fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #define MLOG_MASK_PREFIX ML_XATTR
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44 #include "alloc.h"
  45 #include "blockcheck.h"
  46 #include "dlmglue.h"
  47 #include "file.h"
  48 #include "symlink.h"
  49 #include "sysfile.h"
  50 #include "inode.h"
  51 #include "journal.h"
  52 #include "ocfs2_fs.h"
  53 #include "suballoc.h"
  54 #include "uptodate.h"
  55 #include "buffer_head_io.h"
  56 #include "super.h"
  57 #include "xattr.h"
  58
  59
  60 struct ocfs2_xattr_def_value_root {
  61         struct ocfs2_xattr_value_root   xv;
  62         struct ocfs2_extent_rec         er;
  63 };
  64
  65 struct ocfs2_xattr_bucket {
  66         /* The inode these xattrs are associated with */
  67         struct inode *bu_inode;
  68
  69         /* The actual buffers that make up the bucket */
  70         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  71
  72         /* How many blocks make up one bucket for this filesystem */
  73         int bu_blocks;
  74 };
  75
  76 struct ocfs2_xattr_set_ctxt {
  77         handle_t *handle;
  78         struct ocfs2_alloc_context *meta_ac;
  79         struct ocfs2_alloc_context *data_ac;
  80         struct ocfs2_cached_dealloc_ctxt dealloc;
  81 };
  82
  83 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  84 #define OCFS2_XATTR_INLINE_SIZE 80
  85 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  86                                          - sizeof(struct ocfs2_xattr_header) \
  87                                          - sizeof(__u32))
  88 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  89                                          - sizeof(struct ocfs2_xattr_block) \
  90                                          - sizeof(struct ocfs2_xattr_header) \
  91                                          - sizeof(__u32))
  92
  93 static struct ocfs2_xattr_def_value_root def_xv = {
  94         .xv.xr_list.l_count = cpu_to_le16(1),
  95 };
  96
  97 struct xattr_handler *ocfs2_xattr_handlers[] = {
  98         &ocfs2_xattr_user_handler,
  99 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 100         &ocfs2_xattr_acl_access_handler,
 101         &ocfs2_xattr_acl_default_handler,
 102 #endif
 103         &ocfs2_xattr_trusted_handler,
 104         &ocfs2_xattr_security_handler,
 105         NULL
 106 };
 107
 108 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 109         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 110 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 111         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 112                                         = &ocfs2_xattr_acl_access_handler,
 113         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 114                                         = &ocfs2_xattr_acl_default_handler,
 115 #endif
 116         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 117         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 118 };
 119
 120 struct ocfs2_xattr_info {
 121         int name_index;
 122         const char *name;
 123         const void *value;
 124         size_t value_len;
 125 };
 126
 127 struct ocfs2_xattr_search {
 128         struct buffer_head *inode_bh;
 129         /*
 130          * xattr_bh point to the block buffer head which has extended attribute
 131          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 132          */
 133         struct buffer_head *xattr_bh;
 134         struct ocfs2_xattr_header *header;
 135         struct ocfs2_xattr_bucket *bucket;
 136         void *base;
 137         void *end;
 138         struct ocfs2_xattr_entry *here;
 139         int not_found;
 140 };
 141
 142 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 143                                              struct ocfs2_xattr_header *xh,
 144                                              int index,
 145                                              int *block_off,
 146                                              int *new_offset);
 147
 148 static int ocfs2_xattr_block_find(struct inode *inode,
 149                                   int name_index,
 150                                   const char *name,
 151                                   struct ocfs2_xattr_search *xs);
 152 static int ocfs2_xattr_index_block_find(struct inode *inode,
 153                                         struct buffer_head *root_bh,
 154                                         int name_index,
 155                                         const char *name,
 156                                         struct ocfs2_xattr_search *xs);
 157
 158 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 159                                         struct ocfs2_xattr_tree_root *xt,
 160                                         char *buffer,
 161                                         size_t buffer_size);
 162
 163 static int ocfs2_xattr_create_index_block(struct inode *inode,
 164                                           struct ocfs2_xattr_search *xs,
 165                                           struct ocfs2_xattr_set_ctxt *ctxt);
 166
 167 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 168                                              struct ocfs2_xattr_info *xi,
 169                                              struct ocfs2_xattr_search *xs,
 170                                              struct ocfs2_xattr_set_ctxt *ctxt);
 171
 172 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 173                                           struct buffer_head *xb_bh);
 174 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 175                                   u64 src_blk, u64 last_blk, u64 to_blk,
 176                                   unsigned int start_bucket,
 177                                   u32 *first_hash);
 178
 179 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 180 {
 181         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 182 }
 183
 184 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 185 {
 186         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 187 }
 188
 189 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
 190 {
 191         u16 len = sb->s_blocksize -
 192                  offsetof(struct ocfs2_xattr_header, xh_entries);
 193
 194         return len / sizeof(struct ocfs2_xattr_entry);
 195 }
 196
 197 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 198 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 199 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 200
 201 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 202 {
 203         struct ocfs2_xattr_bucket *bucket;
 204         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 205
 206         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 207
 208         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 209         if (bucket) {
 210                 bucket->bu_inode = inode;
 211                 bucket->bu_blocks = blks;
 212         }
 213
 214         return bucket;
 215 }
 216
 217 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 218 {
 219         int i;
 220
 221         for (i = 0; i < bucket->bu_blocks; i++) {
 222                 brelse(bucket->bu_bhs[i]);
 223                 bucket->bu_bhs[i] = NULL;
 224         }
 225 }
 226
 227 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 228 {
 229         if (bucket) {
 230                 ocfs2_xattr_bucket_relse(bucket);
 231                 bucket->bu_inode = NULL;
 232                 kfree(bucket);
 233         }
 234 }
 235
 236 /*
 237  * A bucket that has never been written to disk doesn't need to be
 238  * read.  We just need the buffer_heads.  Don't call this for
 239  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 240  * them fully.
 241  */
 242 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 243                                    u64 xb_blkno)
 244 {
 245         int i, rc = 0;
 246
 247         for (i = 0; i < bucket->bu_blocks; i++) {
 248                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 249                                               xb_blkno + i);
 250                 if (!bucket->bu_bhs[i]) {
 251                         rc = -EIO;
 252                         mlog_errno(rc);
 253                         break;
 254                 }
 255
 256                 if (!ocfs2_buffer_uptodate(bucket->bu_inode,
 257                                            bucket->bu_bhs[i]))
 258                         ocfs2_set_new_buffer_uptodate(bucket->bu_inode,
 259                                                       bucket->bu_bhs[i]);
 260         }
 261
 262         if (rc)
 263                 ocfs2_xattr_bucket_relse(bucket);
 264         return rc;
 265 }
 266
 267 /* Read the xattr bucket at xb_blkno */
 268 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 269                                    u64 xb_blkno)
 270 {
 271         int rc;
 272
 273         rc = ocfs2_read_blocks(bucket->bu_inode, xb_blkno,
 274                                bucket->bu_blocks, bucket->bu_bhs, 0,
 275                                NULL);
 276         if (!rc) {
 277                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 278                                                  bucket->bu_bhs,
 279                                                  bucket->bu_blocks,
 280                                                  &bucket_xh(bucket)->xh_check);
 281                 if (rc)
 282                         mlog_errno(rc);
 283         }
 284
 285         if (rc)
 286                 ocfs2_xattr_bucket_relse(bucket);
 287         return rc;
 288 }
 289
 290 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 291                                              struct ocfs2_xattr_bucket *bucket,
 292                                              int type)
 293 {
 294         int i, rc = 0;
 295
 296         for (i = 0; i < bucket->bu_blocks; i++) {
 297                 rc = ocfs2_journal_access(handle, bucket->bu_inode,
 298                                           bucket->bu_bhs[i], type);
 299                 if (rc) {
 300                         mlog_errno(rc);
 301                         break;
 302                 }
 303         }
 304
 305         return rc;
 306 }
 307
 308 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 309                                              struct ocfs2_xattr_bucket *bucket)
 310 {
 311         int i;
 312
 313         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 314                                    bucket->bu_bhs, bucket->bu_blocks,
 315                                    &bucket_xh(bucket)->xh_check);
 316
 317         for (i = 0; i < bucket->bu_blocks; i++)
 318                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 319 }
 320
 321 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 322                                          struct ocfs2_xattr_bucket *src)
 323 {
 324         int i;
 325         int blocksize = src->bu_inode->i_sb->s_blocksize;
 326
 327         BUG_ON(dest->bu_blocks != src->bu_blocks);
 328         BUG_ON(dest->bu_inode != src->bu_inode);
 329
 330         for (i = 0; i < src->bu_blocks; i++) {
 331                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 332                        blocksize);
 333         }
 334 }
 335
 336 static int ocfs2_validate_xattr_block(struct super_block *sb,
 337                                       struct buffer_head *bh)
 338 {
 339         int rc;
 340         struct ocfs2_xattr_block *xb =
 341                 (struct ocfs2_xattr_block *)bh->b_data;
 342
 343         mlog(0, "Validating xattr block %llu\n",
 344              (unsigned long long)bh->b_blocknr);
 345
 346         BUG_ON(!buffer_uptodate(bh));
 347
 348         /*
 349          * If the ecc fails, we return the error but otherwise
 350          * leave the filesystem running.  We know any error is
 351          * local to this block.
 352          */
 353         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 354         if (rc)
 355                 return rc;
 356
 357         /*
 358          * Errors after here are fatal
 359          */
 360
 361         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 362                 ocfs2_error(sb,
 363                             "Extended attribute block #%llu has bad "
 364                             "signature %.*s",
 365                             (unsigned long long)bh->b_blocknr, 7,
 366                             xb->xb_signature);
 367                 return -EINVAL;
 368         }
 369
 370         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 371                 ocfs2_error(sb,
 372                             "Extended attribute block #%llu has an "
 373                             "invalid xb_blkno of %llu",
 374                             (unsigned long long)bh->b_blocknr,
 375                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 376                 return -EINVAL;
 377         }
 378
 379         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 380                 ocfs2_error(sb,
 381                             "Extended attribute block #%llu has an invalid "
 382                             "xb_fs_generation of #%u",
 383                             (unsigned long long)bh->b_blocknr,
 384                             le32_to_cpu(xb->xb_fs_generation));
 385                 return -EINVAL;
 386         }
 387
 388         return 0;
 389 }
 390
 391 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 392                                   struct buffer_head **bh)
 393 {
 394         int rc;
 395         struct buffer_head *tmp = *bh;
 396
 397         rc = ocfs2_read_block(inode, xb_blkno, &tmp,
 398                               ocfs2_validate_xattr_block);
 399
 400         /* If ocfs2_read_block() got us a new bh, pass it up. */
 401         if (!rc && !*bh)
 402                 *bh = tmp;
 403
 404         return rc;
 405 }
 406
 407 static inline const char *ocfs2_xattr_prefix(int name_index)
 408 {
 409         struct xattr_handler *handler = NULL;
 410
 411         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 412                 handler = ocfs2_xattr_handler_map[name_index];
 413
 414         return handler ? handler->prefix : NULL;
 415 }
 416
 417 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 418                                  const char *name,
 419                                  int name_len)
 420 {
 421         /* Get hash value of uuid from super block */
 422         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 423         int i;
 424
 425         /* hash extended attribute name */
 426         for (i = 0; i < name_len; i++) {
 427                 hash = (hash << OCFS2_HASH_SHIFT) ^
 428                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 429                        *name++;
 430         }
 431
 432         return hash;
 433 }
 434
 435 /*
 436  * ocfs2_xattr_hash_entry()
 437  *
 438  * Compute the hash of an extended attribute.
 439  */
 440 static void ocfs2_xattr_hash_entry(struct inode *inode,
 441                                    struct ocfs2_xattr_header *header,
 442                                    struct ocfs2_xattr_entry *entry)
 443 {
 444         u32 hash = 0;
 445         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 446
 447         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
 448         entry->xe_name_hash = cpu_to_le32(hash);
 449
 450         return;
 451 }
 452
 453 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 454 {
 455         int size = 0;
 456
 457         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
 458                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 459         else
 460                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 461         size += sizeof(struct ocfs2_xattr_entry);
 462
 463         return size;
 464 }
 465
 466 int ocfs2_calc_security_init(struct inode *dir,
 467                              struct ocfs2_security_xattr_info *si,
 468                              int *want_clusters,
 469                              int *xattr_credits,
 470                              struct ocfs2_alloc_context **xattr_ac)
 471 {
 472         int ret = 0;
 473         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 474         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 475                                                  si->value_len);
 476
 477         /*
 478          * The max space of security xattr taken inline is
 479          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 480          * So reserve one metadata block for it is ok.
 481          */
 482         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 483             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 484                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 485                 if (ret) {
 486                         mlog_errno(ret);
 487                         return ret;
 488                 }
 489                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 490         }
 491
 492         /* reserve clusters for xattr value which will be set in B tree*/
 493         if (si->value_len > OCFS2_XATTR_INLINE_SIZE)
 494                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
 495                                                            si->value_len);
 496         return ret;
 497 }
 498
 499 int ocfs2_calc_xattr_init(struct inode *dir,
 500                           struct buffer_head *dir_bh,
 501                           int mode,
 502                           struct ocfs2_security_xattr_info *si,
 503                           int *want_clusters,
 504                           int *xattr_credits,
 505                           struct ocfs2_alloc_context **xattr_ac)
 506 {
 507         int ret = 0;
 508         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 509         int s_size = 0;
 510         int a_size = 0;
 511         int acl_len = 0;
 512
 513         if (si->enable)
 514                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 515                                                      si->value_len);
 516
 517         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 518                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 519                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 520                                         "", NULL, 0);
 521                 if (acl_len > 0) {
 522                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 523                         if (S_ISDIR(mode))
 524                                 a_size <<= 1;
 525                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 526                         mlog_errno(ret);
 527                         return ret;
 528                 }
 529         }
 530
 531         if (!(s_size + a_size))
 532                 return ret;
 533
 534         /*
 535          * The max space of security xattr taken inline is
 536          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 537          * The max space of acl xattr taken inline is
 538          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 539          * when blocksize = 512, may reserve one more cluser for
 540          * xattr bucket, otherwise reserve one metadata block
 541          * for them is ok.
 542          */
 543         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 544             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 545                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 546                 if (ret) {
 547                         mlog_errno(ret);
 548                         return ret;
 549                 }
 550                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 551         }
 552
 553         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 554             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 555                 *want_clusters += 1;
 556                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 557         }
 558
 559         /* reserve clusters for xattr value which will be set in B tree*/
 560         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE)
 561                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
 562                                                            si->value_len);
 563         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 564             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 565                 *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 566                 if (S_ISDIR(mode))
 567                         *want_clusters += ocfs2_clusters_for_bytes(dir->i_sb,
 568                                                                    acl_len);
 569         }
 570
 571         return ret;
 572 }
 573
 574 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 575                                          u32 clusters_to_add,
 576                                          struct ocfs2_xattr_value_buf *vb,
 577                                          struct ocfs2_xattr_set_ctxt *ctxt)
 578 {
 579         int status = 0;
 580         handle_t *handle = ctxt->handle;
 581         enum ocfs2_alloc_restarted why;
 582         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 583         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 584         struct ocfs2_extent_tree et;
 585
 586         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 587
 588         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 589
 590         status = vb->vb_access(handle, inode, vb->vb_bh,
 591                               OCFS2_JOURNAL_ACCESS_WRITE);
 592         if (status < 0) {
 593                 mlog_errno(status);
 594                 goto leave;
 595         }
 596
 597         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 598         status = ocfs2_add_clusters_in_btree(osb,
 599                                              inode,
 600                                              &logical_start,
 601                                              clusters_to_add,
 602                                              0,
 603                                              &et,
 604                                              handle,
 605                                              ctxt->data_ac,
 606                                              ctxt->meta_ac,
 607                                              &why);
 608         if (status < 0) {
 609                 mlog_errno(status);
 610                 goto leave;
 611         }
 612
 613         status = ocfs2_journal_dirty(handle, vb->vb_bh);
 614         if (status < 0) {
 615                 mlog_errno(status);
 616                 goto leave;
 617         }
 618
 619         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
 620
 621         /*
 622          * We should have already allocated enough space before the transaction,
 623          * so no need to restart.
 624          */
 625         BUG_ON(why != RESTART_NONE || clusters_to_add);
 626
 627 leave:
 628
 629         return status;
 630 }
 631
 632 static int __ocfs2_remove_xattr_range(struct inode *inode,
 633                                       struct ocfs2_xattr_value_buf *vb,
 634                                       u32 cpos, u32 phys_cpos, u32 len,
 635                                       struct ocfs2_xattr_set_ctxt *ctxt)
 636 {
 637         int ret;
 638         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 639         handle_t *handle = ctxt->handle;
 640         struct ocfs2_extent_tree et;
 641
 642         ocfs2_init_xattr_value_extent_tree(&et, inode, vb);
 643
 644         ret = vb->vb_access(handle, inode, vb->vb_bh,
 645                             OCFS2_JOURNAL_ACCESS_WRITE);
 646         if (ret) {
 647                 mlog_errno(ret);
 648                 goto out;
 649         }
 650
 651         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, ctxt->meta_ac,
 652                                   &ctxt->dealloc);
 653         if (ret) {
 654                 mlog_errno(ret);
 655                 goto out;
 656         }
 657
 658         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 659
 660         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
 661         if (ret) {
 662                 mlog_errno(ret);
 663                 goto out;
 664         }
 665
 666         ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc, phys_blkno, len);
 667         if (ret)
 668                 mlog_errno(ret);
 669
 670 out:
 671         return ret;
 672 }
 673
 674 static int ocfs2_xattr_shrink_size(struct inode *inode,
 675                                    u32 old_clusters,
 676                                    u32 new_clusters,
 677                                    struct ocfs2_xattr_value_buf *vb,
 678                                    struct ocfs2_xattr_set_ctxt *ctxt)
 679 {
 680         int ret = 0;
 681         u32 trunc_len, cpos, phys_cpos, alloc_size;
 682         u64 block;
 683
 684         if (old_clusters <= new_clusters)
 685                 return 0;
 686
 687         cpos = new_clusters;
 688         trunc_len = old_clusters - new_clusters;
 689         while (trunc_len) {
 690                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 691                                                &alloc_size,
 692                                                &vb->vb_xv->xr_list);
 693                 if (ret) {
 694                         mlog_errno(ret);
 695                         goto out;
 696                 }
 697
 698                 if (alloc_size > trunc_len)
 699                         alloc_size = trunc_len;
 700
 701                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 702                                                  phys_cpos, alloc_size,
 703                                                  ctxt);
 704                 if (ret) {
 705                         mlog_errno(ret);
 706                         goto out;
 707                 }
 708
 709                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 710                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 711                                                        alloc_size);
 712                 cpos += alloc_size;
 713                 trunc_len -= alloc_size;
 714         }
 715
 716 out:
 717         return ret;
 718 }
 719
 720 static int ocfs2_xattr_value_truncate(struct inode *inode,
 721                                       struct ocfs2_xattr_value_buf *vb,
 722                                       int len,
 723                                       struct ocfs2_xattr_set_ctxt *ctxt)
 724 {
 725         int ret;
 726         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 727         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 728
 729         if (new_clusters == old_clusters)
 730                 return 0;
 731
 732         if (new_clusters > old_clusters)
 733                 ret = ocfs2_xattr_extend_allocation(inode,
 734                                                     new_clusters - old_clusters,
 735                                                     vb, ctxt);
 736         else
 737                 ret = ocfs2_xattr_shrink_size(inode,
 738                                               old_clusters, new_clusters,
 739                                               vb, ctxt);
 740
 741         return ret;
 742 }
 743
 744 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 745                                   size_t *result, const char *prefix,
 746                                   const char *name, int name_len)
 747 {
 748         char *p = buffer + *result;
 749         int prefix_len = strlen(prefix);
 750         int total_len = prefix_len + name_len + 1;
 751
 752         *result += total_len;
 753
 754         /* we are just looking for how big our buffer needs to be */
 755         if (!size)
 756                 return 0;
 757
 758         if (*result > size)
 759                 return -ERANGE;
 760
 761         memcpy(p, prefix, prefix_len);
 762         memcpy(p + prefix_len, name, name_len);
 763         p[prefix_len + name_len] = '\0';
 764
 765         return 0;
 766 }
 767
 768 static int ocfs2_xattr_list_entries(struct inode *inode,
 769                                     struct ocfs2_xattr_header *header,
 770                                     char *buffer, size_t buffer_size)
 771 {
 772         size_t result = 0;
 773         int i, type, ret;
 774         const char *prefix, *name;
 775
 776         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 777                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 778                 type = ocfs2_xattr_get_type(entry);
 779                 prefix = ocfs2_xattr_prefix(type);
 780
 781                 if (prefix) {
 782                         name = (const char *)header +
 783                                 le16_to_cpu(entry->xe_name_offset);
 784
 785                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 786                                                      &result, prefix, name,
 787                                                      entry->xe_name_len);
 788                         if (ret)
 789                                 return ret;
 790                 }
 791         }
 792
 793         return result;
 794 }
 795
 796 static int ocfs2_xattr_ibody_list(struct inode *inode,
 797                                   struct ocfs2_dinode *di,
 798                                   char *buffer,
 799                                   size_t buffer_size)
 800 {
 801         struct ocfs2_xattr_header *header = NULL;
 802         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 803         int ret = 0;
 804
 805         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 806                 return ret;
 807
 808         header = (struct ocfs2_xattr_header *)
 809                  ((void *)di + inode->i_sb->s_blocksize -
 810                  le16_to_cpu(di->i_xattr_inline_size));
 811
 812         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 813
 814         return ret;
 815 }
 816
 817 static int ocfs2_xattr_block_list(struct inode *inode,
 818                                   struct ocfs2_dinode *di,
 819                                   char *buffer,
 820                                   size_t buffer_size)
 821 {
 822         struct buffer_head *blk_bh = NULL;
 823         struct ocfs2_xattr_block *xb;
 824         int ret = 0;
 825
 826         if (!di->i_xattr_loc)
 827                 return ret;
 828
 829         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 830                                      &blk_bh);
 831         if (ret < 0) {
 832                 mlog_errno(ret);
 833                 return ret;
 834         }
 835
 836         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 837         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 838                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 839                 ret = ocfs2_xattr_list_entries(inode, header,
 840                                                buffer, buffer_size);
 841         } else {
 842                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 843                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 844                                                    buffer, buffer_size);
 845         }
 846
 847         brelse(blk_bh);
 848
 849         return ret;
 850 }
 851
 852 ssize_t ocfs2_listxattr(struct dentry *dentry,
 853                         char *buffer,
 854                         size_t size)
 855 {
 856         int ret = 0, i_ret = 0, b_ret = 0;
 857         struct buffer_head *di_bh = NULL;
 858         struct ocfs2_dinode *di = NULL;
 859         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 860
 861         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 862                 return -EOPNOTSUPP;
 863
 864         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 865                 return ret;
 866
 867         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 868         if (ret < 0) {
 869                 mlog_errno(ret);
 870                 return ret;
 871         }
 872
 873         di = (struct ocfs2_dinode *)di_bh->b_data;
 874
 875         down_read(&oi->ip_xattr_sem);
 876         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 877         if (i_ret < 0)
 878                 b_ret = 0;
 879         else {
 880                 if (buffer) {
 881                         buffer += i_ret;
 882                         size -= i_ret;
 883                 }
 884                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 885                                                buffer, size);
 886                 if (b_ret < 0)
 887                         i_ret = 0;
 888         }
 889         up_read(&oi->ip_xattr_sem);
 890         ocfs2_inode_unlock(dentry->d_inode, 0);
 891
 892         brelse(di_bh);
 893
 894         return i_ret + b_ret;
 895 }
 896
 897 static int ocfs2_xattr_find_entry(int name_index,
 898                                   const char *name,
 899                                   struct ocfs2_xattr_search *xs)
 900 {
 901         struct ocfs2_xattr_entry *entry;
 902         size_t name_len;
 903         int i, cmp = 1;
 904
 905         if (name == NULL)
 906                 return -EINVAL;
 907
 908         name_len = strlen(name);
 909         entry = xs->here;
 910         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 911                 cmp = name_index - ocfs2_xattr_get_type(entry);
 912                 if (!cmp)
 913                         cmp = name_len - entry->xe_name_len;
 914                 if (!cmp)
 915                         cmp = memcmp(name, (xs->base +
 916                                      le16_to_cpu(entry->xe_name_offset)),
 917                                      name_len);
 918                 if (cmp == 0)
 919                         break;
 920                 entry += 1;
 921         }
 922         xs->here = entry;
 923
 924         return cmp ? -ENODATA : 0;
 925 }
 926
 927 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 928                                          struct ocfs2_xattr_value_root *xv,
 929                                          void *buffer,
 930                                          size_t len)
 931 {
 932         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 933         u64 blkno;
 934         int i, ret = 0;
 935         size_t cplen, blocksize;
 936         struct buffer_head *bh = NULL;
 937         struct ocfs2_extent_list *el;
 938
 939         el = &xv->xr_list;
 940         clusters = le32_to_cpu(xv->xr_clusters);
 941         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 942         blocksize = inode->i_sb->s_blocksize;
 943
 944         cpos = 0;
 945         while (cpos < clusters) {
 946                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 947                                                &num_clusters, el);
 948                 if (ret) {
 949                         mlog_errno(ret);
 950                         goto out;
 951                 }
 952
 953                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 954                 /* Copy ocfs2_xattr_value */
 955                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 956                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
 957                         if (ret) {
 958                                 mlog_errno(ret);
 959                                 goto out;
 960                         }
 961
 962                         cplen = len >= blocksize ? blocksize : len;
 963                         memcpy(buffer, bh->b_data, cplen);
 964                         len -= cplen;
 965                         buffer += cplen;
 966
 967                         brelse(bh);
 968                         bh = NULL;
 969                         if (len == 0)
 970                                 break;
 971                 }
 972                 cpos += num_clusters;
 973         }
 974 out:
 975         return ret;
 976 }
 977
 978 static int ocfs2_xattr_ibody_get(struct inode *inode,
 979                                  int name_index,
 980                                  const char *name,
 981                                  void *buffer,
 982                                  size_t buffer_size,
 983                                  struct ocfs2_xattr_search *xs)
 984 {
 985         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 986         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 987         struct ocfs2_xattr_value_root *xv;
 988         size_t size;
 989         int ret = 0;
 990
 991         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 992                 return -ENODATA;
 993
 994         xs->end = (void *)di + inode->i_sb->s_blocksize;
 995         xs->header = (struct ocfs2_xattr_header *)
 996                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
 997         xs->base = (void *)xs->header;
 998         xs->here = xs->header->xh_entries;
 999
1000         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1001         if (ret)
1002                 return ret;
1003         size = le64_to_cpu(xs->here->xe_value_size);
1004         if (buffer) {
1005                 if (size > buffer_size)
1006                         return -ERANGE;
1007                 if (ocfs2_xattr_is_local(xs->here)) {
1008                         memcpy(buffer, (void *)xs->base +
1009                                le16_to_cpu(xs->here->xe_name_offset) +
1010                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1011                 } else {
1012                         xv = (struct ocfs2_xattr_value_root *)
1013                                 (xs->base + le16_to_cpu(
1014                                  xs->here->xe_name_offset) +
1015                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1016                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1017                                                             buffer, size);
1018                         if (ret < 0) {
1019                                 mlog_errno(ret);
1020                                 return ret;
1021                         }
1022                 }
1023         }
1024
1025         return size;
1026 }
1027
1028 static int ocfs2_xattr_block_get(struct inode *inode,
1029                                  int name_index,
1030                                  const char *name,
1031                                  void *buffer,
1032                                  size_t buffer_size,
1033                                  struct ocfs2_xattr_search *xs)
1034 {
1035         struct ocfs2_xattr_block *xb;
1036         struct ocfs2_xattr_value_root *xv;
1037         size_t size;
1038         int ret = -ENODATA, name_offset, name_len, block_off, i;
1039
1040         xs->bucket = ocfs2_xattr_bucket_new(inode);
1041         if (!xs->bucket) {
1042                 ret = -ENOMEM;
1043                 mlog_errno(ret);
1044                 goto cleanup;
1045         }
1046
1047         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1048         if (ret) {
1049                 mlog_errno(ret);
1050                 goto cleanup;
1051         }
1052
1053         if (xs->not_found) {
1054                 ret = -ENODATA;
1055                 goto cleanup;
1056         }
1057
1058         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1059         size = le64_to_cpu(xs->here->xe_value_size);
1060         if (buffer) {
1061                 ret = -ERANGE;
1062                 if (size > buffer_size)
1063                         goto cleanup;
1064
1065                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1066                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1067                 i = xs->here - xs->header->xh_entries;
1068
1069                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1070                         ret = ocfs2_xattr_bucket_get_name_value(inode,
1071                                                                 bucket_xh(xs->bucket),
1072                                                                 i,
1073                                                                 &block_off,
1074                                                                 &name_offset);
1075                         xs->base = bucket_block(xs->bucket, block_off);
1076                 }
1077                 if (ocfs2_xattr_is_local(xs->here)) {
1078                         memcpy(buffer, (void *)xs->base +
1079                                name_offset + name_len, size);
1080                 } else {
1081                         xv = (struct ocfs2_xattr_value_root *)
1082                                 (xs->base + name_offset + name_len);
1083                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1084                                                             buffer, size);
1085                         if (ret < 0) {
1086                                 mlog_errno(ret);
1087                                 goto cleanup;
1088                         }
1089                 }
1090         }
1091         ret = size;
1092 cleanup:
1093         ocfs2_xattr_bucket_free(xs->bucket);
1094
1095         brelse(xs->xattr_bh);
1096         xs->xattr_bh = NULL;
1097         return ret;
1098 }
1099
1100 int ocfs2_xattr_get_nolock(struct inode *inode,
1101                            struct buffer_head *di_bh,
1102                            int name_index,
1103                            const char *name,
1104                            void *buffer,
1105                            size_t buffer_size)
1106 {
1107         int ret;
1108         struct ocfs2_dinode *di = NULL;
1109         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1110         struct ocfs2_xattr_search xis = {
1111                 .not_found = -ENODATA,
1112         };
1113         struct ocfs2_xattr_search xbs = {
1114                 .not_found = -ENODATA,
1115         };
1116
1117         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1118                 return -EOPNOTSUPP;
1119
1120         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1121                 ret = -ENODATA;
1122
1123         xis.inode_bh = xbs.inode_bh = di_bh;
1124         di = (struct ocfs2_dinode *)di_bh->b_data;
1125
1126         down_read(&oi->ip_xattr_sem);
1127         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1128                                     buffer_size, &xis);
1129         if (ret == -ENODATA && di->i_xattr_loc)
1130                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1131                                             buffer_size, &xbs);
1132         up_read(&oi->ip_xattr_sem);
1133
1134         return ret;
1135 }
1136
1137 /* ocfs2_xattr_get()
1138  *
1139  * Copy an extended attribute into the buffer provided.
1140  * Buffer is NULL to compute the size of buffer required.
1141  */
1142 static int ocfs2_xattr_get(struct inode *inode,
1143                            int name_index,
1144                            const char *name,
1145                            void *buffer,
1146                            size_t buffer_size)
1147 {
1148         int ret;
1149         struct buffer_head *di_bh = NULL;
1150
1151         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1152         if (ret < 0) {
1153                 mlog_errno(ret);
1154                 return ret;
1155         }
1156         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1157                                      name, buffer, buffer_size);
1158
1159         ocfs2_inode_unlock(inode, 0);
1160
1161         brelse(di_bh);
1162
1163         return ret;
1164 }
1165
1166 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1167                                            handle_t *handle,
1168                                            struct ocfs2_xattr_value_root *xv,
1169                                            const void *value,
1170                                            int value_len)
1171 {
1172         int ret = 0, i, cp_len;
1173         u16 blocksize = inode->i_sb->s_blocksize;
1174         u32 p_cluster, num_clusters;
1175         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1176         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1177         u64 blkno;
1178         struct buffer_head *bh = NULL;
1179
1180         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1181
1182         while (cpos < clusters) {
1183                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1184                                                &num_clusters, &xv->xr_list);
1185                 if (ret) {
1186                         mlog_errno(ret);
1187                         goto out;
1188                 }
1189
1190                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1191
1192                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1193                         ret = ocfs2_read_block(inode, blkno, &bh, NULL);
1194                         if (ret) {
1195                                 mlog_errno(ret);
1196                                 goto out;
1197                         }
1198
1199                         ret = ocfs2_journal_access(handle,
1200                                                    inode,
1201                                                    bh,
1202                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1203                         if (ret < 0) {
1204                                 mlog_errno(ret);
1205                                 goto out;
1206                         }
1207
1208                         cp_len = value_len > blocksize ? blocksize : value_len;
1209                         memcpy(bh->b_data, value, cp_len);
1210                         value_len -= cp_len;
1211                         value += cp_len;
1212                         if (cp_len < blocksize)
1213                                 memset(bh->b_data + cp_len, 0,
1214                                        blocksize - cp_len);
1215
1216                         ret = ocfs2_journal_dirty(handle, bh);
1217                         if (ret < 0) {
1218                                 mlog_errno(ret);
1219                                 goto out;
1220                         }
1221                         brelse(bh);
1222                         bh = NULL;
1223
1224                         /*
1225                          * XXX: do we need to empty all the following
1226                          * blocks in this cluster?
1227                          */
1228                         if (!value_len)
1229                                 break;
1230                 }
1231                 cpos += num_clusters;
1232         }
1233 out:
1234         brelse(bh);
1235
1236         return ret;
1237 }
1238
1239 static int ocfs2_xattr_cleanup(struct inode *inode,
1240                                handle_t *handle,
1241                                struct ocfs2_xattr_info *xi,
1242                                struct ocfs2_xattr_search *xs,
1243                                struct ocfs2_xattr_value_buf *vb,
1244                                size_t offs)
1245 {
1246         int ret = 0;
1247         size_t name_len = strlen(xi->name);
1248         void *val = xs->base + offs;
1249         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1250
1251         ret = vb->vb_access(handle, inode, vb->vb_bh,
1252                             OCFS2_JOURNAL_ACCESS_WRITE);
1253         if (ret) {
1254                 mlog_errno(ret);
1255                 goto out;
1256         }
1257         /* Decrease xattr count */
1258         le16_add_cpu(&xs->header->xh_count, -1);
1259         /* Remove the xattr entry and tree root which has already be set*/
1260         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1261         memset(val, 0, size);
1262
1263         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1264         if (ret < 0)
1265                 mlog_errno(ret);
1266 out:
1267         return ret;
1268 }
1269
1270 static int ocfs2_xattr_update_entry(struct inode *inode,
1271                                     handle_t *handle,
1272                                     struct ocfs2_xattr_info *xi,
1273                                     struct ocfs2_xattr_search *xs,
1274                                     struct ocfs2_xattr_value_buf *vb,
1275                                     size_t offs)
1276 {
1277         int ret;
1278
1279         ret = vb->vb_access(handle, inode, vb->vb_bh,
1280                             OCFS2_JOURNAL_ACCESS_WRITE);
1281         if (ret) {
1282                 mlog_errno(ret);
1283                 goto out;
1284         }
1285
1286         xs->here->xe_name_offset = cpu_to_le16(offs);
1287         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1288         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1289                 ocfs2_xattr_set_local(xs->here, 1);
1290         else
1291                 ocfs2_xattr_set_local(xs->here, 0);
1292         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1293
1294         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1295         if (ret < 0)
1296                 mlog_errno(ret);
1297 out:
1298         return ret;
1299 }
1300
1301 /*
1302  * ocfs2_xattr_set_value_outside()
1303  *
1304  * Set large size value in B tree.
1305  */
1306 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1307                                          struct ocfs2_xattr_info *xi,
1308                                          struct ocfs2_xattr_search *xs,
1309                                          struct ocfs2_xattr_set_ctxt *ctxt,
1310                                          struct ocfs2_xattr_value_buf *vb,
1311                                          size_t offs)
1312 {
1313         size_t name_len = strlen(xi->name);
1314         void *val = xs->base + offs;
1315         struct ocfs2_xattr_value_root *xv = NULL;
1316         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1317         int ret = 0;
1318
1319         memset(val, 0, size);
1320         memcpy(val, xi->name, name_len);
1321         xv = (struct ocfs2_xattr_value_root *)
1322                 (val + OCFS2_XATTR_SIZE(name_len));
1323         xv->xr_clusters = 0;
1324         xv->xr_last_eb_blk = 0;
1325         xv->xr_list.l_tree_depth = 0;
1326         xv->xr_list.l_count = cpu_to_le16(1);
1327         xv->xr_list.l_next_free_rec = 0;
1328         vb->vb_xv = xv;
1329
1330         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1331         if (ret < 0) {
1332                 mlog_errno(ret);
1333                 return ret;
1334         }
1335         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1336         if (ret < 0) {
1337                 mlog_errno(ret);
1338                 return ret;
1339         }
1340         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb->vb_xv,
1341                                               xi->value, xi->value_len);
1342         if (ret < 0)
1343                 mlog_errno(ret);
1344
1345         return ret;
1346 }
1347
1348 /*
1349  * ocfs2_xattr_set_entry_local()
1350  *
1351  * Set, replace or remove extended attribute in local.
1352  */
1353 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1354                                         struct ocfs2_xattr_info *xi,
1355                                         struct ocfs2_xattr_search *xs,
1356                                         struct ocfs2_xattr_entry *last,
1357                                         size_t min_offs)
1358 {
1359         size_t name_len = strlen(xi->name);
1360         int i;
1361
1362         if (xi->value && xs->not_found) {
1363                 /* Insert the new xattr entry. */
1364                 le16_add_cpu(&xs->header->xh_count, 1);
1365                 ocfs2_xattr_set_type(last, xi->name_index);
1366                 ocfs2_xattr_set_local(last, 1);
1367                 last->xe_name_len = name_len;
1368         } else {
1369                 void *first_val;
1370                 void *val;
1371                 size_t offs, size;
1372
1373                 first_val = xs->base + min_offs;
1374                 offs = le16_to_cpu(xs->here->xe_name_offset);
1375                 val = xs->base + offs;
1376
1377                 if (le64_to_cpu(xs->here->xe_value_size) >
1378                     OCFS2_XATTR_INLINE_SIZE)
1379                         size = OCFS2_XATTR_SIZE(name_len) +
1380                                 OCFS2_XATTR_ROOT_SIZE;
1381                 else
1382                         size = OCFS2_XATTR_SIZE(name_len) +
1383                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1384
1385                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1386                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1387                         /* The old and the new value have the
1388                            same size. Just replace the value. */
1389                         ocfs2_xattr_set_local(xs->here, 1);
1390                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1391                         /* Clear value bytes. */
1392                         memset(val + OCFS2_XATTR_SIZE(name_len),
1393                                0,
1394                                OCFS2_XATTR_SIZE(xi->value_len));
1395                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1396                                xi->value,
1397                                xi->value_len);
1398                         return;
1399                 }
1400                 /* Remove the old name+value. */
1401                 memmove(first_val + size, first_val, val - first_val);
1402                 memset(first_val, 0, size);
1403                 xs->here->xe_name_hash = 0;
1404                 xs->here->xe_name_offset = 0;
1405                 ocfs2_xattr_set_local(xs->here, 1);
1406                 xs->here->xe_value_size = 0;
1407
1408                 min_offs += size;
1409
1410                 /* Adjust all value offsets. */
1411                 last = xs->header->xh_entries;
1412                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1413                         size_t o = le16_to_cpu(last->xe_name_offset);
1414
1415                         if (o < offs)
1416                                 last->xe_name_offset = cpu_to_le16(o + size);
1417                         last += 1;
1418                 }
1419
1420                 if (!xi->value) {
1421                         /* Remove the old entry. */
1422                         last -= 1;
1423                         memmove(xs->here, xs->here + 1,
1424                                 (void *)last - (void *)xs->here);
1425                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1426                         le16_add_cpu(&xs->header->xh_count, -1);
1427                 }
1428         }
1429         if (xi->value) {
1430                 /* Insert the new name+value. */
1431                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1432                                 OCFS2_XATTR_SIZE(xi->value_len);
1433                 void *val = xs->base + min_offs - size;
1434
1435                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1436                 memset(val, 0, size);
1437                 memcpy(val, xi->name, name_len);
1438                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1439                        xi->value,
1440                        xi->value_len);
1441                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1442                 ocfs2_xattr_set_local(xs->here, 1);
1443                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1444         }
1445
1446         return;
1447 }
1448
1449 /*
1450  * ocfs2_xattr_set_entry()
1451  *
1452  * Set extended attribute entry into inode or block.
1453  *
1454  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1455  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1456  * then set value in B tree with set_value_outside().
1457  */
1458 static int ocfs2_xattr_set_entry(struct inode *inode,
1459                                  struct ocfs2_xattr_info *xi,
1460                                  struct ocfs2_xattr_search *xs,
1461                                  struct ocfs2_xattr_set_ctxt *ctxt,
1462                                  int flag)
1463 {
1464         struct ocfs2_xattr_entry *last;
1465         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1466         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1467         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1468         size_t size_l = 0;
1469         handle_t *handle = ctxt->handle;
1470         int free, i, ret;
1471         struct ocfs2_xattr_info xi_l = {
1472                 .name_index = xi->name_index,
1473                 .name = xi->name,
1474                 .value = xi->value,
1475                 .value_len = xi->value_len,
1476         };
1477         struct ocfs2_xattr_value_buf vb = {
1478                 .vb_bh = xs->xattr_bh,
1479                 .vb_access = ocfs2_journal_access_di,
1480         };
1481
1482         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1483                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1484                 vb.vb_access = ocfs2_journal_access_xb;
1485         } else
1486                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1487
1488         /* Compute min_offs, last and free space. */
1489         last = xs->header->xh_entries;
1490
1491         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1492                 size_t offs = le16_to_cpu(last->xe_name_offset);
1493                 if (offs < min_offs)
1494                         min_offs = offs;
1495                 last += 1;
1496         }
1497
1498         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1499         if (free < 0)
1500                 return -EIO;
1501
1502         if (!xs->not_found) {
1503                 size_t size = 0;
1504                 if (ocfs2_xattr_is_local(xs->here))
1505                         size = OCFS2_XATTR_SIZE(name_len) +
1506                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1507                 else
1508                         size = OCFS2_XATTR_SIZE(name_len) +
1509                                 OCFS2_XATTR_ROOT_SIZE;
1510                 free += (size + sizeof(struct ocfs2_xattr_entry));
1511         }
1512         /* Check free space in inode or block */
1513         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1514                 if (free < sizeof(struct ocfs2_xattr_entry) +
1515                            OCFS2_XATTR_SIZE(name_len) +
1516                            OCFS2_XATTR_ROOT_SIZE) {
1517                         ret = -ENOSPC;
1518                         goto out;
1519                 }
1520                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1521                 xi_l.value = (void *)&def_xv;
1522                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1523         } else if (xi->value) {
1524                 if (free < sizeof(struct ocfs2_xattr_entry) +
1525                            OCFS2_XATTR_SIZE(name_len) +
1526                            OCFS2_XATTR_SIZE(xi->value_len)) {
1527                         ret = -ENOSPC;
1528                         goto out;
1529                 }
1530         }
1531
1532         if (!xs->not_found) {
1533                 /* For existing extended attribute */
1534                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1535                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1536                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1537                 void *val = xs->base + offs;
1538
1539                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1540                         /* Replace existing local xattr with tree root */
1541                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1542                                                             ctxt, &vb, offs);
1543                         if (ret < 0)
1544                                 mlog_errno(ret);
1545                         goto out;
1546                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1547                         /* For existing xattr which has value outside */
1548                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1549                                 (val + OCFS2_XATTR_SIZE(name_len));
1550
1551                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1552                                 /*
1553                                  * If new value need set outside also,
1554                                  * first truncate old value to new value,
1555                                  * then set new value with set_value_outside().
1556                                  */
1557                                 ret = ocfs2_xattr_value_truncate(inode,
1558                                                                  &vb,
1559                                                                  xi->value_len,
1560                                                                  ctxt);
1561                                 if (ret < 0) {
1562                                         mlog_errno(ret);
1563                                         goto out;
1564                                 }
1565
1566                                 ret = ocfs2_xattr_update_entry(inode,
1567                                                                handle,
1568                                                                xi,
1569                                                                xs,
1570                                                                &vb,
1571                                                                offs);
1572                                 if (ret < 0) {
1573                                         mlog_errno(ret);
1574                                         goto out;
1575                                 }
1576
1577                                 ret = __ocfs2_xattr_set_value_outside(inode,
1578                                                                 handle,
1579                                                                 vb.vb_xv,
1580                                                                 xi->value,
1581                                                                 xi->value_len);
1582                                 if (ret < 0)
1583                                         mlog_errno(ret);
1584                                 goto out;
1585                         } else {
1586                                 /*
1587                                  * If new value need set in local,
1588                                  * just trucate old value to zero.
1589                                  */
1590                                  ret = ocfs2_xattr_value_truncate(inode,
1591                                                                   &vb,
1592                                                                   0,
1593                                                                   ctxt);
1594                                 if (ret < 0)
1595                                         mlog_errno(ret);
1596                         }
1597                 }
1598         }
1599
1600         ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
1601                                       OCFS2_JOURNAL_ACCESS_WRITE);
1602         if (ret) {
1603                 mlog_errno(ret);
1604                 goto out;
1605         }
1606
1607         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1608                 ret = vb.vb_access(handle, inode, vb.vb_bh,
1609                                    OCFS2_JOURNAL_ACCESS_WRITE);
1610                 if (ret) {
1611                         mlog_errno(ret);
1612                         goto out;
1613                 }
1614         }
1615
1616         /*
1617          * Set value in local, include set tree root in local.
1618          * This is the first step for value size >INLINE_SIZE.
1619          */
1620         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1621
1622         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1623                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1624                 if (ret < 0) {
1625                         mlog_errno(ret);
1626                         goto out;
1627                 }
1628         }
1629
1630         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1631             (flag & OCFS2_INLINE_XATTR_FL)) {
1632                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1633                 unsigned int xattrsize = osb->s_xattr_inline_size;
1634
1635                 /*
1636                  * Adjust extent record count or inline data size
1637                  * to reserve space for extended attribute.
1638                  */
1639                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1640                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1641                         le16_add_cpu(&idata->id_count, -xattrsize);
1642                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1643                         struct ocfs2_extent_list *el = &di->id2.i_list;
1644                         le16_add_cpu(&el->l_count, -(xattrsize /
1645                                         sizeof(struct ocfs2_extent_rec)));
1646                 }
1647                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1648         }
1649         /* Update xattr flag */
1650         spin_lock(&oi->ip_lock);
1651         oi->ip_dyn_features |= flag;
1652         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1653         spin_unlock(&oi->ip_lock);
1654
1655         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1656         if (ret < 0)
1657                 mlog_errno(ret);
1658
1659         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1660                 /*
1661                  * Set value outside in B tree.
1662                  * This is the second step for value size > INLINE_SIZE.
1663                  */
1664                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1665                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1666                                                     &vb, offs);
1667                 if (ret < 0) {
1668                         int ret2;
1669
1670                         mlog_errno(ret);
1671                         /*
1672                          * If set value outside failed, we have to clean
1673                          * the junk tree root we have already set in local.
1674                          */
1675                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1676                                                    xi, xs, &vb, offs);
1677                         if (ret2 < 0)
1678                                 mlog_errno(ret2);
1679                 }
1680         }
1681 out:
1682         return ret;
1683 }
1684
1685 static int ocfs2_remove_value_outside(struct inode*inode,
1686                                       struct ocfs2_xattr_value_buf *vb,
1687                                       struct ocfs2_xattr_header *header)
1688 {
1689         int ret = 0, i;
1690         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1691         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1692
1693         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1694
1695         ctxt.handle = ocfs2_start_trans(osb,
1696                                         ocfs2_remove_extent_credits(osb->sb));
1697         if (IS_ERR(ctxt.handle)) {
1698                 ret = PTR_ERR(ctxt.handle);
1699                 mlog_errno(ret);
1700                 goto out;
1701         }
1702
1703         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1704                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1705
1706                 if (!ocfs2_xattr_is_local(entry)) {
1707                         void *val;
1708
1709                         val = (void *)header +
1710                                 le16_to_cpu(entry->xe_name_offset);
1711                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1712                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1713                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1714                         if (ret < 0) {
1715                                 mlog_errno(ret);
1716                                 break;
1717                         }
1718                 }
1719         }
1720
1721         ocfs2_commit_trans(osb, ctxt.handle);
1722         ocfs2_schedule_truncate_log_flush(osb, 1);
1723         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1724 out:
1725         return ret;
1726 }
1727
1728 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1729                                     struct buffer_head *di_bh)
1730 {
1731
1732         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1733         struct ocfs2_xattr_header *header;
1734         int ret;
1735         struct ocfs2_xattr_value_buf vb = {
1736                 .vb_bh = di_bh,
1737                 .vb_access = ocfs2_journal_access_di,
1738         };
1739
1740         header = (struct ocfs2_xattr_header *)
1741                  ((void *)di + inode->i_sb->s_blocksize -
1742                  le16_to_cpu(di->i_xattr_inline_size));
1743
1744         ret = ocfs2_remove_value_outside(inode, &vb, header);
1745
1746         return ret;
1747 }
1748
1749 static int ocfs2_xattr_block_remove(struct inode *inode,
1750                                     struct buffer_head *blk_bh)
1751 {
1752         struct ocfs2_xattr_block *xb;
1753         int ret = 0;
1754         struct ocfs2_xattr_value_buf vb = {
1755                 .vb_bh = blk_bh,
1756                 .vb_access = ocfs2_journal_access_xb,
1757         };
1758
1759         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1760         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1761                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1762                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1763         } else
1764                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1765
1766         return ret;
1767 }
1768
1769 static int ocfs2_xattr_free_block(struct inode *inode,
1770                                   u64 block)
1771 {
1772         struct inode *xb_alloc_inode;
1773         struct buffer_head *xb_alloc_bh = NULL;
1774         struct buffer_head *blk_bh = NULL;
1775         struct ocfs2_xattr_block *xb;
1776         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1777         handle_t *handle;
1778         int ret = 0;
1779         u64 blk, bg_blkno;
1780         u16 bit;
1781
1782         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1783         if (ret < 0) {
1784                 mlog_errno(ret);
1785                 goto out;
1786         }
1787
1788         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1789         if (ret < 0) {
1790                 mlog_errno(ret);
1791                 goto out;
1792         }
1793
1794         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1795         blk = le64_to_cpu(xb->xb_blkno);
1796         bit = le16_to_cpu(xb->xb_suballoc_bit);
1797         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1798
1799         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1800                                 EXTENT_ALLOC_SYSTEM_INODE,
1801                                 le16_to_cpu(xb->xb_suballoc_slot));
1802         if (!xb_alloc_inode) {
1803                 ret = -ENOMEM;
1804                 mlog_errno(ret);
1805                 goto out;
1806         }
1807         mutex_lock(&xb_alloc_inode->i_mutex);
1808
1809         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1810         if (ret < 0) {
1811                 mlog_errno(ret);
1812                 goto out_mutex;
1813         }
1814
1815         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1816         if (IS_ERR(handle)) {
1817                 ret = PTR_ERR(handle);
1818                 mlog_errno(ret);
1819                 goto out_unlock;
1820         }
1821
1822         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1823                                        bit, bg_blkno, 1);
1824         if (ret < 0)
1825                 mlog_errno(ret);
1826
1827         ocfs2_commit_trans(osb, handle);
1828 out_unlock:
1829         ocfs2_inode_unlock(xb_alloc_inode, 1);
1830         brelse(xb_alloc_bh);
1831 out_mutex:
1832         mutex_unlock(&xb_alloc_inode->i_mutex);
1833         iput(xb_alloc_inode);
1834 out:
1835         brelse(blk_bh);
1836         return ret;
1837 }
1838
1839 /*
1840  * ocfs2_xattr_remove()
1841  *
1842  * Free extended attribute resources associated with this inode.
1843  */
1844 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1845 {
1846         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1847         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1848         handle_t *handle;
1849         int ret;
1850
1851         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1852                 return 0;
1853
1854         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1855                 return 0;
1856
1857         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1858                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1859                 if (ret < 0) {
1860                         mlog_errno(ret);
1861                         goto out;
1862                 }
1863         }
1864
1865         if (di->i_xattr_loc) {
1866                 ret = ocfs2_xattr_free_block(inode,
1867                                              le64_to_cpu(di->i_xattr_loc));
1868                 if (ret < 0) {
1869                         mlog_errno(ret);
1870                         goto out;
1871                 }
1872         }
1873
1874         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1875                                    OCFS2_INODE_UPDATE_CREDITS);
1876         if (IS_ERR(handle)) {
1877                 ret = PTR_ERR(handle);
1878                 mlog_errno(ret);
1879                 goto out;
1880         }
1881         ret = ocfs2_journal_access_di(handle, inode, di_bh,
1882                                       OCFS2_JOURNAL_ACCESS_WRITE);
1883         if (ret) {
1884                 mlog_errno(ret);
1885                 goto out_commit;
1886         }
1887
1888         di->i_xattr_loc = 0;
1889
1890         spin_lock(&oi->ip_lock);
1891         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1892         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1893         spin_unlock(&oi->ip_lock);
1894
1895         ret = ocfs2_journal_dirty(handle, di_bh);
1896         if (ret < 0)
1897                 mlog_errno(ret);
1898 out_commit:
1899         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1900 out:
1901         return ret;
1902 }
1903
1904 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1905                                         struct ocfs2_dinode *di)
1906 {
1907         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1908         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1909         int free;
1910
1911         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1912                 return 0;
1913
1914         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1915                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1916                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1917         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1918                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1919                         le64_to_cpu(di->i_size);
1920         } else {
1921                 struct ocfs2_extent_list *el = &di->id2.i_list;
1922                 free = (le16_to_cpu(el->l_count) -
1923                         le16_to_cpu(el->l_next_free_rec)) *
1924                         sizeof(struct ocfs2_extent_rec);
1925         }
1926         if (free >= xattrsize)
1927                 return 1;
1928
1929         return 0;
1930 }
1931
1932 /*
1933  * ocfs2_xattr_ibody_find()
1934  *
1935  * Find extended attribute in inode block and
1936  * fill search info into struct ocfs2_xattr_search.
1937  */
1938 static int ocfs2_xattr_ibody_find(struct inode *inode,
1939                                   int name_index,
1940                                   const char *name,
1941                                   struct ocfs2_xattr_search *xs)
1942 {
1943         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1944         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1945         int ret;
1946         int has_space = 0;
1947
1948         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1949                 return 0;
1950
1951         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1952                 down_read(&oi->ip_alloc_sem);
1953                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1954                 up_read(&oi->ip_alloc_sem);
1955                 if (!has_space)
1956                         return 0;
1957         }
1958
1959         xs->xattr_bh = xs->inode_bh;
1960         xs->end = (void *)di + inode->i_sb->s_blocksize;
1961         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1962                 xs->header = (struct ocfs2_xattr_header *)
1963                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1964         else
1965                 xs->header = (struct ocfs2_xattr_header *)
1966                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1967         xs->base = (void *)xs->header;
1968         xs->here = xs->header->xh_entries;
1969
1970         /* Find the named attribute. */
1971         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1972                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1973                 if (ret && ret != -ENODATA)
1974                         return ret;
1975                 xs->not_found = ret;
1976         }
1977
1978         return 0;
1979 }
1980
1981 /*
1982  * ocfs2_xattr_ibody_set()
1983  *
1984  * Set, replace or remove an extended attribute into inode block.
1985  *
1986  */
1987 static int ocfs2_xattr_ibody_set(struct inode *inode,
1988                                  struct ocfs2_xattr_info *xi,
1989                                  struct ocfs2_xattr_search *xs,
1990                                  struct ocfs2_xattr_set_ctxt *ctxt)
1991 {
1992         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1993         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1994         int ret;
1995
1996         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1997                 return -ENOSPC;
1998
1999         down_write(&oi->ip_alloc_sem);
2000         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2001                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2002                         ret = -ENOSPC;
2003                         goto out;
2004                 }
2005         }
2006
2007         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2008                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2009 out:
2010         up_write(&oi->ip_alloc_sem);
2011
2012         return ret;
2013 }
2014
2015 /*
2016  * ocfs2_xattr_block_find()
2017  *
2018  * Find extended attribute in external block and
2019  * fill search info into struct ocfs2_xattr_search.
2020  */
2021 static int ocfs2_xattr_block_find(struct inode *inode,
2022                                   int name_index,
2023                                   const char *name,
2024                                   struct ocfs2_xattr_search *xs)
2025 {
2026         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2027         struct buffer_head *blk_bh = NULL;
2028         struct ocfs2_xattr_block *xb;
2029         int ret = 0;
2030
2031         if (!di->i_xattr_loc)
2032                 return ret;
2033
2034         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2035                                      &blk_bh);
2036         if (ret < 0) {
2037                 mlog_errno(ret);
2038                 return ret;
2039         }
2040
2041         xs->xattr_bh = blk_bh;
2042         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2043
2044         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2045                 xs->header = &xb->xb_attrs.xb_header;
2046                 xs->base = (void *)xs->header;
2047                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2048                 xs->here = xs->header->xh_entries;
2049
2050                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2051         } else
2052                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2053                                                    name_index,
2054                                                    name, xs);
2055
2056         if (ret && ret != -ENODATA) {
2057                 xs->xattr_bh = NULL;
2058                 goto cleanup;
2059         }
2060         xs->not_found = ret;
2061         return 0;
2062 cleanup:
2063         brelse(blk_bh);
2064
2065         return ret;
2066 }
2067
2068 /*
2069  * ocfs2_xattr_block_set()
2070  *
2071  * Set, replace or remove an extended attribute into external block.
2072  *
2073  */
2074 static int ocfs2_xattr_block_set(struct inode *inode,
2075                                  struct ocfs2_xattr_info *xi,
2076                                  struct ocfs2_xattr_search *xs,
2077                                  struct ocfs2_xattr_set_ctxt *ctxt)
2078 {
2079         struct buffer_head *new_bh = NULL;
2080         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2081         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
2082         handle_t *handle = ctxt->handle;
2083         struct ocfs2_xattr_block *xblk = NULL;
2084         u16 suballoc_bit_start;
2085         u32 num_got;
2086         u64 first_blkno;
2087         int ret;
2088
2089         if (!xs->xattr_bh) {
2090                 ret = ocfs2_journal_access_di(handle, inode, xs->inode_bh,
2091                                               OCFS2_JOURNAL_ACCESS_CREATE);
2092                 if (ret < 0) {
2093                         mlog_errno(ret);
2094                         goto end;
2095                 }
2096
2097                 ret = ocfs2_claim_metadata(osb, handle, ctxt->meta_ac, 1,
2098                                            &suballoc_bit_start, &num_got,
2099                                            &first_blkno);
2100                 if (ret < 0) {
2101                         mlog_errno(ret);
2102                         goto end;
2103                 }
2104
2105                 new_bh = sb_getblk(inode->i_sb, first_blkno);
2106                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
2107
2108                 ret = ocfs2_journal_access_xb(handle, inode, new_bh,
2109                                               OCFS2_JOURNAL_ACCESS_CREATE);
2110                 if (ret < 0) {
2111                         mlog_errno(ret);
2112                         goto end;
2113                 }
2114
2115                 /* Initialize ocfs2_xattr_block */
2116                 xs->xattr_bh = new_bh;
2117                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2118                 memset(xblk, 0, inode->i_sb->s_blocksize);
2119                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2120                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2121                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2122                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2123                 xblk->xb_blkno = cpu_to_le64(first_blkno);
2124
2125                 xs->header = &xblk->xb_attrs.xb_header;
2126                 xs->base = (void *)xs->header;
2127                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2128                 xs->here = xs->header->xh_entries;
2129
2130                 ret = ocfs2_journal_dirty(handle, new_bh);
2131                 if (ret < 0) {
2132                         mlog_errno(ret);
2133                         goto end;
2134                 }
2135                 di->i_xattr_loc = cpu_to_le64(first_blkno);
2136                 ocfs2_journal_dirty(handle, xs->inode_bh);
2137         } else
2138                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2139
2140         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2141                 /* Set extended attribute into external block */
2142                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2143                                             OCFS2_HAS_XATTR_FL);
2144                 if (!ret || ret != -ENOSPC)
2145                         goto end;
2146
2147                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2148                 if (ret)
2149                         goto end;
2150         }
2151
2152         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2153
2154 end:
2155
2156         return ret;
2157 }
2158
2159 /* Check whether the new xattr can be inserted into the inode. */
2160 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2161                                        struct ocfs2_xattr_info *xi,
2162                                        struct ocfs2_xattr_search *xs)
2163 {
2164         u64 value_size;
2165         struct ocfs2_xattr_entry *last;
2166         int free, i;
2167         size_t min_offs = xs->end - xs->base;
2168
2169         if (!xs->header)
2170                 return 0;
2171
2172         last = xs->header->xh_entries;
2173
2174         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2175                 size_t offs = le16_to_cpu(last->xe_name_offset);
2176                 if (offs < min_offs)
2177                         min_offs = offs;
2178                 last += 1;
2179         }
2180
2181         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
2182         if (free < 0)
2183                 return 0;
2184
2185         BUG_ON(!xs->not_found);
2186
2187         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2188                 value_size = OCFS2_XATTR_ROOT_SIZE;
2189         else
2190                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2191
2192         if (free >= sizeof(struct ocfs2_xattr_entry) +
2193                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2194                 return 1;
2195
2196         return 0;
2197 }
2198
2199 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2200                                      struct ocfs2_dinode *di,
2201                                      struct ocfs2_xattr_info *xi,
2202                                      struct ocfs2_xattr_search *xis,
2203                                      struct ocfs2_xattr_search *xbs,
2204                                      int *clusters_need,
2205                                      int *meta_need,
2206                                      int *credits_need)
2207 {
2208         int ret = 0, old_in_xb = 0;
2209         int clusters_add = 0, meta_add = 0, credits = 0;
2210         struct buffer_head *bh = NULL;
2211         struct ocfs2_xattr_block *xb = NULL;
2212         struct ocfs2_xattr_entry *xe = NULL;
2213         struct ocfs2_xattr_value_root *xv = NULL;
2214         char *base = NULL;
2215         int name_offset, name_len = 0;
2216         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2217                                                     xi->value_len);
2218         u64 value_size;
2219
2220         /*
2221          * Calculate the clusters we need to write.
2222          * No matter whether we replace an old one or add a new one,
2223          * we need this for writing.
2224          */
2225         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2226                 credits += new_clusters *
2227                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2228
2229         if (xis->not_found && xbs->not_found) {
2230                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2231
2232                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2233                         clusters_add += new_clusters;
2234                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2235                                                         &def_xv.xv.xr_list,
2236                                                         new_clusters);
2237                 }
2238
2239                 goto meta_guess;
2240         }
2241
2242         if (!xis->not_found) {
2243                 xe = xis->here;
2244                 name_offset = le16_to_cpu(xe->xe_name_offset);
2245                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2246                 base = xis->base;
2247                 credits += OCFS2_INODE_UPDATE_CREDITS;
2248         } else {
2249                 int i, block_off = 0;
2250                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2251                 xe = xbs->here;
2252                 name_offset = le16_to_cpu(xe->xe_name_offset);
2253                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2254                 i = xbs->here - xbs->header->xh_entries;
2255                 old_in_xb = 1;
2256
2257                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2258                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2259                                                         bucket_xh(xbs->bucket),
2260                                                         i, &block_off,
2261                                                         &name_offset);
2262                         base = bucket_block(xbs->bucket, block_off);
2263                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2264                 } else {
2265                         base = xbs->base;
2266                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2267                 }
2268         }
2269
2270         /*
2271          * delete a xattr doesn't need metadata and cluster allocation.
2272          * so just calculate the credits and return.
2273          *
2274          * The credits for removing the value tree will be extended
2275          * by ocfs2_remove_extent itself.
2276          */
2277         if (!xi->value) {
2278                 if (!ocfs2_xattr_is_local(xe))
2279                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2280
2281                 goto out;
2282         }
2283
2284         /* do cluster allocation guess first. */
2285         value_size = le64_to_cpu(xe->xe_value_size);
2286
2287         if (old_in_xb) {
2288                 /*
2289                  * In xattr set, we always try to set the xe in inode first,
2290                  * so if it can be inserted into inode successfully, the old
2291                  * one will be removed from the xattr block, and this xattr
2292                  * will be inserted into inode as a new xattr in inode.
2293                  */
2294                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2295                         clusters_add += new_clusters;
2296                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2297                                     OCFS2_INODE_UPDATE_CREDITS;
2298                         if (!ocfs2_xattr_is_local(xe))
2299                                 credits += ocfs2_calc_extend_credits(
2300                                                         inode->i_sb,
2301                                                         &def_xv.xv.xr_list,
2302                                                         new_clusters);
2303                         goto out;
2304                 }
2305         }
2306
2307         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2308                 /* the new values will be stored outside. */
2309                 u32 old_clusters = 0;
2310
2311                 if (!ocfs2_xattr_is_local(xe)) {
2312                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2313                                                                  value_size);
2314                         xv = (struct ocfs2_xattr_value_root *)
2315                              (base + name_offset + name_len);
2316                         value_size = OCFS2_XATTR_ROOT_SIZE;
2317                 } else
2318                         xv = &def_xv.xv;
2319
2320                 if (old_clusters >= new_clusters) {
2321                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2322                         goto out;
2323                 } else {
2324                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2325                         clusters_add += new_clusters - old_clusters;
2326                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2327                                                              &xv->xr_list,
2328                                                              new_clusters -
2329                                                              old_clusters);
2330                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2331                                 goto out;
2332                 }
2333         } else {
2334                 /*
2335                  * Now the new value will be stored inside. So if the new
2336                  * value is smaller than the size of value root or the old
2337                  * value, we don't need any allocation, otherwise we have
2338                  * to guess metadata allocation.
2339                  */
2340                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2341                     (!ocfs2_xattr_is_local(xe) &&
2342                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2343                         goto out;
2344         }
2345
2346 meta_guess:
2347         /* calculate metadata allocation. */
2348         if (di->i_xattr_loc) {
2349                 if (!xbs->xattr_bh) {
2350                         ret = ocfs2_read_xattr_block(inode,
2351                                                      le64_to_cpu(di->i_xattr_loc),
2352                                                      &bh);
2353                         if (ret) {
2354                                 mlog_errno(ret);
2355                                 goto out;
2356                         }
2357
2358                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2359                 } else
2360                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2361
2362                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2363                         struct ocfs2_extent_list *el =
2364                                  &xb->xb_attrs.xb_root.xt_list;
2365                         meta_add += ocfs2_extend_meta_needed(el);
2366                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2367                                                              el, 1);
2368                 }
2369
2370                 /*
2371                  * This cluster will be used either for new bucket or for
2372                  * new xattr block.
2373                  * If the cluster size is the same as the bucket size, one
2374                  * more is needed since we may need to extend the bucket
2375                  * also.
2376                  */
2377                 clusters_add += 1;
2378                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2379                 if (OCFS2_XATTR_BUCKET_SIZE ==
2380                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2381                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2382                         clusters_add += 1;
2383                 }
2384         } else {
2385                 meta_add += 1;
2386                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2387         }
2388 out:
2389         if (clusters_need)
2390                 *clusters_need = clusters_add;
2391         if (meta_need)
2392                 *meta_need = meta_add;
2393         if (credits_need)
2394                 *credits_need = credits;
2395         brelse(bh);
2396         return ret;
2397 }
2398
2399 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2400                                      struct ocfs2_dinode *di,
2401                                      struct ocfs2_xattr_info *xi,
2402                                      struct ocfs2_xattr_search *xis,
2403                                      struct ocfs2_xattr_search *xbs,
2404                                      struct ocfs2_xattr_set_ctxt *ctxt,
2405                                      int *credits)
2406 {
2407         int clusters_add, meta_add, ret;
2408         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2409
2410         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2411
2412         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2413
2414         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2415                                         &clusters_add, &meta_add, credits);
2416         if (ret) {
2417                 mlog_errno(ret);
2418                 return ret;
2419         }
2420
2421         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2422              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2423
2424         if (meta_add) {
2425                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2426                                                         &ctxt->meta_ac);
2427                 if (ret) {
2428                         mlog_errno(ret);
2429                         goto out;
2430                 }
2431         }
2432
2433         if (clusters_add) {
2434                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2435                 if (ret)
2436                         mlog_errno(ret);
2437         }
2438 out:
2439         if (ret) {
2440                 if (ctxt->meta_ac) {
2441                         ocfs2_free_alloc_context(ctxt->meta_ac);
2442                         ctxt->meta_ac = NULL;
2443                 }
2444
2445                 /*
2446                  * We cannot have an error and a non null ctxt->data_ac.
2447                  */
2448         }
2449
2450         return ret;
2451 }
2452
2453 static int __ocfs2_xattr_set_handle(struct inode *inode,
2454                                     struct ocfs2_dinode *di,
2455                                     struct ocfs2_xattr_info *xi,
2456                                     struct ocfs2_xattr_search *xis,
2457                                     struct ocfs2_xattr_search *xbs,
2458                                     struct ocfs2_xattr_set_ctxt *ctxt)
2459 {
2460         int ret = 0, credits, old_found;
2461
2462         if (!xi->value) {
2463                 /* Remove existing extended attribute */
2464                 if (!xis->not_found)
2465                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2466                 else if (!xbs->not_found)
2467                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2468         } else {
2469                 /* We always try to set extended attribute into inode first*/
2470                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2471                 if (!ret && !xbs->not_found) {
2472                         /*
2473                          * If succeed and that extended attribute existing in
2474                          * external block, then we will remove it.
2475                          */
2476                         xi->value = NULL;
2477                         xi->value_len = 0;
2478
2479                         old_found = xis->not_found;
2480                         xis->not_found = -ENODATA;
2481                         ret = ocfs2_calc_xattr_set_need(inode,
2482                                                         di,
2483                                                         xi,
2484                                                         xis,
2485                                                         xbs,
2486                                                         NULL,
2487                                                         NULL,
2488                                                         &credits);
2489                         xis->not_found = old_found;
2490                         if (ret) {
2491                                 mlog_errno(ret);
2492                                 goto out;
2493                         }
2494
2495                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2496                                         ctxt->handle->h_buffer_credits);
2497                         if (ret) {
2498                                 mlog_errno(ret);
2499                                 goto out;
2500                         }
2501                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2502                 } else if (ret == -ENOSPC) {
2503                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2504                                 ret = ocfs2_xattr_block_find(inode,
2505                                                              xi->name_index,
2506                                                              xi->name, xbs);
2507                                 if (ret)
2508                                         goto out;
2509
2510                                 old_found = xis->not_found;
2511                                 xis->not_found = -ENODATA;
2512                                 ret = ocfs2_calc_xattr_set_need(inode,
2513                                                                 di,
2514                                                                 xi,
2515                                                                 xis,
2516                                                                 xbs,
2517                                                                 NULL,
2518                                                                 NULL,
2519                                                                 &credits);
2520                                 xis->not_found = old_found;
2521                                 if (ret) {
2522                                         mlog_errno(ret);
2523                                         goto out;
2524                                 }
2525
2526                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2527                                         ctxt->handle->h_buffer_credits);
2528                                 if (ret) {
2529                                         mlog_errno(ret);
2530                                         goto out;
2531                                 }
2532                         }
2533                         /*
2534                          * If no space in inode, we will set extended attribute
2535                          * into external block.
2536                          */
2537                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2538                         if (ret)
2539                                 goto out;
2540                         if (!xis->not_found) {
2541                                 /*
2542                                  * If succeed and that extended attribute
2543                                  * existing in inode, we will remove it.
2544                                  */
2545                                 xi->value = NULL;
2546                                 xi->value_len = 0;
2547                                 xbs->not_found = -ENODATA;
2548                                 ret = ocfs2_calc_xattr_set_need(inode,
2549                                                                 di,
2550                                                                 xi,
2551                                                                 xis,
2552                                                                 xbs,
2553                                                                 NULL,
2554                                                                 NULL,
2555                                                                 &credits);
2556                                 if (ret) {
2557                                         mlog_errno(ret);
2558                                         goto out;
2559                                 }
2560
2561                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2562                                                 ctxt->handle->h_buffer_credits);
2563                                 if (ret) {
2564                                         mlog_errno(ret);
2565                                         goto out;
2566                                 }
2567                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2568                                                             xis, ctxt);
2569                         }
2570                 }
2571         }
2572
2573         if (!ret) {
2574                 /* Update inode ctime. */
2575                 ret = ocfs2_journal_access(ctxt->handle, inode, xis->inode_bh,
2576                                            OCFS2_JOURNAL_ACCESS_WRITE);
2577                 if (ret) {
2578                         mlog_errno(ret);
2579                         goto out;
2580                 }
2581
2582                 inode->i_ctime = CURRENT_TIME;
2583                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2584                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2585                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2586         }
2587 out:
2588         return ret;
2589 }
2590
2591 /*
2592  * This function only called duing creating inode
2593  * for init security/acl xattrs of the new inode.
2594  * The xattrs could be put into ibody or extent block,
2595  * xattr bucket would not be use in this case.
2596  * transanction credits also be reserved in here.
2597  */
2598 int ocfs2_xattr_set_handle(handle_t *handle,
2599                            struct inode *inode,
2600                            struct buffer_head *di_bh,
2601                            int name_index,
2602                            const char *name,
2603                            const void *value,
2604                            size_t value_len,
2605                            int flags,
2606                            struct ocfs2_alloc_context *meta_ac,
2607                            struct ocfs2_alloc_context *data_ac)
2608 {
2609         struct ocfs2_dinode *di;
2610         int ret;
2611
2612         struct ocfs2_xattr_info xi = {
2613                 .name_index = name_index,
2614                 .name = name,
2615                 .value = value,
2616                 .value_len = value_len,
2617         };
2618
2619         struct ocfs2_xattr_search xis = {
2620                 .not_found = -ENODATA,
2621         };
2622
2623         struct ocfs2_xattr_search xbs = {
2624                 .not_found = -ENODATA,
2625         };
2626
2627         struct ocfs2_xattr_set_ctxt ctxt = {
2628                 .handle = handle,
2629                 .meta_ac = meta_ac,
2630                 .data_ac = data_ac,
2631         };
2632
2633         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2634                 return -EOPNOTSUPP;
2635
2636         xis.inode_bh = xbs.inode_bh = di_bh;
2637         di = (struct ocfs2_dinode *)di_bh->b_data;
2638
2639         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2640
2641         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2642         if (ret)
2643                 goto cleanup;
2644         if (xis.not_found) {
2645                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2646                 if (ret)
2647                         goto cleanup;
2648         }
2649
2650         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2651
2652 cleanup:
2653         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2654         brelse(xbs.xattr_bh);
2655
2656         return ret;
2657 }
2658
2659 /*
2660  * ocfs2_xattr_set()
2661  *
2662  * Set, replace or remove an extended attribute for this inode.
2663  * value is NULL to remove an existing extended attribute, else either
2664  * create or replace an extended attribute.
2665  */
2666 int ocfs2_xattr_set(struct inode *inode,
2667                     int name_index,
2668                     const char *name,
2669                     const void *value,
2670                     size_t value_len,
2671                     int flags)
2672 {
2673         struct buffer_head *di_bh = NULL;
2674         struct ocfs2_dinode *di;
2675         int ret, credits;
2676         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2677         struct inode *tl_inode = osb->osb_tl_inode;
2678         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2679
2680         struct ocfs2_xattr_info xi = {
2681                 .name_index = name_index,
2682                 .name = name,
2683                 .value = value,
2684                 .value_len = value_len,
2685         };
2686
2687         struct ocfs2_xattr_search xis = {
2688                 .not_found = -ENODATA,
2689         };
2690
2691         struct ocfs2_xattr_search xbs = {
2692                 .not_found = -ENODATA,
2693         };
2694
2695         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2696                 return -EOPNOTSUPP;
2697
2698         /*
2699          * Only xbs will be used on indexed trees.  xis doesn't need a
2700          * bucket.
2701          */
2702         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2703         if (!xbs.bucket) {
2704                 mlog_errno(-ENOMEM);
2705                 return -ENOMEM;
2706         }
2707
2708         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2709         if (ret < 0) {
2710                 mlog_errno(ret);
2711                 goto cleanup_nolock;
2712         }
2713         xis.inode_bh = xbs.inode_bh = di_bh;
2714         di = (struct ocfs2_dinode *)di_bh->b_data;
2715
2716         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2717         /*
2718          * Scan inode and external block to find the same name
2719          * extended attribute and collect search infomation.
2720          */
2721         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2722         if (ret)
2723                 goto cleanup;
2724         if (xis.not_found) {
2725                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2726                 if (ret)
2727                         goto cleanup;
2728         }
2729
2730         if (xis.not_found && xbs.not_found) {
2731                 ret = -ENODATA;
2732                 if (flags & XATTR_REPLACE)
2733                         goto cleanup;
2734                 ret = 0;
2735                 if (!value)
2736                         goto cleanup;
2737         } else {
2738                 ret = -EEXIST;
2739                 if (flags & XATTR_CREATE)
2740                         goto cleanup;
2741         }
2742
2743
2744         mutex_lock(&tl_inode->i_mutex);
2745
2746         if (ocfs2_truncate_log_needs_flush(osb)) {
2747                 ret = __ocfs2_flush_truncate_log(osb);
2748                 if (ret < 0) {
2749                         mutex_unlock(&tl_inode->i_mutex);
2750                         mlog_errno(ret);
2751                         goto cleanup;
2752                 }
2753         }
2754         mutex_unlock(&tl_inode->i_mutex);
2755
2756         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2757                                         &xbs, &ctxt, &credits);
2758         if (ret) {
2759                 mlog_errno(ret);
2760                 goto cleanup;
2761         }
2762
2763         /* we need to update inode's ctime field, so add credit for it. */
2764         credits += OCFS2_INODE_UPDATE_CREDITS;
2765         ctxt.handle = ocfs2_start_trans(osb, credits);
2766         if (IS_ERR(ctxt.handle)) {
2767                 ret = PTR_ERR(ctxt.handle);
2768                 mlog_errno(ret);
2769                 goto cleanup;
2770         }
2771
2772         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2773
2774         ocfs2_commit_trans(osb, ctxt.handle);
2775
2776         if (ctxt.data_ac)
2777                 ocfs2_free_alloc_context(ctxt.data_ac);
2778         if (ctxt.meta_ac)
2779                 ocfs2_free_alloc_context(ctxt.meta_ac);
2780         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2781                 ocfs2_schedule_truncate_log_flush(osb, 1);
2782         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2783 cleanup:
2784         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2785         ocfs2_inode_unlock(inode, 1);
2786 cleanup_nolock:
2787         brelse(di_bh);
2788         brelse(xbs.xattr_bh);
2789         ocfs2_xattr_bucket_free(xbs.bucket);
2790
2791         return ret;
2792 }
2793
2794 /*
2795  * Find the xattr extent rec which may contains name_hash.
2796  * e_cpos will be the first name hash of the xattr rec.
2797  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2798  */
2799 static int ocfs2_xattr_get_rec(struct inode *inode,
2800                                u32 name_hash,
2801                                u64 *p_blkno,
2802                                u32 *e_cpos,
2803                                u32 *num_clusters,
2804                                struct ocfs2_extent_list *el)
2805 {
2806         int ret = 0, i;
2807         struct buffer_head *eb_bh = NULL;
2808         struct ocfs2_extent_block *eb;
2809         struct ocfs2_extent_rec *rec = NULL;
2810         u64 e_blkno = 0;
2811
2812         if (el->l_tree_depth) {
2813                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2814                 if (ret) {
2815                         mlog_errno(ret);
2816                         goto out;
2817                 }
2818
2819                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2820                 el = &eb->h_list;
2821
2822                 if (el->l_tree_depth) {
2823                         ocfs2_error(inode->i_sb,
2824                                     "Inode %lu has non zero tree depth in "
2825                                     "xattr tree block %llu\n", inode->i_ino,
2826                                     (unsigned long long)eb_bh->b_blocknr);
2827                         ret = -EROFS;
2828                         goto out;
2829                 }
2830         }
2831
2832         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2833                 rec = &el->l_recs[i];
2834
2835                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2836                         e_blkno = le64_to_cpu(rec->e_blkno);
2837                         break;
2838                 }
2839         }
2840
2841         if (!e_blkno) {
2842                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2843                             "record (%u, %u, 0) in xattr", inode->i_ino,
2844                             le32_to_cpu(rec->e_cpos),
2845                             ocfs2_rec_clusters(el, rec));
2846                 ret = -EROFS;
2847                 goto out;
2848         }
2849
2850         *p_blkno = le64_to_cpu(rec->e_blkno);
2851         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2852         if (e_cpos)
2853                 *e_cpos = le32_to_cpu(rec->e_cpos);
2854 out:
2855         brelse(eb_bh);
2856         return ret;
2857 }
2858
2859 typedef int (xattr_bucket_func)(struct inode *inode,
2860                                 struct ocfs2_xattr_bucket *bucket,
2861                                 void *para);
2862
2863 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2864                                    struct ocfs2_xattr_bucket *bucket,
2865                                    int name_index,
2866                                    const char *name,
2867                                    u32 name_hash,
2868                                    u16 *xe_index,
2869                                    int *found)
2870 {
2871         int i, ret = 0, cmp = 1, block_off, new_offset;
2872         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2873         size_t name_len = strlen(name);
2874         struct ocfs2_xattr_entry *xe = NULL;
2875         char *xe_name;
2876
2877         /*
2878          * We don't use binary search in the bucket because there
2879          * may be multiple entries with the same name hash.
2880          */
2881         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2882                 xe = &xh->xh_entries[i];
2883
2884                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2885                         continue;
2886                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2887                         break;
2888
2889                 cmp = name_index - ocfs2_xattr_get_type(xe);
2890                 if (!cmp)
2891                         cmp = name_len - xe->xe_name_len;
2892                 if (cmp)
2893                         continue;
2894
2895                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2896                                                         xh,
2897                                                         i,
2898                                                         &block_off,
2899                                                         &new_offset);
2900                 if (ret) {
2901                         mlog_errno(ret);
2902                         break;
2903                 }
2904
2905
2906                 xe_name = bucket_block(bucket, block_off) + new_offset;
2907                 if (!memcmp(name, xe_name, name_len)) {
2908                         *xe_index = i;
2909                         *found = 1;
2910                         ret = 0;
2911                         break;
2912                 }
2913         }
2914
2915         return ret;
2916 }
2917
2918 /*
2919  * Find the specified xattr entry in a series of buckets.
2920  * This series start from p_blkno and last for num_clusters.
2921  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2922  * the num of the valid buckets.
2923  *
2924  * Return the buffer_head this xattr should reside in. And if the xattr's
2925  * hash is in the gap of 2 buckets, return the lower bucket.
2926  */
2927 static int ocfs2_xattr_bucket_find(struct inode *inode,
2928                                    int name_index,
2929                                    const char *name,
2930                                    u32 name_hash,
2931                                    u64 p_blkno,
2932                                    u32 first_hash,
2933                                    u32 num_clusters,
2934                                    struct ocfs2_xattr_search *xs)
2935 {
2936         int ret, found = 0;
2937         struct ocfs2_xattr_header *xh = NULL;
2938         struct ocfs2_xattr_entry *xe = NULL;
2939         u16 index = 0;
2940         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2941         int low_bucket = 0, bucket, high_bucket;
2942         struct ocfs2_xattr_bucket *search;
2943         u32 last_hash;
2944         u64 blkno, lower_blkno = 0;
2945
2946         search = ocfs2_xattr_bucket_new(inode);
2947         if (!search) {
2948                 ret = -ENOMEM;
2949                 mlog_errno(ret);
2950                 goto out;
2951         }
2952
2953         ret = ocfs2_read_xattr_bucket(search, p_blkno);
2954         if (ret) {
2955                 mlog_errno(ret);
2956                 goto out;
2957         }
2958
2959         xh = bucket_xh(search);
2960         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2961         while (low_bucket <= high_bucket) {
2962                 ocfs2_xattr_bucket_relse(search);
2963
2964                 bucket = (low_bucket + high_bucket) / 2;
2965                 blkno = p_blkno + bucket * blk_per_bucket;
2966                 ret = ocfs2_read_xattr_bucket(search, blkno);
2967                 if (ret) {
2968                         mlog_errno(ret);
2969                         goto out;
2970                 }
2971
2972                 xh = bucket_xh(search);
2973                 xe = &xh->xh_entries[0];
2974                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2975                         high_bucket = bucket - 1;
2976                         continue;
2977                 }
2978
2979                 /*
2980                  * Check whether the hash of the last entry in our
2981                  * bucket is larger than the search one. for an empty
2982                  * bucket, the last one is also the first one.
2983                  */
2984                 if (xh->xh_count)
2985                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2986
2987                 last_hash = le32_to_cpu(xe->xe_name_hash);
2988
2989                 /* record lower_blkno which may be the insert place. */
2990                 lower_blkno = blkno;
2991
2992                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2993                         low_bucket = bucket + 1;
2994                         continue;
2995                 }
2996
2997                 /* the searched xattr should reside in this bucket if exists. */
2998                 ret = ocfs2_find_xe_in_bucket(inode, search,
2999                                               name_index, name, name_hash,
3000                                               &index, &found);
3001                 if (ret) {
3002                         mlog_errno(ret);
3003                         goto out;
3004                 }
3005                 break;
3006         }
3007
3008         /*
3009          * Record the bucket we have found.
3010          * When the xattr's hash value is in the gap of 2 buckets, we will
3011          * always set it to the previous bucket.
3012          */
3013         if (!lower_blkno)
3014                 lower_blkno = p_blkno;
3015
3016         /* This should be in cache - we just read it during the search */
3017         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3018         if (ret) {
3019                 mlog_errno(ret);
3020                 goto out;
3021         }
3022
3023         xs->header = bucket_xh(xs->bucket);
3024         xs->base = bucket_block(xs->bucket, 0);
3025         xs->end = xs->base + inode->i_sb->s_blocksize;
3026
3027         if (found) {
3028                 xs->here = &xs->header->xh_entries[index];
3029                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3030                      (unsigned long long)bucket_blkno(xs->bucket), index);
3031         } else
3032                 ret = -ENODATA;
3033
3034 out:
3035         ocfs2_xattr_bucket_free(search);
3036         return ret;
3037 }
3038
3039 static int ocfs2_xattr_index_block_find(struct inode *inode,
3040                                         struct buffer_head *root_bh,
3041                                         int name_index,
3042                                         const char *name,
3043                                         struct ocfs2_xattr_search *xs)
3044 {
3045         int ret;
3046         struct ocfs2_xattr_block *xb =
3047                         (struct ocfs2_xattr_block *)root_bh->b_data;
3048         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3049         struct ocfs2_extent_list *el = &xb_root->xt_list;
3050         u64 p_blkno = 0;
3051         u32 first_hash, num_clusters = 0;
3052         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3053
3054         if (le16_to_cpu(el->l_next_free_rec) == 0)
3055                 return -ENODATA;
3056
3057         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3058              name, name_hash, name_index);
3059
3060         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3061                                   &num_clusters, el);
3062         if (ret) {
3063                 mlog_errno(ret);
3064                 goto out;
3065         }
3066
3067         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3068
3069         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3070              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3071              first_hash);
3072
3073         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3074                                       p_blkno, first_hash, num_clusters, xs);
3075
3076 out:
3077         return ret;
3078 }
3079
3080 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3081                                        u64 blkno,
3082                                        u32 clusters,
3083                                        xattr_bucket_func *func,
3084                                        void *para)
3085 {
3086         int i, ret = 0;
3087         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3088         u32 num_buckets = clusters * bpc;
3089         struct ocfs2_xattr_bucket *bucket;
3090
3091         bucket = ocfs2_xattr_bucket_new(inode);
3092         if (!bucket) {
3093                 mlog_errno(-ENOMEM);
3094                 return -ENOMEM;
3095         }
3096
3097         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3098              clusters, (unsigned long long)blkno);
3099
3100         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3101                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3102                 if (ret) {
3103                         mlog_errno(ret);
3104                         break;
3105                 }
3106
3107                 /*
3108                  * The real bucket num in this series of blocks is stored
3109                  * in the 1st bucket.
3110                  */
3111                 if (i == 0)
3112                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3113
3114                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3115                      (unsigned long long)blkno,
3116                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3117                 if (func) {
3118                         ret = func(inode, bucket, para);
3119                         if (ret)
3120                                 mlog_errno(ret);
3121                         /* Fall through to bucket_relse() */
3122                 }
3123
3124                 ocfs2_xattr_bucket_relse(bucket);
3125                 if (ret)
3126                         break;
3127         }
3128
3129         ocfs2_xattr_bucket_free(bucket);
3130         return ret;
3131 }
3132
3133 struct ocfs2_xattr_tree_list {
3134         char *buffer;
3135         size_t buffer_size;
3136         size_t result;
3137 };
3138
3139 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
3140                                              struct ocfs2_xattr_header *xh,
3141                                              int index,
3142                                              int *block_off,
3143                                              int *new_offset)
3144 {
3145         u16 name_offset;
3146
3147         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3148                 return -EINVAL;
3149
3150         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3151
3152         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
3153         *new_offset = name_offset % inode->i_sb->s_blocksize;
3154
3155         return 0;
3156 }
3157
3158 static int ocfs2_list_xattr_bucket(struct inode *inode,
3159                                    struct ocfs2_xattr_bucket *bucket,
3160                                    void *para)
3161 {
3162         int ret = 0, type;
3163         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3164         int i, block_off, new_offset;
3165         const char *prefix, *name;
3166
3167         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3168                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3169                 type = ocfs2_xattr_get_type(entry);
3170                 prefix = ocfs2_xattr_prefix(type);
3171
3172                 if (prefix) {
3173                         ret = ocfs2_xattr_bucket_get_name_value(inode,
3174                                                                 bucket_xh(bucket),
3175                                                                 i,
3176                                                                 &block_off,
3177                                                                 &new_offset);
3178                         if (ret)
3179                                 break;
3180
3181                         name = (const char *)bucket_block(bucket, block_off) +
3182                                 new_offset;
3183                         ret = ocfs2_xattr_list_entry(xl->buffer,
3184                                                      xl->buffer_size,
3185                                                      &xl->result,
3186                                                      prefix, name,
3187                                                      entry->xe_name_len);
3188                         if (ret)
3189                                 break;
3190                 }
3191         }
3192
3193         return ret;
3194 }
3195
3196 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3197                                              struct ocfs2_xattr_tree_root *xt,
3198                                              char *buffer,
3199                                              size_t buffer_size)
3200 {
3201         struct ocfs2_extent_list *el = &xt->xt_list;
3202         int ret = 0;
3203         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3204         u64 p_blkno = 0;
3205         struct ocfs2_xattr_tree_list xl = {
3206                 .buffer = buffer,
3207                 .buffer_size = buffer_size,
3208                 .result = 0,
3209         };
3210
3211         if (le16_to_cpu(el->l_next_free_rec) == 0)
3212                 return 0;
3213
3214         while (name_hash > 0) {
3215                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3216                                           &e_cpos, &num_clusters, el);
3217                 if (ret) {
3218                         mlog_errno(ret);
3219                         goto out;
3220                 }
3221
3222                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3223                                                   ocfs2_list_xattr_bucket,
3224                                                   &xl);
3225                 if (ret) {
3226                         mlog_errno(ret);
3227                         goto out;
3228                 }
3229
3230                 if (e_cpos == 0)
3231                         break;
3232
3233                 name_hash = e_cpos - 1;
3234         }
3235
3236         ret = xl.result;
3237 out:
3238         return ret;
3239 }
3240
3241 static int cmp_xe(const void *a, const void *b)
3242 {
3243         const struct ocfs2_xattr_entry *l = a, *r = b;
3244         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3245         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3246
3247         if (l_hash > r_hash)
3248                 return 1;
3249         if (l_hash < r_hash)
3250                 return -1;
3251         return 0;
3252 }
3253
3254 static void swap_xe(void *a, void *b, int size)
3255 {
3256         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3257
3258         tmp = *l;
3259         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3260         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3261 }
3262
3263 /*
3264  * When the ocfs2_xattr_block is filled up, new bucket will be created
3265  * and all the xattr entries will be moved to the new bucket.
3266  * The header goes at the start of the bucket, and the names+values are
3267  * filled from the end.  This is why *target starts as the last buffer.
3268  * Note: we need to sort the entries since they are not saved in order
3269  * in the ocfs2_xattr_block.
3270  */
3271 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3272                                            struct buffer_head *xb_bh,
3273                                            struct ocfs2_xattr_bucket *bucket)
3274 {
3275         int i, blocksize = inode->i_sb->s_blocksize;
3276         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3277         u16 offset, size, off_change;
3278         struct ocfs2_xattr_entry *xe;
3279         struct ocfs2_xattr_block *xb =
3280                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3281         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3282         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3283         u16 count = le16_to_cpu(xb_xh->xh_count);
3284         char *src = xb_bh->b_data;
3285         char *target = bucket_block(bucket, blks - 1);
3286
3287         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3288              (unsigned long long)xb_bh->b_blocknr,
3289              (unsigned long long)bucket_blkno(bucket));
3290
3291         for (i = 0; i < blks; i++)
3292                 memset(bucket_block(bucket, i), 0, blocksize);
3293
3294         /*
3295          * Since the xe_name_offset is based on ocfs2_xattr_header,
3296          * there is a offset change corresponding to the change of
3297          * ocfs2_xattr_header's position.
3298          */
3299         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3300         xe = &xb_xh->xh_entries[count - 1];
3301         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3302         size = blocksize - offset;
3303
3304         /* copy all the names and values. */
3305         memcpy(target + offset, src + offset, size);
3306
3307         /* Init new header now. */
3308         xh->xh_count = xb_xh->xh_count;
3309         xh->xh_num_buckets = cpu_to_le16(1);
3310         xh->xh_name_value_len = cpu_to_le16(size);
3311         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3312
3313         /* copy all the entries. */
3314         target = bucket_block(bucket, 0);
3315         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3316         size = count * sizeof(struct ocfs2_xattr_entry);
3317         memcpy(target + offset, (char *)xb_xh + offset, size);
3318
3319         /* Change the xe offset for all the xe because of the move. */
3320         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3321                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3322         for (i = 0; i < count; i++)
3323                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3324
3325         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3326              offset, size, off_change);
3327
3328         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3329              cmp_xe, swap_xe);
3330 }
3331
3332 /*
3333  * After we move xattr from block to index btree, we have to
3334  * update ocfs2_xattr_search to the new xe and base.
3335  *
3336  * When the entry is in xattr block, xattr_bh indicates the storage place.
3337  * While if the entry is in index b-tree, "bucket" indicates the
3338  * real place of the xattr.
3339  */
3340 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3341                                             struct ocfs2_xattr_search *xs,
3342                                             struct buffer_head *old_bh)
3343 {
3344         char *buf = old_bh->b_data;
3345         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3346         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3347         int i;
3348
3349         xs->header = bucket_xh(xs->bucket);
3350         xs->base = bucket_block(xs->bucket, 0);
3351         xs->end = xs->base + inode->i_sb->s_blocksize;
3352
3353         if (xs->not_found)
3354                 return;
3355
3356         i = xs->here - old_xh->xh_entries;
3357         xs->here = &xs->header->xh_entries[i];
3358 }
3359
3360 static int ocfs2_xattr_create_index_block(struct inode *inode,
3361                                           struct ocfs2_xattr_search *xs,
3362                                           struct ocfs2_xattr_set_ctxt *ctxt)
3363 {
3364         int ret;
3365         u32 bit_off, len;
3366         u64 blkno;
3367         handle_t *handle = ctxt->handle;
3368         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3369         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3370         struct buffer_head *xb_bh = xs->xattr_bh;
3371         struct ocfs2_xattr_block *xb =
3372                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3373         struct ocfs2_xattr_tree_root *xr;
3374         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3375
3376         mlog(0, "create xattr index block for %llu\n",
3377              (unsigned long long)xb_bh->b_blocknr);
3378
3379         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3380         BUG_ON(!xs->bucket);
3381
3382         /*
3383          * XXX:
3384          * We can use this lock for now, and maybe move to a dedicated mutex
3385          * if performance becomes a problem later.
3386          */
3387         down_write(&oi->ip_alloc_sem);
3388
3389         ret = ocfs2_journal_access_xb(handle, inode, xb_bh,
3390                                       OCFS2_JOURNAL_ACCESS_WRITE);
3391         if (ret) {
3392                 mlog_errno(ret);
3393                 goto out;
3394         }
3395
3396         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3397                                      1, 1, &bit_off, &len);
3398         if (ret) {
3399                 mlog_errno(ret);
3400                 goto out;
3401         }
3402
3403         /*
3404          * The bucket may spread in many blocks, and
3405          * we will only touch the 1st block and the last block
3406          * in the whole bucket(one for entry and one for data).
3407          */
3408         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3409
3410         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3411              (unsigned long long)blkno);
3412
3413         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3414         if (ret) {
3415                 mlog_errno(ret);
3416                 goto out;
3417         }
3418
3419         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3420                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3421         if (ret) {
3422                 mlog_errno(ret);
3423                 goto out;
3424         }
3425
3426         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3427         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3428
3429         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3430
3431         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3432         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3433                offsetof(struct ocfs2_xattr_block, xb_attrs));
3434
3435         xr = &xb->xb_attrs.xb_root;
3436         xr->xt_clusters = cpu_to_le32(1);
3437         xr->xt_last_eb_blk = 0;
3438         xr->xt_list.l_tree_depth = 0;
3439         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3440         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3441
3442         xr->xt_list.l_recs[0].e_cpos = 0;
3443         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3444         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3445
3446         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3447
3448         ocfs2_journal_dirty(handle, xb_bh);
3449
3450 out:
3451         up_write(&oi->ip_alloc_sem);
3452
3453         return ret;
3454 }
3455
3456 static int cmp_xe_offset(const void *a, const void *b)
3457 {
3458         const struct ocfs2_xattr_entry *l = a, *r = b;
3459         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3460         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3461
3462         if (l_name_offset < r_name_offset)
3463                 return 1;
3464         if (l_name_offset > r_name_offset)
3465                 return -1;
3466         return 0;
3467 }
3468
3469 /*
3470  * defrag a xattr bucket if we find that the bucket has some
3471  * holes beteen name/value pairs.
3472  * We will move all the name/value pairs to the end of the bucket
3473  * so that we can spare some space for insertion.
3474  */
3475 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3476                                      handle_t *handle,
3477                                      struct ocfs2_xattr_bucket *bucket)
3478 {
3479         int ret, i;
3480         size_t end, offset, len, value_len;
3481         struct ocfs2_xattr_header *xh;
3482         char *entries, *buf, *bucket_buf = NULL;
3483         u64 blkno = bucket_blkno(bucket);
3484         u16 xh_free_start;
3485         size_t blocksize = inode->i_sb->s_blocksize;
3486         struct ocfs2_xattr_entry *xe;
3487
3488         /*
3489          * In order to make the operation more efficient and generic,
3490          * we copy all the blocks into a contiguous memory and do the
3491          * defragment there, so if anything is error, we will not touch
3492          * the real block.
3493          */
3494         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3495         if (!bucket_buf) {
3496                 ret = -EIO;
3497                 goto out;
3498         }
3499
3500         buf = bucket_buf;
3501         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3502                 memcpy(buf, bucket_block(bucket, i), blocksize);
3503
3504         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3505                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3506         if (ret < 0) {
3507                 mlog_errno(ret);
3508                 goto out;
3509         }
3510
3511         xh = (struct ocfs2_xattr_header *)bucket_buf;
3512         entries = (char *)xh->xh_entries;
3513         xh_free_start = le16_to_cpu(xh->xh_free_start);
3514
3515         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3516              "xh_free_start = %u, xh_name_value_len = %u.\n",
3517              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3518              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3519
3520         /*
3521          * sort all the entries by their offset.
3522          * the largest will be the first, so that we can
3523          * move them to the end one by one.
3524          */
3525         sort(entries, le16_to_cpu(xh->xh_count),
3526              sizeof(struct ocfs2_xattr_entry),
3527              cmp_xe_offset, swap_xe);
3528
3529         /* Move all name/values to the end of the bucket. */
3530         xe = xh->xh_entries;
3531         end = OCFS2_XATTR_BUCKET_SIZE;
3532         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3533                 offset = le16_to_cpu(xe->xe_name_offset);
3534                 if (ocfs2_xattr_is_local(xe))
3535                         value_len = OCFS2_XATTR_SIZE(
3536                                         le64_to_cpu(xe->xe_value_size));
3537                 else
3538                         value_len = OCFS2_XATTR_ROOT_SIZE;
3539                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3540
3541                 /*
3542                  * We must make sure that the name/value pair
3543                  * exist in the same block. So adjust end to
3544                  * the previous block end if needed.
3545                  */
3546                 if (((end - len) / blocksize !=
3547                         (end - 1) / blocksize))
3548                         end = end - end % blocksize;
3549
3550                 if (end > offset + len) {
3551                         memmove(bucket_buf + end - len,
3552                                 bucket_buf + offset, len);
3553                         xe->xe_name_offset = cpu_to_le16(end - len);
3554                 }
3555
3556                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3557                                 "bucket %llu\n", (unsigned long long)blkno);
3558
3559                 end -= len;
3560         }
3561
3562         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3563                         "bucket %llu\n", (unsigned long long)blkno);
3564
3565         if (xh_free_start == end)
3566                 goto out;
3567
3568         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3569         xh->xh_free_start = cpu_to_le16(end);
3570
3571         /* sort the entries by their name_hash. */
3572         sort(entries, le16_to_cpu(xh->xh_count),
3573              sizeof(struct ocfs2_xattr_entry),
3574              cmp_xe, swap_xe);
3575
3576         buf = bucket_buf;
3577         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3578                 memcpy(bucket_block(bucket, i), buf, blocksize);
3579         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3580
3581 out:
3582         kfree(bucket_buf);
3583         return ret;
3584 }
3585
3586 /*
3587  * prev_blkno points to the start of an existing extent.  new_blkno
3588  * points to a newly allocated extent.  Because we know each of our
3589  * clusters contains more than bucket, we can easily split one cluster
3590  * at a bucket boundary.  So we take the last cluster of the existing
3591  * extent and split it down the middle.  We move the last half of the
3592  * buckets in the last cluster of the existing extent over to the new
3593  * extent.
3594  *
3595  * first_bh is the buffer at prev_blkno so we can update the existing
3596  * extent's bucket count.  header_bh is the bucket were we were hoping
3597  * to insert our xattr.  If the bucket move places the target in the new
3598  * extent, we'll update first_bh and header_bh after modifying the old
3599  * extent.
3600  *
3601  * first_hash will be set as the 1st xe's name_hash in the new extent.
3602  */
3603 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3604                                                handle_t *handle,
3605                                                struct ocfs2_xattr_bucket *first,
3606                                                struct ocfs2_xattr_bucket *target,
3607                                                u64 new_blkno,
3608                                                u32 num_clusters,
3609                                                u32 *first_hash)
3610 {
3611         int ret;
3612         struct super_block *sb = inode->i_sb;
3613         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3614         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3615         int to_move = num_buckets / 2;
3616         u64 src_blkno;
3617         u64 last_cluster_blkno = bucket_blkno(first) +
3618                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3619
3620         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3621         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3622
3623         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3624              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3625
3626         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3627                                      last_cluster_blkno, new_blkno,
3628                                      to_move, first_hash);
3629         if (ret) {
3630                 mlog_errno(ret);
3631                 goto out;
3632         }
3633
3634         /* This is the first bucket that got moved */
3635         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3636
3637         /*
3638          * If the target bucket was part of the moved buckets, we need to
3639          * update first and target.
3640          */
3641         if (bucket_blkno(target) >= src_blkno) {
3642                 /* Find the block for the new target bucket */
3643                 src_blkno = new_blkno +
3644                         (bucket_blkno(target) - src_blkno);
3645
3646                 ocfs2_xattr_bucket_relse(first);
3647                 ocfs2_xattr_bucket_relse(target);
3648
3649                 /*
3650                  * These shouldn't fail - the buffers are in the
3651                  * journal from ocfs2_cp_xattr_bucket().
3652                  */
3653                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3654                 if (ret) {
3655                         mlog_errno(ret);
3656                         goto out;
3657                 }
3658                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3659                 if (ret)
3660                         mlog_errno(ret);
3661
3662         }
3663
3664 out:
3665         return ret;
3666 }
3667
3668 /*
3669  * Find the suitable pos when we divide a bucket into 2.
3670  * We have to make sure the xattrs with the same hash value exist
3671  * in the same bucket.
3672  *
3673  * If this ocfs2_xattr_header covers more than one hash value, find a
3674  * place where the hash value changes.  Try to find the most even split.
3675  * The most common case is that all entries have different hash values,
3676  * and the first check we make will find a place to split.
3677  */
3678 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3679 {
3680         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3681         int count = le16_to_cpu(xh->xh_count);
3682         int delta, middle = count / 2;
3683
3684         /*
3685          * We start at the middle.  Each step gets farther away in both
3686          * directions.  We therefore hit the change in hash value
3687          * nearest to the middle.  Note that this loop does not execute for
3688          * count < 2.
3689          */
3690         for (delta = 0; delta < middle; delta++) {
3691                 /* Let's check delta earlier than middle */
3692                 if (cmp_xe(&entries[middle - delta - 1],
3693                            &entries[middle - delta]))
3694                         return middle - delta;
3695
3696                 /* For even counts, don't walk off the end */
3697                 if ((middle + delta + 1) == count)
3698                         continue;
3699
3700                 /* Now try delta past middle */
3701                 if (cmp_xe(&entries[middle + delta],
3702                            &entries[middle + delta + 1]))
3703                         return middle + delta + 1;
3704         }
3705
3706         /* Every entry had the same hash */
3707         return count;
3708 }
3709
3710 /*
3711  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3712  * first_hash will record the 1st hash of the new bucket.
3713  *
3714  * Normally half of the xattrs will be moved.  But we have to make
3715  * sure that the xattrs with the same hash value are stored in the
3716  * same bucket. If all the xattrs in this bucket have the same hash
3717  * value, the new bucket will be initialized as an empty one and the
3718  * first_hash will be initialized as (hash_value+1).
3719  */
3720 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3721                                     handle_t *handle,
3722                                     u64 blk,
3723                                     u64 new_blk,
3724                                     u32 *first_hash,
3725                                     int new_bucket_head)
3726 {
3727         int ret, i;
3728         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3729         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3730         struct ocfs2_xattr_header *xh;
3731         struct ocfs2_xattr_entry *xe;
3732         int blocksize = inode->i_sb->s_blocksize;
3733
3734         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3735              (unsigned long long)blk, (unsigned long long)new_blk);
3736
3737         s_bucket = ocfs2_xattr_bucket_new(inode);
3738         t_bucket = ocfs2_xattr_bucket_new(inode);
3739         if (!s_bucket || !t_bucket) {
3740                 ret = -ENOMEM;
3741                 mlog_errno(ret);
3742                 goto out;
3743         }
3744
3745         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3746         if (ret) {
3747                 mlog_errno(ret);
3748                 goto out;
3749         }
3750
3751         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3752                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3753         if (ret) {
3754                 mlog_errno(ret);
3755                 goto out;
3756         }
3757
3758         /*
3759          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3760          * there's no need to read it.
3761          */
3762         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3763         if (ret) {
3764                 mlog_errno(ret);
3765                 goto out;
3766         }
3767
3768         /*
3769          * Hey, if we're overwriting t_bucket, what difference does
3770          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3771          * same part of ocfs2_cp_xattr_bucket().
3772          */
3773         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3774                                                 new_bucket_head ?
3775                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3776                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3777         if (ret) {
3778                 mlog_errno(ret);
3779                 goto out;
3780         }
3781
3782         xh = bucket_xh(s_bucket);
3783         count = le16_to_cpu(xh->xh_count);
3784         start = ocfs2_xattr_find_divide_pos(xh);
3785
3786         if (start == count) {
3787                 xe = &xh->xh_entries[start-1];
3788
3789                 /*
3790                  * initialized a new empty bucket here.
3791                  * The hash value is set as one larger than
3792                  * that of the last entry in the previous bucket.
3793                  */
3794                 for (i = 0; i < t_bucket->bu_blocks; i++)
3795                         memset(bucket_block(t_bucket, i), 0, blocksize);
3796
3797                 xh = bucket_xh(t_bucket);
3798                 xh->xh_free_start = cpu_to_le16(blocksize);
3799                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3800                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3801
3802                 goto set_num_buckets;
3803         }
3804
3805         /* copy the whole bucket to the new first. */
3806         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3807
3808         /* update the new bucket. */
3809         xh = bucket_xh(t_bucket);
3810
3811         /*
3812          * Calculate the total name/value len and xh_free_start for
3813          * the old bucket first.
3814          */
3815         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3816         name_value_len = 0;
3817         for (i = 0; i < start; i++) {
3818                 xe = &xh->xh_entries[i];
3819                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3820                 if (ocfs2_xattr_is_local(xe))
3821                         xe_len +=
3822                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3823                 else
3824                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3825                 name_value_len += xe_len;
3826                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3827                         name_offset = le16_to_cpu(xe->xe_name_offset);
3828         }
3829
3830         /*
3831          * Now begin the modification to the new bucket.
3832          *
3833          * In the new bucket, We just move the xattr entry to the beginning
3834          * and don't touch the name/value. So there will be some holes in the
3835          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3836          * called.
3837          */
3838         xe = &xh->xh_entries[start];
3839         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3840         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3841              (int)((char *)xe - (char *)xh),
3842              (int)((char *)xh->xh_entries - (char *)xh));
3843         memmove((char *)xh->xh_entries, (char *)xe, len);
3844         xe = &xh->xh_entries[count - start];
3845         len = sizeof(struct ocfs2_xattr_entry) * start;
3846         memset((char *)xe, 0, len);
3847
3848         le16_add_cpu(&xh->xh_count, -start);
3849         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3850
3851         /* Calculate xh_free_start for the new bucket. */
3852         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3853         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3854                 xe = &xh->xh_entries[i];
3855                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3856                 if (ocfs2_xattr_is_local(xe))
3857                         xe_len +=
3858                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3859                 else
3860                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3861                 if (le16_to_cpu(xe->xe_name_offset) <
3862                     le16_to_cpu(xh->xh_free_start))
3863                         xh->xh_free_start = xe->xe_name_offset;
3864         }
3865
3866 set_num_buckets:
3867         /* set xh->xh_num_buckets for the new xh. */
3868         if (new_bucket_head)
3869                 xh->xh_num_buckets = cpu_to_le16(1);
3870         else
3871                 xh->xh_num_buckets = 0;
3872
3873         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3874
3875         /* store the first_hash of the new bucket. */
3876         if (first_hash)
3877                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3878
3879         /*
3880          * Now only update the 1st block of the old bucket.  If we
3881          * just added a new empty bucket, there is no need to modify
3882          * it.
3883          */
3884         if (start == count)
3885                 goto out;
3886
3887         xh = bucket_xh(s_bucket);
3888         memset(&xh->xh_entries[start], 0,
3889                sizeof(struct ocfs2_xattr_entry) * (count - start));
3890         xh->xh_count = cpu_to_le16(start);
3891         xh->xh_free_start = cpu_to_le16(name_offset);
3892         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3893
3894         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
3895
3896 out:
3897         ocfs2_xattr_bucket_free(s_bucket);
3898         ocfs2_xattr_bucket_free(t_bucket);
3899
3900         return ret;
3901 }
3902
3903 /*
3904  * Copy xattr from one bucket to another bucket.
3905  *
3906  * The caller must make sure that the journal transaction
3907  * has enough space for journaling.
3908  */
3909 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3910                                  handle_t *handle,
3911                                  u64 s_blkno,
3912                                  u64 t_blkno,
3913                                  int t_is_new)
3914 {
3915         int ret;
3916         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3917
3918         BUG_ON(s_blkno == t_blkno);
3919
3920         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3921              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
3922              t_is_new);
3923
3924         s_bucket = ocfs2_xattr_bucket_new(inode);
3925         t_bucket = ocfs2_xattr_bucket_new(inode);
3926         if (!s_bucket || !t_bucket) {
3927                 ret = -ENOMEM;
3928                 mlog_errno(ret);
3929                 goto out;
3930         }
3931
3932         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
3933         if (ret)
3934                 goto out;
3935
3936         /*
3937          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
3938          * there's no need to read it.
3939          */
3940         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
3941         if (ret)
3942                 goto out;
3943
3944         /*
3945          * Hey, if we're overwriting t_bucket, what difference does
3946          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
3947          * cluster to fill, we came here from
3948          * ocfs2_mv_xattr_buckets(), and it is really new -
3949          * ACCESS_CREATE is required.  But we also might have moved data
3950          * out of t_bucket before extending back into it.
3951          * ocfs2_add_new_xattr_bucket() can do this - its call to
3952          * ocfs2_add_new_xattr_cluster() may have created a new extent
3953          * and copied out the end of the old extent.  Then it re-extends
3954          * the old extent back to create space for new xattrs.  That's
3955          * how we get here, and the bucket isn't really new.
3956          */
3957         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3958                                                 t_is_new ?
3959                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3960                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3961         if (ret)
3962                 goto out;
3963
3964         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3965         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3966
3967 out:
3968         ocfs2_xattr_bucket_free(t_bucket);
3969         ocfs2_xattr_bucket_free(s_bucket);
3970
3971         return ret;
3972 }
3973
3974 /*
3975  * src_blk points to the start of an existing extent.  last_blk points to
3976  * last cluster in that extent.  to_blk points to a newly allocated
3977  * extent.  We copy the buckets from the cluster at last_blk to the new
3978  * extent.  If start_bucket is non-zero, we skip that many buckets before
3979  * we start copying.  The new extent's xh_num_buckets gets set to the
3980  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
3981  * by the same amount.
3982  */
3983 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
3984                                   u64 src_blk, u64 last_blk, u64 to_blk,
3985                                   unsigned int start_bucket,
3986                                   u32 *first_hash)
3987 {
3988         int i, ret, credits;
3989         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3990         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3991         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3992         struct ocfs2_xattr_bucket *old_first, *new_first;
3993
3994         mlog(0, "mv xattrs from cluster %llu to %llu\n",
3995              (unsigned long long)last_blk, (unsigned long long)to_blk);
3996
3997         BUG_ON(start_bucket >= num_buckets);
3998         if (start_bucket) {
3999                 num_buckets -= start_bucket;
4000                 last_blk += (start_bucket * blks_per_bucket);
4001         }
4002
4003         /* The first bucket of the original extent */
4004         old_first = ocfs2_xattr_bucket_new(inode);
4005         /* The first bucket of the new extent */
4006         new_first = ocfs2_xattr_bucket_new(inode);
4007         if (!old_first || !new_first) {
4008                 ret = -ENOMEM;
4009                 mlog_errno(ret);
4010                 goto out;
4011         }
4012
4013         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4014         if (ret) {
4015                 mlog_errno(ret);
4016                 goto out;
4017         }
4018
4019         /*
4020          * We need to update the first bucket of the old extent and all
4021          * the buckets going to the new extent.
4022          */
4023         credits = ((num_buckets + 1) * blks_per_bucket) +
4024                 handle->h_buffer_credits;
4025         ret = ocfs2_extend_trans(handle, credits);
4026         if (ret) {
4027                 mlog_errno(ret);
4028                 goto out;
4029         }
4030
4031         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4032                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4033         if (ret) {
4034                 mlog_errno(ret);
4035                 goto out;
4036         }
4037
4038         for (i = 0; i < num_buckets; i++) {
4039                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4040                                             last_blk + (i * blks_per_bucket),
4041                                             to_blk + (i * blks_per_bucket),
4042                                             1);
4043                 if (ret) {
4044                         mlog_errno(ret);
4045                         goto out;
4046                 }
4047         }
4048
4049         /*
4050          * Get the new bucket ready before we dirty anything
4051          * (This actually shouldn't fail, because we already dirtied
4052          * it once in ocfs2_cp_xattr_bucket()).
4053          */
4054         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4055         if (ret) {
4056                 mlog_errno(ret);
4057                 goto out;
4058         }
4059         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4060                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4061         if (ret) {
4062                 mlog_errno(ret);
4063                 goto out;
4064         }
4065
4066         /* Now update the headers */
4067         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4068         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4069
4070         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4071         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4072
4073         if (first_hash)
4074                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4075
4076 out:
4077         ocfs2_xattr_bucket_free(new_first);
4078         ocfs2_xattr_bucket_free(old_first);
4079         return ret;
4080 }
4081
4082 /*
4083  * Move some xattrs in this cluster to the new cluster.
4084  * This function should only be called when bucket size == cluster size.
4085  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4086  */
4087 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4088                                       handle_t *handle,
4089                                       u64 prev_blk,
4090                                       u64 new_blk,
4091                                       u32 *first_hash)
4092 {
4093         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4094         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4095
4096         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4097
4098         ret = ocfs2_extend_trans(handle, credits);
4099         if (ret) {
4100                 mlog_errno(ret);
4101                 return ret;
4102         }
4103
4104         /* Move half of the xattr in start_blk to the next bucket. */
4105         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4106                                           new_blk, first_hash, 1);
4107 }
4108
4109 /*
4110  * Move some xattrs from the old cluster to the new one since they are not
4111  * contiguous in ocfs2 xattr tree.
4112  *
4113  * new_blk starts a new separate cluster, and we will move some xattrs from
4114  * prev_blk to it. v_start will be set as the first name hash value in this
4115  * new cluster so that it can be used as e_cpos during tree insertion and
4116  * don't collide with our original b-tree operations. first_bh and header_bh
4117  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4118  * to extend the insert bucket.
4119  *
4120  * The problem is how much xattr should we move to the new one and when should
4121  * we update first_bh and header_bh?
4122  * 1. If cluster size > bucket size, that means the previous cluster has more
4123  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4124  *    update the first_bh and header_bh if the insert bucket has been moved
4125  *    to the new cluster.
4126  * 2. If cluster_size == bucket_size:
4127  *    a) If the previous extent rec has more than one cluster and the insert
4128  *       place isn't in the last cluster, copy the entire last cluster to the
4129  *       new one. This time, we don't need to upate the first_bh and header_bh
4130  *       since they will not be moved into the new cluster.
4131  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4132  *       the new one. And we set the extend flag to zero if the insert place is
4133  *       moved into the new allocated cluster since no extend is needed.
4134  */
4135 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4136                                             handle_t *handle,
4137                                             struct ocfs2_xattr_bucket *first,
4138                                             struct ocfs2_xattr_bucket *target,
4139                                             u64 new_blk,
4140                                             u32 prev_clusters,
4141                                             u32 *v_start,
4142                                             int *extend)
4143 {
4144         int ret;
4145
4146         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4147              (unsigned long long)bucket_blkno(first), prev_clusters,
4148              (unsigned long long)new_blk);
4149
4150         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4151                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4152                                                           handle,
4153                                                           first, target,
4154                                                           new_blk,
4155                                                           prev_clusters,
4156                                                           v_start);
4157                 if (ret)
4158                         mlog_errno(ret);
4159         } else {
4160                 /* The start of the last cluster in the first extent */
4161                 u64 last_blk = bucket_blkno(first) +
4162                         ((prev_clusters - 1) *
4163                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4164
4165                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4166                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4167                                                      bucket_blkno(first),
4168                                                      last_blk, new_blk, 0,
4169                                                      v_start);
4170                         if (ret)
4171                                 mlog_errno(ret);
4172                 } else {
4173                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4174                                                          last_blk, new_blk,
4175                                                          v_start);
4176                         if (ret)
4177                                 mlog_errno(ret);
4178
4179                         if ((bucket_blkno(target) == last_blk) && extend)
4180                                 *extend = 0;
4181                 }
4182         }
4183
4184         return ret;
4185 }
4186
4187 /*
4188  * Add a new cluster for xattr storage.
4189  *
4190  * If the new cluster is contiguous with the previous one, it will be
4191  * appended to the same extent record, and num_clusters will be updated.
4192  * If not, we will insert a new extent for it and move some xattrs in
4193  * the last cluster into the new allocated one.
4194  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4195  * lose the benefits of hashing because we'll have to search large leaves.
4196  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4197  * if it's bigger).
4198  *
4199  * first_bh is the first block of the previous extent rec and header_bh
4200  * indicates the bucket we will insert the new xattrs. They will be updated
4201  * when the header_bh is moved into the new cluster.
4202  */
4203 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4204                                        struct buffer_head *root_bh,
4205                                        struct ocfs2_xattr_bucket *first,
4206                                        struct ocfs2_xattr_bucket *target,
4207                                        u32 *num_clusters,
4208                                        u32 prev_cpos,
4209                                        int *extend,
4210                                        struct ocfs2_xattr_set_ctxt *ctxt)
4211 {
4212         int ret;
4213         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4214         u32 prev_clusters = *num_clusters;
4215         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4216         u64 block;
4217         handle_t *handle = ctxt->handle;
4218         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4219         struct ocfs2_extent_tree et;
4220
4221         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4222              "previous xattr blkno = %llu\n",
4223              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4224              prev_cpos, (unsigned long long)bucket_blkno(first));
4225
4226         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4227
4228         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4229                                       OCFS2_JOURNAL_ACCESS_WRITE);
4230         if (ret < 0) {
4231                 mlog_errno(ret);
4232                 goto leave;
4233         }
4234
4235         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4236                                      clusters_to_add, &bit_off, &num_bits);
4237         if (ret < 0) {
4238                 if (ret != -ENOSPC)
4239                         mlog_errno(ret);
4240                 goto leave;
4241         }
4242
4243         BUG_ON(num_bits > clusters_to_add);
4244
4245         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4246         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4247              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4248
4249         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4250             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4251              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4252                 /*
4253                  * If this cluster is contiguous with the old one and
4254                  * adding this new cluster, we don't surpass the limit of
4255                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4256                  * initialized and used like other buckets in the previous
4257                  * cluster.
4258                  * So add it as a contiguous one. The caller will handle
4259                  * its init process.
4260                  */
4261                 v_start = prev_cpos + prev_clusters;
4262                 *num_clusters = prev_clusters + num_bits;
4263                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4264                      num_bits);
4265         } else {
4266                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4267                                                        handle,
4268                                                        first,
4269                                                        target,
4270                                                        block,
4271                                                        prev_clusters,
4272                                                        &v_start,
4273                                                        extend);
4274                 if (ret) {
4275                         mlog_errno(ret);
4276                         goto leave;
4277                 }
4278         }
4279
4280         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4281              num_bits, (unsigned long long)block, v_start);
4282         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
4283                                   num_bits, 0, ctxt->meta_ac);
4284         if (ret < 0) {
4285                 mlog_errno(ret);
4286                 goto leave;
4287         }
4288
4289         ret = ocfs2_journal_dirty(handle, root_bh);
4290         if (ret < 0)
4291                 mlog_errno(ret);
4292
4293 leave:
4294         return ret;
4295 }
4296
4297 /*
4298  * We are given an extent.  'first' is the bucket at the very front of
4299  * the extent.  The extent has space for an additional bucket past
4300  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4301  * of the target bucket.  We wish to shift every bucket past the target
4302  * down one, filling in that additional space.  When we get back to the
4303  * target, we split the target between itself and the now-empty bucket
4304  * at target+1 (aka, target_blkno + blks_per_bucket).
4305  */
4306 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4307                                      handle_t *handle,
4308                                      struct ocfs2_xattr_bucket *first,
4309                                      u64 target_blk,
4310                                      u32 num_clusters)
4311 {
4312         int ret, credits;
4313         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4314         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4315         u64 end_blk;
4316         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4317
4318         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4319              "from %llu, len = %u\n", (unsigned long long)target_blk,
4320              (unsigned long long)bucket_blkno(first), num_clusters);
4321
4322         /* The extent must have room for an additional bucket */
4323         BUG_ON(new_bucket >=
4324                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4325
4326         /* end_blk points to the last existing bucket */
4327         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4328
4329         /*
4330          * end_blk is the start of the last existing bucket.
4331          * Thus, (end_blk - target_blk) covers the target bucket and
4332          * every bucket after it up to, but not including, the last
4333          * existing bucket.  Then we add the last existing bucket, the
4334          * new bucket, and the first bucket (3 * blk_per_bucket).
4335          */
4336         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4337                   handle->h_buffer_credits;
4338         ret = ocfs2_extend_trans(handle, credits);
4339         if (ret) {
4340                 mlog_errno(ret);
4341                 goto out;
4342         }
4343
4344         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4345                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4346         if (ret) {
4347                 mlog_errno(ret);
4348                 goto out;
4349         }
4350
4351         while (end_blk != target_blk) {
4352                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4353                                             end_blk + blk_per_bucket, 0);
4354                 if (ret)
4355                         goto out;
4356                 end_blk -= blk_per_bucket;
4357         }
4358
4359         /* Move half of the xattr in target_blkno to the next bucket. */
4360         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4361                                         target_blk + blk_per_bucket, NULL, 0);
4362
4363         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4364         ocfs2_xattr_bucket_journal_dirty(handle, first);
4365
4366 out:
4367         return ret;
4368 }
4369
4370 /*
4371  * Add new xattr bucket in an extent record and adjust the buckets
4372  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4373  * bucket we want to insert into.
4374  *
4375  * In the easy case, we will move all the buckets after target down by
4376  * one. Half of target's xattrs will be moved to the next bucket.
4377  *
4378  * If current cluster is full, we'll allocate a new one.  This may not
4379  * be contiguous.  The underlying calls will make sure that there is
4380  * space for the insert, shifting buckets around if necessary.
4381  * 'target' may be moved by those calls.
4382  */
4383 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4384                                       struct buffer_head *xb_bh,
4385                                       struct ocfs2_xattr_bucket *target,
4386                                       struct ocfs2_xattr_set_ctxt *ctxt)
4387 {
4388         struct ocfs2_xattr_block *xb =
4389                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4390         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4391         struct ocfs2_extent_list *el = &xb_root->xt_list;
4392         u32 name_hash =
4393                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4394         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4395         int ret, num_buckets, extend = 1;
4396         u64 p_blkno;
4397         u32 e_cpos, num_clusters;
4398         /* The bucket at the front of the extent */
4399         struct ocfs2_xattr_bucket *first;
4400
4401         mlog(0, "Add new xattr bucket starting from %llu\n",
4402              (unsigned long long)bucket_blkno(target));
4403
4404         /* The first bucket of the original extent */
4405         first = ocfs2_xattr_bucket_new(inode);
4406         if (!first) {
4407                 ret = -ENOMEM;
4408                 mlog_errno(ret);
4409                 goto out;
4410         }
4411
4412         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4413                                   &num_clusters, el);
4414         if (ret) {
4415                 mlog_errno(ret);
4416                 goto out;
4417         }
4418
4419         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4420         if (ret) {
4421                 mlog_errno(ret);
4422                 goto out;
4423         }
4424
4425         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4426         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4427                 /*
4428                  * This can move first+target if the target bucket moves
4429                  * to the new extent.
4430                  */
4431                 ret = ocfs2_add_new_xattr_cluster(inode,
4432                                                   xb_bh,
4433                                                   first,
4434                                                   target,
4435                                                   &num_clusters,
4436                                                   e_cpos,
4437                                                   &extend,
4438                                                   ctxt);
4439                 if (ret) {
4440                         mlog_errno(ret);
4441                         goto out;
4442                 }
4443         }
4444
4445         if (extend) {
4446                 ret = ocfs2_extend_xattr_bucket(inode,
4447                                                 ctxt->handle,
4448                                                 first,
4449                                                 bucket_blkno(target),
4450                                                 num_clusters);
4451                 if (ret)
4452                         mlog_errno(ret);
4453         }
4454
4455 out:
4456         ocfs2_xattr_bucket_free(first);
4457
4458         return ret;
4459 }
4460
4461 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4462                                         struct ocfs2_xattr_bucket *bucket,
4463                                         int offs)
4464 {
4465         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4466
4467         offs = offs % inode->i_sb->s_blocksize;
4468         return bucket_block(bucket, block_off) + offs;
4469 }
4470
4471 /*
4472  * Handle the normal xattr set, including replace, delete and new.
4473  *
4474  * Note: "local" indicates the real data's locality. So we can't
4475  * just its bucket locality by its length.
4476  */
4477 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4478                                          struct ocfs2_xattr_info *xi,
4479                                          struct ocfs2_xattr_search *xs,
4480                                          u32 name_hash,
4481                                          int local)
4482 {
4483         struct ocfs2_xattr_entry *last, *xe;
4484         int name_len = strlen(xi->name);
4485         struct ocfs2_xattr_header *xh = xs->header;
4486         u16 count = le16_to_cpu(xh->xh_count), start;
4487         size_t blocksize = inode->i_sb->s_blocksize;
4488         char *val;
4489         size_t offs, size, new_size;
4490
4491         last = &xh->xh_entries[count];
4492         if (!xs->not_found) {
4493                 xe = xs->here;
4494                 offs = le16_to_cpu(xe->xe_name_offset);
4495                 if (ocfs2_xattr_is_local(xe))
4496                         size = OCFS2_XATTR_SIZE(name_len) +
4497                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4498                 else
4499                         size = OCFS2_XATTR_SIZE(name_len) +
4500                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4501
4502                 /*
4503                  * If the new value will be stored outside, xi->value has been
4504                  * initalized as an empty ocfs2_xattr_value_root, and the same
4505                  * goes with xi->value_len, so we can set new_size safely here.
4506                  * See ocfs2_xattr_set_in_bucket.
4507                  */
4508                 new_size = OCFS2_XATTR_SIZE(name_len) +
4509                            OCFS2_XATTR_SIZE(xi->value_len);
4510
4511                 le16_add_cpu(&xh->xh_name_value_len, -size);
4512                 if (xi->value) {
4513                         if (new_size > size)
4514                                 goto set_new_name_value;
4515
4516                         /* Now replace the old value with new one. */
4517                         if (local)
4518                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4519                         else
4520                                 xe->xe_value_size = 0;
4521
4522                         val = ocfs2_xattr_bucket_get_val(inode,
4523                                                          xs->bucket, offs);
4524                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4525                                size - OCFS2_XATTR_SIZE(name_len));
4526                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4527                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4528                                        xi->value, xi->value_len);
4529
4530                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4531                         ocfs2_xattr_set_local(xe, local);
4532                         return;
4533                 } else {
4534                         /*
4535                          * Remove the old entry if there is more than one.
4536                          * We don't remove the last entry so that we can
4537                          * use it to indicate the hash value of the empty
4538                          * bucket.
4539                          */
4540                         last -= 1;
4541                         le16_add_cpu(&xh->xh_count, -1);
4542                         if (xh->xh_count) {
4543                                 memmove(xe, xe + 1,
4544                                         (void *)last - (void *)xe);
4545                                 memset(last, 0,
4546                                        sizeof(struct ocfs2_xattr_entry));
4547                         } else
4548                                 xh->xh_free_start =
4549                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4550
4551                         return;
4552                 }
4553         } else {
4554                 /* find a new entry for insert. */
4555                 int low = 0, high = count - 1, tmp;
4556                 struct ocfs2_xattr_entry *tmp_xe;
4557
4558                 while (low <= high && count) {
4559                         tmp = (low + high) / 2;
4560                         tmp_xe = &xh->xh_entries[tmp];
4561
4562                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4563                                 low = tmp + 1;
4564                         else if (name_hash <
4565                                  le32_to_cpu(tmp_xe->xe_name_hash))
4566                                 high = tmp - 1;
4567                         else {
4568                                 low = tmp;
4569                                 break;
4570                         }
4571                 }
4572
4573                 xe = &xh->xh_entries[low];
4574                 if (low != count)
4575                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4576
4577                 le16_add_cpu(&xh->xh_count, 1);
4578                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4579                 xe->xe_name_hash = cpu_to_le32(name_hash);
4580                 xe->xe_name_len = name_len;
4581                 ocfs2_xattr_set_type(xe, xi->name_index);
4582         }
4583
4584 set_new_name_value:
4585         /* Insert the new name+value. */
4586         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4587
4588         /*
4589          * We must make sure that the name/value pair
4590          * exists in the same block.
4591          */
4592         offs = le16_to_cpu(xh->xh_free_start);
4593         start = offs - size;
4594
4595         if (start >> inode->i_sb->s_blocksize_bits !=
4596             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4597                 offs = offs - offs % blocksize;
4598                 xh->xh_free_start = cpu_to_le16(offs);
4599         }
4600
4601         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4602         xe->xe_name_offset = cpu_to_le16(offs - size);
4603
4604         memset(val, 0, size);
4605         memcpy(val, xi->name, name_len);
4606         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4607
4608         xe->xe_value_size = cpu_to_le64(xi->value_len);
4609         ocfs2_xattr_set_local(xe, local);
4610         xs->here = xe;
4611         le16_add_cpu(&xh->xh_free_start, -size);
4612         le16_add_cpu(&xh->xh_name_value_len, size);
4613
4614         return;
4615 }
4616
4617 /*
4618  * Set the xattr entry in the specified bucket.
4619  * The bucket is indicated by xs->bucket and it should have the enough
4620  * space for the xattr insertion.
4621  */
4622 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4623                                            handle_t *handle,
4624                                            struct ocfs2_xattr_info *xi,
4625                                            struct ocfs2_xattr_search *xs,
4626                                            u32 name_hash,
4627                                            int local)
4628 {
4629         int ret;
4630         u64 blkno;
4631
4632         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4633              (unsigned long)xi->value_len, xi->name_index,
4634              (unsigned long long)bucket_blkno(xs->bucket));
4635
4636         if (!xs->bucket->bu_bhs[1]) {
4637                 blkno = bucket_blkno(xs->bucket);
4638                 ocfs2_xattr_bucket_relse(xs->bucket);
4639                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4640                 if (ret) {
4641                         mlog_errno(ret);
4642                         goto out;
4643                 }
4644         }
4645
4646         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4647                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4648         if (ret < 0) {
4649                 mlog_errno(ret);
4650                 goto out;
4651         }
4652
4653         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4654         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4655
4656 out:
4657         return ret;
4658 }
4659
4660 /*
4661  * Truncate the specified xe_off entry in xattr bucket.
4662  * bucket is indicated by header_bh and len is the new length.
4663  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4664  *
4665  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4666  */
4667 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4668                                              struct ocfs2_xattr_bucket *bucket,
4669                                              int xe_off,
4670                                              int len,
4671                                              struct ocfs2_xattr_set_ctxt *ctxt)
4672 {
4673         int ret, offset;
4674         u64 value_blk;
4675         struct ocfs2_xattr_entry *xe;
4676         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4677         size_t blocksize = inode->i_sb->s_blocksize;
4678         struct ocfs2_xattr_value_buf vb = {
4679                 .vb_access = ocfs2_journal_access,
4680         };
4681
4682         xe = &xh->xh_entries[xe_off];
4683
4684         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4685
4686         offset = le16_to_cpu(xe->xe_name_offset) +
4687                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4688
4689         value_blk = offset / blocksize;
4690
4691         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4692         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4693
4694         vb.vb_bh = bucket->bu_bhs[value_blk];
4695         BUG_ON(!vb.vb_bh);
4696
4697         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4698                 (vb.vb_bh->b_data + offset % blocksize);
4699
4700         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4701                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4702         if (ret) {
4703                 mlog_errno(ret);
4704                 goto out;
4705         }
4706
4707         /*
4708          * From here on out we have to dirty the bucket.  The generic
4709          * value calls only modify one of the bucket's bhs, but we need
4710          * to send the bucket at once.  So if they error, they *could* have
4711          * modified something.  We have to assume they did, and dirty
4712          * the whole bucket.  This leaves us in a consistent state.
4713          */
4714         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4715              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4716         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4717         if (ret) {
4718                 mlog_errno(ret);
4719                 goto out_dirty;
4720         }
4721
4722         xe->xe_value_size = cpu_to_le64(len);
4723
4724 out_dirty:
4725         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4726
4727 out:
4728         return ret;
4729 }
4730
4731 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4732                                         struct ocfs2_xattr_search *xs,
4733                                         int len,
4734                                         struct ocfs2_xattr_set_ctxt *ctxt)
4735 {
4736         int ret, offset;
4737         struct ocfs2_xattr_entry *xe = xs->here;
4738         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4739
4740         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4741
4742         offset = xe - xh->xh_entries;
4743         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4744                                                 offset, len, ctxt);
4745         if (ret)
4746                 mlog_errno(ret);
4747
4748         return ret;
4749 }
4750
4751 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4752                                                 handle_t *handle,
4753                                                 struct ocfs2_xattr_search *xs,
4754                                                 char *val,
4755                                                 int value_len)
4756 {
4757         int offset;
4758         struct ocfs2_xattr_value_root *xv;
4759         struct ocfs2_xattr_entry *xe = xs->here;
4760
4761         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4762
4763         offset = le16_to_cpu(xe->xe_name_offset) +
4764                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4765
4766         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4767
4768         return __ocfs2_xattr_set_value_outside(inode, handle,
4769                                                xv, val, value_len);
4770 }
4771
4772 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4773                                   struct buffer_head *root_bh,
4774                                   u64 blkno,
4775                                   u32 cpos,
4776                                   u32 len)
4777 {
4778         int ret;
4779         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4780         struct inode *tl_inode = osb->osb_tl_inode;
4781         handle_t *handle;
4782         struct ocfs2_xattr_block *xb =
4783                         (struct ocfs2_xattr_block *)root_bh->b_data;
4784         struct ocfs2_alloc_context *meta_ac = NULL;
4785         struct ocfs2_cached_dealloc_ctxt dealloc;
4786         struct ocfs2_extent_tree et;
4787
4788         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4789
4790         ocfs2_init_dealloc_ctxt(&dealloc);
4791
4792         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4793              cpos, len, (unsigned long long)blkno);
4794
4795         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4796
4797         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4798         if (ret) {
4799                 mlog_errno(ret);
4800                 return ret;
4801         }
4802
4803         mutex_lock(&tl_inode->i_mutex);
4804
4805         if (ocfs2_truncate_log_needs_flush(osb)) {
4806                 ret = __ocfs2_flush_truncate_log(osb);
4807                 if (ret < 0) {
4808                         mlog_errno(ret);
4809                         goto out;
4810                 }
4811         }
4812
4813         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4814         if (IS_ERR(handle)) {
4815                 ret = -ENOMEM;
4816                 mlog_errno(ret);
4817                 goto out;
4818         }
4819
4820         ret = ocfs2_journal_access_xb(handle, inode, root_bh,
4821                                       OCFS2_JOURNAL_ACCESS_WRITE);
4822         if (ret) {
4823                 mlog_errno(ret);
4824                 goto out_commit;
4825         }
4826
4827         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4828                                   &dealloc);
4829         if (ret) {
4830                 mlog_errno(ret);
4831                 goto out_commit;
4832         }
4833
4834         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4835
4836         ret = ocfs2_journal_dirty(handle, root_bh);
4837         if (ret) {
4838                 mlog_errno(ret);
4839                 goto out_commit;
4840         }
4841
4842         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4843         if (ret)
4844                 mlog_errno(ret);
4845
4846 out_commit:
4847         ocfs2_commit_trans(osb, handle);
4848 out:
4849         ocfs2_schedule_truncate_log_flush(osb, 1);
4850
4851         mutex_unlock(&tl_inode->i_mutex);
4852
4853         if (meta_ac)
4854                 ocfs2_free_alloc_context(meta_ac);
4855
4856         ocfs2_run_deallocs(osb, &dealloc);
4857
4858         return ret;
4859 }
4860
4861 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4862                                          handle_t *handle,
4863                                          struct ocfs2_xattr_search *xs)
4864 {
4865         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4866         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4867                                                 le16_to_cpu(xh->xh_count) - 1];
4868         int ret = 0;
4869
4870         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4871                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4872         if (ret) {
4873                 mlog_errno(ret);
4874                 return;
4875         }
4876
4877         /* Remove the old entry. */
4878         memmove(xs->here, xs->here + 1,
4879                 (void *)last - (void *)xs->here);
4880         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4881         le16_add_cpu(&xh->xh_count, -1);
4882
4883         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4884 }
4885
4886 /*
4887  * Set the xattr name/value in the bucket specified in xs.
4888  *
4889  * As the new value in xi may be stored in the bucket or in an outside cluster,
4890  * we divide the whole process into 3 steps:
4891  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4892  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4893  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4894  * 4. If the clusters for the new outside value can't be allocated, we need
4895  *    to free the xattr we allocated in set.
4896  */
4897 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4898                                      struct ocfs2_xattr_info *xi,
4899                                      struct ocfs2_xattr_search *xs,
4900                                      struct ocfs2_xattr_set_ctxt *ctxt)
4901 {
4902         int ret, local = 1;
4903         size_t value_len;
4904         char *val = (char *)xi->value;
4905         struct ocfs2_xattr_entry *xe = xs->here;
4906         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4907                                               strlen(xi->name));
4908
4909         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4910                 /*
4911                  * We need to truncate the xattr storage first.
4912                  *
4913                  * If both the old and new value are stored to
4914                  * outside block, we only need to truncate
4915                  * the storage and then set the value outside.
4916                  *
4917                  * If the new value should be stored within block,
4918                  * we should free all the outside block first and
4919                  * the modification to the xattr block will be done
4920                  * by following steps.
4921                  */
4922                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4923                         value_len = xi->value_len;
4924                 else
4925                         value_len = 0;
4926
4927                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4928                                                            value_len,
4929                                                            ctxt);
4930                 if (ret)
4931                         goto out;
4932
4933                 if (value_len)
4934                         goto set_value_outside;
4935         }
4936
4937         value_len = xi->value_len;
4938         /* So we have to handle the inside block change now. */
4939         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4940                 /*
4941                  * If the new value will be stored outside of block,
4942                  * initalize a new empty value root and insert it first.
4943                  */
4944                 local = 0;
4945                 xi->value = &def_xv;
4946                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4947         }
4948
4949         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
4950                                               name_hash, local);
4951         if (ret) {
4952                 mlog_errno(ret);
4953                 goto out;
4954         }
4955
4956         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4957                 goto out;
4958
4959         /* allocate the space now for the outside block storage. */
4960         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4961                                                    value_len, ctxt);
4962         if (ret) {
4963                 mlog_errno(ret);
4964
4965                 if (xs->not_found) {
4966                         /*
4967                          * We can't allocate enough clusters for outside
4968                          * storage and we have allocated xattr already,
4969                          * so need to remove it.
4970                          */
4971                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
4972                 }
4973                 goto out;
4974         }
4975
4976 set_value_outside:
4977         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
4978                                                    xs, val, value_len);
4979 out:
4980         return ret;
4981 }
4982
4983 /*
4984  * check whether the xattr bucket is filled up with the same hash value.
4985  * If we want to insert the xattr with the same hash, return -ENOSPC.
4986  * If we want to insert a xattr with different hash value, go ahead
4987  * and ocfs2_divide_xattr_bucket will handle this.
4988  */
4989 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4990                                               struct ocfs2_xattr_bucket *bucket,
4991                                               const char *name)
4992 {
4993         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4994         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
4995
4996         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
4997                 return 0;
4998
4999         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5000             xh->xh_entries[0].xe_name_hash) {
5001                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5002                      "hash = %u\n",
5003                      (unsigned long long)bucket_blkno(bucket),
5004                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5005                 return -ENOSPC;
5006         }
5007
5008         return 0;
5009 }
5010
5011 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5012                                              struct ocfs2_xattr_info *xi,
5013                                              struct ocfs2_xattr_search *xs,
5014                                              struct ocfs2_xattr_set_ctxt *ctxt)
5015 {
5016         struct ocfs2_xattr_header *xh;
5017         struct ocfs2_xattr_entry *xe;
5018         u16 count, header_size, xh_free_start;
5019         int free, max_free, need, old;
5020         size_t value_size = 0, name_len = strlen(xi->name);
5021         size_t blocksize = inode->i_sb->s_blocksize;
5022         int ret, allocation = 0;
5023
5024         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5025
5026 try_again:
5027         xh = xs->header;
5028         count = le16_to_cpu(xh->xh_count);
5029         xh_free_start = le16_to_cpu(xh->xh_free_start);
5030         header_size = sizeof(struct ocfs2_xattr_header) +
5031                         count * sizeof(struct ocfs2_xattr_entry);
5032         max_free = OCFS2_XATTR_BUCKET_SIZE -
5033                 le16_to_cpu(xh->xh_name_value_len) - header_size;
5034
5035         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5036                         "of %u which exceed block size\n",
5037                         (unsigned long long)bucket_blkno(xs->bucket),
5038                         header_size);
5039
5040         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5041                 value_size = OCFS2_XATTR_ROOT_SIZE;
5042         else if (xi->value)
5043                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5044
5045         if (xs->not_found)
5046                 need = sizeof(struct ocfs2_xattr_entry) +
5047                         OCFS2_XATTR_SIZE(name_len) + value_size;
5048         else {
5049                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5050
5051                 /*
5052                  * We only replace the old value if the new length is smaller
5053                  * than the old one. Otherwise we will allocate new space in the
5054                  * bucket to store it.
5055                  */
5056                 xe = xs->here;
5057                 if (ocfs2_xattr_is_local(xe))
5058                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5059                 else
5060                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5061
5062                 if (old >= value_size)
5063                         need = 0;
5064         }
5065
5066         free = xh_free_start - header_size;
5067         /*
5068          * We need to make sure the new name/value pair
5069          * can exist in the same block.
5070          */
5071         if (xh_free_start % blocksize < need)
5072                 free -= xh_free_start % blocksize;
5073
5074         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5075              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5076              " %u\n", xs->not_found,
5077              (unsigned long long)bucket_blkno(xs->bucket),
5078              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5079              le16_to_cpu(xh->xh_name_value_len));
5080
5081         if (free < need ||
5082             (xs->not_found &&
5083              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5084                 if (need <= max_free &&
5085                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5086                         /*
5087                          * We can create the space by defragment. Since only the
5088                          * name/value will be moved, the xe shouldn't be changed
5089                          * in xs.
5090                          */
5091                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5092                                                         xs->bucket);
5093                         if (ret) {
5094                                 mlog_errno(ret);
5095                                 goto out;
5096                         }
5097
5098                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5099                         free = xh_free_start - header_size;
5100                         if (xh_free_start % blocksize < need)
5101                                 free -= xh_free_start % blocksize;
5102
5103                         if (free >= need)
5104                                 goto xattr_set;
5105
5106                         mlog(0, "Can't get enough space for xattr insert by "
5107                              "defragment. Need %u bytes, but we have %d, so "
5108                              "allocate new bucket for it.\n", need, free);
5109                 }
5110
5111                 /*
5112                  * We have to add new buckets or clusters and one
5113                  * allocation should leave us enough space for insert.
5114                  */
5115                 BUG_ON(allocation);
5116
5117                 /*
5118                  * We do not allow for overlapping ranges between buckets. And
5119                  * the maximum number of collisions we will allow for then is
5120                  * one bucket's worth, so check it here whether we need to
5121                  * add a new bucket for the insert.
5122                  */
5123                 ret = ocfs2_check_xattr_bucket_collision(inode,
5124                                                          xs->bucket,
5125                                                          xi->name);
5126                 if (ret) {
5127                         mlog_errno(ret);
5128                         goto out;
5129                 }
5130
5131                 ret = ocfs2_add_new_xattr_bucket(inode,
5132                                                  xs->xattr_bh,
5133                                                  xs->bucket,
5134                                                  ctxt);
5135                 if (ret) {
5136                         mlog_errno(ret);
5137                         goto out;
5138                 }
5139
5140                 /*
5141                  * ocfs2_add_new_xattr_bucket() will have updated
5142                  * xs->bucket if it moved, but it will not have updated
5143                  * any of the other search fields.  Thus, we drop it and
5144                  * re-search.  Everything should be cached, so it'll be
5145                  * quick.
5146                  */
5147                 ocfs2_xattr_bucket_relse(xs->bucket);
5148                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5149                                                    xi->name_index,
5150                                                    xi->name, xs);
5151                 if (ret && ret != -ENODATA)
5152                         goto out;
5153                 xs->not_found = ret;
5154                 allocation = 1;
5155                 goto try_again;
5156         }
5157
5158 xattr_set:
5159         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5160 out:
5161         mlog_exit(ret);
5162         return ret;
5163 }
5164
5165 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5166                                         struct ocfs2_xattr_bucket *bucket,
5167                                         void *para)
5168 {
5169         int ret = 0;
5170         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5171         u16 i;
5172         struct ocfs2_xattr_entry *xe;
5173         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5174         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5175         int credits = ocfs2_remove_extent_credits(osb->sb) +
5176                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5177
5178
5179         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5180
5181         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5182                 xe = &xh->xh_entries[i];
5183                 if (ocfs2_xattr_is_local(xe))
5184                         continue;
5185
5186                 ctxt.handle = ocfs2_start_trans(osb, credits);
5187                 if (IS_ERR(ctxt.handle)) {
5188                         ret = PTR_ERR(ctxt.handle);
5189                         mlog_errno(ret);
5190                         break;
5191                 }
5192
5193                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5194                                                         i, 0, &ctxt);
5195
5196                 ocfs2_commit_trans(osb, ctxt.handle);
5197                 if (ret) {
5198                         mlog_errno(ret);
5199                         break;
5200                 }
5201         }
5202
5203         ocfs2_schedule_truncate_log_flush(osb, 1);
5204         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5205         return ret;
5206 }
5207
5208 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5209                                           struct buffer_head *xb_bh)
5210 {
5211         struct ocfs2_xattr_block *xb =
5212                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5213         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5214         int ret = 0;
5215         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5216         u64 p_blkno;
5217
5218         if (le16_to_cpu(el->l_next_free_rec) == 0)
5219                 return 0;
5220
5221         while (name_hash > 0) {
5222                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5223                                           &e_cpos, &num_clusters, el);
5224                 if (ret) {
5225                         mlog_errno(ret);
5226                         goto out;
5227                 }
5228
5229                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5230                                                   ocfs2_delete_xattr_in_bucket,
5231                                                   NULL);
5232                 if (ret) {
5233                         mlog_errno(ret);
5234                         goto out;
5235                 }
5236
5237                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5238                                              p_blkno, e_cpos, num_clusters);
5239                 if (ret) {
5240                         mlog_errno(ret);
5241                         break;
5242                 }
5243
5244                 if (e_cpos == 0)
5245                         break;
5246
5247                 name_hash = e_cpos - 1;
5248         }
5249
5250 out:
5251         return ret;
5252 }
5253
5254 /*
5255  * 'security' attributes support
5256  */
5257 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5258                                         size_t list_size, const char *name,
5259                                         size_t name_len)
5260 {
5261         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5262         const size_t total_len = prefix_len + name_len + 1;
5263
5264         if (list && total_len <= list_size) {
5265                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5266                 memcpy(list + prefix_len, name, name_len);
5267                 list[prefix_len + name_len] = '\0';
5268         }
5269         return total_len;
5270 }
5271
5272 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5273                                     void *buffer, size_t size)
5274 {
5275         if (strcmp(name, "") == 0)
5276                 return -EINVAL;
5277         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5278                                buffer, size);
5279 }
5280
5281 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5282                                     const void *value, size_t size, int flags)
5283 {
5284         if (strcmp(name, "") == 0)
5285                 return -EINVAL;
5286
5287         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5288                                size, flags);
5289 }
5290
5291 int ocfs2_init_security_get(struct inode *inode,
5292                             struct inode *dir,
5293                             struct ocfs2_security_xattr_info *si)
5294 {
5295         return security_inode_init_security(inode, dir, &si->name, &si->value,
5296                                             &si->value_len);
5297 }
5298
5299 int ocfs2_init_security_set(handle_t *handle,
5300                             struct inode *inode,
5301                             struct buffer_head *di_bh,
5302                             struct ocfs2_security_xattr_info *si,
5303                             struct ocfs2_alloc_context *xattr_ac,
5304                             struct ocfs2_alloc_context *data_ac)
5305 {
5306         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5307                                      OCFS2_XATTR_INDEX_SECURITY,
5308                                      si->name, si->value, si->value_len, 0,
5309                                      xattr_ac, data_ac);
5310 }
5311
5312 struct xattr_handler ocfs2_xattr_security_handler = {
5313         .prefix = XATTR_SECURITY_PREFIX,
5314         .list   = ocfs2_xattr_security_list,
5315         .get    = ocfs2_xattr_security_get,
5316         .set    = ocfs2_xattr_security_set,
5317 };
5318
5319 /*
5320  * 'trusted' attributes support
5321  */
5322 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5323                                        size_t list_size, const char *name,
5324                                        size_t name_len)
5325 {
5326         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5327         const size_t total_len = prefix_len + name_len + 1;
5328
5329         if (list && total_len <= list_size) {
5330                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5331                 memcpy(list + prefix_len, name, name_len);
5332                 list[prefix_len + name_len] = '\0';
5333         }
5334         return total_len;
5335 }
5336
5337 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5338                                    void *buffer, size_t size)
5339 {
5340         if (strcmp(name, "") == 0)
5341                 return -EINVAL;
5342         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5343                                buffer, size);
5344 }
5345
5346 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5347                                    const void *value, size_t size, int flags)
5348 {
5349         if (strcmp(name, "") == 0)
5350                 return -EINVAL;
5351
5352         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5353                                size, flags);
5354 }
5355
5356 struct xattr_handler ocfs2_xattr_trusted_handler = {
5357         .prefix = XATTR_TRUSTED_PREFIX,
5358         .list   = ocfs2_xattr_trusted_list,
5359         .get    = ocfs2_xattr_trusted_get,
5360         .set    = ocfs2_xattr_trusted_set,
5361 };
5362
5363 /*
5364  * 'user' attributes support
5365  */
5366 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5367                                     size_t list_size, const char *name,
5368                                     size_t name_len)
5369 {
5370         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5371         const size_t total_len = prefix_len + name_len + 1;
5372         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5373
5374         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5375                 return 0;
5376
5377         if (list && total_len <= list_size) {
5378                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5379                 memcpy(list + prefix_len, name, name_len);
5380                 list[prefix_len + name_len] = '\0';
5381         }
5382         return total_len;
5383 }
5384
5385 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5386                                 void *buffer, size_t size)
5387 {
5388         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5389
5390         if (strcmp(name, "") == 0)
5391                 return -EINVAL;
5392         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5393                 return -EOPNOTSUPP;
5394         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5395                                buffer, size);
5396 }
5397
5398 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5399                                 const void *value, size_t size, int flags)
5400 {
5401         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5402
5403         if (strcmp(name, "") == 0)
5404                 return -EINVAL;
5405         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5406                 return -EOPNOTSUPP;
5407
5408         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5409                                size, flags);
5410 }
5411
5412 struct xattr_handler ocfs2_xattr_user_handler = {
5413         .prefix = XATTR_USER_PREFIX,
5414         .list   = ocfs2_xattr_user_list,
5415         .get    = ocfs2_xattr_user_get,
5416         .set    = ocfs2_xattr_user_set,
5417 };