len, value_len);
248}
249
250
251static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
252                         struct ocfs2_xattr_header *xh,
253                         int index,
254                         int *block_off,
255                         int *new_offset);
256
257static int ocfs2_xattr_block_find(struct inode *inode,
258                  int name_index,
259                  const char *name,
260                  struct ocfs2_xattr_search *xs);
261static int ocfs2_xattr_index_block_find(struct inode *inode,
262                    struct buffer_head *root_bh,
263                    int name_index,
264                    const char *name,
265                    struct ocfs2_xattr_search *xs);
266
267static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
268                    struct buffer_head *blk_bh,
269                    char *buffer,
270                    size_t buffer_size);
271
272static int ocfs2_xattr_create_index_block(struct inode *inode,
273                      struct ocfs2_xattr_search *xs,
274                      struct ocfs2_xattr_set_ctxt *ctxt);
275
276static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
277                         struct ocfs2_xattr_info *xi,
278                         struct ocfs2_xattr_search *xs,
279                         struct ocfs2_xattr_set_ctxt *ctxt);
280
281typedef int (xattr_tree_rec_func)(struct inode *inode,
282                  struct buffer_head *root_bh,
283                  u64 blkno, u32 cpos, u32 len, void *para);
284static int ocfs2_iterate_xattr_index_block(struct inode *inode,
285                       struct buffer_head *root_bh,
286                       xattr_tree_rec_func *rec_func,
287                       void *para);
288static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
289                    struct ocfs2_xattr_bucket *bucket,
290                    void *para);
291static int ocfs2_rm_xattr_cluster(struct inode *inode,
292                  struct buffer_head *root_bh,
293                  u64 blkno,
294                  u32 cpos,
295                  u32 len,
296                  void *para);
297
298static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
299                  u64 src_blk, u64 last_blk, u64 to_blk,
300                  unsigned int start_bucket,
301                  u32 *first_hash);
302static int ocfs2_prepare_refcount_xattr(struct inode *inode,
303                    struct ocfs2_dinode *di,
304                    struct ocfs2_xattr_info *xi,
305                    struct ocfs2_xattr_search *xis,
306                    struct ocfs2_xattr_search *xbs,
307                    struct ocfs2_refcount_tree **ref_tree,
308                    int *meta_need,
309                    int *credits);
310static int ocfs2_get_xattr_tree_value_root(struct super_block *sb,
311                       struct ocfs2_xattr_bucket *bucket,
312                       int offset,
313                       struct ocfs2_xattr_value_root **xv,
314                       struct buffer_head **bh);
315
316static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
317{
318    return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
319}
320
321static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
322{
323    return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
324}
325
326#define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
327#define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
328#define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
329
330static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
331{
332    struct ocfs2_xattr_bucket *bucket;
333    int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
334
335    BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
336
337    bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
338    if (bucket) {
339        bucket->bu_inode = inode;
340        bucket->bu_blocks = blks;
341    }
342
343    return bucket;
344}
345
346static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
347{
348    int i;
349
350    for (i = 0; i < bucket->bu_blocks; i++) {
351        brelse(bucket->bu_bhs[i]);
352        bucket->bu_bhs[i] = NULL;
353    }
354}
355
356static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
357{
358    if (bucket) {
359        ocfs2_xattr_bucket_relse(bucket);
360        bucket->bu_inode = NULL;
361        kfree(bucket);
362    }
363}
364
365/*
366 * A bucket that has never been written to disk doesn't need to be
367 * read. We just need the buffer_heads. Don't call this for
368 * buckets that are already on disk. ocfs2_read_xattr_bucket() initializes
369 * them fully.
370 */
371static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
372                   u64 xb_blkno)
373{
374    int i, rc = 0;
375
376    for (i = 0; i < bucket->bu_blocks; i++) {
377        bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
378                          xb_blkno + i);
379        if (!bucket->bu_bhs[i]) {
380            rc = -EIO;
381            mlog_errno(rc);
382            break;
383        }
384
385        if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
386                       bucket->bu_bhs[i]))
387            ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
388                              bucket->bu_bhs[i]);
389    }
390
391    if (rc)
392        ocfs2_xattr_bucket_relse(bucket);
393    return rc;
394}
395
396/* Read the xattr bucket at xb_blkno */
397static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
398                   u64 xb_blkno)
399{
400    int rc;
401
402    rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
403                   bucket->bu_blocks, bucket->bu_bhs, 0,
404                   NULL);
405    if (!rc) {
406        spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
407        rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
408                         bucket->bu_bhs,
409                         bucket->bu_blocks,
410                         &bucket_xh(bucket)->xh_check);
411        spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
412        if (rc)
413            mlog_errno(rc);
414    }
415
416    if (rc)
417        ocfs2_xattr_bucket_relse(bucket);
418    return rc;
419}
420
421static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
422                         struct ocfs2_xattr_bucket *bucket,
423                         int type)
424{
425    int i, rc = 0;
426
427    for (i = 0; i < bucket->bu_blocks; i++) {
428        rc = ocfs2_journal_access(handle,
429                      INODE_CACHE(bucket->bu_inode),
430                      bucket->bu_bhs[i], type);
431        if (rc) {
432            mlog_errno(rc);
433            break;
434        }
435    }
436
437    return rc;
438}
439
440static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
441                         struct ocfs2_xattr_bucket *bucket)
442{
443    int i;
444
445    spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
446    ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
447                   bucket->bu_bhs, bucket->bu_blocks,
448                   &bucket_xh(bucket)->xh_check);
449    spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
450
451    for (i = 0; i < bucket->bu_blocks; i++)
452        ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
453}
454
455static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
456                     struct ocfs2_xattr_bucket *src)
457{
458    int i;
459    int blocksize = src->bu_inode->i_sb->s_blocksize;
460
461    BUG_ON(dest->bu_blocks != src->bu_blocks);
462    BUG_ON(dest->bu_inode != src->bu_inode);
463
464    for (i = 0; i < src->bu_blocks; i++) {
465        memcpy(bucket_block(dest, i), bucket_block(src, i),
466               blocksize);
467    }
468}
469
470static int ocfs2_validate_xattr_block(struct super_block *sb,
471                      struct buffer_head *bh)
472{
473    int rc;
474    struct ocfs2_xattr_block *xb =
475        (struct ocfs2_xattr_block *)bh->b_data;
476
477    trace_ocfs2_validate_xattr_block((unsigned long long)bh->b_blocknr);
478
479    BUG_ON(!buffer_uptodate(bh));
480
481    /*
482     * If the ecc fails, we return the error but otherwise
483     * leave the filesystem running. We know any error is
484     * local to this block.
485     */
486    rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
487    if (rc)
488        return rc;
489
490    /*
491     * Errors after here are fatal
492     */
493
494    if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
495        ocfs2_error(sb,
496                "Extended attribute block #%llu has bad "
497                "signature %.*s",
498                (unsigned long long)bh->b_blocknr, 7,
499                xb->xb_signature);
500        return -EINVAL;
501    }
502
503    if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
504        ocfs2_error(sb,
505                "Extended attribute block #%llu has an "
506                "invalid xb_blkno of %llu",
507                (unsigned long long)bh->b_blocknr,
508                (unsigned long long)le64_to_cpu(xb->xb_blkno));
509        return -EINVAL;
510    }
511
512    if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
513        ocfs2_error(sb,
514                "Extended attribute block #%llu has an invalid "
515                "xb_fs_generation of #%u",
516                (unsigned long long)bh->b_blocknr,
517                le32_to_cpu(xb->xb_fs_generation));
518        return -EINVAL;
519    }
520
521    return 0;
522}
523
524static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
525                  struct buffer_head **bh)
526{
527    int rc;
528    struct buffer_head *tmp = *bh;
529
530    rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
531                  ocfs2_validate_xattr_block);
532
533    /* If ocfs2_read_block() got us a new bh, pass it up. */
534    if (!rc && !*bh)
535        *bh = tmp;
536
537    return rc;
538}
539
540static inline const char *ocfs2_xattr_prefix(int name_index)
541{
542    const struct xattr_handler *handler = NULL;
543
544    if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
545        handler = ocfs2_xattr_handler_map[name_index];
546
547    return handler ? handler->prefix : NULL;
548}
549
550static u32 ocfs2_xattr_name_hash(struct inode *inode,
551                 const char *name,
552                 int name_len)
553{
554    /* Get hash value of uuid from super block */
555    u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
556    int i;
557
558    /* hash extended attribute name */
559    for (i = 0; i < name_len; i++) {
560        hash = (hash << OCFS2_HASH_SHIFT) ^
561               (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
562               *name++;
563    }
564
565    return hash;
566}
567
568static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
569{
570    return namevalue_size(name_len, value_len) +
571        sizeof(struct ocfs2_xattr_entry);
572}
573
574static int ocfs2_xi_entry_usage(struct ocfs2_xattr_info *xi)
575{
576    return namevalue_size_xi(xi) +
577        sizeof(struct ocfs2_xattr_entry);
578}
579
580static int ocfs2_xe_entry_usage(struct ocfs2_xattr_entry *xe)
581{
582    return namevalue_size_xe(xe) +
583        sizeof(struct ocfs2_xattr_entry);
584}
585
586int ocfs2_calc_security_init(struct inode *dir,
587                 struct ocfs2_security_xattr_info *si,
588                 int *want_clusters,
589                 int *xattr_credits,
590                 struct ocfs2_alloc_context **xattr_ac)
591{
592    int ret = 0;
593    struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
594    int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
595                         si->value_len);
596
597    /*
598     * The max space of security xattr taken inline is
599     * 256(name) + 80(value) + 16(entry) = 352 bytes,
600     * So reserve one metadata block for it is ok.
601     */
602    if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
603        s_size > OCFS2_XATTR_FREE_IN_IBODY) {
604        ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
605        if (ret) {
606            mlog_errno(ret);
607            return ret;
608        }
609        *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
610    }
611
612    /* reserve clusters for xattr value which will be set in B tree*/
613    if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
614        int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
615                                si->value_len);
616
617        *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
618                               new_clusters);
619        *want_clusters += new_clusters;
620    }
621    return ret;
622}
623
624int ocfs2_calc_xattr_init(struct inode *dir,
625              struct buffer_head *dir_bh,
626              int mode,
627              struct ocfs2_security_xattr_info *si,
628              int *want_clusters,
629              int *xattr_credits,
630              int *want_meta)
631{
632    int ret = 0;
633    struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
634    int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
635
636    if (si->enable)
637        s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
638                             si->value_len);
639
640    if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
641        acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
642                    OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
643                    "", NULL, 0);
644        if (acl_len > 0) {
645            a_size = ocfs2_xattr_entry_real_size(0, acl_len);
646            if (S_ISDIR(mode))
647                a_size <<= 1;
648        } else if (acl_len != 0 && acl_len != -ENODATA) {
649            mlog_errno(ret);
650            return ret;
651        }
652    }
653
654    if (!(s_size + a_size))
655        return ret;
656
657    /*
658     * The max space of security xattr taken inline is
659     * 256(name) + 80(value) + 16(entry) = 352 bytes,
660     * The max space of acl xattr taken inline is
661     * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
662     * when blocksize = 512, may reserve one more cluser for
663     * xattr bucket, otherwise reserve one metadata block
664     * for them is ok.
665     * If this is a new directory with inline data,
666     * we choose to reserve the entire inline area for
667     * directory contents and force an external xattr block.
668     */
669    if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
670        (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
671        (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
672        *want_meta = *want_meta + 1;
673        *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
674    }
675
676    if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
677        (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
678        *want_clusters += 1;
679        *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
680    }
681
682    /*
683     * reserve credits and clusters for xattrs which has large value
684     * and have to be set outside
685     */
686    if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
687        new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
688                            si->value_len);
689        *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
690                               new_clusters);
691        *want_clusters += new_clusters;
692    }
693    if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
694        acl_len > OCFS2_XATTR_INLINE_SIZE) {
695        /* for directory, it has DEFAULT and ACCESS two types of acls */
696        new_clusters = (S_ISDIR(mode) ? 2 : 1) *
697                ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
698        *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
699                               new_clusters);
700        *want_clusters += new_clusters;
701    }
702
703    return ret;
704}
705
706static int ocfs2_xattr_extend_allocation(struct inode *inode,
707                     u32 clusters_to_add,
708                     struct ocfs2_xattr_value_buf *vb,
709                     struct ocfs2_xattr_set_ctxt *ctxt)
710{
711    int status = 0, credits;
712    handle_t *handle = ctxt->handle;
713    enum ocfs2_alloc_restarted why;
714    u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
715    struct ocfs2_extent_tree et;
716
717    ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
718
719    while (clusters_to_add) {
720        trace_ocfs2_xattr_extend_allocation(clusters_to_add);
721
722        status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
723                       OCFS2_JOURNAL_ACCESS_WRITE);
724        if (status < 0) {
725            mlog_errno(status);
726            break;
727        }
728
729        prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
730        status = ocfs2_add_clusters_in_btree(handle,
731                             &et,
732                             &logical_start,
733                             clusters_to_add,
734                             0,
735                             ctxt->data_ac,
736                             ctxt->meta_ac,
737                             &why);
738        if ((status < 0) && (status != -EAGAIN)) {
739            if (status != -ENOSPC)
740                mlog_errno(status);
741            break;
742        }
743
744        ocfs2_journal_dirty(handle, vb->vb_bh);
745
746        clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) -
747                     prev_clusters;
748
749        if (why != RESTART_NONE && clusters_to_add) {
750            /*
751             * We can only fail in case the alloc file doesn't give
752             * up enough clusters.
753             */
754            BUG_ON(why == RESTART_META);
755
756            credits = ocfs2_calc_extend_credits(inode->i_sb,
757                                &vb->vb_xv->xr_list,
758                                clusters_to_add);
759            status = ocfs2_extend_trans(handle, credits);
760            if (status < 0) {
761                status = -ENOMEM;
762                mlog_errno(status);
763                break;
764            }
765        }
766    }
767
768    return status;
769}
770
771static int __ocfs2_remove_xattr_range(struct inode *inode,
772                      struct ocfs2_xattr_value_buf *vb,
773                      u32 cpos, u32 phys_cpos, u32 len,
774                      unsigned int ext_flags,
775                      struct ocfs2_xattr_set_ctxt *ctxt)
776{
777    int ret;
778    u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
779    handle_t *handle = ctxt->handle;
780    struct ocfs2_extent_tree et;
781
782    ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
783
784    ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
785                OCFS2_JOURNAL_ACCESS_WRITE);
786    if (ret) {
787        mlog_errno(ret);
788        goto out;
789    }
790
791    ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
792                  &ctxt->dealloc);
793    if (ret) {
794        mlog_errno(ret);
795        goto out;
796    }
797
798    le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
799    ocfs2_journal_dirty(handle, vb->vb_bh);
800
801    if (ext_flags & OCFS2_EXT_REFCOUNTED)
802        ret = ocfs2_decrease_refcount(inode, handle,
803                    ocfs2_blocks_to_clusters(inode->i_sb,
804                                 phys_blkno),
805                    len, ctxt->meta_ac, &ctxt->dealloc, 1);
806    else
807        ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
808                          phys_blkno, len);
809    if (ret)
810        mlog_errno(ret);
811
812out:
813    return ret;
814}
815
816static int ocfs2_xattr_shrink_size(struct inode *inode,
817                   u32 old_clusters,
818                   u32 new_clusters,
819                   struct ocfs2_xattr_value_buf *vb,
820                   struct ocfs2_xattr_set_ctxt *ctxt)
821{
822    int ret = 0;
823    unsigned int ext_flags;
824    u32 trunc_len, cpos, phys_cpos, alloc_size;
825    u64 block;
826
827    if (old_clusters <= new_clusters)
828        return 0;
829
830    cpos = new_clusters;
831    trunc_len = old_clusters - new_clusters;
832    while (trunc_len) {
833        ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
834                           &alloc_size,
835                           &vb->vb_xv->xr_list, &ext_flags);
836        if (ret) {
837            mlog_errno(ret);
838            goto out;
839        }
840
841        if (alloc_size > trunc_len)
842            alloc_size = trunc_len;
843
844        ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
845                         phys_cpos, alloc_size,
846                         ext_flags, ctxt);
847        if (ret) {
848            mlog_errno(ret);
849            goto out;
850        }
851
852        block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
853        ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
854                               block, alloc_size);
855        cpos += alloc_size;
856        trunc_len -= alloc_size;
857    }
858
859out:
860    return ret;
861}
862
863static int ocfs2_xattr_value_truncate(struct inode *inode,
864                      struct ocfs2_xattr_value_buf *vb,
865                      int len,
866                      struct ocfs2_xattr_set_ctxt *ctxt)
867{
868    int ret;
869    u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
870    u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
871
872    if (new_clusters == old_clusters)
873        return 0;
874
875    if (new_clusters > old_clusters)
876        ret = ocfs2_xattr_extend_allocation(inode,
877                            new_clusters - old_clusters,
878                            vb, ctxt);
879    else
880        ret = ocfs2_xattr_shrink_size(inode,
881                          old_clusters, new_clusters,
882                          vb, ctxt);
883
884    return ret;
885}
886
887static int ocfs2_xattr_list_entry(char *buffer, size_t size,
888                  size_t *result, const char *prefix,
889                  const char *name, int name_len)
890{
891    char *p = buffer + *result;
892    int prefix_len = strlen(prefix);
893    int total_len = prefix_len + name_len + 1;
894
895    *result += total_len;
896
897    /* we are just looking for how big our buffer needs to be */
898    if (!size)
899        return 0;
900
901    if (*result > size)
902        return -ERANGE;
903
904    memcpy(p, prefix, prefix_len);
905    memcpy(p + prefix_len, name, name_len);
906    p[prefix_len + name_len] = '\0';
907
908    return 0;
909}
910
911static int ocfs2_xattr_list_entries(struct inode *inode,
912                    struct ocfs2_xattr_header *header,
913                    char *buffer, size_t buffer_size)
914{
915    size_t result = 0;
916    int i, type, ret;
917    const char *prefix, *name;
918
919    for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
920        struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
921        type = ocfs2_xattr_get_type(entry);
922        prefix = ocfs2_xattr_prefix(type);
923
924        if (prefix) {
925            name = (const char *)header +
926                le16_to_cpu(entry->xe_name_offset);
927
928            ret = ocfs2_xattr_list_entry(buffer, buffer_size,
929                             &result, prefix, name,
930                             entry->xe_name_len);
931            if (ret)
932                return ret;
933        }
934    }
935
936    return result;
937}
938
939int ocfs2_has_inline_xattr_value_outside(struct inode *inode,
940                     struct ocfs2_dinode *di)
941{
942    struct ocfs2_xattr_header *xh;
943    int i;
944
945    xh = (struct ocfs2_xattr_header *)
946         ((void *)di + inode->i_sb->s_blocksize -
947         le16_to_cpu(di->i_xattr_inline_size));
948
949    for (i = 0; i < le16_to_cpu(xh->xh_count); i++)
950        if (!ocfs2_xattr_is_local(&xh->xh_entries[i]))
951            return 1;
952
953    return 0;
954}
955
956static int ocfs2_xattr_ibody_list(struct inode *inode,
957                  struct ocfs2_dinode *di,
958                  char *buffer,
959                  size_t buffer_size)
960{
961    struct ocfs2_xattr_header *header = NULL;
962    struct ocfs2_inode_info *oi = OCFS2_I(inode);
963    int ret = 0;
964
965    if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
966        return ret;
967
968    header = (struct ocfs2_xattr_header *)
969         ((void *)di + inode->i_sb->s_blocksize -
970         le16_to_cpu(di->i_xattr_inline_size));
971
972    ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
973
974    return ret;
975}
976
977static int ocfs2_xattr_block_list(struct inode *inode,
978                  struct ocfs2_dinode *di,
979                  char *buffer,
980                  size_t buffer_size)
981{
982    struct buffer_head *blk_bh = NULL;
983    struct ocfs2_xattr_block *xb;
984    int ret = 0;
985
986    if (!di->i_xattr_loc)
987        return ret;
988
989    ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
990                     &blk_bh);
991    if (ret < 0) {
992        mlog_errno(ret);
993        return ret;
994    }
995
996    xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
997    if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
998        struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
999        ret = ocfs2_xattr_list_entries(inode, header,
1000                           buffer, buffer_size);
1001    } else
1002        ret = ocfs2_xattr_tree_list_index_block(inode, blk_bh,
1003                           buffer, buffer_size);
1004
1005    brelse(blk_bh);
1006
1007    return ret;
1008}
1009
1010ssize_t ocfs2_listxattr(struct dentry *dentry,
1011            char *buffer,
1012            size_t size)
1013{
1014    int ret = 0, i_ret = 0, b_ret = 0;
1015    struct buffer_head *di_bh = NULL;
1016    struct ocfs2_dinode *di = NULL;
1017    struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
1018
1019    if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
1020        return -EOPNOTSUPP;
1021
1022    if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1023        return ret;
1024
1025    ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
1026    if (ret < 0) {
1027        mlog_errno(ret);
1028        return ret;
1029    }
1030
1031    di = (struct ocfs2_dinode *)di_bh->b_data;
1032
1033    down_read(&oi->ip_xattr_sem);
1034    i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
1035    if (i_ret < 0)
1036        b_ret = 0;
1037    else {
1038        if (buffer) {
1039            buffer += i_ret;
1040            size -= i_ret;
1041        }
1042        b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
1043                           buffer, size);
1044        if (b_ret < 0)
1045            i_ret = 0;
1046    }
1047    up_read(&oi->ip_xattr_sem);
1048    ocfs2_inode_unlock(dentry->d_inode, 0);
1049
1050    brelse(di_bh);
1051
1052    return i_ret + b_ret;
1053}
1054
1055static int ocfs2_xattr_find_entry(int name_index,
1056                  const char *name,
1057                  struct ocfs2_xattr_search *xs)
1058{
1059    struct ocfs2_xattr_entry *entry;
1060    size_t name_len;
1061    int i, cmp = 1;
1062
1063    if (name == NULL)
1064        return -EINVAL;
1065
1066    name_len = strlen(name);
1067    entry = xs->here;
1068    for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
1069        cmp = name_index - ocfs2_xattr_get_type(entry);
1070        if (!cmp)
1071            cmp = name_len - entry->xe_name_len;
1072        if (!cmp)
1073            cmp = memcmp(name, (xs->base +
1074                     le16_to_cpu(entry->xe_name_offset)),
1075                     name_len);
1076        if (cmp == 0)
1077            break;
1078        entry += 1;
1079    }
1080    xs->here = entry;
1081
1082    return cmp ? -ENODATA : 0;
1083}
1084
1085static int ocfs2_xattr_get_value_outside(struct inode *inode,
1086                     struct ocfs2_xattr_value_root *xv,
1087                     void *buffer,
1088                     size_t len)
1089{
1090    u32 cpos, p_cluster, num_clusters, bpc, clusters;
1091    u64 blkno;
1092    int i, ret = 0;
1093    size_t cplen, blocksize;
1094    struct buffer_head *bh = NULL;
1095    struct ocfs2_extent_list *el;
1096
1097    el = &xv->xr_list;
1098    clusters = le32_to_cpu(xv->xr_clusters);
1099    bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1100    blocksize = inode->i_sb->s_blocksize;
1101
1102    cpos = 0;
1103    while (cpos < clusters) {
1104        ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1105                           &num_clusters, el, NULL);
1106        if (ret) {
1107            mlog_errno(ret);
1108            goto out;
1109        }
1110
1111        blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1112        /* Copy ocfs2_xattr_value */
1113        for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1114            ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1115                           &bh, NULL);
1116            if (ret) {
1117                mlog_errno(ret);
1118                goto out;
1119            }
1120
1121            cplen = len >= blocksize ? blocksize : len;
1122            memcpy(buffer, bh->b_data, cplen);
1123            len -= cplen;
1124            buffer += cplen;
1125
1126            brelse(bh);
1127            bh = NULL;
1128            if (len == 0)
1129                break;
1130        }
1131        cpos += num_clusters;
1132    }
1133out:
1134    return ret;
1135}
1136
1137static int ocfs2_xattr_ibody_get(struct inode *inode,
1138                 int name_index,
1139                 const char *name,
1140                 void *buffer,
1141                 size_t buffer_size,
1142                 struct ocfs2_xattr_search *xs)
1143{
1144    struct ocfs2_inode_info *oi = OCFS2_I(inode);
1145    struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1146    struct ocfs2_xattr_value_root *xv;
1147    size_t size;
1148    int ret = 0;
1149
1150    if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1151        return -ENODATA;
1152
1153    xs->end = (void *)di + inode->i_sb->s_blocksize;
1154    xs->header = (struct ocfs2_xattr_header *)
1155            (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1156    xs->base = (void *)xs->header;
1157    xs->here = xs->header->xh_entries;
1158
1159    ret = ocfs2_xattr_find_entry(name_index, name, xs);
1160    if (ret)
1161        return ret;
1162    size = le64_to_cpu(xs->here->xe_value_size);
1163    if (buffer) {
1164        if (size > buffer_size)
1165            return -ERANGE;
1166        if (ocfs2_xattr_is_local(xs->here)) {
1167            memcpy(buffer, (void *)xs->base +
1168                   le16_to_cpu(xs->here->xe_name_offset) +
1169                   OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1170        } else {
1171            xv = (struct ocfs2_xattr_value_root *)
1172                (xs->base + le16_to_cpu(
1173                 xs->here->xe_name_offset) +
1174                OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1175            ret = ocfs2_xattr_get_value_outside(inode, xv,
1176                                buffer, size);
1177            if (ret < 0) {
1178                mlog_errno(ret);
1179                return ret;
1180            }
1181        }
1182    }
1183
1184    return size;
1185}
1186
1187static int ocfs2_xattr_block_get(struct inode *inode,
1188                 int name_index,
1189                 const char *name,
1190                 void *buffer,
1191                 size_t buffer_size,
1192                 struct ocfs2_xattr_search *xs)
1193{
1194    struct ocfs2_xattr_block *xb;
1195    struct ocfs2_xattr_value_root *xv;
1196    size_t size;
1197    int ret = -ENODATA, name_offset, name_len, i;
1198    int uninitialized_var(block_off);
1199
1200    xs->bucket = ocfs2_xattr_bucket_new(inode);
1201    if (!xs->bucket) {
1202        ret = -ENOMEM;
1203        mlog_errno(ret);
1204        goto cleanup;
1205    }
1206
1207    ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1208    if (ret) {
1209        mlog_errno(ret);
1210        goto cleanup;
1211    }
1212
1213    if (xs->not_found) {
1214        ret = -ENODATA;
1215        goto cleanup;
1216    }
1217
1218    xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1219    size = le64_to_cpu(xs->here->xe_value_size);
1220    if (buffer) {
1221        ret = -ERANGE;
1222        if (size > buffer_size)
1223            goto cleanup;
1224
1225        name_offset = le16_to_cpu(xs->here->xe_name_offset);
1226        name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1227        i = xs->here - xs->header->xh_entries;
1228
1229        if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1230            ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1231                                bucket_xh(xs->bucket),
1232                                i,
1233                                &block_off,
1234                                &name_offset);
1235            xs->base = bucket_block(xs->bucket, block_off);
1236        }
1237        if (ocfs2_xattr_is_local(xs->here)) {
1238            memcpy(buffer, (void *)xs->base +
1239                   name_offset + name_len, size);
1240        } else {
1241            xv = (struct ocfs2_xattr_value_root *)
1242                (xs->base + name_offset + name_len);
1243            ret = ocfs2_xattr_get_value_outside(inode, xv,
1244                                buffer, size);
1245            if (ret < 0) {
1246                mlog_errno(ret);
1247                goto cleanup;
1248            }
1249        }
1250    }
1251    ret = size;
1252cleanup:
1253    ocfs2_xattr_bucket_free(xs->bucket);
1254
1255    brelse(xs->xattr_bh);
1256    xs->xattr_bh = NULL;
1257    return ret;
1258}
1259
1260int ocfs2_xattr_get_nolock(struct inode *inode,
1261               struct buffer_head *di_bh,
1262               int name_index,
1263               const char *name,
1264               void *buffer,
1265               size_t buffer_size)
1266{
1267    int ret;
1268    struct ocfs2_dinode *di = NULL;
1269    struct ocfs2_inode_info *oi = OCFS2_I(inode);
1270    struct ocfs2_xattr_search xis = {
1271        .not_found = -ENODATA,
1272    };
1273    struct ocfs2_xattr_search xbs = {
1274        .not_found = -ENODATA,
1275    };
1276
1277    if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1278        return -EOPNOTSUPP;
1279
1280    if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1281        ret = -ENODATA;
1282
1283    xis.inode_bh = xbs.inode_bh = di_bh;
1284    di = (struct ocfs2_dinode *)di_bh->b_data;
1285
1286    ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1287                    buffer_size, &xis);
1288    if (ret == -ENODATA && di->i_xattr_loc)
1289        ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1290                        buffer_size, &xbs);
1291
1292    return ret;
1293}
1294
1295/* ocfs2_xattr_get()
1296 *
1297 * Copy an extended attribute into the buffer provided.
1298 * Buffer is NULL to compute the size of buffer required.
1299 */
1300static int ocfs2_xattr_get(struct inode *inode,
1301               int name_index,
1302               const char *name,
1303               void *buffer,
1304               size_t buffer_size)
1305{
1306    int ret;
1307    struct buffer_head *di_bh = NULL;
1308
1309    ret = ocfs2_inode_lock(inode, &di_bh, 0);
1310    if (ret < 0) {
1311        mlog_errno(ret);
1312        return ret;
1313    }
1314    down_read(&OCFS2_I(inode)->ip_xattr_sem);
1315    ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1316                     name, buffer, buffer_size);
1317    up_read(&OCFS2_I(inode)->ip_xattr_sem);
1318
1319    ocfs2_inode_unlock(inode, 0);
1320
1321    brelse(di_bh);
1322
1323    return ret;
1324}
1325
1326static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1327                       handle_t *handle,
1328                       struct ocfs2_xattr_value_buf *vb,
1329                       const void *value,
1330                       int value_len)
1331{
1332    int ret = 0, i, cp_len;
1333    u16 blocksize = inode->i_sb->s_blocksize;
1334    u32 p_cluster, num_clusters;
1335    u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1336    u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1337    u64 blkno;
1338    struct buffer_head *bh = NULL;
1339    unsigned int ext_flags;
1340    struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1341
1342    BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1343
1344    while (cpos < clusters) {
1345        ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1346                           &num_clusters, &xv->xr_list,
1347                           &ext_flags);
1348        if (ret) {
1349            mlog_errno(ret);
1350            goto out;
1351        }
1352
1353        BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1354
1355        blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1356
1357        for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1358            ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1359                           &bh, NULL);
1360            if (ret) {
1361                mlog_errno(ret);
1362                goto out;
1363            }
1364
1365            ret = ocfs2_journal_access(handle,
1366                           INODE_CACHE(inode),
1367                           bh,
1368                           OCFS2_JOURNAL_ACCESS_WRITE);
1369            if (ret < 0) {
1370                mlog_errno(ret);
1371                goto out;
1372            }
1373
1374            cp_len = value_len > blocksize ? blocksize : value_len;
1375            memcpy(bh->b_data, value, cp_len);
1376            value_len -= cp_len;
1377            value += cp_len;
1378            if (cp_len < blocksize)
1379                memset(bh->b_data + cp_len, 0,
1380                       blocksize - cp_len);
1381
1382            ocfs2_journal_dirty(handle, bh);
1383            brelse(bh);
1384            bh = NULL;
1385
1386            /*
1387             * XXX: do we need to empty all the following
1388             * blocks in this cluster?
1389             */
1390            if (!value_len)
1391                break;
1392        }
1393        cpos += num_clusters;
1394    }
1395out:
1396    brelse(bh);
1397
1398    return ret;
1399}
1400
1401static int ocfs2_xa_check_space_helper(int needed_space, int free_start,
1402                       int num_entries)
1403{
1404    int free_space;
1405
1406    if (!needed_space)
1407        return 0;
1408
1409    free_space = free_start -
1410        sizeof(struct ocfs2_xattr_header) -
1411        (num_entries * sizeof(struct ocfs2_xattr_entry)) -
1412        OCFS2_XATTR_HEADER_GAP;
1413    if (free_space < 0)
1414        return -EIO;
1415    if (free_space < needed_space)
1416        return -ENOSPC;
1417
1418    return 0;
1419}
1420
1421static int ocfs2_xa_journal_access(handle_t *handle, struct ocfs2_xa_loc *loc,
1422                   int type)
1423{
1424    return loc->xl_ops->xlo_journal_access(handle, loc, type);
1425}
1426
1427static void ocfs2_xa_journal_dirty(handle_t *handle, struct ocfs2_xa_loc *loc)
1428{
1429    loc->xl_ops->xlo_journal_dirty(handle, loc);
1430}
1431
1432/* Give a pointer into the storage for the given offset */
1433static void *ocfs2_xa_offset_pointer(struct ocfs2_xa_loc *loc, int offset)
1434{
1435    BUG_ON(offset >= loc->xl_size);
1436    return loc->xl_ops->xlo_offset_pointer(loc, offset);
1437}
1438
1439/*
1440 * Wipe the name+value pair and allow the storage to reclaim it. This
1441 * must be followed by either removal of the entry or a call to
1442 * ocfs2_xa_add_namevalue().
1443 */
1444static void ocfs2_xa_wipe_namevalue(struct ocfs2_xa_loc *loc)
1445{
1446    loc->xl_ops->xlo_wipe_namevalue(loc);
1447}
1448
1449/*
1450 * Find lowest offset to a name+value pair. This is the start of our
1451 * downward-growing free space.
1452 */
1453static int ocfs2_xa_get_free_start(struct ocfs2_xa_loc *loc)
1454{
1455    return loc->xl_ops->xlo_get_free_start(loc);
1456}
1457
1458/* Can we reuse loc->xl_entry for xi? */
1459static int ocfs2_xa_can_reuse_entry(struct ocfs2_xa_loc *loc,
1460                    struct ocfs2_xattr_info *xi)
1461{
1462    return loc->xl_ops->xlo_can_reuse(loc, xi);
1463}
1464
1465/* How much free space is needed to set the new value */
1466static int ocfs2_xa_check_space(struct ocfs2_xa_loc *loc,
1467                struct ocfs2_xattr_info *xi)
1468{
1469    return loc->xl_ops->xlo_check_space(loc, xi);
1470}
1471
1472static void ocfs2_xa_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1473{
1474    loc->xl_ops->xlo_add_entry(loc, name_hash);
1475    loc->xl_entry->xe_name_hash = cpu_to_le32(name_hash);
1476    /*
1477     * We can't leave the new entry's xe_name_offset at zero or
1478     * add_namevalue() will go nuts. We set it to the size of our
1479     * storage so that it can never be less than any other entry.
1480     */
1481    loc->xl_entry->xe_name_offset = cpu_to_le16(loc->xl_size);
1482}
1483
1484static void ocfs2_xa_add_namevalue(struct ocfs2_xa_loc *loc,
1485                   struct ocfs2_xattr_info *xi)
1486{
1487    int size = namevalue_size_xi(xi);
1488    int nameval_offset;
1489    char *nameval_buf;
1490
1491    loc->xl_ops->xlo_add_namevalue(loc, size);
1492    loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
1493    loc->xl_entry->xe_name_len = xi->xi_name_len;
1494    ocfs2_xattr_set_type(loc->xl_entry, xi->xi_name_index);
1495    ocfs2_xattr_set_local(loc->xl_entry,
1496                  xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE);
1497
1498    nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1499    nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
1500    memset(nameval_buf, 0, size);
1501    memcpy(nameval_buf, xi->xi_name, xi->xi_name_len);
1502}
1503
1504static void ocfs2_xa_fill_value_buf(struct ocfs2_xa_loc *loc,
1505                    struct ocfs2_xattr_value_buf *vb)
1506{
1507    int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1508    int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
1509
1510    /* Value bufs are for value trees */
1511    BUG_ON(ocfs2_xattr_is_local(loc->xl_entry));
1512    BUG_ON(namevalue_size_xe(loc->xl_entry) !=
1513           (name_size + OCFS2_XATTR_ROOT_SIZE));
1514
1515    loc->xl_ops->xlo_fill_value_buf(loc, vb);
1516    vb->vb_xv =
1517        (struct ocfs2_xattr_value_root *)ocfs2_xa_offset_pointer(loc,
1518                            nameval_offset +
1519                            name_size);
1520}
1521
1522static int ocfs2_xa_block_journal_access(handle_t *handle,
1523                     struct ocfs2_xa_loc *loc, int type)
1524{
1525    struct buffer_head *bh = loc->xl_storage;
1526    ocfs2_journal_access_func access;
1527
1528    if (loc->xl_size == (bh->b_size -
1529                 offsetof(struct ocfs2_xattr_block,
1530                      xb_attrs.xb_header)))
1531        access = ocfs2_journal_access_xb;
1532    else
1533        access = ocfs2_journal_access_di;
1534    return access(handle, INODE_CACHE(loc->xl_inode), bh, type);
1535}
1536
1537static void ocfs2_xa_block_journal_dirty(handle_t *handle,
1538                     struct ocfs2_xa_loc *loc)
1539{
1540    struct buffer_head *bh = loc->xl_storage;
1541
1542    ocfs2_journal_dirty(handle, bh);
1543}
1544
1545static void *ocfs2_xa_block_offset_pointer(struct ocfs2_xa_loc *loc,
1546                       int offset)
1547{
1548    return (char *)loc->xl_header + offset;
1549}
1550
1551static int ocfs2_xa_block_can_reuse(struct ocfs2_xa_loc *loc,
1552                    struct ocfs2_xattr_info *xi)
1553{
1554    /*
1555     * Block storage is strict. If the sizes aren't exact, we will
1556     * remove the old one and reinsert the new.
1557     */
1558    return namevalue_size_xe(loc->xl_entry) ==
1559        namevalue_size_xi(xi);
1560}
1561
1562static int ocfs2_xa_block_get_free_start(struct ocfs2_xa_loc *loc)
1563{
1564    struct ocfs2_xattr_header *xh = loc->xl_header;
1565    int i, count = le16_to_cpu(xh->xh_count);
1566    int offset, free_start = loc->xl_size;
1567
1568    for (i = 0; i < count; i++) {
1569        offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1570        if (offset < free_start)
1571            free_start = offset;
1572    }
1573
1574    return free_start;
1575}
1576
1577static int ocfs2_xa_block_check_space(struct ocfs2_xa_loc *loc,
1578                      struct ocfs2_xattr_info *xi)
1579{
1580    int count = le16_to_cpu(loc->xl_header->xh_count);
1581    int free_start = ocfs2_xa_get_free_start(loc);
1582    int needed_space = ocfs2_xi_entry_usage(xi);
1583
1584    /*
1585     * Block storage will reclaim the original entry before inserting
1586     * the new value, so we only need the difference. If the new
1587     * entry is smaller than the old one, we don't need anything.
1588     */
1589    if (loc->xl_entry) {
1590        /* Don't need space if we're reusing! */
1591        if (ocfs2_xa_can_reuse_entry(loc, xi))
1592            needed_space = 0;
1593        else
1594            needed_space -= ocfs2_xe_entry_usage(loc->xl_entry);
1595    }
1596    if (needed_space < 0)
1597        needed_space = 0;
1598    return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1599}
1600
1601/*
1602 * Block storage for xattrs keeps the name+value pairs compacted. When
1603 * we remove one, we have to shift any that preceded it towards the end.
1604 */
1605static void ocfs2_xa_block_wipe_namevalue(struct ocfs2_xa_loc *loc)
1606{
1607    int i, offset;
1608    int namevalue_offset, first_namevalue_offset, namevalue_size;
1609    struct ocfs2_xattr_entry *entry = loc->xl_entry;
1610    struct ocfs2_xattr_header *xh = loc->xl_header;
1611    int count = le16_to_cpu(xh->xh_count);
1612
1613    namevalue_offset = le16_to_cpu(entry->xe_name_offset);
1614    namevalue_size = namevalue_size_xe(entry);
1615    first_namevalue_offset = ocfs2_xa_get_free_start(loc);
1616
1617    /* Shift the name+value pairs */
1618    memmove((char *)xh + first_namevalue_offset + namevalue_size,
1619        (char *)xh + first_namevalue_offset,
1620        namevalue_offset - first_namevalue_offset);
1621    memset((char *)xh + first_namevalue_offset, 0, namevalue_size);
1622
1623    /* Now tell xh->xh_entries about it */
1624    for (i = 0; i < count; i++) {
1625        offset = le16_to_cpu(xh->xh_entries[i].xe_name_offset);
1626        if (offset <= namevalue_offset)
1627            le16_add_cpu(&xh->xh_entries[i].xe_name_offset,
1628                     namevalue_size);
1629    }
1630
1631    /*
1632     * Note that we don't update xh_free_start or xh_name_value_len
1633     * because they're not used in block-stored xattrs.
1634     */
1635}
1636
1637static void ocfs2_xa_block_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1638{
1639    int count = le16_to_cpu(loc->xl_header->xh_count);
1640    loc->xl_entry = &(loc->xl_header->xh_entries[count]);
1641    le16_add_cpu(&loc->xl_header->xh_count, 1);
1642    memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1643}
1644
1645static void ocfs2_xa_block_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1646{
1647    int free_start = ocfs2_xa_get_free_start(loc);
1648
1649    loc->xl_entry->xe_name_offset = cpu_to_le16(free_start - size);
1650}
1651
1652static void ocfs2_xa_block_fill_value_buf(struct ocfs2_xa_loc *loc,
1653                      struct ocfs2_xattr_value_buf *vb)
1654{
1655    struct buffer_head *bh = loc->xl_storage;
1656
1657    if (loc->xl_size == (bh->b_size -
1658                 offsetof(struct ocfs2_xattr_block,
1659                      xb_attrs.xb_header)))
1660        vb->vb_access = ocfs2_journal_access_xb;
1661    else
1662        vb->vb_access = ocfs2_journal_access_di;
1663    vb->vb_bh = bh;
1664}
1665
1666/*
1667 * Operations for xattrs stored in blocks. This includes inline inode
1668 * storage and unindexed ocfs2_xattr_blocks.
1669 */
1670static const struct ocfs2_xa_loc_operations ocfs2_xa_block_loc_ops = {
1671    .xlo_journal_access = ocfs2_xa_block_journal_access,
1672    .xlo_journal_dirty = ocfs2_xa_block_journal_dirty,
1673    .xlo_offset_pointer = ocfs2_xa_block_offset_pointer,
1674    .xlo_check_space = ocfs2_xa_block_check_space,
1675    .xlo_can_reuse = ocfs2_xa_block_can_reuse,
1676    .xlo_get_free_start = ocfs2_xa_block_get_free_start,
1677    .xlo_wipe_namevalue = ocfs2_xa_block_wipe_namevalue,
1678    .xlo_add_entry = ocfs2_xa_block_add_entry,
1679    .xlo_add_namevalue = ocfs2_xa_block_add_namevalue,
1680    .xlo_fill_value_buf = ocfs2_xa_block_fill_value_buf,
1681};
1682
1683static int ocfs2_xa_bucket_journal_access(handle_t *handle,
1684                      struct ocfs2_xa_loc *loc, int type)
1685{
1686    struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1687
1688    return ocfs2_xattr_bucket_journal_access(handle, bucket, type);
1689}
1690
1691static void ocfs2_xa_bucket_journal_dirty(handle_t *handle,
1692                      struct ocfs2_xa_loc *loc)
1693{
1694    struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1695
1696    ocfs2_xattr_bucket_journal_dirty(handle, bucket);
1697}
1698
1699static void *ocfs2_xa_bucket_offset_pointer(struct ocfs2_xa_loc *loc,
1700                        int offset)
1701{
1702    struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1703    int block, block_offset;
1704
1705    /* The header is at the front of the bucket */
1706    block = offset >> loc->xl_inode->i_sb->s_blocksize_bits;
1707    block_offset = offset % loc->xl_inode->i_sb->s_blocksize;
1708
1709    return bucket_block(bucket, block) + block_offset;
1710}
1711
1712static int ocfs2_xa_bucket_can_reuse(struct ocfs2_xa_loc *loc,
1713                     struct ocfs2_xattr_info *xi)
1714{
1715    return namevalue_size_xe(loc->xl_entry) >=
1716        namevalue_size_xi(xi);
1717}
1718
1719static int ocfs2_xa_bucket_get_free_start(struct ocfs2_xa_loc *loc)
1720{
1721    struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1722    return le16_to_cpu(bucket_xh(bucket)->xh_free_start);
1723}
1724
1725static int ocfs2_bucket_align_free_start(struct super_block *sb,
1726                     int free_start, int size)
1727{
1728    /*
1729     * We need to make sure that the name+value pair fits within
1730     * one block.
1731     */
1732    if (((free_start - size) >> sb->s_blocksize_bits) !=
1733        ((free_start - 1) >> sb->s_blocksize_bits))
1734        free_start -= free_start % sb->s_blocksize;
1735
1736    return free_start;
1737}
1738
1739static int ocfs2_xa_bucket_check_space(struct ocfs2_xa_loc *loc,
1740                       struct ocfs2_xattr_info *xi)
1741{
1742    int rc;
1743    int count = le16_to_cpu(loc->xl_header->xh_count);
1744    int free_start = ocfs2_xa_get_free_start(loc);
1745    int needed_space = ocfs2_xi_entry_usage(xi);
1746    int size = namevalue_size_xi(xi);
1747    struct super_block *sb = loc->xl_inode->i_sb;
1748
1749    /*
1750     * Bucket storage does not reclaim name+value pairs it cannot
1751     * reuse. They live as holes until the bucket fills, and then
1752     * the bucket is defragmented. However, the bucket can reclaim
1753     * the ocfs2_xattr_entry.
1754     */
1755    if (loc->xl_entry) {
1756        /* Don't need space if we're reusing! */
1757        if (ocfs2_xa_can_reuse_entry(loc, xi))
1758            needed_space = 0;
1759        else
1760            needed_space -= sizeof(struct ocfs2_xattr_entry);
1761    }
1762    BUG_ON(needed_space < 0);
1763
1764    if (free_start < size) {
1765        if (needed_space)
1766            return -ENOSPC;
1767    } else {
1768        /*
1769         * First we check if it would fit in the first place.
1770         * Below, we align the free start to a block. This may
1771         * slide us below the minimum gap. By checking unaligned
1772         * first, we avoid that error.
1773         */
1774        rc = ocfs2_xa_check_space_helper(needed_space, free_start,
1775                         count);
1776        if (rc)
1777            return rc;
1778        free_start = ocfs2_bucket_align_free_start(sb, free_start,
1779                               size);
1780    }
1781    return ocfs2_xa_check_space_helper(needed_space, free_start, count);
1782}
1783
1784static void ocfs2_xa_bucket_wipe_namevalue(struct ocfs2_xa_loc *loc)
1785{
1786    le16_add_cpu(&loc->xl_header->xh_name_value_len,
1787             -namevalue_size_xe(loc->xl_entry));
1788}
1789
1790static void ocfs2_xa_bucket_add_entry(struct ocfs2_xa_loc *loc, u32 name_hash)
1791{
1792    struct ocfs2_xattr_header *xh = loc->xl_header;
1793    int count = le16_to_cpu(xh->xh_count);
1794    int low = 0, high = count - 1, tmp;
1795    struct ocfs2_xattr_entry *tmp_xe;
1796
1797    /*
1798     * We keep buckets sorted by name_hash, so we need to find
1799     * our insert place.
1800     */
1801    while (low <= high && count) {
1802        tmp = (low + high) / 2;
1803        tmp_xe = &xh->xh_entries[tmp];
1804
1805        if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
1806            low = tmp + 1;
1807        else if (name_hash < le32_to_cpu(tmp_xe->xe_name_hash))
1808            high = tmp - 1;
1809        else {
1810            low = tmp;
1811            break;
1812        }
1813    }
1814
1815    if (low != count)
1816        memmove(&xh->xh_entries[low + 1],
1817            &xh->xh_entries[low],
1818            ((count - low) * sizeof(struct ocfs2_xattr_entry)));
1819
1820    le16_add_cpu(&xh->xh_count, 1);
1821    loc->xl_entry = &xh->xh_entries[low];
1822    memset(loc->xl_entry, 0, sizeof(struct ocfs2_xattr_entry));
1823}
1824
1825static void ocfs2_xa_bucket_add_namevalue(struct ocfs2_xa_loc *loc, int size)
1826{
1827    int free_start = ocfs2_xa_get_free_start(loc);
1828    struct ocfs2_xattr_header *xh = loc->xl_header;
1829    struct super_block *sb = loc->xl_inode->i_sb;
1830    int nameval_offset;
1831
1832    free_start = ocfs2_bucket_align_free_start(sb, free_start, size);
1833    nameval_offset = free_start - size;
1834    loc->xl_entry->xe_name_offset = cpu_to_le16(nameval_offset);
1835    xh->xh_free_start = cpu_to_le16(nameval_offset);
1836    le16_add_cpu(&xh->xh_name_value_len, size);
1837
1838}
1839
1840static void ocfs2_xa_bucket_fill_value_buf(struct ocfs2_xa_loc *loc,
1841                       struct ocfs2_xattr_value_buf *vb)
1842{
1843    struct ocfs2_xattr_bucket *bucket = loc->xl_storage;
1844    struct super_block *sb = loc->xl_inode->i_sb;
1845    int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
1846    int size = namevalue_size_xe(loc->xl_entry);
1847    int block_offset = nameval_offset >> sb->s_blocksize_bits;
1848
1849    /* Values are not allowed to straddle block boundaries */
1850    BUG_ON(block_offset !=
1851           ((nameval_offset + size - 1) >> sb->s_blocksize_bits));
1852    /* We expect the bucket to be filled in */
1853    BUG_ON(!bucket->bu_bhs[block_offset]);
1854
1855    vb->vb_access = ocfs2_journal_access;
1856    vb->vb_bh = bucket->bu_bhs[block_offset];
1857}
1858
1859/* Operations for xattrs stored in buckets. */
1860static const struct ocfs2_xa_loc_operations ocfs2_xa_bucket_loc_ops = {
1861    .xlo_journal_access = ocfs2_xa_bucket_journal_access,
1862    .xlo_journal_dirty = ocfs2_xa_bucket_journal_dirty,
1863    .xlo_offset_pointer = ocfs2_xa_bucket_offset_pointer,
1864    .xlo_check_space = ocfs2_xa_bucket_check_space,
1865    .xlo_can_reuse = ocfs2_xa_bucket_can_reuse,
1866    .xlo_get_free_start = ocfs2_xa_bucket_get_free_start,
1867    .xlo_wipe_namevalue = ocfs2_xa_bucket_wipe_namevalue,
1868    .xlo_add_entry = ocfs2_xa_bucket_add_entry,
1869    .xlo_add_namevalue = ocfs2_xa_bucket_add_namevalue,
1870    .xlo_fill_value_buf = ocfs2_xa_bucket_fill_value_buf,
1871};
1872
1873static unsigned int ocfs2_xa_value_clusters(struct ocfs2_xa_loc *loc)
1874{
1875    struct ocfs2_xattr_value_buf vb;
1876
1877    if (ocfs2_xattr_is_local(loc->xl_entry))
1878        return 0;
1879
1880    ocfs2_xa_fill_value_buf(loc, &vb);
1881    return le32_to_cpu(vb.vb_xv->xr_clusters);
1882}
1883
1884static int ocfs2_xa_value_truncate(struct ocfs2_xa_loc *loc, u64 bytes,
1885                   struct ocfs2_xattr_set_ctxt *ctxt)
1886{
1887    int trunc_rc, access_rc;
1888    struct ocfs2_xattr_value_buf vb;
1889
1890    ocfs2_xa_fill_value_buf(loc, &vb);
1891    trunc_rc = ocfs2_xattr_value_truncate(loc->xl_inode, &vb, bytes,
1892                          ctxt);
1893
1894    /*
1895     * The caller of ocfs2_xa_value_truncate() has already called
1896     * ocfs2_xa_journal_access on the loc. However, The truncate code
1897     * calls ocfs2_extend_trans(). This may commit the previous
1898     * transaction and open a new one. If this is a bucket, truncate
1899     * could leave only vb->vb_bh set up for journaling. Meanwhile,
1900     * the caller is expecting to dirty the entire bucket. So we must
1901     * reset the journal work. We do this even if truncate has failed,
1902     * as it could have failed after committing the extend.
1903     */
1904    access_rc = ocfs2_xa_journal_access(ctxt->handle, loc,
1905                        OCFS2_JOURNAL_ACCESS_WRITE);
1906
1907    /* Errors in truncate take precedence */
1908    return trunc_rc ? trunc_rc : access_rc;
1909}
1910
1911static void ocfs2_xa_remove_entry(struct ocfs2_xa_loc *loc)
1912{
1913    int index, count;
1914    struct ocfs2_xattr_header *xh = loc->xl_header;
1915    struct ocfs2_xattr_entry *entry = loc->xl_entry;
1916
1917    ocfs2_xa_wipe_namevalue(loc);
1918    loc->xl_entry = NULL;
1919
1920    le16_add_cpu(&xh->xh_count, -1);
1921    count = le16_to_cpu(xh->xh_count);
1922
1923    /*
1924     * Only zero out the entry if there are more remaining. This is
1925     * important for an empty bucket, as it keeps track of the
1926     * bucket's hash value. It doesn't hurt empty block storage.
1927     */
1928    if (count) {
1929        index = ((char *)entry - (char *)&xh->xh_entries) /
1930            sizeof(struct ocfs2_xattr_entry);
1931        memmove(&xh->xh_entries[index], &xh->xh_entries[index + 1],
1932            (count - index) * sizeof(struct ocfs2_xattr_entry));
1933        memset(&xh->xh_entries[count], 0,
1934               sizeof(struct ocfs2_xattr_entry));
1935    }
1936}
1937
1938/*
1939 * If we have a problem adjusting the size of an external value during
1940 * ocfs2_xa_prepare_entry() or ocfs2_xa_remove(), we may have an xattr
1941 * in an intermediate state. For example, the value may be partially
1942 * truncated.
1943 *
1944 * If the value tree hasn't changed, the extend/truncate went nowhere.
1945 * We have nothing to do. The caller can treat it as a straight error.
1946 *
1947 * If the value tree got partially truncated, we now have a corrupted
1948 * extended attribute. We're going to wipe its entry and leak the
1949 * clusters. Better to leak some storage than leave a corrupt entry.
1950 *
1951 * If the value tree grew, it obviously didn't grow enough for the
1952 * new entry. We're not going to try and reclaim those clusters either.
1953 * If there was already an external value there (orig_clusters != 0),
1954 * the new clusters are attached safely and we can just leave the old
1955 * value in place. If there was no external value there, we remove
1956 * the entry.
1957 *
1958 * This way, the xattr block we store in the journal will be consistent.
1959 * If the size change broke because of the journal, no changes will hit
1960 * disk anyway.
1961 */
1962static void ocfs2_xa_cleanup_value_truncate(struct ocfs2_xa_loc *loc,
1963                        const char *what,
1964                        unsigned int orig_clusters)
1965{
1966    unsigned int new_clusters = ocfs2_xa_value_clusters(loc);
1967    char *nameval_buf = ocfs2_xa_offset_pointer(loc,
1968                le16_to_cpu(loc->xl_entry->xe_name_offset));
1969
1970    if (new_clusters < orig_clusters) {
1971        mlog(ML_ERROR,
1972             "Partial truncate while %s xattr %.*s. Leaking "
1973             "%u clusters and removing the entry\n",
1974             what, loc->xl_entry->xe_name_len, nameval_buf,
1975             orig_clusters - new_clusters);
1976        ocfs2_xa_remove_entry(loc);
1977    } else if (!orig_clusters) {
1978        mlog(ML_ERROR,
1979             "Unable to allocate an external value for xattr "
1980             "%.*s safely. Leaking %u clusters and removing the "
1981             "entry\n",
1982             loc->xl_entry->xe_name_len, nameval_buf,
1983             new_clusters - orig_clusters);
1984        ocfs2_xa_remove_entry(loc);
1985    } else if (new_clusters > orig_clusters)
1986        mlog(ML_ERROR,
1987             "Unable to grow xattr %.*s safely. %u new clusters "
1988             "have been added, but the value will not be "
1989             "modified\n",
1990             loc->xl_entry->xe_name_len, nameval_buf,
1991             new_clusters - orig_clusters);
1992}
1993
1994static int ocfs2_xa_remove(struct ocfs2_xa_loc *loc,
1995               struct ocfs2_xattr_set_ctxt *ctxt)
1996{
1997    int rc = 0;
1998    unsigned int orig_clusters;
1999
2000    if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2001        orig_clusters = ocfs2_xa_value_clusters(loc);
2002        rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2003        if (rc) {
2004            mlog_errno(rc);
2005            /*
2006             * Since this is remove, we can return 0 if
2007             * ocfs2_xa_cleanup_value_truncate() is going to
2008             * wipe the entry anyway. So we check the
2009             * cluster count as well.
2010             */
2011            if (orig_clusters != ocfs2_xa_value_clusters(loc))
2012                rc = 0;
2013            ocfs2_xa_cleanup_value_truncate(loc, "removing",
2014                            orig_clusters);
2015            if (rc)
2016                goto out;
2017        }
2018    }
2019
2020    ocfs2_xa_remove_entry(loc);
2021
2022out:
2023    return rc;
2024}
2025
2026static void ocfs2_xa_install_value_root(struct ocfs2_xa_loc *loc)
2027{
2028    int name_size = OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len);
2029    char *nameval_buf;
2030
2031    nameval_buf = ocfs2_xa_offset_pointer(loc,
2032                le16_to_cpu(loc->xl_entry->xe_name_offset));
2033    memcpy(nameval_buf + name_size, &def_xv, OCFS2_XATTR_ROOT_SIZE);
2034}
2035
2036/*
2037 * Take an existing entry and make it ready for the new value. This
2038 * won't allocate space, but it may free space. It should be ready for
2039 * ocfs2_xa_prepare_entry() to finish the work.
2040 */
2041static int ocfs2_xa_reuse_entry(struct ocfs2_xa_loc *loc,
2042                struct ocfs2_xattr_info *xi,
2043                struct ocfs2_xattr_set_ctxt *ctxt)
2044{
2045    int rc = 0;
2046    int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2047    unsigned int orig_clusters;
2048    char *nameval_buf;
2049    int xe_local = ocfs2_xattr_is_local(loc->xl_entry);
2050    int xi_local = xi->xi_value_len <= OCFS2_XATTR_INLINE_SIZE;
2051
2052    BUG_ON(OCFS2_XATTR_SIZE(loc->xl_entry->xe_name_len) !=
2053           name_size);
2054
2055    nameval_buf = ocfs2_xa_offset_pointer(loc,
2056                le16_to_cpu(loc->xl_entry->xe_name_offset));
2057    if (xe_local) {
2058        memset(nameval_buf + name_size, 0,
2059               namevalue_size_xe(loc->xl_entry) - name_size);
2060        if (!xi_local)
2061            ocfs2_xa_install_value_root(loc);
2062    } else {
2063        orig_clusters = ocfs2_xa_value_clusters(loc);
2064        if (xi_local) {
2065            rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2066            if (rc < 0)
2067                mlog_errno(rc);
2068            else
2069                memset(nameval_buf + name_size, 0,
2070                       namevalue_size_xe(loc->xl_entry) -
2071                       name_size);
2072        } else if (le64_to_cpu(loc->xl_entry->xe_value_size) >
2073               xi->xi_value_len) {
2074            rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len,
2075                             ctxt);
2076            if (rc < 0)
2077                mlog_errno(rc);
2078        }
2079
2080        if (rc) {
2081            ocfs2_xa_cleanup_value_truncate(loc, "reusing",
2082                            orig_clusters);
2083            goto out;
2084        }
2085    }
2086
2087    loc->xl_entry->xe_value_size = cpu_to_le64(xi->xi_value_len);
2088    ocfs2_xattr_set_local(loc->xl_entry, xi_local);
2089
2090out:
2091    return rc;
2092}
2093
2094/*
2095 * Prepares loc->xl_entry to receive the new xattr. This includes
2096 * properly setting up the name+value pair region. If loc->xl_entry
2097 * already exists, it will take care of modifying it appropriately.
2098 *
2099 * Note that this modifies the data. You did journal_access already,
2100 * right?
2101 */
2102static int ocfs2_xa_prepare_entry(struct ocfs2_xa_loc *loc,
2103                  struct ocfs2_xattr_info *xi,
2104                  u32 name_hash,
2105                  struct ocfs2_xattr_set_ctxt *ctxt)
2106{
2107    int rc = 0;
2108    unsigned int orig_clusters;
2109    __le64 orig_value_size = 0;
2110
2111    rc = ocfs2_xa_check_space(loc, xi);
2112    if (rc)
2113        goto out;
2114
2115    if (loc->xl_entry) {
2116        if (ocfs2_xa_can_reuse_entry(loc, xi)) {
2117            orig_value_size = loc->xl_entry->xe_value_size;
2118            rc = ocfs2_xa_reuse_entry(loc, xi, ctxt);
2119            if (rc)
2120                goto out;
2121            goto alloc_value;
2122        }
2123
2124        if (!ocfs2_xattr_is_local(loc->xl_entry)) {
2125            orig_clusters = ocfs2_xa_value_clusters(loc);
2126            rc = ocfs2_xa_value_truncate(loc, 0, ctxt);
2127            if (rc) {
2128                mlog_errno(rc);
2129                ocfs2_xa_cleanup_value_truncate(loc,
2130                                "overwriting",
2131                                orig_clusters);
2132                goto out;
2133            }
2134        }
2135        ocfs2_xa_wipe_namevalue(loc);
2136    } else
2137        ocfs2_xa_add_entry(loc, name_hash);
2138
2139    /*
2140     * If we get here, we have a blank entry. Fill it. We grow our
2141     * name+value pair back from the end.
2142     */
2143    ocfs2_xa_add_namevalue(loc, xi);
2144    if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
2145        ocfs2_xa_install_value_root(loc);
2146
2147alloc_value:
2148    if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2149        orig_clusters = ocfs2_xa_value_clusters(loc);
2150        rc = ocfs2_xa_value_truncate(loc, xi->xi_value_len, ctxt);
2151        if (rc < 0) {
2152            ctxt->set_abort = 1;
2153            ocfs2_xa_cleanup_value_truncate(loc, "growing",
2154                            orig_clusters);
2155            /*
2156             * If we were growing an existing value,
2157             * ocfs2_xa_cleanup_value_truncate() won't remove
2158             * the entry. We need to restore the original value
2159             * size.
2160             */
2161            if (loc->xl_entry) {
2162                BUG_ON(!orig_value_size);
2163                loc->xl_entry->xe_value_size = orig_value_size;
2164            }
2165            mlog_errno(rc);
2166        }
2167    }
2168
2169out:
2170    return rc;
2171}
2172
2173/*
2174 * Store the value portion of the name+value pair. This will skip
2175 * values that are stored externally. Their tree roots were set up
2176 * by ocfs2_xa_prepare_entry().
2177 */
2178static int ocfs2_xa_store_value(struct ocfs2_xa_loc *loc,
2179                struct ocfs2_xattr_info *xi,
2180                struct ocfs2_xattr_set_ctxt *ctxt)
2181{
2182    int rc = 0;
2183    int nameval_offset = le16_to_cpu(loc->xl_entry->xe_name_offset);
2184    int name_size = OCFS2_XATTR_SIZE(xi->xi_name_len);
2185    char *nameval_buf;
2186    struct ocfs2_xattr_value_buf vb;
2187
2188    nameval_buf = ocfs2_xa_offset_pointer(loc, nameval_offset);
2189    if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
2190        ocfs2_xa_fill_value_buf(loc, &vb);
2191        rc = __ocfs2_xattr_set_value_outside(loc->xl_inode,
2192                             ctxt->handle, &vb,
2193                             xi->xi_value,
2194                             xi->xi_value_len);
2195    } else
2196        memcpy(nameval_buf + name_size, xi->xi_value, xi->xi_value_len);
2197
2198    return rc;
2199}
2200
2201static int ocfs2_xa_set(struct ocfs2_xa_loc *loc,
2202            struct ocfs2_xattr_info *xi,
2203            struct ocfs2_xattr_set_ctxt *ctxt)
2204{
2205    int ret;
2206    u32 name_hash = ocfs2_xattr_name_hash(loc->xl_inode, xi->xi_name,
2207                          xi->xi_name_len);
2208
2209    ret = ocfs2_xa_journal_access(ctxt->handle, loc,
2210                      OCFS2_JOURNAL_ACCESS_WRITE);
2211    if (ret) {
2212        mlog_errno(ret);
2213        goto out;
2214    }
2215
2216    /*
2217     * From here on out, everything is going to modify the buffer a
2218     * little. Errors are going to leave the xattr header in a
2219     * sane state. Thus, even with errors we dirty the sucker.
2220     */
2221
2222    /* Don't worry, we are never called with !xi_value and !xl_entry */
2223    if (!xi->xi_value) {
2224        ret = ocfs2_xa_remove(loc, ctxt);
2225        goto out_dirty;
2226    }
2227
2228    ret = ocfs2_xa_prepare_entry(loc, xi, name_hash, ctxt);
2229    if (ret) {
2230        if (ret != -ENOSPC)
2231            mlog_errno(ret);
2232        goto out_dirty;
2233    }
2234
2235    ret = ocfs2_xa_store_value(loc, xi, ctxt);
2236    if (ret)
2237        mlog_errno(ret);
2238
2239out_dirty:
2240    ocfs2_xa_journal_dirty(ctxt->handle, loc);
2241
2242out:
2243    return ret;
2244}
2245
2246static void ocfs2_init_dinode_xa_loc(struct ocfs2_xa_loc *loc,
2247                     struct inode *inode,
2248                     struct buffer_head *bh,
2249                     struct ocfs2_xattr_entry *entry)
2250{
2251    struct ocfs2_dinode *di = (struct ocfs2_dinode *)bh->b_data;
2252
2253    BUG_ON(!(OCFS2_I(inode)->ip_dyn_features & OCFS2_INLINE_XATTR_FL));
2254
2255    loc->xl_inode = inode;
2256    loc->xl_ops = &ocfs2_xa_block_loc_ops;
2257    loc->xl_storage = bh;
2258    loc->xl_entry = entry;
2259    loc->xl_size = le16_to_cpu(di->i_xattr_inline_size);
2260    loc->xl_header =
2261        (struct ocfs2_xattr_header *)(bh->b_data + bh->b_size -
2262                          loc->xl_size);
2263}
2264
2265static void ocfs2_init_xattr_block_xa_loc(struct ocfs2_xa_loc *loc,
2266                      struct inode *inode,
2267                      struct buffer_head *bh,
2268                      struct ocfs2_xattr_entry *entry)
2269{
2270    struct ocfs2_xattr_block *xb =
2271        (struct ocfs2_xattr_block *)bh->b_data;
2272
2273    BUG_ON(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED);
2274
2275    loc->xl_inode = inode;
2276    loc->xl_ops = &ocfs2_xa_block_loc_ops;
2277    loc->xl_storage = bh;
2278    loc->xl_header = &(xb->xb_attrs.xb_header);
2279    loc->xl_entry = entry;
2280    loc->xl_size = bh->b_size - offsetof(struct ocfs2_xattr_block,
2281                         xb_attrs.xb_header);
2282}
2283
2284static void ocfs2_init_xattr_bucket_xa_loc(struct ocfs2_xa_loc *loc,
2285                       struct ocfs2_xattr_bucket *bucket,
2286                       struct ocfs2_xattr_entry *entry)
2287{
2288    loc->xl_inode = bucket->bu_inode;
2289    loc->xl_ops = &ocfs2_xa_bucket_loc_ops;
2290    loc->xl_storage = bucket;
2291    loc->xl_header = bucket_xh(bucket);
2292    loc->xl_entry = entry;
2293    loc->xl_size = OCFS2_XATTR_BUCKET_SIZE;
2294}
2295
2296/*
2297 * In xattr remove, if it is stored outside and refcounted, we may have
2298 * the chance to split the refcount tree. So need the allocators.
2299 */
2300static int ocfs2_lock_xattr_remove_allocators(struct inode *inode,
2301                    struct ocfs2_xattr_value_root *xv,
2302                    struct ocfs2_caching_info *ref_ci,
2303                    struct buffer_head *ref_root_bh,
2304                    struct ocfs2_alloc_context **meta_ac,
2305                    int *ref_credits)
2306{
2307    int ret, meta_add = 0;
2308    u32 p_cluster, num_clusters;
2309    unsigned int ext_flags;
2310
2311    *ref_credits = 0;
2312    ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
2313                       &num_clusters,
2314                       &xv->xr_list,
2315                       &ext_flags);
2316    if (ret) {
2317        mlog_errno(ret);
2318        goto out;
2319    }
2320
2321    if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
2322        goto out;
2323
2324    ret = ocfs2_refcounted_xattr_delete_need(inode, ref_ci,
2325                         ref_root_bh, xv,
2326                         &meta_add, ref_credits);
2327    if (ret) {
2328        mlog_errno(ret);
2329        goto out;
2330    }
2331
2332    ret = ocfs2_reserve_new_metadata_blocks(OCFS2_SB(inode->i_sb),
2333                        meta_add, meta_ac);
2334    if (ret)
2335        mlog_errno(ret);
2336
2337out:
2338    return ret;
2339}
2340
2341static int ocfs2_remove_value_outside(struct inode*inode,
2342                      struct ocfs2_xattr_value_buf *vb,
2343                      struct ocfs2_xattr_header *header,
2344                      struct ocfs2_caching_info *ref_ci,
2345                      struct buffer_head *ref_root_bh)
2346{
2347    int ret = 0, i, ref_credits;
2348    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2349    struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2350    void *val;
2351
2352    ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
2353
2354    for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
2355        struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
2356
2357        if (ocfs2_xattr_is_local(entry))
2358            continue;
2359
2360        val = (void *)header +
2361            le16_to_cpu(entry->xe_name_offset);
2362        vb->vb_xv = (struct ocfs2_xattr_value_root *)
2363            (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
2364
2365        ret = ocfs2_lock_xattr_remove_allocators(inode, vb->vb_xv,
2366                             ref_ci, ref_root_bh,
2367                             &ctxt.meta_ac,
2368                             &ref_credits);
2369
2370        ctxt.handle = ocfs2_start_trans(osb, ref_credits +
2371                    ocfs2_remove_extent_credits(osb->sb));
2372        if (IS_ERR(ctxt.handle)) {
2373            ret = PTR_ERR(ctxt.handle);
2374            mlog_errno(ret);
2375            break;
2376        }
2377
2378        ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
2379        if (ret < 0) {
2380            mlog_errno(ret);
2381            break;
2382        }
2383
2384        ocfs2_commit_trans(osb, ctxt.handle);
2385        if (ctxt.meta_ac) {
2386            ocfs2_free_alloc_context(ctxt.meta_ac);
2387            ctxt.meta_ac = NULL;
2388        }
2389    }
2390
2391    if (ctxt.meta_ac)
2392        ocfs2_free_alloc_context(ctxt.meta_ac);
2393    ocfs2_schedule_truncate_log_flush(osb, 1);
2394    ocfs2_run_deallocs(osb, &ctxt.dealloc);
2395    return ret;
2396}
2397
2398static int ocfs2_xattr_ibody_remove(struct inode *inode,
2399                    struct buffer_head *di_bh,
2400                    struct ocfs2_caching_info *ref_ci,
2401                    struct buffer_head *ref_root_bh)
2402{
2403
2404    struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2405    struct ocfs2_xattr_header *header;
2406    int ret;
2407    struct ocfs2_xattr_value_buf vb = {
2408        .vb_bh = di_bh,
2409        .vb_access = ocfs2_journal_access_di,
2410    };
2411
2412    header = (struct ocfs2_xattr_header *)
2413         ((void *)di + inode->i_sb->s_blocksize -
2414         le16_to_cpu(di->i_xattr_inline_size));
2415
2416    ret = ocfs2_remove_value_outside(inode, &vb, header,
2417                     ref_ci, ref_root_bh);
2418
2419    return ret;
2420}
2421
2422struct ocfs2_rm_xattr_bucket_para {
2423    struct ocfs2_caching_info *ref_ci;
2424    struct buffer_head *ref_root_bh;
2425};
2426
2427static int ocfs2_xattr_block_remove(struct inode *inode,
2428                    struct buffer_head *blk_bh,
2429                    struct ocfs2_caching_info *ref_ci,
2430                    struct buffer_head *ref_root_bh)
2431{
2432    struct ocfs2_xattr_block *xb;
2433    int ret = 0;
2434    struct ocfs2_xattr_value_buf vb = {
2435        .vb_bh = blk_bh,
2436        .vb_access = ocfs2_journal_access_xb,
2437    };
2438    struct ocfs2_rm_xattr_bucket_para args = {
2439        .ref_ci = ref_ci,
2440        .ref_root_bh = ref_root_bh,
2441    };
2442
2443    xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2444    if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2445        struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
2446        ret = ocfs2_remove_value_outside(inode, &vb, header,
2447                         ref_ci, ref_root_bh);
2448    } else
2449        ret = ocfs2_iterate_xattr_index_block(inode,
2450                        blk_bh,
2451                        ocfs2_rm_xattr_cluster,
2452                        &args);
2453
2454    return ret;
2455}
2456
2457static int ocfs2_xattr_free_block(struct inode *inode,
2458                  u64 block,
2459                  struct ocfs2_caching_info *ref_ci,
2460                  struct buffer_head *ref_root_bh)
2461{
2462    struct inode *xb_alloc_inode;
2463    struct buffer_head *xb_alloc_bh = NULL;
2464    struct buffer_head *blk_bh = NULL;
2465    struct ocfs2_xattr_block *xb;
2466    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2467    handle_t *handle;
2468    int ret = 0;
2469    u64 blk, bg_blkno;
2470    u16 bit;
2471
2472    ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
2473    if (ret < 0) {
2474        mlog_errno(ret);
2475        goto out;
2476    }
2477
2478    ret = ocfs2_xattr_block_remove(inode, blk_bh, ref_ci, ref_root_bh);
2479    if (ret < 0) {
2480        mlog_errno(ret);
2481        goto out;
2482    }
2483
2484    xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2485    blk = le64_to_cpu(xb->xb_blkno);
2486    bit = le16_to_cpu(xb->xb_suballoc_bit);
2487    if (xb->xb_suballoc_loc)
2488        bg_blkno = le64_to_cpu(xb->xb_suballoc_loc);
2489    else
2490        bg_blkno = ocfs2_which_suballoc_group(blk, bit);
2491
2492    xb_alloc_inode = ocfs2_get_system_file_inode(osb,
2493                EXTENT_ALLOC_SYSTEM_INODE,
2494                le16_to_cpu(xb->xb_suballoc_slot));
2495    if (!xb_alloc_inode) {
2496        ret = -ENOMEM;
2497        mlog_errno(ret);
2498        goto out;
2499    }
2500    mutex_lock(&xb_alloc_inode->i_mutex);
2501
2502    ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
2503    if (ret < 0) {
2504        mlog_errno(ret);
2505        goto out_mutex;
2506    }
2507
2508    handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
2509    if (IS_ERR(handle)) {
2510        ret = PTR_ERR(handle);
2511        mlog_errno(ret);
2512        goto out_unlock;
2513    }
2514
2515    ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
2516                       bit, bg_blkno, 1);
2517    if (ret < 0)
2518        mlog_errno(ret);
2519
2520    ocfs2_commit_trans(osb, handle);
2521out_unlock:
2522    ocfs2_inode_unlock(xb_alloc_inode, 1);
2523    brelse(xb_alloc_bh);
2524out_mutex:
2525    mutex_unlock(&xb_alloc_inode->i_mutex);
2526    iput(xb_alloc_inode);
2527out:
2528    brelse(blk_bh);
2529    return ret;
2530}
2531
2532/*
2533 * ocfs2_xattr_remove()
2534 *
2535 * Free extended attribute resources associated with this inode.
2536 */
2537int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
2538{
2539    struct ocfs2_inode_info *oi = OCFS2_I(inode);
2540    struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2541    struct ocfs2_refcount_tree *ref_tree = NULL;
2542    struct buffer_head *ref_root_bh = NULL;
2543    struct ocfs2_caching_info *ref_ci = NULL;
2544    handle_t *handle;
2545    int ret;
2546
2547    if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2548        return 0;
2549
2550    if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
2551        return 0;
2552
2553    if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL) {
2554        ret = ocfs2_lock_refcount_tree(OCFS2_SB(inode->i_sb),
2555                           le64_to_cpu(di->i_refcount_loc),
2556                           1, &ref_tree, &ref_root_bh);
2557        if (ret) {
2558            mlog_errno(ret);
2559            goto out;
2560        }
2561        ref_ci = &ref_tree->rf_ci;
2562
2563    }
2564
2565    if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2566        ret = ocfs2_xattr_ibody_remove(inode, di_bh,
2567                           ref_ci, ref_root_bh);
2568        if (ret < 0) {
2569            mlog_errno(ret);
2570            goto out;
2571        }
2572    }
2573
2574    if (di->i_xattr_loc) {
2575        ret = ocfs2_xattr_free_block(inode,
2576                         le64_to_cpu(di->i_xattr_loc),
2577                         ref_ci, ref_root_bh);
2578        if (ret < 0) {
2579            mlog_errno(ret);
2580            goto out;
2581        }
2582    }
2583
2584    handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
2585                   OCFS2_INODE_UPDATE_CREDITS);
2586    if (IS_ERR(handle)) {
2587        ret = PTR_ERR(handle);
2588        mlog_errno(ret);
2589        goto out;
2590    }
2591    ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
2592                      OCFS2_JOURNAL_ACCESS_WRITE);
2593    if (ret) {
2594        mlog_errno(ret);
2595        goto out_commit;
2596    }
2597
2598    di->i_xattr_loc = 0;
2599
2600    spin_lock(&oi->ip_lock);
2601    oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
2602    di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2603    spin_unlock(&oi->ip_lock);
2604
2605    ocfs2_journal_dirty(handle, di_bh);
2606out_commit:
2607    ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
2608out:
2609    if (ref_tree)
2610        ocfs2_unlock_refcount_tree(OCFS2_SB(inode->i_sb), ref_tree, 1);
2611    brelse(ref_root_bh);
2612    return ret;
2613}
2614
2615static int ocfs2_xattr_has_space_inline(struct inode *inode,
2616                    struct ocfs2_dinode *di)
2617{
2618    struct ocfs2_inode_info *oi = OCFS2_I(inode);
2619    unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
2620    int free;
2621
2622    if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
2623        return 0;
2624
2625    if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2626        struct ocfs2_inline_data *idata = &di->id2.i_data;
2627        free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
2628    } else if (ocfs2_inode_is_fast_symlink(inode)) {
2629        free = ocfs2_fast_symlink_chars(inode->i_sb) -
2630            le64_to_cpu(di->i_size);
2631    } else {
2632        struct ocfs2_extent_list *el = &di->id2.i_list;
2633        free = (le16_to_cpu(el->l_count) -
2634            le16_to_cpu(el->l_next_free_rec)) *
2635            sizeof(struct ocfs2_extent_rec);
2636    }
2637    if (free >= xattrsize)
2638        return 1;
2639
2640    return 0;
2641}
2642
2643/*
2644 * ocfs2_xattr_ibody_find()
2645 *
2646 * Find extended attribute in inode block and
2647 * fill search info into struct ocfs2_xattr_search.
2648 */
2649static int ocfs2_xattr_ibody_find(struct inode *inode,
2650                  int name_index,
2651                  const char *name,
2652                  struct ocfs2_xattr_search *xs)
2653{
2654    struct ocfs2_inode_info *oi = OCFS2_I(inode);
2655    struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2656    int ret;
2657    int has_space = 0;
2658
2659    if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2660        return 0;
2661
2662    if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2663        down_read(&oi->ip_alloc_sem);
2664        has_space = ocfs2_xattr_has_space_inline(inode, di);
2665        up_read(&oi->ip_alloc_sem);
2666        if (!has_space)
2667            return 0;
2668    }
2669
2670    xs->xattr_bh = xs->inode_bh;
2671    xs->end = (void *)di + inode->i_sb->s_blocksize;
2672    if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2673        xs->header = (struct ocfs2_xattr_header *)
2674            (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2675    else
2676        xs->header = (struct ocfs2_xattr_header *)
2677            (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2678    xs->base = (void *)xs->header;
2679    xs->here = xs->header->xh_entries;
2680
2681    /* Find the named attribute. */
2682    if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2683        ret = ocfs2_xattr_find_entry(name_index, name, xs);
2684        if (ret && ret != -ENODATA)
2685            return ret;
2686        xs->not_found = ret;
2687    }
2688
2689    return 0;
2690}
2691
2692static int ocfs2_xattr_ibody_init(struct inode *inode,
2693                  struct buffer_head *di_bh,
2694                  struct ocfs2_xattr_set_ctxt *ctxt)
2695{
2696    int ret;
2697    struct ocfs2_inode_info *oi = OCFS2_I(inode);
2698    struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
2699    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2700    unsigned int xattrsize = osb->s_xattr_inline_size;
2701
2702    if (!ocfs2_xattr_has_space_inline(inode, di)) {
2703        ret = -ENOSPC;
2704        goto out;
2705    }
2706
2707    ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode), di_bh,
2708                      OCFS2_JOURNAL_ACCESS_WRITE);
2709    if (ret) {
2710        mlog_errno(ret);
2711        goto out;
2712    }
2713
2714    /*
2715     * Adjust extent record count or inline data size
2716     * to reserve space for extended attribute.
2717     */
2718    if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
2719        struct ocfs2_inline_data *idata = &di->id2.i_data;
2720        le16_add_cpu(&idata->id_count, -xattrsize);
2721    } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
2722        struct ocfs2_extent_list *el = &di->id2.i_list;
2723        le16_add_cpu(&el->l_count, -(xattrsize /
2724                         sizeof(struct ocfs2_extent_rec)));
2725    }
2726    di->i_xattr_inline_size = cpu_to_le16(xattrsize);
2727
2728    spin_lock(&oi->ip_lock);
2729    oi->ip_dyn_features |= OCFS2_INLINE_XATTR_FL|OCFS2_HAS_XATTR_FL;
2730    di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
2731    spin_unlock(&oi->ip_lock);
2732
2733    ocfs2_journal_dirty(ctxt->handle, di_bh);
2734
2735out:
2736    return ret;
2737}
2738
2739/*
2740 * ocfs2_xattr_ibody_set()
2741 *
2742 * Set, replace or remove an extended attribute into inode block.
2743 *
2744 */
2745static int ocfs2_xattr_ibody_set(struct inode *inode,
2746                 struct ocfs2_xattr_info *xi,
2747                 struct ocfs2_xattr_search *xs,
2748                 struct ocfs2_xattr_set_ctxt *ctxt)
2749{
2750    int ret;
2751    struct ocfs2_inode_info *oi = OCFS2_I(inode);
2752    struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2753    struct ocfs2_xa_loc loc;
2754
2755    if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2756        return -ENOSPC;
2757
2758    down_write(&oi->ip_alloc_sem);
2759    if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2760        if (!ocfs2_xattr_has_space_inline(inode, di)) {
2761            ret = -ENOSPC;
2762            goto out;
2763        }
2764    }
2765
2766    if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2767        ret = ocfs2_xattr_ibody_init(inode, xs->inode_bh, ctxt);
2768        if (ret) {
2769            if (ret != -ENOSPC)
2770                mlog_errno(ret);
2771            goto out;
2772        }
2773    }
2774
2775    ocfs2_init_dinode_xa_loc(&loc, inode, xs->inode_bh,
2776                 xs->not_found ? NULL : xs->here);
2777    ret = ocfs2_xa_set(&loc, xi, ctxt);
2778    if (ret) {
2779        if (ret != -ENOSPC)
2780            mlog_errno(ret);
2781        goto out;
2782    }
2783    xs->here = loc.xl_entry;
2784
2785out:
2786    up_write(&oi->ip_alloc_sem);
2787
2788    return ret;
2789}
2790
2791/*
2792 * ocfs2_xattr_block_find()
2793 *
2794 * Find extended attribute in external block and
2795 * fill search info into struct ocfs2_xattr_search.
2796 */
2797static int ocfs2_xattr_block_find(struct inode *inode,
2798                  int name_index,
2799                  const char *name,
2800                  struct ocfs2_xattr_search *xs)
2801{
2802    struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2803    struct buffer_head *blk_bh = NULL;
2804    struct ocfs2_xattr_block *xb;
2805    int ret = 0;
2806
2807    if (!di->i_xattr_loc)
2808        return ret;
2809
2810    ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2811                     &blk_bh);
2812    if (ret < 0) {
2813        mlog_errno(ret);
2814        return ret;
2815    }
2816
2817    xs->xattr_bh = blk_bh;
2818    xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2819
2820    if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2821        xs->header = &xb->xb_attrs.xb_header;
2822        xs->base = (void *)xs->header;
2823        xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2824        xs->here = xs->header->xh_entries;
2825
2826        ret = ocfs2_xattr_find_entry(name_index, name, xs);
2827    } else
2828        ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2829                           name_index,
2830                           name, xs);
2831
2832    if (ret && ret != -ENODATA) {
2833        xs->xattr_bh = NULL;
2834        goto cleanup;
2835    }
2836    xs->not_found = ret;
2837    return 0;
2838cleanup:
2839    brelse(blk_bh);
2840
2841    return ret;
2842}
2843
2844static int ocfs2_create_xattr_block(struct inode *inode,
2845                    struct buffer_head *inode_bh,
2846                    struct ocfs2_xattr_set_ctxt *ctxt,
2847                    int indexed,
2848                    struct buffer_head **ret_bh)
2849{
2850    int ret;
2851    u16 suballoc_bit_start;
2852    u32 num_got;
2853    u64 suballoc_loc, first_blkno;
2854    struct ocfs2_dinode *di = (struct ocfs2_dinode *)inode_bh->b_data;
2855    struct buffer_head *new_bh = NULL;
2856    struct ocfs2_xattr_block *xblk;
2857
2858    ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2859                      inode_bh, OCFS2_JOURNAL_ACCESS_CREATE);
2860    if (ret < 0) {
2861        mlog_errno(ret);
2862        goto end;
2863    }
2864
2865    ret = ocfs2_claim_metadata(ctxt->handle, ctxt->meta_ac, 1,
2866                   &suballoc_loc, &suballoc_bit_start,
2867                   &num_got, &first_blkno);
2868    if (ret < 0) {
2869        mlog_errno(ret);
2870        goto end;
2871    }
2872
2873    new_bh = sb_getblk(inode->i_sb, first_blkno);
2874    ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2875
2876    ret = ocfs2_journal_access_xb(ctxt->handle, INODE_CACHE(inode),
2877                      new_bh,
2878                      OCFS2_JOURNAL_ACCESS_CREATE);
2879    if (ret < 0) {
2880        mlog_errno(ret);
2881        goto end;
2882    }
2883
2884    /* Initialize ocfs2_xattr_block */
2885    xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2886    memset(xblk, 0, inode->i_sb->s_blocksize);
2887    strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2888    xblk->xb_suballoc_slot = cpu_to_le16(ctxt->meta_ac->ac_alloc_slot);
2889    xblk->xb_suballoc_loc = cpu_to_le64(suballoc_loc);
2890    xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2891    xblk->xb_fs_generation =
2892        cpu_to_le32(OCFS2_SB(inode->i_sb)->fs_generation);
2893    xblk->xb_blkno = cpu_to_le64(first_blkno);
2894    if (indexed) {
2895        struct ocfs2_xattr_tree_root *xr = &xblk->xb_attrs.xb_root;
2896        xr->xt_clusters = cpu_to_le32(1);
2897        xr->xt_last_eb_blk = 0;
2898        xr->xt_list.l_tree_depth = 0;
2899        xr->xt_list.l_count = cpu_to_le16(
2900                    ocfs2_xattr_recs_per_xb(inode->i_sb));
2901        xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2902        xblk->xb_flags = cpu_to_le16(OCFS2_XATTR_INDEXED);
2903    }
2904    ocfs2_journal_dirty(ctxt->handle, new_bh);
2905
2906    /* Add it to the inode */
2907    di->i_xattr_loc = cpu_to_le64(first_blkno);
2908
2909    spin_lock(&OCFS2_I(inode)->ip_lock);
2910    OCFS2_I(inode)->ip_dyn_features |= OCFS2_HAS_XATTR_FL;
2911    di->i_dyn_features = cpu_to_le16(OCFS2_I(inode)->ip_dyn_features);
2912    spin_unlock(&OCFS2_I(inode)->ip_lock);
2913
2914    ocfs2_journal_dirty(ctxt->handle, inode_bh);
2915
2916    *ret_bh = new_bh;
2917    new_bh = NULL;
2918
2919end:
2920    brelse(new_bh);
2921    return ret;
2922}
2923
2924/*
2925 * ocfs2_xattr_block_set()
2926 *
2927 * Set, replace or remove an extended attribute into external block.
2928 *
2929 */
2930static int ocfs2_xattr_block_set(struct inode *inode,
2931                 struct ocfs2_xattr_info *xi,
2932                 struct ocfs2_xattr_search *xs,
2933                 struct ocfs2_xattr_set_ctxt *ctxt)
2934{
2935    struct buffer_head *new_bh = NULL;
2936    struct ocfs2_xattr_block *xblk = NULL;
2937    int ret;
2938    struct ocfs2_xa_loc loc;
2939
2940    if (!xs->xattr_bh) {
2941        ret = ocfs2_create_xattr_block(inode, xs->inode_bh, ctxt,
2942                           0, &new_bh);
2943        if (ret) {
2944            mlog_errno(ret);
2945            goto end;
2946        }
2947
2948        xs->xattr_bh = new_bh;
2949        xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2950        xs->header = &xblk->xb_attrs.xb_header;
2951        xs->base = (void *)xs->header;
2952        xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2953        xs->here = xs->header->xh_entries;
2954    } else
2955        xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2956
2957    if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2958        ocfs2_init_xattr_block_xa_loc(&loc, inode, xs->xattr_bh,
2959                          xs->not_found ? NULL : xs->here);
2960
2961        ret = ocfs2_xa_set(&loc, xi, ctxt);
2962        if (!ret)
2963            xs->here = loc.xl_entry;
2964        else if ((ret != -ENOSPC) || ctxt->set_abort)
2965            goto end;
2966        else {
2967            ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2968            if (ret)
2969                goto end;
2970        }
2971    }
2972
2973    if (le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)
2974        ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2975
2976end:
2977    return ret;
2978}
2979
2980/* Check whether the new xattr can be inserted into the inode. */
2981static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2982                       struct ocfs2_xattr_info *xi,
2983                       struct ocfs2_xattr_search *xs)
2984{
2985    struct ocfs2_xattr_entry *last;
2986    int free, i;
2987    size_t min_offs = xs->end - xs->base;
2988
2989    if (!xs->header)
2990        return 0;
2991
2992    last = xs->header->xh_entries;
2993
2994    for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2995        size_t offs = le16_to_cpu(last->xe_name_offset);
2996        if (offs < min_offs)
2997            min_offs = offs;
2998        last += 1;
2999    }
3000
3001    free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
3002    if (free < 0)
3003        return 0;
3004
3005    BUG_ON(!xs->not_found);
3006
3007    if (free >= (sizeof(struct ocfs2_xattr_entry) + namevalue_size_xi(xi)))
3008        return 1;
3009
3010    return 0;
3011}
3012
3013static int ocfs2_calc_xattr_set_need(struct inode *inode,
3014                     struct ocfs2_dinode *di,
3015                     struct ocfs2_xattr_info *xi,
3016                     struct ocfs2_xattr_search *xis,
3017                     struct ocfs2_xattr_search *xbs,
3018                     int *clusters_need,
3019                     int *meta_need,
3020                     int *credits_need)
3021{
3022    int ret = 0, old_in_xb = 0;
3023    int clusters_add = 0, meta_add = 0, credits = 0;
3024    struct buffer_head *bh = NULL;
3025    struct ocfs2_xattr_block *xb = NULL;
3026    struct ocfs2_xattr_entry *xe = NULL;
3027    struct ocfs2_xattr_value_root *xv = NULL;
3028    char *base = NULL;
3029    int name_offset, name_len = 0;
3030    u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3031                            xi->xi_value_len);
3032    u64 value_size;
3033
3034    /*
3035     * Calculate the clusters we need to write.
3036     * No matter whether we replace an old one or add a new one,
3037     * we need this for writing.
3038     */
3039    if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE)
3040        credits += new_clusters *
3041               ocfs2_clusters_to_blocks(inode->i_sb, 1);
3042
3043    if (xis->not_found && xbs->not_found) {
3044        credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3045
3046        if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3047            clusters_add += new_clusters;
3048            credits += ocfs2_calc_extend_credits(inode->i_sb,
3049                            &def_xv.xv.xr_list,
3050                            new_clusters);
3051        }
3052
3053        goto meta_guess;
3054    }
3055
3056    if (!xis->not_found) {
3057        xe = xis->here;
3058        name_offset = le16_to_cpu(xe->xe_name_offset);
3059        name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3060        base = xis->base;
3061        credits += OCFS2_INODE_UPDATE_CREDITS;
3062    } else {
3063        int i, block_off = 0;
3064        xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3065        xe = xbs->here;
3066        name_offset = le16_to_cpu(xe->xe_name_offset);
3067        name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3068        i = xbs->here - xbs->header->xh_entries;
3069        old_in_xb = 1;
3070
3071        if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3072            ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3073                            bucket_xh(xbs->bucket),
3074                            i, &block_off,
3075                            &name_offset);
3076            base = bucket_block(xbs->bucket, block_off);
3077            credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3078        } else {
3079            base = xbs->base;
3080            credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
3081        }
3082    }
3083
3084    /*
3085     * delete a xattr doesn't need metadata and cluster allocation.
3086     * so just calculate the credits and return.
3087     *
3088     * The credits for removing the value tree will be extended
3089     * by ocfs2_remove_extent itself.
3090     */
3091    if (!xi->xi_value) {
3092        if (!ocfs2_xattr_is_local(xe))
3093            credits += ocfs2_remove_extent_credits(inode->i_sb);
3094
3095        goto out;
3096    }
3097
3098    /* do cluster allocation guess first. */
3099    value_size = le64_to_cpu(xe->xe_value_size);
3100
3101    if (old_in_xb) {
3102        /*
3103         * In xattr set, we always try to set the xe in inode first,
3104         * so if it can be inserted into inode successfully, the old
3105         * one will be removed from the xattr block, and this xattr
3106         * will be inserted into inode as a new xattr in inode.
3107         */
3108        if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
3109            clusters_add += new_clusters;
3110            credits += ocfs2_remove_extent_credits(inode->i_sb) +
3111                    OCFS2_INODE_UPDATE_CREDITS;
3112            if (!ocfs2_xattr_is_local(xe))
3113                credits += ocfs2_calc_extend_credits(
3114                            inode->i_sb,
3115                            &def_xv.xv.xr_list,
3116                            new_clusters);
3117            goto out;
3118        }
3119    }
3120
3121    if (xi->xi_value_len > OCFS2_XATTR_INLINE_SIZE) {
3122        /* the new values will be stored outside. */
3123        u32 old_clusters = 0;
3124
3125        if (!ocfs2_xattr_is_local(xe)) {
3126            old_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
3127                                 value_size);
3128            xv = (struct ocfs2_xattr_value_root *)
3129                 (base + name_offset + name_len);
3130            value_size = OCFS2_XATTR_ROOT_SIZE;
3131        } else
3132            xv = &def_xv.xv;
3133
3134        if (old_clusters >= new_clusters) {
3135            credits += ocfs2_remove_extent_credits(inode->i_sb);
3136            goto out;
3137        } else {
3138            meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
3139            clusters_add += new_clusters - old_clusters;
3140            credits += ocfs2_calc_extend_credits(inode->i_sb,
3141                                 &xv->xr_list,
3142                                 new_clusters -
3143                                 old_clusters);
3144            if (value_size >= OCFS2_XATTR_ROOT_SIZE)
3145                goto out;
3146        }
3147    } else {
3148        /*
3149         * Now the new value will be stored inside. So if the new
3150         * value is smaller than the size of value root or the old
3151         * value, we don't need any allocation, otherwise we have
3152         * to guess metadata allocation.
3153         */
3154        if ((ocfs2_xattr_is_local(xe) &&
3155             (value_size >= xi->xi_value_len)) ||
3156            (!ocfs2_xattr_is_local(xe) &&
3157             OCFS2_XATTR_ROOT_SIZE >= xi->xi_value_len))
3158            goto out;
3159    }
3160
3161meta_guess:
3162    /* calculate metadata allocation. */
3163    if (di->i_xattr_loc) {
3164        if (!xbs->xattr_bh) {
3165            ret = ocfs2_read_xattr_block(inode,
3166                             le64_to_cpu(di->i_xattr_loc),
3167                             &bh);
3168            if (ret) {
3169                mlog_errno(ret);
3170                goto out;
3171            }
3172
3173            xb = (struct ocfs2_xattr_block *)bh->b_data;
3174        } else
3175            xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
3176
3177        /*
3178         * If there is already an xattr tree, good, we can calculate
3179         * like other b-trees. Otherwise we may have the chance of
3180         * create a tree, the credit calculation is borrowed from
3181         * ocfs2_calc_extend_credits with root_el = NULL. And the
3182         * new tree will be cluster based, so no meta is needed.
3183         */
3184        if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
3185            struct ocfs2_extent_list *el =
3186                 &xb->xb_attrs.xb_root.xt_list;
3187            meta_add += ocfs2_extend_meta_needed(el);
3188            credits += ocfs2_calc_extend_credits(inode->i_sb,
3189                                 el, 1);
3190        } else
3191            credits += OCFS2_SUBALLOC_ALLOC + 1;
3192
3193        /*
3194         * This cluster will be used either for new bucket or for
3195         * new xattr block.
3196         * If the cluster size is the same as the bucket size, one
3197         * more is needed since we may need to extend the bucket
3198         * also.
3199         */
3200        clusters_add += 1;
3201        credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3202        if (OCFS2_XATTR_BUCKET_SIZE ==
3203            OCFS2_SB(inode->i_sb)->s_clustersize) {
3204            credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3205            clusters_add += 1;
3206        }
3207    } else {
3208        meta_add += 1;
3209        credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
3210    }
3211out:
3212    if (clusters_need)
3213        *clusters_need = clusters_add;
3214    if (meta_need)
3215        *meta_need = meta_add;
3216    if (credits_need)
3217        *credits_need = credits;
3218    brelse(bh);
3219    return ret;
3220}
3221
3222static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
3223                     struct ocfs2_dinode *di,
3224                     struct ocfs2_xattr_info *xi,
3225                     struct ocfs2_xattr_search *xis,
3226                     struct ocfs2_xattr_search *xbs,
3227                     struct ocfs2_xattr_set_ctxt *ctxt,
3228                     int extra_meta,
3229                     int *credits)
3230{
3231    int clusters_add, meta_add, ret;
3232    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3233
3234    memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
3235
3236    ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
3237
3238    ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
3239                    &clusters_add, &meta_add, credits);
3240    if (ret) {
3241        mlog_errno(ret);
3242        return ret;
3243    }
3244
3245    meta_add += extra_meta;
3246    trace_ocfs2_init_xattr_set_ctxt(xi->xi_name, meta_add,
3247                    clusters_add, *credits);
3248
3249    if (meta_add) {
3250        ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
3251                            &ctxt->meta_ac);
3252        if (ret) {
3253            mlog_errno(ret);
3254            goto out;
3255        }
3256    }
3257
3258    if (clusters_add) {
3259        ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
3260        if (ret)
3261            mlog_errno(ret);
3262    }
3263out:
3264    if (ret) {
3265        if (ctxt->meta_ac) {
3266            ocfs2_free_alloc_context(ctxt->meta_ac);
3267            ctxt->meta_ac = NULL;
3268        }
3269
3270        /*
3271         * We cannot have an error and a non null ctxt->data_ac.
3272         */
3273    }
3274
3275    return ret;
3276}
3277
3278static int __ocfs2_xattr_set_handle(struct inode *inode,
3279                    struct ocfs2_dinode *di,
3280                    struct ocfs2_xattr_info *xi,
3281                    struct ocfs2_xattr_search *xis,
3282                    struct ocfs2_xattr_search *xbs,
3283                    struct ocfs2_xattr_set_ctxt *ctxt)
3284{
3285    int ret = 0, credits, old_found;
3286
3287    if (!xi->xi_value) {
3288        /* Remove existing extended attribute */
3289        if (!xis->not_found)
3290            ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3291        else if (!xbs->not_found)
3292            ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3293    } else {
3294        /* We always try to set extended attribute into inode first*/
3295        ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
3296        if (!ret && !xbs->not_found) {
3297            /*
3298             * If succeed and that extended attribute existing in
3299             * external block, then we will remove it.
3300             */
3301            xi->xi_value = NULL;
3302            xi->xi_value_len = 0;
3303
3304            old_found = xis->not_found;
3305            xis->not_found = -ENODATA;
3306            ret = ocfs2_calc_xattr_set_need(inode,
3307                            di,
3308                            xi,
3309                            xis,
3310                            xbs,
3311                            NULL,
3312                            NULL,
3313                            &credits);
3314            xis->not_found = old_found;
3315            if (ret) {
3316                mlog_errno(ret);
3317                goto out;
3318            }
3319
3320            ret = ocfs2_extend_trans(ctxt->handle, credits);
3321            if (ret) {
3322                mlog_errno(ret);
3323                goto out;
3324            }
3325            ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3326        } else if ((ret == -ENOSPC) && !ctxt->set_abort) {
3327            if (di->i_xattr_loc && !xbs->xattr_bh) {
3328                ret = ocfs2_xattr_block_find(inode,
3329                                 xi->xi_name_index,
3330                                 xi->xi_name, xbs);
3331                if (ret)
3332                    goto out;
3333
3334                old_found = xis->not_found;
3335                xis->not_found = -ENODATA;
3336                ret = ocfs2_calc_xattr_set_need(inode,
3337                                di,
3338                                xi,
3339                                xis,
3340                                xbs,
3341                                NULL,
3342                                NULL,
3343                                &credits);
3344                xis->not_found = old_found;
3345                if (ret) {
3346                    mlog_errno(ret);
3347                    goto out;
3348                }
3349
3350                ret = ocfs2_extend_trans(ctxt->handle, credits);
3351                if (ret) {
3352                    mlog_errno(ret);
3353                    goto out;
3354                }
3355            }
3356            /*
3357             * If no space in inode, we will set extended attribute
3358             * into external block.
3359             */
3360            ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
3361            if (ret)
3362                goto out;
3363            if (!xis->not_found) {
3364                /*
3365                 * If succeed and that extended attribute
3366                 * existing in inode, we will remove it.
3367                 */
3368                xi->xi_value = NULL;
3369                xi->xi_value_len = 0;
3370                xbs->not_found = -ENODATA;
3371                ret = ocfs2_calc_xattr_set_need(inode,
3372                                di,
3373                                xi,
3374                                xis,
3375                                xbs,
3376                                NULL,
3377                                NULL,
3378                                &credits);
3379                if (ret) {
3380                    mlog_errno(ret);
3381                    goto out;
3382                }
3383
3384                ret = ocfs2_extend_trans(ctxt->handle, credits);
3385                if (ret) {
3386                    mlog_errno(ret);
3387                    goto out;
3388                }
3389                ret = ocfs2_xattr_ibody_set(inode, xi,
3390                                xis, ctxt);
3391            }
3392        }
3393    }
3394
3395    if (!ret) {
3396        /* Update inode ctime. */
3397        ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
3398                          xis->inode_bh,
3399                          OCFS2_JOURNAL_ACCESS_WRITE);
3400        if (ret) {
3401            mlog_errno(ret);
3402            goto out;
3403        }
3404
3405        inode->i_ctime = CURRENT_TIME;
3406        di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
3407        di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
3408        ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
3409    }
3410out:
3411    return ret;
3412}
3413
3414/*
3415 * This function only called duing creating inode
3416 * for init security/acl xattrs of the new inode.
3417 * All transanction credits have been reserved in mknod.
3418 */
3419int ocfs2_xattr_set_handle(handle_t *handle,
3420               struct inode *inode,
3421               struct buffer_head *di_bh,
3422               int name_index,
3423               const char *name,
3424               const void *value,
3425               size_t value_len,
3426               int flags,
3427               struct ocfs2_alloc_context *meta_ac,
3428               struct ocfs2_alloc_context *data_ac)
3429{
3430    struct ocfs2_dinode *di;
3431    int ret;
3432
3433    struct ocfs2_xattr_info xi = {
3434        .xi_name_index = name_index,
3435        .xi_name = name,
3436        .xi_name_len = strlen(name),
3437        .xi_value = value,
3438        .xi_value_len = value_len,
3439    };
3440
3441    struct ocfs2_xattr_search xis = {
3442        .not_found = -ENODATA,
3443    };
3444
3445    struct ocfs2_xattr_search xbs = {
3446        .not_found = -ENODATA,
3447    };
3448
3449    struct ocfs2_xattr_set_ctxt ctxt = {
3450        .handle = handle,
3451        .meta_ac = meta_ac,
3452        .data_ac = data_ac,
3453    };
3454
3455    if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3456        return -EOPNOTSUPP;
3457
3458    /*
3459     * In extreme situation, may need xattr bucket when
3460     * block size is too small. And we have already reserved
3461     * the credits for bucket in mknod.
3462     */
3463    if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
3464        xbs.bucket = ocfs2_xattr_bucket_new(inode);
3465        if (!xbs.bucket) {
3466            mlog_errno(-ENOMEM);
3467            return -ENOMEM;
3468        }
3469    }
3470
3471    xis.inode_bh = xbs.inode_bh = di_bh;
3472    di = (struct ocfs2_dinode *)di_bh->b_data;
3473
3474    down_write(&OCFS2_I(inode)->ip_xattr_sem);
3475
3476    ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3477    if (ret)
3478        goto cleanup;
3479    if (xis.not_found) {
3480        ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3481        if (ret)
3482            goto cleanup;
3483    }
3484
3485    ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3486
3487cleanup:
3488    up_write(&OCFS2_I(inode)->ip_xattr_sem);
3489    brelse(xbs.xattr_bh);
3490    ocfs2_xattr_bucket_free(xbs.bucket);
3491
3492    return ret;
3493}
3494
3495/*
3496 * ocfs2_xattr_set()
3497 *
3498 * Set, replace or remove an extended attribute for this inode.
3499 * value is NULL to remove an existing extended attribute, else either
3500 * create or replace an extended attribute.
3501 */
3502int ocfs2_xattr_set(struct inode *inode,
3503            int name_index,
3504            const char *name,
3505            const void *value,
3506            size_t value_len,
3507            int flags)
3508{
3509    struct buffer_head *di_bh = NULL;
3510    struct ocfs2_dinode *di;
3511    int ret, credits, ref_meta = 0, ref_credits = 0;
3512    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3513    struct inode *tl_inode = osb->osb_tl_inode;
3514    struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
3515    struct ocfs2_refcount_tree *ref_tree = NULL;
3516
3517    struct ocfs2_xattr_info xi = {
3518        .xi_name_index = name_index,
3519        .xi_name = name,
3520        .xi_name_len = strlen(name),
3521        .xi_value = value,
3522        .xi_value_len = value_len,
3523    };
3524
3525    struct ocfs2_xattr_search xis = {
3526        .not_found = -ENODATA,
3527    };
3528
3529    struct ocfs2_xattr_search xbs = {
3530        .not_found = -ENODATA,
3531    };
3532
3533    if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
3534        return -EOPNOTSUPP;
3535
3536    /*
3537     * Only xbs will be used on indexed trees. xis doesn't need a
3538     * bucket.
3539     */
3540    xbs.bucket = ocfs2_xattr_bucket_new(inode);
3541    if (!xbs.bucket) {
3542        mlog_errno(-ENOMEM);
3543        return -ENOMEM;
3544    }
3545
3546    ret = ocfs2_inode_lock(inode, &di_bh, 1);
3547    if (ret < 0) {
3548        mlog_errno(ret);
3549        goto cleanup_nolock;
3550    }
3551    xis.inode_bh = xbs.inode_bh = di_bh;
3552    di = (struct ocfs2_dinode *)di_bh->b_data;
3553
3554    down_write(&OCFS2_I(inode)->ip_xattr_sem);
3555    /*
3556     * Scan inode and external block to find the same name
3557     * extended attribute and collect search information.
3558     */
3559    ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
3560    if (ret)
3561        goto cleanup;
3562    if (xis.not_found) {
3563        ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
3564        if (ret)
3565            goto cleanup;
3566    }
3567
3568    if (xis.not_found && xbs.not_found) {
3569        ret = -ENODATA;
3570        if (flags & XATTR_REPLACE)
3571            goto cleanup;
3572        ret = 0;
3573        if (!value)
3574            goto cleanup;
3575    } else {
3576        ret = -EEXIST;
3577        if (flags & XATTR_CREATE)
3578            goto cleanup;
3579    }
3580
3581    /* Check whether the value is refcounted and do some preparation. */
3582    if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
3583        (!xis.not_found || !xbs.not_found)) {
3584        ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
3585                           &xis, &xbs, &ref_tree,
3586                           &ref_meta, &ref_credits);
3587        if (ret) {
3588            mlog_errno(ret);
3589            goto cleanup;
3590        }
3591    }
3592
3593    mutex_lock(&tl_inode->i_mutex);
3594
3595    if (ocfs2_truncate_log_needs_flush(osb)) {
3596        ret = __ocfs2_flush_truncate_log(osb);
3597        if (ret < 0) {
3598            mutex_unlock(&tl_inode->i_mutex);
3599            mlog_errno(ret);
3600            goto cleanup;
3601        }
3602    }
3603    mutex_unlock(&tl_inode->i_mutex);
3604
3605    ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
3606                    &xbs, &ctxt, ref_meta, &credits);
3607    if (ret) {
3608        mlog_errno(ret);
3609        goto cleanup;
3610    }
3611
3612    /* we need to update inode's ctime field, so add credit for it. */
3613    credits += OCFS2_INODE_UPDATE_CREDITS;
3614    ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
3615    if (IS_ERR(ctxt.handle)) {
3616        ret = PTR_ERR(ctxt.handle);
3617        mlog_errno(ret);
3618        goto cleanup;
3619    }
3620
3621    ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
3622
3623    ocfs2_commit_trans(osb, ctxt.handle);
3624
3625    if (ctxt.data_ac)
3626        ocfs2_free_alloc_context(ctxt.data_ac);
3627    if (ctxt.meta_ac)
3628        ocfs2_free_alloc_context(ctxt.meta_ac);
3629    if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
3630        ocfs2_schedule_truncate_log_flush(osb, 1);
3631    ocfs2_run_deallocs(osb, &ctxt.dealloc);
3632
3633cleanup:
3634    if (ref_tree)
3635        ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
3636    up_write(&OCFS2_I(inode)->ip_xattr_sem);
3637    if (!value && !ret) {
3638        ret = ocfs2_try_remove_refcount_tree(inode, di_bh);
3639        if (ret)
3640            mlog_errno(ret);
3641    }
3642    ocfs2_inode_unlock(inode, 1);
3643cleanup_nolock:
3644    brelse(di_bh);
3645    brelse(xbs.xattr_bh);
3646    ocfs2_xattr_bucket_free(xbs.bucket);
3647
3648    return ret;
3649}
3650
3651/*
3652 * Find the xattr extent rec which may contains name_hash.
3653 * e_cpos will be the first name hash of the xattr rec.
3654 * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
3655 */
3656static int ocfs2_xattr_get_rec(struct inode *inode,
3657                   u32 name_hash,
3658                   u64 *p_blkno,
3659                   u32 *e_cpos,
3660                   u32 *num_clusters,
3661                   struct ocfs2_extent_list *el)
3662{
3663    int ret = 0, i;
3664    struct buffer_head *eb_bh = NULL;
3665    struct ocfs2_extent_block *eb;
3666    struct ocfs2_extent_rec *rec = NULL;
3667    u64 e_blkno = 0;
3668
3669    if (el->l_tree_depth) {
3670        ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
3671                      &eb_bh);
3672        if (ret) {
3673            mlog_errno(ret);
3674            goto out;
3675        }
3676
3677        eb = (struct ocfs2_extent_block *) eb_bh->b_data;
3678        el = &eb->h_list;
3679
3680        if (el->l_tree_depth) {
3681            ocfs2_error(inode->i_sb,
3682                    "Inode %lu has non zero tree depth in "
3683                    "xattr tree block %llu\n", inode->i_ino,
3684                    (unsigned long long)eb_bh->b_blocknr);
3685            ret = -EROFS;
3686            goto out;
3687        }
3688    }
3689
3690    for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
3691        rec = &el->l_recs[i];
3692
3693        if (le32_to_cpu(rec->e_cpos) <= name_hash) {
3694            e_blkno = le64_to_cpu(rec->e_blkno);
3695            break;
3696        }
3697    }
3698
3699    if (!e_blkno) {
3700        ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
3701                "record (%u, %u, 0) in xattr", inode->i_ino,
3702                le32_to_cpu(rec->e_cpos),
3703                ocfs2_rec_clusters(el, rec));
3704        ret = -EROFS;
3705        goto out;
3706    }
3707
3708    *p_blkno = le64_to_cpu(rec->e_blkno);
3709    *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
3710    if (e_cpos)
3711        *e_cpos = le32_to_cpu(rec->e_cpos);
3712out:
3713    brelse(eb_bh);
3714    return ret;
3715}
3716
3717typedef int (xattr_bucket_func)(struct inode *inode,
3718                struct ocfs2_xattr_bucket *bucket,
3719                void *para);
3720
3721static int ocfs2_find_xe_in_bucket(struct inode *inode,
3722                   struct ocfs2_xattr_bucket *bucket,
3723                   int name_index,
3724                   const char *name,
3725                   u32 name_hash,
3726                   u16 *xe_index,
3727                   int *found)
3728{
3729    int i, ret = 0, cmp = 1, block_off, new_offset;
3730    struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3731    size_t name_len = strlen(name);
3732    struct ocfs2_xattr_entry *xe = NULL;
3733    char *xe_name;
3734
3735    /*
3736     * We don't use binary search in the bucket because there
3737     * may be multiple entries with the same name hash.
3738     */
3739    for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3740        xe = &xh->xh_entries[i];
3741
3742        if (name_hash > le32_to_cpu(xe->xe_name_hash))
3743            continue;
3744        else if (name_hash < le32_to_cpu(xe->xe_name_hash))
3745            break;
3746
3747        cmp = name_index - ocfs2_xattr_get_type(xe);
3748        if (!cmp)
3749            cmp = name_len - xe->xe_name_len;
3750        if (cmp)
3751            continue;
3752
3753        ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3754                            xh,
3755                            i,
3756                            &block_off,
3757                            &new_offset);
3758        if (ret) {
3759            mlog_errno(ret);
3760            break;
3761        }
3762
3763
3764        xe_name = bucket_block(bucket, block_off) + new_offset;
3765        if (!memcmp(name, xe_name, name_len)) {
3766            *xe_index = i;
3767            *found = 1;
3768            ret = 0;
3769            break;
3770        }
3771    }
3772
3773    return ret;
3774}
3775
3776/*
3777 * Find the specified xattr entry in a series of buckets.
3778 * This series start from p_blkno and last for num_clusters.
3779 * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3780 * the num of the valid buckets.
3781 *
3782 * Return the buffer_head this xattr should reside in. And if the xattr's
3783 * hash is in the gap of 2 buckets, return the lower bucket.
3784 */
3785static int ocfs2_xattr_bucket_find(struct inode *inode,
3786                   int name_index,
3787                   const char *name,
3788                   u32 name_hash,
3789                   u64 p_blkno,
3790                   u32 first_hash,
3791                   u32 num_clusters,
3792                   struct ocfs2_xattr_search *xs)
3793{
3794    int ret, found = 0;
3795    struct ocfs2_xattr_header *xh = NULL;
3796    struct ocfs2_xattr_entry *xe = NULL;
3797    u16 index = 0;
3798    u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3799    int low_bucket = 0, bucket, high_bucket;
3800    struct ocfs2_xattr_bucket *search;
3801    u32 last_hash;
3802    u64 blkno, lower_blkno = 0;
3803
3804    search = ocfs2_xattr_bucket_new(inode);
3805    if (!search) {
3806        ret = -ENOMEM;
3807        mlog_errno(ret);
3808        goto out;
3809    }
3810
3811    ret = ocfs2_read_xattr_bucket(search, p_blkno);
3812    if (ret) {
3813        mlog_errno(ret);
3814        goto out;
3815    }
3816
3817    xh = bucket_xh(search);
3818    high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3819    while (low_bucket <= high_bucket) {
3820        ocfs2_xattr_bucket_relse(search);
3821
3822        bucket = (low_bucket + high_bucket) / 2;
3823        blkno = p_blkno + bucket * blk_per_bucket;
3824        ret = ocfs2_read_xattr_bucket(search, blkno);
3825        if (ret) {
3826            mlog_errno(ret);
3827            goto out;
3828        }
3829
3830        xh = bucket_xh(search);
3831        xe = &xh->xh_entries[0];
3832        if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3833            high_bucket = bucket - 1;
3834            continue;
3835        }
3836
3837        /*
3838         * Check whether the hash of the last entry in our
3839         * bucket is larger than the search one. for an empty
3840         * bucket, the last one is also the first one.
3841         */
3842        if (xh->xh_count)
3843            xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3844
3845        last_hash = le32_to_cpu(xe->xe_name_hash);
3846
3847        /* record lower_blkno which may be the insert place. */
3848        lower_blkno = blkno;
3849
3850        if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3851            low_bucket = bucket + 1;
3852            continue;
3853        }
3854
3855        /* the searched xattr should reside in this bucket if exists. */
3856        ret = ocfs2_find_xe_in_bucket(inode, search,
3857                          name_index, name, name_hash,
3858                          &index, &found);
3859        if (ret) {
3860            mlog_errno(ret);
3861            goto out;
3862        }
3863        break;
3864    }
3865
3866    /*
3867     * Record the bucket we have found.
3868     * When the xattr's hash value is in the gap of 2 buckets, we will
3869     * always set it to the previous bucket.
3870     */
3871    if (!lower_blkno)
3872        lower_blkno = p_blkno;
3873
3874    /* This should be in cache - we just read it during the search */
3875    ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3876    if (ret) {
3877        mlog_errno(ret);
3878        goto out;
3879    }
3880
3881    xs->header = bucket_xh(xs->bucket);
3882    xs->base = bucket_block(xs->bucket, 0);
3883    xs->end = xs->base + inode->i_sb->s_blocksize;
3884
3885    if (found) {
3886        xs->here = &xs->header->xh_entries[index];
3887        trace_ocfs2_xattr_bucket_find(OCFS2_I(inode)->ip_blkno,
3888            name, name_index, name_hash,
3889            (unsigned long long)bucket_blkno(xs->bucket),
3890            index);
3891    } else
3892        ret = -ENODATA;
3893
3894out:
3895    ocfs2_xattr_bucket_free(search);
3896    return ret;
3897}
3898
3899static int ocfs2_xattr_index_block_find(struct inode *inode,
3900                    struct buffer_head *root_bh,
3901                    int name_index,
3902                    const char *name,
3903                    struct ocfs2_xattr_search *xs)
3904{
3905    int ret;
3906    struct ocfs2_xattr_block *xb =
3907            (struct ocfs2_xattr_block *)root_bh->b_data;
3908    struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3909    struct ocfs2_extent_list *el = &xb_root->xt_list;
3910    u64 p_blkno = 0;
3911    u32 first_hash, num_clusters = 0;
3912    u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3913
3914    if (le16_to_cpu(el->l_next_free_rec) == 0)
3915        return -ENODATA;
3916
3917    trace_ocfs2_xattr_index_block_find(OCFS2_I(inode)->ip_blkno,
3918                    name, name_index, name_hash,
3919                    (unsigned long long)root_bh->b_blocknr,
3920                    -1);
3921
3922    ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3923                  &num_clusters, el);
3924    if (ret) {
3925        mlog_errno(ret);
3926        goto out;
3927    }
3928
3929    BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3930
3931    trace_ocfs2_xattr_index_block_find_rec(OCFS2_I(inode)->ip_blkno,
3932                    name, name_index, first_hash,
3933                    (unsigned long long)p_blkno,
3934                    num_clusters);
3935
3936    ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3937                      p_blkno, first_hash, num_clusters, xs);
3938
3939out:
3940    return ret;
3941}
3942
3943static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3944                       u64 blkno,
3945                       u32 clusters,
3946                       xattr_bucket_func *func,
3947                       void *para)
3948{
3949    int i, ret = 0;
3950    u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3951    u32 num_buckets = clusters * bpc;
3952    struct ocfs2_xattr_bucket *bucket;
3953
3954    bucket = ocfs2_xattr_bucket_new(inode);
3955    if (!bucket) {
3956        mlog_errno(-ENOMEM);
3957        return -ENOMEM;
3958    }
3959
3960    trace_ocfs2_iterate_xattr_buckets(
3961        (unsigned long long)OCFS2_I(inode)->ip_blkno,
3962        (unsigned long long)blkno, clusters);
3963
3964    for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3965        ret = ocfs2_read_xattr_bucket(bucket, blkno);
3966        if (ret) {
3967            mlog_errno(ret);
3968            break;
3969        }
3970
3971        /*
3972         * The real bucket num in this series of blocks is stored
3973         * in the 1st bucket.
3974         */
3975        if (i == 0)
3976            num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3977
3978        trace_ocfs2_iterate_xattr_bucket((unsigned long long)blkno,
3979             le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3980        if (func) {
3981            ret = func(inode, bucket, para);
3982            if (ret && ret != -ERANGE)
3983                mlog_errno(ret);
3984            /* Fall through to bucket_relse() */
3985        }
3986
3987        ocfs2_xattr_bucket_relse(bucket);
3988        if (ret)
3989            break;
3990    }
3991
3992    ocfs2_xattr_bucket_free(bucket);
3993    return ret;
3994}
3995
3996struct ocfs2_xattr_tree_list {
3997    char *buffer;
3998    size_t buffer_size;
3999    size_t result;
4000};
4001
4002static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
4003                         struct ocfs2_xattr_header *xh,
4004                         int index,
4005                         int *block_off,
4006                         int *new_offset)
4007{
4008    u16 name_offset;
4009
4010    if (index < 0 || index >= le16_to_cpu(xh->xh_count))
4011        return -EINVAL;
4012
4013    name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
4014
4015    *block_off = name_offset >> sb->s_blocksize_bits;
4016    *new_offset = name_offset % sb->s_blocksize;
4017
4018    return 0;
4019}
4020
4021static int ocfs2_list_xattr_bucket(struct inode *inode,
4022                   struct ocfs2_xattr_bucket *bucket,
4023                   void *para)
4024{
4025    int ret = 0, type;
4026    struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
4027    int i, block_off, new_offset;
4028    const char *prefix, *name;
4029
4030    for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
4031        struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
4032        type = ocfs2_xattr_get_type(entry);
4033        prefix = ocfs2_xattr_prefix(type);
4034
4035        if (prefix) {
4036            ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
4037                                bucket_xh(bucket),
4038                                i,
4039                                &block_off,
4040                                &new_offset);
4041            if (ret)
4042                break;
4043
4044            name = (const char *)bucket_block(bucket, block_off) +
4045                new_offset;
4046            ret = ocfs2_xattr_list_entry(xl->buffer,
4047                             xl->buffer_size,
4048                             &xl->result,
4049                             prefix, name,
4050                             entry->xe_name_len);
4051            if (ret)
4052                break;
4053        }
4054    }
4055
4056    return ret;
4057}
4058
4059static int ocfs2_iterate_xattr_index_block(struct inode *inode,
4060                       struct buffer_head *blk_bh,
4061                       xattr_tree_rec_func *rec_func,
4062                       void *para)
4063{
4064    struct ocfs2_xattr_block *xb =
4065            (struct ocfs2_xattr_block *)blk_bh->b_data;
4066    struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4067    int ret = 0;
4068    u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
4069    u64 p_blkno = 0;
4070
4071    if (!el->l_next_free_rec || !rec_func)
4072        return 0;
4073
4074    while (name_hash > 0) {
4075        ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4076                      &e_cpos, &num_clusters, el);
4077        if (ret) {
4078            mlog_errno(ret);
4079            break;
4080        }
4081
4082        ret = rec_func(inode, blk_bh, p_blkno, e_cpos,
4083                   num_clusters, para);
4084        if (ret) {
4085            if (ret != -ERANGE)
4086                mlog_errno(ret);
4087            break;
4088        }
4089
4090        if (e_cpos == 0)
4091            break;
4092
4093        name_hash = e_cpos - 1;
4094    }
4095
4096    return ret;
4097
4098}
4099
4100static int ocfs2_list_xattr_tree_rec(struct inode *inode,
4101                     struct buffer_head *root_bh,
4102                     u64 blkno, u32 cpos, u32 len, void *para)
4103{
4104    return ocfs2_iterate_xattr_buckets(inode, blkno, len,
4105                       ocfs2_list_xattr_bucket, para);
4106}
4107
4108static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
4109                         struct buffer_head *blk_bh,
4110                         char *buffer,
4111                         size_t buffer_size)
4112{
4113    int ret;
4114    struct ocfs2_xattr_tree_list xl = {
4115        .buffer = buffer,
4116        .buffer_size = buffer_size,
4117        .result = 0,
4118    };
4119
4120    ret = ocfs2_iterate_xattr_index_block(inode, blk_bh,
4121                          ocfs2_list_xattr_tree_rec, &xl);
4122    if (ret) {
4123        mlog_errno(ret);
4124        goto out;
4125    }
4126
4127    ret = xl.result;
4128out:
4129    return ret;
4130}
4131
4132static int cmp_xe(const void *a, const void *b)
4133{
4134    const struct ocfs2_xattr_entry *l = a, *r = b;
4135    u32 l_hash = le32_to_cpu(l->xe_name_hash);
4136    u32 r_hash = le32_to_cpu(r->xe_name_hash);
4137
4138    if (l_hash > r_hash)
4139        return 1;
4140    if (l_hash < r_hash)
4141        return -1;
4142    return 0;
4143}
4144
4145static void swap_xe(void *a, void *b, int size)
4146{
4147    struct ocfs2_xattr_entry *l = a, *r = b, tmp;
4148
4149    tmp = *l;
4150    memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
4151    memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
4152}
4153
4154/*
4155 * When the ocfs2_xattr_block is filled up, new bucket will be created
4156 * and all the xattr entries will be moved to the new bucket.
4157 * The header goes at the start of the bucket, and the names+values are
4158 * filled from the end. This is why *target starts as the last buffer.
4159 * Note: we need to sort the entries since they are not saved in order
4160 * in the ocfs2_xattr_block.
4161 */
4162static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
4163                       struct buffer_head *xb_bh,
4164                       struct ocfs2_xattr_bucket *bucket)
4165{
4166    int i, blocksize = inode->i_sb->s_blocksize;
4167    int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4168    u16 offset, size, off_change;
4169    struct ocfs2_xattr_entry *xe;
4170    struct ocfs2_xattr_block *xb =
4171                (struct ocfs2_xattr_block *)xb_bh->b_data;
4172    struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
4173    struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4174    u16 count = le16_to_cpu(xb_xh->xh_count);
4175    char *src = xb_bh->b_data;
4176    char *target = bucket_block(bucket, blks - 1);
4177
4178    trace_ocfs2_cp_xattr_block_to_bucket_begin(
4179                (unsigned long long)xb_bh->b_blocknr,
4180                (unsigned long long)bucket_blkno(bucket));
4181
4182    for (i = 0; i < blks; i++)
4183        memset(bucket_block(bucket, i), 0, blocksize);
4184
4185    /*
4186     * Since the xe_name_offset is based on ocfs2_xattr_header,
4187     * there is a offset change corresponding to the change of
4188     * ocfs2_xattr_header's position.
4189     */
4190    off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4191    xe = &xb_xh->xh_entries[count - 1];
4192    offset = le16_to_cpu(xe->xe_name_offset) + off_change;
4193    size = blocksize - offset;
4194
4195    /* copy all the names and values. */
4196    memcpy(target + offset, src + offset, size);
4197
4198    /* Init new header now. */
4199    xh->xh_count = xb_xh->xh_count;
4200    xh->xh_num_buckets = cpu_to_le16(1);
4201    xh->xh_name_value_len = cpu_to_le16(size);
4202    xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
4203
4204    /* copy all the entries. */
4205    target = bucket_block(bucket, 0);
4206    offset = offsetof(struct ocfs2_xattr_header, xh_entries);
4207    size = count * sizeof(struct ocfs2_xattr_entry);
4208    memcpy(target + offset, (char *)xb_xh + offset, size);
4209
4210    /* Change the xe offset for all the xe because of the move. */
4211    off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
4212         offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
4213    for (i = 0; i < count; i++)
4214        le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
4215
4216    trace_ocfs2_cp_xattr_block_to_bucket_end(offset, size, off_change);
4217
4218    sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
4219         cmp_xe, swap_xe);
4220}
4221
4222/*
4223 * After we move xattr from block to index btree, we have to
4224 * update ocfs2_xattr_search to the new xe and base.
4225 *
4226 * When the entry is in xattr block, xattr_bh indicates the storage place.
4227 * While if the entry is in index b-tree, "bucket" indicates the
4228 * real place of the xattr.
4229 */
4230static void ocfs2_xattr_update_xattr_search(struct inode *inode,
4231                        struct ocfs2_xattr_search *xs,
4232                        struct buffer_head *old_bh)
4233{
4234    char *buf = old_bh->b_data;
4235    struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
4236    struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
4237    int i;
4238
4239    xs->header = bucket_xh(xs->bucket);
4240    xs->base = bucket_block(xs->bucket, 0);
4241    xs->end = xs->base + inode->i_sb->s_blocksize;
4242
4243    if (xs->not_found)
4244        return;
4245
4246    i = xs->here - old_xh->xh_entries;
4247    xs->here = &xs->header->xh_entries[i];
4248}
4249
4250static int ocfs2_xattr_create_index_block(struct inode *inode,
4251                      struct ocfs2_xattr_search *xs,
4252                      struct ocfs2_xattr_set_ctxt *ctxt)
4253{
4254    int ret;
4255    u32 bit_off, len;
4256    u64 blkno;
4257    handle_t *handle = ctxt->handle;
4258    struct ocfs2_inode_info *oi = OCFS2_I(inode);
4259    struct buffer_head *xb_bh = xs->xattr_bh;
4260    struct ocfs2_xattr_block *xb =
4261            (struct ocfs2_xattr_block *)xb_bh->b_data;
4262    struct ocfs2_xattr_tree_root *xr;
4263    u16 xb_flags = le16_to_cpu(xb->xb_flags);
4264
4265    trace_ocfs2_xattr_create_index_block_begin(
4266                (unsigned long long)xb_bh->b_blocknr);
4267
4268    BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
4269    BUG_ON(!xs->bucket);
4270
4271    /*
4272     * XXX:
4273     * We can use this lock for now, and maybe move to a dedicated mutex
4274     * if performance becomes a problem later.
4275     */
4276    down_write(&oi->ip_alloc_sem);
4277
4278    ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
4279                      OCFS2_JOURNAL_ACCESS_WRITE);
4280    if (ret) {
4281        mlog_errno(ret);
4282        goto out;
4283    }
4284
4285    ret = __ocfs2_claim_clusters(handle, ctxt->data_ac,
4286                     1, 1, &bit_off, &len);
4287    if (ret) {
4288        mlog_errno(ret);
4289        goto out;
4290    }
4291
4292    /*
4293     * The bucket may spread in many blocks, and
4294     * we will only touch the 1st block and the last block
4295     * in the whole bucket(one for entry and one for data).
4296     */
4297    blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
4298
4299    trace_ocfs2_xattr_create_index_block((unsigned long long)blkno);
4300
4301    ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
4302    if (ret) {
4303        mlog_errno(ret);
4304        goto out;
4305    }
4306
4307    ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4308                        OCFS2_JOURNAL_ACCESS_CREATE);
4309    if (ret) {
4310        mlog_errno(ret);
4311        goto out;
4312    }
4313
4314    ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
4315    ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4316
4317    ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
4318
4319    /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
4320    memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
4321           offsetof(struct ocfs2_xattr_block, xb_attrs));
4322
4323    xr = &xb->xb_attrs.xb_root;
4324    xr->xt_clusters = cpu_to_le32(1);
4325    xr->xt_last_eb_blk = 0;
4326    xr->xt_list.l_tree_depth = 0;
4327    xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
4328    xr->xt_list.l_next_free_rec = cpu_to_le16(1);
4329
4330    xr->xt_list.l_recs[0].e_cpos = 0;
4331    xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
4332    xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
4333
4334    xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
4335
4336    ocfs2_journal_dirty(handle, xb_bh);
4337
4338out:
4339    up_write(&oi->ip_alloc_sem);
4340
4341    return ret;
4342}
4343
4344static int cmp_xe_offset(const void *a, const void *b)
4345{
4346    const struct ocfs2_xattr_entry *l = a, *r = b;
4347    u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
4348    u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
4349
4350    if (l_name_offset < r_name_offset)
4351        return 1;
4352    if (l_name_offset > r_name_offset)
4353        return -1;
4354    return 0;
4355}
4356
4357/*
4358 * defrag a xattr bucket if we find that the bucket has some
4359 * holes beteen name/value pairs.
4360 * We will move all the name/value pairs to the end of the bucket
4361 * so that we can spare some space for insertion.
4362 */
4363static int ocfs2_defrag_xattr_bucket(struct inode *inode,
4364                     handle_t *handle,
4365                     struct ocfs2_xattr_bucket *bucket)
4366{
4367    int ret, i;
4368    size_t end, offset, len;
4369    struct ocfs2_xattr_header *xh;
4370    char *entries, *buf, *bucket_buf = NULL;
4371    u64 blkno = bucket_blkno(bucket);
4372    u16 xh_free_start;
4373    size_t blocksize = inode->i_sb->s_blocksize;
4374    struct ocfs2_xattr_entry *xe;
4375
4376    /*
4377     * In order to make the operation more efficient and generic,
4378     * we copy all the blocks into a contiguous memory and do the
4379     * defragment there, so if anything is error, we will not touch
4380     * the real block.
4381     */
4382    bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
4383    if (!bucket_buf) {
4384        ret = -EIO;
4385        goto out;
4386    }
4387
4388    buf = bucket_buf;
4389    for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4390        memcpy(buf, bucket_block(bucket, i), blocksize);
4391
4392    ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
4393                        OCFS2_JOURNAL_ACCESS_WRITE);
4394    if (ret < 0) {
4395        mlog_errno(ret);
4396        goto out;
4397    }
4398
4399    xh = (struct ocfs2_xattr_header *)bucket_buf;
4400    entries = (char *)xh->xh_entries;
4401    xh_free_start = le16_to_cpu(xh->xh_free_start);
4402
4403    trace_ocfs2_defrag_xattr_bucket(
4404         (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
4405         xh_free_start, le16_to_cpu(xh->xh_name_value_len));
4406
4407    /*
4408     * sort all the entries by their offset.
4409     * the largest will be the first, so that we can
4410     * move them to the end one by one.
4411     */
4412    sort(entries, le16_to_cpu(xh->xh_count),
4413         sizeof(struct ocfs2_xattr_entry),
4414         cmp_xe_offset, swap_xe);
4415
4416    /* Move all name/values to the end of the bucket. */
4417    xe = xh->xh_entries;
4418    end = OCFS2_XATTR_BUCKET_SIZE;
4419    for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
4420        offset = le16_to_cpu(xe->xe_name_offset);
4421        len = namevalue_size_xe(xe);
4422
4423        /*
4424         * We must make sure that the name/value pair
4425         * exist in the same block. So adjust end to
4426         * the previous block end if needed.
4427         */
4428        if (((end - len) / blocksize !=
4429            (end - 1) / blocksize))
4430            end = end - end % blocksize;
4431
4432        if (end > offset + len) {
4433            memmove(bucket_buf + end - len,
4434                bucket_buf + offset, len);
4435            xe->xe_name_offset = cpu_to_le16(end - len);
4436        }
4437
4438        mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
4439                "bucket %llu\n", (unsigned long long)blkno);
4440
4441        end -= len;
4442    }
4443
4444    mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
4445            "bucket %llu\n", (unsigned long long)blkno);
4446
4447    if (xh_free_start == end)
4448        goto out;
4449
4450    memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
4451    xh->xh_free_start = cpu_to_le16(end);
4452
4453    /* sort the entries by their name_hash. */
4454    sort(entries, le16_to_cpu(xh->xh_count),
4455         sizeof(struct ocfs2_xattr_entry),
4456         cmp_xe, swap_xe);
4457
4458    buf = bucket_buf;
4459    for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
4460        memcpy(bucket_block(bucket, i), buf, blocksize);
4461    ocfs2_xattr_bucket_journal_dirty(handle, bucket);
4462
4463out:
4464    kfree(bucket_buf);
4465    return ret;
4466}
4467
4468/*
4469 * prev_blkno points to the start of an existing extent. new_blkno
4470 * points to a newly allocated extent. Because we know each of our
4471 * clusters contains more than bucket, we can easily split one cluster
4472 * at a bucket boundary. So we take the last cluster of the existing
4473 * extent and split it down the middle. We move the last half of the
4474 * buckets in the last cluster of the existing extent over to the new
4475 * extent.
4476 *
4477 * first_bh is the buffer at prev_blkno so we can update the existing
4478 * extent's bucket count. header_bh is the bucket were we were hoping
4479 * to insert our xattr. If the bucket move places the target in the new
4480 * extent, we'll update first_bh and header_bh after modifying the old
4481 * extent.
4482 *
4483 * first_hash will be set as the 1st xe's name_hash in the new extent.
4484 */
4485static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
4486                           handle_t *handle,
4487                           struct ocfs2_xattr_bucket *first,
4488                           struct ocfs2_xattr_bucket *target,
4489                           u64 new_blkno,
4490                           u32 num_clusters,
4491                           u32 *first_hash)
4492{
4493    int ret;
4494    struct super_block *sb = inode->i_sb;
4495    int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
4496    int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
4497    int to_move = num_buckets / 2;
4498    u64 src_blkno;
4499    u64 last_cluster_blkno = bucket_blkno(first) +
4500        ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
4501
4502    BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
4503    BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
4504
4505    trace_ocfs2_mv_xattr_bucket_cross_cluster(
4506                (unsigned long long)last_cluster_blkno,
4507                (unsigned long long)new_blkno);
4508
4509    ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
4510                     last_cluster_blkno, new_blkno,
4511                     to_move, first_hash);
4512    if (ret) {
4513        mlog_errno(ret);
4514        goto out;
4515    }
4516
4517    /* This is the first bucket that got moved */
4518    src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
4519
4520    /*
4521     * If the target bucket was part of the moved buckets, we need to
4522     * update first and target.
4523     */
4524    if (bucket_blkno(target) >= src_blkno) {
4525        /* Find the block for the new target bucket */
4526        src_blkno = new_blkno +
4527            (bucket_blkno(target) - src_blkno);
4528
4529        ocfs2_xattr_bucket_relse(first);
4530        ocfs2_xattr_bucket_relse(target);
4531
4532        /*
4533         * These shouldn't fail - the buffers are in the
4534         * journal from ocfs2_cp_xattr_bucket().
4535         */
4536        ret = ocfs2_read_xattr_bucket(first, new_blkno);
4537        if (ret) {
4538            mlog_errno(ret);
4539            goto out;
4540        }
4541        ret = ocfs2_read_xattr_bucket(target, src_blkno);
4542        if (ret)
4543            mlog_errno(ret);
4544
4545    }
4546
4547out:
4548    return ret;
4549}
4550
4551/*
4552 * Find the suitable pos when we divide a bucket into 2.
4553 * We have to make sure the xattrs with the same hash value exist
4554 * in the same bucket.
4555 *
4556 * If this ocfs2_xattr_header covers more than one hash value, find a
4557 * place where the hash value changes. Try to find the most even split.
4558 * The most common case is that all entries have different hash values,
4559 * and the first check we make will find a place to split.
4560 */
4561static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
4562{
4563    struct ocfs2_xattr_entry *entries = xh->xh_entries;
4564    int count = le16_to_cpu(xh->xh_count);
4565    int delta, middle = count / 2;
4566
4567    /*
4568     * We start at the middle. Each step gets farther away in both
4569     * directions. We therefore hit the change in hash value
4570     * nearest to the middle. Note that this loop does not execute for
4571     * count < 2.
4572     */
4573    for (delta = 0; delta < middle; delta++) {
4574        /* Let's check delta earlier than middle */
4575        if (cmp_xe(&entries[middle - delta - 1],
4576               &entries[middle - delta]))
4577            return middle - delta;
4578
4579        /* For even counts, don't walk off the end */
4580        if ((middle + delta + 1) == count)
4581            continue;
4582
4583        /* Now try delta past middle */
4584        if (cmp_xe(&entries[middle + delta],
4585               &entries[middle + delta + 1]))
4586            return middle + delta + 1;
4587    }
4588
4589    /* Every entry had the same hash */
4590    return count;
4591}
4592
4593/*
4594 * Move some xattrs in old bucket(blk) to new bucket(new_blk).
4595 * first_hash will record the 1st hash of the new bucket.
4596 *
4597 * Normally half of the xattrs will be moved. But we have to make
4598 * sure that the xattrs with the same hash value are stored in the
4599 * same bucket. If all the xattrs in this bucket have the same hash
4600 * value, the new bucket will be initialized as an empty one and the
4601 * first_hash will be initialized as (hash_value+1).
4602 */
4603static int ocfs2_divide_xattr_bucket(struct inode *inode,
4604                    handle_t *handle,
4605                    u64 blk,
4606                    u64 new_blk,
4607                    u32 *first_hash,
4608                    int new_bucket_head)
4609{
4610    int ret, i;
4611    int count, start, len, name_value_len = 0, name_offset = 0;
4612    struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4613    struct ocfs2_xattr_header *xh;
4614    struct ocfs2_xattr_entry *xe;
4615    int blocksize = inode->i_sb->s_blocksize;
4616
4617    trace_ocfs2_divide_xattr_bucket_begin((unsigned long long)blk,
4618                          (unsigned long long)new_blk);
4619
4620    s_bucket = ocfs2_xattr_bucket_new(inode);
4621    t_bucket = ocfs2_xattr_bucket_new(inode);
4622    if (!s_bucket || !t_bucket) {
4623        ret = -ENOMEM;
4624        mlog_errno(ret);
4625        goto out;
4626    }
4627
4628    ret = ocfs2_read_xattr_bucket(s_bucket, blk);
4629    if (ret) {
4630        mlog_errno(ret);
4631        goto out;
4632    }
4633
4634    ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
4635                        OCFS2_JOURNAL_ACCESS_WRITE);
4636    if (ret) {
4637        mlog_errno(ret);
4638        goto out;
4639    }
4640
4641    /*
4642     * Even if !new_bucket_head, we're overwriting t_bucket. Thus,
4643     * there's no need to read it.
4644     */
4645    ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
4646    if (ret) {
4647        mlog_errno(ret);
4648        goto out;
4649    }
4650
4651    /*
4652     * Hey, if we're overwriting t_bucket, what difference does
4653     * ACCESS_CREATE vs ACCESS_WRITE make? See the comment in the
4654     * same part of ocfs2_cp_xattr_bucket().
4655     */
4656    ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4657                        new_bucket_head ?
4658                        OCFS2_JOURNAL_ACCESS_CREATE :
4659                        OCFS2_JOURNAL_ACCESS_WRITE);
4660    if (ret) {
4661        mlog_errno(ret);
4662        goto out;
4663    }
4664
4665    xh = bucket_xh(s_bucket);
4666    count = le16_to_cpu(xh->xh_count);
4667    start = ocfs2_xattr_find_divide_pos(xh);
4668
4669    if (start == count) {
4670        xe = &xh->xh_entries[start-1];
4671
4672        /*
4673         * initialized a new empty bucket here.
4674         * The hash value is set as one larger than
4675         * that of the last entry in the previous bucket.
4676         */
4677        for (i = 0; i < t_bucket->bu_blocks; i++)
4678            memset(bucket_block(t_bucket, i), 0, blocksize);
4679
4680        xh = bucket_xh(t_bucket);
4681        xh->xh_free_start = cpu_to_le16(blocksize);
4682        xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
4683        le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
4684
4685        goto set_num_buckets;
4686    }
4687
4688    /* copy the whole bucket to the new first. */
4689    ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4690
4691    /* update the new bucket. */
4692    xh = bucket_xh(t_bucket);
4693
4694    /*
4695     * Calculate the total name/value len and xh_free_start for
4696     * the old bucket first.
4697     */
4698    name_offset = OCFS2_XATTR_BUCKET_SIZE;
4699    name_value_len = 0;
4700    for (i = 0; i < start; i++) {
4701        xe = &xh->xh_entries[i];
4702        name_value_len += namevalue_size_xe(xe);
4703        if (le16_to_cpu(xe->xe_name_offset) < name_offset)
4704            name_offset = le16_to_cpu(xe->xe_name_offset);
4705    }
4706
4707    /*
4708     * Now begin the modification to the new bucket.
4709     *
4710     * In the new bucket, We just move the xattr entry to the beginning
4711     * and don't touch the name/value. So there will be some holes in the
4712     * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
4713     * called.
4714     */
4715    xe = &xh->xh_entries[start];
4716    len = sizeof(struct ocfs2_xattr_entry) * (count - start);
4717    trace_ocfs2_divide_xattr_bucket_move(len,
4718            (int)((char *)xe - (char *)xh),
4719            (int)((char *)xh->xh_entries - (char *)xh));
4720    memmove((char *)xh->xh_entries, (char *)xe, len);
4721    xe = &xh->xh_entries[count - start];
4722    len = sizeof(struct ocfs2_xattr_entry) * start;
4723    memset((char *)xe, 0, len);
4724
4725    le16_add_cpu(&xh->xh_count, -start);
4726    le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
4727
4728    /* Calculate xh_free_start for the new bucket. */
4729    xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4730    for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4731        xe = &xh->xh_entries[i];
4732        if (le16_to_cpu(xe->xe_name_offset) <
4733            le16_to_cpu(xh->xh_free_start))
4734            xh->xh_free_start = xe->xe_name_offset;
4735    }
4736
4737set_num_buckets:
4738    /* set xh->xh_num_buckets for the new xh. */
4739    if (new_bucket_head)
4740        xh->xh_num_buckets = cpu_to_le16(1);
4741    else
4742        xh->xh_num_buckets = 0;
4743
4744    ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4745
4746    /* store the first_hash of the new bucket. */
4747    if (first_hash)
4748        *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
4749
4750    /*
4751     * Now only update the 1st block of the old bucket. If we
4752     * just added a new empty bucket, there is no need to modify
4753     * it.
4754     */
4755    if (start == count)
4756        goto out;
4757
4758    xh = bucket_xh(s_bucket);
4759    memset(&xh->xh_entries[start], 0,
4760           sizeof(struct ocfs2_xattr_entry) * (count - start));
4761    xh->xh_count = cpu_to_le16(start);
4762    xh->xh_free_start = cpu_to_le16(name_offset);
4763    xh->xh_name_value_len = cpu_to_le16(name_value_len);
4764
4765    ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4766
4767out:
4768    ocfs2_xattr_bucket_free(s_bucket);
4769    ocfs2_xattr_bucket_free(t_bucket);
4770
4771    return ret;
4772}
4773
4774/*
4775 * Copy xattr from one bucket to another bucket.
4776 *
4777 * The caller must make sure that the journal transaction
4778 * has enough space for journaling.
4779 */
4780static int ocfs2_cp_xattr_bucket(struct inode *inode,
4781                 handle_t *handle,
4782                 u64 s_blkno,
4783                 u64 t_blkno,
4784                 int t_is_new)
4785{
4786    int ret;
4787    struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4788
4789    BUG_ON(s_blkno == t_blkno);
4790
4791    trace_ocfs2_cp_xattr_bucket((unsigned long long)s_blkno,
4792                    (unsigned long long)t_blkno,
4793                    t_is_new);
4794
4795    s_bucket = ocfs2_xattr_bucket_new(inode);
4796    t_bucket = ocfs2_xattr_bucket_new(inode);
4797    if (!s_bucket || !t_bucket) {
4798        ret = -ENOMEM;
4799        mlog_errno(ret);
4800        goto out;
4801    }
4802
4803    ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4804    if (ret)
4805        goto out;
4806
4807    /*
4808     * Even if !t_is_new, we're overwriting t_bucket. Thus,
4809     * there's no need to read it.
4810     */
4811    ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4812    if (ret)
4813        goto out;
4814
4815    /*
4816     * Hey, if we're overwriting t_bucket, what difference does
4817     * ACCESS_CREATE vs ACCESS_WRITE make? Well, if we allocated a new
4818     * cluster to fill, we came here from
4819     * ocfs2_mv_xattr_buckets(), and it is really new -
4820     * ACCESS_CREATE is required. But we also might have moved data
4821     * out of t_bucket before extending back into it.
4822     * ocfs2_add_new_xattr_bucket() can do this - its call to
4823     * ocfs2_add_new_xattr_cluster() may have created a new extent
4824     * and copied out the end of the old extent. Then it re-extends
4825     * the old extent back to create space for new xattrs. That's
4826     * how we get here, and the bucket isn't really new.
4827     */
4828    ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4829                        t_is_new ?
4830                        OCFS2_JOURNAL_ACCESS_CREATE :
4831                        OCFS2_JOURNAL_ACCESS_WRITE);
4832    if (ret)
4833        goto out;
4834
4835    ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4836    ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4837
4838out:
4839    ocfs2_xattr_bucket_free(t_bucket);
4840    ocfs2_xattr_bucket_free(s_bucket);
4841
4842    return ret;
4843}
4844
4845/*
4846 * src_blk points to the start of an existing extent. last_blk points to
4847 * last cluster in that extent. to_blk points to a newly allocated
4848 * extent. We copy the buckets from the cluster at last_blk to the new
4849 * extent. If start_bucket is non-zero, we skip that many buckets before
4850 * we start copying. The new extent's xh_num_buckets gets set to the
4851 * number of buckets we copied. The old extent's xh_num_buckets shrinks
4852 * by the same amount.
4853 */
4854static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4855                  u64 src_blk, u64 last_blk, u64 to_blk,
4856                  unsigned int start_bucket,
4857                  u32 *first_hash)
4858{
4859    int i, ret, credits;
4860    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4861    int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4862    int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4863    struct ocfs2_xattr_bucket *old_first, *new_first;
4864
4865    trace_ocfs2_mv_xattr_buckets((unsigned long long)last_blk,
4866                     (unsigned long long)to_blk);
4867
4868    BUG_ON(start_bucket >= num_buckets);
4869    if (start_bucket) {
4870        num_buckets -= start_bucket;
4871        last_blk += (start_bucket * blks_per_bucket);
4872    }
4873
4874    /* The first bucket of the original extent */
4875    old_first = ocfs2_xattr_bucket_new(inode);
4876    /* The first bucket of the new extent */
4877    new_first = ocfs2_xattr_bucket_new(inode);
4878    if (!old_first || !new_first) {
4879        ret = -ENOMEM;
4880        mlog_errno(ret);
4881        goto out;
4882    }
4883
4884    ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4885    if (ret) {
4886        mlog_errno(ret);
4887        goto out;
4888    }
4889
4890    /*
4891     * We need to update the first bucket of the old extent and all
4892     * the buckets going to the new extent.
4893     */
4894    credits = ((num_buckets + 1) * blks_per_bucket);
4895    ret = ocfs2_extend_trans(handle, credits);
4896    if (ret) {
4897        mlog_errno(ret);
4898        goto out;
4899    }
4900
4901    ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4902                        OCFS2_JOURNAL_ACCESS_WRITE);
4903    if (ret) {
4904        mlog_errno(ret);
4905        goto out;
4906    }
4907
4908    for (i = 0; i < num_buckets; i++) {
4909        ret = ocfs2_cp_xattr_bucket(inode, handle,
4910                        last_blk + (i * blks_per_bucket),
4911                        to_blk + (i * blks_per_bucket),
4912                        1);
4913        if (ret) {
4914            mlog_errno(ret);
4915            goto out;
4916        }
4917    }
4918
4919    /*
4920     * Get the new bucket ready before we dirty anything
4921     * (This actually shouldn't fail, because we already dirtied
4922     * it once in ocfs2_cp_xattr_bucket()).
4923     */
4924    ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4925    if (ret) {
4926        mlog_errno(ret);
4927        goto out;
4928    }
4929    ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4930                        OCFS2_JOURNAL_ACCESS_WRITE);
4931    if (ret) {
4932        mlog_errno(ret);
4933        goto out;
4934    }
4935
4936    /* Now update the headers */
4937    le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4938    ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4939
4940    bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4941    ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4942
4943    if (first_hash)
4944        *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4945
4946out:
4947    ocfs2_xattr_bucket_free(new_first);
4948    ocfs2_xattr_bucket_free(old_first);
4949    return ret;
4950}
4951
4952/*
4953 * Move some xattrs in this cluster to the new cluster.
4954 * This function should only be called when bucket size == cluster size.
4955 * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4956 */
4957static int ocfs2_divide_xattr_cluster(struct inode *inode,
4958                      handle_t *handle,
4959                      u64 prev_blk,
4960                      u64 new_blk,
4961                      u32 *first_hash)
4962{
4963    u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4964    int ret, credits = 2 * blk_per_bucket;
4965
4966    BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4967
4968    ret = ocfs2_extend_trans(handle, credits);
4969    if (ret) {
4970        mlog_errno(ret);
4971        return ret;
4972    }
4973
4974    /* Move half of the xattr in start_blk to the next bucket. */
4975    return ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4976                      new_blk, first_hash, 1);
4977}
4978
4979/*
4980 * Move some xattrs from the old cluster to the new one since they are not
4981 * contiguous in ocfs2 xattr tree.
4982 *
4983 * new_blk starts a new separate cluster, and we will move some xattrs from
4984 * prev_blk to it. v_start will be set as the first name hash value in this
4985 * new cluster so that it can be used as e_cpos during tree insertion and
4986 * don't collide with our original b-tree operations. first_bh and header_bh
4987 * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4988 * to extend the insert bucket.
4989 *
4990 * The problem is how much xattr should we move to the new one and when should
4991 * we update first_bh and header_bh?
4992 * 1. If cluster size > bucket size, that means the previous cluster has more
4993 * than 1 bucket, so just move half nums of bucket into the new cluster and
4994 * update the first_bh and header_bh if the insert bucket has been moved
4995 * to the new cluster.
4996 * 2. If cluster_size == bucket_size:
4997 * a) If the previous extent rec has more than one cluster and the insert
4998 * place isn't in the last cluster, copy the entire last cluster to the
4999 * new one. This time, we don't need to upate the first_bh and header_bh
5000 * since they will not be moved into the new cluster.
5001 * b) Otherwise, move the bottom half of the xattrs in the last cluster into
5002 * the new one. And we set the extend flag to zero if the insert place is
5003 * moved into the new allocated cluster since no extend is needed.
5004 */
5005static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
5006                        handle_t *handle,
5007                        struct ocfs2_xattr_bucket *first,
5008                        struct ocfs2_xattr_bucket *target,
5009                        u64 new_blk,
5010                        u32 prev_clusters,
5011                        u32 *v_start,
5012                        int *extend)
5013{
5014    int ret;
5015
5016    trace_ocfs2_adjust_xattr_cross_cluster(
5017            (unsigned long long)bucket_blkno(first),
5018            (unsigned long long)new_blk, prev_clusters);
5019
5020    if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
5021        ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
5022                              handle,
5023                              first, target,
5024                              new_blk,
5025                              prev_clusters,
5026                              v_start);
5027        if (ret)
5028            mlog_errno(ret);
5029    } else {
5030        /* The start of the last cluster in the first extent */
5031        u64 last_blk = bucket_blkno(first) +
5032            ((prev_clusters - 1) *
5033             ocfs2_clusters_to_blocks(inode->i_sb, 1));
5034
5035        if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
5036            ret = ocfs2_mv_xattr_buckets(inode, handle,
5037                             bucket_blkno(first),
5038                             last_blk, new_blk, 0,
5039                             v_start);
5040            if (ret)
5041                mlog_errno(ret);
5042        } else {
5043            ret = ocfs2_divide_xattr_cluster(inode, handle,
5044                             last_blk, new_blk,
5045                             v_start);
5046            if (ret)
5047                mlog_errno(ret);
5048
5049            if ((bucket_blkno(target) == last_blk) && extend)
5050                *extend = 0;
5051        }
5052    }
5053
5054    return ret;
5055}
5056
5057/*
5058 * Add a new cluster for xattr storage.
5059 *
5060 * If the new cluster is contiguous with the previous one, it will be
5061 * appended to the same extent record, and num_clusters will be updated.
5062 * If not, we will insert a new extent for it and move some xattrs in
5063 * the last cluster into the new allocated one.
5064 * We also need to limit the maximum size of a btree leaf, otherwise we'll
5065 * lose the benefits of hashing because we'll have to search large leaves.
5066 * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
5067 * if it's bigger).
5068 *
5069 * first_bh is the first block of the previous extent rec and header_bh
5070 * indicates the bucket we will insert the new xattrs. They will be updated
5071 * when the header_bh is moved into the new cluster.
5072 */
5073static int ocfs2_add_new_xattr_cluster(struct inode *inode,
5074                       struct buffer_head *root_bh,
5075                       struct ocfs2_xattr_bucket *first,
5076                       struct ocfs2_xattr_bucket *target,
5077                       u32 *num_clusters,
5078                       u32 prev_cpos,
5079                       int *extend,
5080                       struct ocfs2_xattr_set_ctxt *ctxt)
5081{
5082    int ret;
5083    u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
5084    u32 prev_clusters = *num_clusters;
5085    u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
5086    u64 block;
5087    handle_t *handle = ctxt->handle;
5088    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5089    struct ocfs2_extent_tree et;
5090
5091    trace_ocfs2_add_new_xattr_cluster_begin(
5092        (unsigned long long)OCFS2_I(inode)->ip_blkno,
5093        (unsigned long long)bucket_blkno(first),
5094        prev_cpos, prev_clusters);
5095
5096    ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5097
5098    ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5099                      OCFS2_JOURNAL_ACCESS_WRITE);
5100    if (ret < 0) {
5101        mlog_errno(ret);
5102        goto leave;
5103    }
5104
5105    ret = __ocfs2_claim_clusters(handle, ctxt->data_ac, 1,
5106                     clusters_to_add, &bit_off, &num_bits);
5107    if (ret < 0) {
5108        if (ret != -ENOSPC)
5109            mlog_errno(ret);
5110        goto leave;
5111    }
5112
5113    BUG_ON(num_bits > clusters_to_add);
5114
5115    block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
5116    trace_ocfs2_add_new_xattr_cluster((unsigned long long)block, num_bits);
5117
5118    if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
5119        (prev_clusters + num_bits) << osb->s_clustersize_bits <=
5120         OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
5121        /*
5122         * If this cluster is contiguous with the old one and
5123         * adding this new cluster, we don't surpass the limit of
5124         * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
5125         * initialized and used like other buckets in the previous
5126         * cluster.
5127         * So add it as a contiguous one. The caller will handle
5128         * its init process.
5129         */
5130        v_start = prev_cpos + prev_clusters;
5131        *num_clusters = prev_clusters + num_bits;
5132    } else {
5133        ret = ocfs2_adjust_xattr_cross_cluster(inode,
5134                               handle,
5135                               first,
5136                               target,
5137                               block,
5138                               prev_clusters,
5139                               &v_start,
5140                               extend);
5141        if (ret) {
5142            mlog_errno(ret);
5143            goto leave;
5144        }
5145    }
5146
5147    trace_ocfs2_add_new_xattr_cluster_insert((unsigned long long)block,
5148                         v_start, num_bits);
5149    ret = ocfs2_insert_extent(handle, &et, v_start, block,
5150                  num_bits, 0, ctxt->meta_ac);
5151    if (ret < 0) {
5152        mlog_errno(ret);
5153        goto leave;
5154    }
5155
5156    ocfs2_journal_dirty(handle, root_bh);
5157
5158leave:
5159    return ret;
5160}
5161
5162/*
5163 * We are given an extent. 'first' is the bucket at the very front of
5164 * the extent. The extent has space for an additional bucket past
5165 * bucket_xh(first)->xh_num_buckets. 'target_blkno' is the block number
5166 * of the target bucket. We wish to shift every bucket past the target
5167 * down one, filling in that additional space. When we get back to the
5168 * target, we split the target between itself and the now-empty bucket
5169 * at target+1 (aka, target_blkno + blks_per_bucket).
5170 */
5171static int ocfs2_extend_xattr_bucket(struct inode *inode,
5172                     handle_t *handle,
5173                     struct ocfs2_xattr_bucket *first,
5174                     u64 target_blk,
5175                     u32 num_clusters)
5176{
5177    int ret, credits;
5178    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5179    u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5180    u64 end_blk;
5181    u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
5182
5183    trace_ocfs2_extend_xattr_bucket((unsigned long long)target_blk,
5184                    (unsigned long long)bucket_blkno(first),
5185                    num_clusters, new_bucket);
5186
5187    /* The extent must have room for an additional bucket */
5188    BUG_ON(new_bucket >=
5189           (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
5190
5191    /* end_blk points to the last existing bucket */
5192    end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
5193
5194    /*
5195     * end_blk is the start of the last existing bucket.
5196     * Thus, (end_blk - target_blk) covers the target bucket and
5197     * every bucket after it up to, but not including, the last
5198     * existing bucket. Then we add the last existing bucket, the
5199     * new bucket, and the first bucket (3 * blk_per_bucket).
5200     */
5201    credits = (end_blk - target_blk) + (3 * blk_per_bucket);
5202    ret = ocfs2_extend_trans(handle, credits);
5203    if (ret) {
5204        mlog_errno(ret);
5205        goto out;
5206    }
5207
5208    ret = ocfs2_xattr_bucket_journal_access(handle, first,
5209                        OCFS2_JOURNAL_ACCESS_WRITE);
5210    if (ret) {
5211        mlog_errno(ret);
5212        goto out;
5213    }
5214
5215    while (end_blk != target_blk) {
5216        ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
5217                        end_blk + blk_per_bucket, 0);
5218        if (ret)
5219            goto out;
5220        end_blk -= blk_per_bucket;
5221    }
5222
5223    /* Move half of the xattr in target_blkno to the next bucket. */
5224    ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
5225                    target_blk + blk_per_bucket, NULL, 0);
5226
5227    le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
5228    ocfs2_xattr_bucket_journal_dirty(handle, first);
5229
5230out:
5231    return ret;
5232}
5233
5234/*
5235 * Add new xattr bucket in an extent record and adjust the buckets
5236 * accordingly. xb_bh is the ocfs2_xattr_block, and target is the
5237 * bucket we want to insert into.
5238 *
5239 * In the easy case, we will move all the buckets after target down by
5240 * one. Half of target's xattrs will be moved to the next bucket.
5241 *
5242 * If current cluster is full, we'll allocate a new one. This may not
5243 * be contiguous. The underlying calls will make sure that there is
5244 * space for the insert, shifting buckets around if necessary.
5245 * 'target' may be moved by those calls.
5246 */
5247static int ocfs2_add_new_xattr_bucket(struct inode *inode,
5248                      struct buffer_head *xb_bh,
5249                      struct ocfs2_xattr_bucket *target,
5250                      struct ocfs2_xattr_set_ctxt *ctxt)
5251{
5252    struct ocfs2_xattr_block *xb =
5253            (struct ocfs2_xattr_block *)xb_bh->b_data;
5254    struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
5255    struct ocfs2_extent_list *el = &xb_root->xt_list;
5256    u32 name_hash =
5257        le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
5258    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5259    int ret, num_buckets, extend = 1;
5260    u64 p_blkno;
5261    u32 e_cpos, num_clusters;
5262    /* The bucket at the front of the extent */
5263    struct ocfs2_xattr_bucket *first;
5264
5265    trace_ocfs2_add_new_xattr_bucket(
5266                (unsigned long long)bucket_blkno(target));
5267
5268    /* The first bucket of the original extent */
5269    first = ocfs2_xattr_bucket_new(inode);
5270    if (!first) {
5271        ret = -ENOMEM;
5272        mlog_errno(ret);
5273        goto out;
5274    }
5275
5276    ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
5277                  &num_clusters, el);
5278    if (ret) {
5279        mlog_errno(ret);
5280        goto out;
5281    }
5282
5283    ret = ocfs2_read_xattr_bucket(first, p_blkno);
5284    if (ret) {
5285        mlog_errno(ret);
5286        goto out;
5287    }
5288
5289    num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
5290    if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
5291        /*
5292         * This can move first+target if the target bucket moves
5293         * to the new extent.
5294         */
5295        ret = ocfs2_add_new_xattr_cluster(inode,
5296                          xb_bh,
5297                          first,
5298                          target,
5299                          &num_clusters,
5300                          e_cpos,
5301                          &extend,
5302                          ctxt);
5303        if (ret) {
5304            mlog_errno(ret);
5305            goto out;
5306        }
5307    }
5308
5309    if (extend) {
5310        ret = ocfs2_extend_xattr_bucket(inode,
5311                        ctxt->handle,
5312                        first,
5313                        bucket_blkno(target),
5314                        num_clusters);
5315        if (ret)
5316            mlog_errno(ret);
5317    }
5318
5319out:
5320    ocfs2_xattr_bucket_free(first);
5321
5322    return ret;
5323}
5324
5325static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
5326                    struct ocfs2_xattr_bucket *bucket,
5327                    int offs)
5328{
5329    int block_off = offs >> inode->i_sb->s_blocksize_bits;
5330
5331    offs = offs % inode->i_sb->s_blocksize;
5332    return bucket_block(bucket, block_off) + offs;
5333}
5334
5335/*
5336 * Truncate the specified xe_off entry in xattr bucket.
5337 * bucket is indicated by header_bh and len is the new length.
5338 * Both the ocfs2_xattr_value_root and the entry will be updated here.
5339 *
5340 * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
5341 */
5342static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
5343                         struct ocfs2_xattr_bucket *bucket,
5344                         int xe_off,
5345                         int len,
5346                         struct ocfs2_xattr_set_ctxt *ctxt)
5347{
5348    int ret, offset;
5349    u64 value_blk;
5350    struct ocfs2_xattr_entry *xe;
5351    struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5352    size_t blocksize = inode->i_sb->s_blocksize;
5353    struct ocfs2_xattr_value_buf vb = {
5354        .vb_access = ocfs2_journal_access,
5355    };
5356
5357    xe = &xh->xh_entries[xe_off];
5358
5359    BUG_ON(!xe || ocfs2_xattr_is_local(xe));
5360
5361    offset = le16_to_cpu(xe->xe_name_offset) +
5362         OCFS2_XATTR_SIZE(xe->xe_name_len);
5363
5364    value_blk = offset / blocksize;
5365
5366    /* We don't allow ocfs2_xattr_value to be stored in different block. */
5367    BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
5368
5369    vb.vb_bh = bucket->bu_bhs[value_blk];
5370    BUG_ON(!vb.vb_bh);
5371
5372    vb.vb_xv = (struct ocfs2_xattr_value_root *)
5373        (vb.vb_bh->b_data + offset % blocksize);
5374
5375    /*
5376     * From here on out we have to dirty the bucket. The generic
5377     * value calls only modify one of the bucket's bhs, but we need
5378     * to send the bucket at once. So if they error, they *could* have
5379     * modified something. We have to assume they did, and dirty
5380     * the whole bucket. This leaves us in a consistent state.
5381     */
5382    trace_ocfs2_xattr_bucket_value_truncate(
5383            (unsigned long long)bucket_blkno(bucket), xe_off, len);
5384    ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
5385    if (ret) {
5386        mlog_errno(ret);
5387        goto out;
5388    }
5389
5390    ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
5391                        OCFS2_JOURNAL_ACCESS_WRITE);
5392    if (ret) {
5393        mlog_errno(ret);
5394        goto out;
5395    }
5396
5397    xe->xe_value_size = cpu_to_le64(len);
5398
5399    ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
5400
5401out:
5402    return ret;
5403}
5404
5405static int ocfs2_rm_xattr_cluster(struct inode *inode,
5406                  struct buffer_head *root_bh,
5407                  u64 blkno,
5408                  u32 cpos,
5409                  u32 len,
5410                  void *para)
5411{
5412    int ret;
5413    struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5414    struct inode *tl_inode = osb->osb_tl_inode;
5415    handle_t *handle;
5416    struct ocfs2_xattr_block *xb =
5417            (struct ocfs2_xattr_block *)root_bh->b_data;
5418    struct ocfs2_alloc_context *meta_ac = NULL;
5419    struct ocfs2_cached_dealloc_ctxt dealloc;
5420    struct ocfs2_extent_tree et;
5421
5422    ret = ocfs2_iterate_xattr_buckets(inode, blkno, len,
5423                      ocfs2_delete_xattr_in_bucket, para);
5424    if (ret) {
5425        mlog_errno(ret);
5426        return ret;
5427    }
5428
5429    ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
5430
5431    ocfs2_init_dealloc_ctxt(&dealloc);
5432
5433    trace_ocfs2_rm_xattr_cluster(
5434            (unsigned long long)OCFS2_I(inode)->ip_blkno,
5435            (unsigned long long)blkno, cpos, len);
5436
5437    ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
5438                           len);
5439
5440    ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
5441    if (ret) {
5442        mlog_errno(ret);
5443        return ret;
5444    }
5445
5446    mutex_lock(&tl_inode->i_mutex);
5447
5448    if (ocfs2_truncate_log_needs_flush(osb)) {
5449        ret = __ocfs2_flush_truncate_log(osb);
5450        if (ret < 0) {
5451            mlog_errno(ret);
5452            goto out;
5453        }
5454    }
5455
5456    handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
5457    if (IS_ERR(handle)) {
5458        ret = -ENOMEM;
5459        mlog_errno(ret);
5460        goto out;
5461    }
5462
5463    ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
5464                      OCFS2_JOURNAL_ACCESS_WRITE);
5465    if (ret) {
5466        mlog_errno(ret);
5467        goto out_commit;
5468    }
5469
5470    ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
5471                  &dealloc);
5472    if (ret) {
5473        mlog_errno(ret);
5474        goto out_commit;
5475    }
5476
5477    le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
5478    ocfs2_journal_dirty(handle, root_bh);
5479
5480    ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
5481    if (ret)
5482        mlog_errno(ret);
5483
5484out_commit:
5485    ocfs2_commit_trans(osb, handle);
5486out:
5487    ocfs2_schedule_truncate_log_flush(osb, 1);
5488
5489    mutex_unlock(&tl_inode->i_mutex);
5490
5491    if (meta_ac)
5492        ocfs2_free_alloc_context(meta_ac);
5493
5494    ocfs2_run_deallocs(osb, &dealloc);
5495
5496    return ret;
5497}
5498
5499/*
5500 * check whether the xattr bucket is filled up with the same hash value.
5501 * If we want to insert the xattr with the same hash, return -ENOSPC.
5502 * If we want to insert a xattr with different hash value, go ahead
5503 * and ocfs2_divide_xattr_bucket will handle this.
5504 */
5505static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5506                          struct ocfs2_xattr_bucket *bucket,
5507                          const char *name)
5508{
5509    struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5510    u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5511
5512    if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5513        return 0;
5514
5515    if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5516        xh->xh_entries[0].xe_name_hash) {
5517        mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5518             "hash = %u\n",
5519             (unsigned long long)bucket_blkno(bucket),
5520             le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5521        return -ENOSPC;
5522    }
5523
5524    return 0;
5525}
5526
5527/*
5528 * Try to set the entry in the current bucket. If we fail, the caller
5529 * will handle getting us another bucket.
5530 */
5531static int ocfs2_xattr_set_entry_bucket(struct inode *inode,
5532                    struct ocfs2_xattr_info *xi,
5533                    struct ocfs2_xattr_search *xs,
5534                    struct ocfs2_xattr_set_ctxt *ctxt)
5535{
5536    int ret;
5537    struct ocfs2_xa_loc loc;
5538
5539    trace_ocfs2_xattr_set_entry_bucket(xi->xi_name);
5540
5541    ocfs2_init_xattr_bucket_xa_loc(&loc, xs->bucket,
5542                       xs->not_found ? NULL : xs->here);
5543    ret = ocfs2_xa_set(&loc, xi, ctxt);
5544    if (!ret) {