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) 2008 Oracle.  All rights reserved.
   7  *
   8  * CREDITS:
   9  * Lots of code in this file is taken from ext3.
  10  *
  11  * This program is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2 of the License, or (at your option) any later version.
  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  * You should have received a copy of the GNU General Public
  22  * License along with this program; if not, write to the
  23  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  24  * Boston, MA 021110-1307, USA.
  25  */
  26
  27 #include <linux/capability.h>
  28 #include <linux/fs.h>
  29 #include <linux/types.h>
  30 #include <linux/slab.h>
  31 #include <linux/highmem.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/uio.h>
  34 #include <linux/sched.h>
  35 #include <linux/splice.h>
  36 #include <linux/mount.h>
  37 #include <linux/writeback.h>
  38 #include <linux/falloc.h>
  39 #include <linux/sort.h>
  40
  41 #define MLOG_MASK_PREFIX ML_XATTR
  42 #include <cluster/masklog.h>
  43
  44 #include "ocfs2.h"
  45 #include "alloc.h"
  46 #include "dlmglue.h"
  47 #include "file.h"
  48 #include "symlink.h"
  49 #include "sysfile.h"
  50 #include "inode.h"
  51 #include "journal.h"
  52 #include "ocfs2_fs.h"
  53 #include "suballoc.h"
  54 #include "uptodate.h"
  55 #include "buffer_head_io.h"
  56 #include "super.h"
  57 #include "xattr.h"
  58
  59
  60 struct ocfs2_xattr_def_value_root {
  61         struct ocfs2_xattr_value_root   xv;
  62         struct ocfs2_extent_rec         er;
  63 };
  64
  65 struct ocfs2_xattr_bucket {
  66         struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  67         struct ocfs2_xattr_header *xh;
  68 };
  69
  70 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  71 #define OCFS2_XATTR_INLINE_SIZE 80
  72
  73 static struct ocfs2_xattr_def_value_root def_xv = {
  74         .xv.xr_list.l_count = cpu_to_le16(1),
  75 };
  76
  77 struct xattr_handler *ocfs2_xattr_handlers[] = {
  78         &ocfs2_xattr_user_handler,
  79         &ocfs2_xattr_trusted_handler,
  80         NULL
  81 };
  82
  83 static struct xattr_handler *ocfs2_xattr_handler_map[] = {
  84         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
  85         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
  86 };
  87
  88 struct ocfs2_xattr_info {
  89         int name_index;
  90         const char *name;
  91         const void *value;
  92         size_t value_len;
  93 };
  94
  95 struct ocfs2_xattr_search {
  96         struct buffer_head *inode_bh;
  97         /*
  98          * xattr_bh point to the block buffer head which has extended attribute
  99          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 100          */
 101         struct buffer_head *xattr_bh;
 102         struct ocfs2_xattr_header *header;
 103         struct ocfs2_xattr_bucket bucket;
 104         void *base;
 105         void *end;
 106         struct ocfs2_xattr_entry *here;
 107         int not_found;
 108 };
 109
 110 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 111                                              struct ocfs2_xattr_header *xh,
 112                                              int index,
 113                                              int *block_off,
 114                                              int *new_offset);
 115
 116 static int ocfs2_xattr_index_block_find(struct inode *inode,
 117                                         struct buffer_head *root_bh,
 118                                         int name_index,
 119                                         const char *name,
 120                                         struct ocfs2_xattr_search *xs);
 121
 122 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 123                                         struct ocfs2_xattr_tree_root *xt,
 124                                         char *buffer,
 125                                         size_t buffer_size);
 126
 127 static int ocfs2_xattr_create_index_block(struct inode *inode,
 128                                           struct ocfs2_xattr_search *xs);
 129
 130 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 131                                              struct ocfs2_xattr_info *xi,
 132                                              struct ocfs2_xattr_search *xs);
 133
 134 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 135                                           struct buffer_head *xb_bh);
 136
 137 static inline struct xattr_handler *ocfs2_xattr_handler(int name_index)
 138 {
 139         struct xattr_handler *handler = NULL;
 140
 141         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 142                 handler = ocfs2_xattr_handler_map[name_index];
 143
 144         return handler;
 145 }
 146
 147 static inline u32 ocfs2_xattr_name_hash(struct inode *inode,
 148                                         char *prefix,
 149                                         int prefix_len,
 150                                         char *name,
 151                                         int name_len)
 152 {
 153         /* Get hash value of uuid from super block */
 154         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 155         int i;
 156
 157         /* hash extended attribute prefix */
 158         for (i = 0; i < prefix_len; i++) {
 159                 hash = (hash << OCFS2_HASH_SHIFT) ^
 160                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 161                        *prefix++;
 162         }
 163         /* hash extended attribute name */
 164         for (i = 0; i < name_len; i++) {
 165                 hash = (hash << OCFS2_HASH_SHIFT) ^
 166                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 167                        *name++;
 168         }
 169
 170         return hash;
 171 }
 172
 173 /*
 174  * ocfs2_xattr_hash_entry()
 175  *
 176  * Compute the hash of an extended attribute.
 177  */
 178 static void ocfs2_xattr_hash_entry(struct inode *inode,
 179                                    struct ocfs2_xattr_header *header,
 180                                    struct ocfs2_xattr_entry *entry)
 181 {
 182         u32 hash = 0;
 183         struct xattr_handler *handler =
 184                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
 185         char *prefix = handler->prefix;
 186         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 187         int prefix_len = strlen(handler->prefix);
 188
 189         hash = ocfs2_xattr_name_hash(inode, prefix, prefix_len, name,
 190                                      entry->xe_name_len);
 191         entry->xe_name_hash = cpu_to_le32(hash);
 192
 193         return;
 194 }
 195
 196 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 197                                          u32 clusters_to_add,
 198                                          struct buffer_head *xattr_bh,
 199                                          struct ocfs2_xattr_value_root *xv)
 200 {
 201         int status = 0;
 202         int restart_func = 0;
 203         int credits = 0;
 204         handle_t *handle = NULL;
 205         struct ocfs2_alloc_context *data_ac = NULL;
 206         struct ocfs2_alloc_context *meta_ac = NULL;
 207         enum ocfs2_alloc_restarted why;
 208         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 209         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
 210         struct ocfs2_extent_tree et;
 211
 212         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 213
 214         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
 215
 216 restart_all:
 217
 218         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 219                                        &data_ac, &meta_ac);
 220         if (status) {
 221                 mlog_errno(status);
 222                 goto leave;
 223         }
 224
 225         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
 226                                             clusters_to_add);
 227         handle = ocfs2_start_trans(osb, credits);
 228         if (IS_ERR(handle)) {
 229                 status = PTR_ERR(handle);
 230                 handle = NULL;
 231                 mlog_errno(status);
 232                 goto leave;
 233         }
 234
 235 restarted_transaction:
 236         status = ocfs2_journal_access(handle, inode, xattr_bh,
 237                                       OCFS2_JOURNAL_ACCESS_WRITE);
 238         if (status < 0) {
 239                 mlog_errno(status);
 240                 goto leave;
 241         }
 242
 243         prev_clusters = le32_to_cpu(xv->xr_clusters);
 244         status = ocfs2_add_clusters_in_btree(osb,
 245                                              inode,
 246                                              &logical_start,
 247                                              clusters_to_add,
 248                                              0,
 249                                              &et,
 250                                              handle,
 251                                              data_ac,
 252                                              meta_ac,
 253                                              &why);
 254         if ((status < 0) && (status != -EAGAIN)) {
 255                 if (status != -ENOSPC)
 256                         mlog_errno(status);
 257                 goto leave;
 258         }
 259
 260         status = ocfs2_journal_dirty(handle, xattr_bh);
 261         if (status < 0) {
 262                 mlog_errno(status);
 263                 goto leave;
 264         }
 265
 266         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
 267
 268         if (why != RESTART_NONE && clusters_to_add) {
 269                 if (why == RESTART_META) {
 270                         mlog(0, "restarting function.\n");
 271                         restart_func = 1;
 272                 } else {
 273                         BUG_ON(why != RESTART_TRANS);
 274
 275                         mlog(0, "restarting transaction.\n");
 276                         /* TODO: This can be more intelligent. */
 277                         credits = ocfs2_calc_extend_credits(osb->sb,
 278                                                             et.et_root_el,
 279                                                             clusters_to_add);
 280                         status = ocfs2_extend_trans(handle, credits);
 281                         if (status < 0) {
 282                                 /* handle still has to be committed at
 283                                  * this point. */
 284                                 status = -ENOMEM;
 285                                 mlog_errno(status);
 286                                 goto leave;
 287                         }
 288                         goto restarted_transaction;
 289                 }
 290         }
 291
 292 leave:
 293         if (handle) {
 294                 ocfs2_commit_trans(osb, handle);
 295                 handle = NULL;
 296         }
 297         if (data_ac) {
 298                 ocfs2_free_alloc_context(data_ac);
 299                 data_ac = NULL;
 300         }
 301         if (meta_ac) {
 302                 ocfs2_free_alloc_context(meta_ac);
 303                 meta_ac = NULL;
 304         }
 305         if ((!status) && restart_func) {
 306                 restart_func = 0;
 307                 goto restart_all;
 308         }
 309
 310         return status;
 311 }
 312
 313 static int __ocfs2_remove_xattr_range(struct inode *inode,
 314                                       struct buffer_head *root_bh,
 315                                       struct ocfs2_xattr_value_root *xv,
 316                                       u32 cpos, u32 phys_cpos, u32 len,
 317                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
 318 {
 319         int ret;
 320         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 321         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 322         struct inode *tl_inode = osb->osb_tl_inode;
 323         handle_t *handle;
 324         struct ocfs2_alloc_context *meta_ac = NULL;
 325         struct ocfs2_extent_tree et;
 326
 327         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
 328
 329         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
 330         if (ret) {
 331                 mlog_errno(ret);
 332                 return ret;
 333         }
 334
 335         mutex_lock(&tl_inode->i_mutex);
 336
 337         if (ocfs2_truncate_log_needs_flush(osb)) {
 338                 ret = __ocfs2_flush_truncate_log(osb);
 339                 if (ret < 0) {
 340                         mlog_errno(ret);
 341                         goto out;
 342                 }
 343         }
 344
 345         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
 346         if (IS_ERR(handle)) {
 347                 ret = PTR_ERR(handle);
 348                 mlog_errno(ret);
 349                 goto out;
 350         }
 351
 352         ret = ocfs2_journal_access(handle, inode, root_bh,
 353                                    OCFS2_JOURNAL_ACCESS_WRITE);
 354         if (ret) {
 355                 mlog_errno(ret);
 356                 goto out_commit;
 357         }
 358
 359         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
 360                                   dealloc);
 361         if (ret) {
 362                 mlog_errno(ret);
 363                 goto out_commit;
 364         }
 365
 366         le32_add_cpu(&xv->xr_clusters, -len);
 367
 368         ret = ocfs2_journal_dirty(handle, root_bh);
 369         if (ret) {
 370                 mlog_errno(ret);
 371                 goto out_commit;
 372         }
 373
 374         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
 375         if (ret)
 376                 mlog_errno(ret);
 377
 378 out_commit:
 379         ocfs2_commit_trans(osb, handle);
 380 out:
 381         mutex_unlock(&tl_inode->i_mutex);
 382
 383         if (meta_ac)
 384                 ocfs2_free_alloc_context(meta_ac);
 385
 386         return ret;
 387 }
 388
 389 static int ocfs2_xattr_shrink_size(struct inode *inode,
 390                                    u32 old_clusters,
 391                                    u32 new_clusters,
 392                                    struct buffer_head *root_bh,
 393                                    struct ocfs2_xattr_value_root *xv)
 394 {
 395         int ret = 0;
 396         u32 trunc_len, cpos, phys_cpos, alloc_size;
 397         u64 block;
 398         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 399         struct ocfs2_cached_dealloc_ctxt dealloc;
 400
 401         ocfs2_init_dealloc_ctxt(&dealloc);
 402
 403         if (old_clusters <= new_clusters)
 404                 return 0;
 405
 406         cpos = new_clusters;
 407         trunc_len = old_clusters - new_clusters;
 408         while (trunc_len) {
 409                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 410                                                &alloc_size, &xv->xr_list);
 411                 if (ret) {
 412                         mlog_errno(ret);
 413                         goto out;
 414                 }
 415
 416                 if (alloc_size > trunc_len)
 417                         alloc_size = trunc_len;
 418
 419                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
 420                                                  phys_cpos, alloc_size,
 421                                                  &dealloc);
 422                 if (ret) {
 423                         mlog_errno(ret);
 424                         goto out;
 425                 }
 426
 427                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 428                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 429                                                        alloc_size);
 430                 cpos += alloc_size;
 431                 trunc_len -= alloc_size;
 432         }
 433
 434 out:
 435         ocfs2_schedule_truncate_log_flush(osb, 1);
 436         ocfs2_run_deallocs(osb, &dealloc);
 437
 438         return ret;
 439 }
 440
 441 static int ocfs2_xattr_value_truncate(struct inode *inode,
 442                                       struct buffer_head *root_bh,
 443                                       struct ocfs2_xattr_value_root *xv,
 444                                       int len)
 445 {
 446         int ret;
 447         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 448         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
 449
 450         if (new_clusters == old_clusters)
 451                 return 0;
 452
 453         if (new_clusters > old_clusters)
 454                 ret = ocfs2_xattr_extend_allocation(inode,
 455                                                     new_clusters - old_clusters,
 456                                                     root_bh, xv);
 457         else
 458                 ret = ocfs2_xattr_shrink_size(inode,
 459                                               old_clusters, new_clusters,
 460                                               root_bh, xv);
 461
 462         return ret;
 463 }
 464
 465 static int ocfs2_xattr_list_entries(struct inode *inode,
 466                                     struct ocfs2_xattr_header *header,
 467                                     char *buffer, size_t buffer_size)
 468 {
 469         size_t rest = buffer_size;
 470         int i;
 471
 472         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 473                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 474                 struct xattr_handler *handler =
 475                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
 476
 477                 if (handler) {
 478                         size_t size = handler->list(inode, buffer, rest,
 479                                         ((char *)header +
 480                                         le16_to_cpu(entry->xe_name_offset)),
 481                                         entry->xe_name_len);
 482                         if (buffer) {
 483                                 if (size > rest)
 484                                         return -ERANGE;
 485                                 buffer += size;
 486                         }
 487                         rest -= size;
 488                 }
 489         }
 490
 491         return buffer_size - rest;
 492 }
 493
 494 static int ocfs2_xattr_ibody_list(struct inode *inode,
 495                                   struct ocfs2_dinode *di,
 496                                   char *buffer,
 497                                   size_t buffer_size)
 498 {
 499         struct ocfs2_xattr_header *header = NULL;
 500         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 501         int ret = 0;
 502
 503         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 504                 return ret;
 505
 506         header = (struct ocfs2_xattr_header *)
 507                  ((void *)di + inode->i_sb->s_blocksize -
 508                  le16_to_cpu(di->i_xattr_inline_size));
 509
 510         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 511
 512         return ret;
 513 }
 514
 515 static int ocfs2_xattr_block_list(struct inode *inode,
 516                                   struct ocfs2_dinode *di,
 517                                   char *buffer,
 518                                   size_t buffer_size)
 519 {
 520         struct buffer_head *blk_bh = NULL;
 521         struct ocfs2_xattr_block *xb;
 522         int ret = 0;
 523
 524         if (!di->i_xattr_loc)
 525                 return ret;
 526
 527         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 528                                le64_to_cpu(di->i_xattr_loc),
 529                                &blk_bh, OCFS2_BH_CACHED, inode);
 530         if (ret < 0) {
 531                 mlog_errno(ret);
 532                 return ret;
 533         }
 534         /*Verify the signature of xattr block*/
 535         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 536                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 537                 ret = -EFAULT;
 538                 goto cleanup;
 539         }
 540
 541         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 542
 543         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 544                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 545                 ret = ocfs2_xattr_list_entries(inode, header,
 546                                                buffer, buffer_size);
 547         } else {
 548                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 549                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 550                                                    buffer, buffer_size);
 551         }
 552 cleanup:
 553         brelse(blk_bh);
 554
 555         return ret;
 556 }
 557
 558 ssize_t ocfs2_listxattr(struct dentry *dentry,
 559                         char *buffer,
 560                         size_t size)
 561 {
 562         int ret = 0, i_ret = 0, b_ret = 0;
 563         struct buffer_head *di_bh = NULL;
 564         struct ocfs2_dinode *di = NULL;
 565         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 566
 567         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 568                 return -EOPNOTSUPP;
 569
 570         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 571                 return ret;
 572
 573         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 574         if (ret < 0) {
 575                 mlog_errno(ret);
 576                 return ret;
 577         }
 578
 579         di = (struct ocfs2_dinode *)di_bh->b_data;
 580
 581         down_read(&oi->ip_xattr_sem);
 582         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 583         if (i_ret < 0)
 584                 b_ret = 0;
 585         else {
 586                 if (buffer) {
 587                         buffer += i_ret;
 588                         size -= i_ret;
 589                 }
 590                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 591                                                buffer, size);
 592                 if (b_ret < 0)
 593                         i_ret = 0;
 594         }
 595         up_read(&oi->ip_xattr_sem);
 596         ocfs2_inode_unlock(dentry->d_inode, 0);
 597
 598         brelse(di_bh);
 599
 600         return i_ret + b_ret;
 601 }
 602
 603 static int ocfs2_xattr_find_entry(int name_index,
 604                                   const char *name,
 605                                   struct ocfs2_xattr_search *xs)
 606 {
 607         struct ocfs2_xattr_entry *entry;
 608         size_t name_len;
 609         int i, cmp = 1;
 610
 611         if (name == NULL)
 612                 return -EINVAL;
 613
 614         name_len = strlen(name);
 615         entry = xs->here;
 616         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 617                 cmp = name_index - ocfs2_xattr_get_type(entry);
 618                 if (!cmp)
 619                         cmp = name_len - entry->xe_name_len;
 620                 if (!cmp)
 621                         cmp = memcmp(name, (xs->base +
 622                                      le16_to_cpu(entry->xe_name_offset)),
 623                                      name_len);
 624                 if (cmp == 0)
 625                         break;
 626                 entry += 1;
 627         }
 628         xs->here = entry;
 629
 630         return cmp ? -ENODATA : 0;
 631 }
 632
 633 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 634                                          struct ocfs2_xattr_value_root *xv,
 635                                          void *buffer,
 636                                          size_t len)
 637 {
 638         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 639         u64 blkno;
 640         int i, ret = 0;
 641         size_t cplen, blocksize;
 642         struct buffer_head *bh = NULL;
 643         struct ocfs2_extent_list *el;
 644
 645         el = &xv->xr_list;
 646         clusters = le32_to_cpu(xv->xr_clusters);
 647         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 648         blocksize = inode->i_sb->s_blocksize;
 649
 650         cpos = 0;
 651         while (cpos < clusters) {
 652                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 653                                                &num_clusters, el);
 654                 if (ret) {
 655                         mlog_errno(ret);
 656                         goto out;
 657                 }
 658
 659                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 660                 /* Copy ocfs2_xattr_value */
 661                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 662                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 663                                                &bh, OCFS2_BH_CACHED, inode);
 664                         if (ret) {
 665                                 mlog_errno(ret);
 666                                 goto out;
 667                         }
 668
 669                         cplen = len >= blocksize ? blocksize : len;
 670                         memcpy(buffer, bh->b_data, cplen);
 671                         len -= cplen;
 672                         buffer += cplen;
 673
 674                         brelse(bh);
 675                         bh = NULL;
 676                         if (len == 0)
 677                                 break;
 678                 }
 679                 cpos += num_clusters;
 680         }
 681 out:
 682         return ret;
 683 }
 684
 685 static int ocfs2_xattr_ibody_get(struct inode *inode,
 686                                  int name_index,
 687                                  const char *name,
 688                                  void *buffer,
 689                                  size_t buffer_size,
 690                                  struct ocfs2_xattr_search *xs)
 691 {
 692         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 693         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 694         struct ocfs2_xattr_value_root *xv;
 695         size_t size;
 696         int ret = 0;
 697
 698         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 699                 return -ENODATA;
 700
 701         xs->end = (void *)di + inode->i_sb->s_blocksize;
 702         xs->header = (struct ocfs2_xattr_header *)
 703                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
 704         xs->base = (void *)xs->header;
 705         xs->here = xs->header->xh_entries;
 706
 707         ret = ocfs2_xattr_find_entry(name_index, name, xs);
 708         if (ret)
 709                 return ret;
 710         size = le64_to_cpu(xs->here->xe_value_size);
 711         if (buffer) {
 712                 if (size > buffer_size)
 713                         return -ERANGE;
 714                 if (ocfs2_xattr_is_local(xs->here)) {
 715                         memcpy(buffer, (void *)xs->base +
 716                                le16_to_cpu(xs->here->xe_name_offset) +
 717                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
 718                 } else {
 719                         xv = (struct ocfs2_xattr_value_root *)
 720                                 (xs->base + le16_to_cpu(
 721                                  xs->here->xe_name_offset) +
 722                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
 723                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 724                                                             buffer, size);
 725                         if (ret < 0) {
 726                                 mlog_errno(ret);
 727                                 return ret;
 728                         }
 729                 }
 730         }
 731
 732         return size;
 733 }
 734
 735 static int ocfs2_xattr_block_get(struct inode *inode,
 736                                  int name_index,
 737                                  const char *name,
 738                                  void *buffer,
 739                                  size_t buffer_size,
 740                                  struct ocfs2_xattr_search *xs)
 741 {
 742         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 743         struct buffer_head *blk_bh = NULL;
 744         struct ocfs2_xattr_block *xb;
 745         struct ocfs2_xattr_value_root *xv;
 746         size_t size;
 747         int ret = -ENODATA, name_offset, name_len, block_off, i;
 748
 749         if (!di->i_xattr_loc)
 750                 return ret;
 751
 752         memset(&xs->bucket, 0, sizeof(xs->bucket));
 753
 754         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 755                                le64_to_cpu(di->i_xattr_loc),
 756                                &blk_bh, OCFS2_BH_CACHED, inode);
 757         if (ret < 0) {
 758                 mlog_errno(ret);
 759                 return ret;
 760         }
 761         /*Verify the signature of xattr block*/
 762         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 763                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 764                 ret = -EFAULT;
 765                 goto cleanup;
 766         }
 767
 768         xs->xattr_bh = blk_bh;
 769         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 770
 771         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 772                 xs->header = &xb->xb_attrs.xb_header;
 773                 xs->base = (void *)xs->header;
 774                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
 775                 xs->here = xs->header->xh_entries;
 776
 777                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
 778         } else
 779                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
 780                                                    name_index,
 781                                                    name, xs);
 782
 783         if (ret)
 784                 goto cleanup;
 785         size = le64_to_cpu(xs->here->xe_value_size);
 786         if (buffer) {
 787                 ret = -ERANGE;
 788                 if (size > buffer_size)
 789                         goto cleanup;
 790
 791                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
 792                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
 793                 i = xs->here - xs->header->xh_entries;
 794
 795                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
 796                         ret = ocfs2_xattr_bucket_get_name_value(inode,
 797                                                                 xs->bucket.xh,
 798                                                                 i,
 799                                                                 &block_off,
 800                                                                 &name_offset);
 801                         xs->base = xs->bucket.bhs[block_off]->b_data;
 802                 }
 803                 if (ocfs2_xattr_is_local(xs->here)) {
 804                         memcpy(buffer, (void *)xs->base +
 805                                name_offset + name_len, size);
 806                 } else {
 807                         xv = (struct ocfs2_xattr_value_root *)
 808                                 (xs->base + name_offset + name_len);
 809                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 810                                                             buffer, size);
 811                         if (ret < 0) {
 812                                 mlog_errno(ret);
 813                                 goto cleanup;
 814                         }
 815                 }
 816         }
 817         ret = size;
 818 cleanup:
 819         for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
 820                 brelse(xs->bucket.bhs[i]);
 821         memset(&xs->bucket, 0, sizeof(xs->bucket));
 822
 823         brelse(blk_bh);
 824         return ret;
 825 }
 826
 827 /* ocfs2_xattr_get()
 828  *
 829  * Copy an extended attribute into the buffer provided.
 830  * Buffer is NULL to compute the size of buffer required.
 831  */
 832 int ocfs2_xattr_get(struct inode *inode,
 833                     int name_index,
 834                     const char *name,
 835                     void *buffer,
 836                     size_t buffer_size)
 837 {
 838         int ret;
 839         struct ocfs2_dinode *di = NULL;
 840         struct buffer_head *di_bh = NULL;
 841         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 842         struct ocfs2_xattr_search xis = {
 843                 .not_found = -ENODATA,
 844         };
 845         struct ocfs2_xattr_search xbs = {
 846                 .not_found = -ENODATA,
 847         };
 848
 849         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
 850                 return -EOPNOTSUPP;
 851
 852         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 853                 ret = -ENODATA;
 854
 855         ret = ocfs2_inode_lock(inode, &di_bh, 0);
 856         if (ret < 0) {
 857                 mlog_errno(ret);
 858                 return ret;
 859         }
 860         xis.inode_bh = xbs.inode_bh = di_bh;
 861         di = (struct ocfs2_dinode *)di_bh->b_data;
 862
 863         down_read(&oi->ip_xattr_sem);
 864         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
 865                                     buffer_size, &xis);
 866         if (ret == -ENODATA)
 867                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
 868                                             buffer_size, &xbs);
 869         up_read(&oi->ip_xattr_sem);
 870         ocfs2_inode_unlock(inode, 0);
 871
 872         brelse(di_bh);
 873
 874         return ret;
 875 }
 876
 877 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
 878                                            struct ocfs2_xattr_value_root *xv,
 879                                            const void *value,
 880                                            int value_len)
 881 {
 882         int ret = 0, i, cp_len, credits;
 883         u16 blocksize = inode->i_sb->s_blocksize;
 884         u32 p_cluster, num_clusters;
 885         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 886         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
 887         u64 blkno;
 888         struct buffer_head *bh = NULL;
 889         handle_t *handle;
 890
 891         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
 892
 893         credits = clusters * bpc;
 894         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
 895         if (IS_ERR(handle)) {
 896                 ret = PTR_ERR(handle);
 897                 mlog_errno(ret);
 898                 goto out;
 899         }
 900
 901         while (cpos < clusters) {
 902                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 903                                                &num_clusters, &xv->xr_list);
 904                 if (ret) {
 905                         mlog_errno(ret);
 906                         goto out_commit;
 907                 }
 908
 909                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 910
 911                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 912                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 913                                                &bh, OCFS2_BH_CACHED, inode);
 914                         if (ret) {
 915                                 mlog_errno(ret);
 916                                 goto out_commit;
 917                         }
 918
 919                         ret = ocfs2_journal_access(handle,
 920                                                    inode,
 921                                                    bh,
 922                                                    OCFS2_JOURNAL_ACCESS_WRITE);
 923                         if (ret < 0) {
 924                                 mlog_errno(ret);
 925                                 goto out_commit;
 926                         }
 927
 928                         cp_len = value_len > blocksize ? blocksize : value_len;
 929                         memcpy(bh->b_data, value, cp_len);
 930                         value_len -= cp_len;
 931                         value += cp_len;
 932                         if (cp_len < blocksize)
 933                                 memset(bh->b_data + cp_len, 0,
 934                                        blocksize - cp_len);
 935
 936                         ret = ocfs2_journal_dirty(handle, bh);
 937                         if (ret < 0) {
 938                                 mlog_errno(ret);
 939                                 goto out_commit;
 940                         }
 941                         brelse(bh);
 942                         bh = NULL;
 943
 944                         /*
 945                          * XXX: do we need to empty all the following
 946                          * blocks in this cluster?
 947                          */
 948                         if (!value_len)
 949                                 break;
 950                 }
 951                 cpos += num_clusters;
 952         }
 953 out_commit:
 954         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 955 out:
 956         brelse(bh);
 957
 958         return ret;
 959 }
 960
 961 static int ocfs2_xattr_cleanup(struct inode *inode,
 962                                struct ocfs2_xattr_info *xi,
 963                                struct ocfs2_xattr_search *xs,
 964                                size_t offs)
 965 {
 966         handle_t *handle = NULL;
 967         int ret = 0;
 968         size_t name_len = strlen(xi->name);
 969         void *val = xs->base + offs;
 970         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 971
 972         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
 973                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
 974         if (IS_ERR(handle)) {
 975                 ret = PTR_ERR(handle);
 976                 mlog_errno(ret);
 977                 goto out;
 978         }
 979         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
 980                                    OCFS2_JOURNAL_ACCESS_WRITE);
 981         if (ret) {
 982                 mlog_errno(ret);
 983                 goto out_commit;
 984         }
 985         /* Decrease xattr count */
 986         le16_add_cpu(&xs->header->xh_count, -1);
 987         /* Remove the xattr entry and tree root which has already be set*/
 988         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
 989         memset(val, 0, size);
 990
 991         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
 992         if (ret < 0)
 993                 mlog_errno(ret);
 994 out_commit:
 995         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 996 out:
 997         return ret;
 998 }
 999
1000 static int ocfs2_xattr_update_entry(struct inode *inode,
1001                                     struct ocfs2_xattr_info *xi,
1002                                     struct ocfs2_xattr_search *xs,
1003                                     size_t offs)
1004 {
1005         handle_t *handle = NULL;
1006         int ret = 0;
1007
1008         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1009                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1010         if (IS_ERR(handle)) {
1011                 ret = PTR_ERR(handle);
1012                 mlog_errno(ret);
1013                 goto out;
1014         }
1015         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1016                                    OCFS2_JOURNAL_ACCESS_WRITE);
1017         if (ret) {
1018                 mlog_errno(ret);
1019                 goto out_commit;
1020         }
1021
1022         xs->here->xe_name_offset = cpu_to_le16(offs);
1023         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1024         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1025                 ocfs2_xattr_set_local(xs->here, 1);
1026         else
1027                 ocfs2_xattr_set_local(xs->here, 0);
1028         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1029
1030         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1031         if (ret < 0)
1032                 mlog_errno(ret);
1033 out_commit:
1034         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1035 out:
1036         return ret;
1037 }
1038
1039 /*
1040  * ocfs2_xattr_set_value_outside()
1041  *
1042  * Set large size value in B tree.
1043  */
1044 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1045                                          struct ocfs2_xattr_info *xi,
1046                                          struct ocfs2_xattr_search *xs,
1047                                          size_t offs)
1048 {
1049         size_t name_len = strlen(xi->name);
1050         void *val = xs->base + offs;
1051         struct ocfs2_xattr_value_root *xv = NULL;
1052         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1053         int ret = 0;
1054
1055         memset(val, 0, size);
1056         memcpy(val, xi->name, name_len);
1057         xv = (struct ocfs2_xattr_value_root *)
1058                 (val + OCFS2_XATTR_SIZE(name_len));
1059         xv->xr_clusters = 0;
1060         xv->xr_last_eb_blk = 0;
1061         xv->xr_list.l_tree_depth = 0;
1062         xv->xr_list.l_count = cpu_to_le16(1);
1063         xv->xr_list.l_next_free_rec = 0;
1064
1065         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1066                                          xi->value_len);
1067         if (ret < 0) {
1068                 mlog_errno(ret);
1069                 return ret;
1070         }
1071         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1072                                               xi->value_len);
1073         if (ret < 0) {
1074                 mlog_errno(ret);
1075                 return ret;
1076         }
1077         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1078         if (ret < 0)
1079                 mlog_errno(ret);
1080
1081         return ret;
1082 }
1083
1084 /*
1085  * ocfs2_xattr_set_entry_local()
1086  *
1087  * Set, replace or remove extended attribute in local.
1088  */
1089 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1090                                         struct ocfs2_xattr_info *xi,
1091                                         struct ocfs2_xattr_search *xs,
1092                                         struct ocfs2_xattr_entry *last,
1093                                         size_t min_offs)
1094 {
1095         size_t name_len = strlen(xi->name);
1096         int i;
1097
1098         if (xi->value && xs->not_found) {
1099                 /* Insert the new xattr entry. */
1100                 le16_add_cpu(&xs->header->xh_count, 1);
1101                 ocfs2_xattr_set_type(last, xi->name_index);
1102                 ocfs2_xattr_set_local(last, 1);
1103                 last->xe_name_len = name_len;
1104         } else {
1105                 void *first_val;
1106                 void *val;
1107                 size_t offs, size;
1108
1109                 first_val = xs->base + min_offs;
1110                 offs = le16_to_cpu(xs->here->xe_name_offset);
1111                 val = xs->base + offs;
1112
1113                 if (le64_to_cpu(xs->here->xe_value_size) >
1114                     OCFS2_XATTR_INLINE_SIZE)
1115                         size = OCFS2_XATTR_SIZE(name_len) +
1116                                 OCFS2_XATTR_ROOT_SIZE;
1117                 else
1118                         size = OCFS2_XATTR_SIZE(name_len) +
1119                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1120
1121                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1122                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1123                         /* The old and the new value have the
1124                            same size. Just replace the value. */
1125                         ocfs2_xattr_set_local(xs->here, 1);
1126                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1127                         /* Clear value bytes. */
1128                         memset(val + OCFS2_XATTR_SIZE(name_len),
1129                                0,
1130                                OCFS2_XATTR_SIZE(xi->value_len));
1131                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1132                                xi->value,
1133                                xi->value_len);
1134                         return;
1135                 }
1136                 /* Remove the old name+value. */
1137                 memmove(first_val + size, first_val, val - first_val);
1138                 memset(first_val, 0, size);
1139                 xs->here->xe_name_hash = 0;
1140                 xs->here->xe_name_offset = 0;
1141                 ocfs2_xattr_set_local(xs->here, 1);
1142                 xs->here->xe_value_size = 0;
1143
1144                 min_offs += size;
1145
1146                 /* Adjust all value offsets. */
1147                 last = xs->header->xh_entries;
1148                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1149                         size_t o = le16_to_cpu(last->xe_name_offset);
1150
1151                         if (o < offs)
1152                                 last->xe_name_offset = cpu_to_le16(o + size);
1153                         last += 1;
1154                 }
1155
1156                 if (!xi->value) {
1157                         /* Remove the old entry. */
1158                         last -= 1;
1159                         memmove(xs->here, xs->here + 1,
1160                                 (void *)last - (void *)xs->here);
1161                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1162                         le16_add_cpu(&xs->header->xh_count, -1);
1163                 }
1164         }
1165         if (xi->value) {
1166                 /* Insert the new name+value. */
1167                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1168                                 OCFS2_XATTR_SIZE(xi->value_len);
1169                 void *val = xs->base + min_offs - size;
1170
1171                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1172                 memset(val, 0, size);
1173                 memcpy(val, xi->name, name_len);
1174                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1175                        xi->value,
1176                        xi->value_len);
1177                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1178                 ocfs2_xattr_set_local(xs->here, 1);
1179                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1180         }
1181
1182         return;
1183 }
1184
1185 /*
1186  * ocfs2_xattr_set_entry()
1187  *
1188  * Set extended attribute entry into inode or block.
1189  *
1190  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1191  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1192  * then set value in B tree with set_value_outside().
1193  */
1194 static int ocfs2_xattr_set_entry(struct inode *inode,
1195                                  struct ocfs2_xattr_info *xi,
1196                                  struct ocfs2_xattr_search *xs,
1197                                  int flag)
1198 {
1199         struct ocfs2_xattr_entry *last;
1200         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1201         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1202         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1203         size_t size_l = 0;
1204         handle_t *handle = NULL;
1205         int free, i, ret;
1206         struct ocfs2_xattr_info xi_l = {
1207                 .name_index = xi->name_index,
1208                 .name = xi->name,
1209                 .value = xi->value,
1210                 .value_len = xi->value_len,
1211         };
1212
1213         /* Compute min_offs, last and free space. */
1214         last = xs->header->xh_entries;
1215
1216         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1217                 size_t offs = le16_to_cpu(last->xe_name_offset);
1218                 if (offs < min_offs)
1219                         min_offs = offs;
1220                 last += 1;
1221         }
1222
1223         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1224         if (free < 0)
1225                 return -EFAULT;
1226
1227         if (!xs->not_found) {
1228                 size_t size = 0;
1229                 if (ocfs2_xattr_is_local(xs->here))
1230                         size = OCFS2_XATTR_SIZE(name_len) +
1231                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1232                 else
1233                         size = OCFS2_XATTR_SIZE(name_len) +
1234                                 OCFS2_XATTR_ROOT_SIZE;
1235                 free += (size + sizeof(struct ocfs2_xattr_entry));
1236         }
1237         /* Check free space in inode or block */
1238         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1239                 if (free < sizeof(struct ocfs2_xattr_entry) +
1240                            OCFS2_XATTR_SIZE(name_len) +
1241                            OCFS2_XATTR_ROOT_SIZE) {
1242                         ret = -ENOSPC;
1243                         goto out;
1244                 }
1245                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1246                 xi_l.value = (void *)&def_xv;
1247                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1248         } else if (xi->value) {
1249                 if (free < sizeof(struct ocfs2_xattr_entry) +
1250                            OCFS2_XATTR_SIZE(name_len) +
1251                            OCFS2_XATTR_SIZE(xi->value_len)) {
1252                         ret = -ENOSPC;
1253                         goto out;
1254                 }
1255         }
1256
1257         if (!xs->not_found) {
1258                 /* For existing extended attribute */
1259                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1260                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1261                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1262                 void *val = xs->base + offs;
1263
1264                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1265                         /* Replace existing local xattr with tree root */
1266                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1267                                                             offs);
1268                         if (ret < 0)
1269                                 mlog_errno(ret);
1270                         goto out;
1271                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1272                         /* For existing xattr which has value outside */
1273                         struct ocfs2_xattr_value_root *xv = NULL;
1274                         xv = (struct ocfs2_xattr_value_root *)(val +
1275                                 OCFS2_XATTR_SIZE(name_len));
1276
1277                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1278                                 /*
1279                                  * If new value need set outside also,
1280                                  * first truncate old value to new value,
1281                                  * then set new value with set_value_outside().
1282                                  */
1283                                 ret = ocfs2_xattr_value_truncate(inode,
1284                                                                  xs->xattr_bh,
1285                                                                  xv,
1286                                                                  xi->value_len);
1287                                 if (ret < 0) {
1288                                         mlog_errno(ret);
1289                                         goto out;
1290                                 }
1291
1292                                 ret = __ocfs2_xattr_set_value_outside(inode,
1293                                                                 xv,
1294                                                                 xi->value,
1295                                                                 xi->value_len);
1296                                 if (ret < 0) {
1297                                         mlog_errno(ret);
1298                                         goto out;
1299                                 }
1300
1301                                 ret = ocfs2_xattr_update_entry(inode,
1302                                                                xi,
1303                                                                xs,
1304                                                                offs);
1305                                 if (ret < 0)
1306                                         mlog_errno(ret);
1307                                 goto out;
1308                         } else {
1309                                 /*
1310                                  * If new value need set in local,
1311                                  * just trucate old value to zero.
1312                                  */
1313                                  ret = ocfs2_xattr_value_truncate(inode,
1314                                                                  xs->xattr_bh,
1315                                                                  xv,
1316                                                                  0);
1317                                 if (ret < 0)
1318                                         mlog_errno(ret);
1319                         }
1320                 }
1321         }
1322
1323         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1324                                    OCFS2_INODE_UPDATE_CREDITS);
1325         if (IS_ERR(handle)) {
1326                 ret = PTR_ERR(handle);
1327                 mlog_errno(ret);
1328                 goto out;
1329         }
1330
1331         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1332                                    OCFS2_JOURNAL_ACCESS_WRITE);
1333         if (ret) {
1334                 mlog_errno(ret);
1335                 goto out_commit;
1336         }
1337
1338         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1339                 /*set extended attribue in external blcok*/
1340                 ret = ocfs2_extend_trans(handle,
1341                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1342                 if (ret) {
1343                         mlog_errno(ret);
1344                         goto out_commit;
1345                 }
1346                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1347                                            OCFS2_JOURNAL_ACCESS_WRITE);
1348                 if (ret) {
1349                         mlog_errno(ret);
1350                         goto out_commit;
1351                 }
1352         }
1353
1354         /*
1355          * Set value in local, include set tree root in local.
1356          * This is the first step for value size >INLINE_SIZE.
1357          */
1358         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1359
1360         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1361                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1362                 if (ret < 0) {
1363                         mlog_errno(ret);
1364                         goto out_commit;
1365                 }
1366         }
1367
1368         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1369             (flag & OCFS2_INLINE_XATTR_FL)) {
1370                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1371                 unsigned int xattrsize = osb->s_xattr_inline_size;
1372
1373                 /*
1374                  * Adjust extent record count or inline data size
1375                  * to reserve space for extended attribute.
1376                  */
1377                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1378                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1379                         le16_add_cpu(&idata->id_count, -xattrsize);
1380                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1381                         struct ocfs2_extent_list *el = &di->id2.i_list;
1382                         le16_add_cpu(&el->l_count, -(xattrsize /
1383                                         sizeof(struct ocfs2_extent_rec)));
1384                 }
1385                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1386         }
1387         /* Update xattr flag */
1388         spin_lock(&oi->ip_lock);
1389         oi->ip_dyn_features |= flag;
1390         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1391         spin_unlock(&oi->ip_lock);
1392         /* Update inode ctime */
1393         inode->i_ctime = CURRENT_TIME;
1394         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1395         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1396
1397         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1398         if (ret < 0)
1399                 mlog_errno(ret);
1400
1401 out_commit:
1402         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1403
1404         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1405                 /*
1406                  * Set value outside in B tree.
1407                  * This is the second step for value size > INLINE_SIZE.
1408                  */
1409                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1410                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1411                 if (ret < 0) {
1412                         int ret2;
1413
1414                         mlog_errno(ret);
1415                         /*
1416                          * If set value outside failed, we have to clean
1417                          * the junk tree root we have already set in local.
1418                          */
1419                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1420                         if (ret2 < 0)
1421                                 mlog_errno(ret2);
1422                 }
1423         }
1424 out:
1425         return ret;
1426
1427 }
1428
1429 static int ocfs2_xattr_free_block(handle_t *handle,
1430                                   struct ocfs2_super *osb,
1431                                   struct ocfs2_xattr_block *xb)
1432 {
1433         struct inode *xb_alloc_inode;
1434         struct buffer_head *xb_alloc_bh = NULL;
1435         u64 blk = le64_to_cpu(xb->xb_blkno);
1436         u16 bit = le16_to_cpu(xb->xb_suballoc_bit);
1437         u64 bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1438         int ret = 0;
1439
1440         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1441                                 EXTENT_ALLOC_SYSTEM_INODE,
1442                                 le16_to_cpu(xb->xb_suballoc_slot));
1443         if (!xb_alloc_inode) {
1444                 ret = -ENOMEM;
1445                 mlog_errno(ret);
1446                 goto out;
1447         }
1448         mutex_lock(&xb_alloc_inode->i_mutex);
1449
1450         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1451         if (ret < 0) {
1452                 mlog_errno(ret);
1453                 goto out_mutex;
1454         }
1455         ret = ocfs2_extend_trans(handle, OCFS2_SUBALLOC_FREE);
1456         if (ret < 0) {
1457                 mlog_errno(ret);
1458                 goto out_unlock;
1459         }
1460         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1461                                        bit, bg_blkno, 1);
1462         if (ret < 0)
1463                 mlog_errno(ret);
1464 out_unlock:
1465         ocfs2_inode_unlock(xb_alloc_inode, 1);
1466         brelse(xb_alloc_bh);
1467 out_mutex:
1468         mutex_unlock(&xb_alloc_inode->i_mutex);
1469         iput(xb_alloc_inode);
1470 out:
1471         return ret;
1472 }
1473
1474 static int ocfs2_remove_value_outside(struct inode*inode,
1475                                       struct buffer_head *bh,
1476                                       struct ocfs2_xattr_header *header)
1477 {
1478         int ret = 0, i;
1479
1480         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1481                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1482
1483                 if (!ocfs2_xattr_is_local(entry)) {
1484                         struct ocfs2_xattr_value_root *xv;
1485                         void *val;
1486
1487                         val = (void *)header +
1488                                 le16_to_cpu(entry->xe_name_offset);
1489                         xv = (struct ocfs2_xattr_value_root *)
1490                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1491                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1492                         if (ret < 0) {
1493                                 mlog_errno(ret);
1494                                 return ret;
1495                         }
1496                 }
1497         }
1498
1499         return ret;
1500 }
1501
1502 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1503                                     struct buffer_head *di_bh)
1504 {
1505
1506         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1507         struct ocfs2_xattr_header *header;
1508         int ret;
1509
1510         header = (struct ocfs2_xattr_header *)
1511                  ((void *)di + inode->i_sb->s_blocksize -
1512                  le16_to_cpu(di->i_xattr_inline_size));
1513
1514         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1515
1516         return ret;
1517 }
1518
1519 static int ocfs2_xattr_block_remove(struct inode *inode,
1520                                     struct buffer_head *blk_bh)
1521 {
1522         struct ocfs2_xattr_block *xb;
1523         int ret = 0;
1524
1525         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1526         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1527                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1528                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1529         } else
1530                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1531
1532         return ret;
1533 }
1534
1535 /*
1536  * ocfs2_xattr_remove()
1537  *
1538  * Free extended attribute resources associated with this inode.
1539  */
1540 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1541 {
1542         struct ocfs2_xattr_block *xb;
1543         struct buffer_head *blk_bh = NULL;
1544         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1545         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1546         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1547         handle_t *handle;
1548         int ret;
1549
1550         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1551                 return 0;
1552
1553         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1554                 return 0;
1555
1556         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1557                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1558                 if (ret < 0) {
1559                         mlog_errno(ret);
1560                         goto out;
1561                 }
1562         }
1563         if (di->i_xattr_loc) {
1564                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1565                                        le64_to_cpu(di->i_xattr_loc),
1566                                        &blk_bh, OCFS2_BH_CACHED, inode);
1567                 if (ret < 0) {
1568                         mlog_errno(ret);
1569                         return ret;
1570                 }
1571                 /*Verify the signature of xattr block*/
1572                 if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1573                            strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1574                         ret = -EFAULT;
1575                         goto out;
1576                 }
1577
1578                 ret = ocfs2_xattr_block_remove(inode, blk_bh);
1579                 if (ret < 0) {
1580                         mlog_errno(ret);
1581                         goto out;
1582                 }
1583         }
1584
1585         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1586                                    OCFS2_INODE_UPDATE_CREDITS);
1587         if (IS_ERR(handle)) {
1588                 ret = PTR_ERR(handle);
1589                 mlog_errno(ret);
1590                 goto out;
1591         }
1592         ret = ocfs2_journal_access(handle, inode, di_bh,
1593                                    OCFS2_JOURNAL_ACCESS_WRITE);
1594         if (ret) {
1595                 mlog_errno(ret);
1596                 goto out_commit;
1597         }
1598
1599         if (di->i_xattr_loc) {
1600                 xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1601                 ocfs2_xattr_free_block(handle, osb, xb);
1602                 di->i_xattr_loc = cpu_to_le64(0);
1603         }
1604
1605         spin_lock(&oi->ip_lock);
1606         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1607         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1608         spin_unlock(&oi->ip_lock);
1609
1610         ret = ocfs2_journal_dirty(handle, di_bh);
1611         if (ret < 0)
1612                 mlog_errno(ret);
1613 out_commit:
1614         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1615 out:
1616         brelse(blk_bh);
1617
1618         return ret;
1619 }
1620
1621 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1622                                         struct ocfs2_dinode *di)
1623 {
1624         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1625         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1626         int free;
1627
1628         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1629                 return 0;
1630
1631         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1632                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1633                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1634         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1635                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1636                         le64_to_cpu(di->i_size);
1637         } else {
1638                 struct ocfs2_extent_list *el = &di->id2.i_list;
1639                 free = (le16_to_cpu(el->l_count) -
1640                         le16_to_cpu(el->l_next_free_rec)) *
1641                         sizeof(struct ocfs2_extent_rec);
1642         }
1643         if (free >= xattrsize)
1644                 return 1;
1645
1646         return 0;
1647 }
1648
1649 /*
1650  * ocfs2_xattr_ibody_find()
1651  *
1652  * Find extended attribute in inode block and
1653  * fill search info into struct ocfs2_xattr_search.
1654  */
1655 static int ocfs2_xattr_ibody_find(struct inode *inode,
1656                                   int name_index,
1657                                   const char *name,
1658                                   struct ocfs2_xattr_search *xs)
1659 {
1660         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1661         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1662         int ret;
1663         int has_space = 0;
1664
1665         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1666                 return 0;
1667
1668         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1669                 down_read(&oi->ip_alloc_sem);
1670                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1671                 up_read(&oi->ip_alloc_sem);
1672                 if (!has_space)
1673                         return 0;
1674         }
1675
1676         xs->xattr_bh = xs->inode_bh;
1677         xs->end = (void *)di + inode->i_sb->s_blocksize;
1678         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1679                 xs->header = (struct ocfs2_xattr_header *)
1680                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1681         else
1682                 xs->header = (struct ocfs2_xattr_header *)
1683                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1684         xs->base = (void *)xs->header;
1685         xs->here = xs->header->xh_entries;
1686
1687         /* Find the named attribute. */
1688         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1689                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1690                 if (ret && ret != -ENODATA)
1691                         return ret;
1692                 xs->not_found = ret;
1693         }
1694
1695         return 0;
1696 }
1697
1698 /*
1699  * ocfs2_xattr_ibody_set()
1700  *
1701  * Set, replace or remove an extended attribute into inode block.
1702  *
1703  */
1704 static int ocfs2_xattr_ibody_set(struct inode *inode,
1705                                  struct ocfs2_xattr_info *xi,
1706                                  struct ocfs2_xattr_search *xs)
1707 {
1708         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1709         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1710         int ret;
1711
1712         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1713                 return -ENOSPC;
1714
1715         down_write(&oi->ip_alloc_sem);
1716         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1717                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1718                         ret = -ENOSPC;
1719                         goto out;
1720                 }
1721         }
1722
1723         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1724                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1725 out:
1726         up_write(&oi->ip_alloc_sem);
1727
1728         return ret;
1729 }
1730
1731 /*
1732  * ocfs2_xattr_block_find()
1733  *
1734  * Find extended attribute in external block and
1735  * fill search info into struct ocfs2_xattr_search.
1736  */
1737 static int ocfs2_xattr_block_find(struct inode *inode,
1738                                   int name_index,
1739                                   const char *name,
1740                                   struct ocfs2_xattr_search *xs)
1741 {
1742         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1743         struct buffer_head *blk_bh = NULL;
1744         struct ocfs2_xattr_block *xb;
1745         int ret = 0;
1746
1747         if (!di->i_xattr_loc)
1748                 return ret;
1749
1750         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1751                                le64_to_cpu(di->i_xattr_loc),
1752                                &blk_bh, OCFS2_BH_CACHED, inode);
1753         if (ret < 0) {
1754                 mlog_errno(ret);
1755                 return ret;
1756         }
1757         /*Verify the signature of xattr block*/
1758         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1759                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1760                         ret = -EFAULT;
1761                         goto cleanup;
1762         }
1763
1764         xs->xattr_bh = blk_bh;
1765         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1766
1767         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1768                 xs->header = &xb->xb_attrs.xb_header;
1769                 xs->base = (void *)xs->header;
1770                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1771                 xs->here = xs->header->xh_entries;
1772
1773                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1774         } else
1775                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1776                                                    name_index,
1777                                                    name, xs);
1778
1779         if (ret && ret != -ENODATA) {
1780                 xs->xattr_bh = NULL;
1781                 goto cleanup;
1782         }
1783         xs->not_found = ret;
1784         return 0;
1785 cleanup:
1786         brelse(blk_bh);
1787
1788         return ret;
1789 }
1790
1791 /*
1792  * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1793  * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1794  * re-initialized.
1795  */
1796 static int ocfs2_restore_xattr_block(struct inode *inode,
1797                                      struct ocfs2_xattr_search *xs)
1798 {
1799         int ret;
1800         handle_t *handle;
1801         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1802         struct ocfs2_xattr_block *xb =
1803                 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1804         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1805         u16 xb_flags = le16_to_cpu(xb->xb_flags);
1806
1807         BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1808                 le16_to_cpu(el->l_next_free_rec) != 0);
1809
1810         handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1811         if (IS_ERR(handle)) {
1812                 ret = PTR_ERR(handle);
1813                 handle = NULL;
1814                 goto out;
1815         }
1816
1817         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1818                                    OCFS2_JOURNAL_ACCESS_WRITE);
1819         if (ret < 0) {
1820                 mlog_errno(ret);
1821                 goto out_commit;
1822         }
1823
1824         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1825                offsetof(struct ocfs2_xattr_block, xb_attrs));
1826
1827         xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1828
1829         ocfs2_journal_dirty(handle, xs->xattr_bh);
1830
1831 out_commit:
1832         ocfs2_commit_trans(osb, handle);
1833 out:
1834         return ret;
1835 }
1836
1837 /*
1838  * ocfs2_xattr_block_set()
1839  *
1840  * Set, replace or remove an extended attribute into external block.
1841  *
1842  */
1843 static int ocfs2_xattr_block_set(struct inode *inode,
1844                                  struct ocfs2_xattr_info *xi,
1845                                  struct ocfs2_xattr_search *xs)
1846 {
1847         struct buffer_head *new_bh = NULL;
1848         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1849         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1850         struct ocfs2_alloc_context *meta_ac = NULL;
1851         handle_t *handle = NULL;
1852         struct ocfs2_xattr_block *xblk = NULL;
1853         u16 suballoc_bit_start;
1854         u32 num_got;
1855         u64 first_blkno;
1856         int ret;
1857
1858         if (!xs->xattr_bh) {
1859                 /*
1860                  * Alloc one external block for extended attribute
1861                  * outside of inode.
1862                  */
1863                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1864                 if (ret < 0) {
1865                         mlog_errno(ret);
1866                         goto out;
1867                 }
1868                 handle = ocfs2_start_trans(osb,
1869                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1870                 if (IS_ERR(handle)) {
1871                         ret = PTR_ERR(handle);
1872                         mlog_errno(ret);
1873                         goto out;
1874                 }
1875                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1876                                            OCFS2_JOURNAL_ACCESS_CREATE);
1877                 if (ret < 0) {
1878                         mlog_errno(ret);
1879                         goto out_commit;
1880                 }
1881
1882                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1883                                            &suballoc_bit_start, &num_got,
1884                                            &first_blkno);
1885                 if (ret < 0) {
1886                         mlog_errno(ret);
1887                         goto out_commit;
1888                 }
1889
1890                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1891                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1892
1893                 ret = ocfs2_journal_access(handle, inode, new_bh,
1894                                            OCFS2_JOURNAL_ACCESS_CREATE);
1895                 if (ret < 0) {
1896                         mlog_errno(ret);
1897                         goto out_commit;
1898                 }
1899
1900                 /* Initialize ocfs2_xattr_block */
1901                 xs->xattr_bh = new_bh;
1902                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1903                 memset(xblk, 0, inode->i_sb->s_blocksize);
1904                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1905                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1906                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1907                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1908                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1909
1910                 xs->header = &xblk->xb_attrs.xb_header;
1911                 xs->base = (void *)xs->header;
1912                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1913                 xs->here = xs->header->xh_entries;
1914
1915
1916                 ret = ocfs2_journal_dirty(handle, new_bh);
1917                 if (ret < 0) {
1918                         mlog_errno(ret);
1919                         goto out_commit;
1920                 }
1921                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1922                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1923                 if (ret < 0)
1924                         mlog_errno(ret);
1925 out_commit:
1926                 ocfs2_commit_trans(osb, handle);
1927 out:
1928                 if (meta_ac)
1929                         ocfs2_free_alloc_context(meta_ac);
1930                 if (ret < 0)
1931                         return ret;
1932         } else
1933                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1934
1935         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1936                 /* Set extended attribute into external block */
1937                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1938                 if (!ret || ret != -ENOSPC)
1939                         goto end;
1940
1941                 ret = ocfs2_xattr_create_index_block(inode, xs);
1942                 if (ret)
1943                         goto end;
1944         }
1945
1946         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1947         if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1948                 ret = ocfs2_restore_xattr_block(inode, xs);
1949
1950 end:
1951
1952         return ret;
1953 }
1954
1955 /*
1956  * ocfs2_xattr_set()
1957  *
1958  * Set, replace or remove an extended attribute for this inode.
1959  * value is NULL to remove an existing extended attribute, else either
1960  * create or replace an extended attribute.
1961  */
1962 int ocfs2_xattr_set(struct inode *inode,
1963                     int name_index,
1964                     const char *name,
1965                     const void *value,
1966                     size_t value_len,
1967                     int flags)
1968 {
1969         struct buffer_head *di_bh = NULL;
1970         struct ocfs2_dinode *di;
1971         int ret;
1972         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1973
1974         struct ocfs2_xattr_info xi = {
1975                 .name_index = name_index,
1976                 .name = name,
1977                 .value = value,
1978                 .value_len = value_len,
1979         };
1980
1981         struct ocfs2_xattr_search xis = {
1982                 .not_found = -ENODATA,
1983         };
1984
1985         struct ocfs2_xattr_search xbs = {
1986                 .not_found = -ENODATA,
1987         };
1988
1989         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1990                 return -EOPNOTSUPP;
1991
1992         ret = ocfs2_inode_lock(inode, &di_bh, 1);
1993         if (ret < 0) {
1994                 mlog_errno(ret);
1995                 return ret;
1996         }
1997         xis.inode_bh = xbs.inode_bh = di_bh;
1998         di = (struct ocfs2_dinode *)di_bh->b_data;
1999
2000         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2001         /*
2002          * Scan inode and external block to find the same name
2003          * extended attribute and collect search infomation.
2004          */
2005         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2006         if (ret)
2007                 goto cleanup;
2008         if (xis.not_found) {
2009                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2010                 if (ret)
2011                         goto cleanup;
2012         }
2013
2014         if (xis.not_found && xbs.not_found) {
2015                 ret = -ENODATA;
2016                 if (flags & XATTR_REPLACE)
2017                         goto cleanup;
2018                 ret = 0;
2019                 if (!value)
2020                         goto cleanup;
2021         } else {
2022                 ret = -EEXIST;
2023                 if (flags & XATTR_CREATE)
2024                         goto cleanup;
2025         }
2026
2027         if (!value) {
2028                 /* Remove existing extended attribute */
2029                 if (!xis.not_found)
2030                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2031                 else if (!xbs.not_found)
2032                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2033         } else {
2034                 /* We always try to set extended attribute into inode first*/
2035                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2036                 if (!ret && !xbs.not_found) {
2037                         /*
2038                          * If succeed and that extended attribute existing in
2039                          * external block, then we will remove it.
2040                          */
2041                         xi.value = NULL;
2042                         xi.value_len = 0;
2043                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2044                 } else if (ret == -ENOSPC) {
2045                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2046                                 ret = ocfs2_xattr_block_find(inode, name_index,
2047                                                              name, &xbs);
2048                                 if (ret)
2049                                         goto cleanup;
2050                         }
2051                         /*
2052                          * If no space in inode, we will set extended attribute
2053                          * into external block.
2054                          */
2055                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2056                         if (ret)
2057                                 goto cleanup;
2058                         if (!xis.not_found) {
2059                                 /*
2060                                  * If succeed and that extended attribute
2061                                  * existing in inode, we will remove it.
2062                                  */
2063                                 xi.value = NULL;
2064                                 xi.value_len = 0;
2065                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2066                         }
2067                 }
2068         }
2069 cleanup:
2070         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2071         ocfs2_inode_unlock(inode, 1);
2072         brelse(di_bh);
2073         brelse(xbs.xattr_bh);
2074         for (i = 0; i < blk_per_bucket; i++)
2075                 brelse(xbs.bucket.bhs[i]);
2076
2077         return ret;
2078 }
2079
2080 static inline u32 ocfs2_xattr_hash_by_name(struct inode *inode,
2081                                            int name_index,
2082                                            const char *suffix_name)
2083 {
2084         struct xattr_handler *handler = ocfs2_xattr_handler(name_index);
2085         char *prefix = handler->prefix;
2086         int prefix_len = strlen(handler->prefix);
2087
2088         return ocfs2_xattr_name_hash(inode, prefix, prefix_len,
2089                                      (char *)suffix_name, strlen(suffix_name));
2090 }
2091
2092 /*
2093  * Find the xattr extent rec which may contains name_hash.
2094  * e_cpos will be the first name hash of the xattr rec.
2095  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2096  */
2097 static int ocfs2_xattr_get_rec(struct inode *inode,
2098                                u32 name_hash,
2099                                u64 *p_blkno,
2100                                u32 *e_cpos,
2101                                u32 *num_clusters,
2102                                struct ocfs2_extent_list *el)
2103 {
2104         int ret = 0, i;
2105         struct buffer_head *eb_bh = NULL;
2106         struct ocfs2_extent_block *eb;
2107         struct ocfs2_extent_rec *rec = NULL;
2108         u64 e_blkno = 0;
2109
2110         if (el->l_tree_depth) {
2111                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2112                 if (ret) {
2113                         mlog_errno(ret);
2114                         goto out;
2115                 }
2116
2117                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2118                 el = &eb->h_list;
2119
2120                 if (el->l_tree_depth) {
2121                         ocfs2_error(inode->i_sb,
2122                                     "Inode %lu has non zero tree depth in "
2123                                     "xattr tree block %llu\n", inode->i_ino,
2124                                     (unsigned long long)eb_bh->b_blocknr);
2125                         ret = -EROFS;
2126                         goto out;
2127                 }
2128         }
2129
2130         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2131                 rec = &el->l_recs[i];
2132
2133                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2134                         e_blkno = le64_to_cpu(rec->e_blkno);
2135                         break;
2136                 }
2137         }
2138
2139         if (!e_blkno) {
2140                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2141                             "record (%u, %u, 0) in xattr", inode->i_ino,
2142                             le32_to_cpu(rec->e_cpos),
2143                             ocfs2_rec_clusters(el, rec));
2144                 ret = -EROFS;
2145                 goto out;
2146         }
2147
2148         *p_blkno = le64_to_cpu(rec->e_blkno);
2149         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2150         if (e_cpos)
2151                 *e_cpos = le32_to_cpu(rec->e_cpos);
2152 out:
2153         brelse(eb_bh);
2154         return ret;
2155 }
2156
2157 typedef int (xattr_bucket_func)(struct inode *inode,
2158                                 struct ocfs2_xattr_bucket *bucket,
2159                                 void *para);
2160
2161 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2162                                    struct buffer_head *header_bh,
2163                                    int name_index,
2164                                    const char *name,
2165                                    u32 name_hash,
2166                                    u16 *xe_index,
2167                                    int *found)
2168 {
2169         int i, ret = 0, cmp = 1, block_off, new_offset;
2170         struct ocfs2_xattr_header *xh =
2171                         (struct ocfs2_xattr_header *)header_bh->b_data;
2172         size_t name_len = strlen(name);
2173         struct ocfs2_xattr_entry *xe = NULL;
2174         struct buffer_head *name_bh = NULL;
2175         char *xe_name;
2176
2177         /*
2178          * We don't use binary search in the bucket because there
2179          * may be multiple entries with the same name hash.
2180          */
2181         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2182                 xe = &xh->xh_entries[i];
2183
2184                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2185                         continue;
2186                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2187                         break;
2188
2189                 cmp = name_index - ocfs2_xattr_get_type(xe);
2190                 if (!cmp)
2191                         cmp = name_len - xe->xe_name_len;
2192                 if (cmp)
2193                         continue;
2194
2195                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2196                                                         xh,
2197                                                         i,
2198                                                         &block_off,
2199                                                         &new_offset);
2200                 if (ret) {
2201                         mlog_errno(ret);
2202                         break;
2203                 }
2204
2205                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2206                                        header_bh->b_blocknr + block_off,
2207                                        &name_bh, OCFS2_BH_CACHED, inode);
2208                 if (ret) {
2209                         mlog_errno(ret);
2210                         break;
2211                 }
2212                 xe_name = name_bh->b_data + new_offset;
2213
2214                 cmp = memcmp(name, xe_name, name_len);
2215                 brelse(name_bh);
2216                 name_bh = NULL;
2217
2218                 if (cmp == 0) {
2219                         *xe_index = i;
2220                         *found = 1;
2221                         ret = 0;
2222                         break;
2223                 }
2224         }
2225
2226         return ret;
2227 }
2228
2229 /*
2230  * Find the specified xattr entry in a series of buckets.
2231  * This series start from p_blkno and last for num_clusters.
2232  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2233  * the num of the valid buckets.
2234  *
2235  * Return the buffer_head this xattr should reside in. And if the xattr's
2236  * hash is in the gap of 2 buckets, return the lower bucket.
2237  */
2238 static int ocfs2_xattr_bucket_find(struct inode *inode,
2239                                    int name_index,
2240                                    const char *name,
2241                                    u32 name_hash,
2242                                    u64 p_blkno,
2243                                    u32 first_hash,
2244                                    u32 num_clusters,
2245                                    struct ocfs2_xattr_search *xs)
2246 {
2247         int ret, found = 0;
2248         struct buffer_head *bh = NULL;
2249         struct buffer_head *lower_bh = NULL;
2250         struct ocfs2_xattr_header *xh = NULL;
2251         struct ocfs2_xattr_entry *xe = NULL;
2252         u16 index = 0;
2253         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2254         int low_bucket = 0, bucket, high_bucket;
2255         u32 last_hash;
2256         u64 blkno;
2257
2258         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
2259                                &bh, OCFS2_BH_CACHED, inode);
2260         if (ret) {
2261                 mlog_errno(ret);
2262                 goto out;
2263         }
2264
2265         xh = (struct ocfs2_xattr_header *)bh->b_data;
2266         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2267
2268         while (low_bucket <= high_bucket) {
2269                 brelse(bh);
2270                 bh = NULL;
2271                 bucket = (low_bucket + high_bucket) / 2;
2272
2273                 blkno = p_blkno + bucket * blk_per_bucket;
2274
2275                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
2276                                        &bh, OCFS2_BH_CACHED, inode);
2277                 if (ret) {
2278                         mlog_errno(ret);
2279                         goto out;
2280                 }
2281
2282                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2283                 xe = &xh->xh_entries[0];
2284                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2285                         high_bucket = bucket - 1;
2286                         continue;
2287                 }
2288
2289                 /*
2290                  * Check whether the hash of the last entry in our
2291                  * bucket is larger than the search one.
2292                  */
2293                 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2294                 last_hash = le32_to_cpu(xe->xe_name_hash);
2295
2296                 /* record lower_bh which may be the insert place. */
2297                 brelse(lower_bh);
2298                 lower_bh = bh;
2299                 bh = NULL;
2300
2301                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2302                         low_bucket = bucket + 1;
2303                         continue;
2304                 }
2305
2306                 /* the searched xattr should reside in this bucket if exists. */
2307                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2308                                               name_index, name, name_hash,
2309                                               &index, &found);
2310                 if (ret) {
2311                         mlog_errno(ret);
2312                         goto out;
2313                 }
2314                 break;
2315         }
2316
2317         /*
2318          * Record the bucket we have found.
2319          * When the xattr's hash value is in the gap of 2 buckets, we will
2320          * always set it to the previous bucket.
2321          */
2322         if (!lower_bh) {
2323                 /*
2324                  * We can't find any bucket whose first name_hash is less
2325                  * than the find name_hash.
2326                  */
2327                 BUG_ON(bh->b_blocknr != p_blkno);
2328                 lower_bh = bh;
2329                 bh = NULL;
2330         }
2331         xs->bucket.bhs[0] = lower_bh;
2332         xs->bucket.xh = (struct ocfs2_xattr_header *)
2333                                         xs->bucket.bhs[0]->b_data;
2334         lower_bh = NULL;
2335
2336         xs->header = xs->bucket.xh;
2337         xs->base = xs->bucket.bhs[0]->b_data;
2338         xs->end = xs->base + inode->i_sb->s_blocksize;
2339
2340         if (found) {
2341                 /*
2342                  * If we have found the xattr enty, read all the blocks in
2343                  * this bucket.
2344                  */
2345                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2346                                         xs->bucket.bhs[0]->b_blocknr + 1,
2347                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2348                                         OCFS2_BH_CACHED, inode);
2349                 if (ret) {
2350                         mlog_errno(ret);
2351                         goto out;
2352                 }
2353
2354                 xs->here = &xs->header->xh_entries[index];
2355                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2356                      (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2357         } else
2358                 ret = -ENODATA;
2359
2360 out:
2361         brelse(bh);
2362         brelse(lower_bh);
2363         return ret;
2364 }
2365
2366 static int ocfs2_xattr_index_block_find(struct inode *inode,
2367                                         struct buffer_head *root_bh,
2368                                         int name_index,
2369                                         const char *name,
2370                                         struct ocfs2_xattr_search *xs)
2371 {
2372         int ret;
2373         struct ocfs2_xattr_block *xb =
2374                         (struct ocfs2_xattr_block *)root_bh->b_data;
2375         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2376         struct ocfs2_extent_list *el = &xb_root->xt_list;
2377         u64 p_blkno = 0;
2378         u32 first_hash, num_clusters = 0;
2379         u32 name_hash = ocfs2_xattr_hash_by_name(inode, name_index, name);
2380
2381         if (le16_to_cpu(el->l_next_free_rec) == 0)
2382                 return -ENODATA;
2383
2384         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2385              name, name_hash, name_index);
2386
2387         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2388                                   &num_clusters, el);
2389         if (ret) {
2390                 mlog_errno(ret);
2391                 goto out;
2392         }
2393
2394         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2395
2396         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2397              "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2398
2399         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2400                                       p_blkno, first_hash, num_clusters, xs);
2401
2402 out:
2403         return ret;
2404 }
2405
2406 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2407                                        u64 blkno,
2408                                        u32 clusters,
2409                                        xattr_bucket_func *func,
2410                                        void *para)
2411 {
2412         int i, j, ret = 0;
2413         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2414         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2415         u32 num_buckets = clusters * bpc;
2416         struct ocfs2_xattr_bucket bucket;
2417
2418         memset(&bucket, 0, sizeof(bucket));
2419
2420         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2421              clusters, blkno);
2422
2423         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2424                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2425                                         blkno, blk_per_bucket,
2426                                         bucket.bhs, OCFS2_BH_CACHED, inode);
2427                 if (ret) {
2428                         mlog_errno(ret);
2429                         goto out;
2430                 }
2431
2432                 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2433                 /*
2434                  * The real bucket num in this series of blocks is stored
2435                  * in the 1st bucket.
2436                  */
2437                 if (i == 0)
2438                         num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2439
2440                 mlog(0, "iterating xattr bucket %llu\n", blkno);
2441                 if (func) {
2442                         ret = func(inode, &bucket, para);
2443                         if (ret) {
2444                                 mlog_errno(ret);
2445                                 break;
2446                         }
2447                 }
2448
2449                 for (j = 0; j < blk_per_bucket; j++)
2450                         brelse(bucket.bhs[j]);
2451                 memset(&bucket, 0, sizeof(bucket));
2452         }
2453
2454 out:
2455         for (j = 0; j < blk_per_bucket; j++)
2456                 brelse(bucket.bhs[j]);
2457
2458         return ret;
2459 }
2460
2461 struct ocfs2_xattr_tree_list {
2462         char *buffer;
2463         size_t buffer_size;
2464 };
2465
2466 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2467                                              struct ocfs2_xattr_header *xh,
2468                                              int index,
2469                                              int *block_off,
2470                                              int *new_offset)
2471 {
2472         u16 name_offset;
2473
2474         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2475                 return -EINVAL;
2476
2477         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2478
2479         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2480         *new_offset = name_offset % inode->i_sb->s_blocksize;
2481
2482         return 0;
2483 }
2484
2485 static int ocfs2_list_xattr_bucket(struct inode *inode,
2486                                    struct ocfs2_xattr_bucket *bucket,
2487                                    void *para)
2488 {
2489         int ret = 0;
2490         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2491         size_t size;
2492         int i, block_off, new_offset;
2493
2494         for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2495                 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2496                 struct xattr_handler *handler =
2497                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
2498
2499                 if (handler) {
2500                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2501                                                                 bucket->xh,
2502                                                                 i,
2503                                                                 &block_off,
2504                                                                 &new_offset);
2505                         if (ret)
2506                                 break;
2507                         size = handler->list(inode, xl->buffer, xl->buffer_size,
2508                                              bucket->bhs[block_off]->b_data +
2509                                              new_offset,
2510                                              entry->xe_name_len);
2511                         if (xl->buffer) {
2512                                 if (size > xl->buffer_size)
2513                                         return -ERANGE;
2514                                 xl->buffer += size;
2515                         }
2516                         xl->buffer_size -= size;
2517                 }
2518         }
2519
2520         return ret;
2521 }
2522
2523 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2524                                              struct ocfs2_xattr_tree_root *xt,
2525                                              char *buffer,
2526                                              size_t buffer_size)
2527 {
2528         struct ocfs2_extent_list *el = &xt->xt_list;
2529         int ret = 0;
2530         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2531         u64 p_blkno = 0;
2532         struct ocfs2_xattr_tree_list xl = {
2533                 .buffer = buffer,
2534                 .buffer_size = buffer_size,
2535         };
2536
2537         if (le16_to_cpu(el->l_next_free_rec) == 0)
2538                 return 0;
2539
2540         while (name_hash > 0) {
2541                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2542                                           &e_cpos, &num_clusters, el);
2543                 if (ret) {
2544                         mlog_errno(ret);
2545                         goto out;
2546                 }
2547
2548                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2549                                                   ocfs2_list_xattr_bucket,
2550                                                   &xl);
2551                 if (ret) {
2552                         mlog_errno(ret);
2553                         goto out;
2554                 }
2555
2556                 if (e_cpos == 0)
2557                         break;
2558
2559                 name_hash = e_cpos - 1;
2560         }
2561
2562         ret = buffer_size - xl.buffer_size;
2563 out:
2564         return ret;
2565 }
2566
2567 static int cmp_xe(const void *a, const void *b)
2568 {
2569         const struct ocfs2_xattr_entry *l = a, *r = b;
2570         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2571         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2572
2573         if (l_hash > r_hash)
2574                 return 1;
2575         if (l_hash < r_hash)
2576                 return -1;
2577         return 0;
2578 }
2579
2580 static void swap_xe(void *a, void *b, int size)
2581 {
2582         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2583
2584         tmp = *l;
2585         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2586         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2587 }
2588
2589 /*
2590  * When the ocfs2_xattr_block is filled up, new bucket will be created
2591  * and all the xattr entries will be moved to the new bucket.
2592  * Note: we need to sort the entries since they are not saved in order
2593  * in the ocfs2_xattr_block.
2594  */
2595 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2596                                            struct buffer_head *xb_bh,
2597                                            struct buffer_head *xh_bh,
2598                                            struct buffer_head *data_bh)
2599 {
2600         int i, blocksize = inode->i_sb->s_blocksize;
2601         u16 offset, size, off_change;
2602         struct ocfs2_xattr_entry *xe;
2603         struct ocfs2_xattr_block *xb =
2604                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2605         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2606         struct ocfs2_xattr_header *xh =
2607                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2608         u16 count = le16_to_cpu(xb_xh->xh_count);
2609         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2610
2611         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2612              (unsigned long long)xb_bh->b_blocknr,
2613              (unsigned long long)xh_bh->b_blocknr);
2614
2615         memset(xh_bh->b_data, 0, blocksize);
2616         if (data_bh)
2617                 memset(data_bh->b_data, 0, blocksize);
2618         /*
2619          * Since the xe_name_offset is based on ocfs2_xattr_header,
2620          * there is a offset change corresponding to the change of
2621          * ocfs2_xattr_header's position.
2622          */
2623         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2624         xe = &xb_xh->xh_entries[count - 1];
2625         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2626         size = blocksize - offset;
2627
2628         /* copy all the names and values. */
2629         if (data_bh)
2630                 target = data_bh->b_data;
2631         memcpy(target + offset, src + offset, size);
2632
2633         /* Init new header now. */
2634         xh->xh_count = xb_xh->xh_count;
2635         xh->xh_num_buckets = cpu_to_le16(1);
2636         xh->xh_name_value_len = cpu_to_le16(size);
2637         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2638
2639         /* copy all the entries. */
2640         target = xh_bh->b_data;
2641         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2642         size = count * sizeof(struct ocfs2_xattr_entry);
2643         memcpy(target + offset, (char *)xb_xh + offset, size);
2644
2645         /* Change the xe offset for all the xe because of the move. */
2646         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2647                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2648         for (i = 0; i < count; i++)
2649                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2650
2651         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2652              offset, size, off_change);
2653
2654         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2655              cmp_xe, swap_xe);
2656 }
2657
2658 /*
2659  * After we move xattr from block to index btree, we have to
2660  * update ocfs2_xattr_search to the new xe and base.
2661  *
2662  * When the entry is in xattr block, xattr_bh indicates the storage place.
2663  * While if the entry is in index b-tree, "bucket" indicates the
2664  * real place of the xattr.
2665  */
2666 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2667                                            struct ocfs2_xattr_search *xs,
2668                                            struct buffer_head *old_bh,
2669                                            struct buffer_head *new_bh)
2670 {
2671         int ret = 0;
2672         char *buf = old_bh->b_data;
2673         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2674         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2675         int i, blocksize = inode->i_sb->s_blocksize;
2676         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2677
2678         xs->bucket.bhs[0] = new_bh;
2679         get_bh(new_bh);
2680         xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2681         xs->header = xs->bucket.xh;
2682
2683         xs->base = new_bh->b_data;
2684         xs->end = xs->base + inode->i_sb->s_blocksize;
2685
2686         if (!xs->not_found) {
2687                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2688                         ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2689                                         xs->bucket.bhs[0]->b_blocknr + 1,
2690                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2691                                         OCFS2_BH_CACHED, inode);
2692                         if (ret) {
2693                                 mlog_errno(ret);
2694                                 return ret;
2695                         }
2696
2697                         i = xs->here - old_xh->xh_entries;
2698                         xs->here = &xs->header->xh_entries[i];
2699                 }
2700         }
2701
2702         return ret;
2703 }
2704
2705 static int ocfs2_xattr_create_index_block(struct inode *inode,
2706                                           struct ocfs2_xattr_search *xs)
2707 {
2708         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2709         u32 bit_off, len;
2710         u64 blkno;
2711         handle_t *handle;
2712         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2713         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2714         struct ocfs2_alloc_context *data_ac;
2715         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2716         struct buffer_head *xb_bh = xs->xattr_bh;
2717         struct ocfs2_xattr_block *xb =
2718                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2719         struct ocfs2_xattr_tree_root *xr;
2720         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2721         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2722
2723         mlog(0, "create xattr index block for %llu\n",
2724              (unsigned long long)xb_bh->b_blocknr);
2725
2726         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2727
2728         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2729         if (ret) {
2730                 mlog_errno(ret);
2731                 goto out;
2732         }
2733
2734         /*
2735          * XXX:
2736          * We can use this lock for now, and maybe move to a dedicated mutex
2737          * if performance becomes a problem later.
2738          */
2739         down_write(&oi->ip_alloc_sem);
2740
2741         /*
2742          * 3 more credits, one for xattr block update, one for the 1st block
2743          * of the new xattr bucket and one for the value/data.
2744          */
2745         credits += 3;
2746         handle = ocfs2_start_trans(osb, credits);
2747         if (IS_ERR(handle)) {
2748                 ret = PTR_ERR(handle);
2749                 mlog_errno(ret);
2750                 goto out_sem;
2751         }
2752
2753         ret = ocfs2_journal_access(handle, inode, xb_bh,
2754                                    OCFS2_JOURNAL_ACCESS_WRITE);
2755         if (ret) {
2756                 mlog_errno(ret);
2757                 goto out_commit;
2758         }
2759
2760         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2761         if (ret) {
2762                 mlog_errno(ret);
2763                 goto out_commit;
2764         }
2765
2766         /*
2767          * The bucket may spread in many blocks, and
2768          * we will only touch the 1st block and the last block
2769          * in the whole bucket(one for entry and one for data).
2770          */
2771         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2772
2773         mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2774
2775         xh_bh = sb_getblk(inode->i_sb, blkno);
2776         if (!xh_bh) {
2777                 ret = -EIO;
2778                 mlog_errno(ret);
2779                 goto out_commit;
2780         }
2781
2782         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2783
2784         ret = ocfs2_journal_access(handle, inode, xh_bh,
2785                                    OCFS2_JOURNAL_ACCESS_CREATE);
2786         if (ret) {
2787                 mlog_errno(ret);
2788                 goto out_commit;
2789         }
2790
2791         if (bpb > 1) {
2792                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2793                 if (!data_bh) {
2794                         ret = -EIO;
2795                         mlog_errno(ret);
2796                         goto out_commit;
2797                 }
2798
2799                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2800
2801                 ret = ocfs2_journal_access(handle, inode, data_bh,
2802                                            OCFS2_JOURNAL_ACCESS_CREATE);
2803                 if (ret) {
2804                         mlog_errno(ret);
2805                         goto out_commit;
2806                 }
2807         }
2808
2809         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2810
2811         ocfs2_journal_dirty(handle, xh_bh);
2812         if (data_bh)
2813                 ocfs2_journal_dirty(handle, data_bh);
2814
2815         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2816
2817         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2818         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2819                offsetof(struct ocfs2_xattr_block, xb_attrs));
2820
2821         xr = &xb->xb_attrs.xb_root;
2822         xr->xt_clusters = cpu_to_le32(1);
2823         xr->xt_last_eb_blk = 0;
2824         xr->xt_list.l_tree_depth = 0;
2825         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2826         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2827
2828         xr->xt_list.l_recs[0].e_cpos = 0;
2829         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2830         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2831
2832         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2833
2834         ret = ocfs2_journal_dirty(handle, xb_bh);
2835         if (ret) {
2836                 mlog_errno(ret);
2837                 goto out_commit;
2838         }
2839
2840 out_commit:
2841         ocfs2_commit_trans(osb, handle);
2842
2843 out_sem:
2844         up_write(&oi->ip_alloc_sem);
2845
2846 out:
2847         if (data_ac)
2848                 ocfs2_free_alloc_context(data_ac);
2849
2850         brelse(xh_bh);
2851         brelse(data_bh);
2852
2853         return ret;
2854 }
2855
2856 static int cmp_xe_offset(const void *a, const void *b)
2857 {
2858         const struct ocfs2_xattr_entry *l = a, *r = b;
2859         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2860         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2861
2862         if (l_name_offset < r_name_offset)
2863                 return 1;
2864         if (l_name_offset > r_name_offset)
2865                 return -1;
2866         return 0;
2867 }
2868
2869 /*
2870  * defrag a xattr bucket if we find that the bucket has some
2871  * holes beteen name/value pairs.
2872  * We will move all the name/value pairs to the end of the bucket
2873  * so that we can spare some space for insertion.
2874  */
2875 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2876                                      struct ocfs2_xattr_bucket *bucket)
2877 {
2878         int ret, i;
2879         size_t end, offset, len, value_len;
2880         struct ocfs2_xattr_header *xh;
2881         char *entries, *buf, *bucket_buf = NULL;
2882         u64 blkno = bucket->bhs[0]->b_blocknr;
2883         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2884         u16 xh_free_start;
2885         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2886         size_t blocksize = inode->i_sb->s_blocksize;
2887         handle_t *handle;
2888         struct buffer_head **bhs;
2889         struct ocfs2_xattr_entry *xe;
2890
2891         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2892                         GFP_NOFS);
2893         if (!bhs)
2894                 return -ENOMEM;
2895
2896         ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs,
2897                                 OCFS2_BH_CACHED, inode);
2898         if (ret)
2899                 goto out;
2900
2901         /*
2902          * In order to make the operation more efficient and generic,
2903          * we copy all the blocks into a contiguous memory and do the
2904          * defragment there, so if anything is error, we will not touch
2905          * the real block.
2906          */
2907         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2908         if (!bucket_buf) {
2909                 ret = -EIO;
2910                 goto out;
2911         }
2912
2913         buf = bucket_buf;
2914         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2915                 memcpy(buf, bhs[i]->b_data, blocksize);
2916
2917         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2918         if (IS_ERR(handle)) {
2919                 ret = PTR_ERR(handle);
2920                 handle = NULL;
2921                 mlog_errno(ret);
2922                 goto out;
2923         }
2924
2925         for (i = 0; i < blk_per_bucket; i++) {
2926                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2927                                            OCFS2_JOURNAL_ACCESS_WRITE);
2928                 if (ret < 0) {
2929                         mlog_errno(ret);
2930                         goto commit;
2931                 }
2932         }
2933
2934         xh = (struct ocfs2_xattr_header *)bucket_buf;
2935         entries = (char *)xh->xh_entries;
2936         xh_free_start = le16_to_cpu(xh->xh_free_start);
2937
2938         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2939              "xh_free_start = %u, xh_name_value_len = %u.\n",
2940              blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2941              le16_to_cpu(xh->xh_name_value_len));
2942
2943         /*
2944          * sort all the entries by their offset.
2945          * the largest will be the first, so that we can
2946          * move them to the end one by one.
2947          */
2948         sort(entries, le16_to_cpu(xh->xh_count),
2949              sizeof(struct ocfs2_xattr_entry),
2950              cmp_xe_offset, swap_xe);
2951
2952         /* Move all name/values to the end of the bucket. */
2953         xe = xh->xh_entries;
2954         end = OCFS2_XATTR_BUCKET_SIZE;
2955         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2956                 offset = le16_to_cpu(xe->xe_name_offset);
2957                 if (ocfs2_xattr_is_local(xe))
2958                         value_len = OCFS2_XATTR_SIZE(
2959                                         le64_to_cpu(xe->xe_value_size));
2960                 else
2961                         value_len = OCFS2_XATTR_ROOT_SIZE;
2962                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2963
2964                 /*
2965                  * We must make sure that the name/value pair
2966                  * exist in the same block. So adjust end to
2967                  * the previous block end if needed.
2968                  */
2969                 if (((end - len) / blocksize !=
2970                         (end - 1) / blocksize))
2971                         end = end - end % blocksize;
2972
2973                 if (end > offset + len) {
2974                         memmove(bucket_buf + end - len,
2975                                 bucket_buf + offset, len);
2976                         xe->xe_name_offset = cpu_to_le16(end - len);
2977                 }
2978
2979                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2980                                 "bucket %llu\n", (unsigned long long)blkno);
2981
2982                 end -= len;
2983         }
2984
2985         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2986                         "bucket %llu\n", (unsigned long long)blkno);
2987
2988         if (xh_free_start == end)
2989                 goto commit;
2990
2991         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
2992         xh->xh_free_start = cpu_to_le16(end);
2993
2994         /* sort the entries by their name_hash. */
2995         sort(entries, le16_to_cpu(xh->xh_count),
2996              sizeof(struct ocfs2_xattr_entry),
2997              cmp_xe, swap_xe);
2998
2999         buf = bucket_buf;
3000         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3001                 memcpy(bhs[i]->b_data, buf, blocksize);
3002                 ocfs2_journal_dirty(handle, bhs[i]);
3003         }
3004
3005 commit:
3006         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3007 out:
3008
3009         if (bhs) {
3010                 for (i = 0; i < blk_per_bucket; i++)
3011                         brelse(bhs[i]);
3012         }
3013         kfree(bhs);
3014
3015         kfree(bucket_buf);
3016         return ret;
3017 }
3018
3019 /*
3020  * Move half nums of the xattr bucket in the previous cluster to this new
3021  * cluster. We only touch the last cluster of the previous extend record.
3022  *
3023  * first_bh is the first buffer_head of a series of bucket in the same
3024  * extent rec and header_bh is the header of one bucket in this cluster.
3025  * They will be updated if we move the data header_bh contains to the new
3026  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3027  */
3028 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3029                                                handle_t *handle,
3030                                                struct buffer_head **first_bh,
3031                                                struct buffer_head **header_bh,
3032                                                u64 new_blkno,
3033                                                u64 prev_blkno,
3034                                                u32 num_clusters,
3035                                                u32 *first_hash)
3036 {
3037         int i, ret, credits;
3038         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3039         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3040         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3041         int blocksize = inode->i_sb->s_blocksize;
3042         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3043         struct ocfs2_xattr_header *new_xh;
3044         struct ocfs2_xattr_header *xh =
3045                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3046
3047         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3048         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3049
3050         prev_bh = *first_bh;
3051         get_bh(prev_bh);
3052         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3053
3054         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3055
3056         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3057              prev_blkno, new_blkno);
3058
3059         /*
3060          * We need to update the 1st half of the new cluster and
3061          * 1 more for the update of the 1st bucket of the previous
3062          * extent record.
3063          */
3064         credits = bpc / 2 + 1;
3065         ret = ocfs2_extend_trans(handle, credits);
3066         if (ret) {
3067                 mlog_errno(ret);
3068                 goto out;
3069         }
3070
3071         ret = ocfs2_journal_access(handle, inode, prev_bh,
3072                                    OCFS2_JOURNAL_ACCESS_WRITE);
3073         if (ret) {
3074                 mlog_errno(ret);
3075                 goto out;
3076         }
3077
3078         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3079                 old_bh = new_bh = NULL;
3080                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3081                 if (!new_bh) {
3082                         ret = -EIO;
3083                         mlog_errno(ret);
3084                         goto out;
3085                 }
3086
3087                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3088
3089                 ret = ocfs2_journal_access(handle, inode, new_bh,
3090                                            OCFS2_JOURNAL_ACCESS_CREATE);
3091                 if (ret < 0) {
3092                         mlog_errno(ret);
3093                         brelse(new_bh);
3094                         goto out;
3095                 }
3096
3097                 ret = ocfs2_read_block(osb, prev_blkno,
3098                                        &old_bh, OCFS2_BH_CACHED, inode);
3099                 if (ret < 0) {
3100                         mlog_errno(ret);
3101                         brelse(new_bh);
3102                         goto out;
3103                 }
3104
3105                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3106
3107                 if (i == 0) {
3108                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3109                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3110
3111                         if (first_hash)
3112                                 *first_hash = le32_to_cpu(
3113                                         new_xh->xh_entries[0].xe_name_hash);
3114                         new_first_bh = new_bh;
3115                         get_bh(new_first_bh);
3116                 }
3117
3118                 ocfs2_journal_dirty(handle, new_bh);
3119
3120                 if (*header_bh == old_bh) {
3121                         brelse(*header_bh);
3122                         *header_bh = new_bh;
3123                         get_bh(*header_bh);
3124
3125                         brelse(*first_bh);
3126                         *first_bh = new_first_bh;
3127                         get_bh(*first_bh);
3128                 }
3129                 brelse(new_bh);
3130                 brelse(old_bh);
3131         }
3132
3133         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3134
3135         ocfs2_journal_dirty(handle, prev_bh);
3136 out:
3137         brelse(prev_bh);
3138         brelse(new_first_bh);
3139         return ret;
3140 }
3141
3142 static int ocfs2_read_xattr_bucket(struct inode *inode,
3143                                    u64 blkno,
3144                                    struct buffer_head **bhs,
3145                                    int new)
3146 {
3147         int ret = 0;
3148         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3149
3150         if (!new)
3151                 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno,
3152                                          blk_per_bucket, bhs,
3153                                          OCFS2_BH_CACHED, inode);
3154
3155         for (i = 0; i < blk_per_bucket; i++) {
3156                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3157                 if (bhs[i] == NULL) {
3158                         ret = -EIO;
3159                         mlog_errno(ret);
3160                         break;
3161                 }
3162                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3163         }
3164
3165         return ret;
3166 }
3167
3168 /*
3169  * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3170  * first_hash will record the 1st hash of the new bucket.
3171  */
3172 static int ocfs2_half_xattr_bucket(struct inode *inode,
3173                                    handle_t *handle,
3174                                    u64 blk,
3175                                    u64 new_blk,
3176                                    u32 *first_hash,
3177                                    int new_bucket_head)
3178 {
3179         int ret, i;
3180         u16 count, start, len, name_value_len, xe_len, name_offset;
3181         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3182         struct buffer_head **s_bhs, **t_bhs = NULL;
3183         struct ocfs2_xattr_header *xh;
3184         struct ocfs2_xattr_entry *xe;
3185         int blocksize = inode->i_sb->s_blocksize;
3186
3187         mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3188              blk, new_blk);
3189
3190         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3191         if (!s_bhs)
3192                 return -ENOMEM;
3193
3194         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3195         if (ret) {
3196                 mlog_errno(ret);
3197                 goto out;
3198         }
3199
3200         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3201                                    OCFS2_JOURNAL_ACCESS_WRITE);
3202         if (ret) {
3203                 mlog_errno(ret);
3204                 goto out;
3205         }
3206
3207         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3208         if (!t_bhs) {
3209                 ret = -ENOMEM;
3210                 goto out;
3211         }
3212
3213         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3214         if (ret) {
3215                 mlog_errno(ret);
3216                 goto out;
3217         }
3218
3219         for (i = 0; i < blk_per_bucket; i++) {
3220                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3221                                            OCFS2_JOURNAL_ACCESS_CREATE);
3222                 if (ret) {
3223                         mlog_errno(ret);
3224                         goto out;
3225                 }
3226         }
3227
3228         /* copy the whole bucket to the new first. */
3229         for (i = 0; i < blk_per_bucket; i++)
3230                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3231
3232         /* update the new bucket. */
3233         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3234         count = le16_to_cpu(xh->xh_count);
3235         start = count / 2;
3236
3237         /*
3238          * Calculate the total name/value len and xh_free_start for
3239          * the old bucket first.
3240          */
3241         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3242         name_value_len = 0;
3243         for (i = 0; i < start; i++) {
3244                 xe = &xh->xh_entries[i];
3245                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3246                 if (ocfs2_xattr_is_local(xe))
3247                         xe_len +=
3248                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3249                 else
3250                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3251                 name_value_len += xe_len;
3252                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3253                         name_offset = le16_to_cpu(xe->xe_name_offset);
3254         }
3255
3256         /*
3257          * Now begin the modification to the new bucket.
3258          *
3259          * In the new bucket, We just move the xattr entry to the beginning
3260          * and don't touch the name/value. So there will be some holes in the
3261          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3262          * called.
3263          */
3264         xe = &xh->xh_entries[start];
3265         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3266         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3267              (int)((char *)xe - (char *)xh),
3268              (int)((char *)xh->xh_entries - (char *)xh));
3269         memmove((char *)xh->xh_entries, (char *)xe, len);
3270         xe = &xh->xh_entries[count - start];
3271         len = sizeof(struct ocfs2_xattr_entry) * start;
3272         memset((char *)xe, 0, len);
3273
3274         le16_add_cpu(&xh->xh_count, -start);
3275         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3276
3277         /* Calculate xh_free_start for the new bucket. */
3278         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3279         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3280                 xe = &xh->xh_entries[i];
3281                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3282                 if (ocfs2_xattr_is_local(xe))
3283                         xe_len +=
3284                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3285                 else
3286                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3287                 if (le16_to_cpu(xe->xe_name_offset) <
3288                     le16_to_cpu(xh->xh_free_start))
3289                         xh->xh_free_start = xe->xe_name_offset;
3290         }
3291
3292         /* set xh->xh_num_buckets for the new xh. */
3293         if (new_bucket_head)
3294                 xh->xh_num_buckets = cpu_to_le16(1);
3295         else
3296                 xh->xh_num_buckets = 0;
3297
3298         for (i = 0; i < blk_per_bucket; i++) {
3299                 ocfs2_journal_dirty(handle, t_bhs[i]);
3300                 if (ret)
3301                         mlog_errno(ret);
3302         }
3303
3304         /* store the first_hash of the new bucket. */
3305         if (first_hash)
3306                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3307
3308         /*
3309          * Now only update the 1st block of the old bucket.
3310          * Please note that the entry has been sorted already above.
3311          */
3312         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3313         memset(&xh->xh_entries[start], 0,
3314                sizeof(struct ocfs2_xattr_entry) * (count - start));
3315         xh->xh_count = cpu_to_le16(start);
3316         xh->xh_free_start = cpu_to_le16(name_offset);
3317         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3318
3319         ocfs2_journal_dirty(handle, s_bhs[0]);
3320         if (ret)
3321                 mlog_errno(ret);
3322
3323 out:
3324         if (s_bhs) {
3325                 for (i = 0; i < blk_per_bucket; i++)
3326                         brelse(s_bhs[i]);
3327         }
3328         kfree(s_bhs);
3329
3330         if (t_bhs) {
3331                 for (i = 0; i < blk_per_bucket; i++)
3332                         brelse(t_bhs[i]);
3333         }
3334         kfree(t_bhs);
3335
3336         return ret;
3337 }
3338
3339 /*
3340  * Copy xattr from one bucket to another bucket.
3341  *
3342  * The caller must make sure that the journal transaction
3343  * has enough space for journaling.
3344  */
3345 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3346                                  handle_t *handle,
3347                                  u64 s_blkno,
3348                                  u64 t_blkno,
3349                                  int t_is_new)
3350 {
3351         int ret, i;
3352         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3353         int blocksize = inode->i_sb->s_blocksize;
3354         struct buffer_head **s_bhs, **t_bhs = NULL;
3355
3356         BUG_ON(s_blkno == t_blkno);
3357
3358         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3359              s_blkno, t_blkno, t_is_new);
3360
3361         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3362                         GFP_NOFS);
3363         if (!s_bhs)
3364                 return -ENOMEM;
3365
3366         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3367         if (ret)
3368                 goto out;
3369
3370         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3371                         GFP_NOFS);
3372         if (!t_bhs) {
3373                 ret = -ENOMEM;
3374                 goto out;
3375         }
3376
3377         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3378         if (ret)
3379                 goto out;
3380
3381         for (i = 0; i < blk_per_bucket; i++) {
3382                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3383                                            OCFS2_JOURNAL_ACCESS_WRITE);
3384                 if (ret)
3385                         goto out;
3386         }
3387
3388         for (i = 0; i < blk_per_bucket; i++) {
3389                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3390                 ocfs2_journal_dirty(handle, t_bhs[i]);
3391         }
3392
3393 out:
3394         if (s_bhs) {
3395                 for (i = 0; i < blk_per_bucket; i++)
3396                         brelse(s_bhs[i]);
3397         }
3398         kfree(s_bhs);
3399
3400         if (t_bhs) {
3401                 for (i = 0; i < blk_per_bucket; i++)
3402                         brelse(t_bhs[i]);
3403         }
3404         kfree(t_bhs);
3405
3406         return ret;
3407 }
3408
3409 /*
3410  * Copy one xattr cluster from src_blk to to_blk.
3411  * The to_blk will become the first bucket header of the cluster, so its
3412  * xh_num_buckets will be initialized as the bucket num in the cluster.
3413  */
3414 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3415                                   handle_t *handle,
3416                                   struct buffer_head *first_bh,
3417                                   u64 src_blk,
3418                                   u64 to_blk,
3419                                   u32 *first_hash)
3420 {
3421         int i, ret, credits;
3422         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3423         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3424         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3425         struct buffer_head *bh = NULL;
3426         struct ocfs2_xattr_header *xh;
3427         u64 to_blk_start = to_blk;
3428
3429         mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3430
3431         /*
3432          * We need to update the new cluster and 1 more for the update of
3433          * the 1st bucket of the previous extent rec.
3434          */
3435         credits = bpc + 1;
3436         ret = ocfs2_extend_trans(handle, credits);
3437         if (ret) {
3438                 mlog_errno(ret);
3439                 goto out;
3440         }
3441
3442         ret = ocfs2_journal_access(handle, inode, first_bh,
3443                                    OCFS2_JOURNAL_ACCESS_WRITE);
3444         if (ret) {
3445                 mlog_errno(ret);
3446                 goto out;
3447         }
3448
3449         for (i = 0; i < num_buckets; i++) {
3450                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3451                                             src_blk, to_blk, 1);
3452                 if (ret) {
3453                         mlog_errno(ret);
3454                         goto out;
3455                 }
3456
3457                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3458                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3459         }
3460
3461         /* update the old bucket header. */
3462         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3463         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3464
3465         ocfs2_journal_dirty(handle, first_bh);
3466
3467         /* update the new bucket header. */
3468         ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode);
3469         if (ret < 0) {
3470                 mlog_errno(ret);
3471                 goto out;
3472         }
3473
3474         ret = ocfs2_journal_access(handle, inode, bh,
3475                                    OCFS2_JOURNAL_ACCESS_WRITE);
3476         if (ret) {
3477                 mlog_errno(ret);
3478                 goto out;
3479         }
3480
3481         xh = (struct ocfs2_xattr_header *)bh->b_data;
3482         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3483
3484         ocfs2_journal_dirty(handle, bh);
3485
3486         if (first_hash)
3487                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3488 out:
3489         brelse(bh);
3490         return ret;
3491 }
3492
3493 /*
3494  * Move half of the xattrs in this cluster to the new cluster.
3495  * This function should only be called when bucket size == cluster size.
3496  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3497  */
3498 static int ocfs2_half_xattr_cluster(struct inode *inode,
3499                                     handle_t *handle,
3500                                     u64 prev_blk,
3501                                     u64 new_blk,
3502                                     u32 *first_hash)
3503 {
3504         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3505         int ret, credits = 2 * blk_per_bucket;
3506
3507         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3508
3509         ret = ocfs2_extend_trans(handle, credits);
3510         if (ret) {
3511                 mlog_errno(ret);
3512                 return ret;
3513         }
3514
3515         /* Move half of the xattr in start_blk to the next bucket. */
3516         return  ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3517                                         new_blk, first_hash, 1);
3518 }
3519
3520 /*
3521  * Move some xattrs from the old cluster to the new one since they are not
3522  * contiguous in ocfs2 xattr tree.
3523  *
3524  * new_blk starts a new separate cluster, and we will move some xattrs from
3525  * prev_blk to it. v_start will be set as the first name hash value in this
3526  * new cluster so that it can be used as e_cpos during tree insertion and
3527  * don't collide with our original b-tree operations. first_bh and header_bh
3528  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3529  * to extend the insert bucket.
3530  *
3531  * The problem is how much xattr should we move to the new one and when should
3532  * we update first_bh and header_bh?
3533  * 1. If cluster size > bucket size, that means the previous cluster has more
3534  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3535  *    update the first_bh and header_bh if the insert bucket has been moved
3536  *    to the new cluster.
3537  * 2. If cluster_size == bucket_size:
3538  *    a) If the previous extent rec has more than one cluster and the insert
3539  *       place isn't in the last cluster, copy the entire last cluster to the
3540  *       new one. This time, we don't need to upate the first_bh and header_bh
3541  *       since they will not be moved into the new cluster.
3542  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3543  *       the new one. And we set the extend flag to zero if the insert place is
3544  *       moved into the new allocated cluster since no extend is needed.
3545  */
3546 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3547                                             handle_t *handle,
3548                                             struct buffer_head **first_bh,
3549                                             struct buffer_head **header_bh,
3550                                             u64 new_blk,
3551                                             u64 prev_blk,
3552                                             u32 prev_clusters,
3553                                             u32 *v_start,
3554                                             int *extend)
3555 {
3556         int ret = 0;
3557         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3558
3559         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3560              prev_blk, prev_clusters, new_blk);
3561
3562         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3563                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3564                                                           handle,
3565                                                           first_bh,
3566                                                           header_bh,
3567                                                           new_blk,
3568                                                           prev_blk,
3569                                                           prev_clusters,
3570                                                           v_start);
3571         else {
3572                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3573
3574                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3575                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3576                                                      last_blk, new_blk,
3577                                                      v_start);
3578                 else {
3579                         ret = ocfs2_half_xattr_cluster(inode, handle,
3580                                                        last_blk, new_blk,
3581                                                        v_start);
3582
3583                         if ((*header_bh)->b_blocknr == last_blk && extend)
3584                                 *extend = 0;
3585                 }
3586         }
3587
3588         return ret;
3589 }
3590
3591 /*
3592  * Add a new cluster for xattr storage.
3593  *
3594  * If the new cluster is contiguous with the previous one, it will be
3595  * appended to the same extent record, and num_clusters will be updated.
3596  * If not, we will insert a new extent for it and move some xattrs in
3597  * the last cluster into the new allocated one.
3598  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3599  * lose the benefits of hashing because we'll have to search large leaves.
3600  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3601  * if it's bigger).
3602  *
3603  * first_bh is the first block of the previous extent rec and header_bh
3604  * indicates the bucket we will insert the new xattrs. They will be updated
3605  * when the header_bh is moved into the new cluster.
3606  */
3607 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3608                                        struct buffer_head *root_bh,
3609                                        struct buffer_head **first_bh,
3610                                        struct buffer_head **header_bh,
3611                                        u32 *num_clusters,
3612                                        u32 prev_cpos,
3613                                        u64 prev_blkno,
3614                                        int *extend)
3615 {
3616         int ret, credits;
3617         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3618         u32 prev_clusters = *num_clusters;
3619         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3620         u64 block;
3621         handle_t *handle = NULL;
3622         struct ocfs2_alloc_context *data_ac = NULL;
3623         struct ocfs2_alloc_context *meta_ac = NULL;
3624         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3625         struct ocfs2_extent_tree et;
3626
3627         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3628              "previous xattr blkno = %llu\n",
3629              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3630              prev_cpos, prev_blkno);
3631
3632         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3633
3634         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3635                                     &data_ac, &meta_ac);
3636         if (ret) {
3637                 mlog_errno(ret);
3638                 goto leave;
3639         }
3640
3641         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3642                                             clusters_to_add);
3643         handle = ocfs2_start_trans(osb, credits);
3644         if (IS_ERR(handle)) {
3645                 ret = PTR_ERR(handle);
3646                 handle = NULL;
3647                 mlog_errno(ret);
3648                 goto leave;
3649         }
3650
3651         ret = ocfs2_journal_access(handle, inode, root_bh,
3652                                    OCFS2_JOURNAL_ACCESS_WRITE);
3653         if (ret < 0) {
3654                 mlog_errno(ret);
3655                 goto leave;
3656         }
3657
3658         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3659                                      clusters_to_add, &bit_off, &num_bits);
3660         if (ret < 0) {
3661                 if (ret != -ENOSPC)
3662                         mlog_errno(ret);
3663                 goto leave;
3664         }
3665
3666         BUG_ON(num_bits > clusters_to_add);
3667
3668         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3669         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3670              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3671
3672         if (prev_blkno + prev_clusters * bpc == block &&
3673             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3674              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3675                 /*
3676                  * If this cluster is contiguous with the old one and
3677                  * adding this new cluster, we don't surpass the limit of
3678                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3679                  * initialized and used like other buckets in the previous
3680                  * cluster.
3681                  * So add it as a contiguous one. The caller will handle
3682                  * its init process.
3683                  */
3684                 v_start = prev_cpos + prev_clusters;
3685                 *num_clusters = prev_clusters + num_bits;
3686                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3687                      num_bits);
3688         } else {
3689                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3690                                                        handle,
3691                                                        first_bh,
3692                                                        header_bh,
3693                                                        block,
3694                                                        prev_blkno,
3695                                                        prev_clusters,
3696                                                        &v_start,
3697                                                        extend);
3698                 if (ret) {
3699                         mlog_errno(ret);
3700                         goto leave;
3701                 }
3702         }
3703
3704         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3705              num_bits, block, v_start);
3706         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3707                                   num_bits, 0, meta_ac);
3708         if (ret < 0) {
3709                 mlog_errno(ret);
3710                 goto leave;
3711         }
3712
3713         ret = ocfs2_journal_dirty(handle, root_bh);
3714         if (ret < 0) {
3715                 mlog_errno(ret);
3716                 goto leave;
3717         }
3718
3719 leave:
3720         if (handle)
3721                 ocfs2_commit_trans(osb, handle);
3722         if (data_ac)
3723                 ocfs2_free_alloc_context(data_ac);
3724         if (meta_ac)
3725                 ocfs2_free_alloc_context(meta_ac);
3726
3727         return ret;
3728 }
3729
3730 /*
3731  * Extend a new xattr bucket and move xattrs to the end one by one until
3732  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3733  */
3734 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3735                                      struct buffer_head *first_bh,
3736                                      struct buffer_head *start_bh,
3737                                      u32 num_clusters)
3738 {
3739         int ret, credits;
3740         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3741         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3742         u64 start_blk = start_bh->b_blocknr, end_blk;
3743         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3744         handle_t *handle;
3745         struct ocfs2_xattr_header *first_xh =
3746                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3747         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3748
3749         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3750              "from %llu, len = %u\n", start_blk,
3751              (unsigned long long)first_bh->b_blocknr, num_clusters);
3752
3753         BUG_ON(bucket >= num_buckets);
3754
3755         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3756
3757         /*
3758          * We will touch all the buckets after the start_bh(include it).
3759          * Add one more bucket and modify the first_bh.
3760          */
3761         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3762         handle = ocfs2_start_trans(osb, credits);
3763         if (IS_ERR(handle)) {
3764                 ret = PTR_ERR(handle);
3765                 handle = NULL;
3766                 mlog_errno(ret);
3767                 goto out;
3768         }
3769
3770         ret = ocfs2_journal_access(handle, inode, first_bh,
3771                                    OCFS2_JOURNAL_ACCESS_WRITE);
3772         if (ret) {
3773                 mlog_errno(ret);
3774                 goto commit;
3775         }
3776
3777         while (end_blk != start_blk) {
3778                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3779                                             end_blk + blk_per_bucket, 0);
3780                 if (ret)
3781                         goto commit;
3782                 end_blk -= blk_per_bucket;
3783         }
3784
3785         /* Move half of the xattr in start_blk to the next bucket. */
3786         ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3787                                       start_blk + blk_per_bucket, NULL, 0);
3788
3789         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3790         ocfs2_journal_dirty(handle, first_bh);
3791
3792 commit:
3793         ocfs2_commit_trans(osb, handle);
3794 out:
3795         return ret;
3796 }
3797
3798 /*
3799  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3800  * xb_bh is the ocfs2_xattr_block.
3801  * We will move all the buckets starting from header_bh to the next place. As
3802  * for this one, half num of its xattrs will be moved to the next one.
3803  *
3804  * We will allocate a new cluster if current cluster is full and adjust
3805  * header_bh and first_bh if the insert place is moved to the new cluster.
3806  */
3807 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3808                                       struct buffer_head *xb_bh,
3809                                       struct buffer_head *header_bh)
3810 {
3811         struct ocfs2_xattr_header *first_xh = NULL;
3812         struct buffer_head *first_bh = NULL;
3813         struct ocfs2_xattr_block *xb =
3814                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3815         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3816         struct ocfs2_extent_list *el = &xb_root->xt_list;
3817         struct ocfs2_xattr_header *xh =
3818                         (struct ocfs2_xattr_header *)header_bh->b_data;
3819         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3820         struct super_block *sb = inode->i_sb;
3821         struct ocfs2_super *osb = OCFS2_SB(sb);
3822         int ret, num_buckets, extend = 1;
3823         u64 p_blkno;
3824         u32 e_cpos, num_clusters;
3825
3826         mlog(0, "Add new xattr bucket starting form %llu\n",
3827              (unsigned long long)header_bh->b_blocknr);
3828
3829         /*
3830          * Add refrence for header_bh here because it may be
3831          * changed in ocfs2_add_new_xattr_cluster and we need
3832          * to free it in the end.
3833          */
3834         get_bh(header_bh);
3835
3836         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3837                                   &num_clusters, el);
3838         if (ret) {
3839                 mlog_errno(ret);
3840                 goto out;
3841         }
3842
3843         ret = ocfs2_read_block(osb, p_blkno,
3844                                &first_bh, OCFS2_BH_CACHED, inode);
3845         if (ret) {
3846                 mlog_errno(ret);
3847                 goto out;
3848         }
3849
3850         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3851         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3852
3853         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3854                 ret = ocfs2_add_new_xattr_cluster(inode,
3855                                                   xb_bh,
3856                                                   &first_bh,
3857                                                   &header_bh,
3858                                                   &num_clusters,
3859                                                   e_cpos,
3860                                                   p_blkno,
3861                                                   &extend);
3862                 if (ret) {
3863                         mlog_errno(ret);
3864                         goto out;
3865                 }
3866         }
3867
3868         if (extend)
3869                 ret = ocfs2_extend_xattr_bucket(inode,
3870                                                 first_bh,
3871                                                 header_bh,
3872                                                 num_clusters);
3873         if (ret)
3874                 mlog_errno(ret);
3875 out:
3876         brelse(first_bh);
3877         brelse(header_bh);
3878         return ret;
3879 }
3880
3881 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3882                                         struct ocfs2_xattr_bucket *bucket,
3883                                         int offs)
3884 {
3885         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3886
3887         offs = offs % inode->i_sb->s_blocksize;
3888         return bucket->bhs[block_off]->b_data + offs;
3889 }
3890
3891 /*
3892  * Handle the normal xattr set, including replace, delete and new.
3893  * When the bucket is empty, "is_empty" is set and the caller can
3894  * free this bucket.
3895  *
3896  * Note: "local" indicates the real data's locality. So we can't
3897  * just its bucket locality by its length.
3898  */
3899 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3900                                          struct ocfs2_xattr_info *xi,
3901                                          struct ocfs2_xattr_search *xs,
3902                                          u32 name_hash,
3903                                          int local,
3904                                          int *is_empty)
3905 {
3906         struct ocfs2_xattr_entry *last, *xe;
3907         int name_len = strlen(xi->name);
3908         struct ocfs2_xattr_header *xh = xs->header;
3909         u16 count = le16_to_cpu(xh->xh_count), start;
3910         size_t blocksize = inode->i_sb->s_blocksize;
3911         char *val;
3912         size_t offs, size, new_size;
3913
3914         last = &xh->xh_entries[count];
3915         if (!xs->not_found) {
3916                 xe = xs->here;
3917                 offs = le16_to_cpu(xe->xe_name_offset);
3918                 if (ocfs2_xattr_is_local(xe))
3919                         size = OCFS2_XATTR_SIZE(name_len) +
3920                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3921                 else
3922                         size = OCFS2_XATTR_SIZE(name_len) +
3923                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3924
3925                 /*
3926                  * If the new value will be stored outside, xi->value has been
3927                  * initalized as an empty ocfs2_xattr_value_root, and the same
3928                  * goes with xi->value_len, so we can set new_size safely here.
3929                  * See ocfs2_xattr_set_in_bucket.
3930                  */
3931                 new_size = OCFS2_XATTR_SIZE(name_len) +
3932                            OCFS2_XATTR_SIZE(xi->value_len);
3933
3934                 le16_add_cpu(&xh->xh_name_value_len, -size);
3935                 if (xi->value) {
3936                         if (new_size > size)
3937                                 goto set_new_name_value;
3938
3939                         /* Now replace the old value with new one. */
3940                         if (local)
3941                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3942                         else
3943                                 xe->xe_value_size = 0;
3944
3945                         val = ocfs2_xattr_bucket_get_val(inode,
3946                                                          &xs->bucket, offs);
3947                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3948                                size - OCFS2_XATTR_SIZE(name_len));
3949                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3950                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3951                                        xi->value, xi->value_len);
3952
3953                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3954                         ocfs2_xattr_set_local(xe, local);
3955                         return;
3956                 } else {
3957                         /* Remove the old entry. */
3958                         last -= 1;
3959                         memmove(xe, xe + 1,
3960                                 (void *)last - (void *)xe);
3961                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
3962                         le16_add_cpu(&xh->xh_count, -1);
3963                         if (xh->xh_count == 0 && is_empty)
3964                                 *is_empty = 1;
3965                         return;
3966                 }
3967         } else {
3968                 /* find a new entry for insert. */
3969                 int low = 0, high = count - 1, tmp;
3970                 struct ocfs2_xattr_entry *tmp_xe;
3971
3972                 while (low <= high) {
3973                         tmp = (low + high) / 2;
3974                         tmp_xe = &xh->xh_entries[tmp];
3975
3976                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
3977                                 low = tmp + 1;
3978                         else if (name_hash <
3979                                  le32_to_cpu(tmp_xe->xe_name_hash))
3980                                 high = tmp - 1;
3981                         else
3982                                 break;
3983                 }
3984
3985                 xe = &xh->xh_entries[low];
3986                 if (low != count)
3987                         memmove(xe + 1, xe, (void *)last - (void *)xe);
3988
3989                 le16_add_cpu(&xh->xh_count, 1);
3990                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
3991                 xe->xe_name_hash = cpu_to_le32(name_hash);
3992                 xe->xe_name_len = name_len;
3993                 ocfs2_xattr_set_type(xe, xi->name_index);
3994         }
3995
3996 set_new_name_value:
3997         /* Insert the new name+value. */
3998         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
3999
4000         /*
4001          * We must make sure that the name/value pair
4002          * exists in the same block.
4003          */
4004         offs = le16_to_cpu(xh->xh_free_start);
4005         start = offs - size;
4006
4007         if (start >> inode->i_sb->s_blocksize_bits !=
4008             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4009                 offs = offs - offs % blocksize;
4010                 xh->xh_free_start = cpu_to_le16(offs);
4011         }
4012
4013         val = ocfs2_xattr_bucket_get_val(inode,
4014                                          &xs->bucket, offs - size);
4015         xe->xe_name_offset = cpu_to_le16(offs - size);
4016
4017         memset(val, 0, size);
4018         memcpy(val, xi->name, name_len);
4019         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4020
4021         xe->xe_value_size = cpu_to_le64(xi->value_len);
4022         ocfs2_xattr_set_local(xe, local);
4023         xs->here = xe;
4024         le16_add_cpu(&xh->xh_free_start, -size);
4025         le16_add_cpu(&xh->xh_name_value_len, size);
4026
4027         return;
4028 }
4029
4030 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4031                                              handle_t *handle,
4032                                              struct ocfs2_xattr_search *xs,
4033                                              struct buffer_head **bhs,
4034                                              u16 bh_num)
4035 {
4036         int ret = 0, off, block_off;
4037         struct ocfs2_xattr_entry *xe = xs->here;
4038
4039         /*
4040          * First calculate all the blocks we should journal_access
4041          * and journal_dirty. The first block should always be touched.
4042          */
4043         ret = ocfs2_journal_dirty(handle, bhs[0]);
4044         if (ret)
4045                 mlog_errno(ret);
4046
4047         /* calc the data. */
4048         off = le16_to_cpu(xe->xe_name_offset);
4049         block_off = off >> inode->i_sb->s_blocksize_bits;
4050         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4051         if (ret)
4052                 mlog_errno(ret);
4053
4054         return ret;
4055 }
4056
4057 /*
4058  * Set the xattr entry in the specified bucket.
4059  * The bucket is indicated by xs->bucket and it should have the enough
4060  * space for the xattr insertion.
4061  */
4062 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4063                                            struct ocfs2_xattr_info *xi,
4064                                            struct ocfs2_xattr_search *xs,
4065                                            u32 name_hash,
4066                                            int local,
4067                                            int *bucket_empty)
4068 {
4069         int i, ret;
4070         handle_t *handle = NULL;
4071         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4072         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4073
4074         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4075              (unsigned long)xi->value_len, xi->name_index,
4076              (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4077
4078         if (!xs->bucket.bhs[1]) {
4079                 ret = ocfs2_read_blocks(osb,
4080                                         xs->bucket.bhs[0]->b_blocknr + 1,
4081                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
4082                                         OCFS2_BH_CACHED, inode);
4083                 if (ret) {
4084                         mlog_errno(ret);
4085                         goto out;
4086                 }
4087         }
4088
4089         handle = ocfs2_start_trans(osb, blk_per_bucket);
4090         if (IS_ERR(handle)) {
4091                 ret = PTR_ERR(handle);
4092                 handle = NULL;
4093                 mlog_errno(ret);
4094                 goto out;
4095         }
4096
4097         for (i = 0; i < blk_per_bucket; i++) {
4098                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4099                                            OCFS2_JOURNAL_ACCESS_WRITE);
4100                 if (ret < 0) {
4101                         mlog_errno(ret);
4102                         goto out;
4103                 }
4104         }
4105
4106         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash,
4107                                      local, bucket_empty);
4108
4109         /*Only dirty the blocks we have touched in set xattr. */
4110         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4111                                                 xs->bucket.bhs, blk_per_bucket);
4112         if (ret)
4113                 mlog_errno(ret);
4114 out:
4115         ocfs2_commit_trans(osb, handle);
4116
4117         return ret;
4118 }
4119
4120 static int ocfs2_xattr_value_update_size(struct inode *inode,
4121                                          struct buffer_head *xe_bh,
4122                                          struct ocfs2_xattr_entry *xe,
4123                                          u64 new_size)
4124 {
4125         int ret;
4126         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4127         handle_t *handle = NULL;
4128
4129         handle = ocfs2_start_trans(osb, 1);
4130         if (handle == NULL) {
4131                 ret = -ENOMEM;
4132                 mlog_errno(ret);
4133                 goto out;
4134         }
4135
4136         ret = ocfs2_journal_access(handle, inode, xe_bh,
4137                                    OCFS2_JOURNAL_ACCESS_WRITE);
4138         if (ret < 0) {
4139                 mlog_errno(ret);
4140                 goto out_commit;
4141         }
4142
4143         xe->xe_value_size = cpu_to_le64(new_size);
4144
4145         ret = ocfs2_journal_dirty(handle, xe_bh);
4146         if (ret < 0)
4147                 mlog_errno(ret);
4148
4149 out_commit:
4150         ocfs2_commit_trans(osb, handle);
4151 out:
4152         return ret;
4153 }
4154
4155 /*
4156  * Truncate the specified xe_off entry in xattr bucket.
4157  * bucket is indicated by header_bh and len is the new length.
4158  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4159  *
4160  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4161  */
4162 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4163                                              struct buffer_head *header_bh,
4164                                              int xe_off,
4165                                              int len)
4166 {
4167         int ret, offset;
4168         u64 value_blk;
4169         struct buffer_head *value_bh = NULL;
4170         struct ocfs2_xattr_value_root *xv;
4171         struct ocfs2_xattr_entry *xe;
4172         struct ocfs2_xattr_header *xh =
4173                         (struct ocfs2_xattr_header *)header_bh->b_data;
4174         size_t blocksize = inode->i_sb->s_blocksize;
4175
4176         xe = &xh->xh_entries[xe_off];
4177
4178         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4179
4180         offset = le16_to_cpu(xe->xe_name_offset) +
4181                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4182
4183         value_blk = offset / blocksize;
4184
4185         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4186         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4187         value_blk += header_bh->b_blocknr;
4188
4189         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk,
4190                                &value_bh, OCFS2_BH_CACHED, inode);
4191         if (ret) {
4192                 mlog_errno(ret);
4193                 goto out;
4194         }
4195
4196         xv = (struct ocfs2_xattr_value_root *)
4197                 (value_bh->b_data + offset % blocksize);
4198
4199         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4200              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4201         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4202         if (ret) {
4203                 mlog_errno(ret);
4204                 goto out;
4205         }
4206
4207         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4208         if (ret) {
4209                 mlog_errno(ret);
4210                 goto out;
4211         }
4212
4213 out:
4214         brelse(value_bh);
4215         return ret;
4216 }
4217
4218 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4219                                                 struct ocfs2_xattr_search *xs,
4220                                                 int len)
4221 {
4222         int ret, offset;
4223         struct ocfs2_xattr_entry *xe = xs->here;
4224         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4225
4226         BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4227
4228         offset = xe - xh->xh_entries;
4229         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4230                                                 offset, len);
4231         if (ret)
4232                 mlog_errno(ret);
4233
4234         return ret;
4235 }
4236
4237 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4238                                                 struct ocfs2_xattr_search *xs,
4239                                                 char *val,
4240                                                 int value_len)
4241 {
4242         int offset;
4243         struct ocfs2_xattr_value_root *xv;
4244         struct ocfs2_xattr_entry *xe = xs->here;
4245
4246         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4247
4248         offset = le16_to_cpu(xe->xe_name_offset) +
4249                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4250
4251         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4252
4253         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4254 }
4255
4256 /*
4257  * Remove the xattr bucket pointed by bucket_bh.
4258  * All the buckets after it in the same xattr extent rec will be
4259  * move forward one by one.
4260  */
4261 static int ocfs2_rm_xattr_bucket(struct inode *inode,
4262                                  struct buffer_head *first_bh,
4263                                  struct ocfs2_xattr_bucket *bucket)
4264 {
4265         int ret = 0, credits;
4266         struct ocfs2_xattr_header *xh =
4267                                 (struct ocfs2_xattr_header *)first_bh->b_data;
4268         u16 bucket_num = le16_to_cpu(xh->xh_num_buckets);
4269         u64 end, start = bucket->bhs[0]->b_blocknr;
4270         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4271         handle_t *handle;
4272         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4273
4274         end = first_bh->b_blocknr + (bucket_num - 1) * blk_per_bucket;
4275
4276         mlog(0, "rm xattr bucket %llu\n", start);
4277         /*
4278          * We need to update the first xattr_header and all the buckets starting
4279          * from start in this xattr rec.
4280          *
4281          * XXX: Should we empty the old last bucket here?
4282          */
4283         credits = 1 + end - start;
4284         handle = ocfs2_start_trans(osb, credits);
4285         if (IS_ERR(handle)) {
4286                 ret = PTR_ERR(handle);
4287                 mlog_errno(ret);
4288                 return ret;
4289         }
4290
4291         ret = ocfs2_journal_access(handle, inode, first_bh,
4292                                    OCFS2_JOURNAL_ACCESS_WRITE);
4293         if (ret) {
4294                 mlog_errno(ret);
4295                 goto out_commit;
4296         }
4297
4298
4299         while (start < end) {
4300                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4301                                             start + blk_per_bucket,
4302                                             start, 0);
4303                 if (ret) {
4304                         mlog_errno(ret);
4305                         goto out_commit;
4306                 }
4307                 start += blk_per_bucket;
4308         }
4309
4310         /* update the first_bh. */
4311         xh->xh_num_buckets = cpu_to_le16(bucket_num - 1);
4312         ocfs2_journal_dirty(handle, first_bh);
4313
4314 out_commit:
4315         ocfs2_commit_trans(osb, handle);
4316         return ret;
4317 }
4318
4319 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4320                                   struct buffer_head *root_bh,
4321                                   u64 blkno,
4322                                   u32 cpos,
4323                                   u32 len)
4324 {
4325         int ret;
4326         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4327         struct inode *tl_inode = osb->osb_tl_inode;
4328         handle_t *handle;
4329         struct ocfs2_xattr_block *xb =
4330                         (struct ocfs2_xattr_block *)root_bh->b_data;
4331         struct ocfs2_alloc_context *meta_ac = NULL;
4332         struct ocfs2_cached_dealloc_ctxt dealloc;
4333         struct ocfs2_extent_tree et;
4334
4335         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4336
4337         ocfs2_init_dealloc_ctxt(&dealloc);
4338
4339         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4340              cpos, len, (unsigned long long)blkno);
4341
4342         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4343
4344         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4345         if (ret) {
4346                 mlog_errno(ret);
4347                 return ret;
4348         }
4349
4350         mutex_lock(&tl_inode->i_mutex);
4351
4352         if (ocfs2_truncate_log_needs_flush(osb)) {
4353                 ret = __ocfs2_flush_truncate_log(osb);
4354                 if (ret < 0) {
4355                         mlog_errno(ret);
4356                         goto out;
4357                 }
4358         }
4359
4360         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4361         if (handle == NULL) {
4362                 ret = -ENOMEM;
4363                 mlog_errno(ret);
4364                 goto out;
4365         }
4366
4367         ret = ocfs2_journal_access(handle, inode, root_bh,
4368                                    OCFS2_JOURNAL_ACCESS_WRITE);
4369         if (ret) {
4370                 mlog_errno(ret);
4371                 goto out_commit;
4372         }
4373
4374         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4375                                   &dealloc);
4376         if (ret) {
4377                 mlog_errno(ret);
4378                 goto out_commit;
4379         }
4380
4381         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4382
4383         ret = ocfs2_journal_dirty(handle, root_bh);
4384         if (ret) {
4385                 mlog_errno(ret);
4386                 goto out_commit;
4387         }
4388
4389         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4390         if (ret)
4391                 mlog_errno(ret);
4392
4393 out_commit:
4394         ocfs2_commit_trans(osb, handle);
4395 out:
4396         ocfs2_schedule_truncate_log_flush(osb, 1);
4397
4398         mutex_unlock(&tl_inode->i_mutex);
4399
4400         if (meta_ac)
4401                 ocfs2_free_alloc_context(meta_ac);
4402
4403         ocfs2_run_deallocs(osb, &dealloc);
4404
4405         return ret;
4406 }
4407
4408 /*
4409  * Free the xattr bucket indicated by xs->bucket and if all the buckets
4410  * in the clusters is free, free the clusters also.
4411  */
4412 static int ocfs2_xattr_bucket_shrink(struct inode *inode,
4413                                      struct ocfs2_xattr_info *xi,
4414                                      struct ocfs2_xattr_search *xs,
4415                                      u32 name_hash)
4416 {
4417         int ret;
4418         u32 e_cpos, num_clusters;
4419         u64 p_blkno;
4420         struct buffer_head *first_bh = NULL;
4421         struct ocfs2_xattr_header *first_xh;
4422         struct ocfs2_xattr_block *xb =
4423                         (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
4424
4425         BUG_ON(xs->header->xh_count != 0);
4426
4427         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4428                                   &e_cpos, &num_clusters,
4429                                   &xb->xb_attrs.xb_root.xt_list);
4430         if (ret) {
4431                 mlog_errno(ret);
4432                 return ret;
4433         }
4434
4435         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
4436                                &first_bh, OCFS2_BH_CACHED, inode);
4437         if (ret) {
4438                 mlog_errno(ret);
4439                 return ret;
4440         }
4441
4442         ret = ocfs2_rm_xattr_bucket(inode, first_bh, &xs->bucket);
4443         if (ret) {
4444                 mlog_errno(ret);
4445                 goto out;
4446         }
4447
4448         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4449         if (first_xh->xh_num_buckets == 0)
4450                 ret = ocfs2_rm_xattr_cluster(inode, xs->xattr_bh,
4451                                              p_blkno, e_cpos,
4452                                              num_clusters);
4453
4454 out:
4455         brelse(first_bh);
4456         return ret;
4457 }
4458
4459 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4460                                          struct ocfs2_xattr_search *xs)
4461 {
4462         handle_t *handle = NULL;
4463         struct ocfs2_xattr_header *xh = xs->bucket.xh;
4464         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4465                                                 le16_to_cpu(xh->xh_count) - 1];
4466         int ret = 0;
4467
4468         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4469         if (IS_ERR(handle)) {
4470                 ret = PTR_ERR(handle);
4471                 mlog_errno(ret);
4472                 return;
4473         }
4474
4475         ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4476                                    OCFS2_JOURNAL_ACCESS_WRITE);
4477         if (ret) {
4478                 mlog_errno(ret);
4479                 goto out_commit;
4480         }
4481
4482         /* Remove the old entry. */
4483         memmove(xs->here, xs->here + 1,
4484                 (void *)last - (void *)xs->here);
4485         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4486         le16_add_cpu(&xh->xh_count, -1);
4487
4488         ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4489         if (ret < 0)
4490                 mlog_errno(ret);
4491 out_commit:
4492         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4493 }
4494
4495 /*
4496  * Set the xattr name/value in the bucket specified in xs.
4497  *
4498  * As the new value in xi may be stored in the bucket or in an outside cluster,
4499  * we divide the whole process into 3 steps:
4500  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4501  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4502  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4503  * 4. If the clusters for the new outside value can't be allocated, we need
4504  *    to free the xattr we allocated in set.
4505  */
4506 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4507                                      struct ocfs2_xattr_info *xi,
4508                                      struct ocfs2_xattr_search *xs)
4509 {
4510         int ret, local = 1, bucket_empty = 0;
4511         size_t value_len;
4512         char *val = (char *)xi->value;
4513         struct ocfs2_xattr_entry *xe = xs->here;
4514         u32 name_hash = ocfs2_xattr_hash_by_name(inode,
4515                                                  xi->name_index, xi->name);
4516
4517         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4518                 /*
4519                  * We need to truncate the xattr storage first.
4520                  *
4521                  * If both the old and new value are stored to
4522                  * outside block, we only need to truncate
4523                  * the storage and then set the value outside.
4524                  *
4525                  * If the new value should be stored within block,
4526                  * we should free all the outside block first and
4527                  * the modification to the xattr block will be done
4528                  * by following steps.
4529                  */
4530                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4531                         value_len = xi->value_len;
4532                 else
4533                         value_len = 0;
4534
4535                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4536                                                            value_len);
4537                 if (ret)
4538                         goto out;
4539
4540                 if (value_len)
4541                         goto set_value_outside;
4542         }
4543
4544         value_len = xi->value_len;
4545         /* So we have to handle the inside block change now. */
4546         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4547                 /*
4548                  * If the new value will be stored outside of block,
4549                  * initalize a new empty value root and insert it first.
4550                  */
4551                 local = 0;
4552                 xi->value = &def_xv;
4553                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4554         }
4555
4556         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash,
4557                                               local, &bucket_empty);
4558         if (ret) {
4559                 mlog_errno(ret);
4560                 goto out;
4561         }
4562
4563         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4564                 /* allocate the space now for the outside block storage. */
4565                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4566                                                            value_len);
4567                 if (ret) {
4568                         mlog_errno(ret);
4569
4570                         if (xs->not_found) {
4571                                 /*
4572                                  * We can't allocate enough clusters for outside
4573                                  * storage and we have allocated xattr already,
4574                                  * so need to remove it.
4575                                  */
4576                                 ocfs2_xattr_bucket_remove_xs(inode, xs);
4577                         }
4578                         goto out;
4579                 }
4580         } else {
4581                 if (bucket_empty)
4582                         ret = ocfs2_xattr_bucket_shrink(inode, xi,
4583                                                         xs, name_hash);
4584                 goto out;
4585         }
4586
4587 set_value_outside:
4588         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4589 out:
4590         return ret;
4591 }
4592
4593 /* check whether the xattr bucket is filled up with the same hash value. */
4594 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4595                                               struct ocfs2_xattr_bucket *bucket)
4596 {
4597         struct ocfs2_xattr_header *xh = bucket->xh;
4598
4599         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4600             xh->xh_entries[0].xe_name_hash) {
4601                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4602                      "hash = %u\n",
4603                      (unsigned long long)bucket->bhs[0]->b_blocknr,
4604                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4605                 return -ENOSPC;
4606         }
4607
4608         return 0;
4609 }
4610
4611 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4612                                              struct ocfs2_xattr_info *xi,
4613                                              struct ocfs2_xattr_search *xs)
4614 {
4615         struct ocfs2_xattr_header *xh;
4616         struct ocfs2_xattr_entry *xe;
4617         u16 count, header_size, xh_free_start;
4618         int i, free, max_free, need, old;
4619         size_t value_size = 0, name_len = strlen(xi->name);
4620         size_t blocksize = inode->i_sb->s_blocksize;
4621         int ret, allocation = 0;
4622         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4623
4624         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4625
4626 try_again:
4627         xh = xs->header;
4628         count = le16_to_cpu(xh->xh_count);
4629         xh_free_start = le16_to_cpu(xh->xh_free_start);
4630         header_size = sizeof(struct ocfs2_xattr_header) +
4631                         count * sizeof(struct ocfs2_xattr_entry);
4632         max_free = OCFS2_XATTR_BUCKET_SIZE -
4633                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4634
4635         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4636                         "of %u which exceed block size\n",
4637                         (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4638                         header_size);
4639
4640         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4641                 value_size = OCFS2_XATTR_ROOT_SIZE;
4642         else if (xi->value)
4643                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4644
4645         if (xs->not_found)
4646                 need = sizeof(struct ocfs2_xattr_entry) +
4647                         OCFS2_XATTR_SIZE(name_len) + value_size;
4648         else {
4649                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4650
4651                 /*
4652                  * We only replace the old value if the new length is smaller
4653                  * than the old one. Otherwise we will allocate new space in the
4654                  * bucket to store it.
4655                  */
4656                 xe = xs->here;
4657                 if (ocfs2_xattr_is_local(xe))
4658                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4659                 else
4660                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4661
4662                 if (old >= value_size)
4663                         need = 0;
4664         }
4665
4666         free = xh_free_start - header_size;
4667         /*
4668          * We need to make sure the new name/value pair
4669          * can exist in the same block.
4670          */
4671         if (xh_free_start % blocksize < need)
4672                 free -= xh_free_start % blocksize;
4673
4674         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4675              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4676              " %u\n", xs->not_found,
4677              (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4678              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4679              le16_to_cpu(xh->xh_name_value_len));
4680
4681         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4682                 if (need <= max_free &&
4683                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4684                         /*
4685                          * We can create the space by defragment. Since only the
4686                          * name/value will be moved, the xe shouldn't be changed
4687                          * in xs.
4688                          */
4689                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4690                         if (ret) {
4691                                 mlog_errno(ret);
4692                                 goto out;
4693                         }
4694
4695                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4696                         free = xh_free_start - header_size;
4697                         if (xh_free_start % blocksize < need)
4698                                 free -= xh_free_start % blocksize;
4699
4700                         if (free >= need)
4701                                 goto xattr_set;
4702
4703                         mlog(0, "Can't get enough space for xattr insert by "
4704                              "defragment. Need %u bytes, but we have %d, so "
4705                              "allocate new bucket for it.\n", need, free);
4706                 }
4707
4708                 /*
4709                  * We have to add new buckets or clusters and one
4710                  * allocation should leave us enough space for insert.
4711                  */
4712                 BUG_ON(allocation);
4713
4714                 /*
4715                  * We do not allow for overlapping ranges between buckets. And
4716                  * the maximum number of collisions we will allow for then is
4717                  * one bucket's worth, so check it here whether we need to
4718                  * add a new bucket for the insert.
4719                  */
4720                 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4721                 if (ret) {
4722                         mlog_errno(ret);
4723                         goto out;
4724                 }
4725
4726                 ret = ocfs2_add_new_xattr_bucket(inode,
4727                                                  xs->xattr_bh,
4728                                                  xs->bucket.bhs[0]);
4729                 if (ret) {
4730                         mlog_errno(ret);
4731                         goto out;
4732                 }
4733
4734                 for (i = 0; i < blk_per_bucket; i++)
4735                         brelse(xs->bucket.bhs[i]);
4736
4737                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4738
4739                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4740                                                    xi->name_index,
4741                                                    xi->name, xs);
4742                 if (ret && ret != -ENODATA)
4743                         goto out;
4744                 xs->not_found = ret;
4745                 allocation = 1;
4746                 goto try_again;
4747         }
4748
4749 xattr_set:
4750         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4751 out:
4752         mlog_exit(ret);
4753         return ret;
4754 }
4755
4756 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4757                                         struct ocfs2_xattr_bucket *bucket,
4758                                         void *para)
4759 {
4760         int ret = 0;
4761         struct ocfs2_xattr_header *xh = bucket->xh;
4762         u16 i;
4763         struct ocfs2_xattr_entry *xe;
4764
4765         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4766                 xe = &xh->xh_entries[i];
4767                 if (ocfs2_xattr_is_local(xe))
4768                         continue;
4769
4770                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4771                                                         bucket->bhs[0],
4772                                                         i, 0);
4773                 if (ret) {
4774                         mlog_errno(ret);
4775                         break;
4776                 }
4777         }
4778
4779         return ret;
4780 }
4781
4782 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4783                                           struct buffer_head *xb_bh)
4784 {
4785         struct ocfs2_xattr_block *xb =
4786                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4787         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4788         int ret = 0;
4789         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4790         u64 p_blkno;
4791
4792         if (le16_to_cpu(el->l_next_free_rec) == 0)
4793                 return 0;
4794
4795         while (name_hash > 0) {
4796                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4797                                           &e_cpos, &num_clusters, el);
4798                 if (ret) {
4799                         mlog_errno(ret);
4800                         goto out;
4801                 }
4802
4803                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4804                                                   ocfs2_delete_xattr_in_bucket,
4805                                                   NULL);
4806                 if (ret) {
4807                         mlog_errno(ret);
4808                         goto out;
4809                 }
4810
4811                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4812                                              p_blkno, e_cpos, num_clusters);
4813                 if (ret) {
4814                         mlog_errno(ret);
4815                         break;
4816                 }
4817
4818                 if (e_cpos == 0)
4819                         break;
4820
4821                 name_hash = e_cpos - 1;
4822         }
4823
4824 out:
4825         return ret;
4826 }