Pileus Git - ~andy/linux/blob - 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 attribute in external block. */
1340                 ret = ocfs2_extend_trans(handle,
1341                                          OCFS2_INODE_UPDATE_CREDITS +
1342                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1343                 if (ret) {
1344                         mlog_errno(ret);
1345                         goto out_commit;
1346                 }
1347                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1348                                            OCFS2_JOURNAL_ACCESS_WRITE);
1349                 if (ret) {
1350                         mlog_errno(ret);
1351                         goto out_commit;
1352                 }
1353         }
1354
1355         /*
1356          * Set value in local, include set tree root in local.
1357          * This is the first step for value size >INLINE_SIZE.
1358          */
1359         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1360
1361         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1362                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1363                 if (ret < 0) {
1364                         mlog_errno(ret);
1365                         goto out_commit;
1366                 }
1367         }
1368
1369         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1370             (flag & OCFS2_INLINE_XATTR_FL)) {
1371                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1372                 unsigned int xattrsize = osb->s_xattr_inline_size;
1373
1374                 /*
1375                  * Adjust extent record count or inline data size
1376                  * to reserve space for extended attribute.
1377                  */
1378                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1379                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1380                         le16_add_cpu(&idata->id_count, -xattrsize);
1381                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1382                         struct ocfs2_extent_list *el = &di->id2.i_list;
1383                         le16_add_cpu(&el->l_count, -(xattrsize /
1384                                         sizeof(struct ocfs2_extent_rec)));
1385                 }
1386                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1387         }
1388         /* Update xattr flag */
1389         spin_lock(&oi->ip_lock);
1390         oi->ip_dyn_features |= flag;
1391         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1392         spin_unlock(&oi->ip_lock);
1393         /* Update inode ctime */
1394         inode->i_ctime = CURRENT_TIME;
1395         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1396         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1397
1398         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1399         if (ret < 0)
1400                 mlog_errno(ret);
1401
1402 out_commit:
1403         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1404
1405         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1406                 /*
1407                  * Set value outside in B tree.
1408                  * This is the second step for value size > INLINE_SIZE.
1409                  */
1410                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1411                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1412                 if (ret < 0) {
1413                         int ret2;
1414
1415                         mlog_errno(ret);
1416                         /*
1417                          * If set value outside failed, we have to clean
1418                          * the junk tree root we have already set in local.
1419                          */
1420                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1421                         if (ret2 < 0)
1422                                 mlog_errno(ret2);
1423                 }
1424         }
1425 out:
1426         return ret;
1427
1428 }
1429
1430 static int ocfs2_remove_value_outside(struct inode*inode,
1431                                       struct buffer_head *bh,
1432                                       struct ocfs2_xattr_header *header)
1433 {
1434         int ret = 0, i;
1435
1436         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1437                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1438
1439                 if (!ocfs2_xattr_is_local(entry)) {
1440                         struct ocfs2_xattr_value_root *xv;
1441                         void *val;
1442
1443                         val = (void *)header +
1444                                 le16_to_cpu(entry->xe_name_offset);
1445                         xv = (struct ocfs2_xattr_value_root *)
1446                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1447                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1448                         if (ret < 0) {
1449                                 mlog_errno(ret);
1450                                 return ret;
1451                         }
1452                 }
1453         }
1454
1455         return ret;
1456 }
1457
1458 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1459                                     struct buffer_head *di_bh)
1460 {
1461
1462         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1463         struct ocfs2_xattr_header *header;
1464         int ret;
1465
1466         header = (struct ocfs2_xattr_header *)
1467                  ((void *)di + inode->i_sb->s_blocksize -
1468                  le16_to_cpu(di->i_xattr_inline_size));
1469
1470         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1471
1472         return ret;
1473 }
1474
1475 static int ocfs2_xattr_block_remove(struct inode *inode,
1476                                     struct buffer_head *blk_bh)
1477 {
1478         struct ocfs2_xattr_block *xb;
1479         int ret = 0;
1480
1481         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1482         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1483                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1484                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1485         } else
1486                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1487
1488         return ret;
1489 }
1490
1491 static int ocfs2_xattr_free_block(struct inode *inode,
1492                                   u64 block)
1493 {
1494         struct inode *xb_alloc_inode;
1495         struct buffer_head *xb_alloc_bh = NULL;
1496         struct buffer_head *blk_bh = NULL;
1497         struct ocfs2_xattr_block *xb;
1498         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1499         handle_t *handle;
1500         int ret = 0;
1501         u64 blk, bg_blkno;
1502         u16 bit;
1503
1504         ret = ocfs2_read_block(osb, block, &blk_bh,
1505                                OCFS2_BH_CACHED, inode);
1506         if (ret < 0) {
1507                 mlog_errno(ret);
1508                 goto out;
1509         }
1510
1511         /*Verify the signature of xattr block*/
1512         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1513                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1514                 ret = -EFAULT;
1515                 goto out;
1516         }
1517
1518         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1519         if (ret < 0) {
1520                 mlog_errno(ret);
1521                 goto out;
1522         }
1523
1524         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1525         blk = le64_to_cpu(xb->xb_blkno);
1526         bit = le16_to_cpu(xb->xb_suballoc_bit);
1527         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1528
1529         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1530                                 EXTENT_ALLOC_SYSTEM_INODE,
1531                                 le16_to_cpu(xb->xb_suballoc_slot));
1532         if (!xb_alloc_inode) {
1533                 ret = -ENOMEM;
1534                 mlog_errno(ret);
1535                 goto out;
1536         }
1537         mutex_lock(&xb_alloc_inode->i_mutex);
1538
1539         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1540         if (ret < 0) {
1541                 mlog_errno(ret);
1542                 goto out_mutex;
1543         }
1544
1545         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1546         if (IS_ERR(handle)) {
1547                 ret = PTR_ERR(handle);
1548                 mlog_errno(ret);
1549                 goto out_unlock;
1550         }
1551
1552         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1553                                        bit, bg_blkno, 1);
1554         if (ret < 0)
1555                 mlog_errno(ret);
1556
1557         ocfs2_commit_trans(osb, handle);
1558 out_unlock:
1559         ocfs2_inode_unlock(xb_alloc_inode, 1);
1560         brelse(xb_alloc_bh);
1561 out_mutex:
1562         mutex_unlock(&xb_alloc_inode->i_mutex);
1563         iput(xb_alloc_inode);
1564 out:
1565         brelse(blk_bh);
1566         return ret;
1567 }
1568
1569 /*
1570  * ocfs2_xattr_remove()
1571  *
1572  * Free extended attribute resources associated with this inode.
1573  */
1574 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1575 {
1576         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1577         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1578         handle_t *handle;
1579         int ret;
1580
1581         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1582                 return 0;
1583
1584         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1585                 return 0;
1586
1587         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1588                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1589                 if (ret < 0) {
1590                         mlog_errno(ret);
1591                         goto out;
1592                 }
1593         }
1594
1595         if (di->i_xattr_loc) {
1596                 ret = ocfs2_xattr_free_block(inode,
1597                                              le64_to_cpu(di->i_xattr_loc));
1598                 if (ret < 0) {
1599                         mlog_errno(ret);
1600                         goto out;
1601                 }
1602         }
1603
1604         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1605                                    OCFS2_INODE_UPDATE_CREDITS);
1606         if (IS_ERR(handle)) {
1607                 ret = PTR_ERR(handle);
1608                 mlog_errno(ret);
1609                 goto out;
1610         }
1611         ret = ocfs2_journal_access(handle, inode, di_bh,
1612                                    OCFS2_JOURNAL_ACCESS_WRITE);
1613         if (ret) {
1614                 mlog_errno(ret);
1615                 goto out_commit;
1616         }
1617
1618         di->i_xattr_loc = 0;
1619
1620         spin_lock(&oi->ip_lock);
1621         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1622         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1623         spin_unlock(&oi->ip_lock);
1624
1625         ret = ocfs2_journal_dirty(handle, di_bh);
1626         if (ret < 0)
1627                 mlog_errno(ret);
1628 out_commit:
1629         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1630 out:
1631         return ret;
1632 }
1633
1634 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1635                                         struct ocfs2_dinode *di)
1636 {
1637         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1638         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1639         int free;
1640
1641         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1642                 return 0;
1643
1644         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1645                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1646                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1647         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1648                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1649                         le64_to_cpu(di->i_size);
1650         } else {
1651                 struct ocfs2_extent_list *el = &di->id2.i_list;
1652                 free = (le16_to_cpu(el->l_count) -
1653                         le16_to_cpu(el->l_next_free_rec)) *
1654                         sizeof(struct ocfs2_extent_rec);
1655         }
1656         if (free >= xattrsize)
1657                 return 1;
1658
1659         return 0;
1660 }
1661
1662 /*
1663  * ocfs2_xattr_ibody_find()
1664  *
1665  * Find extended attribute in inode block and
1666  * fill search info into struct ocfs2_xattr_search.
1667  */
1668 static int ocfs2_xattr_ibody_find(struct inode *inode,
1669                                   int name_index,
1670                                   const char *name,
1671                                   struct ocfs2_xattr_search *xs)
1672 {
1673         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1674         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1675         int ret;
1676         int has_space = 0;
1677
1678         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1679                 return 0;
1680
1681         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1682                 down_read(&oi->ip_alloc_sem);
1683                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1684                 up_read(&oi->ip_alloc_sem);
1685                 if (!has_space)
1686                         return 0;
1687         }
1688
1689         xs->xattr_bh = xs->inode_bh;
1690         xs->end = (void *)di + inode->i_sb->s_blocksize;
1691         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1692                 xs->header = (struct ocfs2_xattr_header *)
1693                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1694         else
1695                 xs->header = (struct ocfs2_xattr_header *)
1696                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1697         xs->base = (void *)xs->header;
1698         xs->here = xs->header->xh_entries;
1699
1700         /* Find the named attribute. */
1701         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1702                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1703                 if (ret && ret != -ENODATA)
1704                         return ret;
1705                 xs->not_found = ret;
1706         }
1707
1708         return 0;
1709 }
1710
1711 /*
1712  * ocfs2_xattr_ibody_set()
1713  *
1714  * Set, replace or remove an extended attribute into inode block.
1715  *
1716  */
1717 static int ocfs2_xattr_ibody_set(struct inode *inode,
1718                                  struct ocfs2_xattr_info *xi,
1719                                  struct ocfs2_xattr_search *xs)
1720 {
1721         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1722         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1723         int ret;
1724
1725         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1726                 return -ENOSPC;
1727
1728         down_write(&oi->ip_alloc_sem);
1729         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1730                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1731                         ret = -ENOSPC;
1732                         goto out;
1733                 }
1734         }
1735
1736         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1737                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1738 out:
1739         up_write(&oi->ip_alloc_sem);
1740
1741         return ret;
1742 }
1743
1744 /*
1745  * ocfs2_xattr_block_find()
1746  *
1747  * Find extended attribute in external block and
1748  * fill search info into struct ocfs2_xattr_search.
1749  */
1750 static int ocfs2_xattr_block_find(struct inode *inode,
1751                                   int name_index,
1752                                   const char *name,
1753                                   struct ocfs2_xattr_search *xs)
1754 {
1755         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1756         struct buffer_head *blk_bh = NULL;
1757         struct ocfs2_xattr_block *xb;
1758         int ret = 0;
1759
1760         if (!di->i_xattr_loc)
1761                 return ret;
1762
1763         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1764                                le64_to_cpu(di->i_xattr_loc),
1765                                &blk_bh, OCFS2_BH_CACHED, inode);
1766         if (ret < 0) {
1767                 mlog_errno(ret);
1768                 return ret;
1769         }
1770         /*Verify the signature of xattr block*/
1771         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1772                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1773                         ret = -EFAULT;
1774                         goto cleanup;
1775         }
1776
1777         xs->xattr_bh = blk_bh;
1778         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1779
1780         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1781                 xs->header = &xb->xb_attrs.xb_header;
1782                 xs->base = (void *)xs->header;
1783                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1784                 xs->here = xs->header->xh_entries;
1785
1786                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1787         } else
1788                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1789                                                    name_index,
1790                                                    name, xs);
1791
1792         if (ret && ret != -ENODATA) {
1793                 xs->xattr_bh = NULL;
1794                 goto cleanup;
1795         }
1796         xs->not_found = ret;
1797         return 0;
1798 cleanup:
1799         brelse(blk_bh);
1800
1801         return ret;
1802 }
1803
1804 /*
1805  * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1806  * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1807  * re-initialized.
1808  */
1809 static int ocfs2_restore_xattr_block(struct inode *inode,
1810                                      struct ocfs2_xattr_search *xs)
1811 {
1812         int ret;
1813         handle_t *handle;
1814         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1815         struct ocfs2_xattr_block *xb =
1816                 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1817         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1818         u16 xb_flags = le16_to_cpu(xb->xb_flags);
1819
1820         BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1821                 le16_to_cpu(el->l_next_free_rec) != 0);
1822
1823         handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1824         if (IS_ERR(handle)) {
1825                 ret = PTR_ERR(handle);
1826                 handle = NULL;
1827                 goto out;
1828         }
1829
1830         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1831                                    OCFS2_JOURNAL_ACCESS_WRITE);
1832         if (ret < 0) {
1833                 mlog_errno(ret);
1834                 goto out_commit;
1835         }
1836
1837         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1838                offsetof(struct ocfs2_xattr_block, xb_attrs));
1839
1840         xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1841
1842         ocfs2_journal_dirty(handle, xs->xattr_bh);
1843
1844 out_commit:
1845         ocfs2_commit_trans(osb, handle);
1846 out:
1847         return ret;
1848 }
1849
1850 /*
1851  * ocfs2_xattr_block_set()
1852  *
1853  * Set, replace or remove an extended attribute into external block.
1854  *
1855  */
1856 static int ocfs2_xattr_block_set(struct inode *inode,
1857                                  struct ocfs2_xattr_info *xi,
1858                                  struct ocfs2_xattr_search *xs)
1859 {
1860         struct buffer_head *new_bh = NULL;
1861         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1862         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1863         struct ocfs2_alloc_context *meta_ac = NULL;
1864         handle_t *handle = NULL;
1865         struct ocfs2_xattr_block *xblk = NULL;
1866         u16 suballoc_bit_start;
1867         u32 num_got;
1868         u64 first_blkno;
1869         int ret;
1870
1871         if (!xs->xattr_bh) {
1872                 /*
1873                  * Alloc one external block for extended attribute
1874                  * outside of inode.
1875                  */
1876                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1877                 if (ret < 0) {
1878                         mlog_errno(ret);
1879                         goto out;
1880                 }
1881                 handle = ocfs2_start_trans(osb,
1882                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1883                 if (IS_ERR(handle)) {
1884                         ret = PTR_ERR(handle);
1885                         mlog_errno(ret);
1886                         goto out;
1887                 }
1888                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1889                                            OCFS2_JOURNAL_ACCESS_CREATE);
1890                 if (ret < 0) {
1891                         mlog_errno(ret);
1892                         goto out_commit;
1893                 }
1894
1895                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1896                                            &suballoc_bit_start, &num_got,
1897                                            &first_blkno);
1898                 if (ret < 0) {
1899                         mlog_errno(ret);
1900                         goto out_commit;
1901                 }
1902
1903                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1904                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1905
1906                 ret = ocfs2_journal_access(handle, inode, new_bh,
1907                                            OCFS2_JOURNAL_ACCESS_CREATE);
1908                 if (ret < 0) {
1909                         mlog_errno(ret);
1910                         goto out_commit;
1911                 }
1912
1913                 /* Initialize ocfs2_xattr_block */
1914                 xs->xattr_bh = new_bh;
1915                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1916                 memset(xblk, 0, inode->i_sb->s_blocksize);
1917                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1918                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1919                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1920                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1921                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1922
1923                 xs->header = &xblk->xb_attrs.xb_header;
1924                 xs->base = (void *)xs->header;
1925                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1926                 xs->here = xs->header->xh_entries;
1927
1928
1929                 ret = ocfs2_journal_dirty(handle, new_bh);
1930                 if (ret < 0) {
1931                         mlog_errno(ret);
1932                         goto out_commit;
1933                 }
1934                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1935                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1936                 if (ret < 0)
1937                         mlog_errno(ret);
1938 out_commit:
1939                 ocfs2_commit_trans(osb, handle);
1940 out:
1941                 if (meta_ac)
1942                         ocfs2_free_alloc_context(meta_ac);
1943                 if (ret < 0)
1944                         return ret;
1945         } else
1946                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1947
1948         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1949                 /* Set extended attribute into external block */
1950                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1951                 if (!ret || ret != -ENOSPC)
1952                         goto end;
1953
1954                 ret = ocfs2_xattr_create_index_block(inode, xs);
1955                 if (ret)
1956                         goto end;
1957         }
1958
1959         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1960         if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1961                 ret = ocfs2_restore_xattr_block(inode, xs);
1962
1963 end:
1964
1965         return ret;
1966 }
1967
1968 /*
1969  * ocfs2_xattr_set()
1970  *
1971  * Set, replace or remove an extended attribute for this inode.
1972  * value is NULL to remove an existing extended attribute, else either
1973  * create or replace an extended attribute.
1974  */
1975 int ocfs2_xattr_set(struct inode *inode,
1976                     int name_index,
1977                     const char *name,
1978                     const void *value,
1979                     size_t value_len,
1980                     int flags)
1981 {
1982         struct buffer_head *di_bh = NULL;
1983         struct ocfs2_dinode *di;
1984         int ret;
1985         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1986
1987         struct ocfs2_xattr_info xi = {
1988                 .name_index = name_index,
1989                 .name = name,
1990                 .value = value,
1991                 .value_len = value_len,
1992         };
1993
1994         struct ocfs2_xattr_search xis = {
1995                 .not_found = -ENODATA,
1996         };
1997
1998         struct ocfs2_xattr_search xbs = {
1999                 .not_found = -ENODATA,
2000         };
2001
2002         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2003                 return -EOPNOTSUPP;
2004
2005         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2006         if (ret < 0) {
2007                 mlog_errno(ret);
2008                 return ret;
2009         }
2010         xis.inode_bh = xbs.inode_bh = di_bh;
2011         di = (struct ocfs2_dinode *)di_bh->b_data;
2012
2013         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2014         /*
2015          * Scan inode and external block to find the same name
2016          * extended attribute and collect search infomation.
2017          */
2018         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2019         if (ret)
2020                 goto cleanup;
2021         if (xis.not_found) {
2022                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2023                 if (ret)
2024                         goto cleanup;
2025         }
2026
2027         if (xis.not_found && xbs.not_found) {
2028                 ret = -ENODATA;
2029                 if (flags & XATTR_REPLACE)
2030                         goto cleanup;
2031                 ret = 0;
2032                 if (!value)
2033                         goto cleanup;
2034         } else {
2035                 ret = -EEXIST;
2036                 if (flags & XATTR_CREATE)
2037                         goto cleanup;
2038         }
2039
2040         if (!value) {
2041                 /* Remove existing extended attribute */
2042                 if (!xis.not_found)
2043                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2044                 else if (!xbs.not_found)
2045                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2046         } else {
2047                 /* We always try to set extended attribute into inode first*/
2048                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2049                 if (!ret && !xbs.not_found) {
2050                         /*
2051                          * If succeed and that extended attribute existing in
2052                          * external block, then we will remove it.
2053                          */
2054                         xi.value = NULL;
2055                         xi.value_len = 0;
2056                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2057                 } else if (ret == -ENOSPC) {
2058                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2059                                 ret = ocfs2_xattr_block_find(inode, name_index,
2060                                                              name, &xbs);
2061                                 if (ret)
2062                                         goto cleanup;
2063                         }
2064                         /*
2065                          * If no space in inode, we will set extended attribute
2066                          * into external block.
2067                          */
2068                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2069                         if (ret)
2070                                 goto cleanup;
2071                         if (!xis.not_found) {
2072                                 /*
2073                                  * If succeed and that extended attribute
2074                                  * existing in inode, we will remove it.
2075                                  */
2076                                 xi.value = NULL;
2077                                 xi.value_len = 0;
2078                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2079                         }
2080                 }
2081         }
2082 cleanup:
2083         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2084         ocfs2_inode_unlock(inode, 1);
2085         brelse(di_bh);
2086         brelse(xbs.xattr_bh);
2087         for (i = 0; i < blk_per_bucket; i++)
2088                 brelse(xbs.bucket.bhs[i]);
2089
2090         return ret;
2091 }
2092
2093 static inline u32 ocfs2_xattr_hash_by_name(struct inode *inode,
2094                                            int name_index,
2095                                            const char *suffix_name)
2096 {
2097         struct xattr_handler *handler = ocfs2_xattr_handler(name_index);
2098         char *prefix = handler->prefix;
2099         int prefix_len = strlen(handler->prefix);
2100
2101         return ocfs2_xattr_name_hash(inode, prefix, prefix_len,
2102                                      (char *)suffix_name, strlen(suffix_name));
2103 }
2104
2105 /*
2106  * Find the xattr extent rec which may contains name_hash.
2107  * e_cpos will be the first name hash of the xattr rec.
2108  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2109  */
2110 static int ocfs2_xattr_get_rec(struct inode *inode,
2111                                u32 name_hash,
2112                                u64 *p_blkno,
2113                                u32 *e_cpos,
2114                                u32 *num_clusters,
2115                                struct ocfs2_extent_list *el)
2116 {
2117         int ret = 0, i;
2118         struct buffer_head *eb_bh = NULL;
2119         struct ocfs2_extent_block *eb;
2120         struct ocfs2_extent_rec *rec = NULL;
2121         u64 e_blkno = 0;
2122
2123         if (el->l_tree_depth) {
2124                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2125                 if (ret) {
2126                         mlog_errno(ret);
2127                         goto out;
2128                 }
2129
2130                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2131                 el = &eb->h_list;
2132
2133                 if (el->l_tree_depth) {
2134                         ocfs2_error(inode->i_sb,
2135                                     "Inode %lu has non zero tree depth in "
2136                                     "xattr tree block %llu\n", inode->i_ino,
2137                                     (unsigned long long)eb_bh->b_blocknr);
2138                         ret = -EROFS;
2139                         goto out;
2140                 }
2141         }
2142
2143         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2144                 rec = &el->l_recs[i];
2145
2146                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2147                         e_blkno = le64_to_cpu(rec->e_blkno);
2148                         break;
2149                 }
2150         }
2151
2152         if (!e_blkno) {
2153                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2154                             "record (%u, %u, 0) in xattr", inode->i_ino,
2155                             le32_to_cpu(rec->e_cpos),
2156                             ocfs2_rec_clusters(el, rec));
2157                 ret = -EROFS;
2158                 goto out;
2159         }
2160
2161         *p_blkno = le64_to_cpu(rec->e_blkno);
2162         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2163         if (e_cpos)
2164                 *e_cpos = le32_to_cpu(rec->e_cpos);
2165 out:
2166         brelse(eb_bh);
2167         return ret;
2168 }
2169
2170 typedef int (xattr_bucket_func)(struct inode *inode,
2171                                 struct ocfs2_xattr_bucket *bucket,
2172                                 void *para);
2173
2174 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2175                                    struct buffer_head *header_bh,
2176                                    int name_index,
2177                                    const char *name,
2178                                    u32 name_hash,
2179                                    u16 *xe_index,
2180                                    int *found)
2181 {
2182         int i, ret = 0, cmp = 1, block_off, new_offset;
2183         struct ocfs2_xattr_header *xh =
2184                         (struct ocfs2_xattr_header *)header_bh->b_data;
2185         size_t name_len = strlen(name);
2186         struct ocfs2_xattr_entry *xe = NULL;
2187         struct buffer_head *name_bh = NULL;
2188         char *xe_name;
2189
2190         /*
2191          * We don't use binary search in the bucket because there
2192          * may be multiple entries with the same name hash.
2193          */
2194         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2195                 xe = &xh->xh_entries[i];
2196
2197                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2198                         continue;
2199                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2200                         break;
2201
2202                 cmp = name_index - ocfs2_xattr_get_type(xe);
2203                 if (!cmp)
2204                         cmp = name_len - xe->xe_name_len;
2205                 if (cmp)
2206                         continue;
2207
2208                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2209                                                         xh,
2210                                                         i,
2211                                                         &block_off,
2212                                                         &new_offset);
2213                 if (ret) {
2214                         mlog_errno(ret);
2215                         break;
2216                 }
2217
2218                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2219                                        header_bh->b_blocknr + block_off,
2220                                        &name_bh, OCFS2_BH_CACHED, inode);
2221                 if (ret) {
2222                         mlog_errno(ret);
2223                         break;
2224                 }
2225                 xe_name = name_bh->b_data + new_offset;
2226
2227                 cmp = memcmp(name, xe_name, name_len);
2228                 brelse(name_bh);
2229                 name_bh = NULL;
2230
2231                 if (cmp == 0) {
2232                         *xe_index = i;
2233                         *found = 1;
2234                         ret = 0;
2235                         break;
2236                 }
2237         }
2238
2239         return ret;
2240 }
2241
2242 /*
2243  * Find the specified xattr entry in a series of buckets.
2244  * This series start from p_blkno and last for num_clusters.
2245  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2246  * the num of the valid buckets.
2247  *
2248  * Return the buffer_head this xattr should reside in. And if the xattr's
2249  * hash is in the gap of 2 buckets, return the lower bucket.
2250  */
2251 static int ocfs2_xattr_bucket_find(struct inode *inode,
2252                                    int name_index,
2253                                    const char *name,
2254                                    u32 name_hash,
2255                                    u64 p_blkno,
2256                                    u32 first_hash,
2257                                    u32 num_clusters,
2258                                    struct ocfs2_xattr_search *xs)
2259 {
2260         int ret, found = 0;
2261         struct buffer_head *bh = NULL;
2262         struct buffer_head *lower_bh = NULL;
2263         struct ocfs2_xattr_header *xh = NULL;
2264         struct ocfs2_xattr_entry *xe = NULL;
2265         u16 index = 0;
2266         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2267         int low_bucket = 0, bucket, high_bucket;
2268         u32 last_hash;
2269         u64 blkno;
2270
2271         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
2272                                &bh, OCFS2_BH_CACHED, inode);
2273         if (ret) {
2274                 mlog_errno(ret);
2275                 goto out;
2276         }
2277
2278         xh = (struct ocfs2_xattr_header *)bh->b_data;
2279         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2280
2281         while (low_bucket <= high_bucket) {
2282                 brelse(bh);
2283                 bh = NULL;
2284                 bucket = (low_bucket + high_bucket) / 2;
2285
2286                 blkno = p_blkno + bucket * blk_per_bucket;
2287
2288                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
2289                                        &bh, OCFS2_BH_CACHED, inode);
2290                 if (ret) {
2291                         mlog_errno(ret);
2292                         goto out;
2293                 }
2294
2295                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2296                 xe = &xh->xh_entries[0];
2297                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2298                         high_bucket = bucket - 1;
2299                         continue;
2300                 }
2301
2302                 /*
2303                  * Check whether the hash of the last entry in our
2304                  * bucket is larger than the search one.
2305                  */
2306                 xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2307                 last_hash = le32_to_cpu(xe->xe_name_hash);
2308
2309                 /* record lower_bh which may be the insert place. */
2310                 brelse(lower_bh);
2311                 lower_bh = bh;
2312                 bh = NULL;
2313
2314                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2315                         low_bucket = bucket + 1;
2316                         continue;
2317                 }
2318
2319                 /* the searched xattr should reside in this bucket if exists. */
2320                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2321                                               name_index, name, name_hash,
2322                                               &index, &found);
2323                 if (ret) {
2324                         mlog_errno(ret);
2325                         goto out;
2326                 }
2327                 break;
2328         }
2329
2330         /*
2331          * Record the bucket we have found.
2332          * When the xattr's hash value is in the gap of 2 buckets, we will
2333          * always set it to the previous bucket.
2334          */
2335         if (!lower_bh) {
2336                 /*
2337                  * We can't find any bucket whose first name_hash is less
2338                  * than the find name_hash.
2339                  */
2340                 BUG_ON(bh->b_blocknr != p_blkno);
2341                 lower_bh = bh;
2342                 bh = NULL;
2343         }
2344         xs->bucket.bhs[0] = lower_bh;
2345         xs->bucket.xh = (struct ocfs2_xattr_header *)
2346                                         xs->bucket.bhs[0]->b_data;
2347         lower_bh = NULL;
2348
2349         xs->header = xs->bucket.xh;
2350         xs->base = xs->bucket.bhs[0]->b_data;
2351         xs->end = xs->base + inode->i_sb->s_blocksize;
2352
2353         if (found) {
2354                 /*
2355                  * If we have found the xattr enty, read all the blocks in
2356                  * this bucket.
2357                  */
2358                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2359                                         xs->bucket.bhs[0]->b_blocknr + 1,
2360                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2361                                         OCFS2_BH_CACHED, inode);
2362                 if (ret) {
2363                         mlog_errno(ret);
2364                         goto out;
2365                 }
2366
2367                 xs->here = &xs->header->xh_entries[index];
2368                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2369                      (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2370         } else
2371                 ret = -ENODATA;
2372
2373 out:
2374         brelse(bh);
2375         brelse(lower_bh);
2376         return ret;
2377 }
2378
2379 static int ocfs2_xattr_index_block_find(struct inode *inode,
2380                                         struct buffer_head *root_bh,
2381                                         int name_index,
2382                                         const char *name,
2383                                         struct ocfs2_xattr_search *xs)
2384 {
2385         int ret;
2386         struct ocfs2_xattr_block *xb =
2387                         (struct ocfs2_xattr_block *)root_bh->b_data;
2388         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2389         struct ocfs2_extent_list *el = &xb_root->xt_list;
2390         u64 p_blkno = 0;
2391         u32 first_hash, num_clusters = 0;
2392         u32 name_hash = ocfs2_xattr_hash_by_name(inode, name_index, name);
2393
2394         if (le16_to_cpu(el->l_next_free_rec) == 0)
2395                 return -ENODATA;
2396
2397         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2398              name, name_hash, name_index);
2399
2400         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2401                                   &num_clusters, el);
2402         if (ret) {
2403                 mlog_errno(ret);
2404                 goto out;
2405         }
2406
2407         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2408
2409         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2410              "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2411
2412         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2413                                       p_blkno, first_hash, num_clusters, xs);
2414
2415 out:
2416         return ret;
2417 }
2418
2419 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2420                                        u64 blkno,
2421                                        u32 clusters,
2422                                        xattr_bucket_func *func,
2423                                        void *para)
2424 {
2425         int i, j, ret = 0;
2426         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2427         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2428         u32 num_buckets = clusters * bpc;
2429         struct ocfs2_xattr_bucket bucket;
2430
2431         memset(&bucket, 0, sizeof(bucket));
2432
2433         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2434              clusters, blkno);
2435
2436         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2437                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2438                                         blkno, blk_per_bucket,
2439                                         bucket.bhs, OCFS2_BH_CACHED, inode);
2440                 if (ret) {
2441                         mlog_errno(ret);
2442                         goto out;
2443                 }
2444
2445                 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2446                 /*
2447                  * The real bucket num in this series of blocks is stored
2448                  * in the 1st bucket.
2449                  */
2450                 if (i == 0)
2451                         num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2452
2453                 mlog(0, "iterating xattr bucket %llu\n", blkno);
2454                 if (func) {
2455                         ret = func(inode, &bucket, para);
2456                         if (ret) {
2457                                 mlog_errno(ret);
2458                                 break;
2459                         }
2460                 }
2461
2462                 for (j = 0; j < blk_per_bucket; j++)
2463                         brelse(bucket.bhs[j]);
2464                 memset(&bucket, 0, sizeof(bucket));
2465         }
2466
2467 out:
2468         for (j = 0; j < blk_per_bucket; j++)
2469                 brelse(bucket.bhs[j]);
2470
2471         return ret;
2472 }
2473
2474 struct ocfs2_xattr_tree_list {
2475         char *buffer;
2476         size_t buffer_size;
2477 };
2478
2479 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2480                                              struct ocfs2_xattr_header *xh,
2481                                              int index,
2482                                              int *block_off,
2483                                              int *new_offset)
2484 {
2485         u16 name_offset;
2486
2487         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2488                 return -EINVAL;
2489
2490         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2491
2492         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2493         *new_offset = name_offset % inode->i_sb->s_blocksize;
2494
2495         return 0;
2496 }
2497
2498 static int ocfs2_list_xattr_bucket(struct inode *inode,
2499                                    struct ocfs2_xattr_bucket *bucket,
2500                                    void *para)
2501 {
2502         int ret = 0;
2503         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2504         size_t size;
2505         int i, block_off, new_offset;
2506
2507         for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2508                 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2509                 struct xattr_handler *handler =
2510                         ocfs2_xattr_handler(ocfs2_xattr_get_type(entry));
2511
2512                 if (handler) {
2513                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2514                                                                 bucket->xh,
2515                                                                 i,
2516                                                                 &block_off,
2517                                                                 &new_offset);
2518                         if (ret)
2519                                 break;
2520                         size = handler->list(inode, xl->buffer, xl->buffer_size,
2521                                              bucket->bhs[block_off]->b_data +
2522                                              new_offset,
2523                                              entry->xe_name_len);
2524                         if (xl->buffer) {
2525                                 if (size > xl->buffer_size)
2526                                         return -ERANGE;
2527                                 xl->buffer += size;
2528                         }
2529                         xl->buffer_size -= size;
2530                 }
2531         }
2532
2533         return ret;
2534 }
2535
2536 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2537                                              struct ocfs2_xattr_tree_root *xt,
2538                                              char *buffer,
2539                                              size_t buffer_size)
2540 {
2541         struct ocfs2_extent_list *el = &xt->xt_list;
2542         int ret = 0;
2543         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2544         u64 p_blkno = 0;
2545         struct ocfs2_xattr_tree_list xl = {
2546                 .buffer = buffer,
2547                 .buffer_size = buffer_size,
2548         };
2549
2550         if (le16_to_cpu(el->l_next_free_rec) == 0)
2551                 return 0;
2552
2553         while (name_hash > 0) {
2554                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2555                                           &e_cpos, &num_clusters, el);
2556                 if (ret) {
2557                         mlog_errno(ret);
2558                         goto out;
2559                 }
2560
2561                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2562                                                   ocfs2_list_xattr_bucket,
2563                                                   &xl);
2564                 if (ret) {
2565                         mlog_errno(ret);
2566                         goto out;
2567                 }
2568
2569                 if (e_cpos == 0)
2570                         break;
2571
2572                 name_hash = e_cpos - 1;
2573         }
2574
2575         ret = buffer_size - xl.buffer_size;
2576 out:
2577         return ret;
2578 }
2579
2580 static int cmp_xe(const void *a, const void *b)
2581 {
2582         const struct ocfs2_xattr_entry *l = a, *r = b;
2583         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2584         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2585
2586         if (l_hash > r_hash)
2587                 return 1;
2588         if (l_hash < r_hash)
2589                 return -1;
2590         return 0;
2591 }
2592
2593 static void swap_xe(void *a, void *b, int size)
2594 {
2595         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2596
2597         tmp = *l;
2598         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2599         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2600 }
2601
2602 /*
2603  * When the ocfs2_xattr_block is filled up, new bucket will be created
2604  * and all the xattr entries will be moved to the new bucket.
2605  * Note: we need to sort the entries since they are not saved in order
2606  * in the ocfs2_xattr_block.
2607  */
2608 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2609                                            struct buffer_head *xb_bh,
2610                                            struct buffer_head *xh_bh,
2611                                            struct buffer_head *data_bh)
2612 {
2613         int i, blocksize = inode->i_sb->s_blocksize;
2614         u16 offset, size, off_change;
2615         struct ocfs2_xattr_entry *xe;
2616         struct ocfs2_xattr_block *xb =
2617                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2618         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2619         struct ocfs2_xattr_header *xh =
2620                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2621         u16 count = le16_to_cpu(xb_xh->xh_count);
2622         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2623
2624         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2625              (unsigned long long)xb_bh->b_blocknr,
2626              (unsigned long long)xh_bh->b_blocknr);
2627
2628         memset(xh_bh->b_data, 0, blocksize);
2629         if (data_bh)
2630                 memset(data_bh->b_data, 0, blocksize);
2631         /*
2632          * Since the xe_name_offset is based on ocfs2_xattr_header,
2633          * there is a offset change corresponding to the change of
2634          * ocfs2_xattr_header's position.
2635          */
2636         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2637         xe = &xb_xh->xh_entries[count - 1];
2638         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2639         size = blocksize - offset;
2640
2641         /* copy all the names and values. */
2642         if (data_bh)
2643                 target = data_bh->b_data;
2644         memcpy(target + offset, src + offset, size);
2645
2646         /* Init new header now. */
2647         xh->xh_count = xb_xh->xh_count;
2648         xh->xh_num_buckets = cpu_to_le16(1);
2649         xh->xh_name_value_len = cpu_to_le16(size);
2650         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2651
2652         /* copy all the entries. */
2653         target = xh_bh->b_data;
2654         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2655         size = count * sizeof(struct ocfs2_xattr_entry);
2656         memcpy(target + offset, (char *)xb_xh + offset, size);
2657
2658         /* Change the xe offset for all the xe because of the move. */
2659         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2660                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2661         for (i = 0; i < count; i++)
2662                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2663
2664         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2665              offset, size, off_change);
2666
2667         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2668              cmp_xe, swap_xe);
2669 }
2670
2671 /*
2672  * After we move xattr from block to index btree, we have to
2673  * update ocfs2_xattr_search to the new xe and base.
2674  *
2675  * When the entry is in xattr block, xattr_bh indicates the storage place.
2676  * While if the entry is in index b-tree, "bucket" indicates the
2677  * real place of the xattr.
2678  */
2679 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2680                                            struct ocfs2_xattr_search *xs,
2681                                            struct buffer_head *old_bh,
2682                                            struct buffer_head *new_bh)
2683 {
2684         int ret = 0;
2685         char *buf = old_bh->b_data;
2686         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2687         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2688         int i, blocksize = inode->i_sb->s_blocksize;
2689         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2690
2691         xs->bucket.bhs[0] = new_bh;
2692         get_bh(new_bh);
2693         xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2694         xs->header = xs->bucket.xh;
2695
2696         xs->base = new_bh->b_data;
2697         xs->end = xs->base + inode->i_sb->s_blocksize;
2698
2699         if (!xs->not_found) {
2700                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2701                         ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2702                                         xs->bucket.bhs[0]->b_blocknr + 1,
2703                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2704                                         OCFS2_BH_CACHED, inode);
2705                         if (ret) {
2706                                 mlog_errno(ret);
2707                                 return ret;
2708                         }
2709
2710                         i = xs->here - old_xh->xh_entries;
2711                         xs->here = &xs->header->xh_entries[i];
2712                 }
2713         }
2714
2715         return ret;
2716 }
2717
2718 static int ocfs2_xattr_create_index_block(struct inode *inode,
2719                                           struct ocfs2_xattr_search *xs)
2720 {
2721         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2722         u32 bit_off, len;
2723         u64 blkno;
2724         handle_t *handle;
2725         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2726         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2727         struct ocfs2_alloc_context *data_ac;
2728         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2729         struct buffer_head *xb_bh = xs->xattr_bh;
2730         struct ocfs2_xattr_block *xb =
2731                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2732         struct ocfs2_xattr_tree_root *xr;
2733         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2734         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2735
2736         mlog(0, "create xattr index block for %llu\n",
2737              (unsigned long long)xb_bh->b_blocknr);
2738
2739         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2740
2741         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2742         if (ret) {
2743                 mlog_errno(ret);
2744                 goto out;
2745         }
2746
2747         /*
2748          * XXX:
2749          * We can use this lock for now, and maybe move to a dedicated mutex
2750          * if performance becomes a problem later.
2751          */
2752         down_write(&oi->ip_alloc_sem);
2753
2754         /*
2755          * 3 more credits, one for xattr block update, one for the 1st block
2756          * of the new xattr bucket and one for the value/data.
2757          */
2758         credits += 3;
2759         handle = ocfs2_start_trans(osb, credits);
2760         if (IS_ERR(handle)) {
2761                 ret = PTR_ERR(handle);
2762                 mlog_errno(ret);
2763                 goto out_sem;
2764         }
2765
2766         ret = ocfs2_journal_access(handle, inode, xb_bh,
2767                                    OCFS2_JOURNAL_ACCESS_WRITE);
2768         if (ret) {
2769                 mlog_errno(ret);
2770                 goto out_commit;
2771         }
2772
2773         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2774         if (ret) {
2775                 mlog_errno(ret);
2776                 goto out_commit;
2777         }
2778
2779         /*
2780          * The bucket may spread in many blocks, and
2781          * we will only touch the 1st block and the last block
2782          * in the whole bucket(one for entry and one for data).
2783          */
2784         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2785
2786         mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2787
2788         xh_bh = sb_getblk(inode->i_sb, blkno);
2789         if (!xh_bh) {
2790                 ret = -EIO;
2791                 mlog_errno(ret);
2792                 goto out_commit;
2793         }
2794
2795         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2796
2797         ret = ocfs2_journal_access(handle, inode, xh_bh,
2798                                    OCFS2_JOURNAL_ACCESS_CREATE);
2799         if (ret) {
2800                 mlog_errno(ret);
2801                 goto out_commit;
2802         }
2803
2804         if (bpb > 1) {
2805                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2806                 if (!data_bh) {
2807                         ret = -EIO;
2808                         mlog_errno(ret);
2809                         goto out_commit;
2810                 }
2811
2812                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2813
2814                 ret = ocfs2_journal_access(handle, inode, data_bh,
2815                                            OCFS2_JOURNAL_ACCESS_CREATE);
2816                 if (ret) {
2817                         mlog_errno(ret);
2818                         goto out_commit;
2819                 }
2820         }
2821
2822         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2823
2824         ocfs2_journal_dirty(handle, xh_bh);
2825         if (data_bh)
2826                 ocfs2_journal_dirty(handle, data_bh);
2827
2828         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2829
2830         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2831         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2832                offsetof(struct ocfs2_xattr_block, xb_attrs));
2833
2834         xr = &xb->xb_attrs.xb_root;
2835         xr->xt_clusters = cpu_to_le32(1);
2836         xr->xt_last_eb_blk = 0;
2837         xr->xt_list.l_tree_depth = 0;
2838         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2839         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2840
2841         xr->xt_list.l_recs[0].e_cpos = 0;
2842         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2843         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2844
2845         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2846
2847         ret = ocfs2_journal_dirty(handle, xb_bh);
2848         if (ret) {
2849                 mlog_errno(ret);
2850                 goto out_commit;
2851         }
2852
2853 out_commit:
2854         ocfs2_commit_trans(osb, handle);
2855
2856 out_sem:
2857         up_write(&oi->ip_alloc_sem);
2858
2859 out:
2860         if (data_ac)
2861                 ocfs2_free_alloc_context(data_ac);
2862
2863         brelse(xh_bh);
2864         brelse(data_bh);
2865
2866         return ret;
2867 }
2868
2869 static int cmp_xe_offset(const void *a, const void *b)
2870 {
2871         const struct ocfs2_xattr_entry *l = a, *r = b;
2872         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2873         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2874
2875         if (l_name_offset < r_name_offset)
2876                 return 1;
2877         if (l_name_offset > r_name_offset)
2878                 return -1;
2879         return 0;
2880 }
2881
2882 /*
2883  * defrag a xattr bucket if we find that the bucket has some
2884  * holes beteen name/value pairs.
2885  * We will move all the name/value pairs to the end of the bucket
2886  * so that we can spare some space for insertion.
2887  */
2888 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2889                                      struct ocfs2_xattr_bucket *bucket)
2890 {
2891         int ret, i;
2892         size_t end, offset, len, value_len;
2893         struct ocfs2_xattr_header *xh;
2894         char *entries, *buf, *bucket_buf = NULL;
2895         u64 blkno = bucket->bhs[0]->b_blocknr;
2896         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2897         u16 xh_free_start;
2898         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2899         size_t blocksize = inode->i_sb->s_blocksize;
2900         handle_t *handle;
2901         struct buffer_head **bhs;
2902         struct ocfs2_xattr_entry *xe;
2903
2904         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2905                         GFP_NOFS);
2906         if (!bhs)
2907                 return -ENOMEM;
2908
2909         ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs,
2910                                 OCFS2_BH_CACHED, inode);
2911         if (ret)
2912                 goto out;
2913
2914         /*
2915          * In order to make the operation more efficient and generic,
2916          * we copy all the blocks into a contiguous memory and do the
2917          * defragment there, so if anything is error, we will not touch
2918          * the real block.
2919          */
2920         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2921         if (!bucket_buf) {
2922                 ret = -EIO;
2923                 goto out;
2924         }
2925
2926         buf = bucket_buf;
2927         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2928                 memcpy(buf, bhs[i]->b_data, blocksize);
2929
2930         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2931         if (IS_ERR(handle)) {
2932                 ret = PTR_ERR(handle);
2933                 handle = NULL;
2934                 mlog_errno(ret);
2935                 goto out;
2936         }
2937
2938         for (i = 0; i < blk_per_bucket; i++) {
2939                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2940                                            OCFS2_JOURNAL_ACCESS_WRITE);
2941                 if (ret < 0) {
2942                         mlog_errno(ret);
2943                         goto commit;
2944                 }
2945         }
2946
2947         xh = (struct ocfs2_xattr_header *)bucket_buf;
2948         entries = (char *)xh->xh_entries;
2949         xh_free_start = le16_to_cpu(xh->xh_free_start);
2950
2951         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2952              "xh_free_start = %u, xh_name_value_len = %u.\n",
2953              blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2954              le16_to_cpu(xh->xh_name_value_len));
2955
2956         /*
2957          * sort all the entries by their offset.
2958          * the largest will be the first, so that we can
2959          * move them to the end one by one.
2960          */
2961         sort(entries, le16_to_cpu(xh->xh_count),
2962              sizeof(struct ocfs2_xattr_entry),
2963              cmp_xe_offset, swap_xe);
2964
2965         /* Move all name/values to the end of the bucket. */
2966         xe = xh->xh_entries;
2967         end = OCFS2_XATTR_BUCKET_SIZE;
2968         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2969                 offset = le16_to_cpu(xe->xe_name_offset);
2970                 if (ocfs2_xattr_is_local(xe))
2971                         value_len = OCFS2_XATTR_SIZE(
2972                                         le64_to_cpu(xe->xe_value_size));
2973                 else
2974                         value_len = OCFS2_XATTR_ROOT_SIZE;
2975                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2976
2977                 /*
2978                  * We must make sure that the name/value pair
2979                  * exist in the same block. So adjust end to
2980                  * the previous block end if needed.
2981                  */
2982                 if (((end - len) / blocksize !=
2983                         (end - 1) / blocksize))
2984                         end = end - end % blocksize;
2985
2986                 if (end > offset + len) {
2987                         memmove(bucket_buf + end - len,
2988                                 bucket_buf + offset, len);
2989                         xe->xe_name_offset = cpu_to_le16(end - len);
2990                 }
2991
2992                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
2993                                 "bucket %llu\n", (unsigned long long)blkno);
2994
2995                 end -= len;
2996         }
2997
2998         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
2999                         "bucket %llu\n", (unsigned long long)blkno);
3000
3001         if (xh_free_start == end)
3002                 goto commit;
3003
3004         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3005         xh->xh_free_start = cpu_to_le16(end);
3006
3007         /* sort the entries by their name_hash. */
3008         sort(entries, le16_to_cpu(xh->xh_count),
3009              sizeof(struct ocfs2_xattr_entry),
3010              cmp_xe, swap_xe);
3011
3012         buf = bucket_buf;
3013         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3014                 memcpy(bhs[i]->b_data, buf, blocksize);
3015                 ocfs2_journal_dirty(handle, bhs[i]);
3016         }
3017
3018 commit:
3019         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3020 out:
3021
3022         if (bhs) {
3023                 for (i = 0; i < blk_per_bucket; i++)
3024                         brelse(bhs[i]);
3025         }
3026         kfree(bhs);
3027
3028         kfree(bucket_buf);
3029         return ret;
3030 }
3031
3032 /*
3033  * Move half nums of the xattr bucket in the previous cluster to this new
3034  * cluster. We only touch the last cluster of the previous extend record.
3035  *
3036  * first_bh is the first buffer_head of a series of bucket in the same
3037  * extent rec and header_bh is the header of one bucket in this cluster.
3038  * They will be updated if we move the data header_bh contains to the new
3039  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3040  */
3041 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3042                                                handle_t *handle,
3043                                                struct buffer_head **first_bh,
3044                                                struct buffer_head **header_bh,
3045                                                u64 new_blkno,
3046                                                u64 prev_blkno,
3047                                                u32 num_clusters,
3048                                                u32 *first_hash)
3049 {
3050         int i, ret, credits;
3051         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3052         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3053         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3054         int blocksize = inode->i_sb->s_blocksize;
3055         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3056         struct ocfs2_xattr_header *new_xh;
3057         struct ocfs2_xattr_header *xh =
3058                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3059
3060         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3061         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3062
3063         prev_bh = *first_bh;
3064         get_bh(prev_bh);
3065         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3066
3067         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3068
3069         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3070              prev_blkno, new_blkno);
3071
3072         /*
3073          * We need to update the 1st half of the new cluster and
3074          * 1 more for the update of the 1st bucket of the previous
3075          * extent record.
3076          */
3077         credits = bpc / 2 + 1;
3078         ret = ocfs2_extend_trans(handle, credits);
3079         if (ret) {
3080                 mlog_errno(ret);
3081                 goto out;
3082         }
3083
3084         ret = ocfs2_journal_access(handle, inode, prev_bh,
3085                                    OCFS2_JOURNAL_ACCESS_WRITE);
3086         if (ret) {
3087                 mlog_errno(ret);
3088                 goto out;
3089         }
3090
3091         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3092                 old_bh = new_bh = NULL;
3093                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3094                 if (!new_bh) {
3095                         ret = -EIO;
3096                         mlog_errno(ret);
3097                         goto out;
3098                 }
3099
3100                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3101
3102                 ret = ocfs2_journal_access(handle, inode, new_bh,
3103                                            OCFS2_JOURNAL_ACCESS_CREATE);
3104                 if (ret < 0) {
3105                         mlog_errno(ret);
3106                         brelse(new_bh);
3107                         goto out;
3108                 }
3109
3110                 ret = ocfs2_read_block(osb, prev_blkno,
3111                                        &old_bh, OCFS2_BH_CACHED, inode);
3112                 if (ret < 0) {
3113                         mlog_errno(ret);
3114                         brelse(new_bh);
3115                         goto out;
3116                 }
3117
3118                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3119
3120                 if (i == 0) {
3121                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3122                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3123
3124                         if (first_hash)
3125                                 *first_hash = le32_to_cpu(
3126                                         new_xh->xh_entries[0].xe_name_hash);
3127                         new_first_bh = new_bh;
3128                         get_bh(new_first_bh);
3129                 }
3130
3131                 ocfs2_journal_dirty(handle, new_bh);
3132
3133                 if (*header_bh == old_bh) {
3134                         brelse(*header_bh);
3135                         *header_bh = new_bh;
3136                         get_bh(*header_bh);
3137
3138                         brelse(*first_bh);
3139                         *first_bh = new_first_bh;
3140                         get_bh(*first_bh);
3141                 }
3142                 brelse(new_bh);
3143                 brelse(old_bh);
3144         }
3145
3146         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3147
3148         ocfs2_journal_dirty(handle, prev_bh);
3149 out:
3150         brelse(prev_bh);
3151         brelse(new_first_bh);
3152         return ret;
3153 }
3154
3155 static int ocfs2_read_xattr_bucket(struct inode *inode,
3156                                    u64 blkno,
3157                                    struct buffer_head **bhs,
3158                                    int new)
3159 {
3160         int ret = 0;
3161         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3162
3163         if (!new)
3164                 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno,
3165                                          blk_per_bucket, bhs,
3166                                          OCFS2_BH_CACHED, inode);
3167
3168         for (i = 0; i < blk_per_bucket; i++) {
3169                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3170                 if (bhs[i] == NULL) {
3171                         ret = -EIO;
3172                         mlog_errno(ret);
3173                         break;
3174                 }
3175                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3176         }
3177
3178         return ret;
3179 }
3180
3181 /*
3182  * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3183  * first_hash will record the 1st hash of the new bucket.
3184  */
3185 static int ocfs2_half_xattr_bucket(struct inode *inode,
3186                                    handle_t *handle,
3187                                    u64 blk,
3188                                    u64 new_blk,
3189                                    u32 *first_hash,
3190                                    int new_bucket_head)
3191 {
3192         int ret, i;
3193         u16 count, start, len, name_value_len, xe_len, name_offset;
3194         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3195         struct buffer_head **s_bhs, **t_bhs = NULL;
3196         struct ocfs2_xattr_header *xh;
3197         struct ocfs2_xattr_entry *xe;
3198         int blocksize = inode->i_sb->s_blocksize;
3199
3200         mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3201              blk, new_blk);
3202
3203         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3204         if (!s_bhs)
3205                 return -ENOMEM;
3206
3207         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3208         if (ret) {
3209                 mlog_errno(ret);
3210                 goto out;
3211         }
3212
3213         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3214                                    OCFS2_JOURNAL_ACCESS_WRITE);
3215         if (ret) {
3216                 mlog_errno(ret);
3217                 goto out;
3218         }
3219
3220         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3221         if (!t_bhs) {
3222                 ret = -ENOMEM;
3223                 goto out;
3224         }
3225
3226         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3227         if (ret) {
3228                 mlog_errno(ret);
3229                 goto out;
3230         }
3231
3232         for (i = 0; i < blk_per_bucket; i++) {
3233                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3234                                            OCFS2_JOURNAL_ACCESS_CREATE);
3235                 if (ret) {
3236                         mlog_errno(ret);
3237                         goto out;
3238                 }
3239         }
3240
3241         /* copy the whole bucket to the new first. */
3242         for (i = 0; i < blk_per_bucket; i++)
3243                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3244
3245         /* update the new bucket. */
3246         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3247         count = le16_to_cpu(xh->xh_count);
3248         start = count / 2;
3249
3250         /*
3251          * Calculate the total name/value len and xh_free_start for
3252          * the old bucket first.
3253          */
3254         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3255         name_value_len = 0;
3256         for (i = 0; i < start; i++) {
3257                 xe = &xh->xh_entries[i];
3258                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3259                 if (ocfs2_xattr_is_local(xe))
3260                         xe_len +=
3261                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3262                 else
3263                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3264                 name_value_len += xe_len;
3265                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3266                         name_offset = le16_to_cpu(xe->xe_name_offset);
3267         }
3268
3269         /*
3270          * Now begin the modification to the new bucket.
3271          *
3272          * In the new bucket, We just move the xattr entry to the beginning
3273          * and don't touch the name/value. So there will be some holes in the
3274          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3275          * called.
3276          */
3277         xe = &xh->xh_entries[start];
3278         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3279         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3280              (int)((char *)xe - (char *)xh),
3281              (int)((char *)xh->xh_entries - (char *)xh));
3282         memmove((char *)xh->xh_entries, (char *)xe, len);
3283         xe = &xh->xh_entries[count - start];
3284         len = sizeof(struct ocfs2_xattr_entry) * start;
3285         memset((char *)xe, 0, len);
3286
3287         le16_add_cpu(&xh->xh_count, -start);
3288         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3289
3290         /* Calculate xh_free_start for the new bucket. */
3291         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3292         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3293                 xe = &xh->xh_entries[i];
3294                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3295                 if (ocfs2_xattr_is_local(xe))
3296                         xe_len +=
3297                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3298                 else
3299                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3300                 if (le16_to_cpu(xe->xe_name_offset) <
3301                     le16_to_cpu(xh->xh_free_start))
3302                         xh->xh_free_start = xe->xe_name_offset;
3303         }
3304
3305         /* set xh->xh_num_buckets for the new xh. */
3306         if (new_bucket_head)
3307                 xh->xh_num_buckets = cpu_to_le16(1);
3308         else
3309                 xh->xh_num_buckets = 0;
3310
3311         for (i = 0; i < blk_per_bucket; i++) {
3312                 ocfs2_journal_dirty(handle, t_bhs[i]);
3313                 if (ret)
3314                         mlog_errno(ret);
3315         }
3316
3317         /* store the first_hash of the new bucket. */
3318         if (first_hash)
3319                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3320
3321         /*
3322          * Now only update the 1st block of the old bucket.
3323          * Please note that the entry has been sorted already above.
3324          */
3325         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3326         memset(&xh->xh_entries[start], 0,
3327                sizeof(struct ocfs2_xattr_entry) * (count - start));
3328         xh->xh_count = cpu_to_le16(start);
3329         xh->xh_free_start = cpu_to_le16(name_offset);
3330         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3331
3332         ocfs2_journal_dirty(handle, s_bhs[0]);
3333         if (ret)
3334                 mlog_errno(ret);
3335
3336 out:
3337         if (s_bhs) {
3338                 for (i = 0; i < blk_per_bucket; i++)
3339                         brelse(s_bhs[i]);
3340         }
3341         kfree(s_bhs);
3342
3343         if (t_bhs) {
3344                 for (i = 0; i < blk_per_bucket; i++)
3345                         brelse(t_bhs[i]);
3346         }
3347         kfree(t_bhs);
3348
3349         return ret;
3350 }
3351
3352 /*
3353  * Copy xattr from one bucket to another bucket.
3354  *
3355  * The caller must make sure that the journal transaction
3356  * has enough space for journaling.
3357  */
3358 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3359                                  handle_t *handle,
3360                                  u64 s_blkno,
3361                                  u64 t_blkno,
3362                                  int t_is_new)
3363 {
3364         int ret, i;
3365         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3366         int blocksize = inode->i_sb->s_blocksize;
3367         struct buffer_head **s_bhs, **t_bhs = NULL;
3368
3369         BUG_ON(s_blkno == t_blkno);
3370
3371         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3372              s_blkno, t_blkno, t_is_new);
3373
3374         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3375                         GFP_NOFS);
3376         if (!s_bhs)
3377                 return -ENOMEM;
3378
3379         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3380         if (ret)
3381                 goto out;
3382
3383         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3384                         GFP_NOFS);
3385         if (!t_bhs) {
3386                 ret = -ENOMEM;
3387                 goto out;
3388         }
3389
3390         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3391         if (ret)
3392                 goto out;
3393
3394         for (i = 0; i < blk_per_bucket; i++) {
3395                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3396                                            OCFS2_JOURNAL_ACCESS_WRITE);
3397                 if (ret)
3398                         goto out;
3399         }
3400
3401         for (i = 0; i < blk_per_bucket; i++) {
3402                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3403                 ocfs2_journal_dirty(handle, t_bhs[i]);
3404         }
3405
3406 out:
3407         if (s_bhs) {
3408                 for (i = 0; i < blk_per_bucket; i++)
3409                         brelse(s_bhs[i]);
3410         }
3411         kfree(s_bhs);
3412
3413         if (t_bhs) {
3414                 for (i = 0; i < blk_per_bucket; i++)
3415                         brelse(t_bhs[i]);
3416         }
3417         kfree(t_bhs);
3418
3419         return ret;
3420 }
3421
3422 /*
3423  * Copy one xattr cluster from src_blk to to_blk.
3424  * The to_blk will become the first bucket header of the cluster, so its
3425  * xh_num_buckets will be initialized as the bucket num in the cluster.
3426  */
3427 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3428                                   handle_t *handle,
3429                                   struct buffer_head *first_bh,
3430                                   u64 src_blk,
3431                                   u64 to_blk,
3432                                   u32 *first_hash)
3433 {
3434         int i, ret, credits;
3435         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3436         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3437         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3438         struct buffer_head *bh = NULL;
3439         struct ocfs2_xattr_header *xh;
3440         u64 to_blk_start = to_blk;
3441
3442         mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3443
3444         /*
3445          * We need to update the new cluster and 1 more for the update of
3446          * the 1st bucket of the previous extent rec.
3447          */
3448         credits = bpc + 1;
3449         ret = ocfs2_extend_trans(handle, credits);
3450         if (ret) {
3451                 mlog_errno(ret);
3452                 goto out;
3453         }
3454
3455         ret = ocfs2_journal_access(handle, inode, first_bh,
3456                                    OCFS2_JOURNAL_ACCESS_WRITE);
3457         if (ret) {
3458                 mlog_errno(ret);
3459                 goto out;
3460         }
3461
3462         for (i = 0; i < num_buckets; i++) {
3463                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3464                                             src_blk, to_blk, 1);
3465                 if (ret) {
3466                         mlog_errno(ret);
3467                         goto out;
3468                 }
3469
3470                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3471                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3472         }
3473
3474         /* update the old bucket header. */
3475         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3476         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3477
3478         ocfs2_journal_dirty(handle, first_bh);
3479
3480         /* update the new bucket header. */
3481         ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode);
3482         if (ret < 0) {
3483                 mlog_errno(ret);
3484                 goto out;
3485         }
3486
3487         ret = ocfs2_journal_access(handle, inode, bh,
3488                                    OCFS2_JOURNAL_ACCESS_WRITE);
3489         if (ret) {
3490                 mlog_errno(ret);
3491                 goto out;
3492         }
3493
3494         xh = (struct ocfs2_xattr_header *)bh->b_data;
3495         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3496
3497         ocfs2_journal_dirty(handle, bh);
3498
3499         if (first_hash)
3500                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3501 out:
3502         brelse(bh);
3503         return ret;
3504 }
3505
3506 /*
3507  * Move half of the xattrs in this cluster to the new cluster.
3508  * This function should only be called when bucket size == cluster size.
3509  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3510  */
3511 static int ocfs2_half_xattr_cluster(struct inode *inode,
3512                                     handle_t *handle,
3513                                     u64 prev_blk,
3514                                     u64 new_blk,
3515                                     u32 *first_hash)
3516 {
3517         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3518         int ret, credits = 2 * blk_per_bucket;
3519
3520         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3521
3522         ret = ocfs2_extend_trans(handle, credits);
3523         if (ret) {
3524                 mlog_errno(ret);
3525                 return ret;
3526         }
3527
3528         /* Move half of the xattr in start_blk to the next bucket. */
3529         return  ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3530                                         new_blk, first_hash, 1);
3531 }
3532
3533 /*
3534  * Move some xattrs from the old cluster to the new one since they are not
3535  * contiguous in ocfs2 xattr tree.
3536  *
3537  * new_blk starts a new separate cluster, and we will move some xattrs from
3538  * prev_blk to it. v_start will be set as the first name hash value in this
3539  * new cluster so that it can be used as e_cpos during tree insertion and
3540  * don't collide with our original b-tree operations. first_bh and header_bh
3541  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3542  * to extend the insert bucket.
3543  *
3544  * The problem is how much xattr should we move to the new one and when should
3545  * we update first_bh and header_bh?
3546  * 1. If cluster size > bucket size, that means the previous cluster has more
3547  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3548  *    update the first_bh and header_bh if the insert bucket has been moved
3549  *    to the new cluster.
3550  * 2. If cluster_size == bucket_size:
3551  *    a) If the previous extent rec has more than one cluster and the insert
3552  *       place isn't in the last cluster, copy the entire last cluster to the
3553  *       new one. This time, we don't need to upate the first_bh and header_bh
3554  *       since they will not be moved into the new cluster.
3555  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3556  *       the new one. And we set the extend flag to zero if the insert place is
3557  *       moved into the new allocated cluster since no extend is needed.
3558  */
3559 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3560                                             handle_t *handle,
3561                                             struct buffer_head **first_bh,
3562                                             struct buffer_head **header_bh,
3563                                             u64 new_blk,
3564                                             u64 prev_blk,
3565                                             u32 prev_clusters,
3566                                             u32 *v_start,
3567                                             int *extend)
3568 {
3569         int ret = 0;
3570         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3571
3572         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3573              prev_blk, prev_clusters, new_blk);
3574
3575         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3576                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3577                                                           handle,
3578                                                           first_bh,
3579                                                           header_bh,
3580                                                           new_blk,
3581                                                           prev_blk,
3582                                                           prev_clusters,
3583                                                           v_start);
3584         else {
3585                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3586
3587                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3588                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3589                                                      last_blk, new_blk,
3590                                                      v_start);
3591                 else {
3592                         ret = ocfs2_half_xattr_cluster(inode, handle,
3593                                                        last_blk, new_blk,
3594                                                        v_start);
3595
3596                         if ((*header_bh)->b_blocknr == last_blk && extend)
3597                                 *extend = 0;
3598                 }
3599         }
3600
3601         return ret;
3602 }
3603
3604 /*
3605  * Add a new cluster for xattr storage.
3606  *
3607  * If the new cluster is contiguous with the previous one, it will be
3608  * appended to the same extent record, and num_clusters will be updated.
3609  * If not, we will insert a new extent for it and move some xattrs in
3610  * the last cluster into the new allocated one.
3611  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3612  * lose the benefits of hashing because we'll have to search large leaves.
3613  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3614  * if it's bigger).
3615  *
3616  * first_bh is the first block of the previous extent rec and header_bh
3617  * indicates the bucket we will insert the new xattrs. They will be updated
3618  * when the header_bh is moved into the new cluster.
3619  */
3620 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3621                                        struct buffer_head *root_bh,
3622                                        struct buffer_head **first_bh,
3623                                        struct buffer_head **header_bh,
3624                                        u32 *num_clusters,
3625                                        u32 prev_cpos,
3626                                        u64 prev_blkno,
3627                                        int *extend)
3628 {
3629         int ret, credits;
3630         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3631         u32 prev_clusters = *num_clusters;
3632         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3633         u64 block;
3634         handle_t *handle = NULL;
3635         struct ocfs2_alloc_context *data_ac = NULL;
3636         struct ocfs2_alloc_context *meta_ac = NULL;
3637         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3638         struct ocfs2_extent_tree et;
3639
3640         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3641              "previous xattr blkno = %llu\n",
3642              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3643              prev_cpos, prev_blkno);
3644
3645         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3646
3647         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3648                                     &data_ac, &meta_ac);
3649         if (ret) {
3650                 mlog_errno(ret);
3651                 goto leave;
3652         }
3653
3654         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3655                                             clusters_to_add);
3656         handle = ocfs2_start_trans(osb, credits);
3657         if (IS_ERR(handle)) {
3658                 ret = PTR_ERR(handle);
3659                 handle = NULL;
3660                 mlog_errno(ret);
3661                 goto leave;
3662         }
3663
3664         ret = ocfs2_journal_access(handle, inode, root_bh,
3665                                    OCFS2_JOURNAL_ACCESS_WRITE);
3666         if (ret < 0) {
3667                 mlog_errno(ret);
3668                 goto leave;
3669         }
3670
3671         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3672                                      clusters_to_add, &bit_off, &num_bits);
3673         if (ret < 0) {
3674                 if (ret != -ENOSPC)
3675                         mlog_errno(ret);
3676                 goto leave;
3677         }
3678
3679         BUG_ON(num_bits > clusters_to_add);
3680
3681         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3682         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3683              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3684
3685         if (prev_blkno + prev_clusters * bpc == block &&
3686             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3687              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3688                 /*
3689                  * If this cluster is contiguous with the old one and
3690                  * adding this new cluster, we don't surpass the limit of
3691                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3692                  * initialized and used like other buckets in the previous
3693                  * cluster.
3694                  * So add it as a contiguous one. The caller will handle
3695                  * its init process.
3696                  */
3697                 v_start = prev_cpos + prev_clusters;
3698                 *num_clusters = prev_clusters + num_bits;
3699                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3700                      num_bits);
3701         } else {
3702                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3703                                                        handle,
3704                                                        first_bh,
3705                                                        header_bh,
3706                                                        block,
3707                                                        prev_blkno,
3708                                                        prev_clusters,
3709                                                        &v_start,
3710                                                        extend);
3711                 if (ret) {
3712                         mlog_errno(ret);
3713                         goto leave;
3714                 }
3715         }
3716
3717         if (handle->h_buffer_credits < credits) {
3718                 /*
3719                  * The journal has been restarted before, and don't
3720                  * have enough space for the insertion, so extend it
3721                  * here.
3722                  */
3723                 ret = ocfs2_extend_trans(handle, credits);
3724                 if (ret) {
3725                         mlog_errno(ret);
3726                         goto leave;
3727                 }
3728         }
3729         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3730              num_bits, block, v_start);
3731         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3732                                   num_bits, 0, meta_ac);
3733         if (ret < 0) {
3734                 mlog_errno(ret);
3735                 goto leave;
3736         }
3737
3738         ret = ocfs2_journal_dirty(handle, root_bh);
3739         if (ret < 0) {
3740                 mlog_errno(ret);
3741                 goto leave;
3742         }
3743
3744 leave:
3745         if (handle)
3746                 ocfs2_commit_trans(osb, handle);
3747         if (data_ac)
3748                 ocfs2_free_alloc_context(data_ac);
3749         if (meta_ac)
3750                 ocfs2_free_alloc_context(meta_ac);
3751
3752         return ret;
3753 }
3754
3755 /*
3756  * Extend a new xattr bucket and move xattrs to the end one by one until
3757  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3758  */
3759 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3760                                      struct buffer_head *first_bh,
3761                                      struct buffer_head *start_bh,
3762                                      u32 num_clusters)
3763 {
3764         int ret, credits;
3765         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3766         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3767         u64 start_blk = start_bh->b_blocknr, end_blk;
3768         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3769         handle_t *handle;
3770         struct ocfs2_xattr_header *first_xh =
3771                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3772         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3773
3774         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3775              "from %llu, len = %u\n", start_blk,
3776              (unsigned long long)first_bh->b_blocknr, num_clusters);
3777
3778         BUG_ON(bucket >= num_buckets);
3779
3780         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3781
3782         /*
3783          * We will touch all the buckets after the start_bh(include it).
3784          * Add one more bucket and modify the first_bh.
3785          */
3786         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3787         handle = ocfs2_start_trans(osb, credits);
3788         if (IS_ERR(handle)) {
3789                 ret = PTR_ERR(handle);
3790                 handle = NULL;
3791                 mlog_errno(ret);
3792                 goto out;
3793         }
3794
3795         ret = ocfs2_journal_access(handle, inode, first_bh,
3796                                    OCFS2_JOURNAL_ACCESS_WRITE);
3797         if (ret) {
3798                 mlog_errno(ret);
3799                 goto commit;
3800         }
3801
3802         while (end_blk != start_blk) {
3803                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3804                                             end_blk + blk_per_bucket, 0);
3805                 if (ret)
3806                         goto commit;
3807                 end_blk -= blk_per_bucket;
3808         }
3809
3810         /* Move half of the xattr in start_blk to the next bucket. */
3811         ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3812                                       start_blk + blk_per_bucket, NULL, 0);
3813
3814         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3815         ocfs2_journal_dirty(handle, first_bh);
3816
3817 commit:
3818         ocfs2_commit_trans(osb, handle);
3819 out:
3820         return ret;
3821 }
3822
3823 /*
3824  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3825  * xb_bh is the ocfs2_xattr_block.
3826  * We will move all the buckets starting from header_bh to the next place. As
3827  * for this one, half num of its xattrs will be moved to the next one.
3828  *
3829  * We will allocate a new cluster if current cluster is full and adjust
3830  * header_bh and first_bh if the insert place is moved to the new cluster.
3831  */
3832 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3833                                       struct buffer_head *xb_bh,
3834                                       struct buffer_head *header_bh)
3835 {
3836         struct ocfs2_xattr_header *first_xh = NULL;
3837         struct buffer_head *first_bh = NULL;
3838         struct ocfs2_xattr_block *xb =
3839                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3840         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3841         struct ocfs2_extent_list *el = &xb_root->xt_list;
3842         struct ocfs2_xattr_header *xh =
3843                         (struct ocfs2_xattr_header *)header_bh->b_data;
3844         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3845         struct super_block *sb = inode->i_sb;
3846         struct ocfs2_super *osb = OCFS2_SB(sb);
3847         int ret, num_buckets, extend = 1;
3848         u64 p_blkno;
3849         u32 e_cpos, num_clusters;
3850
3851         mlog(0, "Add new xattr bucket starting form %llu\n",
3852              (unsigned long long)header_bh->b_blocknr);
3853
3854         /*
3855          * Add refrence for header_bh here because it may be
3856          * changed in ocfs2_add_new_xattr_cluster and we need
3857          * to free it in the end.
3858          */
3859         get_bh(header_bh);
3860
3861         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3862                                   &num_clusters, el);
3863         if (ret) {
3864                 mlog_errno(ret);
3865                 goto out;
3866         }
3867
3868         ret = ocfs2_read_block(osb, p_blkno,
3869                                &first_bh, OCFS2_BH_CACHED, inode);
3870         if (ret) {
3871                 mlog_errno(ret);
3872                 goto out;
3873         }
3874
3875         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3876         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3877
3878         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3879                 ret = ocfs2_add_new_xattr_cluster(inode,
3880                                                   xb_bh,
3881                                                   &first_bh,
3882                                                   &header_bh,
3883                                                   &num_clusters,
3884                                                   e_cpos,
3885                                                   p_blkno,
3886                                                   &extend);
3887                 if (ret) {
3888                         mlog_errno(ret);
3889                         goto out;
3890                 }
3891         }
3892
3893         if (extend)
3894                 ret = ocfs2_extend_xattr_bucket(inode,
3895                                                 first_bh,
3896                                                 header_bh,
3897                                                 num_clusters);
3898         if (ret)
3899                 mlog_errno(ret);
3900 out:
3901         brelse(first_bh);
3902         brelse(header_bh);
3903         return ret;
3904 }
3905
3906 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3907                                         struct ocfs2_xattr_bucket *bucket,
3908                                         int offs)
3909 {
3910         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3911
3912         offs = offs % inode->i_sb->s_blocksize;
3913         return bucket->bhs[block_off]->b_data + offs;
3914 }
3915
3916 /*
3917  * Handle the normal xattr set, including replace, delete and new.
3918  * When the bucket is empty, "is_empty" is set and the caller can
3919  * free this bucket.
3920  *
3921  * Note: "local" indicates the real data's locality. So we can't
3922  * just its bucket locality by its length.
3923  */
3924 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3925                                          struct ocfs2_xattr_info *xi,
3926                                          struct ocfs2_xattr_search *xs,
3927                                          u32 name_hash,
3928                                          int local,
3929                                          int *is_empty)
3930 {
3931         struct ocfs2_xattr_entry *last, *xe;
3932         int name_len = strlen(xi->name);
3933         struct ocfs2_xattr_header *xh = xs->header;
3934         u16 count = le16_to_cpu(xh->xh_count), start;
3935         size_t blocksize = inode->i_sb->s_blocksize;
3936         char *val;
3937         size_t offs, size, new_size;
3938
3939         last = &xh->xh_entries[count];
3940         if (!xs->not_found) {
3941                 xe = xs->here;
3942                 offs = le16_to_cpu(xe->xe_name_offset);
3943                 if (ocfs2_xattr_is_local(xe))
3944                         size = OCFS2_XATTR_SIZE(name_len) +
3945                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3946                 else
3947                         size = OCFS2_XATTR_SIZE(name_len) +
3948                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3949
3950                 /*
3951                  * If the new value will be stored outside, xi->value has been
3952                  * initalized as an empty ocfs2_xattr_value_root, and the same
3953                  * goes with xi->value_len, so we can set new_size safely here.
3954                  * See ocfs2_xattr_set_in_bucket.
3955                  */
3956                 new_size = OCFS2_XATTR_SIZE(name_len) +
3957                            OCFS2_XATTR_SIZE(xi->value_len);
3958
3959                 le16_add_cpu(&xh->xh_name_value_len, -size);
3960                 if (xi->value) {
3961                         if (new_size > size)
3962                                 goto set_new_name_value;
3963
3964                         /* Now replace the old value with new one. */
3965                         if (local)
3966                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3967                         else
3968                                 xe->xe_value_size = 0;
3969
3970                         val = ocfs2_xattr_bucket_get_val(inode,
3971                                                          &xs->bucket, offs);
3972                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3973                                size - OCFS2_XATTR_SIZE(name_len));
3974                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3975                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3976                                        xi->value, xi->value_len);
3977
3978                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3979                         ocfs2_xattr_set_local(xe, local);
3980                         return;
3981                 } else {
3982                         /* Remove the old entry. */
3983                         last -= 1;
3984                         memmove(xe, xe + 1,
3985                                 (void *)last - (void *)xe);
3986                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
3987                         le16_add_cpu(&xh->xh_count, -1);
3988                         if (xh->xh_count == 0 && is_empty)
3989                                 *is_empty = 1;
3990                         return;
3991                 }
3992         } else {
3993                 /* find a new entry for insert. */
3994                 int low = 0, high = count - 1, tmp;
3995                 struct ocfs2_xattr_entry *tmp_xe;
3996
3997                 while (low <= high) {
3998                         tmp = (low + high) / 2;
3999                         tmp_xe = &xh->xh_entries[tmp];
4000
4001                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4002                                 low = tmp + 1;
4003                         else if (name_hash <
4004                                  le32_to_cpu(tmp_xe->xe_name_hash))
4005                                 high = tmp - 1;
4006                         else {
4007                                 low = tmp;
4008                                 break;
4009                         }
4010                 }
4011
4012                 xe = &xh->xh_entries[low];
4013                 if (low != count)
4014                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4015
4016                 le16_add_cpu(&xh->xh_count, 1);
4017                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4018                 xe->xe_name_hash = cpu_to_le32(name_hash);
4019                 xe->xe_name_len = name_len;
4020                 ocfs2_xattr_set_type(xe, xi->name_index);
4021         }
4022
4023 set_new_name_value:
4024         /* Insert the new name+value. */
4025         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4026
4027         /*
4028          * We must make sure that the name/value pair
4029          * exists in the same block.
4030          */
4031         offs = le16_to_cpu(xh->xh_free_start);
4032         start = offs - size;
4033
4034         if (start >> inode->i_sb->s_blocksize_bits !=
4035             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4036                 offs = offs - offs % blocksize;
4037                 xh->xh_free_start = cpu_to_le16(offs);
4038         }
4039
4040         val = ocfs2_xattr_bucket_get_val(inode,
4041                                          &xs->bucket, offs - size);
4042         xe->xe_name_offset = cpu_to_le16(offs - size);
4043
4044         memset(val, 0, size);
4045         memcpy(val, xi->name, name_len);
4046         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4047
4048         xe->xe_value_size = cpu_to_le64(xi->value_len);
4049         ocfs2_xattr_set_local(xe, local);
4050         xs->here = xe;
4051         le16_add_cpu(&xh->xh_free_start, -size);
4052         le16_add_cpu(&xh->xh_name_value_len, size);
4053
4054         return;
4055 }
4056
4057 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4058                                              handle_t *handle,
4059                                              struct ocfs2_xattr_search *xs,
4060                                              struct buffer_head **bhs,
4061                                              u16 bh_num)
4062 {
4063         int ret = 0, off, block_off;
4064         struct ocfs2_xattr_entry *xe = xs->here;
4065
4066         /*
4067          * First calculate all the blocks we should journal_access
4068          * and journal_dirty. The first block should always be touched.
4069          */
4070         ret = ocfs2_journal_dirty(handle, bhs[0]);
4071         if (ret)
4072                 mlog_errno(ret);
4073
4074         /* calc the data. */
4075         off = le16_to_cpu(xe->xe_name_offset);
4076         block_off = off >> inode->i_sb->s_blocksize_bits;
4077         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4078         if (ret)
4079                 mlog_errno(ret);
4080
4081         return ret;
4082 }
4083
4084 /*
4085  * Set the xattr entry in the specified bucket.
4086  * The bucket is indicated by xs->bucket and it should have the enough
4087  * space for the xattr insertion.
4088  */
4089 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4090                                            struct ocfs2_xattr_info *xi,
4091                                            struct ocfs2_xattr_search *xs,
4092                                            u32 name_hash,
4093                                            int local,
4094                                            int *bucket_empty)
4095 {
4096         int i, ret;
4097         handle_t *handle = NULL;
4098         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4099         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4100
4101         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4102              (unsigned long)xi->value_len, xi->name_index,
4103              (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4104
4105         if (!xs->bucket.bhs[1]) {
4106                 ret = ocfs2_read_blocks(osb,
4107                                         xs->bucket.bhs[0]->b_blocknr + 1,
4108                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
4109                                         OCFS2_BH_CACHED, inode);
4110                 if (ret) {
4111                         mlog_errno(ret);
4112                         goto out;
4113                 }
4114         }
4115
4116         handle = ocfs2_start_trans(osb, blk_per_bucket);
4117         if (IS_ERR(handle)) {
4118                 ret = PTR_ERR(handle);
4119                 handle = NULL;
4120                 mlog_errno(ret);
4121                 goto out;
4122         }
4123
4124         for (i = 0; i < blk_per_bucket; i++) {
4125                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4126                                            OCFS2_JOURNAL_ACCESS_WRITE);
4127                 if (ret < 0) {
4128                         mlog_errno(ret);
4129                         goto out;
4130                 }
4131         }
4132
4133         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash,
4134                                      local, bucket_empty);
4135
4136         /*Only dirty the blocks we have touched in set xattr. */
4137         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4138                                                 xs->bucket.bhs, blk_per_bucket);
4139         if (ret)
4140                 mlog_errno(ret);
4141 out:
4142         ocfs2_commit_trans(osb, handle);
4143
4144         return ret;
4145 }
4146
4147 static int ocfs2_xattr_value_update_size(struct inode *inode,
4148                                          struct buffer_head *xe_bh,
4149                                          struct ocfs2_xattr_entry *xe,
4150                                          u64 new_size)
4151 {
4152         int ret;
4153         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4154         handle_t *handle = NULL;
4155
4156         handle = ocfs2_start_trans(osb, 1);
4157         if (handle == NULL) {
4158                 ret = -ENOMEM;
4159                 mlog_errno(ret);
4160                 goto out;
4161         }
4162
4163         ret = ocfs2_journal_access(handle, inode, xe_bh,
4164                                    OCFS2_JOURNAL_ACCESS_WRITE);
4165         if (ret < 0) {
4166                 mlog_errno(ret);
4167                 goto out_commit;
4168         }
4169
4170         xe->xe_value_size = cpu_to_le64(new_size);
4171
4172         ret = ocfs2_journal_dirty(handle, xe_bh);
4173         if (ret < 0)
4174                 mlog_errno(ret);
4175
4176 out_commit:
4177         ocfs2_commit_trans(osb, handle);
4178 out:
4179         return ret;
4180 }
4181
4182 /*
4183  * Truncate the specified xe_off entry in xattr bucket.
4184  * bucket is indicated by header_bh and len is the new length.
4185  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4186  *
4187  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4188  */
4189 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4190                                              struct buffer_head *header_bh,
4191                                              int xe_off,
4192                                              int len)
4193 {
4194         int ret, offset;
4195         u64 value_blk;
4196         struct buffer_head *value_bh = NULL;
4197         struct ocfs2_xattr_value_root *xv;
4198         struct ocfs2_xattr_entry *xe;
4199         struct ocfs2_xattr_header *xh =
4200                         (struct ocfs2_xattr_header *)header_bh->b_data;
4201         size_t blocksize = inode->i_sb->s_blocksize;
4202
4203         xe = &xh->xh_entries[xe_off];
4204
4205         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4206
4207         offset = le16_to_cpu(xe->xe_name_offset) +
4208                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4209
4210         value_blk = offset / blocksize;
4211
4212         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4213         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4214         value_blk += header_bh->b_blocknr;
4215
4216         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk,
4217                                &value_bh, OCFS2_BH_CACHED, inode);
4218         if (ret) {
4219                 mlog_errno(ret);
4220                 goto out;
4221         }
4222
4223         xv = (struct ocfs2_xattr_value_root *)
4224                 (value_bh->b_data + offset % blocksize);
4225
4226         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4227              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4228         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4229         if (ret) {
4230                 mlog_errno(ret);
4231                 goto out;
4232         }
4233
4234         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4235         if (ret) {
4236                 mlog_errno(ret);
4237                 goto out;
4238         }
4239
4240 out:
4241         brelse(value_bh);
4242         return ret;
4243 }
4244
4245 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4246                                                 struct ocfs2_xattr_search *xs,
4247                                                 int len)
4248 {
4249         int ret, offset;
4250         struct ocfs2_xattr_entry *xe = xs->here;
4251         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4252
4253         BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4254
4255         offset = xe - xh->xh_entries;
4256         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4257                                                 offset, len);
4258         if (ret)
4259                 mlog_errno(ret);
4260
4261         return ret;
4262 }
4263
4264 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4265                                                 struct ocfs2_xattr_search *xs,
4266                                                 char *val,
4267                                                 int value_len)
4268 {
4269         int offset;
4270         struct ocfs2_xattr_value_root *xv;
4271         struct ocfs2_xattr_entry *xe = xs->here;
4272
4273         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4274
4275         offset = le16_to_cpu(xe->xe_name_offset) +
4276                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4277
4278         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4279
4280         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4281 }
4282
4283 /*
4284  * Remove the xattr bucket pointed by bucket_bh.
4285  * All the buckets after it in the same xattr extent rec will be
4286  * move forward one by one.
4287  */
4288 static int ocfs2_rm_xattr_bucket(struct inode *inode,
4289                                  struct buffer_head *first_bh,
4290                                  struct ocfs2_xattr_bucket *bucket)
4291 {
4292         int ret = 0, credits;
4293         struct ocfs2_xattr_header *xh =
4294                                 (struct ocfs2_xattr_header *)first_bh->b_data;
4295         u16 bucket_num = le16_to_cpu(xh->xh_num_buckets);
4296         u64 end, start = bucket->bhs[0]->b_blocknr;
4297         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4298         handle_t *handle;
4299         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4300
4301         end = first_bh->b_blocknr + (bucket_num - 1) * blk_per_bucket;
4302
4303         mlog(0, "rm xattr bucket %llu\n", start);
4304         /*
4305          * We need to update the first xattr_header and all the buckets starting
4306          * from start in this xattr rec.
4307          *
4308          * XXX: Should we empty the old last bucket here?
4309          */
4310         credits = 1 + end - start;
4311         handle = ocfs2_start_trans(osb, credits);
4312         if (IS_ERR(handle)) {
4313                 ret = PTR_ERR(handle);
4314                 mlog_errno(ret);
4315                 return ret;
4316         }
4317
4318         ret = ocfs2_journal_access(handle, inode, first_bh,
4319                                    OCFS2_JOURNAL_ACCESS_WRITE);
4320         if (ret) {
4321                 mlog_errno(ret);
4322                 goto out_commit;
4323         }
4324
4325
4326         while (start < end) {
4327                 ret = ocfs2_cp_xattr_bucket(inode, handle,
4328                                             start + blk_per_bucket,
4329                                             start, 0);
4330                 if (ret) {
4331                         mlog_errno(ret);
4332                         goto out_commit;
4333                 }
4334                 start += blk_per_bucket;
4335         }
4336
4337         /* update the first_bh. */
4338         xh->xh_num_buckets = cpu_to_le16(bucket_num - 1);
4339         ocfs2_journal_dirty(handle, first_bh);
4340
4341 out_commit:
4342         ocfs2_commit_trans(osb, handle);
4343         return ret;
4344 }
4345
4346 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4347                                   struct buffer_head *root_bh,
4348                                   u64 blkno,
4349                                   u32 cpos,
4350                                   u32 len)
4351 {
4352         int ret;
4353         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4354         struct inode *tl_inode = osb->osb_tl_inode;
4355         handle_t *handle;
4356         struct ocfs2_xattr_block *xb =
4357                         (struct ocfs2_xattr_block *)root_bh->b_data;
4358         struct ocfs2_alloc_context *meta_ac = NULL;
4359         struct ocfs2_cached_dealloc_ctxt dealloc;
4360         struct ocfs2_extent_tree et;
4361
4362         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4363
4364         ocfs2_init_dealloc_ctxt(&dealloc);
4365
4366         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4367              cpos, len, (unsigned long long)blkno);
4368
4369         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4370
4371         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4372         if (ret) {
4373                 mlog_errno(ret);
4374                 return ret;
4375         }
4376
4377         mutex_lock(&tl_inode->i_mutex);
4378
4379         if (ocfs2_truncate_log_needs_flush(osb)) {
4380                 ret = __ocfs2_flush_truncate_log(osb);
4381                 if (ret < 0) {
4382                         mlog_errno(ret);
4383                         goto out;
4384                 }
4385         }
4386
4387         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4388         if (handle == NULL) {
4389                 ret = -ENOMEM;
4390                 mlog_errno(ret);
4391                 goto out;
4392         }
4393
4394         ret = ocfs2_journal_access(handle, inode, root_bh,
4395                                    OCFS2_JOURNAL_ACCESS_WRITE);
4396         if (ret) {
4397                 mlog_errno(ret);
4398                 goto out_commit;
4399         }
4400
4401         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4402                                   &dealloc);
4403         if (ret) {
4404                 mlog_errno(ret);
4405                 goto out_commit;
4406         }
4407
4408         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4409
4410         ret = ocfs2_journal_dirty(handle, root_bh);
4411         if (ret) {
4412                 mlog_errno(ret);
4413                 goto out_commit;
4414         }
4415
4416         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4417         if (ret)
4418                 mlog_errno(ret);
4419
4420 out_commit:
4421         ocfs2_commit_trans(osb, handle);
4422 out:
4423         ocfs2_schedule_truncate_log_flush(osb, 1);
4424
4425         mutex_unlock(&tl_inode->i_mutex);
4426
4427         if (meta_ac)
4428                 ocfs2_free_alloc_context(meta_ac);
4429
4430         ocfs2_run_deallocs(osb, &dealloc);
4431
4432         return ret;
4433 }
4434
4435 /*
4436  * Free the xattr bucket indicated by xs->bucket and if all the buckets
4437  * in the clusters is free, free the clusters also.
4438  */
4439 static int ocfs2_xattr_bucket_shrink(struct inode *inode,
4440                                      struct ocfs2_xattr_info *xi,
4441                                      struct ocfs2_xattr_search *xs,
4442                                      u32 name_hash)
4443 {
4444         int ret;
4445         u32 e_cpos, num_clusters;
4446         u64 p_blkno;
4447         struct buffer_head *first_bh = NULL;
4448         struct ocfs2_xattr_header *first_xh;
4449         struct ocfs2_xattr_block *xb =
4450                         (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
4451
4452         BUG_ON(xs->header->xh_count != 0);
4453
4454         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4455                                   &e_cpos, &num_clusters,
4456                                   &xb->xb_attrs.xb_root.xt_list);
4457         if (ret) {
4458                 mlog_errno(ret);
4459                 return ret;
4460         }
4461
4462         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
4463                                &first_bh, OCFS2_BH_CACHED, inode);
4464         if (ret) {
4465                 mlog_errno(ret);
4466                 return ret;
4467         }
4468
4469         ret = ocfs2_rm_xattr_bucket(inode, first_bh, &xs->bucket);
4470         if (ret) {
4471                 mlog_errno(ret);
4472                 goto out;
4473         }
4474
4475         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
4476         if (first_xh->xh_num_buckets == 0)
4477                 ret = ocfs2_rm_xattr_cluster(inode, xs->xattr_bh,
4478                                              p_blkno, e_cpos,
4479                                              num_clusters);
4480
4481 out:
4482         brelse(first_bh);
4483         return ret;
4484 }
4485
4486 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4487                                          struct ocfs2_xattr_search *xs)
4488 {
4489         handle_t *handle = NULL;
4490         struct ocfs2_xattr_header *xh = xs->bucket.xh;
4491         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4492                                                 le16_to_cpu(xh->xh_count) - 1];
4493         int ret = 0;
4494
4495         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4496         if (IS_ERR(handle)) {
4497                 ret = PTR_ERR(handle);
4498                 mlog_errno(ret);
4499                 return;
4500         }
4501
4502         ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4503                                    OCFS2_JOURNAL_ACCESS_WRITE);
4504         if (ret) {
4505                 mlog_errno(ret);
4506                 goto out_commit;
4507         }
4508
4509         /* Remove the old entry. */
4510         memmove(xs->here, xs->here + 1,
4511                 (void *)last - (void *)xs->here);
4512         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4513         le16_add_cpu(&xh->xh_count, -1);
4514
4515         ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4516         if (ret < 0)
4517                 mlog_errno(ret);
4518 out_commit:
4519         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4520 }
4521
4522 /*
4523  * Set the xattr name/value in the bucket specified in xs.
4524  *
4525  * As the new value in xi may be stored in the bucket or in an outside cluster,
4526  * we divide the whole process into 3 steps:
4527  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4528  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4529  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4530  * 4. If the clusters for the new outside value can't be allocated, we need
4531  *    to free the xattr we allocated in set.
4532  */
4533 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4534                                      struct ocfs2_xattr_info *xi,
4535                                      struct ocfs2_xattr_search *xs)
4536 {
4537         int ret, local = 1, bucket_empty = 0;
4538         size_t value_len;
4539         char *val = (char *)xi->value;
4540         struct ocfs2_xattr_entry *xe = xs->here;
4541         u32 name_hash = ocfs2_xattr_hash_by_name(inode,
4542                                                  xi->name_index, xi->name);
4543
4544         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4545                 /*
4546                  * We need to truncate the xattr storage first.
4547                  *
4548                  * If both the old and new value are stored to
4549                  * outside block, we only need to truncate
4550                  * the storage and then set the value outside.
4551                  *
4552                  * If the new value should be stored within block,
4553                  * we should free all the outside block first and
4554                  * the modification to the xattr block will be done
4555                  * by following steps.
4556                  */
4557                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4558                         value_len = xi->value_len;
4559                 else
4560                         value_len = 0;
4561
4562                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4563                                                            value_len);
4564                 if (ret)
4565                         goto out;
4566
4567                 if (value_len)
4568                         goto set_value_outside;
4569         }
4570
4571         value_len = xi->value_len;
4572         /* So we have to handle the inside block change now. */
4573         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4574                 /*
4575                  * If the new value will be stored outside of block,
4576                  * initalize a new empty value root and insert it first.
4577                  */
4578                 local = 0;
4579                 xi->value = &def_xv;
4580                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4581         }
4582
4583         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash,
4584                                               local, &bucket_empty);
4585         if (ret) {
4586                 mlog_errno(ret);
4587                 goto out;
4588         }
4589
4590         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4591                 /* allocate the space now for the outside block storage. */
4592                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4593                                                            value_len);
4594                 if (ret) {
4595                         mlog_errno(ret);
4596
4597                         if (xs->not_found) {
4598                                 /*
4599                                  * We can't allocate enough clusters for outside
4600                                  * storage and we have allocated xattr already,
4601                                  * so need to remove it.
4602                                  */
4603                                 ocfs2_xattr_bucket_remove_xs(inode, xs);
4604                         }
4605                         goto out;
4606                 }
4607         } else {
4608                 if (bucket_empty)
4609                         ret = ocfs2_xattr_bucket_shrink(inode, xi,
4610                                                         xs, name_hash);
4611                 goto out;
4612         }
4613
4614 set_value_outside:
4615         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4616 out:
4617         return ret;
4618 }
4619
4620 /* check whether the xattr bucket is filled up with the same hash value. */
4621 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4622                                               struct ocfs2_xattr_bucket *bucket)
4623 {
4624         struct ocfs2_xattr_header *xh = bucket->xh;
4625
4626         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4627             xh->xh_entries[0].xe_name_hash) {
4628                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4629                      "hash = %u\n",
4630                      (unsigned long long)bucket->bhs[0]->b_blocknr,
4631                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4632                 return -ENOSPC;
4633         }
4634
4635         return 0;
4636 }
4637
4638 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4639                                              struct ocfs2_xattr_info *xi,
4640                                              struct ocfs2_xattr_search *xs)
4641 {
4642         struct ocfs2_xattr_header *xh;
4643         struct ocfs2_xattr_entry *xe;
4644         u16 count, header_size, xh_free_start;
4645         int i, free, max_free, need, old;
4646         size_t value_size = 0, name_len = strlen(xi->name);
4647         size_t blocksize = inode->i_sb->s_blocksize;
4648         int ret, allocation = 0;
4649         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4650
4651         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4652
4653 try_again:
4654         xh = xs->header;
4655         count = le16_to_cpu(xh->xh_count);
4656         xh_free_start = le16_to_cpu(xh->xh_free_start);
4657         header_size = sizeof(struct ocfs2_xattr_header) +
4658                         count * sizeof(struct ocfs2_xattr_entry);
4659         max_free = OCFS2_XATTR_BUCKET_SIZE -
4660                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4661
4662         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4663                         "of %u which exceed block size\n",
4664                         (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4665                         header_size);
4666
4667         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4668                 value_size = OCFS2_XATTR_ROOT_SIZE;
4669         else if (xi->value)
4670                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4671
4672         if (xs->not_found)
4673                 need = sizeof(struct ocfs2_xattr_entry) +
4674                         OCFS2_XATTR_SIZE(name_len) + value_size;
4675         else {
4676                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4677
4678                 /*
4679                  * We only replace the old value if the new length is smaller
4680                  * than the old one. Otherwise we will allocate new space in the
4681                  * bucket to store it.
4682                  */
4683                 xe = xs->here;
4684                 if (ocfs2_xattr_is_local(xe))
4685                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4686                 else
4687                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4688
4689                 if (old >= value_size)
4690                         need = 0;
4691         }
4692
4693         free = xh_free_start - header_size;
4694         /*
4695          * We need to make sure the new name/value pair
4696          * can exist in the same block.
4697          */
4698         if (xh_free_start % blocksize < need)
4699                 free -= xh_free_start % blocksize;
4700
4701         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4702              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4703              " %u\n", xs->not_found,
4704              (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4705              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4706              le16_to_cpu(xh->xh_name_value_len));
4707
4708         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4709                 if (need <= max_free &&
4710                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4711                         /*
4712                          * We can create the space by defragment. Since only the
4713                          * name/value will be moved, the xe shouldn't be changed
4714                          * in xs.
4715                          */
4716                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4717                         if (ret) {
4718                                 mlog_errno(ret);
4719                                 goto out;
4720                         }
4721
4722                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4723                         free = xh_free_start - header_size;
4724                         if (xh_free_start % blocksize < need)
4725                                 free -= xh_free_start % blocksize;
4726
4727                         if (free >= need)
4728                                 goto xattr_set;
4729
4730                         mlog(0, "Can't get enough space for xattr insert by "
4731                              "defragment. Need %u bytes, but we have %d, so "
4732                              "allocate new bucket for it.\n", need, free);
4733                 }
4734
4735                 /*
4736                  * We have to add new buckets or clusters and one
4737                  * allocation should leave us enough space for insert.
4738                  */
4739                 BUG_ON(allocation);
4740
4741                 /*
4742                  * We do not allow for overlapping ranges between buckets. And
4743                  * the maximum number of collisions we will allow for then is
4744                  * one bucket's worth, so check it here whether we need to
4745                  * add a new bucket for the insert.
4746                  */
4747                 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4748                 if (ret) {
4749                         mlog_errno(ret);
4750                         goto out;
4751                 }
4752
4753                 ret = ocfs2_add_new_xattr_bucket(inode,
4754                                                  xs->xattr_bh,
4755                                                  xs->bucket.bhs[0]);
4756                 if (ret) {
4757                         mlog_errno(ret);
4758                         goto out;
4759                 }
4760
4761                 for (i = 0; i < blk_per_bucket; i++)
4762                         brelse(xs->bucket.bhs[i]);
4763
4764                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4765
4766                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4767                                                    xi->name_index,
4768                                                    xi->name, xs);
4769                 if (ret && ret != -ENODATA)
4770                         goto out;
4771                 xs->not_found = ret;
4772                 allocation = 1;
4773                 goto try_again;
4774         }
4775
4776 xattr_set:
4777         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4778 out:
4779         mlog_exit(ret);
4780         return ret;
4781 }
4782
4783 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4784                                         struct ocfs2_xattr_bucket *bucket,
4785                                         void *para)
4786 {
4787         int ret = 0;
4788         struct ocfs2_xattr_header *xh = bucket->xh;
4789         u16 i;
4790         struct ocfs2_xattr_entry *xe;
4791
4792         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4793                 xe = &xh->xh_entries[i];
4794                 if (ocfs2_xattr_is_local(xe))
4795                         continue;
4796
4797                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4798                                                         bucket->bhs[0],
4799                                                         i, 0);
4800                 if (ret) {
4801                         mlog_errno(ret);
4802                         break;
4803                 }
4804         }
4805
4806         return ret;
4807 }
4808
4809 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4810                                           struct buffer_head *xb_bh)
4811 {
4812         struct ocfs2_xattr_block *xb =
4813                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4814         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4815         int ret = 0;
4816         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4817         u64 p_blkno;
4818
4819         if (le16_to_cpu(el->l_next_free_rec) == 0)
4820                 return 0;
4821
4822         while (name_hash > 0) {
4823                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4824                                           &e_cpos, &num_clusters, el);
4825                 if (ret) {
4826                         mlog_errno(ret);
4827                         goto out;
4828                 }
4829
4830                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4831                                                   ocfs2_delete_xattr_in_bucket,
4832                                                   NULL);
4833                 if (ret) {
4834                         mlog_errno(ret);
4835                         goto out;
4836                 }
4837
4838                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4839                                              p_blkno, e_cpos, num_clusters);
4840                 if (ret) {
4841                         mlog_errno(ret);
4842                         break;
4843                 }
4844
4845                 if (e_cpos == 0)
4846                         break;
4847
4848                 name_hash = e_cpos - 1;
4849         }
4850
4851 out:
4852         return ret;
4853 }