fs/ocfs2/xattr.c

   1 /* -*- mode: c; c-basic-offset: 8; -*-
   2  * vim: noexpandtab sw=8 ts=8 sts=0:
   3  *
   4  * xattr.c
   5  *
   6  * Copyright (C) 2004, 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is copy from linux/fs/ext3/xattr.c.
  10  * Copyright (C) 2001-2003 Andreas Gruenbacher, <agruen@suse.de>
  11  *
  12  * This program is free software; you can redistribute it and/or
  13  * modify it under the terms of the GNU General Public
  14  * License version 2 as published by the Free Software Foundation.
  15  *
  16  * This program is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * General Public License for more details.
  20  */
  21
  22 #include <linux/capability.h>
  23 #include <linux/fs.h>
  24 #include <linux/types.h>
  25 #include <linux/slab.h>
  26 #include <linux/highmem.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/uio.h>
  29 #include <linux/sched.h>
  30 #include <linux/splice.h>
  31 #include <linux/mount.h>
  32 #include <linux/writeback.h>
  33 #include <linux/falloc.h>
  34 #include <linux/sort.h>
  35 #include <linux/init.h>
  36 #include <linux/module.h>
  37 #include <linux/string.h>
  38 #include <linux/security.h>
  39
  40 #define MLOG_MASK_PREFIX ML_XATTR
  41 #include <cluster/masklog.h>
  42
  43 #include "ocfs2.h"
  44 #include "alloc.h"
  45 #include "blockcheck.h"
  46 #include "dlmglue.h"
  47 #include "file.h"
  48 #include "symlink.h"
  49 #include "sysfile.h"
  50 #include "inode.h"
  51 #include "journal.h"
  52 #include "ocfs2_fs.h"
  53 #include "suballoc.h"
  54 #include "uptodate.h"
  55 #include "buffer_head_io.h"
  56 #include "super.h"
  57 #include "xattr.h"
  58 #include "refcounttree.h"
  59
  60 struct ocfs2_xattr_def_value_root {
  61         struct ocfs2_xattr_value_root   xv;
  62         struct ocfs2_extent_rec         er;
  63 };
  64
  65 struct ocfs2_xattr_bucket {
  66         /* The inode these xattrs are associated with */
  67         struct inode *bu_inode;
  68
  69         /* The actual buffers that make up the bucket */
  70         struct buffer_head *bu_bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  71
  72         /* How many blocks make up one bucket for this filesystem */
  73         int bu_blocks;
  74 };
  75
  76 struct ocfs2_xattr_set_ctxt {
  77         handle_t *handle;
  78         struct ocfs2_alloc_context *meta_ac;
  79         struct ocfs2_alloc_context *data_ac;
  80         struct ocfs2_cached_dealloc_ctxt dealloc;
  81 };
  82
  83 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  84 #define OCFS2_XATTR_INLINE_SIZE 80
  85 #define OCFS2_XATTR_HEADER_GAP  4
  86 #define OCFS2_XATTR_FREE_IN_IBODY       (OCFS2_MIN_XATTR_INLINE_SIZE \
  87                                          - sizeof(struct ocfs2_xattr_header) \
  88                                          - OCFS2_XATTR_HEADER_GAP)
  89 #define OCFS2_XATTR_FREE_IN_BLOCK(ptr)  ((ptr)->i_sb->s_blocksize \
  90                                          - sizeof(struct ocfs2_xattr_block) \
  91                                          - sizeof(struct ocfs2_xattr_header) \
  92                                          - OCFS2_XATTR_HEADER_GAP)
  93
  94 static struct ocfs2_xattr_def_value_root def_xv = {
  95         .xv.xr_list.l_count = cpu_to_le16(1),
  96 };
  97
  98 struct xattr_handler *ocfs2_xattr_handlers[] = {
  99         &ocfs2_xattr_user_handler,
 100 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 101         &ocfs2_xattr_acl_access_handler,
 102         &ocfs2_xattr_acl_default_handler,
 103 #endif
 104         &ocfs2_xattr_trusted_handler,
 105         &ocfs2_xattr_security_handler,
 106         NULL
 107 };
 108
 109 static struct xattr_handler *ocfs2_xattr_handler_map[OCFS2_XATTR_MAX] = {
 110         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
 111 #ifdef CONFIG_OCFS2_FS_POSIX_ACL
 112         [OCFS2_XATTR_INDEX_POSIX_ACL_ACCESS]
 113                                         = &ocfs2_xattr_acl_access_handler,
 114         [OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT]
 115                                         = &ocfs2_xattr_acl_default_handler,
 116 #endif
 117         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
 118         [OCFS2_XATTR_INDEX_SECURITY]    = &ocfs2_xattr_security_handler,
 119 };
 120
 121 struct ocfs2_xattr_info {
 122         int name_index;
 123         const char *name;
 124         const void *value;
 125         size_t value_len;
 126 };
 127
 128 struct ocfs2_xattr_search {
 129         struct buffer_head *inode_bh;
 130         /*
 131          * xattr_bh point to the block buffer head which has extended attribute
 132          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 133          */
 134         struct buffer_head *xattr_bh;
 135         struct ocfs2_xattr_header *header;
 136         struct ocfs2_xattr_bucket *bucket;
 137         void *base;
 138         void *end;
 139         struct ocfs2_xattr_entry *here;
 140         int not_found;
 141 };
 142
 143 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
 144                                              struct ocfs2_xattr_header *xh,
 145                                              int index,
 146                                              int *block_off,
 147                                              int *new_offset);
 148
 149 static int ocfs2_xattr_block_find(struct inode *inode,
 150                                   int name_index,
 151                                   const char *name,
 152                                   struct ocfs2_xattr_search *xs);
 153 static int ocfs2_xattr_index_block_find(struct inode *inode,
 154                                         struct buffer_head *root_bh,
 155                                         int name_index,
 156                                         const char *name,
 157                                         struct ocfs2_xattr_search *xs);
 158
 159 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 160                                         struct ocfs2_xattr_tree_root *xt,
 161                                         char *buffer,
 162                                         size_t buffer_size);
 163
 164 static int ocfs2_xattr_create_index_block(struct inode *inode,
 165                                           struct ocfs2_xattr_search *xs,
 166                                           struct ocfs2_xattr_set_ctxt *ctxt);
 167
 168 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 169                                              struct ocfs2_xattr_info *xi,
 170                                              struct ocfs2_xattr_search *xs,
 171                                              struct ocfs2_xattr_set_ctxt *ctxt);
 172
 173 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 174                                           struct buffer_head *xb_bh);
 175 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
 176                                   u64 src_blk, u64 last_blk, u64 to_blk,
 177                                   unsigned int start_bucket,
 178                                   u32 *first_hash);
 179 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
 180                                         struct ocfs2_dinode *di,
 181                                         struct ocfs2_xattr_info *xi,
 182                                         struct ocfs2_xattr_search *xis,
 183                                         struct ocfs2_xattr_search *xbs,
 184                                         struct ocfs2_refcount_tree **ref_tree,
 185                                         int *meta_need,
 186                                         int *credits);
 187
 188 static inline u16 ocfs2_xattr_buckets_per_cluster(struct ocfs2_super *osb)
 189 {
 190         return (1 << osb->s_clustersize_bits) / OCFS2_XATTR_BUCKET_SIZE;
 191 }
 192
 193 static inline u16 ocfs2_blocks_per_xattr_bucket(struct super_block *sb)
 194 {
 195         return OCFS2_XATTR_BUCKET_SIZE / (1 << sb->s_blocksize_bits);
 196 }
 197
 198 static inline u16 ocfs2_xattr_max_xe_in_bucket(struct super_block *sb)
 199 {
 200         u16 len = sb->s_blocksize -
 201                  offsetof(struct ocfs2_xattr_header, xh_entries);
 202
 203         return len / sizeof(struct ocfs2_xattr_entry);
 204 }
 205
 206 #define bucket_blkno(_b) ((_b)->bu_bhs[0]->b_blocknr)
 207 #define bucket_block(_b, _n) ((_b)->bu_bhs[(_n)]->b_data)
 208 #define bucket_xh(_b) ((struct ocfs2_xattr_header *)bucket_block((_b), 0))
 209
 210 static struct ocfs2_xattr_bucket *ocfs2_xattr_bucket_new(struct inode *inode)
 211 {
 212         struct ocfs2_xattr_bucket *bucket;
 213         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
 214
 215         BUG_ON(blks > OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET);
 216
 217         bucket = kzalloc(sizeof(struct ocfs2_xattr_bucket), GFP_NOFS);
 218         if (bucket) {
 219                 bucket->bu_inode = inode;
 220                 bucket->bu_blocks = blks;
 221         }
 222
 223         return bucket;
 224 }
 225
 226 static void ocfs2_xattr_bucket_relse(struct ocfs2_xattr_bucket *bucket)
 227 {
 228         int i;
 229
 230         for (i = 0; i < bucket->bu_blocks; i++) {
 231                 brelse(bucket->bu_bhs[i]);
 232                 bucket->bu_bhs[i] = NULL;
 233         }
 234 }
 235
 236 static void ocfs2_xattr_bucket_free(struct ocfs2_xattr_bucket *bucket)
 237 {
 238         if (bucket) {
 239                 ocfs2_xattr_bucket_relse(bucket);
 240                 bucket->bu_inode = NULL;
 241                 kfree(bucket);
 242         }
 243 }
 244
 245 /*
 246  * A bucket that has never been written to disk doesn't need to be
 247  * read.  We just need the buffer_heads.  Don't call this for
 248  * buckets that are already on disk.  ocfs2_read_xattr_bucket() initializes
 249  * them fully.
 250  */
 251 static int ocfs2_init_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 252                                    u64 xb_blkno)
 253 {
 254         int i, rc = 0;
 255
 256         for (i = 0; i < bucket->bu_blocks; i++) {
 257                 bucket->bu_bhs[i] = sb_getblk(bucket->bu_inode->i_sb,
 258                                               xb_blkno + i);
 259                 if (!bucket->bu_bhs[i]) {
 260                         rc = -EIO;
 261                         mlog_errno(rc);
 262                         break;
 263                 }
 264
 265                 if (!ocfs2_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 266                                            bucket->bu_bhs[i]))
 267                         ocfs2_set_new_buffer_uptodate(INODE_CACHE(bucket->bu_inode),
 268                                                       bucket->bu_bhs[i]);
 269         }
 270
 271         if (rc)
 272                 ocfs2_xattr_bucket_relse(bucket);
 273         return rc;
 274 }
 275
 276 /* Read the xattr bucket at xb_blkno */
 277 static int ocfs2_read_xattr_bucket(struct ocfs2_xattr_bucket *bucket,
 278                                    u64 xb_blkno)
 279 {
 280         int rc;
 281
 282         rc = ocfs2_read_blocks(INODE_CACHE(bucket->bu_inode), xb_blkno,
 283                                bucket->bu_blocks, bucket->bu_bhs, 0,
 284                                NULL);
 285         if (!rc) {
 286                 spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 287                 rc = ocfs2_validate_meta_ecc_bhs(bucket->bu_inode->i_sb,
 288                                                  bucket->bu_bhs,
 289                                                  bucket->bu_blocks,
 290                                                  &bucket_xh(bucket)->xh_check);
 291                 spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 292                 if (rc)
 293                         mlog_errno(rc);
 294         }
 295
 296         if (rc)
 297                 ocfs2_xattr_bucket_relse(bucket);
 298         return rc;
 299 }
 300
 301 static int ocfs2_xattr_bucket_journal_access(handle_t *handle,
 302                                              struct ocfs2_xattr_bucket *bucket,
 303                                              int type)
 304 {
 305         int i, rc = 0;
 306
 307         for (i = 0; i < bucket->bu_blocks; i++) {
 308                 rc = ocfs2_journal_access(handle,
 309                                           INODE_CACHE(bucket->bu_inode),
 310                                           bucket->bu_bhs[i], type);
 311                 if (rc) {
 312                         mlog_errno(rc);
 313                         break;
 314                 }
 315         }
 316
 317         return rc;
 318 }
 319
 320 static void ocfs2_xattr_bucket_journal_dirty(handle_t *handle,
 321                                              struct ocfs2_xattr_bucket *bucket)
 322 {
 323         int i;
 324
 325         spin_lock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 326         ocfs2_compute_meta_ecc_bhs(bucket->bu_inode->i_sb,
 327                                    bucket->bu_bhs, bucket->bu_blocks,
 328                                    &bucket_xh(bucket)->xh_check);
 329         spin_unlock(&OCFS2_SB(bucket->bu_inode->i_sb)->osb_xattr_lock);
 330
 331         for (i = 0; i < bucket->bu_blocks; i++)
 332                 ocfs2_journal_dirty(handle, bucket->bu_bhs[i]);
 333 }
 334
 335 static void ocfs2_xattr_bucket_copy_data(struct ocfs2_xattr_bucket *dest,
 336                                          struct ocfs2_xattr_bucket *src)
 337 {
 338         int i;
 339         int blocksize = src->bu_inode->i_sb->s_blocksize;
 340
 341         BUG_ON(dest->bu_blocks != src->bu_blocks);
 342         BUG_ON(dest->bu_inode != src->bu_inode);
 343
 344         for (i = 0; i < src->bu_blocks; i++) {
 345                 memcpy(bucket_block(dest, i), bucket_block(src, i),
 346                        blocksize);
 347         }
 348 }
 349
 350 static int ocfs2_validate_xattr_block(struct super_block *sb,
 351                                       struct buffer_head *bh)
 352 {
 353         int rc;
 354         struct ocfs2_xattr_block *xb =
 355                 (struct ocfs2_xattr_block *)bh->b_data;
 356
 357         mlog(0, "Validating xattr block %llu\n",
 358              (unsigned long long)bh->b_blocknr);
 359
 360         BUG_ON(!buffer_uptodate(bh));
 361
 362         /*
 363          * If the ecc fails, we return the error but otherwise
 364          * leave the filesystem running.  We know any error is
 365          * local to this block.
 366          */
 367         rc = ocfs2_validate_meta_ecc(sb, bh->b_data, &xb->xb_check);
 368         if (rc)
 369                 return rc;
 370
 371         /*
 372          * Errors after here are fatal
 373          */
 374
 375         if (!OCFS2_IS_VALID_XATTR_BLOCK(xb)) {
 376                 ocfs2_error(sb,
 377                             "Extended attribute block #%llu has bad "
 378                             "signature %.*s",
 379                             (unsigned long long)bh->b_blocknr, 7,
 380                             xb->xb_signature);
 381                 return -EINVAL;
 382         }
 383
 384         if (le64_to_cpu(xb->xb_blkno) != bh->b_blocknr) {
 385                 ocfs2_error(sb,
 386                             "Extended attribute block #%llu has an "
 387                             "invalid xb_blkno of %llu",
 388                             (unsigned long long)bh->b_blocknr,
 389                             (unsigned long long)le64_to_cpu(xb->xb_blkno));
 390                 return -EINVAL;
 391         }
 392
 393         if (le32_to_cpu(xb->xb_fs_generation) != OCFS2_SB(sb)->fs_generation) {
 394                 ocfs2_error(sb,
 395                             "Extended attribute block #%llu has an invalid "
 396                             "xb_fs_generation of #%u",
 397                             (unsigned long long)bh->b_blocknr,
 398                             le32_to_cpu(xb->xb_fs_generation));
 399                 return -EINVAL;
 400         }
 401
 402         return 0;
 403 }
 404
 405 static int ocfs2_read_xattr_block(struct inode *inode, u64 xb_blkno,
 406                                   struct buffer_head **bh)
 407 {
 408         int rc;
 409         struct buffer_head *tmp = *bh;
 410
 411         rc = ocfs2_read_block(INODE_CACHE(inode), xb_blkno, &tmp,
 412                               ocfs2_validate_xattr_block);
 413
 414         /* If ocfs2_read_block() got us a new bh, pass it up. */
 415         if (!rc && !*bh)
 416                 *bh = tmp;
 417
 418         return rc;
 419 }
 420
 421 static inline const char *ocfs2_xattr_prefix(int name_index)
 422 {
 423         struct xattr_handler *handler = NULL;
 424
 425         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 426                 handler = ocfs2_xattr_handler_map[name_index];
 427
 428         return handler ? handler->prefix : NULL;
 429 }
 430
 431 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 432                                  const char *name,
 433                                  int name_len)
 434 {
 435         /* Get hash value of uuid from super block */
 436         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 437         int i;
 438
 439         /* hash extended attribute name */
 440         for (i = 0; i < name_len; i++) {
 441                 hash = (hash << OCFS2_HASH_SHIFT) ^
 442                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 443                        *name++;
 444         }
 445
 446         return hash;
 447 }
 448
 449 /*
 450  * ocfs2_xattr_hash_entry()
 451  *
 452  * Compute the hash of an extended attribute.
 453  */
 454 static void ocfs2_xattr_hash_entry(struct inode *inode,
 455                                    struct ocfs2_xattr_header *header,
 456                                    struct ocfs2_xattr_entry *entry)
 457 {
 458         u32 hash = 0;
 459         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 460
 461         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
 462         entry->xe_name_hash = cpu_to_le32(hash);
 463
 464         return;
 465 }
 466
 467 static int ocfs2_xattr_entry_real_size(int name_len, size_t value_len)
 468 {
 469         int size = 0;
 470
 471         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
 472                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(value_len);
 473         else
 474                 size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 475         size += sizeof(struct ocfs2_xattr_entry);
 476
 477         return size;
 478 }
 479
 480 int ocfs2_calc_security_init(struct inode *dir,
 481                              struct ocfs2_security_xattr_info *si,
 482                              int *want_clusters,
 483                              int *xattr_credits,
 484                              struct ocfs2_alloc_context **xattr_ac)
 485 {
 486         int ret = 0;
 487         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 488         int s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 489                                                  si->value_len);
 490
 491         /*
 492          * The max space of security xattr taken inline is
 493          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 494          * So reserve one metadata block for it is ok.
 495          */
 496         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 497             s_size > OCFS2_XATTR_FREE_IN_IBODY) {
 498                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, xattr_ac);
 499                 if (ret) {
 500                         mlog_errno(ret);
 501                         return ret;
 502                 }
 503                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 504         }
 505
 506         /* reserve clusters for xattr value which will be set in B tree*/
 507         if (si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 508                 int new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 509                                                             si->value_len);
 510
 511                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 512                                                            new_clusters);
 513                 *want_clusters += new_clusters;
 514         }
 515         return ret;
 516 }
 517
 518 int ocfs2_calc_xattr_init(struct inode *dir,
 519                           struct buffer_head *dir_bh,
 520                           int mode,
 521                           struct ocfs2_security_xattr_info *si,
 522                           int *want_clusters,
 523                           int *xattr_credits,
 524                           int *want_meta)
 525 {
 526         int ret = 0;
 527         struct ocfs2_super *osb = OCFS2_SB(dir->i_sb);
 528         int s_size = 0, a_size = 0, acl_len = 0, new_clusters;
 529
 530         if (si->enable)
 531                 s_size = ocfs2_xattr_entry_real_size(strlen(si->name),
 532                                                      si->value_len);
 533
 534         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL) {
 535                 acl_len = ocfs2_xattr_get_nolock(dir, dir_bh,
 536                                         OCFS2_XATTR_INDEX_POSIX_ACL_DEFAULT,
 537                                         "", NULL, 0);
 538                 if (acl_len > 0) {
 539                         a_size = ocfs2_xattr_entry_real_size(0, acl_len);
 540                         if (S_ISDIR(mode))
 541                                 a_size <<= 1;
 542                 } else if (acl_len != 0 && acl_len != -ENODATA) {
 543                         mlog_errno(ret);
 544                         return ret;
 545                 }
 546         }
 547
 548         if (!(s_size + a_size))
 549                 return ret;
 550
 551         /*
 552          * The max space of security xattr taken inline is
 553          * 256(name) + 80(value) + 16(entry) = 352 bytes,
 554          * The max space of acl xattr taken inline is
 555          * 80(value) + 16(entry) * 2(if directory) = 192 bytes,
 556          * when blocksize = 512, may reserve one more cluser for
 557          * xattr bucket, otherwise reserve one metadata block
 558          * for them is ok.
 559          * If this is a new directory with inline data,
 560          * we choose to reserve the entire inline area for
 561          * directory contents and force an external xattr block.
 562          */
 563         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE ||
 564             (S_ISDIR(mode) && ocfs2_supports_inline_data(osb)) ||
 565             (s_size + a_size) > OCFS2_XATTR_FREE_IN_IBODY) {
 566                 *want_meta = *want_meta + 1;
 567                 *xattr_credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
 568         }
 569
 570         if (dir->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE &&
 571             (s_size + a_size) > OCFS2_XATTR_FREE_IN_BLOCK(dir)) {
 572                 *want_clusters += 1;
 573                 *xattr_credits += ocfs2_blocks_per_xattr_bucket(dir->i_sb);
 574         }
 575
 576         /*
 577          * reserve credits and clusters for xattrs which has large value
 578          * and have to be set outside
 579          */
 580         if (si->enable && si->value_len > OCFS2_XATTR_INLINE_SIZE) {
 581                 new_clusters = ocfs2_clusters_for_bytes(dir->i_sb,
 582                                                         si->value_len);
 583                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 584                                                            new_clusters);
 585                 *want_clusters += new_clusters;
 586         }
 587         if (osb->s_mount_opt & OCFS2_MOUNT_POSIX_ACL &&
 588             acl_len > OCFS2_XATTR_INLINE_SIZE) {
 589                 /* for directory, it has DEFAULT and ACCESS two types of acls */
 590                 new_clusters = (S_ISDIR(mode) ? 2 : 1) *
 591                                 ocfs2_clusters_for_bytes(dir->i_sb, acl_len);
 592                 *xattr_credits += ocfs2_clusters_to_blocks(dir->i_sb,
 593                                                            new_clusters);
 594                 *want_clusters += new_clusters;
 595         }
 596
 597         return ret;
 598 }
 599
 600 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 601                                          u32 clusters_to_add,
 602                                          struct ocfs2_xattr_value_buf *vb,
 603                                          struct ocfs2_xattr_set_ctxt *ctxt)
 604 {
 605         int status = 0;
 606         handle_t *handle = ctxt->handle;
 607         enum ocfs2_alloc_restarted why;
 608         u32 prev_clusters, logical_start = le32_to_cpu(vb->vb_xv->xr_clusters);
 609         struct ocfs2_extent_tree et;
 610
 611         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 612
 613         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 614
 615         status = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 616                               OCFS2_JOURNAL_ACCESS_WRITE);
 617         if (status < 0) {
 618                 mlog_errno(status);
 619                 goto leave;
 620         }
 621
 622         prev_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 623         status = ocfs2_add_clusters_in_btree(handle,
 624                                              &et,
 625                                              &logical_start,
 626                                              clusters_to_add,
 627                                              0,
 628                                              ctxt->data_ac,
 629                                              ctxt->meta_ac,
 630                                              &why);
 631         if (status < 0) {
 632                 mlog_errno(status);
 633                 goto leave;
 634         }
 635
 636         status = ocfs2_journal_dirty(handle, vb->vb_bh);
 637         if (status < 0) {
 638                 mlog_errno(status);
 639                 goto leave;
 640         }
 641
 642         clusters_to_add -= le32_to_cpu(vb->vb_xv->xr_clusters) - prev_clusters;
 643
 644         /*
 645          * We should have already allocated enough space before the transaction,
 646          * so no need to restart.
 647          */
 648         BUG_ON(why != RESTART_NONE || clusters_to_add);
 649
 650 leave:
 651
 652         return status;
 653 }
 654
 655 static int __ocfs2_remove_xattr_range(struct inode *inode,
 656                                       struct ocfs2_xattr_value_buf *vb,
 657                                       u32 cpos, u32 phys_cpos, u32 len,
 658                                       unsigned int ext_flags,
 659                                       struct ocfs2_xattr_set_ctxt *ctxt)
 660 {
 661         int ret;
 662         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 663         handle_t *handle = ctxt->handle;
 664         struct ocfs2_extent_tree et;
 665
 666         ocfs2_init_xattr_value_extent_tree(&et, INODE_CACHE(inode), vb);
 667
 668         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
 669                             OCFS2_JOURNAL_ACCESS_WRITE);
 670         if (ret) {
 671                 mlog_errno(ret);
 672                 goto out;
 673         }
 674
 675         ret = ocfs2_remove_extent(handle, &et, cpos, len, ctxt->meta_ac,
 676                                   &ctxt->dealloc);
 677         if (ret) {
 678                 mlog_errno(ret);
 679                 goto out;
 680         }
 681
 682         le32_add_cpu(&vb->vb_xv->xr_clusters, -len);
 683
 684         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
 685         if (ret) {
 686                 mlog_errno(ret);
 687                 goto out;
 688         }
 689
 690         if (ext_flags & OCFS2_EXT_REFCOUNTED)
 691                 ret = ocfs2_decrease_refcount(inode, handle,
 692                                         ocfs2_blocks_to_clusters(inode->i_sb,
 693                                                                  phys_blkno),
 694                                         len, ctxt->meta_ac, &ctxt->dealloc, 1);
 695         else
 696                 ret = ocfs2_cache_cluster_dealloc(&ctxt->dealloc,
 697                                                   phys_blkno, len);
 698         if (ret)
 699                 mlog_errno(ret);
 700
 701 out:
 702         return ret;
 703 }
 704
 705 static int ocfs2_xattr_shrink_size(struct inode *inode,
 706                                    u32 old_clusters,
 707                                    u32 new_clusters,
 708                                    struct ocfs2_xattr_value_buf *vb,
 709                                    struct ocfs2_xattr_set_ctxt *ctxt)
 710 {
 711         int ret = 0;
 712         unsigned int ext_flags;
 713         u32 trunc_len, cpos, phys_cpos, alloc_size;
 714         u64 block;
 715
 716         if (old_clusters <= new_clusters)
 717                 return 0;
 718
 719         cpos = new_clusters;
 720         trunc_len = old_clusters - new_clusters;
 721         while (trunc_len) {
 722                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 723                                                &alloc_size,
 724                                                &vb->vb_xv->xr_list, &ext_flags);
 725                 if (ret) {
 726                         mlog_errno(ret);
 727                         goto out;
 728                 }
 729
 730                 if (alloc_size > trunc_len)
 731                         alloc_size = trunc_len;
 732
 733                 ret = __ocfs2_remove_xattr_range(inode, vb, cpos,
 734                                                  phys_cpos, alloc_size,
 735                                                  ext_flags, ctxt);
 736                 if (ret) {
 737                         mlog_errno(ret);
 738                         goto out;
 739                 }
 740
 741                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 742                 ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode),
 743                                                        block, alloc_size);
 744                 cpos += alloc_size;
 745                 trunc_len -= alloc_size;
 746         }
 747
 748 out:
 749         return ret;
 750 }
 751
 752 static int ocfs2_xattr_value_truncate(struct inode *inode,
 753                                       struct ocfs2_xattr_value_buf *vb,
 754                                       int len,
 755                                       struct ocfs2_xattr_set_ctxt *ctxt)
 756 {
 757         int ret;
 758         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 759         u32 old_clusters = le32_to_cpu(vb->vb_xv->xr_clusters);
 760
 761         if (new_clusters == old_clusters)
 762                 return 0;
 763
 764         if (new_clusters > old_clusters)
 765                 ret = ocfs2_xattr_extend_allocation(inode,
 766                                                     new_clusters - old_clusters,
 767                                                     vb, ctxt);
 768         else
 769                 ret = ocfs2_xattr_shrink_size(inode,
 770                                               old_clusters, new_clusters,
 771                                               vb, ctxt);
 772
 773         return ret;
 774 }
 775
 776 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 777                                   size_t *result, const char *prefix,
 778                                   const char *name, int name_len)
 779 {
 780         char *p = buffer + *result;
 781         int prefix_len = strlen(prefix);
 782         int total_len = prefix_len + name_len + 1;
 783
 784         *result += total_len;
 785
 786         /* we are just looking for how big our buffer needs to be */
 787         if (!size)
 788                 return 0;
 789
 790         if (*result > size)
 791                 return -ERANGE;
 792
 793         memcpy(p, prefix, prefix_len);
 794         memcpy(p + prefix_len, name, name_len);
 795         p[prefix_len + name_len] = '\0';
 796
 797         return 0;
 798 }
 799
 800 static int ocfs2_xattr_list_entries(struct inode *inode,
 801                                     struct ocfs2_xattr_header *header,
 802                                     char *buffer, size_t buffer_size)
 803 {
 804         size_t result = 0;
 805         int i, type, ret;
 806         const char *prefix, *name;
 807
 808         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 809                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 810                 type = ocfs2_xattr_get_type(entry);
 811                 prefix = ocfs2_xattr_prefix(type);
 812
 813                 if (prefix) {
 814                         name = (const char *)header +
 815                                 le16_to_cpu(entry->xe_name_offset);
 816
 817                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 818                                                      &result, prefix, name,
 819                                                      entry->xe_name_len);
 820                         if (ret)
 821                                 return ret;
 822                 }
 823         }
 824
 825         return result;
 826 }
 827
 828 static int ocfs2_xattr_ibody_list(struct inode *inode,
 829                                   struct ocfs2_dinode *di,
 830                                   char *buffer,
 831                                   size_t buffer_size)
 832 {
 833         struct ocfs2_xattr_header *header = NULL;
 834         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 835         int ret = 0;
 836
 837         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 838                 return ret;
 839
 840         header = (struct ocfs2_xattr_header *)
 841                  ((void *)di + inode->i_sb->s_blocksize -
 842                  le16_to_cpu(di->i_xattr_inline_size));
 843
 844         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 845
 846         return ret;
 847 }
 848
 849 static int ocfs2_xattr_block_list(struct inode *inode,
 850                                   struct ocfs2_dinode *di,
 851                                   char *buffer,
 852                                   size_t buffer_size)
 853 {
 854         struct buffer_head *blk_bh = NULL;
 855         struct ocfs2_xattr_block *xb;
 856         int ret = 0;
 857
 858         if (!di->i_xattr_loc)
 859                 return ret;
 860
 861         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
 862                                      &blk_bh);
 863         if (ret < 0) {
 864                 mlog_errno(ret);
 865                 return ret;
 866         }
 867
 868         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 869         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 870                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 871                 ret = ocfs2_xattr_list_entries(inode, header,
 872                                                buffer, buffer_size);
 873         } else {
 874                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 875                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 876                                                    buffer, buffer_size);
 877         }
 878
 879         brelse(blk_bh);
 880
 881         return ret;
 882 }
 883
 884 ssize_t ocfs2_listxattr(struct dentry *dentry,
 885                         char *buffer,
 886                         size_t size)
 887 {
 888         int ret = 0, i_ret = 0, b_ret = 0;
 889         struct buffer_head *di_bh = NULL;
 890         struct ocfs2_dinode *di = NULL;
 891         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 892
 893         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 894                 return -EOPNOTSUPP;
 895
 896         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 897                 return ret;
 898
 899         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 900         if (ret < 0) {
 901                 mlog_errno(ret);
 902                 return ret;
 903         }
 904
 905         di = (struct ocfs2_dinode *)di_bh->b_data;
 906
 907         down_read(&oi->ip_xattr_sem);
 908         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 909         if (i_ret < 0)
 910                 b_ret = 0;
 911         else {
 912                 if (buffer) {
 913                         buffer += i_ret;
 914                         size -= i_ret;
 915                 }
 916                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 917                                                buffer, size);
 918                 if (b_ret < 0)
 919                         i_ret = 0;
 920         }
 921         up_read(&oi->ip_xattr_sem);
 922         ocfs2_inode_unlock(dentry->d_inode, 0);
 923
 924         brelse(di_bh);
 925
 926         return i_ret + b_ret;
 927 }
 928
 929 static int ocfs2_xattr_find_entry(int name_index,
 930                                   const char *name,
 931                                   struct ocfs2_xattr_search *xs)
 932 {
 933         struct ocfs2_xattr_entry *entry;
 934         size_t name_len;
 935         int i, cmp = 1;
 936
 937         if (name == NULL)
 938                 return -EINVAL;
 939
 940         name_len = strlen(name);
 941         entry = xs->here;
 942         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 943                 cmp = name_index - ocfs2_xattr_get_type(entry);
 944                 if (!cmp)
 945                         cmp = name_len - entry->xe_name_len;
 946                 if (!cmp)
 947                         cmp = memcmp(name, (xs->base +
 948                                      le16_to_cpu(entry->xe_name_offset)),
 949                                      name_len);
 950                 if (cmp == 0)
 951                         break;
 952                 entry += 1;
 953         }
 954         xs->here = entry;
 955
 956         return cmp ? -ENODATA : 0;
 957 }
 958
 959 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 960                                          struct ocfs2_xattr_value_root *xv,
 961                                          void *buffer,
 962                                          size_t len)
 963 {
 964         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 965         u64 blkno;
 966         int i, ret = 0;
 967         size_t cplen, blocksize;
 968         struct buffer_head *bh = NULL;
 969         struct ocfs2_extent_list *el;
 970
 971         el = &xv->xr_list;
 972         clusters = le32_to_cpu(xv->xr_clusters);
 973         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 974         blocksize = inode->i_sb->s_blocksize;
 975
 976         cpos = 0;
 977         while (cpos < clusters) {
 978                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 979                                                &num_clusters, el, NULL);
 980                 if (ret) {
 981                         mlog_errno(ret);
 982                         goto out;
 983                 }
 984
 985                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 986                 /* Copy ocfs2_xattr_value */
 987                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 988                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
 989                                                &bh, NULL);
 990                         if (ret) {
 991                                 mlog_errno(ret);
 992                                 goto out;
 993                         }
 994
 995                         cplen = len >= blocksize ? blocksize : len;
 996                         memcpy(buffer, bh->b_data, cplen);
 997                         len -= cplen;
 998                         buffer += cplen;
 999
1000                         brelse(bh);
1001                         bh = NULL;
1002                         if (len == 0)
1003                                 break;
1004                 }
1005                 cpos += num_clusters;
1006         }
1007 out:
1008         return ret;
1009 }
1010
1011 static int ocfs2_xattr_ibody_get(struct inode *inode,
1012                                  int name_index,
1013                                  const char *name,
1014                                  void *buffer,
1015                                  size_t buffer_size,
1016                                  struct ocfs2_xattr_search *xs)
1017 {
1018         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1019         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1020         struct ocfs2_xattr_value_root *xv;
1021         size_t size;
1022         int ret = 0;
1023
1024         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
1025                 return -ENODATA;
1026
1027         xs->end = (void *)di + inode->i_sb->s_blocksize;
1028         xs->header = (struct ocfs2_xattr_header *)
1029                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1030         xs->base = (void *)xs->header;
1031         xs->here = xs->header->xh_entries;
1032
1033         ret = ocfs2_xattr_find_entry(name_index, name, xs);
1034         if (ret)
1035                 return ret;
1036         size = le64_to_cpu(xs->here->xe_value_size);
1037         if (buffer) {
1038                 if (size > buffer_size)
1039                         return -ERANGE;
1040                 if (ocfs2_xattr_is_local(xs->here)) {
1041                         memcpy(buffer, (void *)xs->base +
1042                                le16_to_cpu(xs->here->xe_name_offset) +
1043                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
1044                 } else {
1045                         xv = (struct ocfs2_xattr_value_root *)
1046                                 (xs->base + le16_to_cpu(
1047                                  xs->here->xe_name_offset) +
1048                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
1049                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1050                                                             buffer, size);
1051                         if (ret < 0) {
1052                                 mlog_errno(ret);
1053                                 return ret;
1054                         }
1055                 }
1056         }
1057
1058         return size;
1059 }
1060
1061 static int ocfs2_xattr_block_get(struct inode *inode,
1062                                  int name_index,
1063                                  const char *name,
1064                                  void *buffer,
1065                                  size_t buffer_size,
1066                                  struct ocfs2_xattr_search *xs)
1067 {
1068         struct ocfs2_xattr_block *xb;
1069         struct ocfs2_xattr_value_root *xv;
1070         size_t size;
1071         int ret = -ENODATA, name_offset, name_len, i;
1072         int uninitialized_var(block_off);
1073
1074         xs->bucket = ocfs2_xattr_bucket_new(inode);
1075         if (!xs->bucket) {
1076                 ret = -ENOMEM;
1077                 mlog_errno(ret);
1078                 goto cleanup;
1079         }
1080
1081         ret = ocfs2_xattr_block_find(inode, name_index, name, xs);
1082         if (ret) {
1083                 mlog_errno(ret);
1084                 goto cleanup;
1085         }
1086
1087         if (xs->not_found) {
1088                 ret = -ENODATA;
1089                 goto cleanup;
1090         }
1091
1092         xb = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1093         size = le64_to_cpu(xs->here->xe_value_size);
1094         if (buffer) {
1095                 ret = -ERANGE;
1096                 if (size > buffer_size)
1097                         goto cleanup;
1098
1099                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
1100                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
1101                 i = xs->here - xs->header->xh_entries;
1102
1103                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
1104                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
1105                                                                 bucket_xh(xs->bucket),
1106                                                                 i,
1107                                                                 &block_off,
1108                                                                 &name_offset);
1109                         xs->base = bucket_block(xs->bucket, block_off);
1110                 }
1111                 if (ocfs2_xattr_is_local(xs->here)) {
1112                         memcpy(buffer, (void *)xs->base +
1113                                name_offset + name_len, size);
1114                 } else {
1115                         xv = (struct ocfs2_xattr_value_root *)
1116                                 (xs->base + name_offset + name_len);
1117                         ret = ocfs2_xattr_get_value_outside(inode, xv,
1118                                                             buffer, size);
1119                         if (ret < 0) {
1120                                 mlog_errno(ret);
1121                                 goto cleanup;
1122                         }
1123                 }
1124         }
1125         ret = size;
1126 cleanup:
1127         ocfs2_xattr_bucket_free(xs->bucket);
1128
1129         brelse(xs->xattr_bh);
1130         xs->xattr_bh = NULL;
1131         return ret;
1132 }
1133
1134 int ocfs2_xattr_get_nolock(struct inode *inode,
1135                            struct buffer_head *di_bh,
1136                            int name_index,
1137                            const char *name,
1138                            void *buffer,
1139                            size_t buffer_size)
1140 {
1141         int ret;
1142         struct ocfs2_dinode *di = NULL;
1143         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1144         struct ocfs2_xattr_search xis = {
1145                 .not_found = -ENODATA,
1146         };
1147         struct ocfs2_xattr_search xbs = {
1148                 .not_found = -ENODATA,
1149         };
1150
1151         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1152                 return -EOPNOTSUPP;
1153
1154         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1155                 ret = -ENODATA;
1156
1157         xis.inode_bh = xbs.inode_bh = di_bh;
1158         di = (struct ocfs2_dinode *)di_bh->b_data;
1159
1160         down_read(&oi->ip_xattr_sem);
1161         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
1162                                     buffer_size, &xis);
1163         if (ret == -ENODATA && di->i_xattr_loc)
1164                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
1165                                             buffer_size, &xbs);
1166         up_read(&oi->ip_xattr_sem);
1167
1168         return ret;
1169 }
1170
1171 /* ocfs2_xattr_get()
1172  *
1173  * Copy an extended attribute into the buffer provided.
1174  * Buffer is NULL to compute the size of buffer required.
1175  */
1176 static int ocfs2_xattr_get(struct inode *inode,
1177                            int name_index,
1178                            const char *name,
1179                            void *buffer,
1180                            size_t buffer_size)
1181 {
1182         int ret;
1183         struct buffer_head *di_bh = NULL;
1184
1185         ret = ocfs2_inode_lock(inode, &di_bh, 0);
1186         if (ret < 0) {
1187                 mlog_errno(ret);
1188                 return ret;
1189         }
1190         ret = ocfs2_xattr_get_nolock(inode, di_bh, name_index,
1191                                      name, buffer, buffer_size);
1192
1193         ocfs2_inode_unlock(inode, 0);
1194
1195         brelse(di_bh);
1196
1197         return ret;
1198 }
1199
1200 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
1201                                            handle_t *handle,
1202                                            struct ocfs2_xattr_value_buf *vb,
1203                                            const void *value,
1204                                            int value_len)
1205 {
1206         int ret = 0, i, cp_len;
1207         u16 blocksize = inode->i_sb->s_blocksize;
1208         u32 p_cluster, num_clusters;
1209         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
1210         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
1211         u64 blkno;
1212         struct buffer_head *bh = NULL;
1213         unsigned int ext_flags;
1214         struct ocfs2_xattr_value_root *xv = vb->vb_xv;
1215
1216         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
1217
1218         while (cpos < clusters) {
1219                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
1220                                                &num_clusters, &xv->xr_list,
1221                                                &ext_flags);
1222                 if (ret) {
1223                         mlog_errno(ret);
1224                         goto out;
1225                 }
1226
1227                 BUG_ON(ext_flags & OCFS2_EXT_REFCOUNTED);
1228
1229                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
1230
1231                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
1232                         ret = ocfs2_read_block(INODE_CACHE(inode), blkno,
1233                                                &bh, NULL);
1234                         if (ret) {
1235                                 mlog_errno(ret);
1236                                 goto out;
1237                         }
1238
1239                         ret = ocfs2_journal_access(handle,
1240                                                    INODE_CACHE(inode),
1241                                                    bh,
1242                                                    OCFS2_JOURNAL_ACCESS_WRITE);
1243                         if (ret < 0) {
1244                                 mlog_errno(ret);
1245                                 goto out;
1246                         }
1247
1248                         cp_len = value_len > blocksize ? blocksize : value_len;
1249                         memcpy(bh->b_data, value, cp_len);
1250                         value_len -= cp_len;
1251                         value += cp_len;
1252                         if (cp_len < blocksize)
1253                                 memset(bh->b_data + cp_len, 0,
1254                                        blocksize - cp_len);
1255
1256                         ret = ocfs2_journal_dirty(handle, bh);
1257                         if (ret < 0) {
1258                                 mlog_errno(ret);
1259                                 goto out;
1260                         }
1261                         brelse(bh);
1262                         bh = NULL;
1263
1264                         /*
1265                          * XXX: do we need to empty all the following
1266                          * blocks in this cluster?
1267                          */
1268                         if (!value_len)
1269                                 break;
1270                 }
1271                 cpos += num_clusters;
1272         }
1273 out:
1274         brelse(bh);
1275
1276         return ret;
1277 }
1278
1279 static int ocfs2_xattr_cleanup(struct inode *inode,
1280                                handle_t *handle,
1281                                struct ocfs2_xattr_info *xi,
1282                                struct ocfs2_xattr_search *xs,
1283                                struct ocfs2_xattr_value_buf *vb,
1284                                size_t offs)
1285 {
1286         int ret = 0;
1287         size_t name_len = strlen(xi->name);
1288         void *val = xs->base + offs;
1289         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1290
1291         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1292                             OCFS2_JOURNAL_ACCESS_WRITE);
1293         if (ret) {
1294                 mlog_errno(ret);
1295                 goto out;
1296         }
1297         /* Decrease xattr count */
1298         le16_add_cpu(&xs->header->xh_count, -1);
1299         /* Remove the xattr entry and tree root which has already be set*/
1300         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1301         memset(val, 0, size);
1302
1303         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1304         if (ret < 0)
1305                 mlog_errno(ret);
1306 out:
1307         return ret;
1308 }
1309
1310 static int ocfs2_xattr_update_entry(struct inode *inode,
1311                                     handle_t *handle,
1312                                     struct ocfs2_xattr_info *xi,
1313                                     struct ocfs2_xattr_search *xs,
1314                                     struct ocfs2_xattr_value_buf *vb,
1315                                     size_t offs)
1316 {
1317         int ret;
1318
1319         ret = vb->vb_access(handle, INODE_CACHE(inode), vb->vb_bh,
1320                             OCFS2_JOURNAL_ACCESS_WRITE);
1321         if (ret) {
1322                 mlog_errno(ret);
1323                 goto out;
1324         }
1325
1326         xs->here->xe_name_offset = cpu_to_le16(offs);
1327         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1328         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1329                 ocfs2_xattr_set_local(xs->here, 1);
1330         else
1331                 ocfs2_xattr_set_local(xs->here, 0);
1332         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1333
1334         ret = ocfs2_journal_dirty(handle, vb->vb_bh);
1335         if (ret < 0)
1336                 mlog_errno(ret);
1337 out:
1338         return ret;
1339 }
1340
1341 /*
1342  * ocfs2_xattr_set_value_outside()
1343  *
1344  * Set large size value in B tree.
1345  */
1346 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1347                                          struct ocfs2_xattr_info *xi,
1348                                          struct ocfs2_xattr_search *xs,
1349                                          struct ocfs2_xattr_set_ctxt *ctxt,
1350                                          struct ocfs2_xattr_value_buf *vb,
1351                                          size_t offs)
1352 {
1353         size_t name_len = strlen(xi->name);
1354         void *val = xs->base + offs;
1355         struct ocfs2_xattr_value_root *xv = NULL;
1356         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1357         int ret = 0;
1358
1359         memset(val, 0, size);
1360         memcpy(val, xi->name, name_len);
1361         xv = (struct ocfs2_xattr_value_root *)
1362                 (val + OCFS2_XATTR_SIZE(name_len));
1363         xv->xr_clusters = 0;
1364         xv->xr_last_eb_blk = 0;
1365         xv->xr_list.l_tree_depth = 0;
1366         xv->xr_list.l_count = cpu_to_le16(1);
1367         xv->xr_list.l_next_free_rec = 0;
1368         vb->vb_xv = xv;
1369
1370         ret = ocfs2_xattr_value_truncate(inode, vb, xi->value_len, ctxt);
1371         if (ret < 0) {
1372                 mlog_errno(ret);
1373                 return ret;
1374         }
1375         ret = ocfs2_xattr_update_entry(inode, ctxt->handle, xi, xs, vb, offs);
1376         if (ret < 0) {
1377                 mlog_errno(ret);
1378                 return ret;
1379         }
1380         ret = __ocfs2_xattr_set_value_outside(inode, ctxt->handle, vb,
1381                                               xi->value, xi->value_len);
1382         if (ret < 0)
1383                 mlog_errno(ret);
1384
1385         return ret;
1386 }
1387
1388 /*
1389  * ocfs2_xattr_set_entry_local()
1390  *
1391  * Set, replace or remove extended attribute in local.
1392  */
1393 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1394                                         struct ocfs2_xattr_info *xi,
1395                                         struct ocfs2_xattr_search *xs,
1396                                         struct ocfs2_xattr_entry *last,
1397                                         size_t min_offs)
1398 {
1399         size_t name_len = strlen(xi->name);
1400         int i;
1401
1402         if (xi->value && xs->not_found) {
1403                 /* Insert the new xattr entry. */
1404                 le16_add_cpu(&xs->header->xh_count, 1);
1405                 ocfs2_xattr_set_type(last, xi->name_index);
1406                 ocfs2_xattr_set_local(last, 1);
1407                 last->xe_name_len = name_len;
1408         } else {
1409                 void *first_val;
1410                 void *val;
1411                 size_t offs, size;
1412
1413                 first_val = xs->base + min_offs;
1414                 offs = le16_to_cpu(xs->here->xe_name_offset);
1415                 val = xs->base + offs;
1416
1417                 if (le64_to_cpu(xs->here->xe_value_size) >
1418                     OCFS2_XATTR_INLINE_SIZE)
1419                         size = OCFS2_XATTR_SIZE(name_len) +
1420                                 OCFS2_XATTR_ROOT_SIZE;
1421                 else
1422                         size = OCFS2_XATTR_SIZE(name_len) +
1423                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1424
1425                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1426                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1427                         /* The old and the new value have the
1428                            same size. Just replace the value. */
1429                         ocfs2_xattr_set_local(xs->here, 1);
1430                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1431                         /* Clear value bytes. */
1432                         memset(val + OCFS2_XATTR_SIZE(name_len),
1433                                0,
1434                                OCFS2_XATTR_SIZE(xi->value_len));
1435                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1436                                xi->value,
1437                                xi->value_len);
1438                         return;
1439                 }
1440                 /* Remove the old name+value. */
1441                 memmove(first_val + size, first_val, val - first_val);
1442                 memset(first_val, 0, size);
1443                 xs->here->xe_name_hash = 0;
1444                 xs->here->xe_name_offset = 0;
1445                 ocfs2_xattr_set_local(xs->here, 1);
1446                 xs->here->xe_value_size = 0;
1447
1448                 min_offs += size;
1449
1450                 /* Adjust all value offsets. */
1451                 last = xs->header->xh_entries;
1452                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1453                         size_t o = le16_to_cpu(last->xe_name_offset);
1454
1455                         if (o < offs)
1456                                 last->xe_name_offset = cpu_to_le16(o + size);
1457                         last += 1;
1458                 }
1459
1460                 if (!xi->value) {
1461                         /* Remove the old entry. */
1462                         last -= 1;
1463                         memmove(xs->here, xs->here + 1,
1464                                 (void *)last - (void *)xs->here);
1465                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1466                         le16_add_cpu(&xs->header->xh_count, -1);
1467                 }
1468         }
1469         if (xi->value) {
1470                 /* Insert the new name+value. */
1471                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1472                                 OCFS2_XATTR_SIZE(xi->value_len);
1473                 void *val = xs->base + min_offs - size;
1474
1475                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1476                 memset(val, 0, size);
1477                 memcpy(val, xi->name, name_len);
1478                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1479                        xi->value,
1480                        xi->value_len);
1481                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1482                 ocfs2_xattr_set_local(xs->here, 1);
1483                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1484         }
1485
1486         return;
1487 }
1488
1489 /*
1490  * ocfs2_xattr_set_entry()
1491  *
1492  * Set extended attribute entry into inode or block.
1493  *
1494  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1495  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1496  * then set value in B tree with set_value_outside().
1497  */
1498 static int ocfs2_xattr_set_entry(struct inode *inode,
1499                                  struct ocfs2_xattr_info *xi,
1500                                  struct ocfs2_xattr_search *xs,
1501                                  struct ocfs2_xattr_set_ctxt *ctxt,
1502                                  int flag)
1503 {
1504         struct ocfs2_xattr_entry *last;
1505         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1506         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1507         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1508         size_t size_l = 0;
1509         handle_t *handle = ctxt->handle;
1510         int free, i, ret;
1511         struct ocfs2_xattr_info xi_l = {
1512                 .name_index = xi->name_index,
1513                 .name = xi->name,
1514                 .value = xi->value,
1515                 .value_len = xi->value_len,
1516         };
1517         struct ocfs2_xattr_value_buf vb = {
1518                 .vb_bh = xs->xattr_bh,
1519                 .vb_access = ocfs2_journal_access_di,
1520         };
1521
1522         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1523                 BUG_ON(xs->xattr_bh == xs->inode_bh);
1524                 vb.vb_access = ocfs2_journal_access_xb;
1525         } else
1526                 BUG_ON(xs->xattr_bh != xs->inode_bh);
1527
1528         /* Compute min_offs, last and free space. */
1529         last = xs->header->xh_entries;
1530
1531         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1532                 size_t offs = le16_to_cpu(last->xe_name_offset);
1533                 if (offs < min_offs)
1534                         min_offs = offs;
1535                 last += 1;
1536         }
1537
1538         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
1539         if (free < 0)
1540                 return -EIO;
1541
1542         if (!xs->not_found) {
1543                 size_t size = 0;
1544                 if (ocfs2_xattr_is_local(xs->here))
1545                         size = OCFS2_XATTR_SIZE(name_len) +
1546                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1547                 else
1548                         size = OCFS2_XATTR_SIZE(name_len) +
1549                                 OCFS2_XATTR_ROOT_SIZE;
1550                 free += (size + sizeof(struct ocfs2_xattr_entry));
1551         }
1552         /* Check free space in inode or block */
1553         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1554                 if (free < sizeof(struct ocfs2_xattr_entry) +
1555                            OCFS2_XATTR_SIZE(name_len) +
1556                            OCFS2_XATTR_ROOT_SIZE) {
1557                         ret = -ENOSPC;
1558                         goto out;
1559                 }
1560                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1561                 xi_l.value = (void *)&def_xv;
1562                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1563         } else if (xi->value) {
1564                 if (free < sizeof(struct ocfs2_xattr_entry) +
1565                            OCFS2_XATTR_SIZE(name_len) +
1566                            OCFS2_XATTR_SIZE(xi->value_len)) {
1567                         ret = -ENOSPC;
1568                         goto out;
1569                 }
1570         }
1571
1572         if (!xs->not_found) {
1573                 /* For existing extended attribute */
1574                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1575                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1576                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1577                 void *val = xs->base + offs;
1578
1579                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1580                         /* Replace existing local xattr with tree root */
1581                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1582                                                             ctxt, &vb, offs);
1583                         if (ret < 0)
1584                                 mlog_errno(ret);
1585                         goto out;
1586                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1587                         /* For existing xattr which has value outside */
1588                         vb.vb_xv = (struct ocfs2_xattr_value_root *)
1589                                 (val + OCFS2_XATTR_SIZE(name_len));
1590
1591                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1592                                 /*
1593                                  * If new value need set outside also,
1594                                  * first truncate old value to new value,
1595                                  * then set new value with set_value_outside().
1596                                  */
1597                                 ret = ocfs2_xattr_value_truncate(inode,
1598                                                                  &vb,
1599                                                                  xi->value_len,
1600                                                                  ctxt);
1601                                 if (ret < 0) {
1602                                         mlog_errno(ret);
1603                                         goto out;
1604                                 }
1605
1606                                 ret = ocfs2_xattr_update_entry(inode,
1607                                                                handle,
1608                                                                xi,
1609                                                                xs,
1610                                                                &vb,
1611                                                                offs);
1612                                 if (ret < 0) {
1613                                         mlog_errno(ret);
1614                                         goto out;
1615                                 }
1616
1617                                 ret = __ocfs2_xattr_set_value_outside(inode,
1618                                                                 handle,
1619                                                                 &vb,
1620                                                                 xi->value,
1621                                                                 xi->value_len);
1622                                 if (ret < 0)
1623                                         mlog_errno(ret);
1624                                 goto out;
1625                         } else {
1626                                 /*
1627                                  * If new value need set in local,
1628                                  * just trucate old value to zero.
1629                                  */
1630                                  ret = ocfs2_xattr_value_truncate(inode,
1631                                                                   &vb,
1632                                                                   0,
1633                                                                   ctxt);
1634                                 if (ret < 0)
1635                                         mlog_errno(ret);
1636                         }
1637                 }
1638         }
1639
1640         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), xs->inode_bh,
1641                                       OCFS2_JOURNAL_ACCESS_WRITE);
1642         if (ret) {
1643                 mlog_errno(ret);
1644                 goto out;
1645         }
1646
1647         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1648                 ret = vb.vb_access(handle, INODE_CACHE(inode), vb.vb_bh,
1649                                    OCFS2_JOURNAL_ACCESS_WRITE);
1650                 if (ret) {
1651                         mlog_errno(ret);
1652                         goto out;
1653                 }
1654         }
1655
1656         /*
1657          * Set value in local, include set tree root in local.
1658          * This is the first step for value size >INLINE_SIZE.
1659          */
1660         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1661
1662         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1663                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1664                 if (ret < 0) {
1665                         mlog_errno(ret);
1666                         goto out;
1667                 }
1668         }
1669
1670         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1671             (flag & OCFS2_INLINE_XATTR_FL)) {
1672                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1673                 unsigned int xattrsize = osb->s_xattr_inline_size;
1674
1675                 /*
1676                  * Adjust extent record count or inline data size
1677                  * to reserve space for extended attribute.
1678                  */
1679                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1680                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1681                         le16_add_cpu(&idata->id_count, -xattrsize);
1682                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1683                         struct ocfs2_extent_list *el = &di->id2.i_list;
1684                         le16_add_cpu(&el->l_count, -(xattrsize /
1685                                         sizeof(struct ocfs2_extent_rec)));
1686                 }
1687                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1688         }
1689         /* Update xattr flag */
1690         spin_lock(&oi->ip_lock);
1691         oi->ip_dyn_features |= flag;
1692         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1693         spin_unlock(&oi->ip_lock);
1694
1695         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1696         if (ret < 0)
1697                 mlog_errno(ret);
1698
1699         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1700                 /*
1701                  * Set value outside in B tree.
1702                  * This is the second step for value size > INLINE_SIZE.
1703                  */
1704                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1705                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, ctxt,
1706                                                     &vb, offs);
1707                 if (ret < 0) {
1708                         int ret2;
1709
1710                         mlog_errno(ret);
1711                         /*
1712                          * If set value outside failed, we have to clean
1713                          * the junk tree root we have already set in local.
1714                          */
1715                         ret2 = ocfs2_xattr_cleanup(inode, ctxt->handle,
1716                                                    xi, xs, &vb, offs);
1717                         if (ret2 < 0)
1718                                 mlog_errno(ret2);
1719                 }
1720         }
1721 out:
1722         return ret;
1723 }
1724
1725 static int ocfs2_remove_value_outside(struct inode*inode,
1726                                       struct ocfs2_xattr_value_buf *vb,
1727                                       struct ocfs2_xattr_header *header)
1728 {
1729         int ret = 0, i;
1730         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1731         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
1732
1733         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
1734
1735         ctxt.handle = ocfs2_start_trans(osb,
1736                                         ocfs2_remove_extent_credits(osb->sb));
1737         if (IS_ERR(ctxt.handle)) {
1738                 ret = PTR_ERR(ctxt.handle);
1739                 mlog_errno(ret);
1740                 goto out;
1741         }
1742
1743         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1744                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1745
1746                 if (!ocfs2_xattr_is_local(entry)) {
1747                         void *val;
1748
1749                         val = (void *)header +
1750                                 le16_to_cpu(entry->xe_name_offset);
1751                         vb->vb_xv = (struct ocfs2_xattr_value_root *)
1752                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1753                         ret = ocfs2_xattr_value_truncate(inode, vb, 0, &ctxt);
1754                         if (ret < 0) {
1755                                 mlog_errno(ret);
1756                                 break;
1757                         }
1758                 }
1759         }
1760
1761         ocfs2_commit_trans(osb, ctxt.handle);
1762         ocfs2_schedule_truncate_log_flush(osb, 1);
1763         ocfs2_run_deallocs(osb, &ctxt.dealloc);
1764 out:
1765         return ret;
1766 }
1767
1768 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1769                                     struct buffer_head *di_bh)
1770 {
1771
1772         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1773         struct ocfs2_xattr_header *header;
1774         int ret;
1775         struct ocfs2_xattr_value_buf vb = {
1776                 .vb_bh = di_bh,
1777                 .vb_access = ocfs2_journal_access_di,
1778         };
1779
1780         header = (struct ocfs2_xattr_header *)
1781                  ((void *)di + inode->i_sb->s_blocksize -
1782                  le16_to_cpu(di->i_xattr_inline_size));
1783
1784         ret = ocfs2_remove_value_outside(inode, &vb, header);
1785
1786         return ret;
1787 }
1788
1789 static int ocfs2_xattr_block_remove(struct inode *inode,
1790                                     struct buffer_head *blk_bh)
1791 {
1792         struct ocfs2_xattr_block *xb;
1793         int ret = 0;
1794         struct ocfs2_xattr_value_buf vb = {
1795                 .vb_bh = blk_bh,
1796                 .vb_access = ocfs2_journal_access_xb,
1797         };
1798
1799         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1800         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1801                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1802                 ret = ocfs2_remove_value_outside(inode, &vb, header);
1803         } else
1804                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1805
1806         return ret;
1807 }
1808
1809 static int ocfs2_xattr_free_block(struct inode *inode,
1810                                   u64 block)
1811 {
1812         struct inode *xb_alloc_inode;
1813         struct buffer_head *xb_alloc_bh = NULL;
1814         struct buffer_head *blk_bh = NULL;
1815         struct ocfs2_xattr_block *xb;
1816         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1817         handle_t *handle;
1818         int ret = 0;
1819         u64 blk, bg_blkno;
1820         u16 bit;
1821
1822         ret = ocfs2_read_xattr_block(inode, block, &blk_bh);
1823         if (ret < 0) {
1824                 mlog_errno(ret);
1825                 goto out;
1826         }
1827
1828         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1829         if (ret < 0) {
1830                 mlog_errno(ret);
1831                 goto out;
1832         }
1833
1834         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1835         blk = le64_to_cpu(xb->xb_blkno);
1836         bit = le16_to_cpu(xb->xb_suballoc_bit);
1837         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1838
1839         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1840                                 EXTENT_ALLOC_SYSTEM_INODE,
1841                                 le16_to_cpu(xb->xb_suballoc_slot));
1842         if (!xb_alloc_inode) {
1843                 ret = -ENOMEM;
1844                 mlog_errno(ret);
1845                 goto out;
1846         }
1847         mutex_lock(&xb_alloc_inode->i_mutex);
1848
1849         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1850         if (ret < 0) {
1851                 mlog_errno(ret);
1852                 goto out_mutex;
1853         }
1854
1855         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1856         if (IS_ERR(handle)) {
1857                 ret = PTR_ERR(handle);
1858                 mlog_errno(ret);
1859                 goto out_unlock;
1860         }
1861
1862         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1863                                        bit, bg_blkno, 1);
1864         if (ret < 0)
1865                 mlog_errno(ret);
1866
1867         ocfs2_commit_trans(osb, handle);
1868 out_unlock:
1869         ocfs2_inode_unlock(xb_alloc_inode, 1);
1870         brelse(xb_alloc_bh);
1871 out_mutex:
1872         mutex_unlock(&xb_alloc_inode->i_mutex);
1873         iput(xb_alloc_inode);
1874 out:
1875         brelse(blk_bh);
1876         return ret;
1877 }
1878
1879 /*
1880  * ocfs2_xattr_remove()
1881  *
1882  * Free extended attribute resources associated with this inode.
1883  */
1884 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1885 {
1886         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1887         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1888         handle_t *handle;
1889         int ret;
1890
1891         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1892                 return 0;
1893
1894         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1895                 return 0;
1896
1897         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1898                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1899                 if (ret < 0) {
1900                         mlog_errno(ret);
1901                         goto out;
1902                 }
1903         }
1904
1905         if (di->i_xattr_loc) {
1906                 ret = ocfs2_xattr_free_block(inode,
1907                                              le64_to_cpu(di->i_xattr_loc));
1908                 if (ret < 0) {
1909                         mlog_errno(ret);
1910                         goto out;
1911                 }
1912         }
1913
1914         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1915                                    OCFS2_INODE_UPDATE_CREDITS);
1916         if (IS_ERR(handle)) {
1917                 ret = PTR_ERR(handle);
1918                 mlog_errno(ret);
1919                 goto out;
1920         }
1921         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), di_bh,
1922                                       OCFS2_JOURNAL_ACCESS_WRITE);
1923         if (ret) {
1924                 mlog_errno(ret);
1925                 goto out_commit;
1926         }
1927
1928         di->i_xattr_loc = 0;
1929
1930         spin_lock(&oi->ip_lock);
1931         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1932         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1933         spin_unlock(&oi->ip_lock);
1934
1935         ret = ocfs2_journal_dirty(handle, di_bh);
1936         if (ret < 0)
1937                 mlog_errno(ret);
1938 out_commit:
1939         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1940 out:
1941         return ret;
1942 }
1943
1944 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1945                                         struct ocfs2_dinode *di)
1946 {
1947         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1948         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1949         int free;
1950
1951         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1952                 return 0;
1953
1954         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1955                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1956                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1957         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1958                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1959                         le64_to_cpu(di->i_size);
1960         } else {
1961                 struct ocfs2_extent_list *el = &di->id2.i_list;
1962                 free = (le16_to_cpu(el->l_count) -
1963                         le16_to_cpu(el->l_next_free_rec)) *
1964                         sizeof(struct ocfs2_extent_rec);
1965         }
1966         if (free >= xattrsize)
1967                 return 1;
1968
1969         return 0;
1970 }
1971
1972 /*
1973  * ocfs2_xattr_ibody_find()
1974  *
1975  * Find extended attribute in inode block and
1976  * fill search info into struct ocfs2_xattr_search.
1977  */
1978 static int ocfs2_xattr_ibody_find(struct inode *inode,
1979                                   int name_index,
1980                                   const char *name,
1981                                   struct ocfs2_xattr_search *xs)
1982 {
1983         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1984         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1985         int ret;
1986         int has_space = 0;
1987
1988         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1989                 return 0;
1990
1991         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1992                 down_read(&oi->ip_alloc_sem);
1993                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1994                 up_read(&oi->ip_alloc_sem);
1995                 if (!has_space)
1996                         return 0;
1997         }
1998
1999         xs->xattr_bh = xs->inode_bh;
2000         xs->end = (void *)di + inode->i_sb->s_blocksize;
2001         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
2002                 xs->header = (struct ocfs2_xattr_header *)
2003                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
2004         else
2005                 xs->header = (struct ocfs2_xattr_header *)
2006                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
2007         xs->base = (void *)xs->header;
2008         xs->here = xs->header->xh_entries;
2009
2010         /* Find the named attribute. */
2011         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
2012                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2013                 if (ret && ret != -ENODATA)
2014                         return ret;
2015                 xs->not_found = ret;
2016         }
2017
2018         return 0;
2019 }
2020
2021 /*
2022  * ocfs2_xattr_ibody_set()
2023  *
2024  * Set, replace or remove an extended attribute into inode block.
2025  *
2026  */
2027 static int ocfs2_xattr_ibody_set(struct inode *inode,
2028                                  struct ocfs2_xattr_info *xi,
2029                                  struct ocfs2_xattr_search *xs,
2030                                  struct ocfs2_xattr_set_ctxt *ctxt)
2031 {
2032         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2033         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2034         int ret;
2035
2036         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
2037                 return -ENOSPC;
2038
2039         down_write(&oi->ip_alloc_sem);
2040         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
2041                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
2042                         ret = -ENOSPC;
2043                         goto out;
2044                 }
2045         }
2046
2047         ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2048                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
2049 out:
2050         up_write(&oi->ip_alloc_sem);
2051
2052         return ret;
2053 }
2054
2055 /*
2056  * ocfs2_xattr_block_find()
2057  *
2058  * Find extended attribute in external block and
2059  * fill search info into struct ocfs2_xattr_search.
2060  */
2061 static int ocfs2_xattr_block_find(struct inode *inode,
2062                                   int name_index,
2063                                   const char *name,
2064                                   struct ocfs2_xattr_search *xs)
2065 {
2066         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
2067         struct buffer_head *blk_bh = NULL;
2068         struct ocfs2_xattr_block *xb;
2069         int ret = 0;
2070
2071         if (!di->i_xattr_loc)
2072                 return ret;
2073
2074         ret = ocfs2_read_xattr_block(inode, le64_to_cpu(di->i_xattr_loc),
2075                                      &blk_bh);
2076         if (ret < 0) {
2077                 mlog_errno(ret);
2078                 return ret;
2079         }
2080
2081         xs->xattr_bh = blk_bh;
2082         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
2083
2084         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
2085                 xs->header = &xb->xb_attrs.xb_header;
2086                 xs->base = (void *)xs->header;
2087                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
2088                 xs->here = xs->header->xh_entries;
2089
2090                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
2091         } else
2092                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
2093                                                    name_index,
2094                                                    name, xs);
2095
2096         if (ret && ret != -ENODATA) {
2097                 xs->xattr_bh = NULL;
2098                 goto cleanup;
2099         }
2100         xs->not_found = ret;
2101         return 0;
2102 cleanup:
2103         brelse(blk_bh);
2104
2105         return ret;
2106 }
2107
2108 static int ocfs2_create_xattr_block(handle_t *handle,
2109                                     struct inode *inode,
2110                                     struct buffer_head *inode_bh,
2111                                     struct ocfs2_alloc_context *meta_ac,
2112                                     struct buffer_head **ret_bh)
2113 {
2114         int ret;
2115         u16 suballoc_bit_start;
2116         u32 num_got;
2117         u64 first_blkno;
2118         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)inode_bh->b_data;
2119         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2120         struct buffer_head *new_bh = NULL;
2121         struct ocfs2_xattr_block *xblk;
2122
2123         ret = ocfs2_journal_access_di(handle, INODE_CACHE(inode), inode_bh,
2124                                       OCFS2_JOURNAL_ACCESS_CREATE);
2125         if (ret < 0) {
2126                 mlog_errno(ret);
2127                 goto end;
2128         }
2129
2130         ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
2131                                    &suballoc_bit_start, &num_got,
2132                                    &first_blkno);
2133         if (ret < 0) {
2134                 mlog_errno(ret);
2135                 goto end;
2136         }
2137
2138         new_bh = sb_getblk(inode->i_sb, first_blkno);
2139         ocfs2_set_new_buffer_uptodate(INODE_CACHE(inode), new_bh);
2140
2141         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode),
2142                                       new_bh,
2143                                       OCFS2_JOURNAL_ACCESS_CREATE);
2144         if (ret < 0) {
2145                 mlog_errno(ret);
2146                 goto end;
2147         }
2148
2149         /* Initialize ocfs2_xattr_block */
2150         xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
2151         memset(xblk, 0, inode->i_sb->s_blocksize);
2152         strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
2153         xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
2154         xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
2155         xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
2156         xblk->xb_blkno = cpu_to_le64(first_blkno);
2157
2158         ret = ocfs2_journal_dirty(handle, new_bh);
2159         if (ret < 0) {
2160                 mlog_errno(ret);
2161                 goto end;
2162         }
2163         di->i_xattr_loc = cpu_to_le64(first_blkno);
2164         ocfs2_journal_dirty(handle, inode_bh);
2165
2166         *ret_bh = new_bh;
2167         new_bh = NULL;
2168
2169 end:
2170         brelse(new_bh);
2171         return ret;
2172 }
2173
2174 /*
2175  * ocfs2_xattr_block_set()
2176  *
2177  * Set, replace or remove an extended attribute into external block.
2178  *
2179  */
2180 static int ocfs2_xattr_block_set(struct inode *inode,
2181                                  struct ocfs2_xattr_info *xi,
2182                                  struct ocfs2_xattr_search *xs,
2183                                  struct ocfs2_xattr_set_ctxt *ctxt)
2184 {
2185         struct buffer_head *new_bh = NULL;
2186         handle_t *handle = ctxt->handle;
2187         struct ocfs2_xattr_block *xblk = NULL;
2188         int ret;
2189
2190         if (!xs->xattr_bh) {
2191                 ret = ocfs2_create_xattr_block(handle, inode, xs->inode_bh,
2192                                                ctxt->meta_ac, &new_bh);
2193                 if (ret) {
2194                         mlog_errno(ret);
2195                         goto end;
2196                 }
2197
2198                 xs->xattr_bh = new_bh;
2199                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2200                 xs->header = &xblk->xb_attrs.xb_header;
2201                 xs->base = (void *)xs->header;
2202                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
2203                 xs->here = xs->header->xh_entries;
2204         } else
2205                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
2206
2207         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
2208                 /* Set extended attribute into external block */
2209                 ret = ocfs2_xattr_set_entry(inode, xi, xs, ctxt,
2210                                             OCFS2_HAS_XATTR_FL);
2211                 if (!ret || ret != -ENOSPC)
2212                         goto end;
2213
2214                 ret = ocfs2_xattr_create_index_block(inode, xs, ctxt);
2215                 if (ret)
2216                         goto end;
2217         }
2218
2219         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs, ctxt);
2220
2221 end:
2222
2223         return ret;
2224 }
2225
2226 /* Check whether the new xattr can be inserted into the inode. */
2227 static int ocfs2_xattr_can_be_in_inode(struct inode *inode,
2228                                        struct ocfs2_xattr_info *xi,
2229                                        struct ocfs2_xattr_search *xs)
2230 {
2231         u64 value_size;
2232         struct ocfs2_xattr_entry *last;
2233         int free, i;
2234         size_t min_offs = xs->end - xs->base;
2235
2236         if (!xs->header)
2237                 return 0;
2238
2239         last = xs->header->xh_entries;
2240
2241         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
2242                 size_t offs = le16_to_cpu(last->xe_name_offset);
2243                 if (offs < min_offs)
2244                         min_offs = offs;
2245                 last += 1;
2246         }
2247
2248         free = min_offs - ((void *)last - xs->base) - OCFS2_XATTR_HEADER_GAP;
2249         if (free < 0)
2250                 return 0;
2251
2252         BUG_ON(!xs->not_found);
2253
2254         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2255                 value_size = OCFS2_XATTR_ROOT_SIZE;
2256         else
2257                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
2258
2259         if (free >= sizeof(struct ocfs2_xattr_entry) +
2260                    OCFS2_XATTR_SIZE(strlen(xi->name)) + value_size)
2261                 return 1;
2262
2263         return 0;
2264 }
2265
2266 static int ocfs2_calc_xattr_set_need(struct inode *inode,
2267                                      struct ocfs2_dinode *di,
2268                                      struct ocfs2_xattr_info *xi,
2269                                      struct ocfs2_xattr_search *xis,
2270                                      struct ocfs2_xattr_search *xbs,
2271                                      int *clusters_need,
2272                                      int *meta_need,
2273                                      int *credits_need)
2274 {
2275         int ret = 0, old_in_xb = 0;
2276         int clusters_add = 0, meta_add = 0, credits = 0;
2277         struct buffer_head *bh = NULL;
2278         struct ocfs2_xattr_block *xb = NULL;
2279         struct ocfs2_xattr_entry *xe = NULL;
2280         struct ocfs2_xattr_value_root *xv = NULL;
2281         char *base = NULL;
2282         int name_offset, name_len = 0;
2283         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb,
2284                                                     xi->value_len);
2285         u64 value_size;
2286
2287         /*
2288          * Calculate the clusters we need to write.
2289          * No matter whether we replace an old one or add a new one,
2290          * we need this for writing.
2291          */
2292         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
2293                 credits += new_clusters *
2294                            ocfs2_clusters_to_blocks(inode->i_sb, 1);
2295
2296         if (xis->not_found && xbs->not_found) {
2297                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2298
2299                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2300                         clusters_add += new_clusters;
2301                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2302                                                         &def_xv.xv.xr_list,
2303                                                         new_clusters);
2304                 }
2305
2306                 goto meta_guess;
2307         }
2308
2309         if (!xis->not_found) {
2310                 xe = xis->here;
2311                 name_offset = le16_to_cpu(xe->xe_name_offset);
2312                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2313                 base = xis->base;
2314                 credits += OCFS2_INODE_UPDATE_CREDITS;
2315         } else {
2316                 int i, block_off = 0;
2317                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2318                 xe = xbs->here;
2319                 name_offset = le16_to_cpu(xe->xe_name_offset);
2320                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
2321                 i = xbs->here - xbs->header->xh_entries;
2322                 old_in_xb = 1;
2323
2324                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2325                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
2326                                                         bucket_xh(xbs->bucket),
2327                                                         i, &block_off,
2328                                                         &name_offset);
2329                         base = bucket_block(xbs->bucket, block_off);
2330                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2331                 } else {
2332                         base = xbs->base;
2333                         credits += OCFS2_XATTR_BLOCK_UPDATE_CREDITS;
2334                 }
2335         }
2336
2337         /*
2338          * delete a xattr doesn't need metadata and cluster allocation.
2339          * so just calculate the credits and return.
2340          *
2341          * The credits for removing the value tree will be extended
2342          * by ocfs2_remove_extent itself.
2343          */
2344         if (!xi->value) {
2345                 if (!ocfs2_xattr_is_local(xe))
2346                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2347
2348                 goto out;
2349         }
2350
2351         /* do cluster allocation guess first. */
2352         value_size = le64_to_cpu(xe->xe_value_size);
2353
2354         if (old_in_xb) {
2355                 /*
2356                  * In xattr set, we always try to set the xe in inode first,
2357                  * so if it can be inserted into inode successfully, the old
2358                  * one will be removed from the xattr block, and this xattr
2359                  * will be inserted into inode as a new xattr in inode.
2360                  */
2361                 if (ocfs2_xattr_can_be_in_inode(inode, xi, xis)) {
2362                         clusters_add += new_clusters;
2363                         credits += ocfs2_remove_extent_credits(inode->i_sb) +
2364                                     OCFS2_INODE_UPDATE_CREDITS;
2365                         if (!ocfs2_xattr_is_local(xe))
2366                                 credits += ocfs2_calc_extend_credits(
2367                                                         inode->i_sb,
2368                                                         &def_xv.xv.xr_list,
2369                                                         new_clusters);
2370                         goto out;
2371                 }
2372         }
2373
2374         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
2375                 /* the new values will be stored outside. */
2376                 u32 old_clusters = 0;
2377
2378                 if (!ocfs2_xattr_is_local(xe)) {
2379                         old_clusters =  ocfs2_clusters_for_bytes(inode->i_sb,
2380                                                                  value_size);
2381                         xv = (struct ocfs2_xattr_value_root *)
2382                              (base + name_offset + name_len);
2383                         value_size = OCFS2_XATTR_ROOT_SIZE;
2384                 } else
2385                         xv = &def_xv.xv;
2386
2387                 if (old_clusters >= new_clusters) {
2388                         credits += ocfs2_remove_extent_credits(inode->i_sb);
2389                         goto out;
2390                 } else {
2391                         meta_add += ocfs2_extend_meta_needed(&xv->xr_list);
2392                         clusters_add += new_clusters - old_clusters;
2393                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2394                                                              &xv->xr_list,
2395                                                              new_clusters -
2396                                                              old_clusters);
2397                         if (value_size >= OCFS2_XATTR_ROOT_SIZE)
2398                                 goto out;
2399                 }
2400         } else {
2401                 /*
2402                  * Now the new value will be stored inside. So if the new
2403                  * value is smaller than the size of value root or the old
2404                  * value, we don't need any allocation, otherwise we have
2405                  * to guess metadata allocation.
2406                  */
2407                 if ((ocfs2_xattr_is_local(xe) && value_size >= xi->value_len) ||
2408                     (!ocfs2_xattr_is_local(xe) &&
2409                      OCFS2_XATTR_ROOT_SIZE >= xi->value_len))
2410                         goto out;
2411         }
2412
2413 meta_guess:
2414         /* calculate metadata allocation. */
2415         if (di->i_xattr_loc) {
2416                 if (!xbs->xattr_bh) {
2417                         ret = ocfs2_read_xattr_block(inode,
2418                                                      le64_to_cpu(di->i_xattr_loc),
2419                                                      &bh);
2420                         if (ret) {
2421                                 mlog_errno(ret);
2422                                 goto out;
2423                         }
2424
2425                         xb = (struct ocfs2_xattr_block *)bh->b_data;
2426                 } else
2427                         xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
2428
2429                 /*
2430                  * If there is already an xattr tree, good, we can calculate
2431                  * like other b-trees. Otherwise we may have the chance of
2432                  * create a tree, the credit calculation is borrowed from
2433                  * ocfs2_calc_extend_credits with root_el = NULL. And the
2434                  * new tree will be cluster based, so no meta is needed.
2435                  */
2436                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
2437                         struct ocfs2_extent_list *el =
2438                                  &xb->xb_attrs.xb_root.xt_list;
2439                         meta_add += ocfs2_extend_meta_needed(el);
2440                         credits += ocfs2_calc_extend_credits(inode->i_sb,
2441                                                              el, 1);
2442                 } else
2443                         credits += OCFS2_SUBALLOC_ALLOC + 1;
2444
2445                 /*
2446                  * This cluster will be used either for new bucket or for
2447                  * new xattr block.
2448                  * If the cluster size is the same as the bucket size, one
2449                  * more is needed since we may need to extend the bucket
2450                  * also.
2451                  */
2452                 clusters_add += 1;
2453                 credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2454                 if (OCFS2_XATTR_BUCKET_SIZE ==
2455                         OCFS2_SB(inode->i_sb)->s_clustersize) {
2456                         credits += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2457                         clusters_add += 1;
2458                 }
2459         } else {
2460                 meta_add += 1;
2461                 credits += OCFS2_XATTR_BLOCK_CREATE_CREDITS;
2462         }
2463 out:
2464         if (clusters_need)
2465                 *clusters_need = clusters_add;
2466         if (meta_need)
2467                 *meta_need = meta_add;
2468         if (credits_need)
2469                 *credits_need = credits;
2470         brelse(bh);
2471         return ret;
2472 }
2473
2474 static int ocfs2_init_xattr_set_ctxt(struct inode *inode,
2475                                      struct ocfs2_dinode *di,
2476                                      struct ocfs2_xattr_info *xi,
2477                                      struct ocfs2_xattr_search *xis,
2478                                      struct ocfs2_xattr_search *xbs,
2479                                      struct ocfs2_xattr_set_ctxt *ctxt,
2480                                      int extra_meta,
2481                                      int *credits)
2482 {
2483         int clusters_add, meta_add, ret;
2484         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2485
2486         memset(ctxt, 0, sizeof(struct ocfs2_xattr_set_ctxt));
2487
2488         ocfs2_init_dealloc_ctxt(&ctxt->dealloc);
2489
2490         ret = ocfs2_calc_xattr_set_need(inode, di, xi, xis, xbs,
2491                                         &clusters_add, &meta_add, credits);
2492         if (ret) {
2493                 mlog_errno(ret);
2494                 return ret;
2495         }
2496
2497         meta_add += extra_meta;
2498         mlog(0, "Set xattr %s, reserve meta blocks = %d, clusters = %d, "
2499              "credits = %d\n", xi->name, meta_add, clusters_add, *credits);
2500
2501         if (meta_add) {
2502                 ret = ocfs2_reserve_new_metadata_blocks(osb, meta_add,
2503                                                         &ctxt->meta_ac);
2504                 if (ret) {
2505                         mlog_errno(ret);
2506                         goto out;
2507                 }
2508         }
2509
2510         if (clusters_add) {
2511                 ret = ocfs2_reserve_clusters(osb, clusters_add, &ctxt->data_ac);
2512                 if (ret)
2513                         mlog_errno(ret);
2514         }
2515 out:
2516         if (ret) {
2517                 if (ctxt->meta_ac) {
2518                         ocfs2_free_alloc_context(ctxt->meta_ac);
2519                         ctxt->meta_ac = NULL;
2520                 }
2521
2522                 /*
2523                  * We cannot have an error and a non null ctxt->data_ac.
2524                  */
2525         }
2526
2527         return ret;
2528 }
2529
2530 static int __ocfs2_xattr_set_handle(struct inode *inode,
2531                                     struct ocfs2_dinode *di,
2532                                     struct ocfs2_xattr_info *xi,
2533                                     struct ocfs2_xattr_search *xis,
2534                                     struct ocfs2_xattr_search *xbs,
2535                                     struct ocfs2_xattr_set_ctxt *ctxt)
2536 {
2537         int ret = 0, credits, old_found;
2538
2539         if (!xi->value) {
2540                 /* Remove existing extended attribute */
2541                 if (!xis->not_found)
2542                         ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2543                 else if (!xbs->not_found)
2544                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2545         } else {
2546                 /* We always try to set extended attribute into inode first*/
2547                 ret = ocfs2_xattr_ibody_set(inode, xi, xis, ctxt);
2548                 if (!ret && !xbs->not_found) {
2549                         /*
2550                          * If succeed and that extended attribute existing in
2551                          * external block, then we will remove it.
2552                          */
2553                         xi->value = NULL;
2554                         xi->value_len = 0;
2555
2556                         old_found = xis->not_found;
2557                         xis->not_found = -ENODATA;
2558                         ret = ocfs2_calc_xattr_set_need(inode,
2559                                                         di,
2560                                                         xi,
2561                                                         xis,
2562                                                         xbs,
2563                                                         NULL,
2564                                                         NULL,
2565                                                         &credits);
2566                         xis->not_found = old_found;
2567                         if (ret) {
2568                                 mlog_errno(ret);
2569                                 goto out;
2570                         }
2571
2572                         ret = ocfs2_extend_trans(ctxt->handle, credits +
2573                                         ctxt->handle->h_buffer_credits);
2574                         if (ret) {
2575                                 mlog_errno(ret);
2576                                 goto out;
2577                         }
2578                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2579                 } else if (ret == -ENOSPC) {
2580                         if (di->i_xattr_loc && !xbs->xattr_bh) {
2581                                 ret = ocfs2_xattr_block_find(inode,
2582                                                              xi->name_index,
2583                                                              xi->name, xbs);
2584                                 if (ret)
2585                                         goto out;
2586
2587                                 old_found = xis->not_found;
2588                                 xis->not_found = -ENODATA;
2589                                 ret = ocfs2_calc_xattr_set_need(inode,
2590                                                                 di,
2591                                                                 xi,
2592                                                                 xis,
2593                                                                 xbs,
2594                                                                 NULL,
2595                                                                 NULL,
2596                                                                 &credits);
2597                                 xis->not_found = old_found;
2598                                 if (ret) {
2599                                         mlog_errno(ret);
2600                                         goto out;
2601                                 }
2602
2603                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2604                                         ctxt->handle->h_buffer_credits);
2605                                 if (ret) {
2606                                         mlog_errno(ret);
2607                                         goto out;
2608                                 }
2609                         }
2610                         /*
2611                          * If no space in inode, we will set extended attribute
2612                          * into external block.
2613                          */
2614                         ret = ocfs2_xattr_block_set(inode, xi, xbs, ctxt);
2615                         if (ret)
2616                                 goto out;
2617                         if (!xis->not_found) {
2618                                 /*
2619                                  * If succeed and that extended attribute
2620                                  * existing in inode, we will remove it.
2621                                  */
2622                                 xi->value = NULL;
2623                                 xi->value_len = 0;
2624                                 xbs->not_found = -ENODATA;
2625                                 ret = ocfs2_calc_xattr_set_need(inode,
2626                                                                 di,
2627                                                                 xi,
2628                                                                 xis,
2629                                                                 xbs,
2630                                                                 NULL,
2631                                                                 NULL,
2632                                                                 &credits);
2633                                 if (ret) {
2634                                         mlog_errno(ret);
2635                                         goto out;
2636                                 }
2637
2638                                 ret = ocfs2_extend_trans(ctxt->handle, credits +
2639                                                 ctxt->handle->h_buffer_credits);
2640                                 if (ret) {
2641                                         mlog_errno(ret);
2642                                         goto out;
2643                                 }
2644                                 ret = ocfs2_xattr_ibody_set(inode, xi,
2645                                                             xis, ctxt);
2646                         }
2647                 }
2648         }
2649
2650         if (!ret) {
2651                 /* Update inode ctime. */
2652                 ret = ocfs2_journal_access_di(ctxt->handle, INODE_CACHE(inode),
2653                                               xis->inode_bh,
2654                                               OCFS2_JOURNAL_ACCESS_WRITE);
2655                 if (ret) {
2656                         mlog_errno(ret);
2657                         goto out;
2658                 }
2659
2660                 inode->i_ctime = CURRENT_TIME;
2661                 di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
2662                 di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
2663                 ocfs2_journal_dirty(ctxt->handle, xis->inode_bh);
2664         }
2665 out:
2666         return ret;
2667 }
2668
2669 /*
2670  * This function only called duing creating inode
2671  * for init security/acl xattrs of the new inode.
2672  * All transanction credits have been reserved in mknod.
2673  */
2674 int ocfs2_xattr_set_handle(handle_t *handle,
2675                            struct inode *inode,
2676                            struct buffer_head *di_bh,
2677                            int name_index,
2678                            const char *name,
2679                            const void *value,
2680                            size_t value_len,
2681                            int flags,
2682                            struct ocfs2_alloc_context *meta_ac,
2683                            struct ocfs2_alloc_context *data_ac)
2684 {
2685         struct ocfs2_dinode *di;
2686         int ret;
2687
2688         struct ocfs2_xattr_info xi = {
2689                 .name_index = name_index,
2690                 .name = name,
2691                 .value = value,
2692                 .value_len = value_len,
2693         };
2694
2695         struct ocfs2_xattr_search xis = {
2696                 .not_found = -ENODATA,
2697         };
2698
2699         struct ocfs2_xattr_search xbs = {
2700                 .not_found = -ENODATA,
2701         };
2702
2703         struct ocfs2_xattr_set_ctxt ctxt = {
2704                 .handle = handle,
2705                 .meta_ac = meta_ac,
2706                 .data_ac = data_ac,
2707         };
2708
2709         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2710                 return -EOPNOTSUPP;
2711
2712         /*
2713          * In extreme situation, may need xattr bucket when
2714          * block size is too small. And we have already reserved
2715          * the credits for bucket in mknod.
2716          */
2717         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE) {
2718                 xbs.bucket = ocfs2_xattr_bucket_new(inode);
2719                 if (!xbs.bucket) {
2720                         mlog_errno(-ENOMEM);
2721                         return -ENOMEM;
2722                 }
2723         }
2724
2725         xis.inode_bh = xbs.inode_bh = di_bh;
2726         di = (struct ocfs2_dinode *)di_bh->b_data;
2727
2728         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2729
2730         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2731         if (ret)
2732                 goto cleanup;
2733         if (xis.not_found) {
2734                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2735                 if (ret)
2736                         goto cleanup;
2737         }
2738
2739         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2740
2741 cleanup:
2742         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2743         brelse(xbs.xattr_bh);
2744         ocfs2_xattr_bucket_free(xbs.bucket);
2745
2746         return ret;
2747 }
2748
2749 /*
2750  * ocfs2_xattr_set()
2751  *
2752  * Set, replace or remove an extended attribute for this inode.
2753  * value is NULL to remove an existing extended attribute, else either
2754  * create or replace an extended attribute.
2755  */
2756 int ocfs2_xattr_set(struct inode *inode,
2757                     int name_index,
2758                     const char *name,
2759                     const void *value,
2760                     size_t value_len,
2761                     int flags)
2762 {
2763         struct buffer_head *di_bh = NULL;
2764         struct ocfs2_dinode *di;
2765         int ret, credits, ref_meta = 0, ref_credits = 0;
2766         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2767         struct inode *tl_inode = osb->osb_tl_inode;
2768         struct ocfs2_xattr_set_ctxt ctxt = { NULL, NULL, };
2769         struct ocfs2_refcount_tree *ref_tree = NULL;
2770
2771         struct ocfs2_xattr_info xi = {
2772                 .name_index = name_index,
2773                 .name = name,
2774                 .value = value,
2775                 .value_len = value_len,
2776         };
2777
2778         struct ocfs2_xattr_search xis = {
2779                 .not_found = -ENODATA,
2780         };
2781
2782         struct ocfs2_xattr_search xbs = {
2783                 .not_found = -ENODATA,
2784         };
2785
2786         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2787                 return -EOPNOTSUPP;
2788
2789         /*
2790          * Only xbs will be used on indexed trees.  xis doesn't need a
2791          * bucket.
2792          */
2793         xbs.bucket = ocfs2_xattr_bucket_new(inode);
2794         if (!xbs.bucket) {
2795                 mlog_errno(-ENOMEM);
2796                 return -ENOMEM;
2797         }
2798
2799         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2800         if (ret < 0) {
2801                 mlog_errno(ret);
2802                 goto cleanup_nolock;
2803         }
2804         xis.inode_bh = xbs.inode_bh = di_bh;
2805         di = (struct ocfs2_dinode *)di_bh->b_data;
2806
2807         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2808         /*
2809          * Scan inode and external block to find the same name
2810          * extended attribute and collect search infomation.
2811          */
2812         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2813         if (ret)
2814                 goto cleanup;
2815         if (xis.not_found) {
2816                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2817                 if (ret)
2818                         goto cleanup;
2819         }
2820
2821         if (xis.not_found && xbs.not_found) {
2822                 ret = -ENODATA;
2823                 if (flags & XATTR_REPLACE)
2824                         goto cleanup;
2825                 ret = 0;
2826                 if (!value)
2827                         goto cleanup;
2828         } else {
2829                 ret = -EEXIST;
2830                 if (flags & XATTR_CREATE)
2831                         goto cleanup;
2832         }
2833
2834         /* Check whether the value is refcounted and do some prepartion. */
2835         if (OCFS2_I(inode)->ip_dyn_features & OCFS2_HAS_REFCOUNT_FL &&
2836             (!xis.not_found || !xbs.not_found)) {
2837                 ret = ocfs2_prepare_refcount_xattr(inode, di, &xi,
2838                                                    &xis, &xbs, &ref_tree,
2839                                                    &ref_meta, &ref_credits);
2840                 if (ret) {
2841                         mlog_errno(ret);
2842                         goto cleanup;
2843                 }
2844         }
2845
2846         mutex_lock(&tl_inode->i_mutex);
2847
2848         if (ocfs2_truncate_log_needs_flush(osb)) {
2849                 ret = __ocfs2_flush_truncate_log(osb);
2850                 if (ret < 0) {
2851                         mutex_unlock(&tl_inode->i_mutex);
2852                         mlog_errno(ret);
2853                         goto cleanup;
2854                 }
2855         }
2856         mutex_unlock(&tl_inode->i_mutex);
2857
2858         ret = ocfs2_init_xattr_set_ctxt(inode, di, &xi, &xis,
2859                                         &xbs, &ctxt, ref_meta, &credits);
2860         if (ret) {
2861                 mlog_errno(ret);
2862                 goto cleanup;
2863         }
2864
2865         /* we need to update inode's ctime field, so add credit for it. */
2866         credits += OCFS2_INODE_UPDATE_CREDITS;
2867         ctxt.handle = ocfs2_start_trans(osb, credits + ref_credits);
2868         if (IS_ERR(ctxt.handle)) {
2869                 ret = PTR_ERR(ctxt.handle);
2870                 mlog_errno(ret);
2871                 goto cleanup;
2872         }
2873
2874         ret = __ocfs2_xattr_set_handle(inode, di, &xi, &xis, &xbs, &ctxt);
2875
2876         ocfs2_commit_trans(osb, ctxt.handle);
2877
2878         if (ctxt.data_ac)
2879                 ocfs2_free_alloc_context(ctxt.data_ac);
2880         if (ctxt.meta_ac)
2881                 ocfs2_free_alloc_context(ctxt.meta_ac);
2882         if (ocfs2_dealloc_has_cluster(&ctxt.dealloc))
2883                 ocfs2_schedule_truncate_log_flush(osb, 1);
2884         ocfs2_run_deallocs(osb, &ctxt.dealloc);
2885 cleanup:
2886         if (ref_tree)
2887                 ocfs2_unlock_refcount_tree(osb, ref_tree, 1);
2888         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2889         ocfs2_inode_unlock(inode, 1);
2890 cleanup_nolock:
2891         brelse(di_bh);
2892         brelse(xbs.xattr_bh);
2893         ocfs2_xattr_bucket_free(xbs.bucket);
2894
2895         return ret;
2896 }
2897
2898 /*
2899  * Find the xattr extent rec which may contains name_hash.
2900  * e_cpos will be the first name hash of the xattr rec.
2901  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2902  */
2903 static int ocfs2_xattr_get_rec(struct inode *inode,
2904                                u32 name_hash,
2905                                u64 *p_blkno,
2906                                u32 *e_cpos,
2907                                u32 *num_clusters,
2908                                struct ocfs2_extent_list *el)
2909 {
2910         int ret = 0, i;
2911         struct buffer_head *eb_bh = NULL;
2912         struct ocfs2_extent_block *eb;
2913         struct ocfs2_extent_rec *rec = NULL;
2914         u64 e_blkno = 0;
2915
2916         if (el->l_tree_depth) {
2917                 ret = ocfs2_find_leaf(INODE_CACHE(inode), el, name_hash,
2918                                       &eb_bh);
2919                 if (ret) {
2920                         mlog_errno(ret);
2921                         goto out;
2922                 }
2923
2924                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2925                 el = &eb->h_list;
2926
2927                 if (el->l_tree_depth) {
2928                         ocfs2_error(inode->i_sb,
2929                                     "Inode %lu has non zero tree depth in "
2930                                     "xattr tree block %llu\n", inode->i_ino,
2931                                     (unsigned long long)eb_bh->b_blocknr);
2932                         ret = -EROFS;
2933                         goto out;
2934                 }
2935         }
2936
2937         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2938                 rec = &el->l_recs[i];
2939
2940                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2941                         e_blkno = le64_to_cpu(rec->e_blkno);
2942                         break;
2943                 }
2944         }
2945
2946         if (!e_blkno) {
2947                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2948                             "record (%u, %u, 0) in xattr", inode->i_ino,
2949                             le32_to_cpu(rec->e_cpos),
2950                             ocfs2_rec_clusters(el, rec));
2951                 ret = -EROFS;
2952                 goto out;
2953         }
2954
2955         *p_blkno = le64_to_cpu(rec->e_blkno);
2956         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2957         if (e_cpos)
2958                 *e_cpos = le32_to_cpu(rec->e_cpos);
2959 out:
2960         brelse(eb_bh);
2961         return ret;
2962 }
2963
2964 typedef int (xattr_bucket_func)(struct inode *inode,
2965                                 struct ocfs2_xattr_bucket *bucket,
2966                                 void *para);
2967
2968 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2969                                    struct ocfs2_xattr_bucket *bucket,
2970                                    int name_index,
2971                                    const char *name,
2972                                    u32 name_hash,
2973                                    u16 *xe_index,
2974                                    int *found)
2975 {
2976         int i, ret = 0, cmp = 1, block_off, new_offset;
2977         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
2978         size_t name_len = strlen(name);
2979         struct ocfs2_xattr_entry *xe = NULL;
2980         char *xe_name;
2981
2982         /*
2983          * We don't use binary search in the bucket because there
2984          * may be multiple entries with the same name hash.
2985          */
2986         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2987                 xe = &xh->xh_entries[i];
2988
2989                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2990                         continue;
2991                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2992                         break;
2993
2994                 cmp = name_index - ocfs2_xattr_get_type(xe);
2995                 if (!cmp)
2996                         cmp = name_len - xe->xe_name_len;
2997                 if (cmp)
2998                         continue;
2999
3000                 ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3001                                                         xh,
3002                                                         i,
3003                                                         &block_off,
3004                                                         &new_offset);
3005                 if (ret) {
3006                         mlog_errno(ret);
3007                         break;
3008                 }
3009
3010
3011                 xe_name = bucket_block(bucket, block_off) + new_offset;
3012                 if (!memcmp(name, xe_name, name_len)) {
3013                         *xe_index = i;
3014                         *found = 1;
3015                         ret = 0;
3016                         break;
3017                 }
3018         }
3019
3020         return ret;
3021 }
3022
3023 /*
3024  * Find the specified xattr entry in a series of buckets.
3025  * This series start from p_blkno and last for num_clusters.
3026  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
3027  * the num of the valid buckets.
3028  *
3029  * Return the buffer_head this xattr should reside in. And if the xattr's
3030  * hash is in the gap of 2 buckets, return the lower bucket.
3031  */
3032 static int ocfs2_xattr_bucket_find(struct inode *inode,
3033                                    int name_index,
3034                                    const char *name,
3035                                    u32 name_hash,
3036                                    u64 p_blkno,
3037                                    u32 first_hash,
3038                                    u32 num_clusters,
3039                                    struct ocfs2_xattr_search *xs)
3040 {
3041         int ret, found = 0;
3042         struct ocfs2_xattr_header *xh = NULL;
3043         struct ocfs2_xattr_entry *xe = NULL;
3044         u16 index = 0;
3045         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3046         int low_bucket = 0, bucket, high_bucket;
3047         struct ocfs2_xattr_bucket *search;
3048         u32 last_hash;
3049         u64 blkno, lower_blkno = 0;
3050
3051         search = ocfs2_xattr_bucket_new(inode);
3052         if (!search) {
3053                 ret = -ENOMEM;
3054                 mlog_errno(ret);
3055                 goto out;
3056         }
3057
3058         ret = ocfs2_read_xattr_bucket(search, p_blkno);
3059         if (ret) {
3060                 mlog_errno(ret);
3061                 goto out;
3062         }
3063
3064         xh = bucket_xh(search);
3065         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
3066         while (low_bucket <= high_bucket) {
3067                 ocfs2_xattr_bucket_relse(search);
3068
3069                 bucket = (low_bucket + high_bucket) / 2;
3070                 blkno = p_blkno + bucket * blk_per_bucket;
3071                 ret = ocfs2_read_xattr_bucket(search, blkno);
3072                 if (ret) {
3073                         mlog_errno(ret);
3074                         goto out;
3075                 }
3076
3077                 xh = bucket_xh(search);
3078                 xe = &xh->xh_entries[0];
3079                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
3080                         high_bucket = bucket - 1;
3081                         continue;
3082                 }
3083
3084                 /*
3085                  * Check whether the hash of the last entry in our
3086                  * bucket is larger than the search one. for an empty
3087                  * bucket, the last one is also the first one.
3088                  */
3089                 if (xh->xh_count)
3090                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
3091
3092                 last_hash = le32_to_cpu(xe->xe_name_hash);
3093
3094                 /* record lower_blkno which may be the insert place. */
3095                 lower_blkno = blkno;
3096
3097                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
3098                         low_bucket = bucket + 1;
3099                         continue;
3100                 }
3101
3102                 /* the searched xattr should reside in this bucket if exists. */
3103                 ret = ocfs2_find_xe_in_bucket(inode, search,
3104                                               name_index, name, name_hash,
3105                                               &index, &found);
3106                 if (ret) {
3107                         mlog_errno(ret);
3108                         goto out;
3109                 }
3110                 break;
3111         }
3112
3113         /*
3114          * Record the bucket we have found.
3115          * When the xattr's hash value is in the gap of 2 buckets, we will
3116          * always set it to the previous bucket.
3117          */
3118         if (!lower_blkno)
3119                 lower_blkno = p_blkno;
3120
3121         /* This should be in cache - we just read it during the search */
3122         ret = ocfs2_read_xattr_bucket(xs->bucket, lower_blkno);
3123         if (ret) {
3124                 mlog_errno(ret);
3125                 goto out;
3126         }
3127
3128         xs->header = bucket_xh(xs->bucket);
3129         xs->base = bucket_block(xs->bucket, 0);
3130         xs->end = xs->base + inode->i_sb->s_blocksize;
3131
3132         if (found) {
3133                 xs->here = &xs->header->xh_entries[index];
3134                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
3135                      (unsigned long long)bucket_blkno(xs->bucket), index);
3136         } else
3137                 ret = -ENODATA;
3138
3139 out:
3140         ocfs2_xattr_bucket_free(search);
3141         return ret;
3142 }
3143
3144 static int ocfs2_xattr_index_block_find(struct inode *inode,
3145                                         struct buffer_head *root_bh,
3146                                         int name_index,
3147                                         const char *name,
3148                                         struct ocfs2_xattr_search *xs)
3149 {
3150         int ret;
3151         struct ocfs2_xattr_block *xb =
3152                         (struct ocfs2_xattr_block *)root_bh->b_data;
3153         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3154         struct ocfs2_extent_list *el = &xb_root->xt_list;
3155         u64 p_blkno = 0;
3156         u32 first_hash, num_clusters = 0;
3157         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
3158
3159         if (le16_to_cpu(el->l_next_free_rec) == 0)
3160                 return -ENODATA;
3161
3162         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
3163              name, name_hash, name_index);
3164
3165         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
3166                                   &num_clusters, el);
3167         if (ret) {
3168                 mlog_errno(ret);
3169                 goto out;
3170         }
3171
3172         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
3173
3174         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
3175              "in the rec is %u\n", num_clusters, (unsigned long long)p_blkno,
3176              first_hash);
3177
3178         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
3179                                       p_blkno, first_hash, num_clusters, xs);
3180
3181 out:
3182         return ret;
3183 }
3184
3185 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
3186                                        u64 blkno,
3187                                        u32 clusters,
3188                                        xattr_bucket_func *func,
3189                                        void *para)
3190 {
3191         int i, ret = 0;
3192         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
3193         u32 num_buckets = clusters * bpc;
3194         struct ocfs2_xattr_bucket *bucket;
3195
3196         bucket = ocfs2_xattr_bucket_new(inode);
3197         if (!bucket) {
3198                 mlog_errno(-ENOMEM);
3199                 return -ENOMEM;
3200         }
3201
3202         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
3203              clusters, (unsigned long long)blkno);
3204
3205         for (i = 0; i < num_buckets; i++, blkno += bucket->bu_blocks) {
3206                 ret = ocfs2_read_xattr_bucket(bucket, blkno);
3207                 if (ret) {
3208                         mlog_errno(ret);
3209                         break;
3210                 }
3211
3212                 /*
3213                  * The real bucket num in this series of blocks is stored
3214                  * in the 1st bucket.
3215                  */
3216                 if (i == 0)
3217                         num_buckets = le16_to_cpu(bucket_xh(bucket)->xh_num_buckets);
3218
3219                 mlog(0, "iterating xattr bucket %llu, first hash %u\n",
3220                      (unsigned long long)blkno,
3221                      le32_to_cpu(bucket_xh(bucket)->xh_entries[0].xe_name_hash));
3222                 if (func) {
3223                         ret = func(inode, bucket, para);
3224                         if (ret && ret != -ERANGE)
3225                                 mlog_errno(ret);
3226                         /* Fall through to bucket_relse() */
3227                 }
3228
3229                 ocfs2_xattr_bucket_relse(bucket);
3230                 if (ret)
3231                         break;
3232         }
3233
3234         ocfs2_xattr_bucket_free(bucket);
3235         return ret;
3236 }
3237
3238 struct ocfs2_xattr_tree_list {
3239         char *buffer;
3240         size_t buffer_size;
3241         size_t result;
3242 };
3243
3244 static int ocfs2_xattr_bucket_get_name_value(struct super_block *sb,
3245                                              struct ocfs2_xattr_header *xh,
3246                                              int index,
3247                                              int *block_off,
3248                                              int *new_offset)
3249 {
3250         u16 name_offset;
3251
3252         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
3253                 return -EINVAL;
3254
3255         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
3256
3257         *block_off = name_offset >> sb->s_blocksize_bits;
3258         *new_offset = name_offset % sb->s_blocksize;
3259
3260         return 0;
3261 }
3262
3263 static int ocfs2_list_xattr_bucket(struct inode *inode,
3264                                    struct ocfs2_xattr_bucket *bucket,
3265                                    void *para)
3266 {
3267         int ret = 0, type;
3268         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
3269         int i, block_off, new_offset;
3270         const char *prefix, *name;
3271
3272         for (i = 0 ; i < le16_to_cpu(bucket_xh(bucket)->xh_count); i++) {
3273                 struct ocfs2_xattr_entry *entry = &bucket_xh(bucket)->xh_entries[i];
3274                 type = ocfs2_xattr_get_type(entry);
3275                 prefix = ocfs2_xattr_prefix(type);
3276
3277                 if (prefix) {
3278                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
3279                                                                 bucket_xh(bucket),
3280                                                                 i,
3281                                                                 &block_off,
3282                                                                 &new_offset);
3283                         if (ret)
3284                                 break;
3285
3286                         name = (const char *)bucket_block(bucket, block_off) +
3287                                 new_offset;
3288                         ret = ocfs2_xattr_list_entry(xl->buffer,
3289                                                      xl->buffer_size,
3290                                                      &xl->result,
3291                                                      prefix, name,
3292                                                      entry->xe_name_len);
3293                         if (ret)
3294                                 break;
3295                 }
3296         }
3297
3298         return ret;
3299 }
3300
3301 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
3302                                              struct ocfs2_xattr_tree_root *xt,
3303                                              char *buffer,
3304                                              size_t buffer_size)
3305 {
3306         struct ocfs2_extent_list *el = &xt->xt_list;
3307         int ret = 0;
3308         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
3309         u64 p_blkno = 0;
3310         struct ocfs2_xattr_tree_list xl = {
3311                 .buffer = buffer,
3312                 .buffer_size = buffer_size,
3313                 .result = 0,
3314         };
3315
3316         if (le16_to_cpu(el->l_next_free_rec) == 0)
3317                 return 0;
3318
3319         while (name_hash > 0) {
3320                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
3321                                           &e_cpos, &num_clusters, el);
3322                 if (ret) {
3323                         mlog_errno(ret);
3324                         goto out;
3325                 }
3326
3327                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
3328                                                   ocfs2_list_xattr_bucket,
3329                                                   &xl);
3330                 if (ret) {
3331                         if (ret != -ERANGE)
3332                                 mlog_errno(ret);
3333                         goto out;
3334                 }
3335
3336                 if (e_cpos == 0)
3337                         break;
3338
3339                 name_hash = e_cpos - 1;
3340         }
3341
3342         ret = xl.result;
3343 out:
3344         return ret;
3345 }
3346
3347 static int cmp_xe(const void *a, const void *b)
3348 {
3349         const struct ocfs2_xattr_entry *l = a, *r = b;
3350         u32 l_hash = le32_to_cpu(l->xe_name_hash);
3351         u32 r_hash = le32_to_cpu(r->xe_name_hash);
3352
3353         if (l_hash > r_hash)
3354                 return 1;
3355         if (l_hash < r_hash)
3356                 return -1;
3357         return 0;
3358 }
3359
3360 static void swap_xe(void *a, void *b, int size)
3361 {
3362         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
3363
3364         tmp = *l;
3365         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
3366         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
3367 }
3368
3369 /*
3370  * When the ocfs2_xattr_block is filled up, new bucket will be created
3371  * and all the xattr entries will be moved to the new bucket.
3372  * The header goes at the start of the bucket, and the names+values are
3373  * filled from the end.  This is why *target starts as the last buffer.
3374  * Note: we need to sort the entries since they are not saved in order
3375  * in the ocfs2_xattr_block.
3376  */
3377 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
3378                                            struct buffer_head *xb_bh,
3379                                            struct ocfs2_xattr_bucket *bucket)
3380 {
3381         int i, blocksize = inode->i_sb->s_blocksize;
3382         int blks = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3383         u16 offset, size, off_change;
3384         struct ocfs2_xattr_entry *xe;
3385         struct ocfs2_xattr_block *xb =
3386                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
3387         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
3388         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
3389         u16 count = le16_to_cpu(xb_xh->xh_count);
3390         char *src = xb_bh->b_data;
3391         char *target = bucket_block(bucket, blks - 1);
3392
3393         mlog(0, "cp xattr from block %llu to bucket %llu\n",
3394              (unsigned long long)xb_bh->b_blocknr,
3395              (unsigned long long)bucket_blkno(bucket));
3396
3397         for (i = 0; i < blks; i++)
3398                 memset(bucket_block(bucket, i), 0, blocksize);
3399
3400         /*
3401          * Since the xe_name_offset is based on ocfs2_xattr_header,
3402          * there is a offset change corresponding to the change of
3403          * ocfs2_xattr_header's position.
3404          */
3405         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3406         xe = &xb_xh->xh_entries[count - 1];
3407         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
3408         size = blocksize - offset;
3409
3410         /* copy all the names and values. */
3411         memcpy(target + offset, src + offset, size);
3412
3413         /* Init new header now. */
3414         xh->xh_count = xb_xh->xh_count;
3415         xh->xh_num_buckets = cpu_to_le16(1);
3416         xh->xh_name_value_len = cpu_to_le16(size);
3417         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
3418
3419         /* copy all the entries. */
3420         target = bucket_block(bucket, 0);
3421         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
3422         size = count * sizeof(struct ocfs2_xattr_entry);
3423         memcpy(target + offset, (char *)xb_xh + offset, size);
3424
3425         /* Change the xe offset for all the xe because of the move. */
3426         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
3427                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
3428         for (i = 0; i < count; i++)
3429                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
3430
3431         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
3432              offset, size, off_change);
3433
3434         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
3435              cmp_xe, swap_xe);
3436 }
3437
3438 /*
3439  * After we move xattr from block to index btree, we have to
3440  * update ocfs2_xattr_search to the new xe and base.
3441  *
3442  * When the entry is in xattr block, xattr_bh indicates the storage place.
3443  * While if the entry is in index b-tree, "bucket" indicates the
3444  * real place of the xattr.
3445  */
3446 static void ocfs2_xattr_update_xattr_search(struct inode *inode,
3447                                             struct ocfs2_xattr_search *xs,
3448                                             struct buffer_head *old_bh)
3449 {
3450         char *buf = old_bh->b_data;
3451         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
3452         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
3453         int i;
3454
3455         xs->header = bucket_xh(xs->bucket);
3456         xs->base = bucket_block(xs->bucket, 0);
3457         xs->end = xs->base + inode->i_sb->s_blocksize;
3458
3459         if (xs->not_found)
3460                 return;
3461
3462         i = xs->here - old_xh->xh_entries;
3463         xs->here = &xs->header->xh_entries[i];
3464 }
3465
3466 static int ocfs2_xattr_create_index_block(struct inode *inode,
3467                                           struct ocfs2_xattr_search *xs,
3468                                           struct ocfs2_xattr_set_ctxt *ctxt)
3469 {
3470         int ret;
3471         u32 bit_off, len;
3472         u64 blkno;
3473         handle_t *handle = ctxt->handle;
3474         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3475         struct ocfs2_inode_info *oi = OCFS2_I(inode);
3476         struct buffer_head *xb_bh = xs->xattr_bh;
3477         struct ocfs2_xattr_block *xb =
3478                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3479         struct ocfs2_xattr_tree_root *xr;
3480         u16 xb_flags = le16_to_cpu(xb->xb_flags);
3481
3482         mlog(0, "create xattr index block for %llu\n",
3483              (unsigned long long)xb_bh->b_blocknr);
3484
3485         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
3486         BUG_ON(!xs->bucket);
3487
3488         /*
3489          * XXX:
3490          * We can use this lock for now, and maybe move to a dedicated mutex
3491          * if performance becomes a problem later.
3492          */
3493         down_write(&oi->ip_alloc_sem);
3494
3495         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), xb_bh,
3496                                       OCFS2_JOURNAL_ACCESS_WRITE);
3497         if (ret) {
3498                 mlog_errno(ret);
3499                 goto out;
3500         }
3501
3502         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac,
3503                                      1, 1, &bit_off, &len);
3504         if (ret) {
3505                 mlog_errno(ret);
3506                 goto out;
3507         }
3508
3509         /*
3510          * The bucket may spread in many blocks, and
3511          * we will only touch the 1st block and the last block
3512          * in the whole bucket(one for entry and one for data).
3513          */
3514         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
3515
3516         mlog(0, "allocate 1 cluster from %llu to xattr block\n",
3517              (unsigned long long)blkno);
3518
3519         ret = ocfs2_init_xattr_bucket(xs->bucket, blkno);
3520         if (ret) {
3521                 mlog_errno(ret);
3522                 goto out;
3523         }
3524
3525         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
3526                                                 OCFS2_JOURNAL_ACCESS_CREATE);
3527         if (ret) {
3528                 mlog_errno(ret);
3529                 goto out;
3530         }
3531
3532         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xs->bucket);
3533         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
3534
3535         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh);
3536
3537         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
3538         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
3539                offsetof(struct ocfs2_xattr_block, xb_attrs));
3540
3541         xr = &xb->xb_attrs.xb_root;
3542         xr->xt_clusters = cpu_to_le32(1);
3543         xr->xt_last_eb_blk = 0;
3544         xr->xt_list.l_tree_depth = 0;
3545         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
3546         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
3547
3548         xr->xt_list.l_recs[0].e_cpos = 0;
3549         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
3550         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
3551
3552         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
3553
3554         ocfs2_journal_dirty(handle, xb_bh);
3555
3556 out:
3557         up_write(&oi->ip_alloc_sem);
3558
3559         return ret;
3560 }
3561
3562 static int cmp_xe_offset(const void *a, const void *b)
3563 {
3564         const struct ocfs2_xattr_entry *l = a, *r = b;
3565         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
3566         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
3567
3568         if (l_name_offset < r_name_offset)
3569                 return 1;
3570         if (l_name_offset > r_name_offset)
3571                 return -1;
3572         return 0;
3573 }
3574
3575 /*
3576  * defrag a xattr bucket if we find that the bucket has some
3577  * holes beteen name/value pairs.
3578  * We will move all the name/value pairs to the end of the bucket
3579  * so that we can spare some space for insertion.
3580  */
3581 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
3582                                      handle_t *handle,
3583                                      struct ocfs2_xattr_bucket *bucket)
3584 {
3585         int ret, i;
3586         size_t end, offset, len, value_len;
3587         struct ocfs2_xattr_header *xh;
3588         char *entries, *buf, *bucket_buf = NULL;
3589         u64 blkno = bucket_blkno(bucket);
3590         u16 xh_free_start;
3591         size_t blocksize = inode->i_sb->s_blocksize;
3592         struct ocfs2_xattr_entry *xe;
3593
3594         /*
3595          * In order to make the operation more efficient and generic,
3596          * we copy all the blocks into a contiguous memory and do the
3597          * defragment there, so if anything is error, we will not touch
3598          * the real block.
3599          */
3600         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
3601         if (!bucket_buf) {
3602                 ret = -EIO;
3603                 goto out;
3604         }
3605
3606         buf = bucket_buf;
3607         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3608                 memcpy(buf, bucket_block(bucket, i), blocksize);
3609
3610         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
3611                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3612         if (ret < 0) {
3613                 mlog_errno(ret);
3614                 goto out;
3615         }
3616
3617         xh = (struct ocfs2_xattr_header *)bucket_buf;
3618         entries = (char *)xh->xh_entries;
3619         xh_free_start = le16_to_cpu(xh->xh_free_start);
3620
3621         mlog(0, "adjust xattr bucket in %llu, count = %u, "
3622              "xh_free_start = %u, xh_name_value_len = %u.\n",
3623              (unsigned long long)blkno, le16_to_cpu(xh->xh_count),
3624              xh_free_start, le16_to_cpu(xh->xh_name_value_len));
3625
3626         /*
3627          * sort all the entries by their offset.
3628          * the largest will be the first, so that we can
3629          * move them to the end one by one.
3630          */
3631         sort(entries, le16_to_cpu(xh->xh_count),
3632              sizeof(struct ocfs2_xattr_entry),
3633              cmp_xe_offset, swap_xe);
3634
3635         /* Move all name/values to the end of the bucket. */
3636         xe = xh->xh_entries;
3637         end = OCFS2_XATTR_BUCKET_SIZE;
3638         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
3639                 offset = le16_to_cpu(xe->xe_name_offset);
3640                 if (ocfs2_xattr_is_local(xe))
3641                         value_len = OCFS2_XATTR_SIZE(
3642                                         le64_to_cpu(xe->xe_value_size));
3643                 else
3644                         value_len = OCFS2_XATTR_ROOT_SIZE;
3645                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
3646
3647                 /*
3648                  * We must make sure that the name/value pair
3649                  * exist in the same block. So adjust end to
3650                  * the previous block end if needed.
3651                  */
3652                 if (((end - len) / blocksize !=
3653                         (end - 1) / blocksize))
3654                         end = end - end % blocksize;
3655
3656                 if (end > offset + len) {
3657                         memmove(bucket_buf + end - len,
3658                                 bucket_buf + offset, len);
3659                         xe->xe_name_offset = cpu_to_le16(end - len);
3660                 }
3661
3662                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3663                                 "bucket %llu\n", (unsigned long long)blkno);
3664
3665                 end -= len;
3666         }
3667
3668         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3669                         "bucket %llu\n", (unsigned long long)blkno);
3670
3671         if (xh_free_start == end)
3672                 goto out;
3673
3674         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3675         xh->xh_free_start = cpu_to_le16(end);
3676
3677         /* sort the entries by their name_hash. */
3678         sort(entries, le16_to_cpu(xh->xh_count),
3679              sizeof(struct ocfs2_xattr_entry),
3680              cmp_xe, swap_xe);
3681
3682         buf = bucket_buf;
3683         for (i = 0; i < bucket->bu_blocks; i++, buf += blocksize)
3684                 memcpy(bucket_block(bucket, i), buf, blocksize);
3685         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
3686
3687 out:
3688         kfree(bucket_buf);
3689         return ret;
3690 }
3691
3692 /*
3693  * prev_blkno points to the start of an existing extent.  new_blkno
3694  * points to a newly allocated extent.  Because we know each of our
3695  * clusters contains more than bucket, we can easily split one cluster
3696  * at a bucket boundary.  So we take the last cluster of the existing
3697  * extent and split it down the middle.  We move the last half of the
3698  * buckets in the last cluster of the existing extent over to the new
3699  * extent.
3700  *
3701  * first_bh is the buffer at prev_blkno so we can update the existing
3702  * extent's bucket count.  header_bh is the bucket were we were hoping
3703  * to insert our xattr.  If the bucket move places the target in the new
3704  * extent, we'll update first_bh and header_bh after modifying the old
3705  * extent.
3706  *
3707  * first_hash will be set as the 1st xe's name_hash in the new extent.
3708  */
3709 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3710                                                handle_t *handle,
3711                                                struct ocfs2_xattr_bucket *first,
3712                                                struct ocfs2_xattr_bucket *target,
3713                                                u64 new_blkno,
3714                                                u32 num_clusters,
3715                                                u32 *first_hash)
3716 {
3717         int ret;
3718         struct super_block *sb = inode->i_sb;
3719         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(sb);
3720         int num_buckets = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(sb));
3721         int to_move = num_buckets / 2;
3722         u64 src_blkno;
3723         u64 last_cluster_blkno = bucket_blkno(first) +
3724                 ((num_clusters - 1) * ocfs2_clusters_to_blocks(sb, 1));
3725
3726         BUG_ON(le16_to_cpu(bucket_xh(first)->xh_num_buckets) < num_buckets);
3727         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == OCFS2_SB(sb)->s_clustersize);
3728
3729         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3730              (unsigned long long)last_cluster_blkno, (unsigned long long)new_blkno);
3731
3732         ret = ocfs2_mv_xattr_buckets(inode, handle, bucket_blkno(first),
3733                                      last_cluster_blkno, new_blkno,
3734                                      to_move, first_hash);
3735         if (ret) {
3736                 mlog_errno(ret);
3737                 goto out;
3738         }
3739
3740         /* This is the first bucket that got moved */
3741         src_blkno = last_cluster_blkno + (to_move * blks_per_bucket);
3742
3743         /*
3744          * If the target bucket was part of the moved buckets, we need to
3745          * update first and target.
3746          */
3747         if (bucket_blkno(target) >= src_blkno) {
3748                 /* Find the block for the new target bucket */
3749                 src_blkno = new_blkno +
3750                         (bucket_blkno(target) - src_blkno);
3751
3752                 ocfs2_xattr_bucket_relse(first);
3753                 ocfs2_xattr_bucket_relse(target);
3754
3755                 /*
3756                  * These shouldn't fail - the buffers are in the
3757                  * journal from ocfs2_cp_xattr_bucket().
3758                  */
3759                 ret = ocfs2_read_xattr_bucket(first, new_blkno);
3760                 if (ret) {
3761                         mlog_errno(ret);
3762                         goto out;
3763                 }
3764                 ret = ocfs2_read_xattr_bucket(target, src_blkno);
3765                 if (ret)
3766                         mlog_errno(ret);
3767
3768         }
3769
3770 out:
3771         return ret;
3772 }
3773
3774 /*
3775  * Find the suitable pos when we divide a bucket into 2.
3776  * We have to make sure the xattrs with the same hash value exist
3777  * in the same bucket.
3778  *
3779  * If this ocfs2_xattr_header covers more than one hash value, find a
3780  * place where the hash value changes.  Try to find the most even split.
3781  * The most common case is that all entries have different hash values,
3782  * and the first check we make will find a place to split.
3783  */
3784 static int ocfs2_xattr_find_divide_pos(struct ocfs2_xattr_header *xh)
3785 {
3786         struct ocfs2_xattr_entry *entries = xh->xh_entries;
3787         int count = le16_to_cpu(xh->xh_count);
3788         int delta, middle = count / 2;
3789
3790         /*
3791          * We start at the middle.  Each step gets farther away in both
3792          * directions.  We therefore hit the change in hash value
3793          * nearest to the middle.  Note that this loop does not execute for
3794          * count < 2.
3795          */
3796         for (delta = 0; delta < middle; delta++) {
3797                 /* Let's check delta earlier than middle */
3798                 if (cmp_xe(&entries[middle - delta - 1],
3799                            &entries[middle - delta]))
3800                         return middle - delta;
3801
3802                 /* For even counts, don't walk off the end */
3803                 if ((middle + delta + 1) == count)
3804                         continue;
3805
3806                 /* Now try delta past middle */
3807                 if (cmp_xe(&entries[middle + delta],
3808                            &entries[middle + delta + 1]))
3809                         return middle + delta + 1;
3810         }
3811
3812         /* Every entry had the same hash */
3813         return count;
3814 }
3815
3816 /*
3817  * Move some xattrs in old bucket(blk) to new bucket(new_blk).
3818  * first_hash will record the 1st hash of the new bucket.
3819  *
3820  * Normally half of the xattrs will be moved.  But we have to make
3821  * sure that the xattrs with the same hash value are stored in the
3822  * same bucket. If all the xattrs in this bucket have the same hash
3823  * value, the new bucket will be initialized as an empty one and the
3824  * first_hash will be initialized as (hash_value+1).
3825  */
3826 static int ocfs2_divide_xattr_bucket(struct inode *inode,
3827                                     handle_t *handle,
3828                                     u64 blk,
3829                                     u64 new_blk,
3830                                     u32 *first_hash,
3831                                     int new_bucket_head)
3832 {
3833         int ret, i;
3834         int count, start, len, name_value_len = 0, xe_len, name_offset = 0;
3835         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
3836         struct ocfs2_xattr_header *xh;
3837         struct ocfs2_xattr_entry *xe;
3838         int blocksize = inode->i_sb->s_blocksize;
3839
3840         mlog(0, "move some of xattrs from bucket %llu to %llu\n",
3841              (unsigned long long)blk, (unsigned long long)new_blk);
3842
3843         s_bucket = ocfs2_xattr_bucket_new(inode);
3844         t_bucket = ocfs2_xattr_bucket_new(inode);
3845         if (!s_bucket || !t_bucket) {
3846                 ret = -ENOMEM;
3847                 mlog_errno(ret);
3848                 goto out;
3849         }
3850
3851         ret = ocfs2_read_xattr_bucket(s_bucket, blk);
3852         if (ret) {
3853                 mlog_errno(ret);
3854                 goto out;
3855         }
3856
3857         ret = ocfs2_xattr_bucket_journal_access(handle, s_bucket,
3858                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3859         if (ret) {
3860                 mlog_errno(ret);
3861                 goto out;
3862         }
3863
3864         /*
3865          * Even if !new_bucket_head, we're overwriting t_bucket.  Thus,
3866          * there's no need to read it.
3867          */
3868         ret = ocfs2_init_xattr_bucket(t_bucket, new_blk);
3869         if (ret) {
3870                 mlog_errno(ret);
3871                 goto out;
3872         }
3873
3874         /*
3875          * Hey, if we're overwriting t_bucket, what difference does
3876          * ACCESS_CREATE vs ACCESS_WRITE make?  See the comment in the
3877          * same part of ocfs2_cp_xattr_bucket().
3878          */
3879         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
3880                                                 new_bucket_head ?
3881                                                 OCFS2_JOURNAL_ACCESS_CREATE :
3882                                                 OCFS2_JOURNAL_ACCESS_WRITE);
3883         if (ret) {
3884                 mlog_errno(ret);
3885                 goto out;
3886         }
3887
3888         xh = bucket_xh(s_bucket);
3889         count = le16_to_cpu(xh->xh_count);
3890         start = ocfs2_xattr_find_divide_pos(xh);
3891
3892         if (start == count) {
3893                 xe = &xh->xh_entries[start-1];
3894
3895                 /*
3896                  * initialized a new empty bucket here.
3897                  * The hash value is set as one larger than
3898                  * that of the last entry in the previous bucket.
3899                  */
3900                 for (i = 0; i < t_bucket->bu_blocks; i++)
3901                         memset(bucket_block(t_bucket, i), 0, blocksize);
3902
3903                 xh = bucket_xh(t_bucket);
3904                 xh->xh_free_start = cpu_to_le16(blocksize);
3905                 xh->xh_entries[0].xe_name_hash = xe->xe_name_hash;
3906                 le32_add_cpu(&xh->xh_entries[0].xe_name_hash, 1);
3907
3908                 goto set_num_buckets;
3909         }
3910
3911         /* copy the whole bucket to the new first. */
3912         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
3913
3914         /* update the new bucket. */
3915         xh = bucket_xh(t_bucket);
3916
3917         /*
3918          * Calculate the total name/value len and xh_free_start for
3919          * the old bucket first.
3920          */
3921         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3922         name_value_len = 0;
3923         for (i = 0; i < start; i++) {
3924                 xe = &xh->xh_entries[i];
3925                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3926                 if (ocfs2_xattr_is_local(xe))
3927                         xe_len +=
3928                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3929                 else
3930                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3931                 name_value_len += xe_len;
3932                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3933                         name_offset = le16_to_cpu(xe->xe_name_offset);
3934         }
3935
3936         /*
3937          * Now begin the modification to the new bucket.
3938          *
3939          * In the new bucket, We just move the xattr entry to the beginning
3940          * and don't touch the name/value. So there will be some holes in the
3941          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3942          * called.
3943          */
3944         xe = &xh->xh_entries[start];
3945         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3946         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3947              (int)((char *)xe - (char *)xh),
3948              (int)((char *)xh->xh_entries - (char *)xh));
3949         memmove((char *)xh->xh_entries, (char *)xe, len);
3950         xe = &xh->xh_entries[count - start];
3951         len = sizeof(struct ocfs2_xattr_entry) * start;
3952         memset((char *)xe, 0, len);
3953
3954         le16_add_cpu(&xh->xh_count, -start);
3955         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3956
3957         /* Calculate xh_free_start for the new bucket. */
3958         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3959         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3960                 xe = &xh->xh_entries[i];
3961                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3962                 if (ocfs2_xattr_is_local(xe))
3963                         xe_len +=
3964                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3965                 else
3966                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3967                 if (le16_to_cpu(xe->xe_name_offset) <
3968                     le16_to_cpu(xh->xh_free_start))
3969                         xh->xh_free_start = xe->xe_name_offset;
3970         }
3971
3972 set_num_buckets:
3973         /* set xh->xh_num_buckets for the new xh. */
3974         if (new_bucket_head)
3975                 xh->xh_num_buckets = cpu_to_le16(1);
3976         else
3977                 xh->xh_num_buckets = 0;
3978
3979         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
3980
3981         /* store the first_hash of the new bucket. */
3982         if (first_hash)
3983                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3984
3985         /*
3986          * Now only update the 1st block of the old bucket.  If we
3987          * just added a new empty bucket, there is no need to modify
3988          * it.
3989          */
3990         if (start == count)
3991                 goto out;
3992
3993         xh = bucket_xh(s_bucket);
3994         memset(&xh->xh_entries[start], 0,
3995                sizeof(struct ocfs2_xattr_entry) * (count - start));
3996         xh->xh_count = cpu_to_le16(start);
3997         xh->xh_free_start = cpu_to_le16(name_offset);
3998         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3999
4000         ocfs2_xattr_bucket_journal_dirty(handle, s_bucket);
4001
4002 out:
4003         ocfs2_xattr_bucket_free(s_bucket);
4004         ocfs2_xattr_bucket_free(t_bucket);
4005
4006         return ret;
4007 }
4008
4009 /*
4010  * Copy xattr from one bucket to another bucket.
4011  *
4012  * The caller must make sure that the journal transaction
4013  * has enough space for journaling.
4014  */
4015 static int ocfs2_cp_xattr_bucket(struct inode *inode,
4016                                  handle_t *handle,
4017                                  u64 s_blkno,
4018                                  u64 t_blkno,
4019                                  int t_is_new)
4020 {
4021         int ret;
4022         struct ocfs2_xattr_bucket *s_bucket = NULL, *t_bucket = NULL;
4023
4024         BUG_ON(s_blkno == t_blkno);
4025
4026         mlog(0, "cp bucket %llu to %llu, target is %d\n",
4027              (unsigned long long)s_blkno, (unsigned long long)t_blkno,
4028              t_is_new);
4029
4030         s_bucket = ocfs2_xattr_bucket_new(inode);
4031         t_bucket = ocfs2_xattr_bucket_new(inode);
4032         if (!s_bucket || !t_bucket) {
4033                 ret = -ENOMEM;
4034                 mlog_errno(ret);
4035                 goto out;
4036         }
4037
4038         ret = ocfs2_read_xattr_bucket(s_bucket, s_blkno);
4039         if (ret)
4040                 goto out;
4041
4042         /*
4043          * Even if !t_is_new, we're overwriting t_bucket.  Thus,
4044          * there's no need to read it.
4045          */
4046         ret = ocfs2_init_xattr_bucket(t_bucket, t_blkno);
4047         if (ret)
4048                 goto out;
4049
4050         /*
4051          * Hey, if we're overwriting t_bucket, what difference does
4052          * ACCESS_CREATE vs ACCESS_WRITE make?  Well, if we allocated a new
4053          * cluster to fill, we came here from
4054          * ocfs2_mv_xattr_buckets(), and it is really new -
4055          * ACCESS_CREATE is required.  But we also might have moved data
4056          * out of t_bucket before extending back into it.
4057          * ocfs2_add_new_xattr_bucket() can do this - its call to
4058          * ocfs2_add_new_xattr_cluster() may have created a new extent
4059          * and copied out the end of the old extent.  Then it re-extends
4060          * the old extent back to create space for new xattrs.  That's
4061          * how we get here, and the bucket isn't really new.
4062          */
4063         ret = ocfs2_xattr_bucket_journal_access(handle, t_bucket,
4064                                                 t_is_new ?
4065                                                 OCFS2_JOURNAL_ACCESS_CREATE :
4066                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4067         if (ret)
4068                 goto out;
4069
4070         ocfs2_xattr_bucket_copy_data(t_bucket, s_bucket);
4071         ocfs2_xattr_bucket_journal_dirty(handle, t_bucket);
4072
4073 out:
4074         ocfs2_xattr_bucket_free(t_bucket);
4075         ocfs2_xattr_bucket_free(s_bucket);
4076
4077         return ret;
4078 }
4079
4080 /*
4081  * src_blk points to the start of an existing extent.  last_blk points to
4082  * last cluster in that extent.  to_blk points to a newly allocated
4083  * extent.  We copy the buckets from the cluster at last_blk to the new
4084  * extent.  If start_bucket is non-zero, we skip that many buckets before
4085  * we start copying.  The new extent's xh_num_buckets gets set to the
4086  * number of buckets we copied.  The old extent's xh_num_buckets shrinks
4087  * by the same amount.
4088  */
4089 static int ocfs2_mv_xattr_buckets(struct inode *inode, handle_t *handle,
4090                                   u64 src_blk, u64 last_blk, u64 to_blk,
4091                                   unsigned int start_bucket,
4092                                   u32 *first_hash)
4093 {
4094         int i, ret, credits;
4095         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4096         int blks_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4097         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
4098         struct ocfs2_xattr_bucket *old_first, *new_first;
4099
4100         mlog(0, "mv xattrs from cluster %llu to %llu\n",
4101              (unsigned long long)last_blk, (unsigned long long)to_blk);
4102
4103         BUG_ON(start_bucket >= num_buckets);
4104         if (start_bucket) {
4105                 num_buckets -= start_bucket;
4106                 last_blk += (start_bucket * blks_per_bucket);
4107         }
4108
4109         /* The first bucket of the original extent */
4110         old_first = ocfs2_xattr_bucket_new(inode);
4111         /* The first bucket of the new extent */
4112         new_first = ocfs2_xattr_bucket_new(inode);
4113         if (!old_first || !new_first) {
4114                 ret = -ENOMEM;
4115                 mlog_errno(ret);
4116                 goto out;
4117         }
4118
4119         ret = ocfs2_read_xattr_bucket(old_first, src_blk);
4120         if (ret) {
4121                 mlog_errno(ret);
4122                 goto out;
4123         }
4124
4125         /*
4126          * We need to update the first bucket of the old extent and all
4127          * the buckets going to the new extent.
4128          */
4129         credits = ((num_buckets + 1) * blks_per_bucket) +
4130                 handle->h_buffer_credits;
4131         ret = ocfs2_extend_trans(handle, credits);
4132         if (ret) {
4133                 mlog_errno(ret);
4134                 goto out;
4135         }
4136
4137         ret = ocfs2_xattr_bucket_journal_access(handle, old_first,
4138                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4139         if (ret) {
4140                 mlog_errno(ret);
4141                 goto out;
4142         }
4143
4144         for (i = 0; i < num_buckets; i++) {
4145                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4146                                             last_blk + (i * blks_per_bucket),
4147                                             to_blk + (i * blks_per_bucket),
4148                                             1);
4149                 if (ret) {
4150                         mlog_errno(ret);
4151                         goto out;
4152                 }
4153         }
4154
4155         /*
4156          * Get the new bucket ready before we dirty anything
4157          * (This actually shouldn't fail, because we already dirtied
4158          * it once in ocfs2_cp_xattr_bucket()).
4159          */
4160         ret = ocfs2_read_xattr_bucket(new_first, to_blk);
4161         if (ret) {
4162                 mlog_errno(ret);
4163                 goto out;
4164         }
4165         ret = ocfs2_xattr_bucket_journal_access(handle, new_first,
4166                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4167         if (ret) {
4168                 mlog_errno(ret);
4169                 goto out;
4170         }
4171
4172         /* Now update the headers */
4173         le16_add_cpu(&bucket_xh(old_first)->xh_num_buckets, -num_buckets);
4174         ocfs2_xattr_bucket_journal_dirty(handle, old_first);
4175
4176         bucket_xh(new_first)->xh_num_buckets = cpu_to_le16(num_buckets);
4177         ocfs2_xattr_bucket_journal_dirty(handle, new_first);
4178
4179         if (first_hash)
4180                 *first_hash = le32_to_cpu(bucket_xh(new_first)->xh_entries[0].xe_name_hash);
4181
4182 out:
4183         ocfs2_xattr_bucket_free(new_first);
4184         ocfs2_xattr_bucket_free(old_first);
4185         return ret;
4186 }
4187
4188 /*
4189  * Move some xattrs in this cluster to the new cluster.
4190  * This function should only be called when bucket size == cluster size.
4191  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
4192  */
4193 static int ocfs2_divide_xattr_cluster(struct inode *inode,
4194                                       handle_t *handle,
4195                                       u64 prev_blk,
4196                                       u64 new_blk,
4197                                       u32 *first_hash)
4198 {
4199         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4200         int ret, credits = 2 * blk_per_bucket + handle->h_buffer_credits;
4201
4202         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
4203
4204         ret = ocfs2_extend_trans(handle, credits);
4205         if (ret) {
4206                 mlog_errno(ret);
4207                 return ret;
4208         }
4209
4210         /* Move half of the xattr in start_blk to the next bucket. */
4211         return  ocfs2_divide_xattr_bucket(inode, handle, prev_blk,
4212                                           new_blk, first_hash, 1);
4213 }
4214
4215 /*
4216  * Move some xattrs from the old cluster to the new one since they are not
4217  * contiguous in ocfs2 xattr tree.
4218  *
4219  * new_blk starts a new separate cluster, and we will move some xattrs from
4220  * prev_blk to it. v_start will be set as the first name hash value in this
4221  * new cluster so that it can be used as e_cpos during tree insertion and
4222  * don't collide with our original b-tree operations. first_bh and header_bh
4223  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
4224  * to extend the insert bucket.
4225  *
4226  * The problem is how much xattr should we move to the new one and when should
4227  * we update first_bh and header_bh?
4228  * 1. If cluster size > bucket size, that means the previous cluster has more
4229  *    than 1 bucket, so just move half nums of bucket into the new cluster and
4230  *    update the first_bh and header_bh if the insert bucket has been moved
4231  *    to the new cluster.
4232  * 2. If cluster_size == bucket_size:
4233  *    a) If the previous extent rec has more than one cluster and the insert
4234  *       place isn't in the last cluster, copy the entire last cluster to the
4235  *       new one. This time, we don't need to upate the first_bh and header_bh
4236  *       since they will not be moved into the new cluster.
4237  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
4238  *       the new one. And we set the extend flag to zero if the insert place is
4239  *       moved into the new allocated cluster since no extend is needed.
4240  */
4241 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
4242                                             handle_t *handle,
4243                                             struct ocfs2_xattr_bucket *first,
4244                                             struct ocfs2_xattr_bucket *target,
4245                                             u64 new_blk,
4246                                             u32 prev_clusters,
4247                                             u32 *v_start,
4248                                             int *extend)
4249 {
4250         int ret;
4251
4252         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
4253              (unsigned long long)bucket_blkno(first), prev_clusters,
4254              (unsigned long long)new_blk);
4255
4256         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1) {
4257                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
4258                                                           handle,
4259                                                           first, target,
4260                                                           new_blk,
4261                                                           prev_clusters,
4262                                                           v_start);
4263                 if (ret)
4264                         mlog_errno(ret);
4265         } else {
4266                 /* The start of the last cluster in the first extent */
4267                 u64 last_blk = bucket_blkno(first) +
4268                         ((prev_clusters - 1) *
4269                          ocfs2_clusters_to_blocks(inode->i_sb, 1));
4270
4271                 if (prev_clusters > 1 && bucket_blkno(target) != last_blk) {
4272                         ret = ocfs2_mv_xattr_buckets(inode, handle,
4273                                                      bucket_blkno(first),
4274                                                      last_blk, new_blk, 0,
4275                                                      v_start);
4276                         if (ret)
4277                                 mlog_errno(ret);
4278                 } else {
4279                         ret = ocfs2_divide_xattr_cluster(inode, handle,
4280                                                          last_blk, new_blk,
4281                                                          v_start);
4282                         if (ret)
4283                                 mlog_errno(ret);
4284
4285                         if ((bucket_blkno(target) == last_blk) && extend)
4286                                 *extend = 0;
4287                 }
4288         }
4289
4290         return ret;
4291 }
4292
4293 /*
4294  * Add a new cluster for xattr storage.
4295  *
4296  * If the new cluster is contiguous with the previous one, it will be
4297  * appended to the same extent record, and num_clusters will be updated.
4298  * If not, we will insert a new extent for it and move some xattrs in
4299  * the last cluster into the new allocated one.
4300  * We also need to limit the maximum size of a btree leaf, otherwise we'll
4301  * lose the benefits of hashing because we'll have to search large leaves.
4302  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
4303  * if it's bigger).
4304  *
4305  * first_bh is the first block of the previous extent rec and header_bh
4306  * indicates the bucket we will insert the new xattrs. They will be updated
4307  * when the header_bh is moved into the new cluster.
4308  */
4309 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
4310                                        struct buffer_head *root_bh,
4311                                        struct ocfs2_xattr_bucket *first,
4312                                        struct ocfs2_xattr_bucket *target,
4313                                        u32 *num_clusters,
4314                                        u32 prev_cpos,
4315                                        int *extend,
4316                                        struct ocfs2_xattr_set_ctxt *ctxt)
4317 {
4318         int ret;
4319         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
4320         u32 prev_clusters = *num_clusters;
4321         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
4322         u64 block;
4323         handle_t *handle = ctxt->handle;
4324         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4325         struct ocfs2_extent_tree et;
4326
4327         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
4328              "previous xattr blkno = %llu\n",
4329              (unsigned long long)OCFS2_I(inode)->ip_blkno,
4330              prev_cpos, (unsigned long long)bucket_blkno(first));
4331
4332         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4333
4334         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4335                                       OCFS2_JOURNAL_ACCESS_WRITE);
4336         if (ret < 0) {
4337                 mlog_errno(ret);
4338                 goto leave;
4339         }
4340
4341         ret = __ocfs2_claim_clusters(osb, handle, ctxt->data_ac, 1,
4342                                      clusters_to_add, &bit_off, &num_bits);
4343         if (ret < 0) {
4344                 if (ret != -ENOSPC)
4345                         mlog_errno(ret);
4346                 goto leave;
4347         }
4348
4349         BUG_ON(num_bits > clusters_to_add);
4350
4351         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
4352         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
4353              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
4354
4355         if (bucket_blkno(first) + (prev_clusters * bpc) == block &&
4356             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
4357              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
4358                 /*
4359                  * If this cluster is contiguous with the old one and
4360                  * adding this new cluster, we don't surpass the limit of
4361                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
4362                  * initialized and used like other buckets in the previous
4363                  * cluster.
4364                  * So add it as a contiguous one. The caller will handle
4365                  * its init process.
4366                  */
4367                 v_start = prev_cpos + prev_clusters;
4368                 *num_clusters = prev_clusters + num_bits;
4369                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
4370                      num_bits);
4371         } else {
4372                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
4373                                                        handle,
4374                                                        first,
4375                                                        target,
4376                                                        block,
4377                                                        prev_clusters,
4378                                                        &v_start,
4379                                                        extend);
4380                 if (ret) {
4381                         mlog_errno(ret);
4382                         goto leave;
4383                 }
4384         }
4385
4386         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
4387              num_bits, (unsigned long long)block, v_start);
4388         ret = ocfs2_insert_extent(handle, &et, v_start, block,
4389                                   num_bits, 0, ctxt->meta_ac);
4390         if (ret < 0) {
4391                 mlog_errno(ret);
4392                 goto leave;
4393         }
4394
4395         ret = ocfs2_journal_dirty(handle, root_bh);
4396         if (ret < 0)
4397                 mlog_errno(ret);
4398
4399 leave:
4400         return ret;
4401 }
4402
4403 /*
4404  * We are given an extent.  'first' is the bucket at the very front of
4405  * the extent.  The extent has space for an additional bucket past
4406  * bucket_xh(first)->xh_num_buckets.  'target_blkno' is the block number
4407  * of the target bucket.  We wish to shift every bucket past the target
4408  * down one, filling in that additional space.  When we get back to the
4409  * target, we split the target between itself and the now-empty bucket
4410  * at target+1 (aka, target_blkno + blks_per_bucket).
4411  */
4412 static int ocfs2_extend_xattr_bucket(struct inode *inode,
4413                                      handle_t *handle,
4414                                      struct ocfs2_xattr_bucket *first,
4415                                      u64 target_blk,
4416                                      u32 num_clusters)
4417 {
4418         int ret, credits;
4419         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4420         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4421         u64 end_blk;
4422         u16 new_bucket = le16_to_cpu(bucket_xh(first)->xh_num_buckets);
4423
4424         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
4425              "from %llu, len = %u\n", (unsigned long long)target_blk,
4426              (unsigned long long)bucket_blkno(first), num_clusters);
4427
4428         /* The extent must have room for an additional bucket */
4429         BUG_ON(new_bucket >=
4430                (num_clusters * ocfs2_xattr_buckets_per_cluster(osb)));
4431
4432         /* end_blk points to the last existing bucket */
4433         end_blk = bucket_blkno(first) + ((new_bucket - 1) * blk_per_bucket);
4434
4435         /*
4436          * end_blk is the start of the last existing bucket.
4437          * Thus, (end_blk - target_blk) covers the target bucket and
4438          * every bucket after it up to, but not including, the last
4439          * existing bucket.  Then we add the last existing bucket, the
4440          * new bucket, and the first bucket (3 * blk_per_bucket).
4441          */
4442         credits = (end_blk - target_blk) + (3 * blk_per_bucket) +
4443                   handle->h_buffer_credits;
4444         ret = ocfs2_extend_trans(handle, credits);
4445         if (ret) {
4446                 mlog_errno(ret);
4447                 goto out;
4448         }
4449
4450         ret = ocfs2_xattr_bucket_journal_access(handle, first,
4451                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4452         if (ret) {
4453                 mlog_errno(ret);
4454                 goto out;
4455         }
4456
4457         while (end_blk != target_blk) {
4458                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
4459                                             end_blk + blk_per_bucket, 0);
4460                 if (ret)
4461                         goto out;
4462                 end_blk -= blk_per_bucket;
4463         }
4464
4465         /* Move half of the xattr in target_blkno to the next bucket. */
4466         ret = ocfs2_divide_xattr_bucket(inode, handle, target_blk,
4467                                         target_blk + blk_per_bucket, NULL, 0);
4468
4469         le16_add_cpu(&bucket_xh(first)->xh_num_buckets, 1);
4470         ocfs2_xattr_bucket_journal_dirty(handle, first);
4471
4472 out:
4473         return ret;
4474 }
4475
4476 /*
4477  * Add new xattr bucket in an extent record and adjust the buckets
4478  * accordingly.  xb_bh is the ocfs2_xattr_block, and target is the
4479  * bucket we want to insert into.
4480  *
4481  * In the easy case, we will move all the buckets after target down by
4482  * one. Half of target's xattrs will be moved to the next bucket.
4483  *
4484  * If current cluster is full, we'll allocate a new one.  This may not
4485  * be contiguous.  The underlying calls will make sure that there is
4486  * space for the insert, shifting buckets around if necessary.
4487  * 'target' may be moved by those calls.
4488  */
4489 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
4490                                       struct buffer_head *xb_bh,
4491                                       struct ocfs2_xattr_bucket *target,
4492                                       struct ocfs2_xattr_set_ctxt *ctxt)
4493 {
4494         struct ocfs2_xattr_block *xb =
4495                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4496         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
4497         struct ocfs2_extent_list *el = &xb_root->xt_list;
4498         u32 name_hash =
4499                 le32_to_cpu(bucket_xh(target)->xh_entries[0].xe_name_hash);
4500         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4501         int ret, num_buckets, extend = 1;
4502         u64 p_blkno;
4503         u32 e_cpos, num_clusters;
4504         /* The bucket at the front of the extent */
4505         struct ocfs2_xattr_bucket *first;
4506
4507         mlog(0, "Add new xattr bucket starting from %llu\n",
4508              (unsigned long long)bucket_blkno(target));
4509
4510         /* The first bucket of the original extent */
4511         first = ocfs2_xattr_bucket_new(inode);
4512         if (!first) {
4513                 ret = -ENOMEM;
4514                 mlog_errno(ret);
4515                 goto out;
4516         }
4517
4518         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
4519                                   &num_clusters, el);
4520         if (ret) {
4521                 mlog_errno(ret);
4522                 goto out;
4523         }
4524
4525         ret = ocfs2_read_xattr_bucket(first, p_blkno);
4526         if (ret) {
4527                 mlog_errno(ret);
4528                 goto out;
4529         }
4530
4531         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
4532         if (num_buckets == le16_to_cpu(bucket_xh(first)->xh_num_buckets)) {
4533                 /*
4534                  * This can move first+target if the target bucket moves
4535                  * to the new extent.
4536                  */
4537                 ret = ocfs2_add_new_xattr_cluster(inode,
4538                                                   xb_bh,
4539                                                   first,
4540                                                   target,
4541                                                   &num_clusters,
4542                                                   e_cpos,
4543                                                   &extend,
4544                                                   ctxt);
4545                 if (ret) {
4546                         mlog_errno(ret);
4547                         goto out;
4548                 }
4549         }
4550
4551         if (extend) {
4552                 ret = ocfs2_extend_xattr_bucket(inode,
4553                                                 ctxt->handle,
4554                                                 first,
4555                                                 bucket_blkno(target),
4556                                                 num_clusters);
4557                 if (ret)
4558                         mlog_errno(ret);
4559         }
4560
4561 out:
4562         ocfs2_xattr_bucket_free(first);
4563
4564         return ret;
4565 }
4566
4567 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
4568                                         struct ocfs2_xattr_bucket *bucket,
4569                                         int offs)
4570 {
4571         int block_off = offs >> inode->i_sb->s_blocksize_bits;
4572
4573         offs = offs % inode->i_sb->s_blocksize;
4574         return bucket_block(bucket, block_off) + offs;
4575 }
4576
4577 /*
4578  * Handle the normal xattr set, including replace, delete and new.
4579  *
4580  * Note: "local" indicates the real data's locality. So we can't
4581  * just its bucket locality by its length.
4582  */
4583 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
4584                                          struct ocfs2_xattr_info *xi,
4585                                          struct ocfs2_xattr_search *xs,
4586                                          u32 name_hash,
4587                                          int local)
4588 {
4589         struct ocfs2_xattr_entry *last, *xe;
4590         int name_len = strlen(xi->name);
4591         struct ocfs2_xattr_header *xh = xs->header;
4592         u16 count = le16_to_cpu(xh->xh_count), start;
4593         size_t blocksize = inode->i_sb->s_blocksize;
4594         char *val;
4595         size_t offs, size, new_size;
4596
4597         last = &xh->xh_entries[count];
4598         if (!xs->not_found) {
4599                 xe = xs->here;
4600                 offs = le16_to_cpu(xe->xe_name_offset);
4601                 if (ocfs2_xattr_is_local(xe))
4602                         size = OCFS2_XATTR_SIZE(name_len) +
4603                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4604                 else
4605                         size = OCFS2_XATTR_SIZE(name_len) +
4606                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4607
4608                 /*
4609                  * If the new value will be stored outside, xi->value has been
4610                  * initalized as an empty ocfs2_xattr_value_root, and the same
4611                  * goes with xi->value_len, so we can set new_size safely here.
4612                  * See ocfs2_xattr_set_in_bucket.
4613                  */
4614                 new_size = OCFS2_XATTR_SIZE(name_len) +
4615                            OCFS2_XATTR_SIZE(xi->value_len);
4616
4617                 le16_add_cpu(&xh->xh_name_value_len, -size);
4618                 if (xi->value) {
4619                         if (new_size > size)
4620                                 goto set_new_name_value;
4621
4622                         /* Now replace the old value with new one. */
4623                         if (local)
4624                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
4625                         else
4626                                 xe->xe_value_size = 0;
4627
4628                         val = ocfs2_xattr_bucket_get_val(inode,
4629                                                          xs->bucket, offs);
4630                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
4631                                size - OCFS2_XATTR_SIZE(name_len));
4632                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
4633                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
4634                                        xi->value, xi->value_len);
4635
4636                         le16_add_cpu(&xh->xh_name_value_len, new_size);
4637                         ocfs2_xattr_set_local(xe, local);
4638                         return;
4639                 } else {
4640                         /*
4641                          * Remove the old entry if there is more than one.
4642                          * We don't remove the last entry so that we can
4643                          * use it to indicate the hash value of the empty
4644                          * bucket.
4645                          */
4646                         last -= 1;
4647                         le16_add_cpu(&xh->xh_count, -1);
4648                         if (xh->xh_count) {
4649                                 memmove(xe, xe + 1,
4650                                         (void *)last - (void *)xe);
4651                                 memset(last, 0,
4652                                        sizeof(struct ocfs2_xattr_entry));
4653                         } else
4654                                 xh->xh_free_start =
4655                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4656
4657                         return;
4658                 }
4659         } else {
4660                 /* find a new entry for insert. */
4661                 int low = 0, high = count - 1, tmp;
4662                 struct ocfs2_xattr_entry *tmp_xe;
4663
4664                 while (low <= high && count) {
4665                         tmp = (low + high) / 2;
4666                         tmp_xe = &xh->xh_entries[tmp];
4667
4668                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4669                                 low = tmp + 1;
4670                         else if (name_hash <
4671                                  le32_to_cpu(tmp_xe->xe_name_hash))
4672                                 high = tmp - 1;
4673                         else {
4674                                 low = tmp;
4675                                 break;
4676                         }
4677                 }
4678
4679                 xe = &xh->xh_entries[low];
4680                 if (low != count)
4681                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4682
4683                 le16_add_cpu(&xh->xh_count, 1);
4684                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4685                 xe->xe_name_hash = cpu_to_le32(name_hash);
4686                 xe->xe_name_len = name_len;
4687                 ocfs2_xattr_set_type(xe, xi->name_index);
4688         }
4689
4690 set_new_name_value:
4691         /* Insert the new name+value. */
4692         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4693
4694         /*
4695          * We must make sure that the name/value pair
4696          * exists in the same block.
4697          */
4698         offs = le16_to_cpu(xh->xh_free_start);
4699         start = offs - size;
4700
4701         if (start >> inode->i_sb->s_blocksize_bits !=
4702             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4703                 offs = offs - offs % blocksize;
4704                 xh->xh_free_start = cpu_to_le16(offs);
4705         }
4706
4707         val = ocfs2_xattr_bucket_get_val(inode, xs->bucket, offs - size);
4708         xe->xe_name_offset = cpu_to_le16(offs - size);
4709
4710         memset(val, 0, size);
4711         memcpy(val, xi->name, name_len);
4712         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4713
4714         xe->xe_value_size = cpu_to_le64(xi->value_len);
4715         ocfs2_xattr_set_local(xe, local);
4716         xs->here = xe;
4717         le16_add_cpu(&xh->xh_free_start, -size);
4718         le16_add_cpu(&xh->xh_name_value_len, size);
4719
4720         return;
4721 }
4722
4723 /*
4724  * Set the xattr entry in the specified bucket.
4725  * The bucket is indicated by xs->bucket and it should have the enough
4726  * space for the xattr insertion.
4727  */
4728 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4729                                            handle_t *handle,
4730                                            struct ocfs2_xattr_info *xi,
4731                                            struct ocfs2_xattr_search *xs,
4732                                            u32 name_hash,
4733                                            int local)
4734 {
4735         int ret;
4736         u64 blkno;
4737
4738         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4739              (unsigned long)xi->value_len, xi->name_index,
4740              (unsigned long long)bucket_blkno(xs->bucket));
4741
4742         if (!xs->bucket->bu_bhs[1]) {
4743                 blkno = bucket_blkno(xs->bucket);
4744                 ocfs2_xattr_bucket_relse(xs->bucket);
4745                 ret = ocfs2_read_xattr_bucket(xs->bucket, blkno);
4746                 if (ret) {
4747                         mlog_errno(ret);
4748                         goto out;
4749                 }
4750         }
4751
4752         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4753                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4754         if (ret < 0) {
4755                 mlog_errno(ret);
4756                 goto out;
4757         }
4758
4759         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4760         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
4761
4762 out:
4763         return ret;
4764 }
4765
4766 /*
4767  * Truncate the specified xe_off entry in xattr bucket.
4768  * bucket is indicated by header_bh and len is the new length.
4769  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4770  *
4771  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4772  */
4773 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4774                                              struct ocfs2_xattr_bucket *bucket,
4775                                              int xe_off,
4776                                              int len,
4777                                              struct ocfs2_xattr_set_ctxt *ctxt)
4778 {
4779         int ret, offset;
4780         u64 value_blk;
4781         struct ocfs2_xattr_entry *xe;
4782         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
4783         size_t blocksize = inode->i_sb->s_blocksize;
4784         struct ocfs2_xattr_value_buf vb = {
4785                 .vb_access = ocfs2_journal_access,
4786         };
4787
4788         xe = &xh->xh_entries[xe_off];
4789
4790         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4791
4792         offset = le16_to_cpu(xe->xe_name_offset) +
4793                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4794
4795         value_blk = offset / blocksize;
4796
4797         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4798         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4799
4800         vb.vb_bh = bucket->bu_bhs[value_blk];
4801         BUG_ON(!vb.vb_bh);
4802
4803         vb.vb_xv = (struct ocfs2_xattr_value_root *)
4804                 (vb.vb_bh->b_data + offset % blocksize);
4805
4806         /*
4807          * From here on out we have to dirty the bucket.  The generic
4808          * value calls only modify one of the bucket's bhs, but we need
4809          * to send the bucket at once.  So if they error, they *could* have
4810          * modified something.  We have to assume they did, and dirty
4811          * the whole bucket.  This leaves us in a consistent state.
4812          */
4813         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4814              xe_off, (unsigned long long)bucket_blkno(bucket), len);
4815         ret = ocfs2_xattr_value_truncate(inode, &vb, len, ctxt);
4816         if (ret) {
4817                 mlog_errno(ret);
4818                 goto out;
4819         }
4820
4821         ret = ocfs2_xattr_bucket_journal_access(ctxt->handle, bucket,
4822                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4823         if (ret) {
4824                 mlog_errno(ret);
4825                 goto out;
4826         }
4827
4828         xe->xe_value_size = cpu_to_le64(len);
4829
4830         ocfs2_xattr_bucket_journal_dirty(ctxt->handle, bucket);
4831
4832 out:
4833         return ret;
4834 }
4835
4836 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4837                                         struct ocfs2_xattr_search *xs,
4838                                         int len,
4839                                         struct ocfs2_xattr_set_ctxt *ctxt)
4840 {
4841         int ret, offset;
4842         struct ocfs2_xattr_entry *xe = xs->here;
4843         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4844
4845         BUG_ON(!xs->bucket->bu_bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4846
4847         offset = xe - xh->xh_entries;
4848         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket,
4849                                                 offset, len, ctxt);
4850         if (ret)
4851                 mlog_errno(ret);
4852
4853         return ret;
4854 }
4855
4856 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4857                                                 handle_t *handle,
4858                                                 struct ocfs2_xattr_search *xs,
4859                                                 char *val,
4860                                                 int value_len)
4861 {
4862         int ret, offset, block_off;
4863         struct ocfs2_xattr_value_root *xv;
4864         struct ocfs2_xattr_entry *xe = xs->here;
4865         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4866         void *base;
4867         struct ocfs2_xattr_value_buf vb = {
4868                 .vb_access = ocfs2_journal_access,
4869         };
4870
4871         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4872
4873         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb, xh,
4874                                                 xe - xh->xh_entries,
4875                                                 &block_off,
4876                                                 &offset);
4877         if (ret) {
4878                 mlog_errno(ret);
4879                 goto out;
4880         }
4881
4882         base = bucket_block(xs->bucket, block_off);
4883         xv = (struct ocfs2_xattr_value_root *)(base + offset +
4884                  OCFS2_XATTR_SIZE(xe->xe_name_len));
4885
4886         vb.vb_xv = xv;
4887         vb.vb_bh = xs->bucket->bu_bhs[block_off];
4888         ret = __ocfs2_xattr_set_value_outside(inode, handle,
4889                                               &vb, val, value_len);
4890         if (ret)
4891                 mlog_errno(ret);
4892 out:
4893         return ret;
4894 }
4895
4896 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4897                                   struct buffer_head *root_bh,
4898                                   u64 blkno,
4899                                   u32 cpos,
4900                                   u32 len)
4901 {
4902         int ret;
4903         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4904         struct inode *tl_inode = osb->osb_tl_inode;
4905         handle_t *handle;
4906         struct ocfs2_xattr_block *xb =
4907                         (struct ocfs2_xattr_block *)root_bh->b_data;
4908         struct ocfs2_alloc_context *meta_ac = NULL;
4909         struct ocfs2_cached_dealloc_ctxt dealloc;
4910         struct ocfs2_extent_tree et;
4911
4912         ocfs2_init_xattr_tree_extent_tree(&et, INODE_CACHE(inode), root_bh);
4913
4914         ocfs2_init_dealloc_ctxt(&dealloc);
4915
4916         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4917              cpos, len, (unsigned long long)blkno);
4918
4919         ocfs2_remove_xattr_clusters_from_cache(INODE_CACHE(inode), blkno,
4920                                                len);
4921
4922         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4923         if (ret) {
4924                 mlog_errno(ret);
4925                 return ret;
4926         }
4927
4928         mutex_lock(&tl_inode->i_mutex);
4929
4930         if (ocfs2_truncate_log_needs_flush(osb)) {
4931                 ret = __ocfs2_flush_truncate_log(osb);
4932                 if (ret < 0) {
4933                         mlog_errno(ret);
4934                         goto out;
4935                 }
4936         }
4937
4938         handle = ocfs2_start_trans(osb, ocfs2_remove_extent_credits(osb->sb));
4939         if (IS_ERR(handle)) {
4940                 ret = -ENOMEM;
4941                 mlog_errno(ret);
4942                 goto out;
4943         }
4944
4945         ret = ocfs2_journal_access_xb(handle, INODE_CACHE(inode), root_bh,
4946                                       OCFS2_JOURNAL_ACCESS_WRITE);
4947         if (ret) {
4948                 mlog_errno(ret);
4949                 goto out_commit;
4950         }
4951
4952         ret = ocfs2_remove_extent(handle, &et, cpos, len, meta_ac,
4953                                   &dealloc);
4954         if (ret) {
4955                 mlog_errno(ret);
4956                 goto out_commit;
4957         }
4958
4959         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4960
4961         ret = ocfs2_journal_dirty(handle, root_bh);
4962         if (ret) {
4963                 mlog_errno(ret);
4964                 goto out_commit;
4965         }
4966
4967         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4968         if (ret)
4969                 mlog_errno(ret);
4970
4971 out_commit:
4972         ocfs2_commit_trans(osb, handle);
4973 out:
4974         ocfs2_schedule_truncate_log_flush(osb, 1);
4975
4976         mutex_unlock(&tl_inode->i_mutex);
4977
4978         if (meta_ac)
4979                 ocfs2_free_alloc_context(meta_ac);
4980
4981         ocfs2_run_deallocs(osb, &dealloc);
4982
4983         return ret;
4984 }
4985
4986 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4987                                          handle_t *handle,
4988                                          struct ocfs2_xattr_search *xs)
4989 {
4990         struct ocfs2_xattr_header *xh = bucket_xh(xs->bucket);
4991         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4992                                                 le16_to_cpu(xh->xh_count) - 1];
4993         int ret = 0;
4994
4995         ret = ocfs2_xattr_bucket_journal_access(handle, xs->bucket,
4996                                                 OCFS2_JOURNAL_ACCESS_WRITE);
4997         if (ret) {
4998                 mlog_errno(ret);
4999                 return;
5000         }
5001
5002         /* Remove the old entry. */
5003         memmove(xs->here, xs->here + 1,
5004                 (void *)last - (void *)xs->here);
5005         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
5006         le16_add_cpu(&xh->xh_count, -1);
5007
5008         ocfs2_xattr_bucket_journal_dirty(handle, xs->bucket);
5009 }
5010
5011 /*
5012  * Set the xattr name/value in the bucket specified in xs.
5013  *
5014  * As the new value in xi may be stored in the bucket or in an outside cluster,
5015  * we divide the whole process into 3 steps:
5016  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
5017  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
5018  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
5019  * 4. If the clusters for the new outside value can't be allocated, we need
5020  *    to free the xattr we allocated in set.
5021  */
5022 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
5023                                      struct ocfs2_xattr_info *xi,
5024                                      struct ocfs2_xattr_search *xs,
5025                                      struct ocfs2_xattr_set_ctxt *ctxt)
5026 {
5027         int ret, local = 1;
5028         size_t value_len;
5029         char *val = (char *)xi->value;
5030         struct ocfs2_xattr_entry *xe = xs->here;
5031         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
5032                                               strlen(xi->name));
5033
5034         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
5035                 /*
5036                  * We need to truncate the xattr storage first.
5037                  *
5038                  * If both the old and new value are stored to
5039                  * outside block, we only need to truncate
5040                  * the storage and then set the value outside.
5041                  *
5042                  * If the new value should be stored within block,
5043                  * we should free all the outside block first and
5044                  * the modification to the xattr block will be done
5045                  * by following steps.
5046                  */
5047                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5048                         value_len = xi->value_len;
5049                 else
5050                         value_len = 0;
5051
5052                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5053                                                            value_len,
5054                                                            ctxt);
5055                 if (ret)
5056                         goto out;
5057
5058                 if (value_len)
5059                         goto set_value_outside;
5060         }
5061
5062         value_len = xi->value_len;
5063         /* So we have to handle the inside block change now. */
5064         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
5065                 /*
5066                  * If the new value will be stored outside of block,
5067                  * initalize a new empty value root and insert it first.
5068                  */
5069                 local = 0;
5070                 xi->value = &def_xv;
5071                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
5072         }
5073
5074         ret = ocfs2_xattr_set_entry_in_bucket(inode, ctxt->handle, xi, xs,
5075                                               name_hash, local);
5076         if (ret) {
5077                 mlog_errno(ret);
5078                 goto out;
5079         }
5080
5081         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
5082                 goto out;
5083
5084         /* allocate the space now for the outside block storage. */
5085         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
5086                                                    value_len, ctxt);
5087         if (ret) {
5088                 mlog_errno(ret);
5089
5090                 if (xs->not_found) {
5091                         /*
5092                          * We can't allocate enough clusters for outside
5093                          * storage and we have allocated xattr already,
5094                          * so need to remove it.
5095                          */
5096                         ocfs2_xattr_bucket_remove_xs(inode, ctxt->handle, xs);
5097                 }
5098                 goto out;
5099         }
5100
5101 set_value_outside:
5102         ret = ocfs2_xattr_bucket_set_value_outside(inode, ctxt->handle,
5103                                                    xs, val, value_len);
5104 out:
5105         return ret;
5106 }
5107
5108 /*
5109  * check whether the xattr bucket is filled up with the same hash value.
5110  * If we want to insert the xattr with the same hash, return -ENOSPC.
5111  * If we want to insert a xattr with different hash value, go ahead
5112  * and ocfs2_divide_xattr_bucket will handle this.
5113  */
5114 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
5115                                               struct ocfs2_xattr_bucket *bucket,
5116                                               const char *name)
5117 {
5118         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5119         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
5120
5121         if (name_hash != le32_to_cpu(xh->xh_entries[0].xe_name_hash))
5122                 return 0;
5123
5124         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
5125             xh->xh_entries[0].xe_name_hash) {
5126                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
5127                      "hash = %u\n",
5128                      (unsigned long long)bucket_blkno(bucket),
5129                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
5130                 return -ENOSPC;
5131         }
5132
5133         return 0;
5134 }
5135
5136 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
5137                                              struct ocfs2_xattr_info *xi,
5138                                              struct ocfs2_xattr_search *xs,
5139                                              struct ocfs2_xattr_set_ctxt *ctxt)
5140 {
5141         struct ocfs2_xattr_header *xh;
5142         struct ocfs2_xattr_entry *xe;
5143         u16 count, header_size, xh_free_start;
5144         int free, max_free, need, old;
5145         size_t value_size = 0, name_len = strlen(xi->name);
5146         size_t blocksize = inode->i_sb->s_blocksize;
5147         int ret, allocation = 0;
5148
5149         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
5150
5151 try_again:
5152         xh = xs->header;
5153         count = le16_to_cpu(xh->xh_count);
5154         xh_free_start = le16_to_cpu(xh->xh_free_start);
5155         header_size = sizeof(struct ocfs2_xattr_header) +
5156                         count * sizeof(struct ocfs2_xattr_entry);
5157         max_free = OCFS2_XATTR_BUCKET_SIZE - header_size -
5158                 le16_to_cpu(xh->xh_name_value_len) - OCFS2_XATTR_HEADER_GAP;
5159
5160         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
5161                         "of %u which exceed block size\n",
5162                         (unsigned long long)bucket_blkno(xs->bucket),
5163                         header_size);
5164
5165         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
5166                 value_size = OCFS2_XATTR_ROOT_SIZE;
5167         else if (xi->value)
5168                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
5169
5170         if (xs->not_found)
5171                 need = sizeof(struct ocfs2_xattr_entry) +
5172                         OCFS2_XATTR_SIZE(name_len) + value_size;
5173         else {
5174                 need = value_size + OCFS2_XATTR_SIZE(name_len);
5175
5176                 /*
5177                  * We only replace the old value if the new length is smaller
5178                  * than the old one. Otherwise we will allocate new space in the
5179                  * bucket to store it.
5180                  */
5181                 xe = xs->here;
5182                 if (ocfs2_xattr_is_local(xe))
5183                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
5184                 else
5185                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
5186
5187                 if (old >= value_size)
5188                         need = 0;
5189         }
5190
5191         free = xh_free_start - header_size - OCFS2_XATTR_HEADER_GAP;
5192         /*
5193          * We need to make sure the new name/value pair
5194          * can exist in the same block.
5195          */
5196         if (xh_free_start % blocksize < need)
5197                 free -= xh_free_start % blocksize;
5198
5199         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
5200              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
5201              " %u\n", xs->not_found,
5202              (unsigned long long)bucket_blkno(xs->bucket),
5203              free, need, max_free, le16_to_cpu(xh->xh_free_start),
5204              le16_to_cpu(xh->xh_name_value_len));
5205
5206         if (free < need ||
5207             (xs->not_found &&
5208              count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb))) {
5209                 if (need <= max_free &&
5210                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
5211                         /*
5212                          * We can create the space by defragment. Since only the
5213                          * name/value will be moved, the xe shouldn't be changed
5214                          * in xs.
5215                          */
5216                         ret = ocfs2_defrag_xattr_bucket(inode, ctxt->handle,
5217                                                         xs->bucket);
5218                         if (ret) {
5219                                 mlog_errno(ret);
5220                                 goto out;
5221                         }
5222
5223                         xh_free_start = le16_to_cpu(xh->xh_free_start);
5224                         free = xh_free_start - header_size
5225                                 - OCFS2_XATTR_HEADER_GAP;
5226                         if (xh_free_start % blocksize < need)
5227                                 free -= xh_free_start % blocksize;
5228
5229                         if (free >= need)
5230                                 goto xattr_set;
5231
5232                         mlog(0, "Can't get enough space for xattr insert by "
5233                              "defragment. Need %u bytes, but we have %d, so "
5234                              "allocate new bucket for it.\n", need, free);
5235                 }
5236
5237                 /*
5238                  * We have to add new buckets or clusters and one
5239                  * allocation should leave us enough space for insert.
5240                  */
5241                 BUG_ON(allocation);
5242
5243                 /*
5244                  * We do not allow for overlapping ranges between buckets. And
5245                  * the maximum number of collisions we will allow for then is
5246                  * one bucket's worth, so check it here whether we need to
5247                  * add a new bucket for the insert.
5248                  */
5249                 ret = ocfs2_check_xattr_bucket_collision(inode,
5250                                                          xs->bucket,
5251                                                          xi->name);
5252                 if (ret) {
5253                         mlog_errno(ret);
5254                         goto out;
5255                 }
5256
5257                 ret = ocfs2_add_new_xattr_bucket(inode,
5258                                                  xs->xattr_bh,
5259                                                  xs->bucket,
5260                                                  ctxt);
5261                 if (ret) {
5262                         mlog_errno(ret);
5263                         goto out;
5264                 }
5265
5266                 /*
5267                  * ocfs2_add_new_xattr_bucket() will have updated
5268                  * xs->bucket if it moved, but it will not have updated
5269                  * any of the other search fields.  Thus, we drop it and
5270                  * re-search.  Everything should be cached, so it'll be
5271                  * quick.
5272                  */
5273                 ocfs2_xattr_bucket_relse(xs->bucket);
5274                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
5275                                                    xi->name_index,
5276                                                    xi->name, xs);
5277                 if (ret && ret != -ENODATA)
5278                         goto out;
5279                 xs->not_found = ret;
5280                 allocation = 1;
5281                 goto try_again;
5282         }
5283
5284 xattr_set:
5285         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs, ctxt);
5286 out:
5287         mlog_exit(ret);
5288         return ret;
5289 }
5290
5291 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
5292                                         struct ocfs2_xattr_bucket *bucket,
5293                                         void *para)
5294 {
5295         int ret = 0;
5296         struct ocfs2_xattr_header *xh = bucket_xh(bucket);
5297         u16 i;
5298         struct ocfs2_xattr_entry *xe;
5299         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5300         struct ocfs2_xattr_set_ctxt ctxt = {NULL, NULL,};
5301         int credits = ocfs2_remove_extent_credits(osb->sb) +
5302                 ocfs2_blocks_per_xattr_bucket(inode->i_sb);
5303
5304
5305         ocfs2_init_dealloc_ctxt(&ctxt.dealloc);
5306
5307         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
5308                 xe = &xh->xh_entries[i];
5309                 if (ocfs2_xattr_is_local(xe))
5310                         continue;
5311
5312                 ctxt.handle = ocfs2_start_trans(osb, credits);
5313                 if (IS_ERR(ctxt.handle)) {
5314                         ret = PTR_ERR(ctxt.handle);
5315                         mlog_errno(ret);
5316                         break;
5317                 }
5318
5319                 ret = ocfs2_xattr_bucket_value_truncate(inode, bucket,
5320                                                         i, 0, &ctxt);
5321
5322                 ocfs2_commit_trans(osb, ctxt.handle);
5323                 if (ret) {
5324                         mlog_errno(ret);
5325                         break;
5326                 }
5327         }
5328
5329         ocfs2_schedule_truncate_log_flush(osb, 1);
5330         ocfs2_run_deallocs(osb, &ctxt.dealloc);
5331         return ret;
5332 }
5333
5334 static int ocfs2_delete_xattr_index_block(struct inode *inode,
5335                                           struct buffer_head *xb_bh)
5336 {
5337         struct ocfs2_xattr_block *xb =
5338                         (struct ocfs2_xattr_block *)xb_bh->b_data;
5339         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
5340         int ret = 0;
5341         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
5342         u64 p_blkno;
5343
5344         if (le16_to_cpu(el->l_next_free_rec) == 0)
5345                 return 0;
5346
5347         while (name_hash > 0) {
5348                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
5349                                           &e_cpos, &num_clusters, el);
5350                 if (ret) {
5351                         mlog_errno(ret);
5352                         goto out;
5353                 }
5354
5355                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
5356                                                   ocfs2_delete_xattr_in_bucket,
5357                                                   NULL);
5358                 if (ret) {
5359                         mlog_errno(ret);
5360                         goto out;
5361                 }
5362
5363                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
5364                                              p_blkno, e_cpos, num_clusters);
5365                 if (ret) {
5366                         mlog_errno(ret);
5367                         break;
5368                 }
5369
5370                 if (e_cpos == 0)
5371                         break;
5372
5373                 name_hash = e_cpos - 1;
5374         }
5375
5376 out:
5377         return ret;
5378 }
5379
5380 /*
5381  * Whenever we modify a xattr value root in the bucket(e.g, CoW
5382  * or change the extent record flag), we need to recalculate
5383  * the metaecc for the whole bucket. So it is done here.
5384  *
5385  * Note:
5386  * We have to give the extra credits for the caller.
5387  */
5388 static int ocfs2_xattr_bucket_post_refcount(struct inode *inode,
5389                                             handle_t *handle,
5390                                             void *para)
5391 {
5392         int ret;
5393         struct ocfs2_xattr_bucket *bucket =
5394                         (struct ocfs2_xattr_bucket *)para;
5395
5396         ret = ocfs2_xattr_bucket_journal_access(handle, bucket,
5397                                                 OCFS2_JOURNAL_ACCESS_WRITE);
5398         if (ret) {
5399                 mlog_errno(ret);
5400                 return ret;
5401         }
5402
5403         ocfs2_xattr_bucket_journal_dirty(handle, bucket);
5404
5405         return 0;
5406 }
5407
5408 /*
5409  * Special action we need if the xattr value is refcounted.
5410  *
5411  * 1. If the xattr is refcounted, lock the tree.
5412  * 2. CoW the xattr if we are setting the new value and the value
5413  *    will be stored outside.
5414  * 3. In other case, decrease_refcount will work for us, so just
5415  *    lock the refcount tree, calculate the meta and credits is OK.
5416  *
5417  * We have to do CoW before ocfs2_init_xattr_set_ctxt since
5418  * currently CoW is a completed transaction, while this function
5419  * will also lock the allocators and let us deadlock. So we will
5420  * CoW the whole xattr value.
5421  */
5422 static int ocfs2_prepare_refcount_xattr(struct inode *inode,
5423                                         struct ocfs2_dinode *di,
5424                                         struct ocfs2_xattr_info *xi,
5425                                         struct ocfs2_xattr_search *xis,
5426                                         struct ocfs2_xattr_search *xbs,
5427                                         struct ocfs2_refcount_tree **ref_tree,
5428                                         int *meta_add,
5429                                         int *credits)
5430 {
5431         int ret = 0;
5432         struct ocfs2_xattr_block *xb;
5433         struct ocfs2_xattr_entry *xe;
5434         char *base;
5435         u32 p_cluster, num_clusters;
5436         unsigned int ext_flags;
5437         int name_offset, name_len;
5438         struct ocfs2_xattr_value_buf vb;
5439         struct ocfs2_xattr_bucket *bucket = NULL;
5440         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5441         struct ocfs2_post_refcount refcount;
5442         struct ocfs2_post_refcount *p = NULL;
5443         struct buffer_head *ref_root_bh = NULL;
5444
5445         if (!xis->not_found) {
5446                 xe = xis->here;
5447                 name_offset = le16_to_cpu(xe->xe_name_offset);
5448                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5449                 base = xis->base;
5450                 vb.vb_bh = xis->inode_bh;
5451                 vb.vb_access = ocfs2_journal_access_di;
5452         } else {
5453                 int i, block_off = 0;
5454                 xb = (struct ocfs2_xattr_block *)xbs->xattr_bh->b_data;
5455                 xe = xbs->here;
5456                 name_offset = le16_to_cpu(xe->xe_name_offset);
5457                 name_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
5458                 i = xbs->here - xbs->header->xh_entries;
5459
5460                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
5461                         ret = ocfs2_xattr_bucket_get_name_value(inode->i_sb,
5462                                                         bucket_xh(xbs->bucket),
5463                                                         i, &block_off,
5464                                                         &name_offset);
5465                         if (ret) {
5466                                 mlog_errno(ret);
5467                                 goto out;
5468                         }
5469                         base = bucket_block(xbs->bucket, block_off);
5470                         vb.vb_bh = xbs->bucket->bu_bhs[block_off];
5471                         vb.vb_access = ocfs2_journal_access;
5472
5473                         if (ocfs2_meta_ecc(osb)) {
5474                                 /*create parameters for ocfs2_post_refcount. */
5475                                 bucket = xbs->bucket;
5476                                 refcount.credits = bucket->bu_blocks;
5477                                 refcount.para = bucket;
5478                                 refcount.func =
5479                                         ocfs2_xattr_bucket_post_refcount;
5480                                 p = &refcount;
5481                         }
5482                 } else {
5483                         base = xbs->base;
5484                         vb.vb_bh = xbs->xattr_bh;
5485                         vb.vb_access = ocfs2_journal_access_xb;
5486                 }
5487         }
5488
5489         if (ocfs2_xattr_is_local(xe))
5490                 goto out;
5491
5492         vb.vb_xv = (struct ocfs2_xattr_value_root *)
5493                                 (base + name_offset + name_len);
5494
5495         ret = ocfs2_xattr_get_clusters(inode, 0, &p_cluster,
5496                                        &num_clusters, &vb.vb_xv->xr_list,
5497                                        &ext_flags);
5498         if (ret) {
5499                 mlog_errno(ret);
5500                 goto out;
5501         }
5502
5503         /*
5504          * We just need to check the 1st extent record, since we always
5505          * CoW the whole xattr. So there shouldn't be a xattr with
5506          * some REFCOUNT extent recs after the 1st one.
5507          */
5508         if (!(ext_flags & OCFS2_EXT_REFCOUNTED))
5509                 goto out;
5510
5511         ret = ocfs2_lock_refcount_tree(osb, le64_to_cpu(di->i_refcount_loc),
5512                                        1, ref_tree, &ref_root_bh);
5513         if (ret) {
5514                 mlog_errno(ret);
5515                 goto out;
5516         }
5517
5518         /*
5519          * If we are deleting the xattr or the new size will be stored inside,
5520          * cool, leave it there, the xattr truncate process will remove them
5521          * for us(it still needs the refcount tree lock and the meta, credits).
5522          * And the worse case is that every cluster truncate will split the
5523          * refcount tree, and make the original extent become 3. So we will need
5524          * 2 * cluster more extent recs at most.
5525          */
5526         if (!xi->value || xi->value_len <= OCFS2_XATTR_INLINE_SIZE) {
5527
5528                 ret = ocfs2_refcounted_xattr_delete_need(inode,
5529                                                          &(*ref_tree)->rf_ci,
5530                                                          ref_root_bh, vb.vb_xv,
5531                                                          meta_add, credits);
5532                 if (ret)
5533                         mlog_errno(ret);
5534                 goto out;
5535         }
5536
5537         ret = ocfs2_refcount_cow_xattr(inode, di, &vb,
5538                                        *ref_tree, ref_root_bh, 0,
5539                                        le32_to_cpu(vb.vb_xv->xr_clusters), p);
5540         if (ret)
5541                 mlog_errno(ret);
5542
5543 out:
5544         brelse(ref_root_bh);
5545         return ret;
5546 }
5547
5548 /*
5549  * 'security' attributes support
5550  */
5551 static size_t ocfs2_xattr_security_list(struct inode *inode, char *list,
5552                                         size_t list_size, const char *name,
5553                                         size_t name_len)
5554 {
5555         const size_t prefix_len = XATTR_SECURITY_PREFIX_LEN;
5556         const size_t total_len = prefix_len + name_len + 1;
5557
5558         if (list && total_len <= list_size) {
5559                 memcpy(list, XATTR_SECURITY_PREFIX, prefix_len);
5560                 memcpy(list + prefix_len, name, name_len);
5561                 list[prefix_len + name_len] = '\0';
5562         }
5563         return total_len;
5564 }
5565
5566 static int ocfs2_xattr_security_get(struct inode *inode, const char *name,
5567                                     void *buffer, size_t size)
5568 {
5569         if (strcmp(name, "") == 0)
5570                 return -EINVAL;
5571         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_SECURITY, name,
5572                                buffer, size);
5573 }
5574
5575 static int ocfs2_xattr_security_set(struct inode *inode, const char *name,
5576                                     const void *value, size_t size, int flags)
5577 {
5578         if (strcmp(name, "") == 0)
5579                 return -EINVAL;
5580
5581         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_SECURITY, name, value,
5582                                size, flags);
5583 }
5584
5585 int ocfs2_init_security_get(struct inode *inode,
5586                             struct inode *dir,
5587                             struct ocfs2_security_xattr_info *si)
5588 {
5589         /* check whether ocfs2 support feature xattr */
5590         if (!ocfs2_supports_xattr(OCFS2_SB(dir->i_sb)))
5591                 return -EOPNOTSUPP;
5592         return security_inode_init_security(inode, dir, &si->name, &si->value,
5593                                             &si->value_len);
5594 }
5595
5596 int ocfs2_init_security_set(handle_t *handle,
5597                             struct inode *inode,
5598                             struct buffer_head *di_bh,
5599                             struct ocfs2_security_xattr_info *si,
5600                             struct ocfs2_alloc_context *xattr_ac,
5601                             struct ocfs2_alloc_context *data_ac)
5602 {
5603         return ocfs2_xattr_set_handle(handle, inode, di_bh,
5604                                      OCFS2_XATTR_INDEX_SECURITY,
5605                                      si->name, si->value, si->value_len, 0,
5606                                      xattr_ac, data_ac);
5607 }
5608
5609 struct xattr_handler ocfs2_xattr_security_handler = {
5610         .prefix = XATTR_SECURITY_PREFIX,
5611         .list   = ocfs2_xattr_security_list,
5612         .get    = ocfs2_xattr_security_get,
5613         .set    = ocfs2_xattr_security_set,
5614 };
5615
5616 /*
5617  * 'trusted' attributes support
5618  */
5619 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
5620                                        size_t list_size, const char *name,
5621                                        size_t name_len)
5622 {
5623         const size_t prefix_len = XATTR_TRUSTED_PREFIX_LEN;
5624         const size_t total_len = prefix_len + name_len + 1;
5625
5626         if (list && total_len <= list_size) {
5627                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
5628                 memcpy(list + prefix_len, name, name_len);
5629                 list[prefix_len + name_len] = '\0';
5630         }
5631         return total_len;
5632 }
5633
5634 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
5635                                    void *buffer, size_t size)
5636 {
5637         if (strcmp(name, "") == 0)
5638                 return -EINVAL;
5639         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
5640                                buffer, size);
5641 }
5642
5643 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
5644                                    const void *value, size_t size, int flags)
5645 {
5646         if (strcmp(name, "") == 0)
5647                 return -EINVAL;
5648
5649         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
5650                                size, flags);
5651 }
5652
5653 struct xattr_handler ocfs2_xattr_trusted_handler = {
5654         .prefix = XATTR_TRUSTED_PREFIX,
5655         .list   = ocfs2_xattr_trusted_list,
5656         .get    = ocfs2_xattr_trusted_get,
5657         .set    = ocfs2_xattr_trusted_set,
5658 };
5659
5660 /*
5661  * 'user' attributes support
5662  */
5663 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
5664                                     size_t list_size, const char *name,
5665                                     size_t name_len)
5666 {
5667         const size_t prefix_len = XATTR_USER_PREFIX_LEN;
5668         const size_t total_len = prefix_len + name_len + 1;
5669         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5670
5671         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5672                 return 0;
5673
5674         if (list && total_len <= list_size) {
5675                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
5676                 memcpy(list + prefix_len, name, name_len);
5677                 list[prefix_len + name_len] = '\0';
5678         }
5679         return total_len;
5680 }
5681
5682 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
5683                                 void *buffer, size_t size)
5684 {
5685         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5686
5687         if (strcmp(name, "") == 0)
5688                 return -EINVAL;
5689         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5690                 return -EOPNOTSUPP;
5691         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
5692                                buffer, size);
5693 }
5694
5695 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
5696                                 const void *value, size_t size, int flags)
5697 {
5698         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
5699
5700         if (strcmp(name, "") == 0)
5701                 return -EINVAL;
5702         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
5703                 return -EOPNOTSUPP;
5704
5705         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
5706                                size, flags);
5707 }
5708
5709 struct xattr_handler ocfs2_xattr_user_handler = {
5710         .prefix = XATTR_USER_PREFIX,
5711         .list   = ocfs2_xattr_user_list,
5712         .get    = ocfs2_xattr_user_get,
5713         .set    = ocfs2_xattr_user_set,
5714 };