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 #include <linux/init.h>
  41 #include <linux/module.h>
  42 #include <linux/string.h>
  43
  44 #define MLOG_MASK_PREFIX ML_XATTR
  45 #include <cluster/masklog.h>
  46
  47 #include "ocfs2.h"
  48 #include "alloc.h"
  49 #include "dlmglue.h"
  50 #include "file.h"
  51 #include "symlink.h"
  52 #include "sysfile.h"
  53 #include "inode.h"
  54 #include "journal.h"
  55 #include "ocfs2_fs.h"
  56 #include "suballoc.h"
  57 #include "uptodate.h"
  58 #include "buffer_head_io.h"
  59 #include "super.h"
  60 #include "xattr.h"
  61
  62
  63 struct ocfs2_xattr_def_value_root {
  64         struct ocfs2_xattr_value_root   xv;
  65         struct ocfs2_extent_rec         er;
  66 };
  67
  68 struct ocfs2_xattr_bucket {
  69         struct buffer_head *bhs[OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET];
  70         struct ocfs2_xattr_header *xh;
  71 };
  72
  73 #define OCFS2_XATTR_ROOT_SIZE   (sizeof(struct ocfs2_xattr_def_value_root))
  74 #define OCFS2_XATTR_INLINE_SIZE 80
  75
  76 static struct ocfs2_xattr_def_value_root def_xv = {
  77         .xv.xr_list.l_count = cpu_to_le16(1),
  78 };
  79
  80 struct xattr_handler *ocfs2_xattr_handlers[] = {
  81         &ocfs2_xattr_user_handler,
  82         &ocfs2_xattr_trusted_handler,
  83         NULL
  84 };
  85
  86 static struct xattr_handler *ocfs2_xattr_handler_map[] = {
  87         [OCFS2_XATTR_INDEX_USER]        = &ocfs2_xattr_user_handler,
  88         [OCFS2_XATTR_INDEX_TRUSTED]     = &ocfs2_xattr_trusted_handler,
  89 };
  90
  91 struct ocfs2_xattr_info {
  92         int name_index;
  93         const char *name;
  94         const void *value;
  95         size_t value_len;
  96 };
  97
  98 struct ocfs2_xattr_search {
  99         struct buffer_head *inode_bh;
 100         /*
 101          * xattr_bh point to the block buffer head which has extended attribute
 102          * when extended attribute in inode, xattr_bh is equal to inode_bh.
 103          */
 104         struct buffer_head *xattr_bh;
 105         struct ocfs2_xattr_header *header;
 106         struct ocfs2_xattr_bucket bucket;
 107         void *base;
 108         void *end;
 109         struct ocfs2_xattr_entry *here;
 110         int not_found;
 111 };
 112
 113 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
 114                                              struct ocfs2_xattr_header *xh,
 115                                              int index,
 116                                              int *block_off,
 117                                              int *new_offset);
 118
 119 static int ocfs2_xattr_index_block_find(struct inode *inode,
 120                                         struct buffer_head *root_bh,
 121                                         int name_index,
 122                                         const char *name,
 123                                         struct ocfs2_xattr_search *xs);
 124
 125 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
 126                                         struct ocfs2_xattr_tree_root *xt,
 127                                         char *buffer,
 128                                         size_t buffer_size);
 129
 130 static int ocfs2_xattr_create_index_block(struct inode *inode,
 131                                           struct ocfs2_xattr_search *xs);
 132
 133 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
 134                                              struct ocfs2_xattr_info *xi,
 135                                              struct ocfs2_xattr_search *xs);
 136
 137 static int ocfs2_delete_xattr_index_block(struct inode *inode,
 138                                           struct buffer_head *xb_bh);
 139
 140 static inline const char *ocfs2_xattr_prefix(int name_index)
 141 {
 142         struct xattr_handler *handler = NULL;
 143
 144         if (name_index > 0 && name_index < OCFS2_XATTR_MAX)
 145                 handler = ocfs2_xattr_handler_map[name_index];
 146
 147         return handler ? handler->prefix : NULL;
 148 }
 149
 150 static u32 ocfs2_xattr_name_hash(struct inode *inode,
 151                                  const char *name,
 152                                  int name_len)
 153 {
 154         /* Get hash value of uuid from super block */
 155         u32 hash = OCFS2_SB(inode->i_sb)->uuid_hash;
 156         int i;
 157
 158         /* hash extended attribute name */
 159         for (i = 0; i < name_len; i++) {
 160                 hash = (hash << OCFS2_HASH_SHIFT) ^
 161                        (hash >> (8*sizeof(hash) - OCFS2_HASH_SHIFT)) ^
 162                        *name++;
 163         }
 164
 165         return hash;
 166 }
 167
 168 /*
 169  * ocfs2_xattr_hash_entry()
 170  *
 171  * Compute the hash of an extended attribute.
 172  */
 173 static void ocfs2_xattr_hash_entry(struct inode *inode,
 174                                    struct ocfs2_xattr_header *header,
 175                                    struct ocfs2_xattr_entry *entry)
 176 {
 177         u32 hash = 0;
 178         char *name = (char *)header + le16_to_cpu(entry->xe_name_offset);
 179
 180         hash = ocfs2_xattr_name_hash(inode, name, entry->xe_name_len);
 181         entry->xe_name_hash = cpu_to_le32(hash);
 182
 183         return;
 184 }
 185
 186 static int ocfs2_xattr_extend_allocation(struct inode *inode,
 187                                          u32 clusters_to_add,
 188                                          struct buffer_head *xattr_bh,
 189                                          struct ocfs2_xattr_value_root *xv)
 190 {
 191         int status = 0;
 192         int restart_func = 0;
 193         int credits = 0;
 194         handle_t *handle = NULL;
 195         struct ocfs2_alloc_context *data_ac = NULL;
 196         struct ocfs2_alloc_context *meta_ac = NULL;
 197         enum ocfs2_alloc_restarted why;
 198         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 199         u32 prev_clusters, logical_start = le32_to_cpu(xv->xr_clusters);
 200         struct ocfs2_extent_tree et;
 201
 202         mlog(0, "(clusters_to_add for xattr= %u)\n", clusters_to_add);
 203
 204         ocfs2_init_xattr_value_extent_tree(&et, inode, xattr_bh, xv);
 205
 206 restart_all:
 207
 208         status = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
 209                                        &data_ac, &meta_ac);
 210         if (status) {
 211                 mlog_errno(status);
 212                 goto leave;
 213         }
 214
 215         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
 216                                             clusters_to_add);
 217         handle = ocfs2_start_trans(osb, credits);
 218         if (IS_ERR(handle)) {
 219                 status = PTR_ERR(handle);
 220                 handle = NULL;
 221                 mlog_errno(status);
 222                 goto leave;
 223         }
 224
 225 restarted_transaction:
 226         status = ocfs2_journal_access(handle, inode, xattr_bh,
 227                                       OCFS2_JOURNAL_ACCESS_WRITE);
 228         if (status < 0) {
 229                 mlog_errno(status);
 230                 goto leave;
 231         }
 232
 233         prev_clusters = le32_to_cpu(xv->xr_clusters);
 234         status = ocfs2_add_clusters_in_btree(osb,
 235                                              inode,
 236                                              &logical_start,
 237                                              clusters_to_add,
 238                                              0,
 239                                              &et,
 240                                              handle,
 241                                              data_ac,
 242                                              meta_ac,
 243                                              &why);
 244         if ((status < 0) && (status != -EAGAIN)) {
 245                 if (status != -ENOSPC)
 246                         mlog_errno(status);
 247                 goto leave;
 248         }
 249
 250         status = ocfs2_journal_dirty(handle, xattr_bh);
 251         if (status < 0) {
 252                 mlog_errno(status);
 253                 goto leave;
 254         }
 255
 256         clusters_to_add -= le32_to_cpu(xv->xr_clusters) - prev_clusters;
 257
 258         if (why != RESTART_NONE && clusters_to_add) {
 259                 if (why == RESTART_META) {
 260                         mlog(0, "restarting function.\n");
 261                         restart_func = 1;
 262                 } else {
 263                         BUG_ON(why != RESTART_TRANS);
 264
 265                         mlog(0, "restarting transaction.\n");
 266                         /* TODO: This can be more intelligent. */
 267                         credits = ocfs2_calc_extend_credits(osb->sb,
 268                                                             et.et_root_el,
 269                                                             clusters_to_add);
 270                         status = ocfs2_extend_trans(handle, credits);
 271                         if (status < 0) {
 272                                 /* handle still has to be committed at
 273                                  * this point. */
 274                                 status = -ENOMEM;
 275                                 mlog_errno(status);
 276                                 goto leave;
 277                         }
 278                         goto restarted_transaction;
 279                 }
 280         }
 281
 282 leave:
 283         if (handle) {
 284                 ocfs2_commit_trans(osb, handle);
 285                 handle = NULL;
 286         }
 287         if (data_ac) {
 288                 ocfs2_free_alloc_context(data_ac);
 289                 data_ac = NULL;
 290         }
 291         if (meta_ac) {
 292                 ocfs2_free_alloc_context(meta_ac);
 293                 meta_ac = NULL;
 294         }
 295         if ((!status) && restart_func) {
 296                 restart_func = 0;
 297                 goto restart_all;
 298         }
 299
 300         return status;
 301 }
 302
 303 static int __ocfs2_remove_xattr_range(struct inode *inode,
 304                                       struct buffer_head *root_bh,
 305                                       struct ocfs2_xattr_value_root *xv,
 306                                       u32 cpos, u32 phys_cpos, u32 len,
 307                                       struct ocfs2_cached_dealloc_ctxt *dealloc)
 308 {
 309         int ret;
 310         u64 phys_blkno = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 311         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 312         struct inode *tl_inode = osb->osb_tl_inode;
 313         handle_t *handle;
 314         struct ocfs2_alloc_context *meta_ac = NULL;
 315         struct ocfs2_extent_tree et;
 316
 317         ocfs2_init_xattr_value_extent_tree(&et, inode, root_bh, xv);
 318
 319         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
 320         if (ret) {
 321                 mlog_errno(ret);
 322                 return ret;
 323         }
 324
 325         mutex_lock(&tl_inode->i_mutex);
 326
 327         if (ocfs2_truncate_log_needs_flush(osb)) {
 328                 ret = __ocfs2_flush_truncate_log(osb);
 329                 if (ret < 0) {
 330                         mlog_errno(ret);
 331                         goto out;
 332                 }
 333         }
 334
 335         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
 336         if (IS_ERR(handle)) {
 337                 ret = PTR_ERR(handle);
 338                 mlog_errno(ret);
 339                 goto out;
 340         }
 341
 342         ret = ocfs2_journal_access(handle, inode, root_bh,
 343                                    OCFS2_JOURNAL_ACCESS_WRITE);
 344         if (ret) {
 345                 mlog_errno(ret);
 346                 goto out_commit;
 347         }
 348
 349         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
 350                                   dealloc);
 351         if (ret) {
 352                 mlog_errno(ret);
 353                 goto out_commit;
 354         }
 355
 356         le32_add_cpu(&xv->xr_clusters, -len);
 357
 358         ret = ocfs2_journal_dirty(handle, root_bh);
 359         if (ret) {
 360                 mlog_errno(ret);
 361                 goto out_commit;
 362         }
 363
 364         ret = ocfs2_truncate_log_append(osb, handle, phys_blkno, len);
 365         if (ret)
 366                 mlog_errno(ret);
 367
 368 out_commit:
 369         ocfs2_commit_trans(osb, handle);
 370 out:
 371         mutex_unlock(&tl_inode->i_mutex);
 372
 373         if (meta_ac)
 374                 ocfs2_free_alloc_context(meta_ac);
 375
 376         return ret;
 377 }
 378
 379 static int ocfs2_xattr_shrink_size(struct inode *inode,
 380                                    u32 old_clusters,
 381                                    u32 new_clusters,
 382                                    struct buffer_head *root_bh,
 383                                    struct ocfs2_xattr_value_root *xv)
 384 {
 385         int ret = 0;
 386         u32 trunc_len, cpos, phys_cpos, alloc_size;
 387         u64 block;
 388         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
 389         struct ocfs2_cached_dealloc_ctxt dealloc;
 390
 391         ocfs2_init_dealloc_ctxt(&dealloc);
 392
 393         if (old_clusters <= new_clusters)
 394                 return 0;
 395
 396         cpos = new_clusters;
 397         trunc_len = old_clusters - new_clusters;
 398         while (trunc_len) {
 399                 ret = ocfs2_xattr_get_clusters(inode, cpos, &phys_cpos,
 400                                                &alloc_size, &xv->xr_list);
 401                 if (ret) {
 402                         mlog_errno(ret);
 403                         goto out;
 404                 }
 405
 406                 if (alloc_size > trunc_len)
 407                         alloc_size = trunc_len;
 408
 409                 ret = __ocfs2_remove_xattr_range(inode, root_bh, xv, cpos,
 410                                                  phys_cpos, alloc_size,
 411                                                  &dealloc);
 412                 if (ret) {
 413                         mlog_errno(ret);
 414                         goto out;
 415                 }
 416
 417                 block = ocfs2_clusters_to_blocks(inode->i_sb, phys_cpos);
 418                 ocfs2_remove_xattr_clusters_from_cache(inode, block,
 419                                                        alloc_size);
 420                 cpos += alloc_size;
 421                 trunc_len -= alloc_size;
 422         }
 423
 424 out:
 425         ocfs2_schedule_truncate_log_flush(osb, 1);
 426         ocfs2_run_deallocs(osb, &dealloc);
 427
 428         return ret;
 429 }
 430
 431 static int ocfs2_xattr_value_truncate(struct inode *inode,
 432                                       struct buffer_head *root_bh,
 433                                       struct ocfs2_xattr_value_root *xv,
 434                                       int len)
 435 {
 436         int ret;
 437         u32 new_clusters = ocfs2_clusters_for_bytes(inode->i_sb, len);
 438         u32 old_clusters = le32_to_cpu(xv->xr_clusters);
 439
 440         if (new_clusters == old_clusters)
 441                 return 0;
 442
 443         if (new_clusters > old_clusters)
 444                 ret = ocfs2_xattr_extend_allocation(inode,
 445                                                     new_clusters - old_clusters,
 446                                                     root_bh, xv);
 447         else
 448                 ret = ocfs2_xattr_shrink_size(inode,
 449                                               old_clusters, new_clusters,
 450                                               root_bh, xv);
 451
 452         return ret;
 453 }
 454
 455 static int ocfs2_xattr_list_entry(char *buffer, size_t size,
 456                                   size_t *result, const char *prefix,
 457                                   const char *name, int name_len)
 458 {
 459         char *p = buffer + *result;
 460         int prefix_len = strlen(prefix);
 461         int total_len = prefix_len + name_len + 1;
 462
 463         *result += total_len;
 464
 465         /* we are just looking for how big our buffer needs to be */
 466         if (!size)
 467                 return 0;
 468
 469         if (*result > size)
 470                 return -ERANGE;
 471
 472         memcpy(p, prefix, prefix_len);
 473         memcpy(p + prefix_len, name, name_len);
 474         p[prefix_len + name_len] = '\0';
 475
 476         return 0;
 477 }
 478
 479 static int ocfs2_xattr_list_entries(struct inode *inode,
 480                                     struct ocfs2_xattr_header *header,
 481                                     char *buffer, size_t buffer_size)
 482 {
 483         size_t result = 0;
 484         int i, type, ret;
 485         const char *prefix, *name;
 486
 487         for (i = 0 ; i < le16_to_cpu(header->xh_count); i++) {
 488                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
 489                 type = ocfs2_xattr_get_type(entry);
 490                 prefix = ocfs2_xattr_prefix(type);
 491
 492                 if (prefix) {
 493                         name = (const char *)header +
 494                                 le16_to_cpu(entry->xe_name_offset);
 495
 496                         ret = ocfs2_xattr_list_entry(buffer, buffer_size,
 497                                                      &result, prefix, name,
 498                                                      entry->xe_name_len);
 499                         if (ret)
 500                                 return ret;
 501                 }
 502         }
 503
 504         return result;
 505 }
 506
 507 static int ocfs2_xattr_ibody_list(struct inode *inode,
 508                                   struct ocfs2_dinode *di,
 509                                   char *buffer,
 510                                   size_t buffer_size)
 511 {
 512         struct ocfs2_xattr_header *header = NULL;
 513         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 514         int ret = 0;
 515
 516         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 517                 return ret;
 518
 519         header = (struct ocfs2_xattr_header *)
 520                  ((void *)di + inode->i_sb->s_blocksize -
 521                  le16_to_cpu(di->i_xattr_inline_size));
 522
 523         ret = ocfs2_xattr_list_entries(inode, header, buffer, buffer_size);
 524
 525         return ret;
 526 }
 527
 528 static int ocfs2_xattr_block_list(struct inode *inode,
 529                                   struct ocfs2_dinode *di,
 530                                   char *buffer,
 531                                   size_t buffer_size)
 532 {
 533         struct buffer_head *blk_bh = NULL;
 534         struct ocfs2_xattr_block *xb;
 535         int ret = 0;
 536
 537         if (!di->i_xattr_loc)
 538                 return ret;
 539
 540         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 541                                le64_to_cpu(di->i_xattr_loc),
 542                                &blk_bh, OCFS2_BH_CACHED, inode);
 543         if (ret < 0) {
 544                 mlog_errno(ret);
 545                 return ret;
 546         }
 547         /*Verify the signature of xattr block*/
 548         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 549                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 550                 ret = -EFAULT;
 551                 goto cleanup;
 552         }
 553
 554         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 555
 556         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 557                 struct ocfs2_xattr_header *header = &xb->xb_attrs.xb_header;
 558                 ret = ocfs2_xattr_list_entries(inode, header,
 559                                                buffer, buffer_size);
 560         } else {
 561                 struct ocfs2_xattr_tree_root *xt = &xb->xb_attrs.xb_root;
 562                 ret = ocfs2_xattr_tree_list_index_block(inode, xt,
 563                                                    buffer, buffer_size);
 564         }
 565 cleanup:
 566         brelse(blk_bh);
 567
 568         return ret;
 569 }
 570
 571 ssize_t ocfs2_listxattr(struct dentry *dentry,
 572                         char *buffer,
 573                         size_t size)
 574 {
 575         int ret = 0, i_ret = 0, b_ret = 0;
 576         struct buffer_head *di_bh = NULL;
 577         struct ocfs2_dinode *di = NULL;
 578         struct ocfs2_inode_info *oi = OCFS2_I(dentry->d_inode);
 579
 580         if (!ocfs2_supports_xattr(OCFS2_SB(dentry->d_sb)))
 581                 return -EOPNOTSUPP;
 582
 583         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 584                 return ret;
 585
 586         ret = ocfs2_inode_lock(dentry->d_inode, &di_bh, 0);
 587         if (ret < 0) {
 588                 mlog_errno(ret);
 589                 return ret;
 590         }
 591
 592         di = (struct ocfs2_dinode *)di_bh->b_data;
 593
 594         down_read(&oi->ip_xattr_sem);
 595         i_ret = ocfs2_xattr_ibody_list(dentry->d_inode, di, buffer, size);
 596         if (i_ret < 0)
 597                 b_ret = 0;
 598         else {
 599                 if (buffer) {
 600                         buffer += i_ret;
 601                         size -= i_ret;
 602                 }
 603                 b_ret = ocfs2_xattr_block_list(dentry->d_inode, di,
 604                                                buffer, size);
 605                 if (b_ret < 0)
 606                         i_ret = 0;
 607         }
 608         up_read(&oi->ip_xattr_sem);
 609         ocfs2_inode_unlock(dentry->d_inode, 0);
 610
 611         brelse(di_bh);
 612
 613         return i_ret + b_ret;
 614 }
 615
 616 static int ocfs2_xattr_find_entry(int name_index,
 617                                   const char *name,
 618                                   struct ocfs2_xattr_search *xs)
 619 {
 620         struct ocfs2_xattr_entry *entry;
 621         size_t name_len;
 622         int i, cmp = 1;
 623
 624         if (name == NULL)
 625                 return -EINVAL;
 626
 627         name_len = strlen(name);
 628         entry = xs->here;
 629         for (i = 0; i < le16_to_cpu(xs->header->xh_count); i++) {
 630                 cmp = name_index - ocfs2_xattr_get_type(entry);
 631                 if (!cmp)
 632                         cmp = name_len - entry->xe_name_len;
 633                 if (!cmp)
 634                         cmp = memcmp(name, (xs->base +
 635                                      le16_to_cpu(entry->xe_name_offset)),
 636                                      name_len);
 637                 if (cmp == 0)
 638                         break;
 639                 entry += 1;
 640         }
 641         xs->here = entry;
 642
 643         return cmp ? -ENODATA : 0;
 644 }
 645
 646 static int ocfs2_xattr_get_value_outside(struct inode *inode,
 647                                          struct ocfs2_xattr_value_root *xv,
 648                                          void *buffer,
 649                                          size_t len)
 650 {
 651         u32 cpos, p_cluster, num_clusters, bpc, clusters;
 652         u64 blkno;
 653         int i, ret = 0;
 654         size_t cplen, blocksize;
 655         struct buffer_head *bh = NULL;
 656         struct ocfs2_extent_list *el;
 657
 658         el = &xv->xr_list;
 659         clusters = le32_to_cpu(xv->xr_clusters);
 660         bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 661         blocksize = inode->i_sb->s_blocksize;
 662
 663         cpos = 0;
 664         while (cpos < clusters) {
 665                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 666                                                &num_clusters, el);
 667                 if (ret) {
 668                         mlog_errno(ret);
 669                         goto out;
 670                 }
 671
 672                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 673                 /* Copy ocfs2_xattr_value */
 674                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 675                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 676                                                &bh, OCFS2_BH_CACHED, inode);
 677                         if (ret) {
 678                                 mlog_errno(ret);
 679                                 goto out;
 680                         }
 681
 682                         cplen = len >= blocksize ? blocksize : len;
 683                         memcpy(buffer, bh->b_data, cplen);
 684                         len -= cplen;
 685                         buffer += cplen;
 686
 687                         brelse(bh);
 688                         bh = NULL;
 689                         if (len == 0)
 690                                 break;
 691                 }
 692                 cpos += num_clusters;
 693         }
 694 out:
 695         return ret;
 696 }
 697
 698 static int ocfs2_xattr_ibody_get(struct inode *inode,
 699                                  int name_index,
 700                                  const char *name,
 701                                  void *buffer,
 702                                  size_t buffer_size,
 703                                  struct ocfs2_xattr_search *xs)
 704 {
 705         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 706         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 707         struct ocfs2_xattr_value_root *xv;
 708         size_t size;
 709         int ret = 0;
 710
 711         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL))
 712                 return -ENODATA;
 713
 714         xs->end = (void *)di + inode->i_sb->s_blocksize;
 715         xs->header = (struct ocfs2_xattr_header *)
 716                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
 717         xs->base = (void *)xs->header;
 718         xs->here = xs->header->xh_entries;
 719
 720         ret = ocfs2_xattr_find_entry(name_index, name, xs);
 721         if (ret)
 722                 return ret;
 723         size = le64_to_cpu(xs->here->xe_value_size);
 724         if (buffer) {
 725                 if (size > buffer_size)
 726                         return -ERANGE;
 727                 if (ocfs2_xattr_is_local(xs->here)) {
 728                         memcpy(buffer, (void *)xs->base +
 729                                le16_to_cpu(xs->here->xe_name_offset) +
 730                                OCFS2_XATTR_SIZE(xs->here->xe_name_len), size);
 731                 } else {
 732                         xv = (struct ocfs2_xattr_value_root *)
 733                                 (xs->base + le16_to_cpu(
 734                                  xs->here->xe_name_offset) +
 735                                 OCFS2_XATTR_SIZE(xs->here->xe_name_len));
 736                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 737                                                             buffer, size);
 738                         if (ret < 0) {
 739                                 mlog_errno(ret);
 740                                 return ret;
 741                         }
 742                 }
 743         }
 744
 745         return size;
 746 }
 747
 748 static int ocfs2_xattr_block_get(struct inode *inode,
 749                                  int name_index,
 750                                  const char *name,
 751                                  void *buffer,
 752                                  size_t buffer_size,
 753                                  struct ocfs2_xattr_search *xs)
 754 {
 755         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
 756         struct buffer_head *blk_bh = NULL;
 757         struct ocfs2_xattr_block *xb;
 758         struct ocfs2_xattr_value_root *xv;
 759         size_t size;
 760         int ret = -ENODATA, name_offset, name_len, block_off, i;
 761
 762         if (!di->i_xattr_loc)
 763                 return ret;
 764
 765         memset(&xs->bucket, 0, sizeof(xs->bucket));
 766
 767         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
 768                                le64_to_cpu(di->i_xattr_loc),
 769                                &blk_bh, OCFS2_BH_CACHED, inode);
 770         if (ret < 0) {
 771                 mlog_errno(ret);
 772                 return ret;
 773         }
 774         /*Verify the signature of xattr block*/
 775         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
 776                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
 777                 ret = -EFAULT;
 778                 goto cleanup;
 779         }
 780
 781         xs->xattr_bh = blk_bh;
 782         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
 783
 784         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
 785                 xs->header = &xb->xb_attrs.xb_header;
 786                 xs->base = (void *)xs->header;
 787                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
 788                 xs->here = xs->header->xh_entries;
 789
 790                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
 791         } else
 792                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
 793                                                    name_index,
 794                                                    name, xs);
 795
 796         if (ret)
 797                 goto cleanup;
 798         size = le64_to_cpu(xs->here->xe_value_size);
 799         if (buffer) {
 800                 ret = -ERANGE;
 801                 if (size > buffer_size)
 802                         goto cleanup;
 803
 804                 name_offset = le16_to_cpu(xs->here->xe_name_offset);
 805                 name_len = OCFS2_XATTR_SIZE(xs->here->xe_name_len);
 806                 i = xs->here - xs->header->xh_entries;
 807
 808                 if (le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED) {
 809                         ret = ocfs2_xattr_bucket_get_name_value(inode,
 810                                                                 xs->bucket.xh,
 811                                                                 i,
 812                                                                 &block_off,
 813                                                                 &name_offset);
 814                         xs->base = xs->bucket.bhs[block_off]->b_data;
 815                 }
 816                 if (ocfs2_xattr_is_local(xs->here)) {
 817                         memcpy(buffer, (void *)xs->base +
 818                                name_offset + name_len, size);
 819                 } else {
 820                         xv = (struct ocfs2_xattr_value_root *)
 821                                 (xs->base + name_offset + name_len);
 822                         ret = ocfs2_xattr_get_value_outside(inode, xv,
 823                                                             buffer, size);
 824                         if (ret < 0) {
 825                                 mlog_errno(ret);
 826                                 goto cleanup;
 827                         }
 828                 }
 829         }
 830         ret = size;
 831 cleanup:
 832         for (i = 0; i < OCFS2_XATTR_MAX_BLOCKS_PER_BUCKET; i++)
 833                 brelse(xs->bucket.bhs[i]);
 834         memset(&xs->bucket, 0, sizeof(xs->bucket));
 835
 836         brelse(blk_bh);
 837         return ret;
 838 }
 839
 840 /* ocfs2_xattr_get()
 841  *
 842  * Copy an extended attribute into the buffer provided.
 843  * Buffer is NULL to compute the size of buffer required.
 844  */
 845 int ocfs2_xattr_get(struct inode *inode,
 846                     int name_index,
 847                     const char *name,
 848                     void *buffer,
 849                     size_t buffer_size)
 850 {
 851         int ret;
 852         struct ocfs2_dinode *di = NULL;
 853         struct buffer_head *di_bh = NULL;
 854         struct ocfs2_inode_info *oi = OCFS2_I(inode);
 855         struct ocfs2_xattr_search xis = {
 856                 .not_found = -ENODATA,
 857         };
 858         struct ocfs2_xattr_search xbs = {
 859                 .not_found = -ENODATA,
 860         };
 861
 862         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
 863                 return -EOPNOTSUPP;
 864
 865         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
 866                 ret = -ENODATA;
 867
 868         ret = ocfs2_inode_lock(inode, &di_bh, 0);
 869         if (ret < 0) {
 870                 mlog_errno(ret);
 871                 return ret;
 872         }
 873         xis.inode_bh = xbs.inode_bh = di_bh;
 874         di = (struct ocfs2_dinode *)di_bh->b_data;
 875
 876         down_read(&oi->ip_xattr_sem);
 877         ret = ocfs2_xattr_ibody_get(inode, name_index, name, buffer,
 878                                     buffer_size, &xis);
 879         if (ret == -ENODATA)
 880                 ret = ocfs2_xattr_block_get(inode, name_index, name, buffer,
 881                                             buffer_size, &xbs);
 882         up_read(&oi->ip_xattr_sem);
 883         ocfs2_inode_unlock(inode, 0);
 884
 885         brelse(di_bh);
 886
 887         return ret;
 888 }
 889
 890 static int __ocfs2_xattr_set_value_outside(struct inode *inode,
 891                                            struct ocfs2_xattr_value_root *xv,
 892                                            const void *value,
 893                                            int value_len)
 894 {
 895         int ret = 0, i, cp_len, credits;
 896         u16 blocksize = inode->i_sb->s_blocksize;
 897         u32 p_cluster, num_clusters;
 898         u32 cpos = 0, bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
 899         u32 clusters = ocfs2_clusters_for_bytes(inode->i_sb, value_len);
 900         u64 blkno;
 901         struct buffer_head *bh = NULL;
 902         handle_t *handle;
 903
 904         BUG_ON(clusters > le32_to_cpu(xv->xr_clusters));
 905
 906         credits = clusters * bpc;
 907         handle = ocfs2_start_trans(OCFS2_SB(inode->i_sb), credits);
 908         if (IS_ERR(handle)) {
 909                 ret = PTR_ERR(handle);
 910                 mlog_errno(ret);
 911                 goto out;
 912         }
 913
 914         while (cpos < clusters) {
 915                 ret = ocfs2_xattr_get_clusters(inode, cpos, &p_cluster,
 916                                                &num_clusters, &xv->xr_list);
 917                 if (ret) {
 918                         mlog_errno(ret);
 919                         goto out_commit;
 920                 }
 921
 922                 blkno = ocfs2_clusters_to_blocks(inode->i_sb, p_cluster);
 923
 924                 for (i = 0; i < num_clusters * bpc; i++, blkno++) {
 925                         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
 926                                                &bh, OCFS2_BH_CACHED, inode);
 927                         if (ret) {
 928                                 mlog_errno(ret);
 929                                 goto out_commit;
 930                         }
 931
 932                         ret = ocfs2_journal_access(handle,
 933                                                    inode,
 934                                                    bh,
 935                                                    OCFS2_JOURNAL_ACCESS_WRITE);
 936                         if (ret < 0) {
 937                                 mlog_errno(ret);
 938                                 goto out_commit;
 939                         }
 940
 941                         cp_len = value_len > blocksize ? blocksize : value_len;
 942                         memcpy(bh->b_data, value, cp_len);
 943                         value_len -= cp_len;
 944                         value += cp_len;
 945                         if (cp_len < blocksize)
 946                                 memset(bh->b_data + cp_len, 0,
 947                                        blocksize - cp_len);
 948
 949                         ret = ocfs2_journal_dirty(handle, bh);
 950                         if (ret < 0) {
 951                                 mlog_errno(ret);
 952                                 goto out_commit;
 953                         }
 954                         brelse(bh);
 955                         bh = NULL;
 956
 957                         /*
 958                          * XXX: do we need to empty all the following
 959                          * blocks in this cluster?
 960                          */
 961                         if (!value_len)
 962                                 break;
 963                 }
 964                 cpos += num_clusters;
 965         }
 966 out_commit:
 967         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
 968 out:
 969         brelse(bh);
 970
 971         return ret;
 972 }
 973
 974 static int ocfs2_xattr_cleanup(struct inode *inode,
 975                                struct ocfs2_xattr_info *xi,
 976                                struct ocfs2_xattr_search *xs,
 977                                size_t offs)
 978 {
 979         handle_t *handle = NULL;
 980         int ret = 0;
 981         size_t name_len = strlen(xi->name);
 982         void *val = xs->base + offs;
 983         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
 984
 985         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
 986                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
 987         if (IS_ERR(handle)) {
 988                 ret = PTR_ERR(handle);
 989                 mlog_errno(ret);
 990                 goto out;
 991         }
 992         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
 993                                    OCFS2_JOURNAL_ACCESS_WRITE);
 994         if (ret) {
 995                 mlog_errno(ret);
 996                 goto out_commit;
 997         }
 998         /* Decrease xattr count */
 999         le16_add_cpu(&xs->header->xh_count, -1);
1000         /* Remove the xattr entry and tree root which has already be set*/
1001         memset((void *)xs->here, 0, sizeof(struct ocfs2_xattr_entry));
1002         memset(val, 0, size);
1003
1004         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1005         if (ret < 0)
1006                 mlog_errno(ret);
1007 out_commit:
1008         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1009 out:
1010         return ret;
1011 }
1012
1013 static int ocfs2_xattr_update_entry(struct inode *inode,
1014                                     struct ocfs2_xattr_info *xi,
1015                                     struct ocfs2_xattr_search *xs,
1016                                     size_t offs)
1017 {
1018         handle_t *handle = NULL;
1019         int ret = 0;
1020
1021         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1022                                    OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1023         if (IS_ERR(handle)) {
1024                 ret = PTR_ERR(handle);
1025                 mlog_errno(ret);
1026                 goto out;
1027         }
1028         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1029                                    OCFS2_JOURNAL_ACCESS_WRITE);
1030         if (ret) {
1031                 mlog_errno(ret);
1032                 goto out_commit;
1033         }
1034
1035         xs->here->xe_name_offset = cpu_to_le16(offs);
1036         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1037         if (xi->value_len <= OCFS2_XATTR_INLINE_SIZE)
1038                 ocfs2_xattr_set_local(xs->here, 1);
1039         else
1040                 ocfs2_xattr_set_local(xs->here, 0);
1041         ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1042
1043         ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1044         if (ret < 0)
1045                 mlog_errno(ret);
1046 out_commit:
1047         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1048 out:
1049         return ret;
1050 }
1051
1052 /*
1053  * ocfs2_xattr_set_value_outside()
1054  *
1055  * Set large size value in B tree.
1056  */
1057 static int ocfs2_xattr_set_value_outside(struct inode *inode,
1058                                          struct ocfs2_xattr_info *xi,
1059                                          struct ocfs2_xattr_search *xs,
1060                                          size_t offs)
1061 {
1062         size_t name_len = strlen(xi->name);
1063         void *val = xs->base + offs;
1064         struct ocfs2_xattr_value_root *xv = NULL;
1065         size_t size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1066         int ret = 0;
1067
1068         memset(val, 0, size);
1069         memcpy(val, xi->name, name_len);
1070         xv = (struct ocfs2_xattr_value_root *)
1071                 (val + OCFS2_XATTR_SIZE(name_len));
1072         xv->xr_clusters = 0;
1073         xv->xr_last_eb_blk = 0;
1074         xv->xr_list.l_tree_depth = 0;
1075         xv->xr_list.l_count = cpu_to_le16(1);
1076         xv->xr_list.l_next_free_rec = 0;
1077
1078         ret = ocfs2_xattr_value_truncate(inode, xs->xattr_bh, xv,
1079                                          xi->value_len);
1080         if (ret < 0) {
1081                 mlog_errno(ret);
1082                 return ret;
1083         }
1084         ret = __ocfs2_xattr_set_value_outside(inode, xv, xi->value,
1085                                               xi->value_len);
1086         if (ret < 0) {
1087                 mlog_errno(ret);
1088                 return ret;
1089         }
1090         ret = ocfs2_xattr_update_entry(inode, xi, xs, offs);
1091         if (ret < 0)
1092                 mlog_errno(ret);
1093
1094         return ret;
1095 }
1096
1097 /*
1098  * ocfs2_xattr_set_entry_local()
1099  *
1100  * Set, replace or remove extended attribute in local.
1101  */
1102 static void ocfs2_xattr_set_entry_local(struct inode *inode,
1103                                         struct ocfs2_xattr_info *xi,
1104                                         struct ocfs2_xattr_search *xs,
1105                                         struct ocfs2_xattr_entry *last,
1106                                         size_t min_offs)
1107 {
1108         size_t name_len = strlen(xi->name);
1109         int i;
1110
1111         if (xi->value && xs->not_found) {
1112                 /* Insert the new xattr entry. */
1113                 le16_add_cpu(&xs->header->xh_count, 1);
1114                 ocfs2_xattr_set_type(last, xi->name_index);
1115                 ocfs2_xattr_set_local(last, 1);
1116                 last->xe_name_len = name_len;
1117         } else {
1118                 void *first_val;
1119                 void *val;
1120                 size_t offs, size;
1121
1122                 first_val = xs->base + min_offs;
1123                 offs = le16_to_cpu(xs->here->xe_name_offset);
1124                 val = xs->base + offs;
1125
1126                 if (le64_to_cpu(xs->here->xe_value_size) >
1127                     OCFS2_XATTR_INLINE_SIZE)
1128                         size = OCFS2_XATTR_SIZE(name_len) +
1129                                 OCFS2_XATTR_ROOT_SIZE;
1130                 else
1131                         size = OCFS2_XATTR_SIZE(name_len) +
1132                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1133
1134                 if (xi->value && size == OCFS2_XATTR_SIZE(name_len) +
1135                                 OCFS2_XATTR_SIZE(xi->value_len)) {
1136                         /* The old and the new value have the
1137                            same size. Just replace the value. */
1138                         ocfs2_xattr_set_local(xs->here, 1);
1139                         xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1140                         /* Clear value bytes. */
1141                         memset(val + OCFS2_XATTR_SIZE(name_len),
1142                                0,
1143                                OCFS2_XATTR_SIZE(xi->value_len));
1144                         memcpy(val + OCFS2_XATTR_SIZE(name_len),
1145                                xi->value,
1146                                xi->value_len);
1147                         return;
1148                 }
1149                 /* Remove the old name+value. */
1150                 memmove(first_val + size, first_val, val - first_val);
1151                 memset(first_val, 0, size);
1152                 xs->here->xe_name_hash = 0;
1153                 xs->here->xe_name_offset = 0;
1154                 ocfs2_xattr_set_local(xs->here, 1);
1155                 xs->here->xe_value_size = 0;
1156
1157                 min_offs += size;
1158
1159                 /* Adjust all value offsets. */
1160                 last = xs->header->xh_entries;
1161                 for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1162                         size_t o = le16_to_cpu(last->xe_name_offset);
1163
1164                         if (o < offs)
1165                                 last->xe_name_offset = cpu_to_le16(o + size);
1166                         last += 1;
1167                 }
1168
1169                 if (!xi->value) {
1170                         /* Remove the old entry. */
1171                         last -= 1;
1172                         memmove(xs->here, xs->here + 1,
1173                                 (void *)last - (void *)xs->here);
1174                         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
1175                         le16_add_cpu(&xs->header->xh_count, -1);
1176                 }
1177         }
1178         if (xi->value) {
1179                 /* Insert the new name+value. */
1180                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1181                                 OCFS2_XATTR_SIZE(xi->value_len);
1182                 void *val = xs->base + min_offs - size;
1183
1184                 xs->here->xe_name_offset = cpu_to_le16(min_offs - size);
1185                 memset(val, 0, size);
1186                 memcpy(val, xi->name, name_len);
1187                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
1188                        xi->value,
1189                        xi->value_len);
1190                 xs->here->xe_value_size = cpu_to_le64(xi->value_len);
1191                 ocfs2_xattr_set_local(xs->here, 1);
1192                 ocfs2_xattr_hash_entry(inode, xs->header, xs->here);
1193         }
1194
1195         return;
1196 }
1197
1198 /*
1199  * ocfs2_xattr_set_entry()
1200  *
1201  * Set extended attribute entry into inode or block.
1202  *
1203  * If extended attribute value size > OCFS2_XATTR_INLINE_SIZE,
1204  * We first insert tree root(ocfs2_xattr_value_root) with set_entry_local(),
1205  * then set value in B tree with set_value_outside().
1206  */
1207 static int ocfs2_xattr_set_entry(struct inode *inode,
1208                                  struct ocfs2_xattr_info *xi,
1209                                  struct ocfs2_xattr_search *xs,
1210                                  int flag)
1211 {
1212         struct ocfs2_xattr_entry *last;
1213         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1214         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1215         size_t min_offs = xs->end - xs->base, name_len = strlen(xi->name);
1216         size_t size_l = 0;
1217         handle_t *handle = NULL;
1218         int free, i, ret;
1219         struct ocfs2_xattr_info xi_l = {
1220                 .name_index = xi->name_index,
1221                 .name = xi->name,
1222                 .value = xi->value,
1223                 .value_len = xi->value_len,
1224         };
1225
1226         /* Compute min_offs, last and free space. */
1227         last = xs->header->xh_entries;
1228
1229         for (i = 0 ; i < le16_to_cpu(xs->header->xh_count); i++) {
1230                 size_t offs = le16_to_cpu(last->xe_name_offset);
1231                 if (offs < min_offs)
1232                         min_offs = offs;
1233                 last += 1;
1234         }
1235
1236         free = min_offs - ((void *)last - xs->base) - sizeof(__u32);
1237         if (free < 0)
1238                 return -EFAULT;
1239
1240         if (!xs->not_found) {
1241                 size_t size = 0;
1242                 if (ocfs2_xattr_is_local(xs->here))
1243                         size = OCFS2_XATTR_SIZE(name_len) +
1244                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1245                 else
1246                         size = OCFS2_XATTR_SIZE(name_len) +
1247                                 OCFS2_XATTR_ROOT_SIZE;
1248                 free += (size + sizeof(struct ocfs2_xattr_entry));
1249         }
1250         /* Check free space in inode or block */
1251         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1252                 if (free < sizeof(struct ocfs2_xattr_entry) +
1253                            OCFS2_XATTR_SIZE(name_len) +
1254                            OCFS2_XATTR_ROOT_SIZE) {
1255                         ret = -ENOSPC;
1256                         goto out;
1257                 }
1258                 size_l = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_ROOT_SIZE;
1259                 xi_l.value = (void *)&def_xv;
1260                 xi_l.value_len = OCFS2_XATTR_ROOT_SIZE;
1261         } else if (xi->value) {
1262                 if (free < sizeof(struct ocfs2_xattr_entry) +
1263                            OCFS2_XATTR_SIZE(name_len) +
1264                            OCFS2_XATTR_SIZE(xi->value_len)) {
1265                         ret = -ENOSPC;
1266                         goto out;
1267                 }
1268         }
1269
1270         if (!xs->not_found) {
1271                 /* For existing extended attribute */
1272                 size_t size = OCFS2_XATTR_SIZE(name_len) +
1273                         OCFS2_XATTR_SIZE(le64_to_cpu(xs->here->xe_value_size));
1274                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1275                 void *val = xs->base + offs;
1276
1277                 if (ocfs2_xattr_is_local(xs->here) && size == size_l) {
1278                         /* Replace existing local xattr with tree root */
1279                         ret = ocfs2_xattr_set_value_outside(inode, xi, xs,
1280                                                             offs);
1281                         if (ret < 0)
1282                                 mlog_errno(ret);
1283                         goto out;
1284                 } else if (!ocfs2_xattr_is_local(xs->here)) {
1285                         /* For existing xattr which has value outside */
1286                         struct ocfs2_xattr_value_root *xv = NULL;
1287                         xv = (struct ocfs2_xattr_value_root *)(val +
1288                                 OCFS2_XATTR_SIZE(name_len));
1289
1290                         if (xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1291                                 /*
1292                                  * If new value need set outside also,
1293                                  * first truncate old value to new value,
1294                                  * then set new value with set_value_outside().
1295                                  */
1296                                 ret = ocfs2_xattr_value_truncate(inode,
1297                                                                  xs->xattr_bh,
1298                                                                  xv,
1299                                                                  xi->value_len);
1300                                 if (ret < 0) {
1301                                         mlog_errno(ret);
1302                                         goto out;
1303                                 }
1304
1305                                 ret = __ocfs2_xattr_set_value_outside(inode,
1306                                                                 xv,
1307                                                                 xi->value,
1308                                                                 xi->value_len);
1309                                 if (ret < 0) {
1310                                         mlog_errno(ret);
1311                                         goto out;
1312                                 }
1313
1314                                 ret = ocfs2_xattr_update_entry(inode,
1315                                                                xi,
1316                                                                xs,
1317                                                                offs);
1318                                 if (ret < 0)
1319                                         mlog_errno(ret);
1320                                 goto out;
1321                         } else {
1322                                 /*
1323                                  * If new value need set in local,
1324                                  * just trucate old value to zero.
1325                                  */
1326                                  ret = ocfs2_xattr_value_truncate(inode,
1327                                                                  xs->xattr_bh,
1328                                                                  xv,
1329                                                                  0);
1330                                 if (ret < 0)
1331                                         mlog_errno(ret);
1332                         }
1333                 }
1334         }
1335
1336         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1337                                    OCFS2_INODE_UPDATE_CREDITS);
1338         if (IS_ERR(handle)) {
1339                 ret = PTR_ERR(handle);
1340                 mlog_errno(ret);
1341                 goto out;
1342         }
1343
1344         ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1345                                    OCFS2_JOURNAL_ACCESS_WRITE);
1346         if (ret) {
1347                 mlog_errno(ret);
1348                 goto out_commit;
1349         }
1350
1351         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1352                 /* set extended attribute in external block. */
1353                 ret = ocfs2_extend_trans(handle,
1354                                          OCFS2_INODE_UPDATE_CREDITS +
1355                                          OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1356                 if (ret) {
1357                         mlog_errno(ret);
1358                         goto out_commit;
1359                 }
1360                 ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1361                                            OCFS2_JOURNAL_ACCESS_WRITE);
1362                 if (ret) {
1363                         mlog_errno(ret);
1364                         goto out_commit;
1365                 }
1366         }
1367
1368         /*
1369          * Set value in local, include set tree root in local.
1370          * This is the first step for value size >INLINE_SIZE.
1371          */
1372         ocfs2_xattr_set_entry_local(inode, &xi_l, xs, last, min_offs);
1373
1374         if (!(flag & OCFS2_INLINE_XATTR_FL)) {
1375                 ret = ocfs2_journal_dirty(handle, xs->xattr_bh);
1376                 if (ret < 0) {
1377                         mlog_errno(ret);
1378                         goto out_commit;
1379                 }
1380         }
1381
1382         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) &&
1383             (flag & OCFS2_INLINE_XATTR_FL)) {
1384                 struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1385                 unsigned int xattrsize = osb->s_xattr_inline_size;
1386
1387                 /*
1388                  * Adjust extent record count or inline data size
1389                  * to reserve space for extended attribute.
1390                  */
1391                 if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1392                         struct ocfs2_inline_data *idata = &di->id2.i_data;
1393                         le16_add_cpu(&idata->id_count, -xattrsize);
1394                 } else if (!(ocfs2_inode_is_fast_symlink(inode))) {
1395                         struct ocfs2_extent_list *el = &di->id2.i_list;
1396                         le16_add_cpu(&el->l_count, -(xattrsize /
1397                                         sizeof(struct ocfs2_extent_rec)));
1398                 }
1399                 di->i_xattr_inline_size = cpu_to_le16(xattrsize);
1400         }
1401         /* Update xattr flag */
1402         spin_lock(&oi->ip_lock);
1403         oi->ip_dyn_features |= flag;
1404         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1405         spin_unlock(&oi->ip_lock);
1406         /* Update inode ctime */
1407         inode->i_ctime = CURRENT_TIME;
1408         di->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
1409         di->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
1410
1411         ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1412         if (ret < 0)
1413                 mlog_errno(ret);
1414
1415 out_commit:
1416         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1417
1418         if (!ret && xi->value_len > OCFS2_XATTR_INLINE_SIZE) {
1419                 /*
1420                  * Set value outside in B tree.
1421                  * This is the second step for value size > INLINE_SIZE.
1422                  */
1423                 size_t offs = le16_to_cpu(xs->here->xe_name_offset);
1424                 ret = ocfs2_xattr_set_value_outside(inode, xi, xs, offs);
1425                 if (ret < 0) {
1426                         int ret2;
1427
1428                         mlog_errno(ret);
1429                         /*
1430                          * If set value outside failed, we have to clean
1431                          * the junk tree root we have already set in local.
1432                          */
1433                         ret2 = ocfs2_xattr_cleanup(inode, xi, xs, offs);
1434                         if (ret2 < 0)
1435                                 mlog_errno(ret2);
1436                 }
1437         }
1438 out:
1439         return ret;
1440
1441 }
1442
1443 static int ocfs2_remove_value_outside(struct inode*inode,
1444                                       struct buffer_head *bh,
1445                                       struct ocfs2_xattr_header *header)
1446 {
1447         int ret = 0, i;
1448
1449         for (i = 0; i < le16_to_cpu(header->xh_count); i++) {
1450                 struct ocfs2_xattr_entry *entry = &header->xh_entries[i];
1451
1452                 if (!ocfs2_xattr_is_local(entry)) {
1453                         struct ocfs2_xattr_value_root *xv;
1454                         void *val;
1455
1456                         val = (void *)header +
1457                                 le16_to_cpu(entry->xe_name_offset);
1458                         xv = (struct ocfs2_xattr_value_root *)
1459                                 (val + OCFS2_XATTR_SIZE(entry->xe_name_len));
1460                         ret = ocfs2_xattr_value_truncate(inode, bh, xv, 0);
1461                         if (ret < 0) {
1462                                 mlog_errno(ret);
1463                                 return ret;
1464                         }
1465                 }
1466         }
1467
1468         return ret;
1469 }
1470
1471 static int ocfs2_xattr_ibody_remove(struct inode *inode,
1472                                     struct buffer_head *di_bh)
1473 {
1474
1475         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1476         struct ocfs2_xattr_header *header;
1477         int ret;
1478
1479         header = (struct ocfs2_xattr_header *)
1480                  ((void *)di + inode->i_sb->s_blocksize -
1481                  le16_to_cpu(di->i_xattr_inline_size));
1482
1483         ret = ocfs2_remove_value_outside(inode, di_bh, header);
1484
1485         return ret;
1486 }
1487
1488 static int ocfs2_xattr_block_remove(struct inode *inode,
1489                                     struct buffer_head *blk_bh)
1490 {
1491         struct ocfs2_xattr_block *xb;
1492         int ret = 0;
1493
1494         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1495         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1496                 struct ocfs2_xattr_header *header = &(xb->xb_attrs.xb_header);
1497                 ret = ocfs2_remove_value_outside(inode, blk_bh, header);
1498         } else
1499                 ret = ocfs2_delete_xattr_index_block(inode, blk_bh);
1500
1501         return ret;
1502 }
1503
1504 static int ocfs2_xattr_free_block(struct inode *inode,
1505                                   u64 block)
1506 {
1507         struct inode *xb_alloc_inode;
1508         struct buffer_head *xb_alloc_bh = NULL;
1509         struct buffer_head *blk_bh = NULL;
1510         struct ocfs2_xattr_block *xb;
1511         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1512         handle_t *handle;
1513         int ret = 0;
1514         u64 blk, bg_blkno;
1515         u16 bit;
1516
1517         ret = ocfs2_read_block(osb, block, &blk_bh,
1518                                OCFS2_BH_CACHED, inode);
1519         if (ret < 0) {
1520                 mlog_errno(ret);
1521                 goto out;
1522         }
1523
1524         /*Verify the signature of xattr block*/
1525         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1526                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1527                 ret = -EFAULT;
1528                 goto out;
1529         }
1530
1531         ret = ocfs2_xattr_block_remove(inode, blk_bh);
1532         if (ret < 0) {
1533                 mlog_errno(ret);
1534                 goto out;
1535         }
1536
1537         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1538         blk = le64_to_cpu(xb->xb_blkno);
1539         bit = le16_to_cpu(xb->xb_suballoc_bit);
1540         bg_blkno = ocfs2_which_suballoc_group(blk, bit);
1541
1542         xb_alloc_inode = ocfs2_get_system_file_inode(osb,
1543                                 EXTENT_ALLOC_SYSTEM_INODE,
1544                                 le16_to_cpu(xb->xb_suballoc_slot));
1545         if (!xb_alloc_inode) {
1546                 ret = -ENOMEM;
1547                 mlog_errno(ret);
1548                 goto out;
1549         }
1550         mutex_lock(&xb_alloc_inode->i_mutex);
1551
1552         ret = ocfs2_inode_lock(xb_alloc_inode, &xb_alloc_bh, 1);
1553         if (ret < 0) {
1554                 mlog_errno(ret);
1555                 goto out_mutex;
1556         }
1557
1558         handle = ocfs2_start_trans(osb, OCFS2_SUBALLOC_FREE);
1559         if (IS_ERR(handle)) {
1560                 ret = PTR_ERR(handle);
1561                 mlog_errno(ret);
1562                 goto out_unlock;
1563         }
1564
1565         ret = ocfs2_free_suballoc_bits(handle, xb_alloc_inode, xb_alloc_bh,
1566                                        bit, bg_blkno, 1);
1567         if (ret < 0)
1568                 mlog_errno(ret);
1569
1570         ocfs2_commit_trans(osb, handle);
1571 out_unlock:
1572         ocfs2_inode_unlock(xb_alloc_inode, 1);
1573         brelse(xb_alloc_bh);
1574 out_mutex:
1575         mutex_unlock(&xb_alloc_inode->i_mutex);
1576         iput(xb_alloc_inode);
1577 out:
1578         brelse(blk_bh);
1579         return ret;
1580 }
1581
1582 /*
1583  * ocfs2_xattr_remove()
1584  *
1585  * Free extended attribute resources associated with this inode.
1586  */
1587 int ocfs2_xattr_remove(struct inode *inode, struct buffer_head *di_bh)
1588 {
1589         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1590         struct ocfs2_dinode *di = (struct ocfs2_dinode *)di_bh->b_data;
1591         handle_t *handle;
1592         int ret;
1593
1594         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
1595                 return 0;
1596
1597         if (!(oi->ip_dyn_features & OCFS2_HAS_XATTR_FL))
1598                 return 0;
1599
1600         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1601                 ret = ocfs2_xattr_ibody_remove(inode, di_bh);
1602                 if (ret < 0) {
1603                         mlog_errno(ret);
1604                         goto out;
1605                 }
1606         }
1607
1608         if (di->i_xattr_loc) {
1609                 ret = ocfs2_xattr_free_block(inode,
1610                                              le64_to_cpu(di->i_xattr_loc));
1611                 if (ret < 0) {
1612                         mlog_errno(ret);
1613                         goto out;
1614                 }
1615         }
1616
1617         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)),
1618                                    OCFS2_INODE_UPDATE_CREDITS);
1619         if (IS_ERR(handle)) {
1620                 ret = PTR_ERR(handle);
1621                 mlog_errno(ret);
1622                 goto out;
1623         }
1624         ret = ocfs2_journal_access(handle, inode, di_bh,
1625                                    OCFS2_JOURNAL_ACCESS_WRITE);
1626         if (ret) {
1627                 mlog_errno(ret);
1628                 goto out_commit;
1629         }
1630
1631         di->i_xattr_loc = 0;
1632
1633         spin_lock(&oi->ip_lock);
1634         oi->ip_dyn_features &= ~(OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL);
1635         di->i_dyn_features = cpu_to_le16(oi->ip_dyn_features);
1636         spin_unlock(&oi->ip_lock);
1637
1638         ret = ocfs2_journal_dirty(handle, di_bh);
1639         if (ret < 0)
1640                 mlog_errno(ret);
1641 out_commit:
1642         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
1643 out:
1644         return ret;
1645 }
1646
1647 static int ocfs2_xattr_has_space_inline(struct inode *inode,
1648                                         struct ocfs2_dinode *di)
1649 {
1650         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1651         unsigned int xattrsize = OCFS2_SB(inode->i_sb)->s_xattr_inline_size;
1652         int free;
1653
1654         if (xattrsize < OCFS2_MIN_XATTR_INLINE_SIZE)
1655                 return 0;
1656
1657         if (oi->ip_dyn_features & OCFS2_INLINE_DATA_FL) {
1658                 struct ocfs2_inline_data *idata = &di->id2.i_data;
1659                 free = le16_to_cpu(idata->id_count) - le64_to_cpu(di->i_size);
1660         } else if (ocfs2_inode_is_fast_symlink(inode)) {
1661                 free = ocfs2_fast_symlink_chars(inode->i_sb) -
1662                         le64_to_cpu(di->i_size);
1663         } else {
1664                 struct ocfs2_extent_list *el = &di->id2.i_list;
1665                 free = (le16_to_cpu(el->l_count) -
1666                         le16_to_cpu(el->l_next_free_rec)) *
1667                         sizeof(struct ocfs2_extent_rec);
1668         }
1669         if (free >= xattrsize)
1670                 return 1;
1671
1672         return 0;
1673 }
1674
1675 /*
1676  * ocfs2_xattr_ibody_find()
1677  *
1678  * Find extended attribute in inode block and
1679  * fill search info into struct ocfs2_xattr_search.
1680  */
1681 static int ocfs2_xattr_ibody_find(struct inode *inode,
1682                                   int name_index,
1683                                   const char *name,
1684                                   struct ocfs2_xattr_search *xs)
1685 {
1686         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1687         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1688         int ret;
1689         int has_space = 0;
1690
1691         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1692                 return 0;
1693
1694         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1695                 down_read(&oi->ip_alloc_sem);
1696                 has_space = ocfs2_xattr_has_space_inline(inode, di);
1697                 up_read(&oi->ip_alloc_sem);
1698                 if (!has_space)
1699                         return 0;
1700         }
1701
1702         xs->xattr_bh = xs->inode_bh;
1703         xs->end = (void *)di + inode->i_sb->s_blocksize;
1704         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)
1705                 xs->header = (struct ocfs2_xattr_header *)
1706                         (xs->end - le16_to_cpu(di->i_xattr_inline_size));
1707         else
1708                 xs->header = (struct ocfs2_xattr_header *)
1709                         (xs->end - OCFS2_SB(inode->i_sb)->s_xattr_inline_size);
1710         xs->base = (void *)xs->header;
1711         xs->here = xs->header->xh_entries;
1712
1713         /* Find the named attribute. */
1714         if (oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL) {
1715                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1716                 if (ret && ret != -ENODATA)
1717                         return ret;
1718                 xs->not_found = ret;
1719         }
1720
1721         return 0;
1722 }
1723
1724 /*
1725  * ocfs2_xattr_ibody_set()
1726  *
1727  * Set, replace or remove an extended attribute into inode block.
1728  *
1729  */
1730 static int ocfs2_xattr_ibody_set(struct inode *inode,
1731                                  struct ocfs2_xattr_info *xi,
1732                                  struct ocfs2_xattr_search *xs)
1733 {
1734         struct ocfs2_inode_info *oi = OCFS2_I(inode);
1735         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1736         int ret;
1737
1738         if (inode->i_sb->s_blocksize == OCFS2_MIN_BLOCKSIZE)
1739                 return -ENOSPC;
1740
1741         down_write(&oi->ip_alloc_sem);
1742         if (!(oi->ip_dyn_features & OCFS2_INLINE_XATTR_FL)) {
1743                 if (!ocfs2_xattr_has_space_inline(inode, di)) {
1744                         ret = -ENOSPC;
1745                         goto out;
1746                 }
1747         }
1748
1749         ret = ocfs2_xattr_set_entry(inode, xi, xs,
1750                                 (OCFS2_INLINE_XATTR_FL | OCFS2_HAS_XATTR_FL));
1751 out:
1752         up_write(&oi->ip_alloc_sem);
1753
1754         return ret;
1755 }
1756
1757 /*
1758  * ocfs2_xattr_block_find()
1759  *
1760  * Find extended attribute in external block and
1761  * fill search info into struct ocfs2_xattr_search.
1762  */
1763 static int ocfs2_xattr_block_find(struct inode *inode,
1764                                   int name_index,
1765                                   const char *name,
1766                                   struct ocfs2_xattr_search *xs)
1767 {
1768         struct ocfs2_dinode *di = (struct ocfs2_dinode *)xs->inode_bh->b_data;
1769         struct buffer_head *blk_bh = NULL;
1770         struct ocfs2_xattr_block *xb;
1771         int ret = 0;
1772
1773         if (!di->i_xattr_loc)
1774                 return ret;
1775
1776         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
1777                                le64_to_cpu(di->i_xattr_loc),
1778                                &blk_bh, OCFS2_BH_CACHED, inode);
1779         if (ret < 0) {
1780                 mlog_errno(ret);
1781                 return ret;
1782         }
1783         /*Verify the signature of xattr block*/
1784         if (memcmp((void *)blk_bh->b_data, OCFS2_XATTR_BLOCK_SIGNATURE,
1785                    strlen(OCFS2_XATTR_BLOCK_SIGNATURE))) {
1786                         ret = -EFAULT;
1787                         goto cleanup;
1788         }
1789
1790         xs->xattr_bh = blk_bh;
1791         xb = (struct ocfs2_xattr_block *)blk_bh->b_data;
1792
1793         if (!(le16_to_cpu(xb->xb_flags) & OCFS2_XATTR_INDEXED)) {
1794                 xs->header = &xb->xb_attrs.xb_header;
1795                 xs->base = (void *)xs->header;
1796                 xs->end = (void *)(blk_bh->b_data) + blk_bh->b_size;
1797                 xs->here = xs->header->xh_entries;
1798
1799                 ret = ocfs2_xattr_find_entry(name_index, name, xs);
1800         } else
1801                 ret = ocfs2_xattr_index_block_find(inode, blk_bh,
1802                                                    name_index,
1803                                                    name, xs);
1804
1805         if (ret && ret != -ENODATA) {
1806                 xs->xattr_bh = NULL;
1807                 goto cleanup;
1808         }
1809         xs->not_found = ret;
1810         return 0;
1811 cleanup:
1812         brelse(blk_bh);
1813
1814         return ret;
1815 }
1816
1817 /*
1818  * When all the xattrs are deleted from index btree, the ocfs2_xattr_tree
1819  * will be erased and ocfs2_xattr_block will have its ocfs2_xattr_header
1820  * re-initialized.
1821  */
1822 static int ocfs2_restore_xattr_block(struct inode *inode,
1823                                      struct ocfs2_xattr_search *xs)
1824 {
1825         int ret;
1826         handle_t *handle;
1827         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1828         struct ocfs2_xattr_block *xb =
1829                 (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1830         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
1831         u16 xb_flags = le16_to_cpu(xb->xb_flags);
1832
1833         BUG_ON(!(xb_flags & OCFS2_XATTR_INDEXED) ||
1834                 le16_to_cpu(el->l_next_free_rec) != 0);
1835
1836         handle = ocfs2_start_trans(osb, OCFS2_XATTR_BLOCK_UPDATE_CREDITS);
1837         if (IS_ERR(handle)) {
1838                 ret = PTR_ERR(handle);
1839                 handle = NULL;
1840                 goto out;
1841         }
1842
1843         ret = ocfs2_journal_access(handle, inode, xs->xattr_bh,
1844                                    OCFS2_JOURNAL_ACCESS_WRITE);
1845         if (ret < 0) {
1846                 mlog_errno(ret);
1847                 goto out_commit;
1848         }
1849
1850         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
1851                offsetof(struct ocfs2_xattr_block, xb_attrs));
1852
1853         xb->xb_flags = cpu_to_le16(xb_flags & ~OCFS2_XATTR_INDEXED);
1854
1855         ocfs2_journal_dirty(handle, xs->xattr_bh);
1856
1857 out_commit:
1858         ocfs2_commit_trans(osb, handle);
1859 out:
1860         return ret;
1861 }
1862
1863 /*
1864  * ocfs2_xattr_block_set()
1865  *
1866  * Set, replace or remove an extended attribute into external block.
1867  *
1868  */
1869 static int ocfs2_xattr_block_set(struct inode *inode,
1870                                  struct ocfs2_xattr_info *xi,
1871                                  struct ocfs2_xattr_search *xs)
1872 {
1873         struct buffer_head *new_bh = NULL;
1874         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
1875         struct ocfs2_dinode *di =  (struct ocfs2_dinode *)xs->inode_bh->b_data;
1876         struct ocfs2_alloc_context *meta_ac = NULL;
1877         handle_t *handle = NULL;
1878         struct ocfs2_xattr_block *xblk = NULL;
1879         u16 suballoc_bit_start;
1880         u32 num_got;
1881         u64 first_blkno;
1882         int ret;
1883
1884         if (!xs->xattr_bh) {
1885                 /*
1886                  * Alloc one external block for extended attribute
1887                  * outside of inode.
1888                  */
1889                 ret = ocfs2_reserve_new_metadata_blocks(osb, 1, &meta_ac);
1890                 if (ret < 0) {
1891                         mlog_errno(ret);
1892                         goto out;
1893                 }
1894                 handle = ocfs2_start_trans(osb,
1895                                            OCFS2_XATTR_BLOCK_CREATE_CREDITS);
1896                 if (IS_ERR(handle)) {
1897                         ret = PTR_ERR(handle);
1898                         mlog_errno(ret);
1899                         goto out;
1900                 }
1901                 ret = ocfs2_journal_access(handle, inode, xs->inode_bh,
1902                                            OCFS2_JOURNAL_ACCESS_CREATE);
1903                 if (ret < 0) {
1904                         mlog_errno(ret);
1905                         goto out_commit;
1906                 }
1907
1908                 ret = ocfs2_claim_metadata(osb, handle, meta_ac, 1,
1909                                            &suballoc_bit_start, &num_got,
1910                                            &first_blkno);
1911                 if (ret < 0) {
1912                         mlog_errno(ret);
1913                         goto out_commit;
1914                 }
1915
1916                 new_bh = sb_getblk(inode->i_sb, first_blkno);
1917                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
1918
1919                 ret = ocfs2_journal_access(handle, inode, new_bh,
1920                                            OCFS2_JOURNAL_ACCESS_CREATE);
1921                 if (ret < 0) {
1922                         mlog_errno(ret);
1923                         goto out_commit;
1924                 }
1925
1926                 /* Initialize ocfs2_xattr_block */
1927                 xs->xattr_bh = new_bh;
1928                 xblk = (struct ocfs2_xattr_block *)new_bh->b_data;
1929                 memset(xblk, 0, inode->i_sb->s_blocksize);
1930                 strcpy((void *)xblk, OCFS2_XATTR_BLOCK_SIGNATURE);
1931                 xblk->xb_suballoc_slot = cpu_to_le16(osb->slot_num);
1932                 xblk->xb_suballoc_bit = cpu_to_le16(suballoc_bit_start);
1933                 xblk->xb_fs_generation = cpu_to_le32(osb->fs_generation);
1934                 xblk->xb_blkno = cpu_to_le64(first_blkno);
1935
1936                 xs->header = &xblk->xb_attrs.xb_header;
1937                 xs->base = (void *)xs->header;
1938                 xs->end = (void *)xblk + inode->i_sb->s_blocksize;
1939                 xs->here = xs->header->xh_entries;
1940
1941
1942                 ret = ocfs2_journal_dirty(handle, new_bh);
1943                 if (ret < 0) {
1944                         mlog_errno(ret);
1945                         goto out_commit;
1946                 }
1947                 di->i_xattr_loc = cpu_to_le64(first_blkno);
1948                 ret = ocfs2_journal_dirty(handle, xs->inode_bh);
1949                 if (ret < 0)
1950                         mlog_errno(ret);
1951 out_commit:
1952                 ocfs2_commit_trans(osb, handle);
1953 out:
1954                 if (meta_ac)
1955                         ocfs2_free_alloc_context(meta_ac);
1956                 if (ret < 0)
1957                         return ret;
1958         } else
1959                 xblk = (struct ocfs2_xattr_block *)xs->xattr_bh->b_data;
1960
1961         if (!(le16_to_cpu(xblk->xb_flags) & OCFS2_XATTR_INDEXED)) {
1962                 /* Set extended attribute into external block */
1963                 ret = ocfs2_xattr_set_entry(inode, xi, xs, OCFS2_HAS_XATTR_FL);
1964                 if (!ret || ret != -ENOSPC)
1965                         goto end;
1966
1967                 ret = ocfs2_xattr_create_index_block(inode, xs);
1968                 if (ret)
1969                         goto end;
1970         }
1971
1972         ret = ocfs2_xattr_set_entry_index_block(inode, xi, xs);
1973         if (!ret && xblk->xb_attrs.xb_root.xt_list.l_next_free_rec == 0)
1974                 ret = ocfs2_restore_xattr_block(inode, xs);
1975
1976 end:
1977
1978         return ret;
1979 }
1980
1981 /*
1982  * ocfs2_xattr_set()
1983  *
1984  * Set, replace or remove an extended attribute for this inode.
1985  * value is NULL to remove an existing extended attribute, else either
1986  * create or replace an extended attribute.
1987  */
1988 int ocfs2_xattr_set(struct inode *inode,
1989                     int name_index,
1990                     const char *name,
1991                     const void *value,
1992                     size_t value_len,
1993                     int flags)
1994 {
1995         struct buffer_head *di_bh = NULL;
1996         struct ocfs2_dinode *di;
1997         int ret;
1998         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
1999
2000         struct ocfs2_xattr_info xi = {
2001                 .name_index = name_index,
2002                 .name = name,
2003                 .value = value,
2004                 .value_len = value_len,
2005         };
2006
2007         struct ocfs2_xattr_search xis = {
2008                 .not_found = -ENODATA,
2009         };
2010
2011         struct ocfs2_xattr_search xbs = {
2012                 .not_found = -ENODATA,
2013         };
2014
2015         if (!ocfs2_supports_xattr(OCFS2_SB(inode->i_sb)))
2016                 return -EOPNOTSUPP;
2017
2018         ret = ocfs2_inode_lock(inode, &di_bh, 1);
2019         if (ret < 0) {
2020                 mlog_errno(ret);
2021                 return ret;
2022         }
2023         xis.inode_bh = xbs.inode_bh = di_bh;
2024         di = (struct ocfs2_dinode *)di_bh->b_data;
2025
2026         down_write(&OCFS2_I(inode)->ip_xattr_sem);
2027         /*
2028          * Scan inode and external block to find the same name
2029          * extended attribute and collect search infomation.
2030          */
2031         ret = ocfs2_xattr_ibody_find(inode, name_index, name, &xis);
2032         if (ret)
2033                 goto cleanup;
2034         if (xis.not_found) {
2035                 ret = ocfs2_xattr_block_find(inode, name_index, name, &xbs);
2036                 if (ret)
2037                         goto cleanup;
2038         }
2039
2040         if (xis.not_found && xbs.not_found) {
2041                 ret = -ENODATA;
2042                 if (flags & XATTR_REPLACE)
2043                         goto cleanup;
2044                 ret = 0;
2045                 if (!value)
2046                         goto cleanup;
2047         } else {
2048                 ret = -EEXIST;
2049                 if (flags & XATTR_CREATE)
2050                         goto cleanup;
2051         }
2052
2053         if (!value) {
2054                 /* Remove existing extended attribute */
2055                 if (!xis.not_found)
2056                         ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2057                 else if (!xbs.not_found)
2058                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2059         } else {
2060                 /* We always try to set extended attribute into inode first*/
2061                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2062                 if (!ret && !xbs.not_found) {
2063                         /*
2064                          * If succeed and that extended attribute existing in
2065                          * external block, then we will remove it.
2066                          */
2067                         xi.value = NULL;
2068                         xi.value_len = 0;
2069                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2070                 } else if (ret == -ENOSPC) {
2071                         if (di->i_xattr_loc && !xbs.xattr_bh) {
2072                                 ret = ocfs2_xattr_block_find(inode, name_index,
2073                                                              name, &xbs);
2074                                 if (ret)
2075                                         goto cleanup;
2076                         }
2077                         /*
2078                          * If no space in inode, we will set extended attribute
2079                          * into external block.
2080                          */
2081                         ret = ocfs2_xattr_block_set(inode, &xi, &xbs);
2082                         if (ret)
2083                                 goto cleanup;
2084                         if (!xis.not_found) {
2085                                 /*
2086                                  * If succeed and that extended attribute
2087                                  * existing in inode, we will remove it.
2088                                  */
2089                                 xi.value = NULL;
2090                                 xi.value_len = 0;
2091                                 ret = ocfs2_xattr_ibody_set(inode, &xi, &xis);
2092                         }
2093                 }
2094         }
2095 cleanup:
2096         up_write(&OCFS2_I(inode)->ip_xattr_sem);
2097         ocfs2_inode_unlock(inode, 1);
2098         brelse(di_bh);
2099         brelse(xbs.xattr_bh);
2100         for (i = 0; i < blk_per_bucket; i++)
2101                 brelse(xbs.bucket.bhs[i]);
2102
2103         return ret;
2104 }
2105
2106 /*
2107  * Find the xattr extent rec which may contains name_hash.
2108  * e_cpos will be the first name hash of the xattr rec.
2109  * el must be the ocfs2_xattr_header.xb_attrs.xb_root.xt_list.
2110  */
2111 static int ocfs2_xattr_get_rec(struct inode *inode,
2112                                u32 name_hash,
2113                                u64 *p_blkno,
2114                                u32 *e_cpos,
2115                                u32 *num_clusters,
2116                                struct ocfs2_extent_list *el)
2117 {
2118         int ret = 0, i;
2119         struct buffer_head *eb_bh = NULL;
2120         struct ocfs2_extent_block *eb;
2121         struct ocfs2_extent_rec *rec = NULL;
2122         u64 e_blkno = 0;
2123
2124         if (el->l_tree_depth) {
2125                 ret = ocfs2_find_leaf(inode, el, name_hash, &eb_bh);
2126                 if (ret) {
2127                         mlog_errno(ret);
2128                         goto out;
2129                 }
2130
2131                 eb = (struct ocfs2_extent_block *) eb_bh->b_data;
2132                 el = &eb->h_list;
2133
2134                 if (el->l_tree_depth) {
2135                         ocfs2_error(inode->i_sb,
2136                                     "Inode %lu has non zero tree depth in "
2137                                     "xattr tree block %llu\n", inode->i_ino,
2138                                     (unsigned long long)eb_bh->b_blocknr);
2139                         ret = -EROFS;
2140                         goto out;
2141                 }
2142         }
2143
2144         for (i = le16_to_cpu(el->l_next_free_rec) - 1; i >= 0; i--) {
2145                 rec = &el->l_recs[i];
2146
2147                 if (le32_to_cpu(rec->e_cpos) <= name_hash) {
2148                         e_blkno = le64_to_cpu(rec->e_blkno);
2149                         break;
2150                 }
2151         }
2152
2153         if (!e_blkno) {
2154                 ocfs2_error(inode->i_sb, "Inode %lu has bad extent "
2155                             "record (%u, %u, 0) in xattr", inode->i_ino,
2156                             le32_to_cpu(rec->e_cpos),
2157                             ocfs2_rec_clusters(el, rec));
2158                 ret = -EROFS;
2159                 goto out;
2160         }
2161
2162         *p_blkno = le64_to_cpu(rec->e_blkno);
2163         *num_clusters = le16_to_cpu(rec->e_leaf_clusters);
2164         if (e_cpos)
2165                 *e_cpos = le32_to_cpu(rec->e_cpos);
2166 out:
2167         brelse(eb_bh);
2168         return ret;
2169 }
2170
2171 typedef int (xattr_bucket_func)(struct inode *inode,
2172                                 struct ocfs2_xattr_bucket *bucket,
2173                                 void *para);
2174
2175 static int ocfs2_find_xe_in_bucket(struct inode *inode,
2176                                    struct buffer_head *header_bh,
2177                                    int name_index,
2178                                    const char *name,
2179                                    u32 name_hash,
2180                                    u16 *xe_index,
2181                                    int *found)
2182 {
2183         int i, ret = 0, cmp = 1, block_off, new_offset;
2184         struct ocfs2_xattr_header *xh =
2185                         (struct ocfs2_xattr_header *)header_bh->b_data;
2186         size_t name_len = strlen(name);
2187         struct ocfs2_xattr_entry *xe = NULL;
2188         struct buffer_head *name_bh = NULL;
2189         char *xe_name;
2190
2191         /*
2192          * We don't use binary search in the bucket because there
2193          * may be multiple entries with the same name hash.
2194          */
2195         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
2196                 xe = &xh->xh_entries[i];
2197
2198                 if (name_hash > le32_to_cpu(xe->xe_name_hash))
2199                         continue;
2200                 else if (name_hash < le32_to_cpu(xe->xe_name_hash))
2201                         break;
2202
2203                 cmp = name_index - ocfs2_xattr_get_type(xe);
2204                 if (!cmp)
2205                         cmp = name_len - xe->xe_name_len;
2206                 if (cmp)
2207                         continue;
2208
2209                 ret = ocfs2_xattr_bucket_get_name_value(inode,
2210                                                         xh,
2211                                                         i,
2212                                                         &block_off,
2213                                                         &new_offset);
2214                 if (ret) {
2215                         mlog_errno(ret);
2216                         break;
2217                 }
2218
2219                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb),
2220                                        header_bh->b_blocknr + block_off,
2221                                        &name_bh, OCFS2_BH_CACHED, inode);
2222                 if (ret) {
2223                         mlog_errno(ret);
2224                         break;
2225                 }
2226                 xe_name = name_bh->b_data + new_offset;
2227
2228                 cmp = memcmp(name, xe_name, name_len);
2229                 brelse(name_bh);
2230                 name_bh = NULL;
2231
2232                 if (cmp == 0) {
2233                         *xe_index = i;
2234                         *found = 1;
2235                         ret = 0;
2236                         break;
2237                 }
2238         }
2239
2240         return ret;
2241 }
2242
2243 /*
2244  * Find the specified xattr entry in a series of buckets.
2245  * This series start from p_blkno and last for num_clusters.
2246  * The ocfs2_xattr_header.xh_num_buckets of the first bucket contains
2247  * the num of the valid buckets.
2248  *
2249  * Return the buffer_head this xattr should reside in. And if the xattr's
2250  * hash is in the gap of 2 buckets, return the lower bucket.
2251  */
2252 static int ocfs2_xattr_bucket_find(struct inode *inode,
2253                                    int name_index,
2254                                    const char *name,
2255                                    u32 name_hash,
2256                                    u64 p_blkno,
2257                                    u32 first_hash,
2258                                    u32 num_clusters,
2259                                    struct ocfs2_xattr_search *xs)
2260 {
2261         int ret, found = 0;
2262         struct buffer_head *bh = NULL;
2263         struct buffer_head *lower_bh = NULL;
2264         struct ocfs2_xattr_header *xh = NULL;
2265         struct ocfs2_xattr_entry *xe = NULL;
2266         u16 index = 0;
2267         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2268         int low_bucket = 0, bucket, high_bucket;
2269         u32 last_hash;
2270         u64 blkno;
2271
2272         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), p_blkno,
2273                                &bh, OCFS2_BH_CACHED, inode);
2274         if (ret) {
2275                 mlog_errno(ret);
2276                 goto out;
2277         }
2278
2279         xh = (struct ocfs2_xattr_header *)bh->b_data;
2280         high_bucket = le16_to_cpu(xh->xh_num_buckets) - 1;
2281
2282         while (low_bucket <= high_bucket) {
2283                 brelse(bh);
2284                 bh = NULL;
2285                 bucket = (low_bucket + high_bucket) / 2;
2286
2287                 blkno = p_blkno + bucket * blk_per_bucket;
2288
2289                 ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), blkno,
2290                                        &bh, OCFS2_BH_CACHED, inode);
2291                 if (ret) {
2292                         mlog_errno(ret);
2293                         goto out;
2294                 }
2295
2296                 xh = (struct ocfs2_xattr_header *)bh->b_data;
2297                 xe = &xh->xh_entries[0];
2298                 if (name_hash < le32_to_cpu(xe->xe_name_hash)) {
2299                         high_bucket = bucket - 1;
2300                         continue;
2301                 }
2302
2303                 /*
2304                  * Check whether the hash of the last entry in our
2305                  * bucket is larger than the search one. for an empty
2306                  * bucket, the last one is also the first one.
2307                  */
2308                 if (xh->xh_count)
2309                         xe = &xh->xh_entries[le16_to_cpu(xh->xh_count) - 1];
2310
2311                 last_hash = le32_to_cpu(xe->xe_name_hash);
2312
2313                 /* record lower_bh which may be the insert place. */
2314                 brelse(lower_bh);
2315                 lower_bh = bh;
2316                 bh = NULL;
2317
2318                 if (name_hash > le32_to_cpu(xe->xe_name_hash)) {
2319                         low_bucket = bucket + 1;
2320                         continue;
2321                 }
2322
2323                 /* the searched xattr should reside in this bucket if exists. */
2324                 ret = ocfs2_find_xe_in_bucket(inode, lower_bh,
2325                                               name_index, name, name_hash,
2326                                               &index, &found);
2327                 if (ret) {
2328                         mlog_errno(ret);
2329                         goto out;
2330                 }
2331                 break;
2332         }
2333
2334         /*
2335          * Record the bucket we have found.
2336          * When the xattr's hash value is in the gap of 2 buckets, we will
2337          * always set it to the previous bucket.
2338          */
2339         if (!lower_bh) {
2340                 /*
2341                  * We can't find any bucket whose first name_hash is less
2342                  * than the find name_hash.
2343                  */
2344                 BUG_ON(bh->b_blocknr != p_blkno);
2345                 lower_bh = bh;
2346                 bh = NULL;
2347         }
2348         xs->bucket.bhs[0] = lower_bh;
2349         xs->bucket.xh = (struct ocfs2_xattr_header *)
2350                                         xs->bucket.bhs[0]->b_data;
2351         lower_bh = NULL;
2352
2353         xs->header = xs->bucket.xh;
2354         xs->base = xs->bucket.bhs[0]->b_data;
2355         xs->end = xs->base + inode->i_sb->s_blocksize;
2356
2357         if (found) {
2358                 /*
2359                  * If we have found the xattr enty, read all the blocks in
2360                  * this bucket.
2361                  */
2362                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2363                                         xs->bucket.bhs[0]->b_blocknr + 1,
2364                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2365                                         OCFS2_BH_CACHED, inode);
2366                 if (ret) {
2367                         mlog_errno(ret);
2368                         goto out;
2369                 }
2370
2371                 xs->here = &xs->header->xh_entries[index];
2372                 mlog(0, "find xattr %s in bucket %llu, entry = %u\n", name,
2373                      (unsigned long long)xs->bucket.bhs[0]->b_blocknr, index);
2374         } else
2375                 ret = -ENODATA;
2376
2377 out:
2378         brelse(bh);
2379         brelse(lower_bh);
2380         return ret;
2381 }
2382
2383 static int ocfs2_xattr_index_block_find(struct inode *inode,
2384                                         struct buffer_head *root_bh,
2385                                         int name_index,
2386                                         const char *name,
2387                                         struct ocfs2_xattr_search *xs)
2388 {
2389         int ret;
2390         struct ocfs2_xattr_block *xb =
2391                         (struct ocfs2_xattr_block *)root_bh->b_data;
2392         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
2393         struct ocfs2_extent_list *el = &xb_root->xt_list;
2394         u64 p_blkno = 0;
2395         u32 first_hash, num_clusters = 0;
2396         u32 name_hash = ocfs2_xattr_name_hash(inode, name, strlen(name));
2397
2398         if (le16_to_cpu(el->l_next_free_rec) == 0)
2399                 return -ENODATA;
2400
2401         mlog(0, "find xattr %s, hash = %u, index = %d in xattr tree\n",
2402              name, name_hash, name_index);
2403
2404         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &first_hash,
2405                                   &num_clusters, el);
2406         if (ret) {
2407                 mlog_errno(ret);
2408                 goto out;
2409         }
2410
2411         BUG_ON(p_blkno == 0 || num_clusters == 0 || first_hash > name_hash);
2412
2413         mlog(0, "find xattr extent rec %u clusters from %llu, the first hash "
2414              "in the rec is %u\n", num_clusters, p_blkno, first_hash);
2415
2416         ret = ocfs2_xattr_bucket_find(inode, name_index, name, name_hash,
2417                                       p_blkno, first_hash, num_clusters, xs);
2418
2419 out:
2420         return ret;
2421 }
2422
2423 static int ocfs2_iterate_xattr_buckets(struct inode *inode,
2424                                        u64 blkno,
2425                                        u32 clusters,
2426                                        xattr_bucket_func *func,
2427                                        void *para)
2428 {
2429         int i, j, ret = 0;
2430         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2431         u32 bpc = ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb));
2432         u32 num_buckets = clusters * bpc;
2433         struct ocfs2_xattr_bucket bucket;
2434
2435         memset(&bucket, 0, sizeof(bucket));
2436
2437         mlog(0, "iterating xattr buckets in %u clusters starting from %llu\n",
2438              clusters, blkno);
2439
2440         for (i = 0; i < num_buckets; i++, blkno += blk_per_bucket) {
2441                 ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2442                                         blkno, blk_per_bucket,
2443                                         bucket.bhs, OCFS2_BH_CACHED, inode);
2444                 if (ret) {
2445                         mlog_errno(ret);
2446                         goto out;
2447                 }
2448
2449                 bucket.xh = (struct ocfs2_xattr_header *)bucket.bhs[0]->b_data;
2450                 /*
2451                  * The real bucket num in this series of blocks is stored
2452                  * in the 1st bucket.
2453                  */
2454                 if (i == 0)
2455                         num_buckets = le16_to_cpu(bucket.xh->xh_num_buckets);
2456
2457                 mlog(0, "iterating xattr bucket %llu, first hash %u\n", blkno,
2458                      le32_to_cpu(bucket.xh->xh_entries[0].xe_name_hash));
2459                 if (func) {
2460                         ret = func(inode, &bucket, para);
2461                         if (ret) {
2462                                 mlog_errno(ret);
2463                                 break;
2464                         }
2465                 }
2466
2467                 for (j = 0; j < blk_per_bucket; j++)
2468                         brelse(bucket.bhs[j]);
2469                 memset(&bucket, 0, sizeof(bucket));
2470         }
2471
2472 out:
2473         for (j = 0; j < blk_per_bucket; j++)
2474                 brelse(bucket.bhs[j]);
2475
2476         return ret;
2477 }
2478
2479 struct ocfs2_xattr_tree_list {
2480         char *buffer;
2481         size_t buffer_size;
2482         size_t result;
2483 };
2484
2485 static int ocfs2_xattr_bucket_get_name_value(struct inode *inode,
2486                                              struct ocfs2_xattr_header *xh,
2487                                              int index,
2488                                              int *block_off,
2489                                              int *new_offset)
2490 {
2491         u16 name_offset;
2492
2493         if (index < 0 || index >= le16_to_cpu(xh->xh_count))
2494                 return -EINVAL;
2495
2496         name_offset = le16_to_cpu(xh->xh_entries[index].xe_name_offset);
2497
2498         *block_off = name_offset >> inode->i_sb->s_blocksize_bits;
2499         *new_offset = name_offset % inode->i_sb->s_blocksize;
2500
2501         return 0;
2502 }
2503
2504 static int ocfs2_list_xattr_bucket(struct inode *inode,
2505                                    struct ocfs2_xattr_bucket *bucket,
2506                                    void *para)
2507 {
2508         int ret = 0, type;
2509         struct ocfs2_xattr_tree_list *xl = (struct ocfs2_xattr_tree_list *)para;
2510         int i, block_off, new_offset;
2511         const char *prefix, *name;
2512
2513         for (i = 0 ; i < le16_to_cpu(bucket->xh->xh_count); i++) {
2514                 struct ocfs2_xattr_entry *entry = &bucket->xh->xh_entries[i];
2515                 type = ocfs2_xattr_get_type(entry);
2516                 prefix = ocfs2_xattr_prefix(type);
2517
2518                 if (prefix) {
2519                         ret = ocfs2_xattr_bucket_get_name_value(inode,
2520                                                                 bucket->xh,
2521                                                                 i,
2522                                                                 &block_off,
2523                                                                 &new_offset);
2524                         if (ret)
2525                                 break;
2526
2527                         name = (const char *)bucket->bhs[block_off]->b_data +
2528                                 new_offset;
2529                         ret = ocfs2_xattr_list_entry(xl->buffer,
2530                                                      xl->buffer_size,
2531                                                      &xl->result,
2532                                                      prefix, name,
2533                                                      entry->xe_name_len);
2534                         if (ret)
2535                                 break;
2536                 }
2537         }
2538
2539         return ret;
2540 }
2541
2542 static int ocfs2_xattr_tree_list_index_block(struct inode *inode,
2543                                              struct ocfs2_xattr_tree_root *xt,
2544                                              char *buffer,
2545                                              size_t buffer_size)
2546 {
2547         struct ocfs2_extent_list *el = &xt->xt_list;
2548         int ret = 0;
2549         u32 name_hash = UINT_MAX, e_cpos = 0, num_clusters = 0;
2550         u64 p_blkno = 0;
2551         struct ocfs2_xattr_tree_list xl = {
2552                 .buffer = buffer,
2553                 .buffer_size = buffer_size,
2554                 .result = 0,
2555         };
2556
2557         if (le16_to_cpu(el->l_next_free_rec) == 0)
2558                 return 0;
2559
2560         while (name_hash > 0) {
2561                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
2562                                           &e_cpos, &num_clusters, el);
2563                 if (ret) {
2564                         mlog_errno(ret);
2565                         goto out;
2566                 }
2567
2568                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
2569                                                   ocfs2_list_xattr_bucket,
2570                                                   &xl);
2571                 if (ret) {
2572                         mlog_errno(ret);
2573                         goto out;
2574                 }
2575
2576                 if (e_cpos == 0)
2577                         break;
2578
2579                 name_hash = e_cpos - 1;
2580         }
2581
2582         ret = xl.result;
2583 out:
2584         return ret;
2585 }
2586
2587 static int cmp_xe(const void *a, const void *b)
2588 {
2589         const struct ocfs2_xattr_entry *l = a, *r = b;
2590         u32 l_hash = le32_to_cpu(l->xe_name_hash);
2591         u32 r_hash = le32_to_cpu(r->xe_name_hash);
2592
2593         if (l_hash > r_hash)
2594                 return 1;
2595         if (l_hash < r_hash)
2596                 return -1;
2597         return 0;
2598 }
2599
2600 static void swap_xe(void *a, void *b, int size)
2601 {
2602         struct ocfs2_xattr_entry *l = a, *r = b, tmp;
2603
2604         tmp = *l;
2605         memcpy(l, r, sizeof(struct ocfs2_xattr_entry));
2606         memcpy(r, &tmp, sizeof(struct ocfs2_xattr_entry));
2607 }
2608
2609 /*
2610  * When the ocfs2_xattr_block is filled up, new bucket will be created
2611  * and all the xattr entries will be moved to the new bucket.
2612  * Note: we need to sort the entries since they are not saved in order
2613  * in the ocfs2_xattr_block.
2614  */
2615 static void ocfs2_cp_xattr_block_to_bucket(struct inode *inode,
2616                                            struct buffer_head *xb_bh,
2617                                            struct buffer_head *xh_bh,
2618                                            struct buffer_head *data_bh)
2619 {
2620         int i, blocksize = inode->i_sb->s_blocksize;
2621         u16 offset, size, off_change;
2622         struct ocfs2_xattr_entry *xe;
2623         struct ocfs2_xattr_block *xb =
2624                                 (struct ocfs2_xattr_block *)xb_bh->b_data;
2625         struct ocfs2_xattr_header *xb_xh = &xb->xb_attrs.xb_header;
2626         struct ocfs2_xattr_header *xh =
2627                                 (struct ocfs2_xattr_header *)xh_bh->b_data;
2628         u16 count = le16_to_cpu(xb_xh->xh_count);
2629         char *target = xh_bh->b_data, *src = xb_bh->b_data;
2630
2631         mlog(0, "cp xattr from block %llu to bucket %llu\n",
2632              (unsigned long long)xb_bh->b_blocknr,
2633              (unsigned long long)xh_bh->b_blocknr);
2634
2635         memset(xh_bh->b_data, 0, blocksize);
2636         if (data_bh)
2637                 memset(data_bh->b_data, 0, blocksize);
2638         /*
2639          * Since the xe_name_offset is based on ocfs2_xattr_header,
2640          * there is a offset change corresponding to the change of
2641          * ocfs2_xattr_header's position.
2642          */
2643         off_change = offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2644         xe = &xb_xh->xh_entries[count - 1];
2645         offset = le16_to_cpu(xe->xe_name_offset) + off_change;
2646         size = blocksize - offset;
2647
2648         /* copy all the names and values. */
2649         if (data_bh)
2650                 target = data_bh->b_data;
2651         memcpy(target + offset, src + offset, size);
2652
2653         /* Init new header now. */
2654         xh->xh_count = xb_xh->xh_count;
2655         xh->xh_num_buckets = cpu_to_le16(1);
2656         xh->xh_name_value_len = cpu_to_le16(size);
2657         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE - size);
2658
2659         /* copy all the entries. */
2660         target = xh_bh->b_data;
2661         offset = offsetof(struct ocfs2_xattr_header, xh_entries);
2662         size = count * sizeof(struct ocfs2_xattr_entry);
2663         memcpy(target + offset, (char *)xb_xh + offset, size);
2664
2665         /* Change the xe offset for all the xe because of the move. */
2666         off_change = OCFS2_XATTR_BUCKET_SIZE - blocksize +
2667                  offsetof(struct ocfs2_xattr_block, xb_attrs.xb_header);
2668         for (i = 0; i < count; i++)
2669                 le16_add_cpu(&xh->xh_entries[i].xe_name_offset, off_change);
2670
2671         mlog(0, "copy entry: start = %u, size = %u, offset_change = %u\n",
2672              offset, size, off_change);
2673
2674         sort(target + offset, count, sizeof(struct ocfs2_xattr_entry),
2675              cmp_xe, swap_xe);
2676 }
2677
2678 /*
2679  * After we move xattr from block to index btree, we have to
2680  * update ocfs2_xattr_search to the new xe and base.
2681  *
2682  * When the entry is in xattr block, xattr_bh indicates the storage place.
2683  * While if the entry is in index b-tree, "bucket" indicates the
2684  * real place of the xattr.
2685  */
2686 static int ocfs2_xattr_update_xattr_search(struct inode *inode,
2687                                            struct ocfs2_xattr_search *xs,
2688                                            struct buffer_head *old_bh,
2689                                            struct buffer_head *new_bh)
2690 {
2691         int ret = 0;
2692         char *buf = old_bh->b_data;
2693         struct ocfs2_xattr_block *old_xb = (struct ocfs2_xattr_block *)buf;
2694         struct ocfs2_xattr_header *old_xh = &old_xb->xb_attrs.xb_header;
2695         int i, blocksize = inode->i_sb->s_blocksize;
2696         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2697
2698         xs->bucket.bhs[0] = new_bh;
2699         get_bh(new_bh);
2700         xs->bucket.xh = (struct ocfs2_xattr_header *)xs->bucket.bhs[0]->b_data;
2701         xs->header = xs->bucket.xh;
2702
2703         xs->base = new_bh->b_data;
2704         xs->end = xs->base + inode->i_sb->s_blocksize;
2705
2706         if (!xs->not_found) {
2707                 if (OCFS2_XATTR_BUCKET_SIZE != blocksize) {
2708                         ret = ocfs2_read_blocks(OCFS2_SB(inode->i_sb),
2709                                         xs->bucket.bhs[0]->b_blocknr + 1,
2710                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
2711                                         OCFS2_BH_CACHED, inode);
2712                         if (ret) {
2713                                 mlog_errno(ret);
2714                                 return ret;
2715                         }
2716
2717                         i = xs->here - old_xh->xh_entries;
2718                         xs->here = &xs->header->xh_entries[i];
2719                 }
2720         }
2721
2722         return ret;
2723 }
2724
2725 static int ocfs2_xattr_create_index_block(struct inode *inode,
2726                                           struct ocfs2_xattr_search *xs)
2727 {
2728         int ret, credits = OCFS2_SUBALLOC_ALLOC;
2729         u32 bit_off, len;
2730         u64 blkno;
2731         handle_t *handle;
2732         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2733         struct ocfs2_inode_info *oi = OCFS2_I(inode);
2734         struct ocfs2_alloc_context *data_ac;
2735         struct buffer_head *xh_bh = NULL, *data_bh = NULL;
2736         struct buffer_head *xb_bh = xs->xattr_bh;
2737         struct ocfs2_xattr_block *xb =
2738                         (struct ocfs2_xattr_block *)xb_bh->b_data;
2739         struct ocfs2_xattr_tree_root *xr;
2740         u16 xb_flags = le16_to_cpu(xb->xb_flags);
2741         u16 bpb = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2742
2743         mlog(0, "create xattr index block for %llu\n",
2744              (unsigned long long)xb_bh->b_blocknr);
2745
2746         BUG_ON(xb_flags & OCFS2_XATTR_INDEXED);
2747
2748         ret = ocfs2_reserve_clusters(osb, 1, &data_ac);
2749         if (ret) {
2750                 mlog_errno(ret);
2751                 goto out;
2752         }
2753
2754         /*
2755          * XXX:
2756          * We can use this lock for now, and maybe move to a dedicated mutex
2757          * if performance becomes a problem later.
2758          */
2759         down_write(&oi->ip_alloc_sem);
2760
2761         /*
2762          * 3 more credits, one for xattr block update, one for the 1st block
2763          * of the new xattr bucket and one for the value/data.
2764          */
2765         credits += 3;
2766         handle = ocfs2_start_trans(osb, credits);
2767         if (IS_ERR(handle)) {
2768                 ret = PTR_ERR(handle);
2769                 mlog_errno(ret);
2770                 goto out_sem;
2771         }
2772
2773         ret = ocfs2_journal_access(handle, inode, xb_bh,
2774                                    OCFS2_JOURNAL_ACCESS_WRITE);
2775         if (ret) {
2776                 mlog_errno(ret);
2777                 goto out_commit;
2778         }
2779
2780         ret = ocfs2_claim_clusters(osb, handle, data_ac, 1, &bit_off, &len);
2781         if (ret) {
2782                 mlog_errno(ret);
2783                 goto out_commit;
2784         }
2785
2786         /*
2787          * The bucket may spread in many blocks, and
2788          * we will only touch the 1st block and the last block
2789          * in the whole bucket(one for entry and one for data).
2790          */
2791         blkno = ocfs2_clusters_to_blocks(inode->i_sb, bit_off);
2792
2793         mlog(0, "allocate 1 cluster from %llu to xattr block\n", blkno);
2794
2795         xh_bh = sb_getblk(inode->i_sb, blkno);
2796         if (!xh_bh) {
2797                 ret = -EIO;
2798                 mlog_errno(ret);
2799                 goto out_commit;
2800         }
2801
2802         ocfs2_set_new_buffer_uptodate(inode, xh_bh);
2803
2804         ret = ocfs2_journal_access(handle, inode, xh_bh,
2805                                    OCFS2_JOURNAL_ACCESS_CREATE);
2806         if (ret) {
2807                 mlog_errno(ret);
2808                 goto out_commit;
2809         }
2810
2811         if (bpb > 1) {
2812                 data_bh = sb_getblk(inode->i_sb, blkno + bpb - 1);
2813                 if (!data_bh) {
2814                         ret = -EIO;
2815                         mlog_errno(ret);
2816                         goto out_commit;
2817                 }
2818
2819                 ocfs2_set_new_buffer_uptodate(inode, data_bh);
2820
2821                 ret = ocfs2_journal_access(handle, inode, data_bh,
2822                                            OCFS2_JOURNAL_ACCESS_CREATE);
2823                 if (ret) {
2824                         mlog_errno(ret);
2825                         goto out_commit;
2826                 }
2827         }
2828
2829         ocfs2_cp_xattr_block_to_bucket(inode, xb_bh, xh_bh, data_bh);
2830
2831         ocfs2_journal_dirty(handle, xh_bh);
2832         if (data_bh)
2833                 ocfs2_journal_dirty(handle, data_bh);
2834
2835         ocfs2_xattr_update_xattr_search(inode, xs, xb_bh, xh_bh);
2836
2837         /* Change from ocfs2_xattr_header to ocfs2_xattr_tree_root */
2838         memset(&xb->xb_attrs, 0, inode->i_sb->s_blocksize -
2839                offsetof(struct ocfs2_xattr_block, xb_attrs));
2840
2841         xr = &xb->xb_attrs.xb_root;
2842         xr->xt_clusters = cpu_to_le32(1);
2843         xr->xt_last_eb_blk = 0;
2844         xr->xt_list.l_tree_depth = 0;
2845         xr->xt_list.l_count = cpu_to_le16(ocfs2_xattr_recs_per_xb(inode->i_sb));
2846         xr->xt_list.l_next_free_rec = cpu_to_le16(1);
2847
2848         xr->xt_list.l_recs[0].e_cpos = 0;
2849         xr->xt_list.l_recs[0].e_blkno = cpu_to_le64(blkno);
2850         xr->xt_list.l_recs[0].e_leaf_clusters = cpu_to_le16(1);
2851
2852         xb->xb_flags = cpu_to_le16(xb_flags | OCFS2_XATTR_INDEXED);
2853
2854         ret = ocfs2_journal_dirty(handle, xb_bh);
2855         if (ret) {
2856                 mlog_errno(ret);
2857                 goto out_commit;
2858         }
2859
2860 out_commit:
2861         ocfs2_commit_trans(osb, handle);
2862
2863 out_sem:
2864         up_write(&oi->ip_alloc_sem);
2865
2866 out:
2867         if (data_ac)
2868                 ocfs2_free_alloc_context(data_ac);
2869
2870         brelse(xh_bh);
2871         brelse(data_bh);
2872
2873         return ret;
2874 }
2875
2876 static int cmp_xe_offset(const void *a, const void *b)
2877 {
2878         const struct ocfs2_xattr_entry *l = a, *r = b;
2879         u32 l_name_offset = le16_to_cpu(l->xe_name_offset);
2880         u32 r_name_offset = le16_to_cpu(r->xe_name_offset);
2881
2882         if (l_name_offset < r_name_offset)
2883                 return 1;
2884         if (l_name_offset > r_name_offset)
2885                 return -1;
2886         return 0;
2887 }
2888
2889 /*
2890  * defrag a xattr bucket if we find that the bucket has some
2891  * holes beteen name/value pairs.
2892  * We will move all the name/value pairs to the end of the bucket
2893  * so that we can spare some space for insertion.
2894  */
2895 static int ocfs2_defrag_xattr_bucket(struct inode *inode,
2896                                      struct ocfs2_xattr_bucket *bucket)
2897 {
2898         int ret, i;
2899         size_t end, offset, len, value_len;
2900         struct ocfs2_xattr_header *xh;
2901         char *entries, *buf, *bucket_buf = NULL;
2902         u64 blkno = bucket->bhs[0]->b_blocknr;
2903         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
2904         u16 xh_free_start;
2905         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
2906         size_t blocksize = inode->i_sb->s_blocksize;
2907         handle_t *handle;
2908         struct buffer_head **bhs;
2909         struct ocfs2_xattr_entry *xe;
2910
2911         bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
2912                         GFP_NOFS);
2913         if (!bhs)
2914                 return -ENOMEM;
2915
2916         ret = ocfs2_read_blocks(osb, blkno, blk_per_bucket, bhs,
2917                                 OCFS2_BH_CACHED, inode);
2918         if (ret)
2919                 goto out;
2920
2921         /*
2922          * In order to make the operation more efficient and generic,
2923          * we copy all the blocks into a contiguous memory and do the
2924          * defragment there, so if anything is error, we will not touch
2925          * the real block.
2926          */
2927         bucket_buf = kmalloc(OCFS2_XATTR_BUCKET_SIZE, GFP_NOFS);
2928         if (!bucket_buf) {
2929                 ret = -EIO;
2930                 goto out;
2931         }
2932
2933         buf = bucket_buf;
2934         for (i = 0; i < blk_per_bucket; i++, buf += blocksize)
2935                 memcpy(buf, bhs[i]->b_data, blocksize);
2936
2937         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), blk_per_bucket);
2938         if (IS_ERR(handle)) {
2939                 ret = PTR_ERR(handle);
2940                 handle = NULL;
2941                 mlog_errno(ret);
2942                 goto out;
2943         }
2944
2945         for (i = 0; i < blk_per_bucket; i++) {
2946                 ret = ocfs2_journal_access(handle, inode, bhs[i],
2947                                            OCFS2_JOURNAL_ACCESS_WRITE);
2948                 if (ret < 0) {
2949                         mlog_errno(ret);
2950                         goto commit;
2951                 }
2952         }
2953
2954         xh = (struct ocfs2_xattr_header *)bucket_buf;
2955         entries = (char *)xh->xh_entries;
2956         xh_free_start = le16_to_cpu(xh->xh_free_start);
2957
2958         mlog(0, "adjust xattr bucket in %llu, count = %u, "
2959              "xh_free_start = %u, xh_name_value_len = %u.\n",
2960              blkno, le16_to_cpu(xh->xh_count), xh_free_start,
2961              le16_to_cpu(xh->xh_name_value_len));
2962
2963         /*
2964          * sort all the entries by their offset.
2965          * the largest will be the first, so that we can
2966          * move them to the end one by one.
2967          */
2968         sort(entries, le16_to_cpu(xh->xh_count),
2969              sizeof(struct ocfs2_xattr_entry),
2970              cmp_xe_offset, swap_xe);
2971
2972         /* Move all name/values to the end of the bucket. */
2973         xe = xh->xh_entries;
2974         end = OCFS2_XATTR_BUCKET_SIZE;
2975         for (i = 0; i < le16_to_cpu(xh->xh_count); i++, xe++) {
2976                 offset = le16_to_cpu(xe->xe_name_offset);
2977                 if (ocfs2_xattr_is_local(xe))
2978                         value_len = OCFS2_XATTR_SIZE(
2979                                         le64_to_cpu(xe->xe_value_size));
2980                 else
2981                         value_len = OCFS2_XATTR_ROOT_SIZE;
2982                 len = OCFS2_XATTR_SIZE(xe->xe_name_len) + value_len;
2983
2984                 /*
2985                  * We must make sure that the name/value pair
2986                  * exist in the same block. So adjust end to
2987                  * the previous block end if needed.
2988                  */
2989                 if (((end - len) / blocksize !=
2990                         (end - 1) / blocksize))
2991                         end = end - end % blocksize;
2992
2993                 if (end > offset + len) {
2994                         memmove(bucket_buf + end - len,
2995                                 bucket_buf + offset, len);
2996                         xe->xe_name_offset = cpu_to_le16(end - len);
2997                 }
2998
2999                 mlog_bug_on_msg(end < offset + len, "Defrag check failed for "
3000                                 "bucket %llu\n", (unsigned long long)blkno);
3001
3002                 end -= len;
3003         }
3004
3005         mlog_bug_on_msg(xh_free_start > end, "Defrag check failed for "
3006                         "bucket %llu\n", (unsigned long long)blkno);
3007
3008         if (xh_free_start == end)
3009                 goto commit;
3010
3011         memset(bucket_buf + xh_free_start, 0, end - xh_free_start);
3012         xh->xh_free_start = cpu_to_le16(end);
3013
3014         /* sort the entries by their name_hash. */
3015         sort(entries, le16_to_cpu(xh->xh_count),
3016              sizeof(struct ocfs2_xattr_entry),
3017              cmp_xe, swap_xe);
3018
3019         buf = bucket_buf;
3020         for (i = 0; i < blk_per_bucket; i++, buf += blocksize) {
3021                 memcpy(bhs[i]->b_data, buf, blocksize);
3022                 ocfs2_journal_dirty(handle, bhs[i]);
3023         }
3024
3025 commit:
3026         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
3027 out:
3028
3029         if (bhs) {
3030                 for (i = 0; i < blk_per_bucket; i++)
3031                         brelse(bhs[i]);
3032         }
3033         kfree(bhs);
3034
3035         kfree(bucket_buf);
3036         return ret;
3037 }
3038
3039 /*
3040  * Move half nums of the xattr bucket in the previous cluster to this new
3041  * cluster. We only touch the last cluster of the previous extend record.
3042  *
3043  * first_bh is the first buffer_head of a series of bucket in the same
3044  * extent rec and header_bh is the header of one bucket in this cluster.
3045  * They will be updated if we move the data header_bh contains to the new
3046  * cluster. first_hash will be set as the 1st xe's name_hash of the new cluster.
3047  */
3048 static int ocfs2_mv_xattr_bucket_cross_cluster(struct inode *inode,
3049                                                handle_t *handle,
3050                                                struct buffer_head **first_bh,
3051                                                struct buffer_head **header_bh,
3052                                                u64 new_blkno,
3053                                                u64 prev_blkno,
3054                                                u32 num_clusters,
3055                                                u32 *first_hash)
3056 {
3057         int i, ret, credits;
3058         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3059         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3060         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3061         int blocksize = inode->i_sb->s_blocksize;
3062         struct buffer_head *old_bh, *new_bh, *prev_bh, *new_first_bh = NULL;
3063         struct ocfs2_xattr_header *new_xh;
3064         struct ocfs2_xattr_header *xh =
3065                         (struct ocfs2_xattr_header *)((*first_bh)->b_data);
3066
3067         BUG_ON(le16_to_cpu(xh->xh_num_buckets) < num_buckets);
3068         BUG_ON(OCFS2_XATTR_BUCKET_SIZE == osb->s_clustersize);
3069
3070         prev_bh = *first_bh;
3071         get_bh(prev_bh);
3072         xh = (struct ocfs2_xattr_header *)prev_bh->b_data;
3073
3074         prev_blkno += (num_clusters - 1) * bpc + bpc / 2;
3075
3076         mlog(0, "move half of xattrs in cluster %llu to %llu\n",
3077              prev_blkno, new_blkno);
3078
3079         /*
3080          * We need to update the 1st half of the new cluster and
3081          * 1 more for the update of the 1st bucket of the previous
3082          * extent record.
3083          */
3084         credits = bpc / 2 + 1;
3085         ret = ocfs2_extend_trans(handle, credits);
3086         if (ret) {
3087                 mlog_errno(ret);
3088                 goto out;
3089         }
3090
3091         ret = ocfs2_journal_access(handle, inode, prev_bh,
3092                                    OCFS2_JOURNAL_ACCESS_WRITE);
3093         if (ret) {
3094                 mlog_errno(ret);
3095                 goto out;
3096         }
3097
3098         for (i = 0; i < bpc / 2; i++, prev_blkno++, new_blkno++) {
3099                 old_bh = new_bh = NULL;
3100                 new_bh = sb_getblk(inode->i_sb, new_blkno);
3101                 if (!new_bh) {
3102                         ret = -EIO;
3103                         mlog_errno(ret);
3104                         goto out;
3105                 }
3106
3107                 ocfs2_set_new_buffer_uptodate(inode, new_bh);
3108
3109                 ret = ocfs2_journal_access(handle, inode, new_bh,
3110                                            OCFS2_JOURNAL_ACCESS_CREATE);
3111                 if (ret < 0) {
3112                         mlog_errno(ret);
3113                         brelse(new_bh);
3114                         goto out;
3115                 }
3116
3117                 ret = ocfs2_read_block(osb, prev_blkno,
3118                                        &old_bh, OCFS2_BH_CACHED, inode);
3119                 if (ret < 0) {
3120                         mlog_errno(ret);
3121                         brelse(new_bh);
3122                         goto out;
3123                 }
3124
3125                 memcpy(new_bh->b_data, old_bh->b_data, blocksize);
3126
3127                 if (i == 0) {
3128                         new_xh = (struct ocfs2_xattr_header *)new_bh->b_data;
3129                         new_xh->xh_num_buckets = cpu_to_le16(num_buckets / 2);
3130
3131                         if (first_hash)
3132                                 *first_hash = le32_to_cpu(
3133                                         new_xh->xh_entries[0].xe_name_hash);
3134                         new_first_bh = new_bh;
3135                         get_bh(new_first_bh);
3136                 }
3137
3138                 ocfs2_journal_dirty(handle, new_bh);
3139
3140                 if (*header_bh == old_bh) {
3141                         brelse(*header_bh);
3142                         *header_bh = new_bh;
3143                         get_bh(*header_bh);
3144
3145                         brelse(*first_bh);
3146                         *first_bh = new_first_bh;
3147                         get_bh(*first_bh);
3148                 }
3149                 brelse(new_bh);
3150                 brelse(old_bh);
3151         }
3152
3153         le16_add_cpu(&xh->xh_num_buckets, -(num_buckets / 2));
3154
3155         ocfs2_journal_dirty(handle, prev_bh);
3156 out:
3157         brelse(prev_bh);
3158         brelse(new_first_bh);
3159         return ret;
3160 }
3161
3162 static int ocfs2_read_xattr_bucket(struct inode *inode,
3163                                    u64 blkno,
3164                                    struct buffer_head **bhs,
3165                                    int new)
3166 {
3167         int ret = 0;
3168         u16 i, blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3169
3170         if (!new)
3171                 return ocfs2_read_blocks(OCFS2_SB(inode->i_sb), blkno,
3172                                          blk_per_bucket, bhs,
3173                                          OCFS2_BH_CACHED, inode);
3174
3175         for (i = 0; i < blk_per_bucket; i++) {
3176                 bhs[i] = sb_getblk(inode->i_sb, blkno + i);
3177                 if (bhs[i] == NULL) {
3178                         ret = -EIO;
3179                         mlog_errno(ret);
3180                         break;
3181                 }
3182                 ocfs2_set_new_buffer_uptodate(inode, bhs[i]);
3183         }
3184
3185         return ret;
3186 }
3187
3188 /*
3189  * Move half num of the xattrs in old bucket(blk) to new bucket(new_blk).
3190  * first_hash will record the 1st hash of the new bucket.
3191  */
3192 static int ocfs2_half_xattr_bucket(struct inode *inode,
3193                                    handle_t *handle,
3194                                    u64 blk,
3195                                    u64 new_blk,
3196                                    u32 *first_hash,
3197                                    int new_bucket_head)
3198 {
3199         int ret, i;
3200         u16 count, start, len, name_value_len, xe_len, name_offset;
3201         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3202         struct buffer_head **s_bhs, **t_bhs = NULL;
3203         struct ocfs2_xattr_header *xh;
3204         struct ocfs2_xattr_entry *xe;
3205         int blocksize = inode->i_sb->s_blocksize;
3206
3207         mlog(0, "move half of xattrs from bucket %llu to %llu\n",
3208              blk, new_blk);
3209
3210         s_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3211         if (!s_bhs)
3212                 return -ENOMEM;
3213
3214         ret = ocfs2_read_xattr_bucket(inode, blk, s_bhs, 0);
3215         if (ret) {
3216                 mlog_errno(ret);
3217                 goto out;
3218         }
3219
3220         ret = ocfs2_journal_access(handle, inode, s_bhs[0],
3221                                    OCFS2_JOURNAL_ACCESS_WRITE);
3222         if (ret) {
3223                 mlog_errno(ret);
3224                 goto out;
3225         }
3226
3227         t_bhs = kcalloc(blk_per_bucket, sizeof(struct buffer_head *), GFP_NOFS);
3228         if (!t_bhs) {
3229                 ret = -ENOMEM;
3230                 goto out;
3231         }
3232
3233         ret = ocfs2_read_xattr_bucket(inode, new_blk, t_bhs, new_bucket_head);
3234         if (ret) {
3235                 mlog_errno(ret);
3236                 goto out;
3237         }
3238
3239         for (i = 0; i < blk_per_bucket; i++) {
3240                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3241                                            OCFS2_JOURNAL_ACCESS_CREATE);
3242                 if (ret) {
3243                         mlog_errno(ret);
3244                         goto out;
3245                 }
3246         }
3247
3248         /* copy the whole bucket to the new first. */
3249         for (i = 0; i < blk_per_bucket; i++)
3250                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3251
3252         /* update the new bucket. */
3253         xh = (struct ocfs2_xattr_header *)t_bhs[0]->b_data;
3254         count = le16_to_cpu(xh->xh_count);
3255         start = count / 2;
3256
3257         /*
3258          * Calculate the total name/value len and xh_free_start for
3259          * the old bucket first.
3260          */
3261         name_offset = OCFS2_XATTR_BUCKET_SIZE;
3262         name_value_len = 0;
3263         for (i = 0; i < start; i++) {
3264                 xe = &xh->xh_entries[i];
3265                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3266                 if (ocfs2_xattr_is_local(xe))
3267                         xe_len +=
3268                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3269                 else
3270                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3271                 name_value_len += xe_len;
3272                 if (le16_to_cpu(xe->xe_name_offset) < name_offset)
3273                         name_offset = le16_to_cpu(xe->xe_name_offset);
3274         }
3275
3276         /*
3277          * Now begin the modification to the new bucket.
3278          *
3279          * In the new bucket, We just move the xattr entry to the beginning
3280          * and don't touch the name/value. So there will be some holes in the
3281          * bucket, and they will be removed when ocfs2_defrag_xattr_bucket is
3282          * called.
3283          */
3284         xe = &xh->xh_entries[start];
3285         len = sizeof(struct ocfs2_xattr_entry) * (count - start);
3286         mlog(0, "mv xattr entry len %d from %d to %d\n", len,
3287              (int)((char *)xe - (char *)xh),
3288              (int)((char *)xh->xh_entries - (char *)xh));
3289         memmove((char *)xh->xh_entries, (char *)xe, len);
3290         xe = &xh->xh_entries[count - start];
3291         len = sizeof(struct ocfs2_xattr_entry) * start;
3292         memset((char *)xe, 0, len);
3293
3294         le16_add_cpu(&xh->xh_count, -start);
3295         le16_add_cpu(&xh->xh_name_value_len, -name_value_len);
3296
3297         /* Calculate xh_free_start for the new bucket. */
3298         xh->xh_free_start = cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
3299         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
3300                 xe = &xh->xh_entries[i];
3301                 xe_len = OCFS2_XATTR_SIZE(xe->xe_name_len);
3302                 if (ocfs2_xattr_is_local(xe))
3303                         xe_len +=
3304                            OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3305                 else
3306                         xe_len += OCFS2_XATTR_ROOT_SIZE;
3307                 if (le16_to_cpu(xe->xe_name_offset) <
3308                     le16_to_cpu(xh->xh_free_start))
3309                         xh->xh_free_start = xe->xe_name_offset;
3310         }
3311
3312         /* set xh->xh_num_buckets for the new xh. */
3313         if (new_bucket_head)
3314                 xh->xh_num_buckets = cpu_to_le16(1);
3315         else
3316                 xh->xh_num_buckets = 0;
3317
3318         for (i = 0; i < blk_per_bucket; i++) {
3319                 ocfs2_journal_dirty(handle, t_bhs[i]);
3320                 if (ret)
3321                         mlog_errno(ret);
3322         }
3323
3324         /* store the first_hash of the new bucket. */
3325         if (first_hash)
3326                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3327
3328         /*
3329          * Now only update the 1st block of the old bucket.
3330          * Please note that the entry has been sorted already above.
3331          */
3332         xh = (struct ocfs2_xattr_header *)s_bhs[0]->b_data;
3333         memset(&xh->xh_entries[start], 0,
3334                sizeof(struct ocfs2_xattr_entry) * (count - start));
3335         xh->xh_count = cpu_to_le16(start);
3336         xh->xh_free_start = cpu_to_le16(name_offset);
3337         xh->xh_name_value_len = cpu_to_le16(name_value_len);
3338
3339         ocfs2_journal_dirty(handle, s_bhs[0]);
3340         if (ret)
3341                 mlog_errno(ret);
3342
3343 out:
3344         if (s_bhs) {
3345                 for (i = 0; i < blk_per_bucket; i++)
3346                         brelse(s_bhs[i]);
3347         }
3348         kfree(s_bhs);
3349
3350         if (t_bhs) {
3351                 for (i = 0; i < blk_per_bucket; i++)
3352                         brelse(t_bhs[i]);
3353         }
3354         kfree(t_bhs);
3355
3356         return ret;
3357 }
3358
3359 /*
3360  * Copy xattr from one bucket to another bucket.
3361  *
3362  * The caller must make sure that the journal transaction
3363  * has enough space for journaling.
3364  */
3365 static int ocfs2_cp_xattr_bucket(struct inode *inode,
3366                                  handle_t *handle,
3367                                  u64 s_blkno,
3368                                  u64 t_blkno,
3369                                  int t_is_new)
3370 {
3371         int ret, i;
3372         int blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3373         int blocksize = inode->i_sb->s_blocksize;
3374         struct buffer_head **s_bhs, **t_bhs = NULL;
3375
3376         BUG_ON(s_blkno == t_blkno);
3377
3378         mlog(0, "cp bucket %llu to %llu, target is %d\n",
3379              s_blkno, t_blkno, t_is_new);
3380
3381         s_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3382                         GFP_NOFS);
3383         if (!s_bhs)
3384                 return -ENOMEM;
3385
3386         ret = ocfs2_read_xattr_bucket(inode, s_blkno, s_bhs, 0);
3387         if (ret)
3388                 goto out;
3389
3390         t_bhs = kzalloc(sizeof(struct buffer_head *) * blk_per_bucket,
3391                         GFP_NOFS);
3392         if (!t_bhs) {
3393                 ret = -ENOMEM;
3394                 goto out;
3395         }
3396
3397         ret = ocfs2_read_xattr_bucket(inode, t_blkno, t_bhs, t_is_new);
3398         if (ret)
3399                 goto out;
3400
3401         for (i = 0; i < blk_per_bucket; i++) {
3402                 ret = ocfs2_journal_access(handle, inode, t_bhs[i],
3403                                            OCFS2_JOURNAL_ACCESS_WRITE);
3404                 if (ret)
3405                         goto out;
3406         }
3407
3408         for (i = 0; i < blk_per_bucket; i++) {
3409                 memcpy(t_bhs[i]->b_data, s_bhs[i]->b_data, blocksize);
3410                 ocfs2_journal_dirty(handle, t_bhs[i]);
3411         }
3412
3413 out:
3414         if (s_bhs) {
3415                 for (i = 0; i < blk_per_bucket; i++)
3416                         brelse(s_bhs[i]);
3417         }
3418         kfree(s_bhs);
3419
3420         if (t_bhs) {
3421                 for (i = 0; i < blk_per_bucket; i++)
3422                         brelse(t_bhs[i]);
3423         }
3424         kfree(t_bhs);
3425
3426         return ret;
3427 }
3428
3429 /*
3430  * Copy one xattr cluster from src_blk to to_blk.
3431  * The to_blk will become the first bucket header of the cluster, so its
3432  * xh_num_buckets will be initialized as the bucket num in the cluster.
3433  */
3434 static int ocfs2_cp_xattr_cluster(struct inode *inode,
3435                                   handle_t *handle,
3436                                   struct buffer_head *first_bh,
3437                                   u64 src_blk,
3438                                   u64 to_blk,
3439                                   u32 *first_hash)
3440 {
3441         int i, ret, credits;
3442         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3443         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3444         int num_buckets = ocfs2_xattr_buckets_per_cluster(osb);
3445         struct buffer_head *bh = NULL;
3446         struct ocfs2_xattr_header *xh;
3447         u64 to_blk_start = to_blk;
3448
3449         mlog(0, "cp xattrs from cluster %llu to %llu\n", src_blk, to_blk);
3450
3451         /*
3452          * We need to update the new cluster and 1 more for the update of
3453          * the 1st bucket of the previous extent rec.
3454          */
3455         credits = bpc + 1;
3456         ret = ocfs2_extend_trans(handle, credits);
3457         if (ret) {
3458                 mlog_errno(ret);
3459                 goto out;
3460         }
3461
3462         ret = ocfs2_journal_access(handle, inode, first_bh,
3463                                    OCFS2_JOURNAL_ACCESS_WRITE);
3464         if (ret) {
3465                 mlog_errno(ret);
3466                 goto out;
3467         }
3468
3469         for (i = 0; i < num_buckets; i++) {
3470                 ret = ocfs2_cp_xattr_bucket(inode, handle,
3471                                             src_blk, to_blk, 1);
3472                 if (ret) {
3473                         mlog_errno(ret);
3474                         goto out;
3475                 }
3476
3477                 src_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3478                 to_blk += ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3479         }
3480
3481         /* update the old bucket header. */
3482         xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3483         le16_add_cpu(&xh->xh_num_buckets, -num_buckets);
3484
3485         ocfs2_journal_dirty(handle, first_bh);
3486
3487         /* update the new bucket header. */
3488         ret = ocfs2_read_block(osb, to_blk_start, &bh, OCFS2_BH_CACHED, inode);
3489         if (ret < 0) {
3490                 mlog_errno(ret);
3491                 goto out;
3492         }
3493
3494         ret = ocfs2_journal_access(handle, inode, bh,
3495                                    OCFS2_JOURNAL_ACCESS_WRITE);
3496         if (ret) {
3497                 mlog_errno(ret);
3498                 goto out;
3499         }
3500
3501         xh = (struct ocfs2_xattr_header *)bh->b_data;
3502         xh->xh_num_buckets = cpu_to_le16(num_buckets);
3503
3504         ocfs2_journal_dirty(handle, bh);
3505
3506         if (first_hash)
3507                 *first_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3508 out:
3509         brelse(bh);
3510         return ret;
3511 }
3512
3513 /*
3514  * Move half of the xattrs in this cluster to the new cluster.
3515  * This function should only be called when bucket size == cluster size.
3516  * Otherwise ocfs2_mv_xattr_bucket_cross_cluster should be used instead.
3517  */
3518 static int ocfs2_half_xattr_cluster(struct inode *inode,
3519                                     handle_t *handle,
3520                                     u64 prev_blk,
3521                                     u64 new_blk,
3522                                     u32 *first_hash)
3523 {
3524         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3525         int ret, credits = 2 * blk_per_bucket;
3526
3527         BUG_ON(OCFS2_XATTR_BUCKET_SIZE < OCFS2_SB(inode->i_sb)->s_clustersize);
3528
3529         ret = ocfs2_extend_trans(handle, credits);
3530         if (ret) {
3531                 mlog_errno(ret);
3532                 return ret;
3533         }
3534
3535         /* Move half of the xattr in start_blk to the next bucket. */
3536         return  ocfs2_half_xattr_bucket(inode, handle, prev_blk,
3537                                         new_blk, first_hash, 1);
3538 }
3539
3540 /*
3541  * Move some xattrs from the old cluster to the new one since they are not
3542  * contiguous in ocfs2 xattr tree.
3543  *
3544  * new_blk starts a new separate cluster, and we will move some xattrs from
3545  * prev_blk to it. v_start will be set as the first name hash value in this
3546  * new cluster so that it can be used as e_cpos during tree insertion and
3547  * don't collide with our original b-tree operations. first_bh and header_bh
3548  * will also be updated since they will be used in ocfs2_extend_xattr_bucket
3549  * to extend the insert bucket.
3550  *
3551  * The problem is how much xattr should we move to the new one and when should
3552  * we update first_bh and header_bh?
3553  * 1. If cluster size > bucket size, that means the previous cluster has more
3554  *    than 1 bucket, so just move half nums of bucket into the new cluster and
3555  *    update the first_bh and header_bh if the insert bucket has been moved
3556  *    to the new cluster.
3557  * 2. If cluster_size == bucket_size:
3558  *    a) If the previous extent rec has more than one cluster and the insert
3559  *       place isn't in the last cluster, copy the entire last cluster to the
3560  *       new one. This time, we don't need to upate the first_bh and header_bh
3561  *       since they will not be moved into the new cluster.
3562  *    b) Otherwise, move the bottom half of the xattrs in the last cluster into
3563  *       the new one. And we set the extend flag to zero if the insert place is
3564  *       moved into the new allocated cluster since no extend is needed.
3565  */
3566 static int ocfs2_adjust_xattr_cross_cluster(struct inode *inode,
3567                                             handle_t *handle,
3568                                             struct buffer_head **first_bh,
3569                                             struct buffer_head **header_bh,
3570                                             u64 new_blk,
3571                                             u64 prev_blk,
3572                                             u32 prev_clusters,
3573                                             u32 *v_start,
3574                                             int *extend)
3575 {
3576         int ret = 0;
3577         int bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3578
3579         mlog(0, "adjust xattrs from cluster %llu len %u to %llu\n",
3580              prev_blk, prev_clusters, new_blk);
3581
3582         if (ocfs2_xattr_buckets_per_cluster(OCFS2_SB(inode->i_sb)) > 1)
3583                 ret = ocfs2_mv_xattr_bucket_cross_cluster(inode,
3584                                                           handle,
3585                                                           first_bh,
3586                                                           header_bh,
3587                                                           new_blk,
3588                                                           prev_blk,
3589                                                           prev_clusters,
3590                                                           v_start);
3591         else {
3592                 u64 last_blk = prev_blk + bpc * (prev_clusters - 1);
3593
3594                 if (prev_clusters > 1 && (*header_bh)->b_blocknr != last_blk)
3595                         ret = ocfs2_cp_xattr_cluster(inode, handle, *first_bh,
3596                                                      last_blk, new_blk,
3597                                                      v_start);
3598                 else {
3599                         ret = ocfs2_half_xattr_cluster(inode, handle,
3600                                                        last_blk, new_blk,
3601                                                        v_start);
3602
3603                         if ((*header_bh)->b_blocknr == last_blk && extend)
3604                                 *extend = 0;
3605                 }
3606         }
3607
3608         return ret;
3609 }
3610
3611 /*
3612  * Add a new cluster for xattr storage.
3613  *
3614  * If the new cluster is contiguous with the previous one, it will be
3615  * appended to the same extent record, and num_clusters will be updated.
3616  * If not, we will insert a new extent for it and move some xattrs in
3617  * the last cluster into the new allocated one.
3618  * We also need to limit the maximum size of a btree leaf, otherwise we'll
3619  * lose the benefits of hashing because we'll have to search large leaves.
3620  * So now the maximum size is OCFS2_MAX_XATTR_TREE_LEAF_SIZE(or clustersize,
3621  * if it's bigger).
3622  *
3623  * first_bh is the first block of the previous extent rec and header_bh
3624  * indicates the bucket we will insert the new xattrs. They will be updated
3625  * when the header_bh is moved into the new cluster.
3626  */
3627 static int ocfs2_add_new_xattr_cluster(struct inode *inode,
3628                                        struct buffer_head *root_bh,
3629                                        struct buffer_head **first_bh,
3630                                        struct buffer_head **header_bh,
3631                                        u32 *num_clusters,
3632                                        u32 prev_cpos,
3633                                        u64 prev_blkno,
3634                                        int *extend)
3635 {
3636         int ret, credits;
3637         u16 bpc = ocfs2_clusters_to_blocks(inode->i_sb, 1);
3638         u32 prev_clusters = *num_clusters;
3639         u32 clusters_to_add = 1, bit_off, num_bits, v_start = 0;
3640         u64 block;
3641         handle_t *handle = NULL;
3642         struct ocfs2_alloc_context *data_ac = NULL;
3643         struct ocfs2_alloc_context *meta_ac = NULL;
3644         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3645         struct ocfs2_extent_tree et;
3646
3647         mlog(0, "Add new xattr cluster for %llu, previous xattr hash = %u, "
3648              "previous xattr blkno = %llu\n",
3649              (unsigned long long)OCFS2_I(inode)->ip_blkno,
3650              prev_cpos, prev_blkno);
3651
3652         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
3653
3654         ret = ocfs2_lock_allocators(inode, &et, clusters_to_add, 0,
3655                                     &data_ac, &meta_ac);
3656         if (ret) {
3657                 mlog_errno(ret);
3658                 goto leave;
3659         }
3660
3661         credits = ocfs2_calc_extend_credits(osb->sb, et.et_root_el,
3662                                             clusters_to_add);
3663         handle = ocfs2_start_trans(osb, credits);
3664         if (IS_ERR(handle)) {
3665                 ret = PTR_ERR(handle);
3666                 handle = NULL;
3667                 mlog_errno(ret);
3668                 goto leave;
3669         }
3670
3671         ret = ocfs2_journal_access(handle, inode, root_bh,
3672                                    OCFS2_JOURNAL_ACCESS_WRITE);
3673         if (ret < 0) {
3674                 mlog_errno(ret);
3675                 goto leave;
3676         }
3677
3678         ret = __ocfs2_claim_clusters(osb, handle, data_ac, 1,
3679                                      clusters_to_add, &bit_off, &num_bits);
3680         if (ret < 0) {
3681                 if (ret != -ENOSPC)
3682                         mlog_errno(ret);
3683                 goto leave;
3684         }
3685
3686         BUG_ON(num_bits > clusters_to_add);
3687
3688         block = ocfs2_clusters_to_blocks(osb->sb, bit_off);
3689         mlog(0, "Allocating %u clusters at block %u for xattr in inode %llu\n",
3690              num_bits, bit_off, (unsigned long long)OCFS2_I(inode)->ip_blkno);
3691
3692         if (prev_blkno + prev_clusters * bpc == block &&
3693             (prev_clusters + num_bits) << osb->s_clustersize_bits <=
3694              OCFS2_MAX_XATTR_TREE_LEAF_SIZE) {
3695                 /*
3696                  * If this cluster is contiguous with the old one and
3697                  * adding this new cluster, we don't surpass the limit of
3698                  * OCFS2_MAX_XATTR_TREE_LEAF_SIZE, cool. We will let it be
3699                  * initialized and used like other buckets in the previous
3700                  * cluster.
3701                  * So add it as a contiguous one. The caller will handle
3702                  * its init process.
3703                  */
3704                 v_start = prev_cpos + prev_clusters;
3705                 *num_clusters = prev_clusters + num_bits;
3706                 mlog(0, "Add contiguous %u clusters to previous extent rec.\n",
3707                      num_bits);
3708         } else {
3709                 ret = ocfs2_adjust_xattr_cross_cluster(inode,
3710                                                        handle,
3711                                                        first_bh,
3712                                                        header_bh,
3713                                                        block,
3714                                                        prev_blkno,
3715                                                        prev_clusters,
3716                                                        &v_start,
3717                                                        extend);
3718                 if (ret) {
3719                         mlog_errno(ret);
3720                         goto leave;
3721                 }
3722         }
3723
3724         if (handle->h_buffer_credits < credits) {
3725                 /*
3726                  * The journal has been restarted before, and don't
3727                  * have enough space for the insertion, so extend it
3728                  * here.
3729                  */
3730                 ret = ocfs2_extend_trans(handle, credits);
3731                 if (ret) {
3732                         mlog_errno(ret);
3733                         goto leave;
3734                 }
3735         }
3736         mlog(0, "Insert %u clusters at block %llu for xattr at %u\n",
3737              num_bits, block, v_start);
3738         ret = ocfs2_insert_extent(osb, handle, inode, &et, v_start, block,
3739                                   num_bits, 0, meta_ac);
3740         if (ret < 0) {
3741                 mlog_errno(ret);
3742                 goto leave;
3743         }
3744
3745         ret = ocfs2_journal_dirty(handle, root_bh);
3746         if (ret < 0) {
3747                 mlog_errno(ret);
3748                 goto leave;
3749         }
3750
3751 leave:
3752         if (handle)
3753                 ocfs2_commit_trans(osb, handle);
3754         if (data_ac)
3755                 ocfs2_free_alloc_context(data_ac);
3756         if (meta_ac)
3757                 ocfs2_free_alloc_context(meta_ac);
3758
3759         return ret;
3760 }
3761
3762 /*
3763  * Extend a new xattr bucket and move xattrs to the end one by one until
3764  * We meet with start_bh. Only move half of the xattrs to the bucket after it.
3765  */
3766 static int ocfs2_extend_xattr_bucket(struct inode *inode,
3767                                      struct buffer_head *first_bh,
3768                                      struct buffer_head *start_bh,
3769                                      u32 num_clusters)
3770 {
3771         int ret, credits;
3772         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
3773         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
3774         u64 start_blk = start_bh->b_blocknr, end_blk;
3775         u32 num_buckets = num_clusters * ocfs2_xattr_buckets_per_cluster(osb);
3776         handle_t *handle;
3777         struct ocfs2_xattr_header *first_xh =
3778                                 (struct ocfs2_xattr_header *)first_bh->b_data;
3779         u16 bucket = le16_to_cpu(first_xh->xh_num_buckets);
3780
3781         mlog(0, "extend xattr bucket in %llu, xattr extend rec starting "
3782              "from %llu, len = %u\n", start_blk,
3783              (unsigned long long)first_bh->b_blocknr, num_clusters);
3784
3785         BUG_ON(bucket >= num_buckets);
3786
3787         end_blk = first_bh->b_blocknr + (bucket - 1) * blk_per_bucket;
3788
3789         /*
3790          * We will touch all the buckets after the start_bh(include it).
3791          * Add one more bucket and modify the first_bh.
3792          */
3793         credits = end_blk - start_blk + 2 * blk_per_bucket + 1;
3794         handle = ocfs2_start_trans(osb, credits);
3795         if (IS_ERR(handle)) {
3796                 ret = PTR_ERR(handle);
3797                 handle = NULL;
3798                 mlog_errno(ret);
3799                 goto out;
3800         }
3801
3802         ret = ocfs2_journal_access(handle, inode, first_bh,
3803                                    OCFS2_JOURNAL_ACCESS_WRITE);
3804         if (ret) {
3805                 mlog_errno(ret);
3806                 goto commit;
3807         }
3808
3809         while (end_blk != start_blk) {
3810                 ret = ocfs2_cp_xattr_bucket(inode, handle, end_blk,
3811                                             end_blk + blk_per_bucket, 0);
3812                 if (ret)
3813                         goto commit;
3814                 end_blk -= blk_per_bucket;
3815         }
3816
3817         /* Move half of the xattr in start_blk to the next bucket. */
3818         ret = ocfs2_half_xattr_bucket(inode, handle, start_blk,
3819                                       start_blk + blk_per_bucket, NULL, 0);
3820
3821         le16_add_cpu(&first_xh->xh_num_buckets, 1);
3822         ocfs2_journal_dirty(handle, first_bh);
3823
3824 commit:
3825         ocfs2_commit_trans(osb, handle);
3826 out:
3827         return ret;
3828 }
3829
3830 /*
3831  * Add new xattr bucket in an extent record and adjust the buckets accordingly.
3832  * xb_bh is the ocfs2_xattr_block.
3833  * We will move all the buckets starting from header_bh to the next place. As
3834  * for this one, half num of its xattrs will be moved to the next one.
3835  *
3836  * We will allocate a new cluster if current cluster is full and adjust
3837  * header_bh and first_bh if the insert place is moved to the new cluster.
3838  */
3839 static int ocfs2_add_new_xattr_bucket(struct inode *inode,
3840                                       struct buffer_head *xb_bh,
3841                                       struct buffer_head *header_bh)
3842 {
3843         struct ocfs2_xattr_header *first_xh = NULL;
3844         struct buffer_head *first_bh = NULL;
3845         struct ocfs2_xattr_block *xb =
3846                         (struct ocfs2_xattr_block *)xb_bh->b_data;
3847         struct ocfs2_xattr_tree_root *xb_root = &xb->xb_attrs.xb_root;
3848         struct ocfs2_extent_list *el = &xb_root->xt_list;
3849         struct ocfs2_xattr_header *xh =
3850                         (struct ocfs2_xattr_header *)header_bh->b_data;
3851         u32 name_hash = le32_to_cpu(xh->xh_entries[0].xe_name_hash);
3852         struct super_block *sb = inode->i_sb;
3853         struct ocfs2_super *osb = OCFS2_SB(sb);
3854         int ret, num_buckets, extend = 1;
3855         u64 p_blkno;
3856         u32 e_cpos, num_clusters;
3857
3858         mlog(0, "Add new xattr bucket starting form %llu\n",
3859              (unsigned long long)header_bh->b_blocknr);
3860
3861         /*
3862          * Add refrence for header_bh here because it may be
3863          * changed in ocfs2_add_new_xattr_cluster and we need
3864          * to free it in the end.
3865          */
3866         get_bh(header_bh);
3867
3868         ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno, &e_cpos,
3869                                   &num_clusters, el);
3870         if (ret) {
3871                 mlog_errno(ret);
3872                 goto out;
3873         }
3874
3875         ret = ocfs2_read_block(osb, p_blkno,
3876                                &first_bh, OCFS2_BH_CACHED, inode);
3877         if (ret) {
3878                 mlog_errno(ret);
3879                 goto out;
3880         }
3881
3882         num_buckets = ocfs2_xattr_buckets_per_cluster(osb) * num_clusters;
3883         first_xh = (struct ocfs2_xattr_header *)first_bh->b_data;
3884
3885         if (num_buckets == le16_to_cpu(first_xh->xh_num_buckets)) {
3886                 ret = ocfs2_add_new_xattr_cluster(inode,
3887                                                   xb_bh,
3888                                                   &first_bh,
3889                                                   &header_bh,
3890                                                   &num_clusters,
3891                                                   e_cpos,
3892                                                   p_blkno,
3893                                                   &extend);
3894                 if (ret) {
3895                         mlog_errno(ret);
3896                         goto out;
3897                 }
3898         }
3899
3900         if (extend)
3901                 ret = ocfs2_extend_xattr_bucket(inode,
3902                                                 first_bh,
3903                                                 header_bh,
3904                                                 num_clusters);
3905         if (ret)
3906                 mlog_errno(ret);
3907 out:
3908         brelse(first_bh);
3909         brelse(header_bh);
3910         return ret;
3911 }
3912
3913 static inline char *ocfs2_xattr_bucket_get_val(struct inode *inode,
3914                                         struct ocfs2_xattr_bucket *bucket,
3915                                         int offs)
3916 {
3917         int block_off = offs >> inode->i_sb->s_blocksize_bits;
3918
3919         offs = offs % inode->i_sb->s_blocksize;
3920         return bucket->bhs[block_off]->b_data + offs;
3921 }
3922
3923 /*
3924  * Handle the normal xattr set, including replace, delete and new.
3925  *
3926  * Note: "local" indicates the real data's locality. So we can't
3927  * just its bucket locality by its length.
3928  */
3929 static void ocfs2_xattr_set_entry_normal(struct inode *inode,
3930                                          struct ocfs2_xattr_info *xi,
3931                                          struct ocfs2_xattr_search *xs,
3932                                          u32 name_hash,
3933                                          int local)
3934 {
3935         struct ocfs2_xattr_entry *last, *xe;
3936         int name_len = strlen(xi->name);
3937         struct ocfs2_xattr_header *xh = xs->header;
3938         u16 count = le16_to_cpu(xh->xh_count), start;
3939         size_t blocksize = inode->i_sb->s_blocksize;
3940         char *val;
3941         size_t offs, size, new_size;
3942
3943         last = &xh->xh_entries[count];
3944         if (!xs->not_found) {
3945                 xe = xs->here;
3946                 offs = le16_to_cpu(xe->xe_name_offset);
3947                 if (ocfs2_xattr_is_local(xe))
3948                         size = OCFS2_XATTR_SIZE(name_len) +
3949                         OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
3950                 else
3951                         size = OCFS2_XATTR_SIZE(name_len) +
3952                         OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
3953
3954                 /*
3955                  * If the new value will be stored outside, xi->value has been
3956                  * initalized as an empty ocfs2_xattr_value_root, and the same
3957                  * goes with xi->value_len, so we can set new_size safely here.
3958                  * See ocfs2_xattr_set_in_bucket.
3959                  */
3960                 new_size = OCFS2_XATTR_SIZE(name_len) +
3961                            OCFS2_XATTR_SIZE(xi->value_len);
3962
3963                 le16_add_cpu(&xh->xh_name_value_len, -size);
3964                 if (xi->value) {
3965                         if (new_size > size)
3966                                 goto set_new_name_value;
3967
3968                         /* Now replace the old value with new one. */
3969                         if (local)
3970                                 xe->xe_value_size = cpu_to_le64(xi->value_len);
3971                         else
3972                                 xe->xe_value_size = 0;
3973
3974                         val = ocfs2_xattr_bucket_get_val(inode,
3975                                                          &xs->bucket, offs);
3976                         memset(val + OCFS2_XATTR_SIZE(name_len), 0,
3977                                size - OCFS2_XATTR_SIZE(name_len));
3978                         if (OCFS2_XATTR_SIZE(xi->value_len) > 0)
3979                                 memcpy(val + OCFS2_XATTR_SIZE(name_len),
3980                                        xi->value, xi->value_len);
3981
3982                         le16_add_cpu(&xh->xh_name_value_len, new_size);
3983                         ocfs2_xattr_set_local(xe, local);
3984                         return;
3985                 } else {
3986                         /*
3987                          * Remove the old entry if there is more than one.
3988                          * We don't remove the last entry so that we can
3989                          * use it to indicate the hash value of the empty
3990                          * bucket.
3991                          */
3992                         last -= 1;
3993                         le16_add_cpu(&xh->xh_count, -1);
3994                         if (xh->xh_count) {
3995                                 memmove(xe, xe + 1,
3996                                         (void *)last - (void *)xe);
3997                                 memset(last, 0,
3998                                        sizeof(struct ocfs2_xattr_entry));
3999                         } else
4000                                 xh->xh_free_start =
4001                                         cpu_to_le16(OCFS2_XATTR_BUCKET_SIZE);
4002
4003                         return;
4004                 }
4005         } else {
4006                 /* find a new entry for insert. */
4007                 int low = 0, high = count - 1, tmp;
4008                 struct ocfs2_xattr_entry *tmp_xe;
4009
4010                 while (low <= high && count) {
4011                         tmp = (low + high) / 2;
4012                         tmp_xe = &xh->xh_entries[tmp];
4013
4014                         if (name_hash > le32_to_cpu(tmp_xe->xe_name_hash))
4015                                 low = tmp + 1;
4016                         else if (name_hash <
4017                                  le32_to_cpu(tmp_xe->xe_name_hash))
4018                                 high = tmp - 1;
4019                         else {
4020                                 low = tmp;
4021                                 break;
4022                         }
4023                 }
4024
4025                 xe = &xh->xh_entries[low];
4026                 if (low != count)
4027                         memmove(xe + 1, xe, (void *)last - (void *)xe);
4028
4029                 le16_add_cpu(&xh->xh_count, 1);
4030                 memset(xe, 0, sizeof(struct ocfs2_xattr_entry));
4031                 xe->xe_name_hash = cpu_to_le32(name_hash);
4032                 xe->xe_name_len = name_len;
4033                 ocfs2_xattr_set_type(xe, xi->name_index);
4034         }
4035
4036 set_new_name_value:
4037         /* Insert the new name+value. */
4038         size = OCFS2_XATTR_SIZE(name_len) + OCFS2_XATTR_SIZE(xi->value_len);
4039
4040         /*
4041          * We must make sure that the name/value pair
4042          * exists in the same block.
4043          */
4044         offs = le16_to_cpu(xh->xh_free_start);
4045         start = offs - size;
4046
4047         if (start >> inode->i_sb->s_blocksize_bits !=
4048             (offs - 1) >> inode->i_sb->s_blocksize_bits) {
4049                 offs = offs - offs % blocksize;
4050                 xh->xh_free_start = cpu_to_le16(offs);
4051         }
4052
4053         val = ocfs2_xattr_bucket_get_val(inode,
4054                                          &xs->bucket, offs - size);
4055         xe->xe_name_offset = cpu_to_le16(offs - size);
4056
4057         memset(val, 0, size);
4058         memcpy(val, xi->name, name_len);
4059         memcpy(val + OCFS2_XATTR_SIZE(name_len), xi->value, xi->value_len);
4060
4061         xe->xe_value_size = cpu_to_le64(xi->value_len);
4062         ocfs2_xattr_set_local(xe, local);
4063         xs->here = xe;
4064         le16_add_cpu(&xh->xh_free_start, -size);
4065         le16_add_cpu(&xh->xh_name_value_len, size);
4066
4067         return;
4068 }
4069
4070 static int ocfs2_xattr_bucket_handle_journal(struct inode *inode,
4071                                              handle_t *handle,
4072                                              struct ocfs2_xattr_search *xs,
4073                                              struct buffer_head **bhs,
4074                                              u16 bh_num)
4075 {
4076         int ret = 0, off, block_off;
4077         struct ocfs2_xattr_entry *xe = xs->here;
4078
4079         /*
4080          * First calculate all the blocks we should journal_access
4081          * and journal_dirty. The first block should always be touched.
4082          */
4083         ret = ocfs2_journal_dirty(handle, bhs[0]);
4084         if (ret)
4085                 mlog_errno(ret);
4086
4087         /* calc the data. */
4088         off = le16_to_cpu(xe->xe_name_offset);
4089         block_off = off >> inode->i_sb->s_blocksize_bits;
4090         ret = ocfs2_journal_dirty(handle, bhs[block_off]);
4091         if (ret)
4092                 mlog_errno(ret);
4093
4094         return ret;
4095 }
4096
4097 /*
4098  * Set the xattr entry in the specified bucket.
4099  * The bucket is indicated by xs->bucket and it should have the enough
4100  * space for the xattr insertion.
4101  */
4102 static int ocfs2_xattr_set_entry_in_bucket(struct inode *inode,
4103                                            struct ocfs2_xattr_info *xi,
4104                                            struct ocfs2_xattr_search *xs,
4105                                            u32 name_hash,
4106                                            int local)
4107 {
4108         int i, ret;
4109         handle_t *handle = NULL;
4110         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4111         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4112
4113         mlog(0, "Set xattr entry len = %lu index = %d in bucket %llu\n",
4114              (unsigned long)xi->value_len, xi->name_index,
4115              (unsigned long long)xs->bucket.bhs[0]->b_blocknr);
4116
4117         if (!xs->bucket.bhs[1]) {
4118                 ret = ocfs2_read_blocks(osb,
4119                                         xs->bucket.bhs[0]->b_blocknr + 1,
4120                                         blk_per_bucket - 1, &xs->bucket.bhs[1],
4121                                         OCFS2_BH_CACHED, inode);
4122                 if (ret) {
4123                         mlog_errno(ret);
4124                         goto out;
4125                 }
4126         }
4127
4128         handle = ocfs2_start_trans(osb, blk_per_bucket);
4129         if (IS_ERR(handle)) {
4130                 ret = PTR_ERR(handle);
4131                 handle = NULL;
4132                 mlog_errno(ret);
4133                 goto out;
4134         }
4135
4136         for (i = 0; i < blk_per_bucket; i++) {
4137                 ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[i],
4138                                            OCFS2_JOURNAL_ACCESS_WRITE);
4139                 if (ret < 0) {
4140                         mlog_errno(ret);
4141                         goto out;
4142                 }
4143         }
4144
4145         ocfs2_xattr_set_entry_normal(inode, xi, xs, name_hash, local);
4146
4147         /*Only dirty the blocks we have touched in set xattr. */
4148         ret = ocfs2_xattr_bucket_handle_journal(inode, handle, xs,
4149                                                 xs->bucket.bhs, blk_per_bucket);
4150         if (ret)
4151                 mlog_errno(ret);
4152 out:
4153         ocfs2_commit_trans(osb, handle);
4154
4155         return ret;
4156 }
4157
4158 static int ocfs2_xattr_value_update_size(struct inode *inode,
4159                                          struct buffer_head *xe_bh,
4160                                          struct ocfs2_xattr_entry *xe,
4161                                          u64 new_size)
4162 {
4163         int ret;
4164         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4165         handle_t *handle = NULL;
4166
4167         handle = ocfs2_start_trans(osb, 1);
4168         if (handle == NULL) {
4169                 ret = -ENOMEM;
4170                 mlog_errno(ret);
4171                 goto out;
4172         }
4173
4174         ret = ocfs2_journal_access(handle, inode, xe_bh,
4175                                    OCFS2_JOURNAL_ACCESS_WRITE);
4176         if (ret < 0) {
4177                 mlog_errno(ret);
4178                 goto out_commit;
4179         }
4180
4181         xe->xe_value_size = cpu_to_le64(new_size);
4182
4183         ret = ocfs2_journal_dirty(handle, xe_bh);
4184         if (ret < 0)
4185                 mlog_errno(ret);
4186
4187 out_commit:
4188         ocfs2_commit_trans(osb, handle);
4189 out:
4190         return ret;
4191 }
4192
4193 /*
4194  * Truncate the specified xe_off entry in xattr bucket.
4195  * bucket is indicated by header_bh and len is the new length.
4196  * Both the ocfs2_xattr_value_root and the entry will be updated here.
4197  *
4198  * Copy the new updated xe and xe_value_root to new_xe and new_xv if needed.
4199  */
4200 static int ocfs2_xattr_bucket_value_truncate(struct inode *inode,
4201                                              struct buffer_head *header_bh,
4202                                              int xe_off,
4203                                              int len)
4204 {
4205         int ret, offset;
4206         u64 value_blk;
4207         struct buffer_head *value_bh = NULL;
4208         struct ocfs2_xattr_value_root *xv;
4209         struct ocfs2_xattr_entry *xe;
4210         struct ocfs2_xattr_header *xh =
4211                         (struct ocfs2_xattr_header *)header_bh->b_data;
4212         size_t blocksize = inode->i_sb->s_blocksize;
4213
4214         xe = &xh->xh_entries[xe_off];
4215
4216         BUG_ON(!xe || ocfs2_xattr_is_local(xe));
4217
4218         offset = le16_to_cpu(xe->xe_name_offset) +
4219                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4220
4221         value_blk = offset / blocksize;
4222
4223         /* We don't allow ocfs2_xattr_value to be stored in different block. */
4224         BUG_ON(value_blk != (offset + OCFS2_XATTR_ROOT_SIZE - 1) / blocksize);
4225         value_blk += header_bh->b_blocknr;
4226
4227         ret = ocfs2_read_block(OCFS2_SB(inode->i_sb), value_blk,
4228                                &value_bh, OCFS2_BH_CACHED, inode);
4229         if (ret) {
4230                 mlog_errno(ret);
4231                 goto out;
4232         }
4233
4234         xv = (struct ocfs2_xattr_value_root *)
4235                 (value_bh->b_data + offset % blocksize);
4236
4237         mlog(0, "truncate %u in xattr bucket %llu to %d bytes.\n",
4238              xe_off, (unsigned long long)header_bh->b_blocknr, len);
4239         ret = ocfs2_xattr_value_truncate(inode, value_bh, xv, len);
4240         if (ret) {
4241                 mlog_errno(ret);
4242                 goto out;
4243         }
4244
4245         ret = ocfs2_xattr_value_update_size(inode, header_bh, xe, len);
4246         if (ret) {
4247                 mlog_errno(ret);
4248                 goto out;
4249         }
4250
4251 out:
4252         brelse(value_bh);
4253         return ret;
4254 }
4255
4256 static int ocfs2_xattr_bucket_value_truncate_xs(struct inode *inode,
4257                                                 struct ocfs2_xattr_search *xs,
4258                                                 int len)
4259 {
4260         int ret, offset;
4261         struct ocfs2_xattr_entry *xe = xs->here;
4262         struct ocfs2_xattr_header *xh = (struct ocfs2_xattr_header *)xs->base;
4263
4264         BUG_ON(!xs->bucket.bhs[0] || !xe || ocfs2_xattr_is_local(xe));
4265
4266         offset = xe - xh->xh_entries;
4267         ret = ocfs2_xattr_bucket_value_truncate(inode, xs->bucket.bhs[0],
4268                                                 offset, len);
4269         if (ret)
4270                 mlog_errno(ret);
4271
4272         return ret;
4273 }
4274
4275 static int ocfs2_xattr_bucket_set_value_outside(struct inode *inode,
4276                                                 struct ocfs2_xattr_search *xs,
4277                                                 char *val,
4278                                                 int value_len)
4279 {
4280         int offset;
4281         struct ocfs2_xattr_value_root *xv;
4282         struct ocfs2_xattr_entry *xe = xs->here;
4283
4284         BUG_ON(!xs->base || !xe || ocfs2_xattr_is_local(xe));
4285
4286         offset = le16_to_cpu(xe->xe_name_offset) +
4287                  OCFS2_XATTR_SIZE(xe->xe_name_len);
4288
4289         xv = (struct ocfs2_xattr_value_root *)(xs->base + offset);
4290
4291         return __ocfs2_xattr_set_value_outside(inode, xv, val, value_len);
4292 }
4293
4294 static int ocfs2_rm_xattr_cluster(struct inode *inode,
4295                                   struct buffer_head *root_bh,
4296                                   u64 blkno,
4297                                   u32 cpos,
4298                                   u32 len)
4299 {
4300         int ret;
4301         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4302         struct inode *tl_inode = osb->osb_tl_inode;
4303         handle_t *handle;
4304         struct ocfs2_xattr_block *xb =
4305                         (struct ocfs2_xattr_block *)root_bh->b_data;
4306         struct ocfs2_alloc_context *meta_ac = NULL;
4307         struct ocfs2_cached_dealloc_ctxt dealloc;
4308         struct ocfs2_extent_tree et;
4309
4310         ocfs2_init_xattr_tree_extent_tree(&et, inode, root_bh);
4311
4312         ocfs2_init_dealloc_ctxt(&dealloc);
4313
4314         mlog(0, "rm xattr extent rec at %u len = %u, start from %llu\n",
4315              cpos, len, (unsigned long long)blkno);
4316
4317         ocfs2_remove_xattr_clusters_from_cache(inode, blkno, len);
4318
4319         ret = ocfs2_lock_allocators(inode, &et, 0, 1, NULL, &meta_ac);
4320         if (ret) {
4321                 mlog_errno(ret);
4322                 return ret;
4323         }
4324
4325         mutex_lock(&tl_inode->i_mutex);
4326
4327         if (ocfs2_truncate_log_needs_flush(osb)) {
4328                 ret = __ocfs2_flush_truncate_log(osb);
4329                 if (ret < 0) {
4330                         mlog_errno(ret);
4331                         goto out;
4332                 }
4333         }
4334
4335         handle = ocfs2_start_trans(osb, OCFS2_REMOVE_EXTENT_CREDITS);
4336         if (handle == NULL) {
4337                 ret = -ENOMEM;
4338                 mlog_errno(ret);
4339                 goto out;
4340         }
4341
4342         ret = ocfs2_journal_access(handle, inode, root_bh,
4343                                    OCFS2_JOURNAL_ACCESS_WRITE);
4344         if (ret) {
4345                 mlog_errno(ret);
4346                 goto out_commit;
4347         }
4348
4349         ret = ocfs2_remove_extent(inode, &et, cpos, len, handle, meta_ac,
4350                                   &dealloc);
4351         if (ret) {
4352                 mlog_errno(ret);
4353                 goto out_commit;
4354         }
4355
4356         le32_add_cpu(&xb->xb_attrs.xb_root.xt_clusters, -len);
4357
4358         ret = ocfs2_journal_dirty(handle, root_bh);
4359         if (ret) {
4360                 mlog_errno(ret);
4361                 goto out_commit;
4362         }
4363
4364         ret = ocfs2_truncate_log_append(osb, handle, blkno, len);
4365         if (ret)
4366                 mlog_errno(ret);
4367
4368 out_commit:
4369         ocfs2_commit_trans(osb, handle);
4370 out:
4371         ocfs2_schedule_truncate_log_flush(osb, 1);
4372
4373         mutex_unlock(&tl_inode->i_mutex);
4374
4375         if (meta_ac)
4376                 ocfs2_free_alloc_context(meta_ac);
4377
4378         ocfs2_run_deallocs(osb, &dealloc);
4379
4380         return ret;
4381 }
4382
4383 static void ocfs2_xattr_bucket_remove_xs(struct inode *inode,
4384                                          struct ocfs2_xattr_search *xs)
4385 {
4386         handle_t *handle = NULL;
4387         struct ocfs2_xattr_header *xh = xs->bucket.xh;
4388         struct ocfs2_xattr_entry *last = &xh->xh_entries[
4389                                                 le16_to_cpu(xh->xh_count) - 1];
4390         int ret = 0;
4391
4392         handle = ocfs2_start_trans((OCFS2_SB(inode->i_sb)), 1);
4393         if (IS_ERR(handle)) {
4394                 ret = PTR_ERR(handle);
4395                 mlog_errno(ret);
4396                 return;
4397         }
4398
4399         ret = ocfs2_journal_access(handle, inode, xs->bucket.bhs[0],
4400                                    OCFS2_JOURNAL_ACCESS_WRITE);
4401         if (ret) {
4402                 mlog_errno(ret);
4403                 goto out_commit;
4404         }
4405
4406         /* Remove the old entry. */
4407         memmove(xs->here, xs->here + 1,
4408                 (void *)last - (void *)xs->here);
4409         memset(last, 0, sizeof(struct ocfs2_xattr_entry));
4410         le16_add_cpu(&xh->xh_count, -1);
4411
4412         ret = ocfs2_journal_dirty(handle, xs->bucket.bhs[0]);
4413         if (ret < 0)
4414                 mlog_errno(ret);
4415 out_commit:
4416         ocfs2_commit_trans(OCFS2_SB(inode->i_sb), handle);
4417 }
4418
4419 /*
4420  * Set the xattr name/value in the bucket specified in xs.
4421  *
4422  * As the new value in xi may be stored in the bucket or in an outside cluster,
4423  * we divide the whole process into 3 steps:
4424  * 1. insert name/value in the bucket(ocfs2_xattr_set_entry_in_bucket)
4425  * 2. truncate of the outside cluster(ocfs2_xattr_bucket_value_truncate_xs)
4426  * 3. Set the value to the outside cluster(ocfs2_xattr_bucket_set_value_outside)
4427  * 4. If the clusters for the new outside value can't be allocated, we need
4428  *    to free the xattr we allocated in set.
4429  */
4430 static int ocfs2_xattr_set_in_bucket(struct inode *inode,
4431                                      struct ocfs2_xattr_info *xi,
4432                                      struct ocfs2_xattr_search *xs)
4433 {
4434         int ret, local = 1;
4435         size_t value_len;
4436         char *val = (char *)xi->value;
4437         struct ocfs2_xattr_entry *xe = xs->here;
4438         u32 name_hash = ocfs2_xattr_name_hash(inode, xi->name,
4439                                               strlen(xi->name));
4440
4441         if (!xs->not_found && !ocfs2_xattr_is_local(xe)) {
4442                 /*
4443                  * We need to truncate the xattr storage first.
4444                  *
4445                  * If both the old and new value are stored to
4446                  * outside block, we only need to truncate
4447                  * the storage and then set the value outside.
4448                  *
4449                  * If the new value should be stored within block,
4450                  * we should free all the outside block first and
4451                  * the modification to the xattr block will be done
4452                  * by following steps.
4453                  */
4454                 if (xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4455                         value_len = xi->value_len;
4456                 else
4457                         value_len = 0;
4458
4459                 ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4460                                                            value_len);
4461                 if (ret)
4462                         goto out;
4463
4464                 if (value_len)
4465                         goto set_value_outside;
4466         }
4467
4468         value_len = xi->value_len;
4469         /* So we have to handle the inside block change now. */
4470         if (value_len > OCFS2_XATTR_INLINE_SIZE) {
4471                 /*
4472                  * If the new value will be stored outside of block,
4473                  * initalize a new empty value root and insert it first.
4474                  */
4475                 local = 0;
4476                 xi->value = &def_xv;
4477                 xi->value_len = OCFS2_XATTR_ROOT_SIZE;
4478         }
4479
4480         ret = ocfs2_xattr_set_entry_in_bucket(inode, xi, xs, name_hash, local);
4481         if (ret) {
4482                 mlog_errno(ret);
4483                 goto out;
4484         }
4485
4486         if (value_len <= OCFS2_XATTR_INLINE_SIZE)
4487                 goto out;
4488
4489         /* allocate the space now for the outside block storage. */
4490         ret = ocfs2_xattr_bucket_value_truncate_xs(inode, xs,
4491                                                    value_len);
4492         if (ret) {
4493                 mlog_errno(ret);
4494
4495                 if (xs->not_found) {
4496                         /*
4497                          * We can't allocate enough clusters for outside
4498                          * storage and we have allocated xattr already,
4499                          * so need to remove it.
4500                          */
4501                         ocfs2_xattr_bucket_remove_xs(inode, xs);
4502                 }
4503                 goto out;
4504         }
4505
4506 set_value_outside:
4507         ret = ocfs2_xattr_bucket_set_value_outside(inode, xs, val, value_len);
4508 out:
4509         return ret;
4510 }
4511
4512 /* check whether the xattr bucket is filled up with the same hash value. */
4513 static int ocfs2_check_xattr_bucket_collision(struct inode *inode,
4514                                               struct ocfs2_xattr_bucket *bucket)
4515 {
4516         struct ocfs2_xattr_header *xh = bucket->xh;
4517
4518         if (xh->xh_entries[le16_to_cpu(xh->xh_count) - 1].xe_name_hash ==
4519             xh->xh_entries[0].xe_name_hash) {
4520                 mlog(ML_ERROR, "Too much hash collision in xattr bucket %llu, "
4521                      "hash = %u\n",
4522                      (unsigned long long)bucket->bhs[0]->b_blocknr,
4523                      le32_to_cpu(xh->xh_entries[0].xe_name_hash));
4524                 return -ENOSPC;
4525         }
4526
4527         return 0;
4528 }
4529
4530 static int ocfs2_xattr_set_entry_index_block(struct inode *inode,
4531                                              struct ocfs2_xattr_info *xi,
4532                                              struct ocfs2_xattr_search *xs)
4533 {
4534         struct ocfs2_xattr_header *xh;
4535         struct ocfs2_xattr_entry *xe;
4536         u16 count, header_size, xh_free_start;
4537         int i, free, max_free, need, old;
4538         size_t value_size = 0, name_len = strlen(xi->name);
4539         size_t blocksize = inode->i_sb->s_blocksize;
4540         int ret, allocation = 0;
4541         u16 blk_per_bucket = ocfs2_blocks_per_xattr_bucket(inode->i_sb);
4542
4543         mlog_entry("Set xattr %s in xattr index block\n", xi->name);
4544
4545 try_again:
4546         xh = xs->header;
4547         count = le16_to_cpu(xh->xh_count);
4548         xh_free_start = le16_to_cpu(xh->xh_free_start);
4549         header_size = sizeof(struct ocfs2_xattr_header) +
4550                         count * sizeof(struct ocfs2_xattr_entry);
4551         max_free = OCFS2_XATTR_BUCKET_SIZE -
4552                 le16_to_cpu(xh->xh_name_value_len) - header_size;
4553
4554         mlog_bug_on_msg(header_size > blocksize, "bucket %llu has header size "
4555                         "of %u which exceed block size\n",
4556                         (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4557                         header_size);
4558
4559         if (xi->value && xi->value_len > OCFS2_XATTR_INLINE_SIZE)
4560                 value_size = OCFS2_XATTR_ROOT_SIZE;
4561         else if (xi->value)
4562                 value_size = OCFS2_XATTR_SIZE(xi->value_len);
4563
4564         if (xs->not_found)
4565                 need = sizeof(struct ocfs2_xattr_entry) +
4566                         OCFS2_XATTR_SIZE(name_len) + value_size;
4567         else {
4568                 need = value_size + OCFS2_XATTR_SIZE(name_len);
4569
4570                 /*
4571                  * We only replace the old value if the new length is smaller
4572                  * than the old one. Otherwise we will allocate new space in the
4573                  * bucket to store it.
4574                  */
4575                 xe = xs->here;
4576                 if (ocfs2_xattr_is_local(xe))
4577                         old = OCFS2_XATTR_SIZE(le64_to_cpu(xe->xe_value_size));
4578                 else
4579                         old = OCFS2_XATTR_SIZE(OCFS2_XATTR_ROOT_SIZE);
4580
4581                 if (old >= value_size)
4582                         need = 0;
4583         }
4584
4585         free = xh_free_start - header_size;
4586         /*
4587          * We need to make sure the new name/value pair
4588          * can exist in the same block.
4589          */
4590         if (xh_free_start % blocksize < need)
4591                 free -= xh_free_start % blocksize;
4592
4593         mlog(0, "xs->not_found = %d, in xattr bucket %llu: free = %d, "
4594              "need = %d, max_free = %d, xh_free_start = %u, xh_name_value_len ="
4595              " %u\n", xs->not_found,
4596              (unsigned long long)xs->bucket.bhs[0]->b_blocknr,
4597              free, need, max_free, le16_to_cpu(xh->xh_free_start),
4598              le16_to_cpu(xh->xh_name_value_len));
4599
4600         if (free < need || count == ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4601                 if (need <= max_free &&
4602                     count < ocfs2_xattr_max_xe_in_bucket(inode->i_sb)) {
4603                         /*
4604                          * We can create the space by defragment. Since only the
4605                          * name/value will be moved, the xe shouldn't be changed
4606                          * in xs.
4607                          */
4608                         ret = ocfs2_defrag_xattr_bucket(inode, &xs->bucket);
4609                         if (ret) {
4610                                 mlog_errno(ret);
4611                                 goto out;
4612                         }
4613
4614                         xh_free_start = le16_to_cpu(xh->xh_free_start);
4615                         free = xh_free_start - header_size;
4616                         if (xh_free_start % blocksize < need)
4617                                 free -= xh_free_start % blocksize;
4618
4619                         if (free >= need)
4620                                 goto xattr_set;
4621
4622                         mlog(0, "Can't get enough space for xattr insert by "
4623                              "defragment. Need %u bytes, but we have %d, so "
4624                              "allocate new bucket for it.\n", need, free);
4625                 }
4626
4627                 /*
4628                  * We have to add new buckets or clusters and one
4629                  * allocation should leave us enough space for insert.
4630                  */
4631                 BUG_ON(allocation);
4632
4633                 /*
4634                  * We do not allow for overlapping ranges between buckets. And
4635                  * the maximum number of collisions we will allow for then is
4636                  * one bucket's worth, so check it here whether we need to
4637                  * add a new bucket for the insert.
4638                  */
4639                 ret = ocfs2_check_xattr_bucket_collision(inode, &xs->bucket);
4640                 if (ret) {
4641                         mlog_errno(ret);
4642                         goto out;
4643                 }
4644
4645                 ret = ocfs2_add_new_xattr_bucket(inode,
4646                                                  xs->xattr_bh,
4647                                                  xs->bucket.bhs[0]);
4648                 if (ret) {
4649                         mlog_errno(ret);
4650                         goto out;
4651                 }
4652
4653                 for (i = 0; i < blk_per_bucket; i++)
4654                         brelse(xs->bucket.bhs[i]);
4655
4656                 memset(&xs->bucket, 0, sizeof(xs->bucket));
4657
4658                 ret = ocfs2_xattr_index_block_find(inode, xs->xattr_bh,
4659                                                    xi->name_index,
4660                                                    xi->name, xs);
4661                 if (ret && ret != -ENODATA)
4662                         goto out;
4663                 xs->not_found = ret;
4664                 allocation = 1;
4665                 goto try_again;
4666         }
4667
4668 xattr_set:
4669         ret = ocfs2_xattr_set_in_bucket(inode, xi, xs);
4670 out:
4671         mlog_exit(ret);
4672         return ret;
4673 }
4674
4675 static int ocfs2_delete_xattr_in_bucket(struct inode *inode,
4676                                         struct ocfs2_xattr_bucket *bucket,
4677                                         void *para)
4678 {
4679         int ret = 0;
4680         struct ocfs2_xattr_header *xh = bucket->xh;
4681         u16 i;
4682         struct ocfs2_xattr_entry *xe;
4683
4684         for (i = 0; i < le16_to_cpu(xh->xh_count); i++) {
4685                 xe = &xh->xh_entries[i];
4686                 if (ocfs2_xattr_is_local(xe))
4687                         continue;
4688
4689                 ret = ocfs2_xattr_bucket_value_truncate(inode,
4690                                                         bucket->bhs[0],
4691                                                         i, 0);
4692                 if (ret) {
4693                         mlog_errno(ret);
4694                         break;
4695                 }
4696         }
4697
4698         return ret;
4699 }
4700
4701 static int ocfs2_delete_xattr_index_block(struct inode *inode,
4702                                           struct buffer_head *xb_bh)
4703 {
4704         struct ocfs2_xattr_block *xb =
4705                         (struct ocfs2_xattr_block *)xb_bh->b_data;
4706         struct ocfs2_extent_list *el = &xb->xb_attrs.xb_root.xt_list;
4707         int ret = 0;
4708         u32 name_hash = UINT_MAX, e_cpos, num_clusters;
4709         u64 p_blkno;
4710
4711         if (le16_to_cpu(el->l_next_free_rec) == 0)
4712                 return 0;
4713
4714         while (name_hash > 0) {
4715                 ret = ocfs2_xattr_get_rec(inode, name_hash, &p_blkno,
4716                                           &e_cpos, &num_clusters, el);
4717                 if (ret) {
4718                         mlog_errno(ret);
4719                         goto out;
4720                 }
4721
4722                 ret = ocfs2_iterate_xattr_buckets(inode, p_blkno, num_clusters,
4723                                                   ocfs2_delete_xattr_in_bucket,
4724                                                   NULL);
4725                 if (ret) {
4726                         mlog_errno(ret);
4727                         goto out;
4728                 }
4729
4730                 ret = ocfs2_rm_xattr_cluster(inode, xb_bh,
4731                                              p_blkno, e_cpos, num_clusters);
4732                 if (ret) {
4733                         mlog_errno(ret);
4734                         break;
4735                 }
4736
4737                 if (e_cpos == 0)
4738                         break;
4739
4740                 name_hash = e_cpos - 1;
4741         }
4742
4743 out:
4744         return ret;
4745 }
4746
4747 /*
4748  * 'trusted' attributes support
4749  */
4750
4751 #define XATTR_TRUSTED_PREFIX "trusted."
4752
4753 static size_t ocfs2_xattr_trusted_list(struct inode *inode, char *list,
4754                                        size_t list_size, const char *name,
4755                                        size_t name_len)
4756 {
4757         const size_t prefix_len = sizeof(XATTR_TRUSTED_PREFIX) - 1;
4758         const size_t total_len = prefix_len + name_len + 1;
4759
4760         if (list && total_len <= list_size) {
4761                 memcpy(list, XATTR_TRUSTED_PREFIX, prefix_len);
4762                 memcpy(list + prefix_len, name, name_len);
4763                 list[prefix_len + name_len] = '\0';
4764         }
4765         return total_len;
4766 }
4767
4768 static int ocfs2_xattr_trusted_get(struct inode *inode, const char *name,
4769                                    void *buffer, size_t size)
4770 {
4771         if (strcmp(name, "") == 0)
4772                 return -EINVAL;
4773         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_TRUSTED, name,
4774                                buffer, size);
4775 }
4776
4777 static int ocfs2_xattr_trusted_set(struct inode *inode, const char *name,
4778                                    const void *value, size_t size, int flags)
4779 {
4780         if (strcmp(name, "") == 0)
4781                 return -EINVAL;
4782
4783         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_TRUSTED, name, value,
4784                                size, flags);
4785 }
4786
4787 struct xattr_handler ocfs2_xattr_trusted_handler = {
4788         .prefix = XATTR_TRUSTED_PREFIX,
4789         .list   = ocfs2_xattr_trusted_list,
4790         .get    = ocfs2_xattr_trusted_get,
4791         .set    = ocfs2_xattr_trusted_set,
4792 };
4793
4794
4795 /*
4796  * 'user' attributes support
4797  */
4798
4799 #define XATTR_USER_PREFIX "user."
4800
4801 static size_t ocfs2_xattr_user_list(struct inode *inode, char *list,
4802                                     size_t list_size, const char *name,
4803                                     size_t name_len)
4804 {
4805         const size_t prefix_len = sizeof(XATTR_USER_PREFIX) - 1;
4806         const size_t total_len = prefix_len + name_len + 1;
4807         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4808
4809         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4810                 return 0;
4811
4812         if (list && total_len <= list_size) {
4813                 memcpy(list, XATTR_USER_PREFIX, prefix_len);
4814                 memcpy(list + prefix_len, name, name_len);
4815                 list[prefix_len + name_len] = '\0';
4816         }
4817         return total_len;
4818 }
4819
4820 static int ocfs2_xattr_user_get(struct inode *inode, const char *name,
4821                                 void *buffer, size_t size)
4822 {
4823         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4824
4825         if (strcmp(name, "") == 0)
4826                 return -EINVAL;
4827         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4828                 return -EOPNOTSUPP;
4829         return ocfs2_xattr_get(inode, OCFS2_XATTR_INDEX_USER, name,
4830                                buffer, size);
4831 }
4832
4833 static int ocfs2_xattr_user_set(struct inode *inode, const char *name,
4834                                 const void *value, size_t size, int flags)
4835 {
4836         struct ocfs2_super *osb = OCFS2_SB(inode->i_sb);
4837
4838         if (strcmp(name, "") == 0)
4839                 return -EINVAL;
4840         if (osb->s_mount_opt & OCFS2_MOUNT_NOUSERXATTR)
4841                 return -EOPNOTSUPP;
4842
4843         return ocfs2_xattr_set(inode, OCFS2_XATTR_INDEX_USER, name, value,
4844                                size, flags);
4845 }
4846
4847 struct xattr_handler ocfs2_xattr_user_handler = {
4848         .prefix = XATTR_USER_PREFIX,
4849         .list   = ocfs2_xattr_user_list,
4850         .get    = ocfs2_xattr_user_get,
4851         .set    = ocfs2_xattr_user_set,
4852 };